UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays on economic development in India Paul, Sourabh Bikas 2011

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2011_fall_paul_sourabh.pdf [ 1.66MB ]
Metadata
JSON: 24-1.0072200.json
JSON-LD: 24-1.0072200-ld.json
RDF/XML (Pretty): 24-1.0072200-rdf.xml
RDF/JSON: 24-1.0072200-rdf.json
Turtle: 24-1.0072200-turtle.txt
N-Triples: 24-1.0072200-rdf-ntriples.txt
Original Record: 24-1.0072200-source.json
Full Text
24-1.0072200-fulltext.txt
Citation
24-1.0072200.ris

Full Text

Essays on Economic Development in India by Sourabh Bikas Paul B.Sc., The University of Calcutta, India, 1998 M.S.(Q.E.), Indian Statistical Institute, India, 2000 M.Phil., Indira Gandhi Institute of Development Research, India, 2004 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in The Faculty of Graduate Studies (Economics) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) September 2011 c© Sourabh Bikas Paul 2011 Abstract My research is an empirical investigation of how some recent changes in the Indian economy have affected the most vulnerable sections of Indian society. The thesis has three chapters. The first chapter examines the impact of the tariff reductions undertaken in 1991 across different consumption groups. I evaluate the distributional impact of tariff reforms in India using household survey data. I estimate the overall gains coming from general equilibrium effects of the commodity market and labour market adjustment; all consumption groups have significant welfare gains. In addition, it appears that tariff reforms have a pro-poor distributional effect in rural areas and a pro-rich distributional effect in urban areas. The second chapter deals with income opportunities of underprivileged classes in India. Can large macroeconomic changes also alter the historical economic mobility patterns of various social groups? We examine this question by contrasting the fortunes of the historically disadvantaged scheduled castes and tribes (SC/ST) in India with the rest of the workforce in terms of their education attainment, occupation choices and wages. Our key findings are that wages have been converging across the two groups with rising education attainments accounting for the majority of this convergence. SC/STs have also been switching occupations at increasing rates during this period. Moreover, inter-generational education and income mobility rates of SC/STs have converged to non-SC/ST levels. In the third chapter, I present some estimates for India that demonstrate that struc- tural changes in the impact of income on food demand have been significant factors driving the changes in dietary patterns in this rapidly growing economy. A Quadratic Almost Ideal Demand System is estimated for six food groups. The estimation results confirm a shift in taste of both rural and urban households that explains low demand for nutrient-rich inexpensive food and a greater variety of expensive sources of nutri- ents. The quality of diet has been falling in terms of excessive fat intake with no sign of significant improvement in diet quality in terms of other nutrients. ii Preface Chapter 3, titled ”Castes and Labour Mobility” is based on collaborative research with Prof. Amartya Lahiri and Prof. Viktoria Hnatkovska at the department of economics, University of British Columbia. I was responsible for extracting and arranging the data for statistical analysis. The data analysis, interpretation and presentation of the research are equally shared by the co-authors. iii Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 Distributional Effects of Tariff Reforms in India . . . . . . . . . . . . . 5 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.2 Trade reform in India . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.3 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.5 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.5.1 Price change of traded goods . . . . . . . . . . . . . . . . . . . . 17 2.5.2 Consumption effects of traded goods . . . . . . . . . . . . . . . . 18 2.5.3 Consumption effects of non-traded goods . . . . . . . . . . . . . 21 2.5.4 Labour income effects . . . . . . . . . . . . . . . . . . . . . . . . 26 2.5.5 Total distributional effect . . . . . . . . . . . . . . . . . . . . . . 31 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3 Castes and Labour Mobility . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 iv 3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 3.3 Intragenerational cohort comparison . . . . . . . . . . . . . . . . . . . . 43 3.3.1 Education attainment . . . . . . . . . . . . . . . . . . . . . . . . 44 3.3.2 Occupation choices . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.3.3 Industry choices . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 3.3.4 Wages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.3.5 Sample and robustness . . . . . . . . . . . . . . . . . . . . . . . 65 3.4 Intergenerational mobility . . . . . . . . . . . . . . . . . . . . . . . . . . 66 3.4.1 Education mobility . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3.4.2 Occupation mobility . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.4.3 Industry mobility . . . . . . . . . . . . . . . . . . . . . . . . . . 72 3.4.4 Income mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4 Food Preference and Nutrition in India . . . . . . . . . . . . . . . . . . 79 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 4.3 Food share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 4.4 Food demand system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 4.5 Nutrition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Appendix A: Appendix to chapter 3 . . . . . . . . . . . . . . . . . . . . . . . 131 Appendix B: Appendix to chapter 4 . . . . . . . . . . . . . . . . . . . . . . . 137 v List of Tables 2.1 Tariff structure in India . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.2 Tariff structure in 1990 and 2005 . . . . . . . . . . . . . . . . . . . . . . 18 2.3 Response of prices of non-traded goods . . . . . . . . . . . . . . . . . . . 23 2.4 Regression of log wage on prices interacted with education dummies . . . 29 3.1 Sample summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.2 Education attainment levels and gaps . . . . . . . . . . . . . . . . . . . 45 3.3 Education attainment levels and gaps by occupations . . . . . . . . . . . 51 3.4 Wage gaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 3.5 Conditional wage regressions . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.6 Oaxaca-Blinder decomposition . . . . . . . . . . . . . . . . . . . . . . . . 64 3.7 Intergenerational occupation transition probabilities . . . . . . . . . . . . 71 3.8 Intergenerational industry transition probabilities . . . . . . . . . . . . . 74 4.1 Mean and SD of food shares, real MPCE and ln(realMPCE) . . . . . . . 85 4.2 Multivariate model of food share . . . . . . . . . . . . . . . . . . . . . . 94 4.3 Multivariate model of food share - pooled cross section . . . . . . . . . . 96 4.4 Estimation of expenditure and own price elasticities - rural . . . . . . . . 106 4.5 Estimation of expenditure and own price elasticities - urban . . . . . . . 107 4.6 Proportion of households meeting RDA . . . . . . . . . . . . . . . . . . . 114 4.7 Average diet quality index . . . . . . . . . . . . . . . . . . . . . . . . . . 116 4.8 Censored regression model of diet quality index . . . . . . . . . . . . . . 119 A.1 Occupation categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 A.2 Industry categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 A.3 Intergenerational education improvements and reductions . . . . . . . . 136 B.1 Frequency distribution (%) of households classified by food share . . . . . 137 B.2 Food share by selected states in top and bottom quartile - rural . . . . . 138 B.3 Food share by selected states in top and bottom quartile - urban . . . . . 139 B.4 Mean and SD of food group shares . . . . . . . . . . . . . . . . . . . . . 140 vi B.5 Multivariate model of food share . . . . . . . . . . . . . . . . . . . . . . 141 B.6 Multivariate model of food share with region dummies . . . . . . . . . . 142 B.7 Mean and SD of food group quantity (per capita, per month) . . . . . . . 144 B.8 Number of households with non-zero share . . . . . . . . . . . . . . . . . 144 B.9 Estimation of QAIDS model . . . . . . . . . . . . . . . . . . . . . . . . . 146 B.10 Per capita per diem intake . . . . . . . . . . . . . . . . . . . . . . . . . . 152 B.11 Nutrient share from different food groups - rural . . . . . . . . . . . . . . 153 B.12 Nutrient share from different food groups - urban . . . . . . . . . . . . . 154 B.13 Prices per nutrient from different food groups - rural . . . . . . . . . . . 155 B.14 Prices per nutrient from different food groups - urban . . . . . . . . . . . 156 vii List of Figures 2.1 Simple and weighted average tariff of different consumption item groups . 9 2.2 Share of export, import and merchandise trade in GDP . . . . . . . . . . 10 2.3 Consumption effects of traded goods . . . . . . . . . . . . . . . . . . . . 20 2.4 Consumption effects of traded goods -rural, urban . . . . . . . . . . . . . 21 2.5 Consumption effects of non-traded goods . . . . . . . . . . . . . . . . . . 24 2.6 Consumption effects of non-traded goods -rural, urban . . . . . . . . . . 25 2.7 Labour income effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 2.8 Labour income effects -rural, urban . . . . . . . . . . . . . . . . . . . . . 30 2.9 Total effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.10 Total effect -rural, urban . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.1 Education distribution of children and parents . . . . . . . . . . . . . . 46 3.2 Education gaps by age cohorts . . . . . . . . . . . . . . . . . . . . . . . 48 3.3 Occupation distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 3.4 Education gaps by age cohorts and occupations . . . . . . . . . . . . . . 52 3.5 Industry distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.6 Wage density for non-SC/STs and SC/STs . . . . . . . . . . . . . . . . 56 3.7 Differences in percentiles for non-SC/STs and SC/STs for log wages and log consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.8 Wage gaps by age cohorts . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3.9 Wage gaps by age cohorts and occupations . . . . . . . . . . . . . . . . 61 3.10 Robustness to sample choice . . . . . . . . . . . . . . . . . . . . . . . . 66 3.11 Intergenerational education switches . . . . . . . . . . . . . . . . . . . . 68 3.12 Intergenerational occupation switches . . . . . . . . . . . . . . . . . . . 70 3.13 Intergenerational industry switches . . . . . . . . . . . . . . . . . . . . . 73 3.14 Intergenerational income mobility . . . . . . . . . . . . . . . . . . . . . . 76 4.1 MPCE distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 4.2 Food share distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 viii 4.3 Food share by state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 4.4 Joint distribution of MPCE and food share . . . . . . . . . . . . . . . . . 88 4.5 Engel curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.6 Engel curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 4.7 Partial effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 4.8 Relative change in food groups share (1983 to 2004-05) . . . . . . . . . . 97 4.9 Relative change (%) in food consumption (1983 to 2004-05) . . . . . . . . 98 4.10 Composition of food groups (1983 to 2004-05) . . . . . . . . . . . . . . . 98 4.11 Non-parametric and quadratic Engel curves for food groups . . . . . . . . 102 4.12 Trend in energy and macronutrients intake in India . . . . . . . . . . . . 110 4.13 Calorie Engel curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.14 Calorie elasticity curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.15 MPCE and diet quality index . . . . . . . . . . . . . . . . . . . . . . . . 118 B.1 Engel curve - household composition . . . . . . . . . . . . . . . . . . . . 143 B.2 Non-parametric and quadratic Engel curves for food groups . . . . . . . . 145 B.3 Distribution of households by ratio of nutrients intake to RDA . . . . . . 150 B.4 MPCE and nutrients intake as proportion to RDA . . . . . . . . . . . . . 151 ix Acknowledgements I would like to express my gratitude to thesis supervisor, Prof. Ashok Kotwal for constant support and advice at every stage of my graduate studies at the University of British Columbia. My thesis advisory committee members, Prof. Amartya Lahiri and Prof. Patrick Francois have always been sources of helpful suggestions. I would like to thank my co-authors Prof. Viktoria Hnatkovska and Prof. Amartya Lahiri for giving me this opportunity to undertake collaborative research. I am grateful to Prof. Bharat Ramaswamy at the Indian Statistical Institute, New Delhi for helpful suggestions at the early stage of this thesis and for providing excellent research facility at the institute. I had many illuminating discussions with my friends here at the University of British Columbia which helped me in different ways in writing this thesis. I owe special thanks to Souvik Datta, Anirban Mukherjee, Subrata Sarker, Nishant Chadha and Arka Roy Chaudhuri. I am grateful to my friend Nabin Jana for his support at various stages of this thesis. Finally, I must acknowledge the contribution made towards this thesis by my family, especially my wife Sohini Paul, who had to bear with all the family responsibilities during my graduate studies. x Dedication To My Gurudeva, Shrii Shrii Anandamurtiji. xi 1. Introduction In the past several decades, India has experienced rapid economic growth and struc- tural transformation. Considering a sluggish annual rate of growth around 3 to 4 per- cent for three decades after independence in 1947, the jump to almost 6 percent annual growth rate for the period 1980-2005 is quite remarkable. Since 1980, the annual per capita GDP growth rate became more than double, rising from 1.7 percent during 1950- 1980 to 3.98 percent during 1980-2005. In a country as poor as India, income growth is desirable because it has a significant impact on human development indicators includ- ing poverty, earning opportunities and health status. There is considerable evidence of significant progress in some of these aspects. In addition to achieving rapid growth rate, poverty levels declined from 44.48 percent in 1983-84 to 27.5 percent in 2004-05. But little is known about the distributional consequences of rapid growth and structural shift. The broader question is: Have all segments of Indian society benefitted equally from these changes? The rapid economic growth and poverty reduction in the past quarter of a century were accompanied by structural transformations directly linked to some of the key policy measures undertaken during the same period. On the economic front we have seen a quick succession of several measures to open up the economy which led to increasing the rate of transfer of goods, technology and ideas from developed countries. We also observe the Indian urban lifestyle rapidly embracing a globalised culture befitting its emerging status. As a democratic state with socialist goals, Indian policymakers have always been concerned with distributional consequences from the early years of planning. My thesis is an attempt to examine how some of the recent changes in the Indian economy have affected different social groups, with special emphasis on the most vulnerable sections. In particular, I examine three important aspects of the Indian development experience, viz., 1) How did trade reforms in 1991 affect different income groups across geographical regions? 2) Do large-scale macroeconomic changes tend to accentuate or dampen historical inequities between caste groups? and 3) Do we observe any progress in the nutritional status of different income groups due to rapid economic transformation? 1 The literature on the distributional impact of tariff reforms and macroeconomic struc- tural changes is far from conclusive. Trade theory suggests that in developing countries abundant unskilled labour would benefit most from trade, and thus inequality would fall. However, these predictions are challenged by new theories(Davis, 1996) and ex- tensive empirical studies produce contradictory results. Therefore, the distributional impact of trade policy reform cannot be generalised to all developing countries. The period of rapid economic liberalisation in India also witnessed erratic trends in con- sumption inequality. A number of studies reveal mixed evidence on consumption and income inequality during this period(Pal and Ghosh, 2007). The studies on the dis- tributional effect of trade reforms in India is remarkably scant and the primary focus of these studies is mainly on labour market adjustment. Kumar and Mishra (2008) evaluate the effects of tariff reform on industry wages. One of their main conclusions is that liberalisation has led to decreased wage inequality between skilled and unskilled workers in India. However, there are other channels through which trade reforms can potentially affect income distribution. Topalova (2007) found that tariff reforms led to a higher poverty rate and a higher poverty gap in rural areas; she also found that inequal- ity is unaffected by tariff reform. The main drawback of the empirical methodology in Topalova (2007) is that it does not clearly track all possible general equilibrium effects of tariff reforms which eventually affect income distribution. My exercise in this thesis estimates all possible links between tariff reforms and income distribution. India has had a long history of social division due to the traditional institution of caste that created a social stratification along education, occupation and income lines. These factors motivate our focus on contemporary India in addressing the questions related to distributional impact of macroeconomic structural changes. I, along with my co-authors, study the evolution of the economic well-being of individuals belonging to historically disadvantaged castes between 1983 and 2005. There is a large literature that has investigated the existence and extent of labour market discrimination in India. Among others, Banerjee and Knight (1985) and Madheswaran and Attewell (2007) have studied the extent of wage discrimination faced by Scheduled Castes and Scheduled Tribes (SC/ST) in the urban Indian labour market. Borooah (2005) has studied the extent of discrimination in employment in the urban labour market. Ito (2009) studies both wage and employment discrimination simultaneously by examining data from two Indian states, namely, Bihar and Uttar Pradesh. Our study differs from these in that we examine the data for all states and for both rural and urban areas. Moreover, as 2 opposed to most of these studies, our study controls for the presence of occupation and industry effects on wage outcomes. We also provide a time series perspective on the evolution of SC/ST fortunes in India, a feature that other studies have typically not examined. To the best of our knowledge, this is the first study to jointly analyse caste differences in education, occupation, industry and wage outcomes in a single study, track the time series evolution of these outcomes, and do so using data that covers the entire country. Despite rapid economic growth in India, we observe disappointing performance in health status. Waves of National Family Health Surveys in India reveal that malnu- trition continues to be a significant problem for all age groups in India (IIPS (2007)). Moreover, Indians suffer from the dual burden of poor nutrition - more than 33% of adults are underweight while more than 10% are obese. Only 57% of males and 52% of females were considered to be healthy in terms of the weight for height index in 2005- 06. Income growth has accompanied many changes in society, such as ideas of social status and perception of diet patterns associated with social status. Due to these social transformations, changes in tastes and living standards significantly influence the com- position of food demand and diet quality. Powerful marketing strategies, developments in communications and media, rapid urbanisation and an irresistible demonstration ef- fect give rise to changes in both rural and urban areas. A closer look at the composition of food groups reveals that a drop in expenditure share on cheap sources of carbohy- drates is not properly compensated by more balanced diets, particularly from sources of proteins. This is more prominent among the urban rich. The compelling question is: why do we not see a significant development in nutrition status despite higher economic growth and poverty alleviation? Does it mean that changes in tastes and lifestyles im- pede improvement in nutrition status which should be engendered from more economic fortunes? I look into this question in the last chapter of this thesis. The limited source of good quality observational data in developing countries is one of many serious problems faced by researchers. However, India is an exception, as it has a pioneering household socio-economic survey system from early years of planning. Since 1999, the National Sample Survey Organisation of India (NSSO) has made unit-level survey data available to researchers. I use the last five quinquennial rounds of NSSO data on household consumption and employment & unemployment in this thesis. 3 The thesis is organised as follows. The next chapter examines the distributional im- pact of tariff reforms using household survey data. Chapter 3 analyses caste differences in education, occupation, industry and wage outcomes. Chapter 4 attempts to estimate structural changes in food preferences and the nutritional consequences. 4 2. Distributional Effects of Tariff Reforms in India 2.1 Introduction As a democratic state with socialist goals, Indian policy-makers have always been concerned with income distribution from the early years of planning. The government has favoured more state intervention with the sole intention of more equitable distribu- tion and rapid poverty reduction. However, the ”Hindu growth rate” in the first three decades of planning were not adequate to pull the masses out of the poverty trap. When- ever some (un)favourable conditions led to policy change, the debate centred around the distributional impact. It was during the mid-eighties that the government gradually adopted market-oriented economic reform and begun to loosen its grip on import licens- ing. However, the process was slow and ad hoc. In 1991, India initiated comprehensive measures of global economic integration compelled by international monitoring bodies to bridge the huge gap in the internal as well as the external balance. The debate on economic policy reform, which started in the eighties, continues to draw our attention today. The left wing of policy analysts perceive the reform as a shift of focus away from state intervention for more equitable distribution towards capitalistic exploitation under the free market economy. However, the literature is unable to deliver an unequivocal verdict. Tariff reduction and elimination of some non-tariff barriers were two important com- ponents of the liberalising process in the 1990s. The general presumption about trade liberalisation is that it would lead to higher GDP growth rate, productivity and effi- ciency. Apart from these macroeconomic impacts, the estimation of the distributional effects is also important to understand whether higher macroeconomic performances are realised at greater social costs. If gains from reform do not reach all sections of the population, we need some complementary strategy to combat the widening gap be- tween the rich and the poor. The distributional impact of trade policy reforms in India 5 is an under-studied problem. In the literature where distributional impact has been considered, the main focus remains on labour market outcomes (Kumar and Mishra, 2008). The general equilibrium effects of trade reform on income( or expenditure) dis- tribution is difficult to predict in a complex real world. Trade theory suggests that in developing countries, abundant unskilled labour would benefit most from trade, and thus inequality would fall. However, these predictions are challenged by new theories (Davis, 1996) and extensive empirical studies produce contradictory results. Therefore, the distributional impact of trade policy reform cannot be generalised to all developing countries. The period of rapid economic liberalisation in India also witnessed erratic trends in consumption inequality. A number of studies reveal mixed evidence on con- sumption and income inequality during this period (Pal and Ghosh, 2007). My own estimates of district-level Gini, based on National Sample Survey consumption data, show a downward trend from 1988 to 2000 that sharply increased in 2004-05. Thus, it is important to ask whether the increasing consumption inequality could be attributed to trade openness. The causal impact of trade reform on poverty and inequality is largely understudied in developing countries for two reasons: 1) the lack of data and 2) the difficulty of finding a suitable empirical strategy to identify the causal relation. Recently, Topalova (2007) solved these two problems by constructing a unique district- level panel of trade exposure for India. It is found that tariff reforms led to higher poverty rates and a higher poverty gap in rural districts. She also found that district inequality is unaffected by tariff reform. The main drawback of this approach is that it does not clearly track all possible general equilibrium effects of tariff reforms which eventually affect income distribution. One of the possible links between income dis- tribution and tariff changes is expressed as a change in relative sectoral factor returns due to a change in relative output prices. As pointed out by Topalova (2007), it is mainly labour market adjustment that is identified as the primary mechanism in India. However, there are other general equilibrium effects, viz., the effect of relative output prices on consumption, which can potentially affect poverty and inequality measures. Therefore, it is important to consider all possible links, and not only the labour market adjustment, to estimate the overall distributional impact of tariff policy changes. In this chapter I estimate the general equilibrium distributional impact of tariff re- forms in India using the empirical method of Porto (2006). This method enables me to find each channel separately through which poverty and inequality could be affected by tariff changes. The reform causes domestic prices of traded goods to change and it 6 triggers two general equilibrium effects: 1) changes in the prices of non-traded goods and 2) changes in factor incomes. By combining the estimates of consumption effects due to changes in the prices of traded and non-traded goods and labour income effects due to labour market adjustment, I am able to find the overall distributional impact. Using National Sample Survey data I find that all income groups have a significantly positive welfare gain, but the overall distribution is affected differently in the rural and urban sectors. Unlike Topalova (2007), I found that gains for the poor are relatively higher in rural areas and that for the rich is higher in urban areas. Since I estimate each effect separately, this method also helps me to explain why inequality is affected by tariff reforms and the relative contribution of labour market adjustment and adjustment in consumption spending. The consumption effects and labour income effects in the rural sector have offsetting effects along income distribution. But the labour market effect is pro-poor and relatively stronger in the rural sector. Therefore, the aggregate effect of tariff reforms is pro-poor in the rural sector. On the other hand, the labour market effect is similar across all groups and the consumption effect is relatively higher for richer households in the urban sector. Thus, the total gain as a percentage of household expenditure is higher for richer households. The remainder the chapter is organised as follows. In Section 2.2, I briefly describe the trade liberalisation process and some important aspects of the tariff structure. Section 2.3 describes the empirical strategy developed by Porto (2006), while section 4.2 presents the data used for this exercise. In Section 2.5, the results are derived. Section 4.6 concludes. 2.2 Trade reform in India During the first three decades (1950-80) of planning, India grew at a slow but steady rate of three and a half percent. After that, the growth rate almost doubled. The blame for the relatively slow growth during the first half is attributed to microeconomic dis- tortions and state intervention that severely restricted private entrepreneurship (Pana- gariya, 2004). Investment licensing restricted competition in the domestic market and import licensing eliminated foreign competition. This was during the mid-eighties, when India started some ad hoc external reform measures followed by comprehensive economic liberalisation in 1991. The reform in India in 1991 was unique in the sense that it was drastic and came as a surprise to policy-makers. The exogenous nature of 7 the trade liberalisation measures in 1991 helps analysts to establish a causal relationship between reform measures and economic outcomes (Topalova, 2007). The earlier phases of external sector reforms were driven by the economic situation. The literature traces three distinct phases of trade policy in India (Panagariya, 2004). The first period is identified as a trend towards protectionism culminating in virtual autarky (1950-1975). The proportion of licences going to traders (not actual users of imported goods) had diminished from 61% of all licenses issued in 1951-52 to less than 3% in 1970-71 (Bhag- wati and Srinivasan, 1975). The structure of imports had shifted towards capital goods, intermediate goods and raw materials through actual user licences (food grains were the only significant major consumer goods imported). According to Panagariya (2004), the trade regime was so restrictive by the mid-1970s that the share of import of non-oil and non-food grains in GDP fell to 3% in 1975-76. In the late seventies, the healthy accumulation of foreign exchange reserves due to remittances from the Middle East and increased export performances raised the comfort level of policymakers. Industrialists began to lobby for less restrictive import licensing for capital goods. Against the back- drop of this development, the second phase of the trade regime started in 1976 with the re-introduction of Open General Licensing (OGL). The articles scheduled in OGL no longer required a licence from the Ministry of Commerce. More and more articles were included in this list and by April 1990 the value of OGL imports was approximately 30 percent of total imports. Improved agricultural productivity and the discovery of oil fields made it possible to expand non-oil and non-food imports. However, tariff rates were raised substantially during this period. Moreover, the tariff codes were so complex and obscure that even trade specialists had problems interpreting the information. By the end of 1990, the average tariff rate was 83.7% and the maximum tariff rate was 521% (Table 2.1). The liberalisation process in the 1980s was complemented by an ex- pansionary fiscal policy which was supported by internal and external borrowing. The external borrowing was unsustainable and the Gulf War in 1990 led to a swelled import bill for oil. The balance of payment deficit was so severe that the total foreign exchange reserve could merely finance two weeks’ imports. Political uncertainty, mainly due to the short span of two consecutive coalition governments, undermined the confidence level of investors. The newly elected Congress government in 1991 used the crisis as an opportunity to led the country towards a new phase of trade regime (third phase). The budget in July 1991 was a clear shift towards an outward-oriented, market-based economy. Import licensing on all but consumer goods, some intermediate inputs and capital goods were removed. After rounds of deliberation at the WTO, consumer goods 8 were freed of licensing in 2001. Table 2.1: Tariff structure in India Year Mean SD Max 1990 83.7 51.99 520.93 1992 58.08 22.99 355.00 1997 30.63 14.63 260.00 1999 33.67 12.55 230.00 2001 34.87 26.54 586.91 2004 30.38 15.04 232.39 2005 19.45 16.85 232.39 All statistics are calculated using import weighted average MFN tariff rates of 6 digit HS item group. (source: Trade Analysis and Information System, UNCTAD) After gradual de-licensing in the external sector during the 1980s, high import tariffs were an effective source of trade protection before 1991. Therefore, tariff rates were drastically reduced as part of a comprehensive liberalisation process in 1991. Table 1 shows the significant change in MFN (Most Favoured Nations) tariff rates during the early phase. The mean tariff rate fell from 83.7% in 1990 to 58.1% in 1992 and gradually reduced to 19.5% in 2005. The standard deviation of the import weighted average tariff rates of 6-digit (HS code) item groups is only 16.9% in 2005 compared to a very high degree of dispersion in 1990. 0 20 40 60 80 Im po rt w ei gh te d av er ag e ta rif f 1990 1995 2000 2005 Trade Year Food, beverages, tobocco Fuel Textiles Other manufactured goods Average tariff (weighted) rates of different item groups (a) weighted 20 40 60 80 10 0 Si m pl e av er ag e ta rif f 1990 1995 2000 2005 Trade Year Food, beverages, tobocco Fuel Textiles Other manufactured goods Average tariff (simple) rates of different item groups (b) simple Figure 2.1: Simple and weighted average tariff of different consumption item groups 9 Figure 2.1(a) shows the weighted average tariff rates of different consumption item groups. From 1990 to 1992, both the simple and weighted averages of food, textiles and other manufactured goods fell sharply and then followed a steady declining trend except for food, beverages and tobacco products. The simple average tariff rate for fuel and fuel products had a declining trend, but import weighted tariffs did not show a uniform trend. It is clear from the plots that the first part of the 1990s witnessed the sharpest drop in tariff rates. Though there are some reversal in later periods, the general direction of tariff reform is towards liberalisation. As a result, trade shares in GDP have increased at a much higher rate after 1991 (fig. 2.2). The drastic change in tariff rates in 1991 and subsequent gradual changes have several general equilibrium effects on prices and wages. The next section outlines the empirical strategy to identify welfare effects of tariff changes incorporating all these general equilibrium effects. 0 10 20 30 40 Sh ar e in  G D P (% ) 1960 1970 1980 1990 2000 2010 Year export/gdp import/gdp merchandise trade/gdp Share of export, import and merchandise trade in GDP Figure 2.2: Share of export, import and merchandise trade in GDP 2.3 Empirical strategy Trade liberalisation is a broad concept encompassing variety of phenomena that reflect increased interdependence between countries, flow of goods and services across border, movement of capital and labour, etc. Some aspects of trade reforms are easier to measure and therefore, received more attention in empirical research. Detailed information of trade barrier is often not easily available for developing countries. The episodes of reductions in tariff barriers are commonly studied mechanism of trade liberalisation 10 for various reasons. First, tariffs are relatively easier to measure and readily available. Second, it is mostly ad valorem and therefore, it reflects price based trade protection. My study focuses on tariff reform. Co-evolution of income or consumption distribution and trade policy is a common phenomenon. However, establishing a causal link between reform and inequality by pro- viding credible evidences poses several challenges apart from the issue of endogeneity of reform measures. The main challenge is to isolate the effects of tariff reform from other contemporary changes that might have influenced change in the consumption (income) distribution. Governments in developing countries often undertake several reforms in external sector as well as in internal sector simultaneously. Any study that attempts to identify the overall impact of tariff reform thus demands a strong empirical method- ology with proper identification assumptions. Goldberg and Pavcnik (2007) discuss two broad approaches to the problem: 1) general equilibrium approach (Porto (2006)) and 2) differential exposure approach (Topalova (2007)). The main advantage of the general equilibrium approach is that it explicitly accounts for all possible major chan- nels through which reform affects distribution of income or consumption expenditure in a country. However, the predictions of this approach depend on reliability of esti- mates of certain crucial parameters that are typically unknown: 1) elasticity of wages with respect to prices, 2) elasticity of prices of non-traded goods with respect to prices of traded goods and 3) degree of pass-through from tariff changes to product prices. These parameters are difficult to estimate when many other policies change contempo- raneously with tariff reform. The second approach exploits the cross sectional variation in trade protection to examine whether a region or industry that were more exposed to trade protection experienced higher/lower change in inequality (wage, income or con- sumption) compared to a less exposed region or industry. Given a trade policy at the industry level within a country, exposure to trade protection of a geographical region depends on industry concentration in that area. The main advantage of this approach is that it requires much weaker identification assumptions than the general equilibrium approach. However, such approaches can only identify the extent of region/industry specific deviation from the aggregate trend that could be attributed to the reform mea- sures. It can not identify the role of reform in explaining the trend itself. It also can not shed light on composition of the welfare change due to reforms - whether the driving force of such changes is labour market adjustment or strong inter-linkages between the traded and the non-traded industries. I take the general equilibrium approach for two 11 reasons: 1) explicitly account for all linkages and 2) comparing results derived from my exercise with that of differential exposure approach. I follow the Porto (2006) to identify the distributional impact of tariff reforms at the household level. I briefly discuss the main elements of the Porto (2006) model which is adapted from the small open economy models of Dixit and Norman (1980) and Woodland (1982). Let us assume that total family income is equal to total family expenditure on different consumption goods and services (no saving). I also assume that total family income consists of factor incomes and some exogenous income. Therefore, ej(PT,PNT, u j) = xj0 + ∑ m wjm + k j (2.1) where ej(.) is the expenditure function of household j. The expenditure function de- pends on price vectors of traded goods(PT), non-traded goods (PNT) and required household utility uj. Household’s total income consists of household capital income (kj), sum of individual labour incomes (wjm) and some exogenous income (x j 0). In a small open economy the domestic price of traded goods (pi) depends on exogenous international price (p∗i ) and imposed tariff rate (τi). I can express the domestic price of traded goods as pi = p ∗ i (1 + τi). (2.2) It is assumed that domestic firms in the traded good sector produce under constant returns to scale and competitive market. Thus, the prices of these goods are equal to unit production costs pi = ci(w), (2.3) where ci(.) is the unit production cost and w is the vector of factor prices. The system of equations in 2.3 determines the general equilibrium relationship between factor prices and commodity prices. When prices of traded goods change, the factor reallocation takes place for the given economy-wide factor endowment. As a result, factor prices adjust. In a two-good and two-factor model the relationship is described by the Stolper-Samuelson theorem. However, for a multi-dimensional set-up, there is a correlation among prices of goods and prices of factors (Dixit and Norman, 1980). The total expenditure of a household consists of spending on traded goods and spend- ing on non-traded goods. The prices of non-traded goods is derived from demand and 12 supply interaction in the domestic market. Using Roy’s identity the demand for non- traded good is derived as a derivative of expenditure function ej(.) with respect to the price of that good. Similarly, using Hotelling’s lemma I get the supply of non-traded good as a derivative of GDP with respect to the price of that good. That is, ∑ j ∂ ∂pk ej(pT,pNT, u j) = ∂ ∂pk r(pT,pNT,v, φ) (2.4) where r(.) is GDP function of the economy, v is factor endowments and φ is some measure of technology. The equilibrium prices of non-traded good is determined by pk = pk(pT,v, φ,u) (2.5) where u is a vector of utilities for all households. When there are changes in tariff rates, they affect the domestic prices of traded goods (equation 2.2). The changes in domestic prices of traded goods induce two adjustments: changes in factor returns (equation 2.3) and changes in prices of non-traded goods (equa- tion 2.5). These adjustments capture the general equilibrium effects of tariff reform. I assume that capital income for the majority of households is negligible. Thus, I consider labour income as the only source of factor income. The welfare effects caused by the change in prices of traded goods and non-traded goods are called consumption effects and the welfare impact caused by changes in labour income is called labour income effects (Porto, 2006). The change in household welfare due to change in tariff rates is computed using com- pensating variation measures (CV). CV is the amount of money needed to compensate a household to achieve the same level of utility before the price change. From the house- hold budget (equation 2.1), I derive the change in exogenous income xj0 so that the family gets the same pre-reform utility level. Taking the total differential of Equation 2.1 for an exogenous change in domestic price of traded good i (pi) and assuming zero capital income, dxj0 = ∂ej(.) ∂pi dpi + ∑ k∈NT ∂ej(.) ∂pk ∂pk ∂pi dpi − ∑ m ∂wjm ∂pi dpi. (2.6) 13 The change in domestic price of traded good is induced by an exogenous change in tariff rate. Therefore, dpi = ∂pi ∂τi dτi. (2.7) Dividing both sides of equation 2.6 by total expenditure ej, CV is expressed as a share of total household expenditure, dxj0 ej = ( sji + ∑ k∈NT sjk ∂ln(pk) ∂ln(pi) − ∑ m θjmε j wmpi ) ∂ln(pi) ∂ln(τi) dln(τi) (2.8) where sji is budget share spent on traded good i by household j, s j k is budget share spent on non-traded good k, θjm is the wage income share of member m in total family expenditure and εjwmpi is wage-price elasticity of household member m. The wage-price elasticity, εjwmpi captures the proportional change in wage earned by member m due to a change in the price of traded good, pi.Since I am measuring CV for a reduction in tariff rate, the welfare effect in Equation 2.8 is negative of compensating variation. As a result, a positive estimate of CV in Equation 2.8 means a welfare gain. The equation 2.8 shows that the total welfare effect of each household has three components: a direct effect through the consumption of traded goods and general equi- librium effects through non-traded sectors and the labour market. To estimate the overall impact of trade policy reform, I estimate each of these three components. It is also noted that Equation 2.8 captures only first order effects. The higher order effects deal with estimating own price and cross-price elasticities. Following Porto (2006), I ignore higher order effects because the price elasticities will be irrelevant for distribu- tional effects if price elasticities are assumed to be same for all income levels. Since all higher order effects are ignored, the absolute measures of welfare effects are distorted. However, the relative distributional effect will not be affected by ignoring higher order effects. Similarly, the assumption of a competitive market is crucial for estimating the absolute measure of welfare gain. I assume that if these distortions affect the general equilibrium prices and wages, it will affect all income groups equally. Using consump- tion survey data, I estimate the consumption effects of traded and non-traded goods. Employment survey data is used to estimate household-level labour income effects. The assumptions considered above need some caveat. I have assumed that there is perfect pass-through from tariff rate changes to domestic price changes and there is 14 no change in world price due to domestic policy change. In a small open economy this is quite reasonable to assume fixed international price. However, there are some empirical evidences from developing countries that demonstrate significant change in price behaviour of international exporters due to tariff and exchange rate reforms. If pass-through rates are imperfect and industry specific, the results derived in this paper need more qualification. I do not consider imperfect pass-through analysis due to data limitation. I also assume that household saving is negligible. This assumption need not be valid if there is strong presence of financial market. In presence of significant linkages between tariff rates and capital market adjustment, the results are biased. Data limitation prohibits us to estimate elasticity of rate of return from capital with respect to product prices. As I have mentioned before, there are several contemporaneous changes in Indian economy. The combined effect of such changes could lead to more competitive labour and capital markets. How could we minimise the bias introduced by such contemporaneous changes? One important aspect of these policy changes is that they occurred during same period, i.e. after 1991. While estimating the parameters (to be used in general equilibrium analysis) we explicitly control for time dummy to minimise such biases. 2.4 Data The purpose of the chapter is to estimate household-level welfare effects and com- pare them across different income groups. To estimate the welfare effects of tariff reform, I need tariff data over the reform period as well as information on disaggre- gated household-level consumption and individual wages. The National Sample Survey Organisation (NSSO), set up by the Government of India, conducts large quinquennial rounds of survey to collect socio-economic data. The consumption schedule collects in- formation on household consumptions on different goods and services. The employment and unemployment schedule collects individual information on education and wages. I use the 61st consumption schedule as the base survey to estimate household budget shares of traded and non-traded goods. The cross-section of households gives detailed information on the value of consumption of disaggregated items. I classify those items in the traded and non-traded categories based on import data (HS code and its de- scription). The employment and unemployment survey rounds are used to estimate wage-price elasticity. I use the five latest quinquennial rounds (1983, 1987-88, 1993-94, 1999-2000 and 2004-05) to get pooled cross-section data of individual wage, age, sex, 15 education and other characteristics. These surveys cover the whole geographical area of India except for some areas of Jammu & Kashmir and Andaman & Nicobar Islands. The NSSO uses a complex stratified sampling design to select the ultimate stage unit (households) in both the urban and rural sectors. The tariff data is extracted from the UNCTAD - TRAINS (Trade Analysis and In- formation System) database. Detailed information by sub-headings on tariffs and value of imports is available on the World Integrated Trade Solution (WITS) website. The data portal obtained Indian tariff rates from the Customs Tariff Schedule 2004-2005, Directorate of Publicity and Public Relations, and Customs and Central Excise. I use MFN (Most Favoured Nation) tariff rates to find the import weighted average tariff rates from 1990 to 2005 of different item groups. It also reports the calculation of ad valorem equivalents of non-ad valorem tariffs. The domestic prices are collected from various publications of the Central Statistical Organisation (CSO). The CSO publishes wholesale price indices (WPI) of disaggregated items for the all-India level. The prices of non-traded goods (mostly services) are not covered in the WPI data published by the CSO. However, I use GDP deflator to estimate the price indices of non-traded goods. The nominal and real GDP are collected from various publications of the CSO. I use sector-wise GDP deflators to find the price indices of different non-traded categories. Therefore, the creation of non-traded item groups is based on the availability of GDP de- flator data for that category. I discuss each of these variables in detail in the estimation section (section 2.5). 2.5 Estimation The household-level welfare effects of tariff policy are summarised by Equation 2.8. In this section I first estimate the change in domestic prices of traded goods due to tariff reform. Then, I estimate the general equilibrium effects: consumption effects in traded goods, non-traded goods and labour income effects. The impact of tariff reduction on domestic prices is estimated as the change in import weighted tariff rates. The TRAINS dataset provides a 6-digit level classification of MFN tariff rates with import values. The goods are classified in four traded groups: 1) Food (primary and manufactured), beverages and tobacco products (FFBT), 2) Fuel, power, light & lubricants(FPLL), 3) Textile(TXTL) and 4) Other manufactured goods(OMFG). Although the classification is based on the availability of time series of wholesale prices, this is quite a reasonable 16 classification given household budget shares and import values. 2.5.1 Price change of traded goods I estimate the average tariff rates of the four groups using import values as weights. τi,t = ∑ s∈i µs,tτs,t (2.9) where τi,t is the average tariff of group i(FFBT, FPLL, TXTL and OMFG) in year t, µs is the import share(value) of good s in group i and τs,t is the tariff rate of good s in year t. The average tariff is estimated using this formula for all four traded goods. I use the import weighted average instead of the simple average because India maintained non-tariff barriers for a host of goods in the early period of liberalisation. The Indian authorities first lifted the regulations on capital goods, basic goods and intermediate goods. In the late nineties consumer non-durables and agricultural products were slowly liberalised. Since liberalisation started in 1991 and still continues, my period of analysis is from 1990 to 2005. The earliest available tariff data on TRAINS is 1990 and the latest round of the NSSO survey was conducted in 2004-05. Therefore, I consider the total change in tariffs from 1990 to 2005. I assume a unitary pass-through rate from tariff to price so that any change in tariff will have a proportional impact on prices. The exogenous change in the price of traded good i due to tariff reduction is given by the following formula: dln(pi) = ln(1 + τi,2005)− ln(1 + τi,1990). (2.10) Table 2.2 shows the structure of tariffs in 1990 and 2005. I report both the simple average as well as the weighted average of tariff rates. There is an excessive degree of protection just before the reforms in 1990. The average tariff rates (simple) for all traded goods declined drastically within this period. However, the weighted tariff rates actually increased for fuel, power, light & lubricants. Textiles and other manufactured goods show the maximum change in tariff rates. Food items, beverages and tobacco products are still protected (41% average tariff) compared to other categories. Using the formula 2.10, I compute the average change in prices of four traded goods. The price of FFBT decreased by 6.4% whereas the prices of TXTL and OMFG have gone down by 174% and 143%, respectively. However, the price of FPLL has increased by 90% within this period. All these changes are measured in terms of weighted average 17 Table 2.2: Tariff structure in 1990 and 2005 Traded good 1990 2005 Weighted Simple Weighted Simple Food, beverages & tobacco 61.17 (77.14) 94.71 57.31 (28.61) 40.63 Fuel, power, light & lubricants 3.86 (20.05) 45.89 10.9 (8.22) 14.82 Textile 85.26 (33.8) 96.37 19.7 (24.72) 25.47 Other manufactured goods 76.66 (51.78) 81.97 12.66 (6.93) 15.18 The numbers in parenthesis report standard deviation of tariff rates within each groups. to take care of the importance of each of the goods in total import value. Since the import values of different agricultural products are less due to quantitative restrictions, the actual impact of tariff reduction in such commodities will be realised less in terms of changes in domestic prices. During the reform period, not only did the average tariff rates decrease, but the dispersion was also reduced to a large extent (standard deviations are reported in parentheses). This is the estimate of first order or direct impact of tariff reduction on domestic prices of traded goods. As I discussed earlier, the higher order effects will be realised through own price and cross-price elasticities. The limitation of data will hinder us in estimating such higher order effects. Since I am concerned with distributional impact, the higher order effects are not important. The changes in the prices of traded goods affect the prices of non-traded goods and wages through labour market adjustment. The general equilibrium relationships are captured in Equations 2.5 and 2.3. In section 2.5.2 and 2.5.3, I estimate the consumption effects of traded and non-traded goods. The labour income effects are estimated in Section 2.5.4. 2.5.2 Consumption effects of traded goods The consumption effects of traded goods depend on expenditure shares and changes in prices. The term sji ∂ln(pi) ∂ln(τi) dln(τi) captures the consumption effect of household j due to a change in tariff of good i. I need household-level data on consumption items to estimate the budget shares sji . The NSSO 61 st round provides detailed household-level expenditure data for the period 2004 to 2005. I categorise the goods and services in four different traded groups. Multiplying the budget share by change in price due to tariff change, I estimate household-level welfare change. Summing over all traded goods, I get the total consumption effect of all traded goods. I have 124,644 households in the cross- 18 section. To get the distributional effect, I need to summarise the household welfare effects in different income groups. A non-parametric method is used to summarise the welfare effects as a function of household expenditure. The Fan (1992) method of locally weighted regression is used to estimate welfare along different points of monthly per capita expenditure. Practically, I run a linear regression of the estimated household welfare effects (y) on the log of monthly per capita household expenditures at 50 grid points (x) instead of all values of the independent variable (Deaton, 1997). For each grid point, first I calculate a series of weights for each data point (MPCE) within a given bandwidth (h). This is done using a suitable kernel function θj(x) = ωjK ( x− xj h ) (2.11) where θj(x) is the weight for household j at grid point x which depends on household survey weights ωj and a kernel functionK(.). The estimated parameters of the weighted regression at each grid point x is given by β̂(x) = [X ′Θ(x)X]−1X ′Θ(x)Y (2.12) where Θ(x) is a diagonal matrix of weights of each households(θj(x)), X is matrix of two columns (ones in first column and MPCE in second column) and Y is the vector of welfare effects of households. . For each grid point, I estimate two parameters (intercept and slope). The predicted value of welfare effect at each grid point is thus given by m̂(x) = β̂1(x) + β̂2(x)x. (2.13) The summarised consumption effects of traded goods are plotted in Figure 2.3. I measure the average welfare gain as a percentage of MPCE along the vertical axis and the log of per capita expenditure along the horizontal axis. The dotted lines give a 95% confidence band. The standard errors are calculated using the bootstrap method. I replicate the estimation 200 times, taking random samples with replacement. The randomness comes only from household consumption expenditures. I get an upward sloping distributional effect for traded goods with some exception at the lowest range. The horizontal axis measures the log of real monthly per capita expenditure (MPCE) and the vertical axis measures compensating variation as a pro- portion of MPCE. The vertical lines give all-India poverty lines for the rural (PLr) and 19 PLr PLu.2 . 3 . 4 . 5 . 6 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure smth_ce_traded upper_limit lower_limit The solid line shows average consumption effect of traded goods using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Consumption effects of traded goods Figure 2.3: Consumption effects of traded goods urban (PLu) sectors. All expenditure groups have welfare gain due to a reduction in tariff rates. The extreme poor section has a downward sloping welfare gain, i.e., as the household gets richer, the welfare gain is lower. But this downward trend is associated with higher standard errors. The richest section has the highest welfare gain. It should be noted that the welfare gain is expressed as a percentage of the monthly per capita household expenditure. The consumption effect ranges from 0.23% to 0.47%. The rural-urban breakup of consumption effect is given in Figures 2.4(a) and 2.4(b). All geographical regions have a similar pattern of welfare gains coming from adjustment in the consumption of traded goods. In general, percentage welfare gains are higher for richer households except for extremely poor households. As noticed earlier, this segment of the downward sloping relation is associated with higher standard errors due to fewer observations. Households below the poverty line have welfare gains between 0.2% and 0.25%. The important aspect is that these gains are statistically significant as depicted by the confidence band around the line. The shape of the curve is determined by the budget shares of income groups. As it is known from Engel’s law, the proportion of expenditure on food falls as the household gets richer. The poorest section has the highest budget shares on FFBT, but the price has reduced by only 6.4%, whereas textiles and other manufactured goods have larger budget shares for higher income groups. The prices of these two groups of articles have been affected mostly due to tariff reforms. 20 PL.2 . 4 . 6 . 8 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure smth_ce_traded upper_limit lower_limit The solid line shows average consumption effect of traded goods using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Consumption effects of traded goods − Rural (a) Rural PL.2 . 3 . 4 . 5 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure smth_ce_traded upper_limit lower_limit The solid line shows average consumption effect of traded goods using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Consumption effects of traded goods − Urban (b) Urban Figure 2.4: Consumption effects of traded goods -rural, urban 2.5.3 Consumption effects of non-traded goods The change in prices of traded goods have a general equilibrium effect on the prices of non-traded goods. In this section, I measure the compensating variation due to change in prices of non-traded goods. The term ∑ k∈NT sjk ∂ln(pk) ∂ln(pi) ∂ln(pi) ∂ln(τi) dln(τi) (2.14) in Equation 2.8 captures the consumption effect(non-traded) of household j for change in tariff of good i. Summing over all traded goods, I estimate the total change in welfare of non-traded goods. I use the NSSO consumption survey to estimate the expenditure shares of non-traded goods (sjk). The items are classified in three non-traded goods based on the availability of price data. I use sector-specific GDP deflators to estimate the price series of non-traded goods. The GDP deflators are calculated from nominal and real GDP figures in three main sectors: 1) Real estate, ownership of dwelling & business services (ROB), 2) Transport, storage & communication (TSC) and 3) Health, education and other services (HEO). These sectors are mainly non-traded sectors except for the later part of the reform period when the government allowed some foreign capital. Despite that, the classification is a close approximation of non-traded sectors. Using the matching description of items in the consumption survey, I classify three broad non-traded goods (ROB, TSC and HEO). 21 In Equation 2.5, I show that the endogenous prices of non-traded goods have a general equilibrium relationship with the exogenous prices of traded goods. Therefore, pk = pk(pFFBT , pFPLL, pTXTL, pOMFG,X) (2.15) where pFFBT , pFPLL, pTXTL and pOMFG are the prices of traded goods and the vector X is all other exogenous variables. I assume a simple distributed-lag model to estimate the relationship. In what follows, I estimate ln(pkt) = a0 + ∑ i∈T a0iln(pit) + ∑ i∈T a1iln(pit−1) + ∑ i∈T a2iln(pit−2) + ∑ i∈T a3iln(pit−3) + ∑ i∈T a4iln(pit−4) +Xtγ + µt. (2.16) where µt is the error term and X is the vector of other control variables. I have regressed the annual prices of each non-traded good on current as well as lagged prices of all traded goods. Time trend is included to capture technological changes over the years. I also include a liberalisation dummy to capture any differences between the pre- liberalisation and post-liberalisation periods. The annual price series of traded goods are taken from the wholesale price index published by the CSO. The non-traded goods’ prices are calculated using GDP deflator. I use the period 1953 to 2008 to estimate the relationship. The raw series show the characteristics of integrated of order one; thus I estimate Equation 2.16 in the first difference. Theoretically, the relationship should follow homogeneity of degree one in prices. Therefore, I impose the following restriction 4∑ l=0 ∑ i∈T ali = 1. (2.17) Table 2.3 shows the response of prices of non-traded goods for exogenous changes in prices of traded goods. Since all prices are expressed in logarithm, the coefficients are interpreted as elasticities. Once I estimate Equation 2.16, the elasticities are calculated as the sum of coefficients of all lagged prices and the current price for each traded good. Therefore, the significance of the elasticities are tested using F-test statistics of the sum of the coefficients of the lagged prices and the current price. Since the estimated equation captures a complex, general equilibrium relationship, there are no theoretical predictions (Porto, 2006). In a multidimensional model, there are no clear predictions about the signs of these elasticities (Dixit and Norman, 1980). 22 Table 2.3: Response of prices of non-traded goods Non-traded goods Traded goods TSC ROB HEO Food, beverages & tobacco 0.40 0.36 0.14 (1.58) (1.85) (0.97) Fuel, power, light & lubricants -0.71 -0.23 0.09 (-2.27) (-0.94) (0.47) Textile -0.49 -0.49 0.45 (-1.35) (-1.76) (2.11) Other manufactured goods 1.80 1.36 0.32 (2.64) (2.59) (0.79) Price change due to tariff reform -2.38 -1.31 -1.17 (t-test for price change) -2.98 -2.13 -2.47 R2 0.73 0.26 0.79 The F test statistics of the significance of all lagged prices are in parenthesis. The price of Transport, storage & communications (TSC) is positively associated with other manufactured goods (OMFG) and negatively associated with Fuel, power, lights & lubricants (FPLL). The prices of other traded goods are not correlated with TSC. The relationship between fuel price and transport price is surprising. Usually, they should be positively correlated: as fuel price goes up, transport price should go up as well. The intuitive interpretation is that the transport and communications sector is highly subsidised by government agencies. Therefore, the prices households face do not really reflect the actual cost of operations. Textile price is also negatively associated with TSC though it is insignificant. The second column gives the elasticities of ROB (real estate, ownership dwelling & business services). Textile price is negatively correlated, but other manufactured goods is positively correlated. The price of FFBT (food, beverages and tobacco products) has a positive effect on the price of ROB. The effect of fuel price is not significant for this sector. The price of HEO (health, education and other services) is positively associated with textiles, but all other prices do not have any significant correlation with the price of this sector. The intuitive reason is that health and education are very much state-controlled except in some urban areas. Therefore, the costs of health and education do not really reflect the prices of all other goods. I should emphasise that these elasticities are merely correlations in the general equilibrium sense, and are not causal relationships. Therefore, the signs are not always interpreted using theoretical explanations. The total changes in prices of non-traded 23 goods due to changes in tariff rates in all traded goods are reported with t-statistics. Though there are some opposing effects on prices, the total changes in prices of all non-traded goods are negative. The highest reduction is in TSC (2.38%) and the lowest is in HEO (1.17%). The elasticities in Table 2.3 give the percent changes in prices of non-traded goods for a one percent change in the prices of traded goods. Therefore, the total change in prices of non-traded goods is measured by the product of these elasticities and the change in prices of traded food due to tariff reform (Section 2.5.1). Using Expression 2.14, I estimate household-level consumption effects of non-traded goods. I use the same base survey of NSSO 2004-05 to estimate the welfare effects of non-traded goods so that all the effects are comparable. Once household-level consumption effects of non- traded goods are estimated, I use the same non-parametric procedure in Section 2.5.2 to summarise the welfare change along expenditure lines. The locally weighted regression gives the average distributional impact of non-traded goods in Figure 2.5. PLr PLu0 . 5 1 1. 5 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure smth_ce_non_traded upper_limit lower_limit The solid line shows average consumption effect of non−traded goods using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Consumption effects of non−traded goods Figure 2.5: Consumption effects of non-traded goods The solid line in Figure 2.5 gives the average effect and the dotted lines give the 95% confidence band. The standard errors are calculated using bootstrapping. The random- ness in this estimation has two sources: 1) price elasticities, and 2) estimated household budget shares. I replicate the estimation of expenditure shares 200 times using random samples with replacement, keeping the rural-urban stratification from the cross-section 24 of households. To get the randomness in price elasticities I select random samples from a normal distribution with estimated mean and standard errors of price elasticities. Theoretically, the estimated parameters β̂ in Equation 2.16 follow an asymptotic nor- mal distribution, so that β̂ −→d N(β,Ω), where β is the true vector of parameters and Ω is its asymptotic variance. In practical computation, I do not have the true vector of parameters and its variance matrix. Therefore, I use a Cholesky decompo- sition of estimated Ω̂ variance-covariance matrix. If β0 is a randomly drawn vector from N(0, I) and R is the Cholesky decomposition of Ω̂, then Rβ0+ β̂ ∼ N(β̂, Ω̂). In each replication, I computed the welfare effects of randomly selected households, using randomly selected elasticities and then summarised the average welfare effects along the log of monthly per capital household expenditure using locally weighted regression. The standard errors are calculated using these 200 estimated values. The consumption effects of non-traded goods show a monotonous relationship with per capita expenditure. As a family gets richer, the welfare effect (as a percentage of household expenditure) is higher. However, the standard errors also increase with household expenditure. The shape is somewhat similar to that of traded goods (Figure 2.3). Households below the poverty line do not have statistically significant welfare gains from a change in the prices of non-traded goods. Figure 2.6(a) and 2.6(b) show a similar trend across all geographical regions. The top quantiles in the urban sector have relatively higher welfare gains compared to the rural sector, whereas in the rural sector, the richest quantiles have falling welfare gains. It is also noted that the 95% PL0 . 2 . 4 . 6 . 8 1 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure smth_ce_non_traded upper_limit lower_limit The solid line shows average consumption effect of non−traded goods using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Consumption effects of non−traded goods − Rural (a) Rural PL0 . 5 1 1. 5 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure smth_ce_non_traded upper_limit lower_limit The solid line shows average consumption effect of non−traded goods using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Consumption effects of non−traded goods − Urban (b) Urban Figure 2.6: Consumption effects of non-traded goods -rural, urban confidence band is wider compared to that of traded goods. In this case, there are two sources of randomness, whereas for traded goods there is only one source of randomness. 25 The non-traded goods are mostly services and, therefore, the budget shares are larger for the richer class. Tariff reduction in other manufactured goods (OMFG) leads to lower prices in Transportation, storage and communication (TSC) and Real estate, ownership dwelling and business services (ROB). Since the budget shares for these two groups are higher for the richer class, the welfare gain is also higher for them. Lower textile prices also lead to a higher welfare gain for them through the price adjustment in health, education and other services (HEO). The average compensating variations of non-traded goods are positive and significantly different from zero for all income levels above the poverty line. However, I emphasise the distributional effect, and not its magnitude. When I compare the consumption effects of traded and non-traded goods with the labour income effect, the opposite trends offset each other and the total effects are subdued. Now I need to compute labour income effects to complete the estimation of the overall distributional impact. 2.5.4 Labour income effects The labour income effects are measured by the expression ∑ m θjmε j wmpi ∂ln(pi) ∂ln(τi) dln(τi) (2.18) in Equation 2.8. To estimate the labour income effects, I first estimate the wage- price elasticities (εjwmpi). The NSSO employment and unemployment surveys collect information on individual wage, education and other socio-economic characteristics. I regress wage on prices of traded goods and other characteristics of the individual to find the response of wages due to changes in price levels. Equation 2.3 gives a general equilibrium relationship between factor prices and the prices of traded goods. Therefore, wages could be expressed as a function of the prices of traded good (pT ) and other individual characteristics (Z): wjm = w j m(p j T ,Zjm) (2.19) The NSSO employment rounds 38, 43, 50, 55 and 61 give a pooled cross-section of individual wages and other individual characteristics. I exploit the time variability of prices over the years to find the response of wages for different skill levels. The 26 estimating equation is logwjm = ∑ i∈T (β1ilogpi ∗ edu1 + β2ilogpi ∗ edu2 + β3ilogpi ∗ edu3) + γ2edu2 + γ3edu3 + δ1age j m + δ2age j2 m + σgender j m + ε j m (2.20) where wjm is wage of individual m in household j, pis are prices of traded goods, edu1, edu2, edu3 are three skill dummies based on level of education. I also include gender, age and age squared in the regression. The usual error terms are represented by εjm. The coefficient β1i is the elasticity of wage with respect to prices of traded good i for individuals with skill level edu1. Similarly, β2i and β3i are the elasticities for the other two skill levels. The lowest skill edu1 is for unskilled labours (below primary), edu2 is for semi-skilled labour (below secondary) and edu3 is for skill labour (secondary and above). I impose homogeneity of degree one in prices for each skill. The restrictions are ∑ i∈T β1i = 1, ∑ i∈T β2i = 1, ∑ i∈T β3i = 1. (2.21) The estimated results are shown in Table 2.4. Model 1 does not control for individual characteristics, whereas Models 2 and 3 control for individual characteristics. The standard errors are corrected for year-specific cluster effects in Model 3. I use the estimated results in Model 3 in my analysis. The table gives the total 12 wage-price elasticities, three for each traded good. All the elasticities are significantly different from zero when I take care of age and gender effects. However, the prices of fuel, power, lights and lubricants (FPLL) and textiles (TXTL) become insignificant for all skill demands when I correct the standard errors for clustering effects. The wages for all skill types have a positive correlation with the price of Food, beverages and tobacco (FFBT). When the prices of goods in this group increase, the wages for unskilled labour increases the most. On the other hand, an increase in the prices of other manufactured goods has a negative effect on wages for unskilled labour. Interestingly, wages for all skill types are negatively correlated with prices of all other manufactured goods. I have already mentioned that this regression captures the complex, general equilibrium effects of prices on wages, and thus there is no theoretical prediction. However, there are some intuitive explanations for the negative correlation. Topalova (2007) argues that trade liberalisation has led to higher productivity growth in the manufacturing sector. However, this increased productivity does not led to higher reduction in poverty level in the urban sector where the share of employment in the manufacturing sector is 27 larger. The hypothesis put forward by Topalova (2007) is that the gain in productivity did not increase the wage share; instead it helped the capital income share to grow. In my estimation, in fact, the wages of all types of skills have a negative correlation with the prices of traded goods. When the prices of manufacturing goods fall, productivity rises in the manufacturing sector and it helps wages to grow. The only channel through which prices have a positive significant effect on wages is the FFBT sector. However, the magnitude of tariff reduction in this sector is very low. Therefore, the negative impact of tariff reform on wages of all skill types (and all sectors) is dampened. The labour income effects of households are estimated using Expression 2.18. For each member of a household I compute the share of wage in total family income1. The wage-price elasticities (εjwmpi) from the regression Equation 2.20 is multiplied by wage share (θjm) and change in price of traded good i to get the labour income effect of each household member. Summing over all household members and all traded goods I find household-level income effects. Locally weighted regression is used to find the average effects for different income groups. The estimated average income effects are derived from different surveys which are not comparable with the base household expenditure survey 2004-05. To make it comparable I find matching percentile-wise distributional effect in the base survey. The result is shown in Figure 2.7. The standard errors are calculated using bootstrapping. The sources of variation are: 1) estimated share of wage in total family income and 2) the estimated wage-price elasticities. To deal with variance in share of wage, I re-sample the households 200 times keeping the cluster structure of the original pooled cross-section. The estimated coefficients of Equation 2.20 follow an asymptotic normal distribution, i.e., β̂ −→d N(β,Ω). I use the same procedure as in consumption effects to get random samples of wage-price elasticities. Figure 2.7 plots the average labour income effects as a percentage of total expenditure for change in prices of all traded goods. I find that tariff reform has a significantly positive effect for all income groups. However, the relationship is not monotonous. The poorest group shows a rising labour income effect. Households below the poverty line, on average, show the maximum gains from labour market adjustment. After some 1I approximate the share of wage income in monthly per capita household expenditure by wjm/ ∑ m∈j w j m because I do not have comparable expenditure data in the employment survey. If total household expenditure is not less than total household labour income, the approximated θjm will be upper bound for true θjm. Therefore, the estimated labour income effects overestimate the welfare effects. 28 Table 2.4: Regression of log wage on prices interacted with education dummies Model 1 Model 2 Model 3 b/se b/se b/se edu1 ∗ ln(pffbt) 2.348*** 2.279*** 2.279*** (0.03) (0.03) (0.17) edu2 ∗ ln(pffbt) 2.035*** 2.047*** 2.047*** (0.06) (0.05) (0.39) edu3 ∗ ln(pffbt) 1.741*** 1.698*** 1.698*** (0.06) (0.06) (0.36) edu1 ∗ ln(pfpll) 0.231*** 0.085*** 0.085 (0.02) (0.02) (0.21) edu2 ∗ ln(pfpll) -0.163 -0.084* -0.084 (0.04) (0.03) (0.47) edu3 ∗ ln(pfpll) 0.259 0.315*** 0.315 (0.04) (0.04) (0.43) edu1 ∗ ln(ptxtl) -0.041 -0.143*** -0.143 (0.03) (0.03) (0.26) edu2 ∗ ln(ptxtl) -0.206 -0.165*** -0.165 (0.05) (0.05) (0.58) edu3 ∗ ln(ptxtl) 0.220 0.218*** 0.218 (0.06) (0.05) (0.54) edu1 ∗ ln(pomfg) -1.538*** -1.221*** -1.221*** (0.04) (0.03) (0.30) edu2 ∗ ln(pomfg) -0.665 -0.798*** -0.798 (0.06) (0.06) (0.67) edu3 ∗ ln(pomfg) -1.219* -1.230*** -1.230* (0.06) (0.06) (0.62) edu2 0.747*** 0.555*** 0.555*** (0.01) (0.01) (0.01) edu3 1.697*** 1.433*** 1.433*** (0.01) (0.02) (0.01) age 0.055*** 0.055*** (0.00) (0.00) age2 -0.001*** -0.001*** (0.00) (0.00) gender 0.529*** 0.529*** (0.00) (0.02) constant -5.147*** -6.515*** -6.515*** (0.00) (0.01) (0.06) R2 0.5865 0.643 0.643 * p<0.05, ** p<0.01, *** p<0.001 The standard errors are in parenthesis. threshold, the labour income effects fall as the household gets richer. At the higher tail of income distribution, the income effects have an increasing trend though higher standard errors are observed. The maximum gain is more than one percent of the total monthly household expenditure, whereas the minimum is approximately 0.75 percent 29 PLr PLu.7 . 8 . 9 1 1. 1 1. 2 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure matching_smth_ie_wage upper_limit lower_limit The solid line shows average total effect using locally weighted regression. The broken lines report 5% confidence band using bootstrap method labour income effects (matching grid) Figure 2.7: Labour income effects for middle income groups. All income groups have statistically significant labour income effects. The labour income effects are different in the rural and urban sectors (Figure 2.8). Poor households in the rural sector have significantly higher gains compared to their urban counterparts, whereas in the urban sector, all income groups have almost similar gains. Moreover, labour income effects are significantly greater compared to consump- tion effects. It appears that poor households in the rural sector, classified by the official poverty lines (depicted as vertical lines in Figure 2.8), have greater gains from labour market adjustment that help to offset the uneven gains from consumption effects. PL0 . 5 1 1. 5 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure matching_smth_ie_wage upper_limit lower_limit The solid line shows average total effect using locally weighted regression. The broken lines report 5% confidence band using bootstrap method labour income effects (matching grid) − Rural (a) Rural PL1 1. 2 1. 4 1. 6 1. 8 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure matching_smth_ie_wage upper_limit lower_limit The solid line shows average total effect using locally weighted regression. The broken lines report 5% confidence band using bootstrap method labour income effects (matching grid) − Urban (b) Urban Figure 2.8: Labour income effects -rural, urban 30 2.5.5 Total distributional effect The main result of this exercise is depicted in Figure 2.9. All the effects are added to find the overall distributional impact of tariff reforms in India. The welfare gains are statistically significant for all income groups. However, the total effects are non- monotonous with income level. The richest section has a rising trend but with higher standard errors. The figure also gives the estimated kernel density function of monthly per capita household expenditure. Households below the poverty line have almost simi- lar percentage gains except for the extreme lower quantiles. Figure 2.10(a) and 2.10(b) PLr PLu0 . 5 1 1. 5 2 2. 5 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure Total effect Upper limit Lower limit Density: log(MPCE) The solid line shows average total effect using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Total effects Figure 2.9: Total effects depict the rural-urban breakup. In the rural sector poor households have relatively higher welfare gains compared to middle income groups. Extremely rich households in the rural sector have upward rising welfare gains, but it comes with higher stan- dard errors. The welfare gain monotonically increases with income level in the urban sector. Therefore, the tariff reforms have a pro-poor effect in the rural sector and a pro-rich effect in the urban sector. These differentiated effects are linked to the fact that labour market adjustment in the rural sector has resulted in relatively higher gains for the poor. Given the upward trend in inequality in recent years, this result must be put in the proper context. My analysis only explains that part of inequality which could be attributed to tariff reforms from 1990 to 2005. There are several other reasons for increasing inequality, particularly in the urban sector. Topalova (2007) finds that district-level inequality is unaffected by tariff reforms. My results explain that distri- butional impact could be different for rural and urban areas. In fact, the inequality 31 PL0 . 5 1 1. 5 2 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure Total effect Upper limit Lower limit Density: log(MPCE) The solid line shows average total effect using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Total effects − Rural (a) Rural PL0 1 2 3 4 co m pe ns at in g va ria tio n %  o f h ou se ho ld  e xp en di tu re 5 6 7 8 9 log per capita expenditure Total effect Upper limit Lower limit Density: log(MPCE) The solid line shows average total effect using locally weighted regression. The broken lines report 5% confidence band using bootstrap method Total effects − Urban (b) Urban Figure 2.10: Total effect -rural, urban measures are affected by tariff reforms. Topalova (2007) used two measures of inequality in her analysis: 1) standard deviation of log consumption expenditure and 2) the mean logarithmic deviation of consumption. Both these inequality measures are unchanged if consumption level for each household increases proportionately. In my result, compensating variations as a percentage of initial household expenditure for all households are not similar across income groups. If CV j/ej = c (some constant) for all households, the welfare gain expressed in terms of monetary value for each household is constant, i.e., dx/x = c for any consumption expenditure x. It is easy to show that the new inequality measures remain the same as long as welfare gain as a percentage of total income is the same across income groups. If change in consumption expenditure is dln(x), the variance of log of consumption expenditure after reform is given by var (ln(x) + dln(x)) = var (ln(x)) + var (dln(x)) + 2cov (ln(x), dln(x)) = var (ln(x)) since dln(x) = dx/x = c (constant). Similarly, the other measure (generalised entropy coefficients) of inequality is unchanged when dx/x is constant. The mean logarithmic deviation of consumption is given by I(0) = ∫ x µ ln (µ x ) f(x)dx 32 where µ is mean expenditure. After tariff reforms, if the log of consumption expenditure changes to ln(x) + dln(x), the new expenditure is mx and the new mean expenditure is mµ where m = ec. It is straight forward to see that the inequality measure I(0) is unchanged for a proportionate increase in income. In this analysis, I find some explanations for increasing inequality, specifically in the urban sector, due to tariff reforms in India. The welfare gain as a percentage of initial expenditure is different across income levels. In other words, tariff reforms lead to uneven gains. I find that all income groups have positive and statistically significant welfare gains. This is contrary to the findings in Topalova (2007). The most interesting result of Topalova (2007) is that the urban poverty is unaffected, but the rural poverty ratio and poverty gap reduced less rapidly due to tariff reforms. The districts which were more exposed to trade reforms have a negative correlation between district poverty and reform measures in rural areas. The contradictory outcome in my exercise demands fur- ther research on this question. My approach (general equilibrium) and that of Topalova (2007) are completely different and therefore, it is hard to identify the sources of this contradictory result. As I have mentioned earlier, the general equilibrium methodology of evaluating policy change has several advantages. It can explicitly show the gains and losses originating from product market and factor market adjustments. However, the results crucially depend on precise estimates of certain parameters. In my approach, I have ignored some factors and that might be a reason of differentiating results. The passthrough rates of tariff adjustment are different for different product groups depend- ing on market structure and other policies like exchange rate regime. If passthrough rates are very low and it is even lower for food group then my estimate of total welfare gain is over estimated. Similarly, the estimates of wage price elasticity and elasticity of the price of non-traded goods with respect to the price of traded goods crucially depend on estimating assumptions. For example, while estimating the elasticities, I have ignored all other internal and external policy changes (exchange rate, non-tariff barriers, deregulation in internal sector, promotion of competitiveness through deli- cencing, etc). All these contemporaneous changes may be correlated with some of the explanatory variables. Though I have used a dummy for the liberalisation period, this mis-specification leads to biased estimates of elasticities. Suppose all the excluded vari- ables have co-movement with tariff rate in same direction. Theoretically, delicencing may improve labour productivity through competitiveness and improved technological investment. This in turn increases wage rate. Therefore, under the above postulations, 33 omitted variable specification leads to underestimation of the wage price elasticity. Sim- ilarly, if the prices of non-traded goods are positively related with the excluded variables (i.e. more deregulation leads to lower prices of non-traded goods) then the estimates of the price elasticities of non-traded goods with respect to the price of traded goods will be biased upward. I find that there are positive welfare gains for all income groups. This does not mean that the gains are large enough to pull the large mass above the poverty line. The gains shown in the figures are in percentage of total expenditure. Therefore, in terms of absolute gain the poorer section still gets the least advantage from tariff reforms. It is also important to note that the district level poverty measures are sensitive to the choice of poverty lines. In the differential exposure approach of Topalova (2007), the results, therefore, may suffer from this problem. Whereas, my approach is more robust. 2.6 Conclusion In this chapter I have used household survey data to find the distributional impact of tariff reforms in India. In 1991 India adopted comprehensive measures of internal and external reforms. The most important aspect of external reform was drastic reduction in tariff rates with some other liberalisation measures. The literature suggests that growth, productivity, competitiveness and efficiency are positively affected by reform measures. However, the question of distributional impact is largely unanswered. If gains from reform do not reach all sections in equal proportions, we need some complementary strategy to combat the widening gap between the rich and poor. The modelling and econometric techniques, which are useful to find the macroeconomic impact of trade policy, are sometimes hard to apply in this context. I estimate general equilibrium consumption effects and labour income effects at the household level. The distributional effects are derived for different income groups using locally weighted regression. The individual consumption effects of traded goods and non-traded goods and labour income effects have different magnitudes for the richer and poorer groups. The overall effects show quite a different pattern in the rural and urban sectors. The distributional effect of tariff reforms in India is pro-poor in the rural sector and pro-rich in the urban sector. The contradictory results in the literature require further analysis on this question. 34 3. Castes and Labour Mobility “If there is no enthusiasm, life becomes drudgery - a mere burden to be dragged. Nothing can be achieved if there is no enthusiasm. The main reason for this lack of enthusiasm on the part of a man is that an individual loses the hope of getting an opportunity to elevate himself. Hopelessness leads to lack of enthusiasm. The mind in such cases becomes deceased...When is enthusiasm created? When one breathes an atmosphere where one is sure of getting the legitimate reward for one’s labour, only then one feels enriched by enthusiasm and inspiration” “This condition obtains even where there is no slavery in the legal sense. It is found where as in caste system, some persons are forced to carry on the prescribed callings which are not their choice...” B. R. Ambedkar (Chief architect of the Indian Constitution.) 3.1 Introduction Large macroeconomic changes and major structural changes of the economy often go hand-in-hand. These phases are often associated with winners and losers at the level of individuals, sectors or social groups. Hence, managing the microeconomic distri- butional consequence of macroeconomic changes is often a key challenge for policy- makers. But do large-scale macroeconomic changes tend to accentuate or dampen historical inequities? Do these economic redistributions necessarily benefit the eco- nomically stronger sections of society or can they also lead to a reduction in economic inequality? What are the key margins which account for these distributional changes? The Indian economy provides a natural environment to investigate these questions due to a dramatic process of structural changes, implementation of a series of compre- hensive reforms and rapid economic growth since the 1980s. At the same time, India has had a long history of social division due to a traditional institution of caste that created a social stratification along education, occupation and income lines. These fac- tors motivate our focus on contemporary India in addressing the questions above. We 35 study the evolution of the economic well-being of individuals belonging to historically disadvantaged castes between 1983 and 2005. We show that there has been a significant narrowing of economic backwardness of this group relative to the rest. We find that a large part of this economic catch-up has occurred through a catch-up in the relative education attainment level of this group. The past 25-30 years have been a period of massive changes in the Indian econ- omy. Average annual GDP growth rates have climbed rapidly from the anaemic 3-3.5 percent that characterised the first 35 years since 1947 to between 8 and 10 percent. Accompanying this growth take-off has been a hastening process of structural trans- formation of the economy. The agricultural sector, which historically had the largest employment and output share, has rapidly lost ground in both during this period. Such rapid structural changes often deeply affect the lives of people in these economies by redistributing income and economic opportunities from some groups to others. As an example, a particularly emotive issue that has been debated energetically with regard to the Indian experience is the effect of this economic take-off on the fortunes of the poor. In this chapter we study the impact of the recent rapid transformation of the In- dian economy on one such historically disadvantaged group: the Scheduled Castes and Scheduled Tribes (SC/STs). SC/STs were historically economically backward, mostly very poor, concentrated in low-skill (mostly agricultural) occupations and primarily rural2. Moreover, they were also subject to centuries of systematic caste-based discrim- ination both economically and socially. This was so endemic that the constitution of India aggregated these castes into a Schedule of the constitution and provided them with affirmative action cover in both education and public sector employment. Indeed, this was viewed as a key component of attaining the ultimate policy goal of raising the social and economic mobility of the SC/STs to the levels of the non-SC/STs. The existence of caste-based frictions in labour market allocations and social match- ing processes have been documented by a number of micro-level studies. Indeed, a key goal of the reservations policy was to make it easier for, say, the child of an illiterate SC or ST farm worker living below the poverty line to get educated and find productive employment in a better paying occupation. How have the tectonic changes in India 2The Scheduled Castes (SC), also known as Dalit, and the Scheduled Tribes (ST) are two historically disadvantaged social groups based on their caste identity. The Constitution of India has given express recognition to these groups in the form of a schedule. There are 15% SC and 7.5% ST population in India according to 2001 census. The caste based indentity of a person is easily recognisable by their last names. Change of caste (that is self seclection) is very hard to undertake because caste indetity of a household is a community knowledge and it is determined by patriarchal lineage. 36 since the early 1980s affected this goal? What has been the net effect on the fortunes of SC/STs of the interplay between these micro-level frictions and the massive aggregate macroeconomic changes in India over the past two decades? Has the rapid growth per- colated down to the SC/STs in terms of tangible changes in their economic and social conditions? Is the primary reason for the economic deprivation of these underprivileged castes the types of occupations they tend to work in, i.e., do successive generations of SC/STs tend to get stuck in low wage jobs? Alternatively, is the key impediment the lack of education, i.e., do they get stuck in low wage jobs due to the lack of education? Or, is ongoing discrimination in occupations and wages the primary problem facing these groups? This chapter attempts to answer some of these questions. We use data from five successive rounds of the National Sample Survey (NSS) from 1983 to 2004-05 to analyse patterns of occupation choices, education attainment and wages of both SC/ST and non-SC/ST households. We conduct our analysis along two dimensions. First, we contrast the time-series evolution of education, occupation and industry choices and wages of SC/STs with their non-SC/ST counterparts from the same age cohort. We conduct this cohort-level analysis both at an aggregated generation level of parents and children as well as at a more disaggregated level of five different age cohorts.3 Second, we contrast the time series behaviour of the intergenerational persistence of education, occupation, industry of employment and wage levels of SC/ST and non-SC/ST households. Our analysis yields four main results. First, while SC/ST households are, on average, less educated than their non-SC/ST counterparts throughout the sample period and across cohorts, the education attainment levels of SC/STs have been converging toward the level of their non-SC/ST cohort. This trend is particularly pronounced for SC/ST children. Moreover, the trend towards education convergence of the two groups emerges both in rural and urban sectors but is sharper in urban areas. The trend also shows up clearly across occupations. Second, there have been similar compositional changes in the occupational distri- butions of SC/STs and non-SC/STs between 1983 and 2004-05. Children and parents of both SC/ST and non-SC/ST households have been moving out of low skill agrarian occupations into relatively higher skill occupations. However, these changes have oc- curred slightly faster for SC/STs. As a result, the occupation distribution of the two groups appear to be converging during this period. We also study trends in industry 3We look at aggregated cohorts of parents and children to set the stage for the intergenerational mobility analysis. 37 mobility of the two groups and find that the results for industry mobility are broadly similar to those for occupation mobility. Third, we find a clear trend of convergence of the relative wage of the two groups towards one, i.e., the median wage premium of non-SC/STs relative to SC/STs has declined systematically from 17 percent in 1983 to 3 percent 2004-05. The trend is particularly strong for children where the wage premium of non-SC/STs has declined from 14 percent to approximately zero. For the parent cohort, the non-SC/ST wage premium has declined from 25 percent to 10 percent during this period. This pattern of relative wage convergence also emerges in mean wages and across more disaggregated age cohorts. Overall, our conditional wage regressions suggest that less than 5 percent of the observed wage gap is attributable to SC/ST factors alone (independent of the other correlates of wages like education and occupation). To put these wage gaps in perspective, the median white male to black male wage premium in the US has hovered stubbornly between 25 and 40 percent over the past 35 years, which makes the SC/ST relative wage behaviour in India even more striking. Fourth, we find that intergenerational mobility of SC/STs has risen faster than that of non-SC/ST households in both education attainment rates and wages. The probabil- ity of an SC/ST child changing his level of education attainment relative to the parent was just 42 percent in 1983 but rose sharply to 67 percent by 2004-05. The correspond- ing probabilities of a change in education attainment for a non-SC/ST child were 57 percent and 67 percent. Hence, there has been a clear convergence of intergenerational education mobility rates between SC/STs and non-SC/STs. Correspondingly, the elas- ticity of wages of children with respect to the wages of their parent has declined from 88 percent to 45 percent for SC/ST households and from 76 to 58 percent for non-SC/ST households. Clearly, the intergenerational income mobility rates have also converged. Lastly, intergenerational occupational and industry mobility rates have increased for both groups during this period. However, these changes in occupational and industry mobility rates have been relatively similar across the two groups. As a result, children in non-SC/ST households continue to be more likely to work in a different occupation and/or different industry than their parent relative to children from SC/ST households. In summary, these results suggest some uplifting answers to the questions we set out to answer. Over the last 20 years SC/STs have sharply narrowed both their education and wage gaps relative to non-SC/STs. The fact that these trends are sharpest amongst younger age-cohorts and amongst urban households suggests that the overall statistics are likely to improve even more sharply in their favour in the coming years as these 38 cohorts become older and as the country becomes more urbanised. Moreover, children from the historically disadvantaged SC/ST households are increasingly raising their ed- ucation attainments levels, switching occupations and improving their income positions relative to their parents. Crucially, intergenerational income and educational mobility of SC/ST households has been rising faster than for non-SC/STs. Overall, we conclude that neither the lack of occupational mobility nor the lack of education have been a ma- jor impediment toward the SC/STs taking advantage of the rapid structural changes in India during this period to better their economic position. This period of rapid structural changes appears to have been very beneficial for SC/STs who have used this period to rapidly narrow their huge historical economic disparities with non-SC/STs. To the best of our knowledge, our’s is the first study to jointly analyse caste dif- ferences in education, occupation, industry and wage outcomes in a single study, track the time series evolution of these outcomes, and do so using data that covers the en- tire country. It is worth reiterating that we do this using the NSS data which has the broadest coverage for India both spatially and over time. There exists a large literature which has investigated the existence and extent of labour market discrimination in India. Amongst others, Banerjee and Knight (1985) and Madheswaran and Attewell (2007) have studied the extent of wage discrimination faced by SC/STs in the urban Indian labour market. Borooah (2005) has studied the extent of discrimination in employment in the urban labour market. Ito (2009) studies both wage and employment discrimination simultaneously by examining data from two Indian states – Bihar and Uttar Pradesh. Our study differs from these in that we examine the data for all states and for both rural and urban areas. Moreover, as opposed to most of these studies, our study controls for the presence of occupation and industry effects on wage outcomes. Lastly, by using data for five rounds of the National Sample Survey of households we are also able to provide a time series perspective on the evolution of SC/ST fortunes in India, a feature that other studies have typically not examined. While there has been considerable work on intergenerational mobility in the US and other western countries (see Becker and Tomes (1986), Behrman and Taubman (1985), Haider and Solon (2006) amongst others), this issue has received remarkably little attention in the work on India. The two notable exceptions are Jalan and Murgai (2009) and Maitra and Sharma (2009) both of which focus on intergenerational mobility in education attainment. The biggest difference between our work and these other studies is that we examine intergenerational mobility patterns not just in education 39 attainment but also in occupation choices, industry of employment, and income. We are not aware of any other study that documents intergenerational mobility patterns in occupation, industry, and income. Our work also differs from Jalan and Murgai (2009) and Maitra and Sharma (2009) in two other respects: (a) we use a much larger sample of households due to our use of the NSS data; and (b) by examining multiple rounds of the NSS data we are also able to study the time-series evolution of intergenerational mobility patterns in India.4 In the next section we describe the data and our constructed measures as well as some summary statistics. Section 3.3 contrasts SC/STs with their non-SC/ST cohorts in terms of the evolution of the distributions of education attainment rates, occupations, industry of employment and wages. Section 3.4 presents and discusses the evidence on intergenerational mobility, while the last section concludes. 3.2 Data Our data comes from the National Sample Survey (NSS) of India and its various rounds. In particular, we use the NSS Rounds 38 (1983), 43 (1987-88), 50 (1993-94), 55 (1999- 2000) and 61 (2004-05). The survey covers the whole country except for a few remote and inaccessible pockets. The rounds that we use include detailed information on over 120,000 households and 600,000 individuals. Our working sample consists of all house- holds heads and their children/grandchildren who provided their 3-digit occupation code information and their education information. We restrict our sample to males whose age is between 16 and 65.5 Our focus is on full-time working individuals who are defined as those that worked at least 2.5 days per week, and who are not currently enrolled in any education institution. We conduct all our data work using a sample in which the criteria above are satisfied for both household’s head and at least one child or grandchild in that household. This restriction is necessitated by our interest in ex- amining inter-generational mobility trends. We choose to work with this sample for our 4In related work Munshi and Rosenzweig (2009) document the lack of labour mobility in India. Also, Munshi and Rosenzweig (2006) show how caste-based network effects affect education choices by gender. 5We also consider a broader sample in which we do not restrict the gender of the children and find that our results remain robust (in fact, majority of the children working full-time in our sample are male). We choose the restriction to only males for two reasons. First, female led households are few and usually special in that those households are likely to have undergone some special circumstances. Second, since there are a number of societal issues surrounding the female labour force participation decision which can vary both across states and between rural and urban areas, focusing only on males allows us to avoid having to deal with these complications. 40 intra-generational exercises as well in order to retain comparability of the samples and the results. This selection leaves us with a sample of around 43,000-51,000 individuals, depending on the survey round and we refer to this sample as “working” sample. If we do not restrict the sample to households with working heads and at least one working child or grandchild, the sample size grows to between 136,000-152,000 individuals, de- pending on the round. We refer to this sample as“extended sample” and in the later sections verify the robustness of our key results to these alternative sample restrictions.6 Data on wages are more limited. The sub-sample with complete wage data for both the head of household and at least one child or grandchild in the same household consists of, on average across rounds, about 7,000-9,000 individuals which is considerably smaller than our working sample but large enough to facilitate formal analysis. In the extended sample, we have wage data for about 55,000 individuals across rounds. Wages are obtained as the daily wage/salaried income received for the work done by respondents during the previous week (relative to the survey week). Wages can be paid in cash or kind, where the latter are evaluated by the current retail prices. We convert wages into real terms using state-level poverty lines that differ for rural and urban sectors. We express all wages in 1983 Maharashtra prices. Details regarding the dataset are contained in the Appendix. Our education variable contains 5 categories: not-literate; literate but below pri- mary; primary education; middle education; and secondary and above education (which includes higher secondary, diploma/certificate course, graduate and above in different professional fields, postgraduate and above). These categories are coded as education categories 1, 2, 3, 4 and 5 respectively. We also calculate years of education from more disaggregated education codes in the questionnaires. The conversion of education years from education codes maintains uniformity across rounds. Our dataset also contains information about occupation choices of individuals. In particular, we know the three- digit occupation code associated with the work that each individual performed over the last year (relative to the survey year). We use only those individuals for whom the occupation code reported for the last year coincided with the occupation code for which wages over the last week were collected (relative to the survey week). Our dataset also contains information on the four-digit industry of employment for each individual. 6Both the number of households with co-residing generations as well as the total number of individ- uals living in such households are not too different across rounds. This suggests to us that co-residence patterns have not changed too dramatically during the period under study. Hence the representative- ness of the sample under this identification should have remained comparable across rounds. 41 Table 3.1: Sample summary statistics (a) children (b) parents All age edu year %married age edu year %married %rural hh size 1983 22.83 4.14 0.53 51.67 1.90 0.92 0.81 7.18 (0.04) (0.03) (0.00) (0.07) (0.03) (0.00) (0.00) (0.02) 1987-88 23.13 4.42 0.53 51.65 2.11 0.92 0.83 6.98 (0.04) (0.03) (0.00) (0.06) (0.03) (0.00) (0.00) (0.02) 1993-94 23.17 5.21 0.48 51.78 2.54 0.94 0.82 6.51 (0.04) (0.03) (0.00) (0.06) (0.03) (0.00) (0.00) (0.02) 1999-00 23.51 5.91 0.46 51.60 3.08 0.94 0.81 6.56 (0.05) (0.04) (0.00) (0.07) (0.04) (0.00) (0.00) (0.02) 2004-05 23.77 6.40 0.46 51.63 3.43 0.94 0.80 6.39 (0.05) (0.04) (0.00) (0.07) (0.04) (0.00) (0.00) (0.02) Non-SC/ST 1983 23.00 4.70 0.52 52.04 2.28 0.92 0.79 7.29 (0.05) (0.04) (0.00) (0.08) (0.03) (0.00) (0.00) (0.03) 1987-88 23.30 4.99 0.51 51.98 2.49 0.93 0.80 7.06 (0.05) (0.04) (0.00) (0.08) (0.03) (0.00) (0.00) (0.02) 1993-94 23.36 5.78 0.47 52.10 3.01 0.94 0.79 6.6 (0.05) (0.04) (0.00) (0.07) (0.04) (0.00) (0.00) (0.02) 1999-00 23.76 6.53 0.47 52.01 3.64 0.95 0.78 6.62 (0.05) (0.04) (0.00) (0.08) (0.05) (0.00) (0.00) (0.03) 2004-05 24.04 6.87 0.46 52.01 3.93 0.95 0.77 6.42 (0.06) (0.04) (0.01) (0.08) (0.05) (0.00) (0.00) (0.03) SC/ST 1983 22.30 2.38 0.56 50.59 0.81 0.92 0.89 6.86 (0.08) (0.06) (0.01) (0.13) (0.03) (0.01) (0.01) (0.04) 1987-88 22.63 2.67 0.56 50.72 1.01 0.91 0.90 6.76 (0.08) (0.05) (0.01) (0.12) (0.04) (0.00) (0.00) (0.04) 1993-94 22.61 3.57 0.49 50.92 1.29 0.92 0.90 6.25 (0.08) (0.06) (0.01) (0.13) (0.04) (0.00) (0.00) (0.04) 1999-00 22.85 4.32 0.46 50.61 1.72 0.94 0.88 6.41 (0.09) (0.07) (0.01) (0.13) (0.06) (0.00) (0.01) (0.04) 2004-05 23.05 5.13 0.45 50.66 2.16 0.94 0.87 6.30 (0.09) (0.07) (0.01) (0.14) (0.07) (0.00) (0.01) (0.05) Notes: This table reports summary statistics for our sample. Panel (a) gives the statistics for the generational subsample of children, while panel (b) gives the statistics for the household heads (parents). Standard errors are reported in parenthesis. Table 3.1 gives some summary statistics of the data. Panel (a) reports average age, years of education, share of males and married individuals among children; while panel (b) reports the corresponding statistics for household heads (parents). Panel (b) 42 also reports the percentage of rural households in our sample, as well as the average household size. Note that “All” refers to the full working sample, while the “Non- SC/ST” and “SC/ST” panels refer to the corresponding sub-samples. Household-heads are around 52 years of age while their male working children are typically around 23 years old. Around 81 percent of surveyed households are rural and engaged in farming/pastoral activities. This number is slightly higher for SC/ST households, 88-89 percent of whom live in rural areas on average. Of the working children living with the Household-head, 49 percent are married on average. While the percent of married children has declined over time, this change was characteristic of both non-SC/ST and SC/ST children. Finally, the average education level of children is greater than that of parents for both SC/STs and non-SC/STs, and has increased over time. Non-SC/STs are also consistently more educated than SC/ST. The proportion of SC/ST households in the sample across the different rounds is around 24 percent. 3.3 Intragenerational cohort comparison: how have the scheduled castes fared? We start our analysis by comparing SC/STs with non-SC/STs across age and genera- tional cohorts. We construct cohorts in two ways. First, for every round of the survey we split our sample into five broad age cohorts: 16-25, 26-35, 36-45, 46-55 and 56- 65. Second, in each round we split the sample into household heads and children in co-residence with a household head. We call these generational cohorts “Parents” and “Children”, respectively. For each age and generational cohort we compute the occu- pation distribution, the industry distribution, the average education attainment level and the average daily wage earned for the entire group as well as for SC/STs and non- SC/STs separately. Issues of particular interest to us are: (a) whether the education attainment levels of SC/ST children and parents are converging to the levels of their non-SC/ST cohorts? (b) whether their occupation and industry choices are converging over time; and (c) whether wages of SC/STs are converging to non-SC/ST levels. A few notes on our generational cohort classification are in order. First, we refer to household heads as parents. In a literal sense household heads are not always the parents of younger working members in the household since there are a few households with a grandparent as the head of a household that also contains his working children and grandchildren.7 More generally, our terminology is meant as a stand-in for parent- 7Due to the 16-65 age restriction, however, the share of such households is small in our sample. 43 figures. Second, since we evaluate the performances of parents and children in successive rounds of the survey, there will definitely be cases where children in one round become household heads, and therefore “parents”, in later rounds. However, across the different survey rounds the mean age of parents remains relatively stable at around 52 years while the mean age of children remains around 23 years. Thus, all children under the age of 30 in 1983 would still be less than the mean parent age in the last round of the sample in 2004-05. This suggests that while there definitely is some movement of people from one cohort into another over time, it doesn’t appear to be a large share of the sample. Hence, the changes over time in the statistics of parents are not solely attributable to the changing age composition of the cohort, i.e., due to children in earlier rounds becoming parents in later rounds. Third, we choose to work with age cohorts rather than birth cohorts. This is a deliberate choice which reflects our interest in determining the effects of changing ag- gregate conditions and how they alter the incentives of agents over time. The age-cohort approach allows us to contrast the behaviour of 16-25 year olds in 2004-05 with 16-25 year olds in 1983. If the behaviour is different then it would indicate that the incentives underlying the choices being made by this age cohort have changed over this period. While some of the dynamics of the age-cohorts may potentially include the cohort effects related to birth, the constant and historically determined caste identity of the groups combined with the impossibility of changing caste identities makes us less concerned about a big “cohort” effect underlying our results. The alternative of examining birth cohorts and tracking them over time makes it harder to make this deduction since some of the changes over time would also reflect the ageing process. 3.3.1 Education attainment We start with the record on education attainment rates. Panel (a) of Table 3.2 shows the average education attainment level in years of education of the overall population as well as those for working children and parents separately. Both generational groups increased their education attainment levels over the sample period. Panel (b) of Table 3.2 shows the relative education gap between non-SC/STs and SC/STs, computed as the ratio of their corresponding education attainments. In 1983, non-SC/STs had two years more education than SC/STs8 -a 120 percent relative discrepancy. However, over the sample period, there was a clear trend towards convergence in education levels of 8The level of education attainments for SC/STs in 1983 was 1.65 years, while for non-SC/STs it was 3.64 years. 44 SC/STs toward their non-SC/ST counterparts as the gap declined to just 47 percent by 2004-05. This trend is particularly pronounced for the cohort of children. While the difference in 1983 was almost one year, by 2004-05 this had narrowed to one third of a year.9 Both groups increased their education attainment levels over the period with the SC/STs rising faster. Table 3.2: Education attainment levels and gaps (a). Levels (b). Gaps all children parents all children parents 1983 3.15 4.14 1.90 2.20 1.98 2.80 (0.02) (0.03) (0.03) (0.05) (0.05) (0.12) 1987-88 3.39 4.42 2.11 2.03 1.87 2.47 (0.02) (0.03) (0.03) (0.04) (0.04) (0.09) 1993-94 4.04 5.21 2.54 1.80 1.62 2.33 (0.02) (0.03) (0.03) (0.03) (0.03) (0.08) 1999-00 4.68 5.91 3.08 1.67 1.51 2.12 (0.03) (0.04) (0.04) (0.03) (0.03) (0.07) 2004-05 5.11 6.40 3.43 1.47 1.34 1.82 (0.03) (0.04) (0.04) (0.02) (0.02) (0.07) Notes: This table presents the average education attainment levels in years for our overall benchmark sample and separately for two generational groups – parents and children. Gaps refer to the ratio of average education attainment levels of non-SC/STs to SC/STs for the same three groups. The reported statistics are obtained for each NSS survey round which is shown in the first column. Standard errors are in parenthesis. The increase in average education levels has been more tepid for parents relative to their children. However, there is a clear trend toward convergence in the average levels across the parents: the relative gap declined from 180% in 1983 to just 82% in 2004-05. A related issue of interest is the distribution of parents and children across the five education categories. In particular, is the change in the average attainment level due to more illiterates beginning to go to primary school or is it primarily due to more people going on to middle school or higher? We answer this question using Figure 3.1. Panels (a) and (b) of Figure 3.1 show the distribution of the workforce across educa- tion categories and the corresponding non-SC/ST–SC/ST differences for children and parents respectively. The top graph of Panel (a) shows the distribution of non-SC/ST 9In 1983, education attainment levels of SC/ST and non-SC/ST children were 2.38 and 4.70 years, respectively. By 2004-05 these levels have increased to 5.13 and 6.87 years, respectively. 45 children across the five education categories (left set of bars) and the corresponding distribution of SC/ST children (right set of bars). It is clear that SC/ST children are systematically less educated than their non-SC/ST counterparts. The difference is most glaring in the lowest and highest categories. In category 1 (the illiterate groups) SC/STs are hugely over-represented while in category 5 (secondary education or above) they are strongly under-represented. Figure 3.1: Education distribution of children and parents 0 20 40 60 80 10 0  Non−SC/ST SC/ST 1983 1987−88 1993−94 1999−00 2004−05 1983 1987−88 1993−94 1999−00 2004−05 Distribution of workforce across edu, children Edu1 Edu2 Edu3 Edu4 Edu5 0 20 40 60 80 10 0  Non−SC/ST SC/ST 1983 1987−88 1993−94 1999−00 2004−05 1983 1987−88 1993−94 1999−00 2004−05 Distribution of workforce across edu, parents Edu1 Edu2 Edu3 Edu4 Edu5 − 20 − 10 0 10 20 1983 1987−88 1993−94 1999−00 2004−05 Gap in workforce distribution across edu, children Edu1 Edu2 Edu3 Edu4 Edu5 − 20 − 10 0 10 20 1983 1987−88 1993−94 1999−00 2004−05 Gap in workforce distribution across edu, parents Edu1 Edu2 Edu3 Edu4 Edu5 (a) (b) Notes: Panel (a) of this figure presents the distribution of workforce of children and parents across five education categories across different NSS rounds. The left set of bars on each figure refers to non-SC/STs, while the right set is for SC/STs. Panel (b) presents absolute gaps in the distribution of non-SC/STs relative to SC/STs across five education categories. The gaps are also reported for children and parents. See the text for the description of how education categories are defined (category 1 is the lowest education level - illiterate). The scale of the lack of education in India, both in general and amongst SC/STs, is probably best summarised by the fact that as recently as in 1983, about 64 percent of SC/ST children were either illiterate or had below primary level education while the corresponding number for non-SC/STs was 40 percent. These numbers declined to 30 percent for SC/ST children and 19 percent for non-SC/ST children by 2004-05. 46 The figure also makes clear that there has been a sustained decrease over time in the share of illiterates amongst both SC/STs and non-SC/STs. Thus, by 2004-05 the proportion of illiterate SC/ST children (category 1) fell from 50 percent to 17 percent. This was by far the sharpest change amongst all education categories. Correspondingly, the sharpest increases occurred in education categories 4 (middle school) and 5 (sec- ondary or above). The pattern for non-SC/STs is broadly similar except for the fact that their sharpest increase occurred in the secondary or above education category 5. The top graph of Panel (b) of Figure 3.1 shows the difference between the percentage of non-SC/ST children and SC/ST children within each education category. Thus, the first bar from the left in the top graph of Panel (b) shows that the percentage of all SC/ST children who belonged to education category 1 in 1983 exceeded the correspond- ing percentage of non-SC/ST children in category 1 in that year by over 20 percentage points. This panel captures, to some extent, the tendency toward convergence of pat- terns across the two groups. With one exception, the differences in the proportion of children in the different education categories either stayed constant or tended towards convergence for the two groups. The exception was in category 5 (secondary or higher). In 1983 about 19 percent of non-SC/ST children had secondary school or higher levels of education while the number was just around 6 percent for SC/STs. By 2004-05, 39 percent of non-SC/ST children had secondary or higher levels of education while the number for SC/STs had risen to 22 percent. Clearly, for both groups there has been a significant increase in the share of children with secondary or higher education but the difference between the two groups has continued to remain very high. The bottom panel of Figure 3.1 shows the same information for parents. There are a few key differences between the education distribution patterns for parents and their children. First, the share of illiterates and those with less than primary education (edu- cation categories 1 and 2) is higher for both SC/ST and non-SC/ST parents throughout and declined at a slower rate than that of their children. Thus, in 1983 the combined share of categories 1 and 2 was 69 percent for non-SC/ST parents and 88 percent for SC/ST parents. These numbers fell over time but still remained at a very high 48 per- cent and 66 percent, respectively, in 2004-05. Second, at the high end of the education distribution the changes have been much more tepid for parents of both groups relative to that of their children. The share of secondary or higher educated parents amongst non-SC/STs rose from 7 percent in 1983 to 21 percent in 2004-05. Correspondingly, the share for SC/ST parents rose from 2 percent to 9 percent. Third, in contrast to the pattern amongst children, there is no clear trend towards homogenisation of the two 47 groups in their educational composition. This feature is clearly brought out in Panel (b) of Figure 3.1, in particular, by the heights of the bars depicting the shares for education categories 1 and 5.10 Figure 3.2: Education gaps by age cohorts 1 1. 5 2 2. 5 3 3. 5 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 overall 1 1. 5 2 2. 5 3 3. 5 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 rural 1 1. 5 2 2. 5 3 3. 5 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 urban Notes: The figures show the evolution of the relative education gap between non-SC/STs and SC/STs over time for different age groups. Panel (a) presents the results for the overall sample, while panels (b) and (c) report the results for rural and urban households separately. Do these trends in aggregated generational cohorts mask key differences in the rel- ative movements within more disaggregated age cohorts? To investigate this we com- pute the education attainment levels of non-SC/STs relative to SC/STs within five age-cohorts for each survey round. Figure 3.2 plots the result. Panel (a) reveals a clear pattern of education convergence across the different age-cohorts over time. Impor- tantly, the convergence appears to be the sharpest amongst the younger cohorts. Given the large concentration of households in rural areas, a related question is whether the trends in education attainment rates are different between rural and urban households. To address this, we split the different age-cohorts into rural and urban households and then plot the education attainment gaps for the two sectors separately in panels (b) and (c) of Figure 3.2. Three features of the figure are noteworthy. First, the variation in education levels across cohorts is much smaller in rural areas than in urban areas. Second, except for the oldest rural cohorts, education attainment levels have been con- verging in both rural and urban areas. Third, the convergence rates are, on average, faster in urban areas and for younger cohorts. Overall, the data suggests that there has been a universal trend toward convergence in education levels of SC/STs toward the levels of non-SC/STs. While this trend is 10The education attainment gaps of the “parents” cohort can change over successive rounds for two reasons. First, as some children become parents in subsequent rounds, the education composition of the parents cohort will clearly change. Second, since 1951 India has introduced a series of literacy initiatives (such the National Literacy Mission) with a special focus on adult literacy. In as much as these programs had a positive effect on adult literacy, the education composition of the parents cohort would change over time due to them as well. 48 common across generations, ages and rural-urban locations, it is sharpest amongst the younger cohorts and in the urban areas. These trend patterns are likely to get sharper in the future as more uneducated parents drop out and more educated children become parents. 3.3.2 Occupation choices We now turn to the occupation choices of the two groups. In order to facilitate ease of presentation, we aggregate the 3-digit occupation codes that individuals report into a one-digit code. This leaves us with ten categories which are then grouped further into three broad occupation categories.11 Our groupings, while subjective, are based on com- bining occupations with similar skill requirements. Thus, Occ 1 comprises white collar administrators, executives, managers, professionals, technical and clerical workers; Occ 2 collects blue collar workers such as sales workers, service workers and production work- ers; while Occ 3 collects farmers, fishermen, loggers, hunters etc.. The groupings also reflect differences in the returns to skills in the Indian economy: Occ 1 is characterised by the highest mean wage in our sample, followed by Occ 2, and Occ 3. Figure 3.3 shows the occupation distribution for the working population in our sample, and the differences between non-SC/STs and SC/STs in this distribution. The top panel of Figure 3.1 refers to children, while the bottom panel refers to parents. There are three features to note. First, there has been a systematic decline in Occ 3 (farming/pastoral activities) between 1983 and 2004-05 across all groups. This decline has been marginally sharper for SC/STs – both children and parents. This reflects the structural transformation at the aggregate level for India wherein there has been a gradual decline in the output and employment share of the agricultural sector. Second, the largest expansion in the employment share has been in Occ 2 which comprises mostly low skill blue collar and service sector jobs. This phenomenon too has been common to both groups. Third, the share of Occ 1 (white collar/high skill) has risen slightly faster for SC/STs than non-SC/STs. This is possibly a sign of increasing mobility for SC/STs and an indicator of possibly faster future improvements. Since SC/STs were over-represented in Occ 3 and under-represented in Occ 1 and 2 in 1983, the trends in occupation shares of the two groups imply that the overall occupation distribution has become more similar over the sample period for both parents and children, i.e., the distributions have been converging. We should point out though that the occupation distribution appears to be converging marginally faster for parents 11See Appendix 5 for more details on the occupation categories. 49 than for children. Figure 3.3: Occupation distribution 0 20 40 60 80 10 0  Non−SC/ST SC/ST 1983 1987−88 1993−94 1999−00 2004−05 1983 1987−88 1993−94 1999−00 2004−05 Distribution of workforce across occ, children Occ1 Occ2 Occ3 0 20 40 60 80 10 0  Non−SC/ST SC/ST 1983 1987−88 1993−94 1999−00 2004−05 1983 1987−88 1993−94 1999−00 2004−05 Distribution of workforce across occ, parents Occ1 Occ2 Occ3 − 15 − 10 − 5 0 5 10 1983 1987−88 1993−94 1999−00 2004−05 Gap in workforce distribution across occ, children Occ1 Occ2 Occ3 − 15 − 10 − 5 0 5 10 1983 1987−88 1993−94 1999−00 2004−05 Gap in workforce distribution across occ, parents Occ1 Occ2 Occ3 (a) (b) Notes: Panel (a) of this figure presents the distribution of workforce of children and parents across three occupation categories for different NSS rounds. The left set of bars on each figure refers to non-SC/STs, while the right set is for SC/STs. Panel (b) presents absolute gaps in the distribution of non-SC/STs relative to SC/STs across three occupation categories. The gaps are also reported for children and parents. Occ 1 collects white collar workers, Occ 2 collects blue collar workers, while Occ 3 refers to farmers and other agricultural workers. Having documented the large changes in the sectoral distribution of occupations as well as differences in educational attainment levels of SC/STs and non-SC/STs, we now look at two additional aspects of the occupation distribution. First, how different are these occupation categories in terms of their educational requirements? Second, are there systematic differences in the educational levels of SC/STs and non-SC/STs even within occupations? Panel (a) of Table 3.3 shows the average educational attainment level of children and parents working in each of the occupations. Clearly, children working in occupation 1 have the highest education level while occupations 2 and 3 employ children with progressively lesser education, on average. 50 Table 3.3: Education attainment levels and gaps by occupations Panel (a). Education attainment levels children parents Occ 1 Occ 2 Occ 3 Occ 1 Occ 2 Occ 3 1983 10.3 5.06 3.16 7.17 2.67 1.20 (0.18) (0.06) (0.04) (0.17) (0.06) (0.03) 1987-88 10.35 5.14 3.52 7.58 2.63 1.41 (0.12) (0.05) (0.04) (0.14) (0.05) (0.02) 1993-94 10.82 5.99 4.28 8.37 3.17 1.75 (0.13) (0.05) (0.04) (0.16) (0.06) (0.03) 1999-00 10.75 6.49 4.99 8.35 3.61 2.18 (0.12) (0.06) (0.05) (0.15) (0.07) (0.04) 2004-05 10.66 6.70 5.52 8.37 3.94 2.51 (0.11) (0.05) (0.05) (0.16) (0.07) (0.05) Panel (b). Education attainment gaps children parents Occ 1 Occ 2 Occ 3 Occ 1 Occ 2 Occ 3 1983 1.20 1.66 1.98 1.72 2.35 2.60 (0.05) (0.07) (0.06) (0.17) (0.17) (0.15) 1987-88 1.19 1.70 1.8 1.46 2.49 2.15 (0.05) (0.06) (0.05) (0.12) (0.16) (0.11) 1993-94 1.17 1.43 1.56 1.55 1.87 2.21 (0.05) (0.04) (0.04) (0.12) (0.11) (0.10) 1999-00 1.08 1.31 1.55 1.37 1.65 2.16 (0.04) (0.03) (0.03) (0.09) (0.10) (0.10) 2004-05 1.13 1.29 1.31 1.27 1.61 1.84 (0.04) (0.03) (0.03) (0.08) (0.08) (0.11) Notes: Panel (a) of this Table presents the average education attainment levels in years for the two generational groups – parents and children – by occupations. Panel (b) summaries the relative education gaps for parents and children computed as a ratio of education attainments levels of non-SC/STs to SC/STs. The reported statistics are obtained for each NSS survey round which is shown in the first column. Occ 1 collects white collar workers, Occ 2 collects blue collar workers, while Occ 3 refers to farmers and other agricultural workers. Standard errors are in parenthesis. Moreover, the average level of education in all occupations has risen throughout the period with the sharpest increase in education levels being in blue collar low-skill jobs (Occ 2) and farming/agricultural jobs (Occ 3). The pattern of average education attainment levels of parents across occupations is similar to that for their children – occupation 1 employs parents with the highest education, with occupations 2 and 3 51 following in that order. Similar to the pattern for children, the average education levels have been rising in all occupations. However, the rise in education levels of parents in occupations 2 and 3 have been much more muted than the corresponding increase for children. Panel (b) of Table 3.3 shows the relative gap in the average education levels of non- SC/STs and SC/STs within the same occupation. Two features are noteworthy here. First, SC/ST children are less educated than non-SC/STs of the same cohort even within the same occupation. Second, gaps in education attainment levels are lowest in the high-skill white collar occupations. Third, education gaps amongst children within the same occupations have declined over time in all occupations. The trends in gaps for parents are broadly similar to those we uncovered for children with one key difference. The education gaps between non-SC/ST and SC/ST parents are much larger in white collar, high-skill occupations (occupation 1). This is in sharp contrast to the pattern for children where the difference are the smallest in these occupations. Finally, we study the more disaggregated age cohorts and document the evolution of education attainments within each occupation. Panel (a) of Figure 3.4 shows the relative education gap between non-SC/STs and SC/STs working in occupation 1 for different age cohorts. Similarly, panel (b) summaries the corresponding gap for those employed in occupation 2; and panel (c) for those working in occupation 3. Figure 3.4: Education gaps by age cohorts and occupations 1 1. 2 1. 4 1. 6 1. 8 2 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 occupation 1 1 1. 5 2 2. 5 3 3. 5 4 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 occupation 2 1 1. 5 2 2. 5 3 3. 5 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 occupation 3 Notes: The figures show the evolution of the relative education gap between non-SC/STs and SC/STs over time for different age and occupation groups. Occ 1 collects white collar workers, Occ 2 collects blue collar workers, while Occ 3 refers to farmers and other agricultural workers. These results confirm our earlier findings: education attainment levels are converging between non-SC/STs and SC/STs. They are converging faster for younger age cohorts and for higher-skill occupations 1 and 2. Education gaps in occupation 3 have declined for the youngest age cohorts while remaining relatively unchanged or even increasing 52 slightly for the older age cohorts.12 3.3.3 Industry choices Next, we look at the industry of employment choices of households. In order to facilitate the presentation, we aggregate the 4-digit industry code that individuals report into a one-digit code. This gives us seventeen categories. We then group these seventeen cate- gories into three broad industry categories: Ind 1, Ind 2 and Ind 3. Ind 1 comprises the Agricultural sector, Ind 2 collects tradable industries while Ind 3 comprises non-tradable industries. Our grouping reflects the tradition of classifying industries into tradables and non-tradables. Incorporating Agriculture as a distinct category is intended to take account of the traditional reliance of the Indian economy on agriculture. See Appendix 5 for more details on the industry grouping. Figure 3.5 reports the industry distribution of parents and children, and the absolute gaps in this distribution. Consistent with the results for occupation choice, SC/STs were and remain more likely to be employed in agriculture and other farming activities (Ind 1) than non-SC/STs, however the gap has somewhat narrowed in the last ten years of our sample. Interestingly, SC/STs are also more likely to work in non-tradable industries (Ind 3) relative to non-SC/STs - with the gap becoming more pronounced over the past twenty years for children and emerging in 2004-05 round for parents. In contrast, a larger share of non-SC/STs population is employed in tradable industries (Ind 2), but the data suggests that SC/STs are gradually moving into those industries as well. 12One other interesting feature of our data is that the dispersion in education gaps across age cohorts is the highest in occupation 1 with lower dispersions in occupations 2 and 3. This probably reflects the heterogeneity of skills underlying the occupation groups we constructed. Occ 1 combines a variety of high-skill occupations that can lead to more heterogeneity in education gaps. Such skill heterogeneity is lower in occupations 2 and 3. 53 Figure 3.5: Industry distribution 0 20 40 60 80 10 0  Non−SC/ST SC/ST 1983 1987−88 1993−94 1999−00 2004−05 1983 1987−88 1993−94 1999−00 2004−05 Distribution of workforce across ind, children Ind1 Ind2 Ind3 0 20 40 60 80 10 0  Non−SC/ST SC/ST 1983 1987−88 1993−94 1999−00 2004−05 1983 1987−88 1993−94 1999−00 2004−05 Distribution of workforce across ind, parents Ind1 Ind2 Ind3 − 20 − 10 0 10 20 1983 1987−88 1993−94 1999−00 2004−05 Gap in workforce distribution across ind, children Ind1 Ind2 Ind3 − 20 − 10 0 10 20 1983 1987−88 1993−94 1999−00 2004−05 Gap in workforce distribution across ind, parents Ind1 Ind2 Ind3 (a) (b) Notes: Panel (a) of this Figure presents the distribution of workforce of children and parents across three industry categories for different NSS rounds. The left set of bars on each Figure refers to non-SC/STs, while the right set is for SC/STs. Panel (b) presents absolute gaps in the distribution of non-SC/STs relative to SC/STs across three industry categories. The gaps are also reported for children and parents. Ind 1 refers to Agriculture and other farming activities, Ind 2 collects tradable industries, while Ind 3 refers to nontradable industries. 3.3.4 Wages The third issue of interest is the evolution of wages of SC/STs and their non-SC/ST cohorts. We are particularly interested in determining whether the rising educational attainment rates and changing occupation distribution of SC/STs towards relatively higher skilled occupations have also resulted in a change in the wage gap relative to non-SC/STs. Before describing our results on wages it is important to reiterate that the sample size for the wage data is, on average, a third of the sample size for the education and occupation distribution data due to a large number of households with missing wage observations. The missing wage observations are primarily due to the large segment of 54 the rural population who identify themselves as being self-employed and correspondingly do not report any wage data. Across the rounds, on average, about 65 percent of the sample are self-employed with 76 percent of them residing in rural areas. The missing wage data raises a natural concern about sample selection. In particular, if non-SC/ST rural households are more likely to be land-owning and hence self-employed, then the wage data (particularly for rural households) would be skewed towards landless SC/ST households. The problem would be compounded by the fact that the wage earning non-SC/ST households may also be the most worse off amongst the non-SC/STs. In this event we would be biasing our results toward finding low wage gaps between the two groups. We examined this issue in two ways. First, on average, 21 percent of the self- employed belong to SC/ST households. This is comparable to the 24 percent share of SC/STs in our working sample. Clearly, SC/STs are not disproportionately under- represented amongst the self-employed. Second, to assess the seriousness of the potential sample selection problem, we computed the per capita household consumption expendi- ture of non-SC/STs relative to SC/STs for self-employed households and wage earning households separately. Averaged across rounds, the ratio was 1.24 for both. Hence, self-employed households do not appear to be distinctly different from wage earning households. Based on these two findings, we feel that the sample selection issues raised by the missing wage observations are not too serious and that the patterns of inter-group welfare dynamics indicated by the wage data are likely to generalise to the self-employed as well. It is also important to note one important oddity in the 1987-88 data generated by the 43th round. In particular, the number of observations for wages in this round falls precipitously to about half the level of the other rounds. This occurs due to a very large and disproportionate decline in the rural wage observations for this round. We are not sure as to the reasons for this sudden increase in the number of missing observations in the 43th round. For the sake of completeness though, we report the results for all rounds. However, the results for the 43th round should be treated with caution on account of the missing rural wage observations. 55 Figure 3.6: Wage density for non-SC/STs and SC/STs 0 . 2 . 4 . 6 . 8 1 de ns ity 0 1 2 3 4 5 log wage (real) non−SCST − 1983 SCST − 1983 non−SCST − 2004−05 SCST − 2004−05 Notes: This figure shows the estimated kernel density of log real wages for non-SC/STs and SC/STs over 1983 and 2004-05 NSS rounds. It is instructive to start our analysis of the wage data by presenting the distribution of wages for the first and last rounds of our sample, i.e., for 1983 and for 2004-05. Figure 3.6 plots the kernel densities of the wage distribution for SC/STs and non- SC/STs separately for both these rounds. Two features emerge clearly from the figure. First, for both groups the wage distribution has shifted sharply to the right. This is to be expected as the period 1983-2005 coincides with the rapid takeoff of the Indian economy. Second, the density functions for the two groups have come much closer together in 2004-05 relative to 1983.13 We can examine the changes in wage inequality more closely by looking at the differences in log wages between non-SC/STs and SC/STs for different percentiles of their wage distributions. Panel (a) of Figure 3.7 shows this for two survey rounds: 13We should note though that a formal Kolmogorov-Smirnov test of the equality of SC/ST and non- SC/ST wage distributions rejects the null hypothesis of equality both for 1983 and 2004-05. Moreover, the test also rejects the null hypothesis of the SC/ST distribution in 1983 being the same as the SC/ST distribution in 2004-05. This conclusion carries over to a comparison of the non-SC/ST wage distributions in these two rounds as well. 56 1983 and 2004-05. Several features are worth pointing out from the panel (a) of Figure 3.7. First, the Figure shows first-order stochastic dominance of the non-SC/ST wage distribution relative to the SC/ST wage distribution since wages are almost uniformly higher for non-SC/STs than for SC/STs for every percentile. However, the degree of the stochastic dominance has declined over time as the line for 2005-05 is much closer to zero for almost all percentiles relative to the earlier round. Second, both lines slope up and to the right, indicating that the wage distribution of non-SC/STs is more unequal that the wage distribution of SC/STs. Figure 3.7: Differences in percentiles for non-SC/STs and SC/STs for log wages and log consumption − . 1 0 . 1 . 2 . 3 . 4 . 5 . 6 . 7 ln w ag e(N on SC /S T) −ln wa ge (S C/ ST ) 0 10 20 30 40 50 60 70 80 90 100 percentile 1983 2004−05 − . 1 0 . 1 . 2 . 3 . 4 . 5 . 6 . 7 ln m pc e(N on SC /S T) −ln mp ce (S C/ ST ) 0 10 20 30 40 50 60 70 80 90 100 percentile 1983 2004−05 Notes: Figure (a) shows the difference in percentiles of log-wages between non-SC/STs and SC/STs plotted against the percentile, while Figure (b) does the same for real consumption expenditures. The plots are for 1983 and 2004-05 NSS rounds. The line that slopes upward and to the right indicates more unequal distribution for non-SC/STs compared to SC/STs. The lines that are above the horizontal axis indicate stochastic dominance in non-SC/STs wage distribution. An upward sloping line indicates that the difference in wages of the two groups is smaller for lower percentiles than for higher percentiles. But this implies that higher percentile non-SC/STs must earn not only more than higher percentile SC/STs, but their wage mark-up relative to lower percentile non-SC/STs must also be greater than the wage mark-up of higher percentile SC/STs relative to their lower percentile coun- terparts. Hence, an upward sloping line indicates a more unequal wage distribution for non-SC/STs than SC/STs. The flattening out of the lines over time indicates a decrease in the wage inequality of the two distributions even though the sharp positive slope towards the right tail indicates continued wage inequality at the top-end of the income distribution. Overall, the plot confirms our earlier finding of convergence in the two distributions over time as the line for 2004-05 round is well below the line for 1983 57 round. As we mentioned earlier, the sample of individuals for whom wages are available is significantly smaller than our working sample. We therefore, verify the robustness of the wage inequality results uncovered above using the data on consumption expenditures which are available for a larger sample than the wage sample. We convert consumption expenditures into real terms using the same deflators that we used for wages and com- pute the differences in percentiles of consumption distributions between non-SC/STs and SC/STs in the same way as we did for wages. Panel (b) of Figure 3.7 shows the results for 1983 and 2004-05 survey rounds. The consumption results confirm our find- ings for wages. The plots indicate stochastic dominance of the non-SC/ST consumption distribution relative to the SC/ST consumption distribution, but show a significant de- cline in consumption gap between non-SC/STs and SC/STs over time for the most part of the distribution, except at the very highest end. The wage distributions plotted in Figures 3.6 and 3.7 appear to indicate a decline in wage inequality between SC/STs and non-SC/STs between 1983 and 2004-05. We now examine this impression more closely by contrasting the wage evolution of SC/STs with non-SC/STs over finer sub-groups of age and generation cohorts as well as for all the survey rounds under study. We start with the wage evolution of the generational cohorts of parents and children across the survey rounds. Table 3.4 shows the daily wage earned by working children and parents of non-SC/ST households relative to their SC/ST cohorts over the period 1983 to 2004-05. Overall, the wage premium of non-SC/STs has declined from 42 percent to 23 percent during this period. The Table reveals a contrast between the children and parents in terms of the evolution of the wage gap between SC/STs and non-SC/STs during this period. There has been a clear convergence of wages between children of these two groups. The wage premium of non-SC/ST children has secularly declined from around 34 percent in 1983 to 14 percent by 2004-05. For parents too the non-SC/ST wage premium has fallen from 51 percent in 1983 to 31 percent in 2004-05. The trends we uncover are even more dramatic if we look at wage gaps computed using median wages. In particular, we find that the median wage premium of non- SC/STs relative to SC/STs has declined from 17 percent in 1983 to 3 percent in 2004- 05. This decrease is especially pronounced for children for whom the relative wage premium of non-SC/STs relative to SC/STs essentially disappears during this period – falling from 14 percent in 1983 to approximately zero in 2004-05. For parents too the premium fell from 25 percent in 1983 to 10 percent in 2004-05. Clearly, both mean and 58 median wages have been converging across the two groups during this period. Table 3.4: Wage gaps overall children parents mean median mean median mean median 1983 1.38 1.15 1.28 1.13 1.48 1.25 (0.04) (0.00) (0.06) (0.02) (0.05) (0.03) 1987-88 1.48 1.24 1.48 1.10 1.49 1.40 (0.09) (0.00) (0.18) (0.03) (0.07) (0.03) 1993-94 1.30 1.07 1.16 1.04 1.45 1.18 (0.03) (0.00) (0.03) (0.01) (0.05) (0.02) 1999-00 1.30 1.07 1.18 1.02 1.42 1.15 (0.03) (0.00) (0.04) (0.01) (0.06) (0.02) 2004-05 1.23 1.03 1.14 0.99 1.31 1.10 (0.03) (0.00) (0.03) (0.01) (0.05) (0.02) Notes: This Table presents the relative mean and median wage gaps for our overall bench- mark sample (columns “overall”) and separately for the two generational groups – children (columns “children”) and parents (columns “parents”). The mean gaps are obtained as the ratios of average real wages of non-SC/STs to SC/STs; while median wage gaps are computed as the ratios of median real wages of the two groups. The reported statistics are obtained for each NSS survey round which is shown in the first column. Next, we switch from the aggregated generational cohorts to the more disaggregated age-cohorts. Our principal interest here is to determine whether the relative wage gap behaviour at the aggregate generational level is masking significant variation across different age cohorts. Figure 3.8 plots the wage gaps for our five age cohorts for the different survey rounds. Panel (a) reveals a general pattern of wage convergence across the cohorts with the younger cohorts, on average, closer to parity than the older ones. Do these overall wage gaps between non-SC/STs and SC/STs reflect significant differ- ences between rural and urban areas? Panels (b) and (c) of Figure 3.8 shows that the evidence is mixed on this. Relative wages have tended to converge for younger cohorts in both sectors but have often widened for the older ones. Thus, the wage gaps have widened for the 46-55 age group in rural areas and for the 56-65 age group in urban areas. Overall, we view this evidence to be along the same lines as the evidence on education, albeit more volatile due to the smaller sample size. 59 Figure 3.8: Wage gaps by age cohorts . 8 1 1. 2 1. 4 1. 6 1. 8 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 overall . 8 1 1. 2 1. 4 1. 6 1. 8 1983 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 rural . 8 1 1. 2 1. 4 1. 6 1. 8 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 urban Notes: The figures show the evolution of the relative wage gap between non-SC/STs and SC/STs over time for different age groups. Panel (a) presents the results for the overall sample, while panels (b) and (c) report the results for rural and urban households separately. We also examine the behaviour of relative wages of non-SC/STs relative to SC/STs by age cohorts across different occupations. Figure 3.9 presents the results. Panel (a) is for occupation 1, panel (b) is for occupation 2, while panel (c) is for occupation 3. It can be seen that the relative wage premia in occupations 1 and 2 have declined across most age cohorts. In agricultural jobs (occupation 3) the convergence is a bit muted, but the wage gaps there are very small to begin with. As with educations gaps, we see that wage gaps are the most spread out in occupation 1, and to a smaller extent in occupations 2 and 3. Overall, the data suggests that SC/ST wages have been converging toward non-SC/ST levels, and this trend is most pronounced for higher-skill occupations. The evolution of the wage gaps between SC/STs and non-SC/STs provides an in- teresting counterpoint to the racial wage gaps that are typically reported in the USA. Thus, during the period 1980-2006 the median wage of black males relative to white male workers has fluctuated between 70 and 80 percent with an average of around 75 percent. During the same period the median wage of Hispanic men relative to white men has declined from 71 percent to under 60 percent.14 In contrast, our computations above imply that the median wage of SC/STs relative to non-SC/STs has increased from 80 percent in 1983 to 95 percent in 2004-05.15 14These numbers are from US Current Population Survey. 15For ease of comparison with the typical wage gap numbers reported for the USA, the SC/ST wage gaps reported here are the inverses of the non-SC/ST to SC/ST relative wage gaps we reported above. 60 Figure 3.9: Wage gaps by age cohorts and occupations . 6 . 8 1 1. 2 1. 4 1. 6 1. 8 2 2. 2 2. 4 2. 6 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 occupation 1 . 6 . 8 1 1. 2 1. 4 1. 6 1. 8 2 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 occupation 2 . 6 . 8 1 1. 2 1. 4 1. 6 1. 8 2 1983 1987−88 1993−94 1999−00 2004−05  16−25 26−35 36−45 46−55 56−65 occupation 3 Notes: The figures show the evolution of the relative wage gap between non-SC/STs and SC/STs over time for different age and occupation groups. Occ 1 collects white collar workers, Occ 2 collects blue collar workers, while Occ 3 refers to farmers and other agricultural workers. Amongst the younger cohorts the wage catch-up has been even faster with the relative median wages of SC/ST children having risen from 88 percent to 101 percent during this period. Clearly, the rate of wage convergence for SC/STs since 1983 has been quite striking both at an absolute level as well as in comparison to historically disadvantaged minority groups in more developed countries like the USA. Conditional wages The trends we documented above suggest that the wage gap between SC/STs and non- SC/STs has been declining over the past 22 years. We now examine this impression more closely using formal statistical tests. In particular, for each survey round we estimate a linear log wage regression on the following characteristics: individual age and age squared, dummies for his education category, SC/ST dummy, Muslim dummy, region and occupation specific dummies. We control for differences in reservation policies across states by including state-level SC/ST reservation quotas (quota SC/ST ). The introduction of reservations for SC/STs in public sector employment and in higher education institutions was a key policy initiative in India. The reservations were provided in proportion to the population shares of SCs and STs.16 We also include a Muslim dummy in our regression specification. This is intended as a control for the fact that Muslims, on average, have done poorly 16State-level reservations can change over time due to changes in SC/ST population shares. In 1991 the Indian government extended the reservation policy to include other backward castes (OBCs). In our analysis we focus only on the group of SC/STs while OBCs are included in the non-SC/ST reference group. If reservations increased OBC relative wages then our results potentially understate the true degree of convergence between SC/STs and non-SC/STs (excluding OBCs), especially since the extension of reservations to OBCs in 1991. 61 in modern India (post independence in 1947). If we do not control for a Muslim fixed factor explicitly, then part of the decline in wage and economic inequality that we find in the data may be attributed to the poor performance of Muslims who would be assigned into non-SC/ST group. We control for regional differences by grouping states into five regions – North, South, East, West, Central and North-East – and include region dummies in the regression specification.17 In combination with the state-level reservation policy, this allows us to decompose state-level differences into those attributable to reservations policy, and those due to other time-invariant factors that are common to all states within a given region. The identifying assumption behind this strategy is that the states within a region are broadly similar but differ in terms of the reservation quota they implement. Table 3.5 reports the key results. We find that the coefficient on the SC/ST dummy variable is negative and significant throughout except for the 1993-94 and 1999-2000 rounds. The negative estimates for the SC/ST dummy indicate that the conditional wages of SC/STs were significantly lower than similarly endowed non-Muslim non- SC/STs. Interestingly, the size of this negative SC/ST effect declined over the first four rounds – in fact becoming insignificant in 1993-94 and 1999-2000 before returning to its initial 1983 level in the last round. Our results also suggest that reservations have been associated with lower average wages for all groups.18 Table 3.5 also shows a significant positive coefficient on the rural dummy for all rounds except for the first which indicates a positive wage effect of living in rural areas after one accounts for all our controls. We find this to be an interesting feature of the Indian experience during this period. From the perspective of this study though, it is worth noting that this differential rural wage effect is common to both SC/STs and non-SC/STs.19 17This grouping reflects similarities across states along their geographic characteristics, and charac- teristics that are shared based on proximity. 18Prakash (2009) focuses on the role of reservations for India’s lower castes. He finds broadly in- significant effects of reservations on wages of all except the very poorly educated SC/STs. 19In particular, when we include an interaction term between rural dummy and SC/STs dummy, we find the coefficient on it to be insignificant for all survey rounds. These results are available upon request. 62 Table 3.5: Conditional wage regressions 1983 1987-88 1993-94 1999-00 2004-05 (i) (ii) (iii) (iv) (v) age 0.0350*** 0.0732*** 0.0358*** 0.0327*** 0.0289*** (0.0034) (0.0067) (0.0046) (0.0031) (0.0035) age sqr -0.0004*** -0.0007*** -0.0004*** -0.0003*** -0.0002*** (0.0000) (0.0001) (0.0001) (0.0000) (0.0000) 1-SC/ST, 0-non SC/ST -0.0522*** -0.0664** -0.0303 -0.0013 -0.0586*** (0.0162) (0.0336) (0.0208) (0.0144) (0.0161) 1-rural, 0-urban 0.0473** 0.2659*** 0.1670*** 0.1930*** 0.2497*** (0.0235) (0.0677) (0.0274) (0.0223) (0.0234) edu 2 dummy 0.0855*** 0.1586*** 0.0861*** 0.1025*** 0.1296*** (0.0192) (0.0428) (0.0273) (0.0184) (0.0209) edu 3 dummy 0.1825*** 0.1765*** 0.1270*** 0.1468*** 0.1106*** (0.0237) (0.0372) (0.0296) (0.0206) (0.0217) edu 4 dummy 0.2424*** 0.2921*** 0.1894*** 0.1492*** 0.1251*** (0.0292) (0.0459) (0.0335) (0.0210) (0.0253) edu 5 dummy 0.5553*** 0.6409*** 0.3207*** 0.3715*** 0.3527*** (0.0388) (0.0479) (0.0376) (0.0277) (0.0311) 1-muslim, 0-other -0.0041 -0.0302 -0.0560 0.0159 -0.0844*** (0.0235) (0.0344) (0.0365) (0.0212) (0.0228) quota SC/ST -0.0145*** -0.0118*** -0.0032 -0.0140*** -0.0094*** (0.0013) (0.0026) (0.0020) (0.0013) (0.0012) R-sqr 0.359 0.376 0.201 0.356 0.320 N 8313 3807 8287 9354 8453 Notes: This table presents estimation results from a regression of log real wages on a set of individual-level, household-level and aggregate control variables for five NSS survey rounds ((i)-(iv)). Refer to the text for model specification details. Standard errors are in parenthesis. * p-value≤0.10, ** p-value≤0.05, *** p-value≤0.01. While a more detailed investigation of this issue is beyond the scope of this chap- ter, it is also worth noting that both the rural-urban wage differential and rural-urban education differential declined during the period under study. However, the wage differ- ential declined at a faster rate than the education differential. Put differently, relative to the corresponding urban levels, wages in rural areas grew faster than education levels in rural areas. What factors can account for this feature of the data is an intriguing issue which we intend to study in greater detail in future work. As we saw earlier, relative wage gaps have been declining and this decline has co- incided with a decline in the gaps in education attainment levels between SC/STs and non-SC/STs. So, how much of the wage gap between the two groups arises due to differences in education? We answer this question by using the Oaxaca-Blinder decom- 63 positions. We employ a two-fold Oaxaca-Blinder procedure which involves running wage re- gressions separately for the two groups on a list of controls including education levels. One then decomposes the wage gaps into the part coming from the different coefficients on the controls for the two groups, and the part due to differences in endowments be- tween the two groups. To obtain the reference coefficients we use a pooled approach which allows for a group membership indicator (as in Fortin, 2006). Our controls are the same as in the regression specification above. Table 3.6 reports the results for the overall sample. Table 3.6: Oaxaca-Blinder decomposition non-SC/ST SC/ST difference explained unexplained fraction to edu (i) (ii) (iii) (iv) (v) (vi) 1983 2.13 1.93 0.20 0.15 0.05 0.59 (0.01) (0.01) (0.02) (0.01) (0.02) 1987-88 2.46 2.21 0.25 0.18 0.07 0.77 (0.02) (0.03) (0.04) (0.03) (0.03) 1993-94 2.32 2.15 0.17 0.14 0.03 0.37 (0.01) (0.02) (0.02) (0.01) (0.02) 1999-00 2.54 2.38 0.16 0.16 0.00 0.40 (0.01) (0.01) (0.02) (0.01) (0.01) 2004-05 2.60 2.49 0.10 0.05 0.06 0.93 (0.01) (0.01) (0.02) (0.01) (0.02) Notes: This Table presents a two-fold Oaxaca-Blinder decomposition of the log-wage gap between non-SC/STs and SC/STs for the five NSS survey rounds, as identified in the first column. Columns (i) ‘non-SC/ST’, (ii) ‘SC/ST’ and (iii) ‘difference’ present the average real log-wages for non-SC/STs, SC/STs and the gap between them, respectively. Columns (iv) ‘explained’ and (v) ‘unexplained’ refer to the size of the wage gap attributable to differences in endowments between non-SC/STs and SC/STs, and to the differences in the returns to those endowments, respectively. Column (vi) ‘fraction to edu’ reports the share of the explained wage gap coming from education attainment differences between the two groups. Standard errors are in parenthesis. Columns (i) and (ii) report average log wages of non-SC/STs and SC/STs, respec- tively; while column (iii) reports the wage gap for the two groups over different survey rounds.20 Column (iv) attributes a fraction of this gap to group differences in measured 20Wage gap reported in Oaxaca-Blinder decomposition is computed as a difference between average log wages of non-SC/STs and SC/STs. The relative wage gaps we reported earlier were obtained as the ratios of average wages (in levels) of non-SC/STs to SC/STs. As a result, the magnitudes of the gaps from these two approaches are different, but the trends are the same. 64 endowments, while column (v) reports the size of the gap usually attributable to dis- crimination or potentially to group differences in unobserved characteristics. Finally, the last column of Table 3.6 reports the fraction of the explained log wage difference that is accounted for by differences in education endowments alone. The column shows that differences in education accounted for 56 percent of the explained wage gap in 1983 which increased to 95 percent in 2004-05. The detailed results from Oaxaca-Blinder decomposition and for the regressions that will follow are available upon request. These results, in conjunction with the facts that both wage and education differences have been declining over time, suggest that the major part of the decline in the wage differences between SC/STs and non-SC/STs between 1983 and 2004-05 is due to a decline in the education differences between them. 3.3.5 Sample and robustness A key restriction underlying our working sample is that we have only considered joint households consisting of a head of household and at least one child or grandchild. This restriction allowed us to construct the groups of Parents and Children which will allow us to study the time trends in inter-generational mobility patterns within SC/ST and non-SC/ST households (see below). The imposed restriction did however reduce the sample size by dropping brothers/cousins of the head of the household as well as their children. Moreover, the restriction also dropped households with only one generation of full-time working males. Does this sample restriction matter for our results? To check this we compared the relative education and wage gaps from the extended sample (without the joint-household restrictions) with the results reported above using the working sample (which reflects the restrictions). Panel (a) of Figure 3.10 shows the average education attainment levels of non- SC/STs relative to SC/STs for both the extended sample and the working sample. Similarly, Panel (b) of Figure 3.10 plots the median wages of non-SC/STs relative to median SC/ST wages for the two samples. Clearly, the same pattern of convergence emerges in both samples. Both figures show, however, that the non-SC/ST to SC/ST gaps are larger for all years in the extended sample. This is due to the fact that the larger sample includes a number of older individuals (like brothers of the head of household living in a joint household) who get dropped in the working sample. Recall that the measured gaps in both education and wages are larger for older cohorts. We conclude from this exercise that our results on education and wage convergence are not an artifact of the sample selection imposed by our joint household condition. 65 Figure 3.10: Robustness to sample choice 1 1. 2 1. 4 1. 6 1. 8 2 2. 2 2. 4 1983 1987−88 1993−94 1999−00 2004−05  working sample extended sample Education attainment gap 1 1. 1 1. 2 1. 3 1. 4 1. 5 1983 1987−88 1993−94 1999−00 2004−05  working sample extended sample Wage gap: Median Notes: Panel (a) depicts the ratio of the mean education attainment level of non-SC/STs to SC/STs for the working and extended samples. Panel (b) presents the ratio of the median wages of non-SC/STs to SC/STs for the same two samples. We also checked for the robustness across samples of our regression estimates from the conditional wage regression by running the wage regressions on the extended sample as well. We found that the coefficient on the SC/ST dummy declines over the five rounds from -0.07 to -0.03. Hence, the trend in the extended sample is the same as in the working sample. However, as in the unconditional gaps plotted above, the greater presence of older individuals with larger wage gaps implies that the overall estimate for the SC/ST dummy is, on average, slightly larger in the extended sample relative to that in the working sample. 3.4 Intergenerational mobility We now turn to the key question that we started with: how have the patterns of intergenerational mobility in India changed between 1983 and 2004-05? Are children changing occupation, industry, education and income status relative to their parents more frequently than before? Our primary interest is in studying how the occupation and industry choices, education attainment levels and wages of children compare with the corresponding levels for their parents. We shall look at each of these in turn. In the foregoing analysis we shall define the intergenerational education/ occupation/ industry switch as a binary variable that takes a value of one if the child’s or grand- child’s education level/ occupation/ industry of employment is different from his par- ent’s (who is the head of the household) education achievements/ occupation/ industry of employment; and zero otherwise. We label education switch variable as switch-edu; 66 occupation switch variable as switch-occ; and industry switch variable as switch-ind. We also distinguish education improvement, which is another binary variable equal to one if the child’s education is higher than that of his parent and zero otherwise, from education reduction which is a binary variable that takes a value of one if the child’s education is below his parent’s education and zero otherwise. 3.4.1 Education mobility We begin by analysing intergenerational education switches. Our main interest is in determining the degree to which children are changing their education levels relative to their parents and by how much. We are also interested in determining whether or not the switches reflect increases in educational attainment by the children. To obtain average probabilities of education switches we posit the following probit model: Pi ≡ Pr(yi = 1|xi) = E (yi|xi) = ψ(xiβ), where ψ(xiβ) = Φ(xiβ), with Φ(.) representing the cumulative standard normal distri- bution function, yi is a binary variable for education switch as defined above (switch- edu), and xi is a vector of controls. We allow the education switch for individual i to depend on his individual characteristics, such as age, age squared, belonging to an SC/ST group (SC/ST ), and religion (muslim); household-level characteristics, such as household size (hh size), his rural location (rural); and state-level characteristics, such as state-level reservation quota for SC/STs, and region-specific fixed effects. Thus, xiβ = β0 + β1agei + β2age 2 i + β3SC/STi + β4muslimi +β5rurali + β6hh sizei + β7quota scstj + 6∑ j=1 αjregion dummyj. (3.1) We estimate the model for each survey round separately and use it to obtain fitted values for each individual. These fitted values are used to compute the average prob- ability of intergenerational education switch. We compute these probabilities for the overall sample as well as for SC/STs and non-SC/STs separately.21,22 Panel (a) of Figure 3.11 depicts the computed probabilities of intergenerational 21The detailed regression results for this Section are available in supplemental tables from http://faculty.arts.ubc.ca/vhnatkovska/research.htm. 22We choose to proceed with the regression approach as we are also interested in the effect of caste on the probability of switching conditional on other controls. As we show in the Supplemental Tables, the marginal effects of caste on these probabilities are always significant. 67 switches in education attainment together with the ± 2 std error confidence bands (dashed lines).23 The remarkable feature highlighted by the Figure is that the switch probabilities of the two groups have converged at 67 percent by the end of our sample period in 2004-05. This is particularly impressive once one notes that in 1983, the prob- ability of an intergenerational education switch for SC/ST households was a meagre 42 percent relative to the 57 percent corresponding probability of non-SC/ST households. A related question is about the degree or size of the change in education levels. In particular, amongst the children who switch education levels relative to their parent, how large is the change? How has this evolved over our sample period? Panel (b) of Figure 3.11 reveals that the average size of the switch has been increasing over time for both groups. Moreover, by the end of our sample, the switch sizes for the two groups not only converged, but SC/STs were in fact switching education levels by more than non-SC/STs. This again is noteworthy since the average size of a switch for SC/STs was significantly lower at 0.6 in 1983 relative to 0.84 for the non-SC/ST households. Note that positive numbers for the size of the switch indicate improvements in education categories. Figure 3.11: Intergenerational education switches . 4 . 45 . 5 . 55 . 6 . 65 . 7 1983 1987−88 1993−94 1999−00 2004−05  overall non−SC/ST SC/ST Avg prob of edu switch . 5 . 6 . 7 . 8 . 9 1 1. 1 1. 2 1983 1987−88 1993−94 1999−00 2004−05  overall non−SC/ST SC/ST Avg size of edu switches Notes: Panel (a) of this figure presents the average predicted probability of intergenerational education switch, while panel (b) reports the average size of the intergenerational education switches for our overall sample, for SC/STs and non-SC/STs. The numbers are reported for the five NSS survey rounds. Dotted lines are ±2 std error bands. We also find that most of the intergenerational education switches are in fact in- creases in educational attainment levels. The estimated probability of an SC/ST child increasing his level of education attainment relative to the parent was just 36 percent 23Confidence bands around the probability of education switch are very narrow and do not appear on the graph. 68 in 1983 but rose sharply to 59 percent by 2004-05. The corresponding probabilities of an increase in education attainment for a non-SC/ST child were 49 percent and 58 percent. The probability of an education reduction is around 9 percent for non-SC/STs and 7 percent for SC/STs. Both these probabilities have remained stable over the sam- ple period. Hence, a majority of the increase in the education switch probability for SC/STs relative to the non-SC/STs is accounted for by an increase in the probability of an improvement in the education attainment level. Detailed summary of these results is available in the Appendix Table A.3. 3.4.2 Occupation mobility We now turn to intergenerational occupation switches. The conditional probability of an occupation switch is obtained in a similar manner to the education switch probabilities. Now, yi is a binary variable for occupation switch as defined above (switch-occ) while xi is a vector of controls: xiβ = β0 + β1agei + β2age 2 i + β3SC/STi + β4muslimi +β5rurali + β6hh sizei + β7quota scstj + 4∑ j=1 θjedu dumj + 6∑ j=1 αjregion dummyj + 9∑ j=1 γjoccup dummyj. (3.2) In our model, the occupation switch for individual i depends on three sets of controls. The first set includes individual characteristics such as age, age squared, belonging to an SC/ST group (SCST ), and religion (muslim). Second, we control for household-level characteristics such as household size (hh sizei), and his rural location (rurali). Third, we allow for occupation-specific fixed effects, region-level fixed effects, and state-level SC/ST reservation quotas. The model is estimated for each sample round separately and then used to obtain fitted values for each individual. These fitted values provide us with estimates of the probability of occupation switches in each round. We compute this measure of inter- generational occupational mobility for the overall sample as well as for SC/STs and non-SC/STs separately. Figure 3.12 depicts the computed probabilities of occupation switches at the three- digit level (dotted lines plot the ± 2 std error confidence bands). As the Figure shows, the overall probability of an occupation switch by the next generation relative to the 69 household-head has steadily increased from 32 percent in 1983 to 41 percent in 2004-05. This increase has been mirrored in the two sub-groups with the switch probabilities rising for both. For non-SC/STs the switch probability has risen from 33 to 42 percent while for SC/STs it has gone from 30 to 39 percent. Crucially, there is no trend towards convergence of these probabilities across the two groups which indicates that differences in intergenerational mobility between them has not changed over this period.24 Figure 3.12: Intergenerational occupation switches . 28 . 33 . 38 . 43 1983 1987−88 1993−94 1999−00 2004−05  overall non−SC/ST SC/ST Avg prob of occ switch − 3 digit Notes: This figure presents the average predicted probability of intergenerational occupation switch for our overall sample, for SC/STs and non-SC/STs. The numbers are reported for the five NSS survey rounds. Dotted lines are ±2 std error bands. Occupation transition matrix While the overall probability of switches indicates the degree of mobility across occu- pations, we are also interested in determining the pattern of movements within occu- pations: children who are switching are most likely to have parents working in which sector? Which sectors are absorbing most of the intergenerational switchers? Have these trends varied over time? Are there any differences between SC/STs and non-SC/STs in these patterns? To address these issues, we compute the transition probabilities across occupations. Thus, for each NSS round we compute pij where i denotes the occupation of the house- hold head and j denotes the occupation of the child. Thus, pij is the probability of a household head working in occupation i having a child working in occupation j. Clearly, 24We also estimated the occupation switch probabilities at the one-digit and two-digit levels and found that the patterns are similar to the three-digit probabilities. The main difference is that the probability of an occupation switch is universally lower at the two-digit and more so at the one-digit level. The results for the one- and two-digit occupation categories are available upon request. 70 high pii would reflect relatively little intergenerational occupational mobility while large pij where i 6= j would indicate high mobility. Table 3.7: Intergenerational occupation transition probabilities (a). Average mobility in the 1983 round Non-SC/ST To SC/ST To From Occ 1 Occ 2 Occ 3 size From Occ 1 Occ 2 Occ 3 size Occ 1 0.49 0.33 0.18 0.06 Occ 1 0.29 0.40 0.31 0.03 (0.02) (0.01) (0.01) (0.00) (0.05) (0.06) (0.05) (0.00) Occ 2 0.06 0.82 0.12 0.26 Occ 2 0.04 0.77 0.19 0.20 (0.00) (0.01) (0.01) (0.00) (0.01) (0.01) (0.01) (0.01) Occ 3 0.03 0.10 0.86 0.67 Occ 3 0.02 0.09 0.90 0.78 (0.00) (0.00) (0.01) (0.00) (0.00) (0.01) (0.01) (0.01) (b). Average mobility in the 2004-05 round Non-SC/ST To SC/ST To From Occ 1 Occ 2 Occ 3 size From Occ 1 Occ 2 Occ 3 size Occ 1 0.48 0.38 0.14 0.10 Occ 1 0.35 0.45 0.20 0.05 (0.01) (0.01) (0.01) (0.00) (0.03) (0.03) (0.03) (0.00) Occ 2 0.07 0.84 0.09 0.30 Occ 2 0.04 0.85 0.11 0.27 (0.00) (0.01) (0.00) (0.00) (0.01) (0.01) (0.01) (0.01) Occ 3 0.04 0.19 0.77 0.60 Occ 3 0.03 0.18 0.79 0.68 (0.00) (0.01) (0.01) (0.00) (0.00) (0.01) (0.01) (0.01) Notes: Each cell ij represents the average probability (for a given NSS survey round) of a household head working in occupation i having a child working in occupation j. Occ 1 collects white collar workers, Occ 2 collects blue collar workers, while Occ 3 refers to farmers and other agricultural workers. Column titled ‘size’ reports the fraction of parents employed in occupation 1, 2, or 3 in a given survey round. Standard errors are in parenthesis. We report the results for the three broad occupation categories we defined earlier in Table 3.7. Each row of the Table denotes the occupation of the parent while columns indicate the occupation of the child. Thus, going across columns along any row i would indicate the probability of a household head working in occupation i to have a child working in the relevant occupation column. Clearly, off-diagonal elements measure the degree of intergenerational occupational mobility. Column “size” reports the average share of parents employed in each of the occupations in a given round. The Table has two panels: Panel (a) gives the numbers for 1983 and Panel (b) for 2004-05. Table 3.7 reveals a few interesting features. First, the diagonal elements of both Panel (a) and (b) are quite high, indicating relatively little intergenerational occupa- tion mobility over this period. The highest persistence rates (or the least mobility) 71 in 1983 was in occupation 3 (agriculture) for both SC/STs and non-SC/STs with the persistence rate being slightly higher for SC/STs. In 2004-05, the persistence rate in occupation 3 was significantly lower for both caste groups, though the SC/ST rate re- mained larger. The intergenerational persistence in occupation 2, in contrast, increased, and significantly so for SC/STs. In fact, in the 2004-05 round, occupation 2 shows the most intergenerational persistence among all occupations. Interestingly, SC/STs also experienced a large increase in intergenerational persistence in occupation 1, while non- SC/STs saw a reduction in that persistence. These trends imply a dramatic convergence in the intergenerational persistence of all occupations between the two caste groups. Second, the probability of the son of a farmer (working in occupation 3) switching to occupations 1 or 2 has risen for both groups. This probability is of interest to us as it indicates an improvement in the quality of jobs across generations. In 1983 the probability of an intergenerational switch from occupation 3 to occupations 1 or 2 was 13% for non-SC/STs and 11% for SC/STs. By 2004-05 these numbers had risen to 23% for non-SC/STs and 21% for SC/STs. We interpret these findings as evidence of conver- gence in upward occupation mobility of both caste groups, with SC/STs experiencing larger positive changes. Third, the probability of a child working in occupation 3 conditional on his father being employed in occupation 1 or 2 has declined from 50% to 31% for SC/STs and from 30% to 23% for non-SC/STs over our sample period. We believe that this reflects a significant reduction in regress prospects of SC/ST households during this period. 3.4.3 Industry mobility Given the large sectoral changes in India during the period under study, an issue of independent interest is the degree of industry mobility in India between 1983 and 2004- 05. We define intergenerational industry switch in the same manner as occupation switches and estimate the conditional probability of industry switches using equation (3.2). Figure 3.13 presents the overall probability of industry switches at the four-digit level as well as the probability of switches for SC/STs and non-SC/STs (dotted lines plot the ± 2 std error confidence bands). The figure shows that the overall probability of children switching the industry of employment relative to their parent has risen from 26 percent in 1983 to 36 percent in 2004-05 period. The industry mobility trends of both SC/STs and non-SC/STs have been similar although the level of the switching probability has remained significantly higher for non-SC/STs. We also estimated the 72 probability of industry switching at the three-, two-, and one-digit levels and found similar time-series trends, with little convergence across the two groups. Figure 3.13: Intergenerational industry switches . 2 . 25 . 3 . 35 . 4 1983 1987−88 1993−94 1999−00 2004−05  overall non−SC/ST SC/ST Avg prob of ind switch − 4 digit Notes: This figure presents the average predicted probability of intergenerational industry switch for our overall sample, for SC/STs and non-SC/STs. The numbers are reported for the five NSS survey rounds. Dotted lines are ±2 std error bands. As with the occupation mobility estimates, the main difference when considering more aggregated industry categories is that the probability of an industry switch is universally lower.25 Industry transition matrix We now turn to the industry choices of the two groups. Using the same approach that we employed to evaluate occupation mobility, we compute industry transition probabilities and summarise them in Table 3.8. As with occupation transition probabilities, each row of the Table denotes the industry of the parent’s employment while columns indicate the industry of the child’s employment. Thus, going across columns along any row i would indicate the probability that a household-head working in industry i has a child working in the relevant industry column. Off-diagonal elements measure the degree of intergenerational industry mobility. Column “size” reports the average share of parents employed in each of the industries in a given round. Panel (a) gives the numbers for 1983 and Panel (b) for 2004-05. Not surprisingly, Ind 1 (agriculture) has remained the primary industry of employ- ment for both SC/STs and non-SC/STs throughout, although its share has declined 25The results for three-, two- and one-digit industry categories are available upon request. 73 Table 3.8: Intergenerational industry transition probabilities (a). Average mobility in the 1983 round Non-SC/ST To SC/ST To From Ind 1 Ind 2 Ind 3 size From Ind 1 Ind 2 Ind 3 size Ind 1 0.87 0.08 0.05 0.67 Ind 1 0.90 0.06 0.04 0.78 (0.00) (0.00) (0.00) (0.00) (0.01) (0.01) (0.00) (0.01) Ind 2 0.11 0.83 0.06 0.24 Ind 2 0.17 0.75 0.08 0.13 (0.01) (0.01) (0.00) (0.00) (0.02) (0.02) (0.01) (0.01) Ind 3 0.19 0.32 0.49 0.08 Ind 3 0.24 0.18 0.58 0.09 (0.01) (0.01) (0.01) (0.00) (0.02) (0.02) (0.03) (0.00) (b). Average mobility in the 2004-05 round Non-SC/ST To SC/ST To From Ind 1 Ind 2 Ind 3 size From Ind 1 Ind 2 Ind 3 size Ind 1 0.77 0.17 0.07 0.60 Ind 1 0.79 0.12 0.09 0.68 (0.01) (0.00) (0.00) (0.00) (0.01) (0.01) (0.01) (0.01) Ind 2 0.08 0.82 0.10 0.29 Ind 2 0.13 0.72 0.15 0.18 (0.00) (0.01) (0.01) (0.00) (0.01) (0.02) (0.01) (0.01) Ind 3 0.16 0.36 0.49 0.11 Ind 3 0.12 0.29 0.60 0.14 (0.01) (0.01) (0.01) (0.00) (0.01) (0.02) (0.02) (0.01) Notes: Each cell ij represents the average probability (for a given NSS survey round) of a household head working in industry i having a child working in industry j. Ind 1 refers to agriculture, Ind 2 collects all traded industries, while Ind 3 refers to all nontraded industries. Column titled ‘size’ reports the fraction of parents employed in industry 1, 2, or 3 in a given survey round. Standard errors are in parenthesis. significantly between 1983 and 2004-05. Ind 1 also has the highest persistence of the three industry groups. The numbers indicate that intergenerational industry persistence has decreased sharply for Ind 1. Children are switching from agriculture into other in- dustries more frequently in 2004-05 in comparison with 1983. While most of this move is primarily into tradable industries, the probability of moving into non-tradable in- dustries has increased, especially for SC/STs. At the same time, the probabilities of moving from Ind 2 or Ind 3 into Ind 1 have declined and more so for SC/STs. We interpret these results as evidence of upward industry mobility, especially for SC/STs. 3.4.4 Income mobility Our fourth, and probably the most typical, measure of intergenerational mobility is on income. We turn to this issue next. The goal of this exercise to provide a measure of the degree to which the long run income of a child of a family is correlated with the long 74 run income of his father. The intergenerational elasticity of long run income is typically estimated as the slope coefficient in a regression of the log of the long run income (relative to the mean) of the child on the log of the parents’ long run income (relative to the mean for the parents’ generation). The estimated coefficient indicates the degree to which income status in one generation gets transmitted to the next generation. The typical problem surrounding income mobility regression specifications is the absence of measures of long run income. The standard procedure is to use short run measures of income as proxies for long run income. We face the same problem since our income data is the daily wage during the survey period. Clearly, the daily wage may be a very noisy measure of long run income with significant associated measurement error. Moreover, as pointed out by Haider and Solon (2006), an additional problem with using short run measures for children’s income is the systematic heterogeneity in income growth over the life cycle. In particular, individuals with higher lifetime income also tend to have steeper income trajectories. As a result, early in the life cycle, current income gaps between those with high lifetime incomes and those with low lifetime incomes tend to understate their lifetime income differences while current income gaps later in the life cycle overstate the lifetime income gaps. We follow Lee and Solon (2009) to address these issues by (a) introducing controls for children’s age to account for the stage of the life-cycle at which the income is observed; (b) introduce an interaction between parents’s income and children’s age to account for the systematic heterogeneity in the profiles; and (c) by instrumenting parents’s income with household consumption expenditure and household size to mitigate the measure- ment error associated with using daily wage data. Hence, our regression specification is wic = α + βwip + γ1Aip + γ2A 2 ip ++γ3A 3 ip + δ1Ãic + δ2à 2 ic +δ3à 3 ic + θ1wipÃic + θ2wipà 2 ic + θ3wipà 3 ic + εi (3.3) where wic denotes the log daily wage of the child of household i and wip is the log daily wage of the male head of the same household. Aip denotes the head of household i’s age while Ãic is the child’s age, which we normalised to equal zero at age 23 which is the mean age of children in our sample. The control for a cubic in parents’ age is to account for differences in the ages of parents in the sample at the time of observing their child’s income. As pointed out in Haider and Solon (2006), the short run proxy for long run income of parents will bias 75 the estimated β downward. However, as long as the bias is stable over time it will not alter the interpretation of how the intergenerational elasticity of income has evolved over time. We run this regression separately for each NSS sample year. Note that the constant α picks up any sample year specific factors. The key parameter of interest is β. We estimate a different β for each NSS round and focus on how the estimated β’s have changed over the sample period. Figure 3.14: Intergenerational income mobility . 3 . 4 . 5 . 6 . 7 . 8 . 9 1 1983 1987−88 1993−94 1999−00 2004−05  non−SC/ST SC/ST Income mobility: OLS . 3 . 4 . 5 . 6 . 7 . 8 . 9 1 1983 1987−88 1993−94 1999−00 2004−05  non−SC/ST SC/ST Income mobility: IV Notes: Figures (a) and (b) present the results from the OLS and IV regressions, respectively, of child’s per day log real wage on parent’s per day log real wage and a set of controls. The figure plot the coefficients on the parent’s wage from those regressions estimated separately for non-SC/STs and SC/STs. All estimated coefficients are statistically significant. Detailed estimation results are presented in the Appendix. We plot the OLS estimates in panel (a) of Figure 3.14 below, while panel (b) of the Figure presents our estimates from an instrumental variable (IV) regression.26 We should note that all the point estimates in both figures are significant at the 1 percent level except for the OLS estimate for 1987-88 which is significant at the 5 percent level. There are three features of the results worth noting. First, the income persistence across generations has declined sharply over the period 1983 and 2004-05 for both SC/STs and non-SC/STs. In fact by the end of our sample period the estimates are much closer to the typical numbers around 0.45 that are reported for the USA by a number of different studies (see Solon, 2002). Second, there has been a clear convergence in intergenerational income persistence across the two groups. Third, the IV estimates are uniformly higher than the OLS estimates. This is similar to the findings of Solon (1992) for the US. 26As with the other regressions, the complete estimation results are available in supplemental tables from http://faculty.arts.ubc.ca/vhnatkovska/research.htm. 76 More importantly, they confirm our findings from the OLS estimation. In fact, IV estimates suggest that SC/STs’ intergenerational income persistence has declined from a whopping 0.87 to 0.45 and, by the end of our sample period, was below that for non-SC/STs. Overall, our results suggest that there has indeed been an upward trend in the degree of intergenerational mobility in education, occupation, industry and income. However, there are significant differences in the convergence patterns of these mobility indicators for SC/STs relative to non-SC/STs. While intergenerational educational and wage mobility of the two groups have tended to converge, occupational and industry mobility rates have not converged similarly. 3.5 Conclusion In this chapter we have studied the evolution of occupation and industry choices, ed- ucation attainment rates and wages in India between 1983 and 2004-05 with a special focus on the fortunes of scheduled castes and scheduled tribes (SC/STs). We have found that the 22-year period under study has been a period of dramatic changes for these historically disadvantaged groups. SC/STs have systematically reduced the gap with non-SC/STs in education attainment levels and have been changing occupation and industry of employment at increasingly faster rates. Moreover, the wage gap between SC/STs and non-SC/STs has narrowed sharply during this period. We have also found that the majority of the wage gap is accounted for by differences in education whose contribution has been rising over time. The caste effect on wages appears to have almost disappeared. Crucially, we find that these trends are the sharpest amongst the younger cohorts and in urban areas. The last two features are especially uplifting since they are potentially indicative of the types of changes one may expect in the future since India has been becoming increasingly urbanised and younger over time. It is worth reiterating that SC/ST wages have been converging toward non-SC/ST levels across cohorts, education and occupation categories. Moreover, the speed of this convergence is impressive not just at an absolute level but also when compared to the wage convergence experienced by historically disadvantaged minority groups elsewhere such as Blacks and Hispanics in the USA. We find this evidence particularly reassuring in terms of the future prospects of SC/STs in India. What explains these significant changes in the Indian social landscape? We believe that the rapid structural changes in the Indian economy over the past 25 years are at 77 the heart of this progress. The liberalisation of the previously restricted economy has opened up new opportunities for the private sector. While the increase in potential opportunities is common to all segments of the population, the more rapid response of SC/STs probably reflects a confluence of factors. One factor may be that the rapidly changing socio-economic environment in India has presented SC/STs socio-economicc opportunity to break out of a centuries-old cycle of illiteracy and poverty, and they have been acting proactively to take advantage of it. A strengthening of community based networks of SC/STs along the lines suggested in Munshi (2010) may have also been at play in accelerating this process. The second possibility is that the reservations policy in place since 1950 for public sector jobs and higher education seats may have played a role in the declining wage gaps. The first possibility implies that caste may be becoming a less important factor in economic allocations in India while the second factor would put caste-based policies at the center of the explanation. We intend to examine these potential explanations in future work. 78 4. Food Preference and Nutrition in India 4.1 Introduction In a country as poor as India, income growth is desirable because it has a significant impact on human development indicators including health status. One of the major determinants of health status is nutritional intake. As India has exhibited dramatic economic growth in recent times, one would have expected a significant improvement in nutritional status. This chapter tries to take stock of the nutritional status of various income groups and how it has changed over time. In the past several decades, India has experienced rapid economic growth and struc- tural transformation. Often these changes in transitional economies appear to coincide with social changes. Indian urban lifestyles are rapidly embracing globalised culture befitting its emerging status. Despite these changes in India, we observe disappointing performance in health status. Waves of national family health surveys in India reveal that malnutrition continues to be a significant problem for all age groups in India (IIPS (2007)). Moreover, Indians suffer from the dual burden of poor nutrition: more than 33% of adults are underweight while more than 10% are obese. Only 57% of males and 52% of females were considered to be healthy in terms of the weight for height index in 2005-06. According to World Health Organisation, obesity is a major risk factor for several chronic illnesses, including diabetes, heart diseases and cancer. Once it was believed that obesity is a problem only in high income countries. But, this phenomenon is now dramatically on the rise in low-income and middle-income countries, especially in urban areas. Obesity has been more than double in total world population since 1980. According to WHO fact sheet, “65% of the world’s population live in countries where overweight and obesity kills more people than underweight”27. As the economy develops, the share of expenditure on food falls and that on non-essential commodities 27Data retrieved on 6 June, 2011 from http://www.who.int/mediacentre/factsheets/fs311/en/index.html 79 rises -an idea popularly known as Engel’s law. But income growth has accompanied many changes in society, such as ideas of social status and perception of diet pattern associated with social status. Due to these social transformations, changes in tastes and living standards significantly influence the composition of food demand and diet quality. Powerful marketing strategies, developments in communications and media, rapid urbanisation and the irresistible demonstration effect give rise to changes in both rural and urban areas. Food share in total expenditure dropped from 70% to 56% in rural areas and from 63% to 46% in urban areas between 1983 and 2004-05. However, a closer look at the composition of food groups reveals that a drop in expenditure share on cheap sources of carbohydrates is not properly compensated by more balanced diets, particularly from sources of proteins. This is more prominent among the urban rich. The compelling question is: why do we not see a significant development in nutrition status despite higher economic growth and poverty alleviation? Does it mean that changes in tastes and lifestyles impede improvement in nutrition status which should be engendered by better economic fortunes? The National Sample Surveys of India collect household-level consumption data on several food items. It allows me to examine structural changes in food preferences and changes in nutritional status during the period of rapid economic growth. I start by pre- senting evidence of changes in expenditure share on food and its composition between 1983 and 2004-05. During this period, the distribution of monthly per capita real ex- penditure has shifted significantly in both rural and urban areas, showing a favourable improvement in poverty. We observe an optimistic change in the economic fortunes of socially and economically deprived groups (Hnatkovska et al. (2010)). Therefore, it is important to ask whether such changes in the Engel ratio are related to observable changes in income, price and related shifts in demography. Or is it primarily due to change in the structure of the behaviour that affects income elasticities? I answer this question by modelling the demand system of six food groups using household consump- tion survey data. I find that during this period an increase in income coincided with a shift in income elasticities, but not in price elasticities (except for milk and milk products). Inferior goods became more inferior during this period, whereas households appear to be less selective in choosing superior quality nutrient-rich foods. They shifted expenditure share to fatty products instead of protein-rich products. Moreover, income elasticities for protein-rich foods decreased over time. As the composition of food groups changed, the nutrient share and price per nutrients also changed during this period. In 80 rural areas 78% of total calorie intake was from cereals and cereal products in 1983 and it decreased to 71% in 2004-05. However, this drop in share of calories from staples was mostly matched by a rise in share from poor-quality diet, namely, beverages, ed- ible oils, spices and sugar (defined as ”other”). I estimate the calorie elasticity (with respect to total expenditure) and calorie Engel curve to show that nutritional intake does not necessarily improve with income in India. Calorie elasticity was low in 1983 and it is even lower now in both rural and urban areas, posing a serious policy concern. Mere economic development does not lead to sufficiently high nutritional intake which is important from the perspective of the efficiency wage argument. I also compute a household-level diet quality index based on recommended daily allowances (RDA) of 12 nutrients and examine the factors affecting it. A significant proportion of the rich (top 25%) households do not meet the recommended daily allowances of major nutrients, let alone poor households. The proportion of rural households that meet the recom- mended daily intake of calories reduced by 16 percentage points while that of protein intake dropped by 8.5 percentage points during this period. In contrast, the proportion of rural households that meets the RDA of fat intake has increased by 18 percentage points. The trend is similar in urban areas as well. Almost 100% of the urban rich (top 25%) meet the RDA of fat intake; however, they lack calorie intake in general and protein intake specifically. When we place these findings in perspective we apprehend an alarming situation in India if proper policies are not implemented. Studies on food demand and nutrition in India have been sufficiently large. There are two strands of literature: one set of studies focuses on demand system estimation using either aggregate data or micro-data and another set focuses on nutrition. Ray (1980) estimates Almost Ideal Demand System of food groups pooling aggregate data in India from 1952 to 1969. Coondoo and Majumder (1987), Majumder (1992) and Chattopad- hyay et al. (2009) are some of the studies that dealt with the demand system in India. Meenakshi and Ray (1999) analyse regional differences in food preferences during the seventies and eighties. Two important studies that examine the relationship between nutrition and income in India are Behrman and Deolalikar (1987) and Subramanian and Deaton (1996). For a sample of relatively poor rural households from South India during 1976-78, Behrman and Deolalikar (1987) find that nutrient expenditure elas- ticities are not significantly positive. Using a rural sample from the western state of Maharashtra in 1983, Subramanian and Deaton (1996) find a relatively higher elasticity of calorie intake with respect to total expenditure. All these studies cover the period 81 before liberalisation in India. In a recent study, Deaton and Dreze (2009) examine some evidence of recent changes in food intake and nutrition in India. My study is primarily motivated by some puzzles put forward by the authors. There is substantial evidence of a sustained decline in per capita calorie, protein and many other nutrients except fat consumption during the period from 1983 to 2004-05. Their study also points out that it is difficult to relate the decline in calorie consumption to observable changes in income or relative prices of foods since there is clear evidence of a right-ward shift of real per capita consumption distribution and little change in food price relative to other commodities. Therefore, we find a downward shift of the calorie Engel curve. Deaton and Dreze (2009) draw our attention to several plausible explanations for the falling calorie intake with rising income over time. Two interesting explanations are changes in food habits and declining needs for calories. While all these arguments need to be substantiated, my study focuses on the changes in the structure of the behaviour that affects income elasticities, i.e., changes in tastes and preferences. This chapter serves three purposes: 1) a detailed study of the changes in Engel ratios and nutrient intake between 1983 and 2004-05, 2) modelling the food demand system in India and identify- ing any changes in the preference structure and 3) the determinants of diet quality in India. The rest of the chapter is organised as follows. The next section describes the data. Section 4.3 3 looks at food share and its determinants. Section 4.4 presents the results of the food demand system estimation. Results on nutrition and diet quality are presented in Section 4.5. Section 4.6 concludes. 4.2 Data The data used in this chapter is taken from the unit-level record of the consumption schedule of National Sample Surveys (NSS) in India. The National Sample Survey Organisation (NSSO), set up by the Government of India, conducts rounds of sample survey to collect socio-economic data. Each round is earmarked for a particular subject coverage. I selected two specific ’thick’ rounds, viz., 38th and 61st, to reflect any changes in a long-run period spanning before and after the economic reforms undertaken in 1991. There are two types of samples in each NSS round: one collected by the NSS (central sample) and the other collected by the state (state sample). I use the central sample collected from all geographical parts of India except some remote pockets of Jammu & 82 Kashmir, Andaman & Nicober Islands and the north-eastern states. After cleaning the outliers, I had a total of 102,191 and 124,541 households in the 38th and 61st rounds, respectively. The NSS follows multi-stage stratified sampling, rural and urban being the first stage strata in each state. Almost 70% households in my working sample are from rural areas in 1983 and that reduced to 64% in 2004-05. The consumption schedule col- lects information on several household-level demographic and economic characteristics. It also collects household member’s age, sex, marital status, education, etc. Respon- dents are asked to recall consumption spending in rupees as well as physical quantity whenever appropriate on more than 300 items, of which more than 145 are food items, for a recall window of 30 days and/or 365 days depending on spending frequency. Total consumption spending on each item consists of consumption from the market and con- sumption from home-grown production valued appropriately at locally prevailing prices. The lists of items are very similar in all rounds, and therefore consistency over time is fairly well maintained. The field survey is conducted in four sub-rounds spanning the year, which helps to minimise any seasonality in consumption spending. A list of items and their broader grouping is provided in the appendix. The surveys do not collect data on household income since it is believed that any such attempt will introduce response error. Therefore, we follow the standard practice in the literature that allows the use of total consumption spending or outlay instead of total income. The data files of 38th round consumption schedule provided by the NSSO has serious problems in its clarity of data presentation and documentation. The data extraction process itself is strenuous. I checked the validity of my extraction process by comparing my estimates with the published results (NSS and other independent work). Other than the above problem, the consumption schedule data files of this particular round do not provide information on age28. The only viable option to get this information is to match households of the Consumption Schedule with the Employment and Unemployment schedule which reports household member’s demographic characteristics. My sample for the 38th round consists of only matching households. The nutritional content of all food items are calculated using the conversion factor reported in Gopalan et al. (1974). Some food items, for example, other beverages, other processed foods and other fresh fruits, are dropped from this analysis due to ambiguous 28The questionnaire collects the age of household members. However, for unknown reasons, the NSSO does not provide this variable in the unit-level dataset. Instead, they report some aggregate-level age groups. 83 measurement units and the nutrient conversion factor. The dropped items have a very small expenditure share. However, it appears that these items are mostly consumed by the rich and therefore if this omission introduces some error, it will underestimate the nutritional intake only for the rich. As we are interested in changes in preferences and nutritional intake, these biases should not affect our results since we apply the same procedure for all rounds. Consumption spending on food does not necessarily reflect actual consumption. House- holds buy food for their own consumption as well as for guests and other non-members. Similarly, household members eat meals outside the home. Therefore, I adjust the ac- tual nutritional intake using a conversion factor derived from number of meals to guests and number of meals from outside. The details of this methodology are discussed in Section 4.5. The NSS collects information on the number of ceremonies in the recall period and how many guest meals are served during ceremonies and any other day. It also collects information on the number of meals each household member has taken outside. I use poverty lines in terms of monthly per capita expenditure in rupees, published by the Planning Commission of India, as a deflator to convert nominal quantity to real quantity. The poverty line of rural Maharashtra in 1983 is used as the base and nominal quantities are converted to real quantities in terms of rural Maharashtra in 1983. 4.3 Food share In less developed countries, food expenditure share typically ranges between 50 to 80 percent. Food is regarded as the foremost necessity and overall welfare is closely related to share of expenditure on food. As the economy develops this share falls and expen- diture on other amenities like education, health, housing, clothing, and entertainment increases. During the past three decades the Indian economy performed well in terms of GDP growth and moderately well in reducing the poverty level. The average real monthly per capita expenditure (MPCE) in rural areas has increased from Rs. 107 in 1983 to Rs. 142 in 2004-05, which is a steady increase by almost 33%. In the urban sector we observe a sharper increase by 42% from Rs. 127 during the same period. However, inequality, measured by standard deviation of the logarithm of real MPCE, has increased in the urban sector between 1983 and 2004-05 (Table 4.1). Kernel den- 84 PL0 . 2 . 4 . 6 . 8 1 D en sit y 3 4 5 6 7 log(real MPCE) 1983: Rural 2004−05: Rural Log real MPCE distribution in rural sector (c) Rural PL0 . 2 . 4 . 6 . 8 D en sit y 3 4 5 6 7 log(real MPCE) 1983: Urban 2004−05: Urban Log real MPCE distribution in urban sector (d) Urban Note: The vertical lines show All India poverty lines for rural and urban sectors. Figure 4.1: MPCE distribution (1983 to 2004-05) sity functions of real MPCE in 1983 and 2004-05 are plotted in Figure 4.1. In both sectors, the distributions have shifted to the right. The vertical lines show all-India poverty lines for the rural and urban sectors in 1983. The area below the poverty line under the curve depicts the head count ratio. During this period the poverty ratio has declined substantially. In 1983, the mode of the rural MPCE distribution was close to the poverty line, and that for urban was below the poverty line. Though there is substantial improvement in the poverty ratio, the urban population still concentrates around the poverty line. Therefore, it is evident that the poverty ratio is very sensitive to the choice of the poverty line. Another interesting observation is that the right tail of the urban distribution is heavier in 2004-05. Table 4.1: Mean and SD of food shares, real MPCE and ln(realMPCE) 1983 2004-05 Rural Urban Rural Urban Variables Mean SD Mean SD Mean SD Mean SD Real MPCE 106.91 64.32 127.01 88.95 141.97 95.36 180.16 151.86 ln(real MPCE) 4.55 0.48 4.68 0.55 4.83 0.46 4.98 0.62 Food 69.69 9.91 62.75 11.14 55.79 10.36 45.87 12.06 Cereals 53.03 18.07 36.83 16.66 36.14 12.89 27.36 10.73 The stochastic shift in MPCE distributions between 1983 and 2004-05 is reflected in food share distributions. Food share dropped from 70% to 56% in the rural sector, 85 whereas it dropped from 63% to 46% in the urban sector (Table 4.1). More than one- half of total food expenditure was on cereals and cereal substitute in rural areas in 1983. Over time, it reduced to 36% - a drop of almost 32%. Urban population on average spent only 37% and 27% of total food budget on cereals and cereal substitutes in 1983 and 2004-05, respectively. About 50% of rural Indians spent more than 70% of their monthly budget on food items, whereas only 26% of urban Indians spent more than 70% on food items in 1983. The corresponding figures dropped to 5% and 2.4%, respectively in 2004-05 (Table B.1). I plot the Kernel density of expenditure share on food in Figure 4.2. Over time, food share distribution has shifted to the left in both the rural and urban sectors, showing significant improvement in welfare. It appears that the change is larger in urban areas. In 1983, both the rural and urban distributions are somewhat negatively skewed. However, the urban distribution appears to be ’normal’ in 2004-04, whereas rural distribution is still negatively skewed in 2004-05. The thicker lower tail of urban food share distribution in 2004-05 corresponds to the heavier upper tail of MPCE distribution consisting of well-off families who have smaller food budget shares. Almost 60% of urban households spent less than 50% of their budget on food in 2004-05. 0 1 2 3 4 D en sit y 0 .2 .4 .6 .8 1 Foodshare 1983: Rural 2004−05: Rural Foodshare distribution in rural sector in 1983 and 2003−04 (a) Rural 0 1 2 3 4 D en sit y 0 .2 .4 .6 .8 1 Foodshare 1983: Urban 2004−05: Urban Foodshare distribution in urban sector in 1983 and 2003−04 (b) Urban Figure 4.2: Food share distribution (1983 to 2004-05) The trend of falling food share is apparent across geographical regions in India (Table B.2, B.3). Punjab and Goa had the lowest food share (64%) in 1983, whereas West Bengal and Orissa had the highest food share (76%) during that time in the rural sector. In 2004-05 state29-level average food shares ranged between 48% (Puducherry) 29State refers to state as well as union territories. 86 and 63% (Assam). It is interesting to note that the average gap between the food shares of the top and bottom quartiles of MPCE distribution in the rural sector has increased over time. In general, the eastern part of India (Assam, West Bengal, Bihar and Orissa) have the highest food share in 1983 as well as in 2004-05. Figure 4.3 shows the changes in food share for four representative states across four geographical parts in India.      
  
  Punjab     	 (a) Punjab      

	    Maharashtra      	  (b) Maharashtra      

	    Tamilnadu      	  (c) Tamilnadu      

	    Assam      	  (d) Assam Note: The scales are in fraction. Figure 4.3: Change in food share for selected states (1983 to 2004-05) In general, food share and income are negatively related. As a household becomes richer, the share of expenditure on food falls -an idea widely known as ”Engel curve” and often used in the literature as a measure of household welfare. I present a contour plot of the joint distribution of the logarithm of MPCE and food share in Figure 4.4. The relationship between MPCE and food share is non-linear and after transforming 87 Foodshare and lnMPCE(real) − 1983: Rural log(real MPCE) Fo o ds ha re  0.5  1  1.5  2  2.5  3  3.5 3 4 5 6 7 0. 2 0. 4 0. 6 0. 8 1. 0 (a) Rural, 1983 Foodshare and lnMPCE(real) − 1983: Urban log(real MPCE) Fo o ds ha re  0.5  1  1.5  2  2.5  3 3 4 5 6 7 0. 2 0. 4 0. 6 0. 8 1. 0 (b) Urban, 1983 Foodshare and lnMPCE(real) − 2004−05: Rural log(real MPCE) Fo o ds ha re  0.5  1  1.5  2  2.5  3  3.5  4 3 4 5 6 7 0. 2 0. 4 0. 6 0. 8 1. 0 (c) Rural, 2004-05 Foodshare and lnMPCE(real) − 2004−05: Urban log(real MPCE) Fo o ds ha re  0.2  0.4  0.6  0.8  1  1.2  1.4  1.6  1.8  2  2  2.2 3 4 5 6 7 0. 2 0. 4 0. 6 0. 8 1. 0 (d) Urban, 2004-05 Note: Inner contour lines indicate higher densities. Figure 4.4: Joint distribution of food share and log(real MPCE) per capita expenditure to logarithm scale, I get a reasonable linear relationship. The contour lines indicate different levels of densities with inner lines for higher densities. In 1983, the densities were highly concentrated around the mode. However, in 2004-05 we see a greater span of food share and lower kurtosis (consistent with Figure4.2(b)), particularly in urban areas. The range of food share is wide for any given level of per capita income. We also observe that the contours have been shifted to the south-east corner over time, indicating an improvement in household welfare in terms of rising 88 MPCE and falling food share. In particular, in 2004-05 more urban households are spread towards very low food share and higher MPCE. All these figures indicate that the empirical relationship between food share and per capita income is strongly inverse. As the marginal distribution of log(MPCE) (Figure 4.1) is fairly positively skewed and that for food share (Figure 4.2) is slightly negatively skewed, we expect a considerable curvature in the relationship between log(MPCE) and food share. Let us now study more closely the relation between food share and per capita in- come and any changes in this relationship over time. It is quite plausible that very poor households, that do not meet subsistence level of food consumption, may actually increase the share of expenditure on food with increasing income. Once they meet the subsistence level, food share falls. Therefore, we expect some concavity of the Engel curve at lower income levels. Possible non-linearity in this relationship could be ad- dressed by non-parametric estimation. I use the locally weighted regression method of Fan (1992) to plot estimated food share along real monthly per capita expenditure. Let us denote the regression function of food share on log(MPCE) as m(x) = E(y|x), (4.1) where x is the logarithm of MPCE and y is food share. In the small neighbourhood of a point x0, m(x) could be approximated as m(x) ≈ m(x0) +m ′(x0)(x− x0) = a+ b(x− x0). Therefore, the estimation of the regression function m(x0) is equivalent to the intercept parameter of a local regression in the neighbourhood of x0. If we repeat this exercise for all possible values of x0 in the sample, we find a smooth non-linear relationship between per capita expenditure and expenditure share on food. The methodology is simplified using the procedure described in Deaton (1997); Subramanian and Deaton (1996). Practically, I run a linear regression of household expenditure share on food (y) on the logarithm of monthly per capita household expenditures at 50 grid points (x) instead of all values of the independent variable. First, I calculate a series of weights for each data point (log(MPCE)) within a given bandwidth (h) around each grid point. 89 This is done using a quartic kernel function θi(x) = ωi 15 16 [ 1− ( x− xi h )2]2 (4.2) where θi(x) is the weight for household i in the neighbourhood of grid point x, which depends on household survey weights ωi and bandwidth h. If xi lies beyond the h- neighbourhood, a zero weight is assigned. Then, the estimated parameters of the weighted regression at each grid point x is given by β̂(x) = [X ′Θ(x)X]−1X ′Θ(x)Y (4.3) where Θ(x) is a diagonal matrix of weights of each households(θj(x)), X is matrix of two columns (ones in first column and log(MPCE) in second column) and Y is the vector of food expenditure shares.For each grid point, I estimate two parameters (intercept and slope). The predicted value of food share at each grid point is thus given by m̂(x) = β̂1(x) + β̂2(x)x. (4.4) Standard errors are calculated using bootstrapping. I repeatedly (100 times) select random sample of households with replacement, using rural-urban stratification, in- dependently for each round. Estimates from these re-samplings are used to obtain standard errors. Figure 4.5 shows the local regression estimates of the bi-variate regression function of food share on log(MPCE). As expected, the estimated lines are upward rising at extremely low per capita expenditure levels. It then starts falling and takes a linear form. The 95% confidence band is very tight except for the extreme tails. Two important observations should be explained here. First, the Engel curves have shifted down for all income (expenditure) groups, with a larger drift for richer households. Second, food share starts falling with MPCE even for households living well below the subsistence level defined by the poverty line. Households with the same real MPCE in 1983 and 2004-05 spend lower expenditure share on food in 2004-05. Most importantly, this is also true for subsistence households. This implies that food price has to be cheaper in 2004-04 to meet subsistence demand for food consumption of households with the same real income level in 1983 and 2004-05. The falling expenditure share on food over time alongside an increase in total expenditure (downward drift) is in sharp contrast 90 PL.2 . 4 . 6 . 8 Fo od ha re 3 4 5 6 7 ln(realMPCE) 1983 2004−05 Rural Foodshare and MPCE (a) Rural PL.2 . 3 . 4 . 5 . 6 . 7 Fo od ha re 3 4 5 6 7 ln(realMPCE) 1983 2004−05 Urban Foodshare and MPCE (b) Urban Note: I select 50 grid points and h = 1. The dotted lines are 95% confidence bandwidth and the vertical lines indicate poverty lines. Figure 4.5: Local regression estimate of Engel curve to a positive correlation between the two when we look across very poor households at a point in time (upward sloping segment of the Engel curve). We should note that the rising relationship at the lower tail of the MPCE distribution has higher standard errors. If the relative price of food is steady over time (I do not consider the recent time of global food price inflation after 2008), then this implies that calorie intake is lower in 2004-05 for any given real expenditure level in 1983 and 2004-05. I will discuss this in greater detail in Section 4.5. We also observe that the upward rising segment of the Engel curve in the rural sector is prominent in 2004-05 compared to that of 1983. As mentioned before, for a given point of time, households just below the poverty line have a falling food expenditure share with rising per capita expenditure. This falling segment is sharper in urban areas in 2004-05. The Indian poverty line, defined in terms of real expenditure on a basket of commodities of which food is a major component, is considered to be a conservative estimate of subsistence level. Why do households, being so poor that they cannot afford a subsistence basket divert their expenditure share to non-food consumption goods? While this question is important on its own merit, we focus only on food consumption, nutrition and changes in food preferences over time. So far I have considered per capita expenditure as the only determinant of food share. Household composition also affects the shape of the Engel curve. I plot a non- parametric Engel curve for varying household composition in Figure B.1. There is considerable variation in the shape of Engel curves for different household sizes and 91 number of children in the household. However, all shapes indicate that there is a quadratic relationship between food share and per capita expenditure. Figure 4.6 shows how closely the quadratic regression function fits the non-parametric Engel curve. The . 5 . 55 . 6 . 65 . 7 . 75 Fo od ha re 3 4 5 6 7 ln(realMPCE) Quadratic fit Non−parametric fit Rural − 1983 Quadratic Engel curve of food (a) Rural . 4 . 5 . 6 . 7 Fo od ha re 3 4 5 6 7 ln(realMPCE) Quadratic fit Non−parametric fit Urban − 1983 Quadratic Engel curve of food (b) Urban . 3 . 4 . 5 . 6 . 7 Fo od ha re 3 4 5 6 7 ln(realMPCE) Quadratic fit Non−parametric fit Rural − 2004−05 Quadratic Engel curve of food (c) Rural . 3 . 4 . 5 . 6 Fo od ha re 3 4 5 6 7 ln(realMPCE) Quadratic fit Non−parametric fit Urban − 2004−05 Quadratic Engel curve of food (d) Urban Figure 4.6: Comparison of quadratic fit and non-parametric fit of Engel curve quadratic function is a close approximation of locally regression function except for extremely low MPCE households. Non-parametric estimation is useful to explore the bi-variate relationship between MPCE and food share, but it becomes problematic when we recognise the potential role of other variables. In what follows I consider a quadratic model of the Engel curve including household composition and geographical dummies. 92 Table 4.2 reports an estimation of the following model: wi = β0 + β1lmpcei + β2lmpce 2 i + β3adultmalei + β4adultfemalei + β5childreni + β6adultmale 2 i + β7adultfemale 2 i + β8children 2 i + β9adultmalei ∗ lmpcei + β10adultfemalei ∗ lmpcei + β11childreni ∗ lmpcei + β12adultmalei ∗ childreni + β13adultfemalei ∗ childreni + 3∑ j=1 γjsubroundij + 30∑ k=1 δkstateik + i (4.5) where wi and lmpcei refer to expenditure share on food and log(MPCE) of household i, respectively. I control for number of adult males, females and children below age 15 years. I also include the interaction of these variables with log(MPCE) and dummies for sub-rounds and states. The sub-round dummies capture any seasonality effects in food demand. Estimation is carried out for each sector and round separately. The concav- ity of the Engel curve is confirmed, except for the urban sector in 2004-05, controlling for demographic characteristics and geographic dummies. The number of children is a significant positive determinant of household’s food share. However, for any given level of real per capita expenditure, expenditure elasticity of food share decreases with number of children (Table B.5). Price varies considerably across geographical locations. All households residing in the same place face similar market conditions, and therefore location dummies would potentially capture some of these market effects. The NSS divides each state into many regions based on population densities and geographical similarities. Model (4.5) is estimated with region dummies to control for market condi- tions which potentially include aggregate-level price effects. The estimation results are reported in Table B.6. In fact, there is no significant change in estimated coefficients, i.e., state dummies and region dummies capture similar aggregate effects. It is impor- tant to note that geographical dummies do not capture the quality effect embedded in prices. When we exploit price variation at the household level, the quality information embedded in prices are also taken care of. I will use household-level price in demand system estimation in Section 4.4. Overtime the coefficient of lmpce has increased in the rural sector, but it decreased in the urban sector. In fact, this variable becomes negative (insignificant) in 2004- 05 for the urban sector. The quadratic term of MPCE remains significantly negative throughout the years in both sectors. However, the effect of this quadratic term becomes stronger in the rural sector, whereas it becomes weaker in the urban sector in 2004-05. 93 Table 4.2: Multivariate model of food share 1983 2004-05 Rural Urban All Rural Urban All lmpce 0.255*** 0.361*** 0.288*** 0.324*** -0.012 0.278*** (12.72) (8.51) (16.72) (14.28) (-0.60) (17.33) lmpce2 -0.031*** -0.045*** -0.036*** -0.040*** -0.010*** -0.038*** (-14.52) (-10.85) (-20.34) (-19.41) (-5.81) (-26.09) Adult males -0.003 -0.018 -0.011* 0.015* 0.024** 0.013* (-0.52) (-1.55) (-2.02) (2.20) (2.94) (2.51) Adult females 0.003 0.027 0.012 0.021** 0.021** 0.023*** (0.41) (1.62) (1.73) (2.84) (2.69) (4.14) Children 0.019*** 0.034*** 0.012*** 0.031*** 0.012* 0.018*** (4.97) (4.63) (3.52) (6.82) (2.08) (4.85) Sub-round dummy Yes Yes Yes Yes Yes Yes State dummy Yes Yes Yes Yes Yes Yes R-sqr 0.204 0.271 0.223 0.337 0.555 0.407 RMSE 0.088 0.095 0.092 0.084 0.080 0.090 N 72404 29785 102189 79253 45288 124541 Note: This table reports selected coefficients. Please refer to table B.5 for detail result. t-values are in parenthesis. Standard errors are calculated correcting for heteroskedasticity as suggested by Davidson and MacKinnon (1993). In effect, I calculate standard errors using vce(hc3) option in STATA regression command. ∗p < 0.05, ∗ ∗ p < 0.01, ∗ ∗ ∗p < 0.001. I plot the partial effect of lmpce and lmpce2 in Figure 4.7. The predicted food share is plotted against the log of real MPCE in Figure 4.7(b). There is a downward drift in predicted food share for rural areas and this change is higher for richer households. However, for urban households the change in predicted food share is not similar across income groups. Overall the slopes of the curves have been changed over time. Figure 4.7(a) shows that the marginal effect of per capita expenditure on food share is lower in 2004-05 for rural households except for the extremely poor section. The drop in marginal effect is remarkable for the urban poor. The marginal effect of log(MPCE) is negative for urban households throughout all income groups in 2004-05. It is clear from Figure 4.7(a) that the marginal effect is negative except for very poor households whose per capita expenditure (log scale) is below 4 (approximately). As pointed out earlier, there are a large number of households that are just below the poverty line and prefer to decrease their share of food expenditure with a higher MPCE (negative marginal effect for 4 ≤ log(MPCE) ≤ PL). The marginal effect schedule of rural households in 2004- 05 is similar to that for urban households in 1983. In other words, rural expenditure elasticity today looks much like urban expenditure elasticity 20 years ago. 94 PLr PLu − . 2 − . 1 0 . 1 M ar gi na l e ffe ct 3 4 5 6 7 ln(realMPCE) 1983, rural 1983, urban 2004−05, rural 2004−05, urban Note: The marginal effects are derived from ∂ƒ(lmpce)⁄∂lmpce = β1 + 2β2lmpce at 50 grid points of lmpce. (no family composition effect) Marginal effect of lmpce on food share (a) Marginal effect PLr PLu.3 . 4 . 5 . 6 . 7 Pr ed ic te d sh ar e 3 4 5 6 7 ln(realMPCE) 1983, rural 1983, urban 2004−05, rural 2004−05, urban Note: The predicted values are derived from w = β0 + β1lmpce + β2lmpce2 at 50 grid points of lmpce. (partial effect of lmpce) Predicted food share (b) Predicted share Figure 4.7: Partial effect of per capita expenditure on food share These results indicate that there is substantial change in the impact of income on food consumption in India. I test it formally using pooled cross-section data for the years 1983 and 2004-05. Equation 4.5 is estimated for the rural and urban sectors separately including year dummy and the interaction of the year dummy with all explanatory variables. The results are reported in Table 4.3. The year dummy, interactions of year and per capita expenditure and the square of per capita expenditure are significant in the urban sector. I test the joint significance of year ∗ lmpce and year ∗ lmpce2. The F values are high in both sectors, and therefore we can reject the hypothesis that there is no significant change in the relationship between food share and per capita expenditure from 1983 to 2005. With substantial changes in food share and the marginal effect of per capita expenditure on food share, we expect a similar shift in the composition of food demand and income and price elasticities. As mentioned earlier, recent socio- economic developments in India may have some effect on tastes and preferences for food consumption. There are several potential ways, other than any changes coming from higher income or price movement, through which structural changes in the economy may affect the food demand pattern. First, there may be less requirement for calories due to changes in occupational structure. Particularly in urban areas workers may engage in sedentary occupations. Second, the energy requirement is less due to improved health and hygiene. Third, there may be a wider choice of food items available in the market and, therefore, the composition of food demand changes over time. Fourth, the demand pattern may change due to the demonstration effect from foreign cultures. Fifth, fast food is becoming increasingly popular in urban centres because they are convenient to prepare and eat. 95 Table 4.3: Multivariate model of food share - pooled cross section Rural Urban year=2004-05 -0.039 0.776*** (-0.47) (6.26) lmpce 0.251*** 0.349*** (12.57) (8.40) lmpce2 -0.030*** -0.044*** (-14.33) (-10.84) year*lmpce 0.058 -0.369*** (1.92) (-7.92) year*lmpce2 -0.009** 0.034*** (-3.09) (7.70) year*adult male*lmpce -0.003 -0.009** (-1.65) (-3.15) year*adult female*lmpce -0.005* -0.000 (-2.20) (-0.06) year*children*lmpce -0.003** 0.005* (-2.90) (2.53) Sub-round dummy Yes Yes Region dummy Yes Yes R-sqr 0.517 0.638 RMSE 0.085 0.085 N 154308 78246 Chow test F( 2,151478) F( 2, 74896) year*lmpce, year*lmpce2: = 31.20 = 32.33 Note: This table reports selected coefficients. The detail results are available upon request. t-values are in parenthesis. Standard errors are calculated correcting for heteroskedasticity as suggested by Davidson and MacKinnon (1993). ∗p < 0.05, ∗ ∗ p < 0.01, ∗ ∗ ∗p < 0.001. Whatever the reasons for such changes in preference, one could expect a recent shift in the composition of food groups in India. Figure 4.8 shows the percentage changes in average budget shares in the urban and rural sectors. It is quite clear that during this period both rural and urban Indians reduced their consumption share of staples and substituted that with dairy, meat, eggs, fish, fruits, vegetables and beverages. In general, rural households show a higher percentage change in almost all food items. Table B.4 shows that the average share of cereal expenditure in total food expenditure has decreased by more than 30% in rural areas and that in urban areas it decreased by more than 25%. Expenditure share on meat, eggs and fish has increased by 35% and 6%, respectively, in rural and urban areas. Per capita per month consumption of different food groups are reported in Table B.7. Average rural households consumed 15 kg of cereals per month per capita in 1983 and 96      Relative change in food groups share (1983 to 2004-05) Rural Urban         Figure 4.8: Relative change in food groups share (1983 to 2004-05) this dropped by 20% to 12 kg. Per capita pulse consumption also decreased during this period. Dairy, meat and fish consumption remains almost the same. However, we observe a substantial increase in vegetable consumption for both rural and urban households (Figure 4.9). When we compare changes in expenditure share and per capita consumption of dif- ferent food groups (Figure 4.8 and 4.9) we see a complex pattern of changes in food demand. While cereal consumption dropped in both expenditure share and quantity, pulse consumption dropped in quantity without any substantial changes in expenditure share. We notice opposite patterns for dairy, meant and fish in the rural sector. While the decline in the cereal and pulse expenditure share occurred in both the rural and urban sectors, the magnitude of the decline was different across the expenditure spec- trum. Figure 4.10 shows the composition of the food group for the top and bottom expenditure quartiles. In fact, expenditure share on meat, eggs, fish, fruits, vegetables, beverages and others of the poorest quartile have converged to that of the top quartile. Apparently, while some of these shifts in consumption pattern are due to increase in in- come and changes in the relative prices of food groups, some parts cannot be attributed to visible changes in income and prices. The complexity of these changes could be fully understood in a complete demand system analysis. In the next section I estimate the food demand system to find changes in preferences. 97       Relative (%) change in consumption (per capita per month)        Note: All items in a food group are not necessarily have same nutritional values. Quantities are converted to same unit for all items in same food group. Milk includes all dairy products. I assume 1 litre = 1kg of milk. Figure 4.9: Relative change (%) in food consumption (1983 to 2004-05) 0 . 2 . 4 . 6 . 8 1 Sh ar e Poorest 25% Richest 25% 1983 2004−05 1983 2004−05 Rural Composition of food Cereal Pulses Dairy Meat, egg and fish Fruits and vegetables Other (a) Rural 0 . 2 . 4 . 6 . 8 1 Sh ar e Poorest 25% Richest 25% 1983 2004−05 1983 2004−05 Urban Composition of food Cereal Pulses Dairy Meat, egg and fish Fruits and vegetables Other (b) Urban Figure 4.10: Composition of food groups (1983 to 2004-05) 98 4.4 Food demand system The primary motivation of this study is to provide evidence of changes in food prefer- ences in India in recent years. While the Engel curves derived in the previous section give some idea about the relationship between food expenditure and total outlay and their changes in recent years, a complete demand system consisting of both income and price effects consistent with consumer theory is necessary to identify shifts in food preferences. The literature on demand analysis in developing countries in quite small for several reasons, scarcity of good quality data being one of them. However, there are a number of studies on consumer demand in India based on household expenditure data (for example Ray (1980, 1985); Coondoo and Majumder (1987)). There are two approaches to demand system estimation: 1) conventional time se- ries approach to estimate price and income elasticities exploiting variation in prices of commodity groups and aggregate consumption expenditure across income groups over time; and 2) micro-data approach making use of household level variation in expendi- ture. The first approach introduces biases due to interaction of household characteristics with income and price effects. The second approach explicitly model household level composition. However, in second approach, we need enough price variation across geo- graphical regions to identify the price effects. It is quite plausible to estimate both price and budget elasticities in micro-data approach pooling repeated cross section of house- holds. The estimation of demand system using household data is challenging for several reasons. First, for any given commodity, there could be zero consumption for some households implying censored dependent variable. However, in a demand system with fewer aggregated commodity groups, zero expenditure cases are minimised. Second, measurement errors in recall consumption data poses serious problem. Ahmed et al. (2006) point out that measurement errors are substantial and they characteristically differ from classical measurement error. Third, prices at household level could poten- tially include quality effects, thus it is hard to isolate pure price effects. Fourth, total expenditure of households could be endogenous. Moreover, if households consume home produced goods, particularly agricultural products, distinguishing price effect from in- come effect is difficult. Despite these challenges, micro-level demand estimation has several attractive features. The price and income elasticities estimated on individual household data can be used to evaluate welfare changes at more disaggregated level. Aggregation of individual elasticities is readily comparable to that derived from aggre- gated data. My goal in this section is to examine any change in the preference structure 99 that can not be attributed to visible changes in prices and income. Thus distinguishing pure price effects from quality effects or from income effect is not so important. How- ever, the measurement error causes serious problem. The extent of the measurement error is estimable only if we collect more accurate consumption data. It is believed that food diary data (households are asked to fill a diary) are more accurate than food recall data. India conducts consumption survey using recall technique only. There is no comparable food diary survey to estimate the bias. The methodology of estimation of the demand system incorporating both price effect and income effect using household budget data, is developed by a series of works by Barten (1964), Muellbauer (1976), Deaton and Muellbauer (1980), Jorgenson et al. (1982), Deaton (1987), Blundell et al. (1993), Banks et al. (1997) and others. The Almost Ideal Demand System (AIDS) developed by Deaton and Muellbauer (1980) is very popular in the literature because it allows exact aggregation over households and is easier to estimate in its linear form. The AIDS expenditure share equations have the following form wih = αi + ∑ j γij ln pj + βi ln [ mh P (p) ] (4.6) where wih is share of expenditure on food item i by household h, mh is household total outlay, pj is the price of food item j and P (p) is a price index defined over the vector of all prices p as follows: lnP (p) ≡ α0 + ∑ k αk ln pk + 1 2 ∑ j ∑ k γjk ln pj ln pk. (4.7) The share equation 4.6 can be expressed in a more general form as wih = Ai(p) +Bi(p) ln xh (4.8) where xh ≡ mh/a(p) and a(p) is some general form of price index similar to Equation 4.7. This class of demand system has expenditure shares that are linear in a logarithm of total expenditure (lnm) and derived from the indirect utility function that is also linear in lnm. This class of demand systems is called Price-Independent Generalised Logarithmic (PIGLOG) (Muellbauer (1976)). Different food items may have different income effects at different points across income distribution. Therefore, it is important now to examine empirically whether the expenditure share of different food groups has a linear form in lnm. Figure 4.11 shows non-parametric local regressions, quadratic 100 polynomial regressions and confidence bands for non-parametric Engel curves for two important food groups, namely, cereals and pulses. The specification of locally weighted regression is exactly the same as described in Section 4.3. Although the linear form of the Engel curve appears to be a reasonable approximation for cereals (cereals and cereal substitutes), a non-linear relationship is evident from Figures 4.11 and B.2 for other food groups, particularly pulses, dairy, fruits and vegetables. We define six distinct food groups, viz.,“cereal”, “pulses”, “milk”, “meat, egg and fish”, “fruits and vegetables” and all remaining food items grouped as “others”30. The expenditure share of cereals and cereal substitutes follows a downward sloping relationship with total food expenditure with a downward drift in 2004-05. In contrast, pulses and pulse products became downward sloping in 2004-05 in rural areas and remained fairly stable in urban areas. The divergence between the non-parametric line and the quadratic line is prominent when the confidence band is large. The above results suggest that while the Engel curve for some food groups can be summarised by a linear model, for certain other food groups higher order terms of log(MPCE) are necessary. To serve this purpose, I used the Quadratic Almost Ideal Demand System (QAIDS), developed by Banks et al. (1997) as an extension of AIDS. Following Banks et al. (1997), ), a general form of demand, consistent with my empirical Engel curves in Figure 4.11 is given by wih = Ai(p) +Bi(p) ln xh + Ci(p)g(xh). (4.9) where g(xh) is a differentiable function. The additional term Ci(p)g(xh) supplements Equation 4.8 with a smooth function of total expenditure, g(xh) that allows non- linearities. Any demand system is said to be exactly aggregable as long as the system is linear in functions of total income (expenditure), m. According to Gorman (1981), the maximum possible rank of any exactly aggregable demand system is 3, where rank refers to the number of independent terms on the right-hand side of the expenditure share equation. Thus, the AIDS demand system has rank 2 (Equation 4.8) and it is ex- actly aggregable. Banks et al. (1997) show that all exactly aggregable demand systems, consistent with utility maximisation, in the form of Equation 4.9 must have indirect 30See Table B.8 for grouping of food items 101 0 . 2 . 4 . 6 . 8 ce re al _a ll sh ar e 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Rural cereal_all share and total expenditure in food (a) Cereal - Rural 0 . 2 . 4 . 6 ce re al _a ll sh ar e 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Urban cereal_all share and total expenditure in food (b) Cereal - Urban . 03 . 04 . 05 . 06 . 07 pu lse  sh ar e 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Rural pulse share and total expenditure in food (c) Pulses - Rural 0 . 02 . 04 . 06 . 08 pu lse  sh ar e 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Urban pulse share and total expenditure in food (d) Pulses - Urban Note: Other food groups are plotted in figure B.2. Figure 4.11: Non-parametric and quadratic Engel curves for food groups utility functions of the form ln v(p,mh) = {[ lnmh − ln a(p) b(p) ]−1 + λ(p) }−1 (4.10) where the term λ is a differentiable, homogeneous function of degree zero in prices p and the function g(xh) must take a quadratic form in ln xh, i.e. g(xh) ≡ (ln xh) 2. Following Banks et al. (1997), I use the following specification for a(p), b(p), λ(p) and wih. ln a(p) ≡ lnP (p) as in AIDS model (Equation 4.7), b(p) ≡ ∏ i=1 pβii , (4.11) 102 λ(p) ≡ ∑ i λi ln pi (4.12) and wih = αi + ∑ j γij ln pj + βi ln [ mh P (p) ] + λi b(p) { ln [ mh P (p) ]}2 . (4.13) Theory imposes several restrictions on the parameters αi, βi, λi and γij. Adding-up condition, i.e. ∑ iwi = 0 implies that∑ i αi = 1, ∑ i βi = 0, ∑ i γi = 0, and ∑ i γij = 0 ∀j. (4.14) As demand functions are homogeneous of degree zero in (p,m), we must have ∑ j γij = 0 ∀i. (4.15) Moreover, Slutsky symmetry requires γij = γji. (4.16) I estimate the system of equations (4.13) for six food groups (i = 1 to 6) using a non- linear seemingly unrelated regression method. A multivariate normal error structure is assumed with the system of equations (4.13) for estimation purpose. Adding up implies that the variance-covariance matrix of the errors is singular. Therefore, one of the six demand equations is dropped from our estimation and the parameters of this dropped equation are recovered from the parametric restrictions stated above. I also include two demographic variables, namely, the number of adults and number of children below age 15 years, in Model (4.13). The demographically extended QAIDS is, therefore, given by wih = αi + θiah + ηich + ∑ j γij ln pj + βi ln [ mh P (p) ] + λi b(p) { ln [ mh P (p) ]}2 + ih (4.17) where ah and ch are number of adults and number of children in household h, respec- tively. I have two cross-sections in two different time points. I include a time dummy and interaction of the time dummy with all explanatory variables on the right-hand side to allow taste changes between 1983 and 2004-05. The adding-up condition must be satisfied by the additional coefficients of ah, ch, time dummy and interaction terms. 103 Therefore, the summation of each additional parameter should be equal to zero. The parameter α0 in Equation (4.7) is usually difficult to identify and, therefore, following Deaton and Muellbauer (1980), it is set to an a priori value. I assume α0 = 5 in estimation of system of non-linear equations (4.17) using the Feasible Generalised Non-linear Least Squares (FGNLS) method. Utilising parametric restrictions, we have a total of 80 parameters to estimate (including interaction terms). The results are given in Table B.9. Differentiating the budget share equation with respect to lnm and pj, the following expressions are derived, respectively. µi ≡ ∂wi ∂ lnm = βi + 2λi b(p) { ln [ m a(p) ]} (4.18) µij ≡ ∂wi ∂ ln pj = γij − µi ( αj + ∑ k γjk ln pk ) − λiβj b(p) { ln [ m a(p) ]}2 (4.19) The expenditure and uncompensated price elasticities are given by ei = µi wi + 1 (4.20) euij = muij wi − δij (4.21) where δij is the Kronecker delta. Estimation of Equation 4.17 requires data on household-level budget share on food groups, number of adults, number of children aged below 15 years, total nominal expen- diture on all food items and prices. Price variation at the household level also captures the quality of food. In addition to these variables we have a year dummy for 2004-05 and the interaction of the year dummy with all explanatory variables. Following Poi (2002) and Poi (2008), I use Stata nlsur estimation routine implementing Feasible Generalised Non-linear Least Squares. Unlike the AIDS model where λ = 0, the expenditure and price elasticities are functions of total expenditure, m. Table 4.4 and 4.5 report the estimated coefficients and elasticities at mean values for the rural and urban sectors, respectively. As expected, expenditure elasticities are positive and own price elasticities are negative. The expenditure inelastic nature of cereals helps explain the fall in share of expenditure on cereals during this time. Moreover, the falling expenditure elasticity of cereals implies faster changes in diet composition. As the share of cereals and pulses falls, more variety is introduced in the diet. 104 In the rural sector, expenditure elasticities have decreased and price elasticities are stable over time for all food items except dairy products. Own price elasticity (un- compensated) for dairy products is double in 2004-05 and expenditure elasticity has increased from 2.13 to 2.40. In the urban sector, we observe substantial changes in both price and expenditure elasticities. Own price elasticities (absolute) of cereals and dairy products have increased, whereas meat, eggs and fish has become less price elastic during this time period. With the exception of meat, eggs and fish and ’other’, all price elasticities are below unity (absolute value). It is clear that expenditure elasticities for dairy products and meat, eggs and fish are larger than unity in both periods. Expen- diture elasticities for pulses, fruits and vegetables and ’others’ dropped below one in the rural sector, whereas expenditure elasticity for fruits and vegetables in the urban sector declined from 1.13 to 0.99. While expenditure elasticity in the urban sector has dropped significantly for almost all food groups, it has increased for dairy products and remains almost the same for beverages, processed food and other food stuff. There is a rural-urban gap in elasticities. Cereals and pulses are more expenditure inelastic in urban areas compared to the rural sector. Dairy products are less expenditure elastic, whereas meat, eggs, fish, fruits and vegetables are more expenditure elastic in the urban sector compared to that in the rural sector. My primary interest is to examine any changes in tastes and preferences. Table B.9 reports the estimated coefficients and their difference between 1983 and 2004-05. The intercept parameter αi is interpreted as the budget share of item i of a subsistence household at base year price (Meenakshi and Ray (1999)). The estimated values of αi and βi in 2004-05 are significantly different from 1983. The parameter λ captures non- linearities in demand system with respect to expenditure, m. The estimated values of λ confirm that confirm that the demand system is in fact non-linear in total expenditure for several food items. The marginal effects of number of adults and number of children are significant in all sectors. If we define structural changes in preference as changes in consumption behaviour that appear to be arising separately from price and income effects, then differences in the estimated values of α, θ and η between 1983 and 2004-05 capture such changes. Table B.9 shows that the values of α are significantly different in two periods except for pulses in the rural sector and meat, egg and fish in the urban sector. 105 Table 4.4: Estimation of expenditure and own price elasticities - rural Food groups Year α θ η β λ χ2(10) Expenditure elasticity Own price elasticity cereal all 1983 0.437*** 0.0207*** 0.0215*** -0.175*** -0.0475*** 16664.24 0.73 -0.59 (98.51) (25.09) (33.89) (-46.91) (-19.31) (0.0048) (0.0053) 2004-05 0.197*** 0.0238*** 0.0269*** -0.202*** -0.0512*** 0.57 -0.60 (51.97) (39.47) (53.77) (-53.75) (-22.21) (0.0068) (0.0085) pulse 1983 0.0281*** 0.000322 0.0000460 0.00129 0.00113 1644.69 1.01 -0.89 (22.83) (1.73) (0.28) (1.64) (1.82) (0.0108) (0.0168) 2004-05 0.0250*** 0.00199*** 0.00111*** -0.0169*** -0.00420*** 0.78 -0.85 (21.32) (13.68) (8.66) (-20.96) (-12.12) (0.0085) (0.0117) milk 1983 0.158*** -0.0116*** -0.00917*** 0.117*** 0.0195*** 5797.78 2.13 -0.47 (57.45) (-22.18) (-24.19) (49.47) (15.34) (0.0199) (0.0094) 2004-05 0.294*** -0.0222*** -0.0172*** 0.220*** 0.0401*** 2.40 -0.94 (75.44) (-35.86) (-34.69) (46.59) (16.46) (0.0215) (0.0082) mef 1983 0.0625*** -0.00246*** -0.00277*** 0.0168*** 0.00143 2815.17 1.39 -1.10 (38.26) (-8.14) (-16.27) (12.66) (1.87) (0.0302) (0.0094) 2004-05 0.0740*** -0.00141*** -0.00482*** 0.0131*** -0.00205*** 1.27 -1.02 (45.38) (-5.25) (-20.83) (9.03) (-3.45) (0.0191) (0.0072) vegfruit 1983 0.0920*** -0.000549** -0.00192*** 0.00388*** 0.00618*** 13894.45 1.00 -0.89 (76.29) (-2.79) (-11.74) (3.63) (7.08) (0.0089) (0.0059) 2004-05 0.131*** 0.000659** -0.00180*** -0.0146*** 0.00345 0.88 -0.85 (95.89) (2.87) (-9.81) (-7.99) (1.94) (0.0054) (0.0048) other 1983 0.223*** -0.00644*** -0.00767*** 0.0364*** 0.0192*** 8112.55 1.13 -1.04 (99.39) (-17.63) (-25.53) (19.07) (14.86) (0.0082) (0.0062) 2004-05 0.278*** -0.00283*** -0.00426*** -0.0000663 0.0139*** 0.95 -1.09 (98.61) (-8.16) (-14.68) (-0.02) (6.49) (0.0058) (0.0084) System chi2(40) =36260.30 Note: t statistics in parentheses for estimated parameters. Standard errors are in parenthesis for elasticities. * p<0.05, ** p<0.01, *** p<0.001. chi2(10) and chi2(40) give chi test statistics for significance of each demand equation and whole demand system respectively. See Table B.9 for detail results. 106 Table 4.5: Estimation of expenditure and own price elasticities - urban Food groups Year α θ η β λ χ2(10) Expenditure elasticity Own price elasticity cereal all 1983 0.229*** 0.0239*** 0.0258*** -0.169*** -0.0411*** 2933.44 0.62 -0.51 (27.99) (16.86) (25.68) (-13.38) (-5.10) (0.0206) (0.0153) 2004-05 0.0877*** 0.0246*** 0.0257*** -0.179*** -0.0468*** 0.51 -0.69 (17.79) (28.74) (34.73) (-39.48) (-26.46) (0.0124) (0.014) pulse 1983 0.0370*** 0.000945*** 0.00101*** -0.00758*** -0.000498 658.17 0.88 -0.82 (27.12) (4.89) (5.40) (-3.91) (-0.18) (0.0119) (0.021) 2004-05 0.0282*** 0.00279*** 0.00195*** -0.0225*** -0.00691*** 0.73 -0.79 (25.11) (16.84) (12.03) (-23.47) (-20.05) (0.0114) (0.0169) milk 1983 0.197*** -0.0112*** -0.00906*** 0.0965*** 0.00737 1622.42 1.67 -0.74 (43.85) (-14.07) (-15.98) (12.96) (1.60) (0.0322) (0.0109) 2004-05 0.285*** -0.0142*** -0.0135*** 0.144*** 0.0182*** 1.73 -0.90 (59.30) (-17.37) (-20.93) (30.43) (12.73) (0.0221) (0.0081) mef 1983 0.0783*** -0.00313*** -0.00294*** 0.0336*** 0.00360* 204.96 1.52 -1.00 (33.19) (-7.10) (-7.84) (10.72) (2.08) (0.0347) (0.0137) 2004-05 0.0769*** -0.00278*** -0.00356*** 0.0257*** 0.00296*** 1.37 -0.89 (27.22) (-6.08) (-8.95) (9.11) (3.38) (0.0344) (0.0076) vegfruit 1983 0.132*** -0.00185*** -0.00446*** 0.0237*** 0.0121** 2298.53 1.13 -0.82 (48.64) (-4.14) (-14.29) (5.44) (3.11) (0.0194) (0.009) 2004-05 0.162*** -0.00152*** -0.00343*** 0.0000473 0.00190 0.99 -0.83 (68.84) (-4.15) (-10.05) (0.02) (1.06) (0.0115) (0.0059) other 1983 0.326*** -0.00858*** -0.0103*** 0.0224*** 0.0185*** 496.42 1.04 -1.21 (72.51) (-13.02) (-17.49) (5.08) (4.21) (0.0108) (0.0144) 2004-05 0.360*** -0.00893*** -0.00719*** 0.0315*** 0.0306*** 1.01 -1.19 (69.82) (-13.30) (-12.82) (7.33) (13.95) (0.0101) (0.0157) System chi2(40) = 6112.10 Note: t statistics in parentheses for estimated parameters. Standard errors are in parenthesis for elasticities. * p<0.05, ** p<0.01, *** p<0.001. chi2(10) and chi2(40) give chi test statistics for significance of each demand equation and whole demand system respectively. See Table B.9 for detail results. 107 The coefficients for number of adults in household (θ) in two time periods are signif- icantly different for all food groups in rural areas, whereas in urban areas the differences are significant only for pulses and dairy products. The marginal effects of number of children (η) are different in two time points, except for fruits and vegetables in the rural sector and cereals and meat, eggs and fish in the urban sector. While the estimated structural shift in food preference is statistically significant and this result is important from an economic policy perspective (for example, tax policy), such a transformation is important from the health policy perspective too. In summary, the results confirm a structural shift in food preferences in favour of more variety in the diet. A general implication from the above results can be drawn. Improved income status may have led to varied food habits of Indians with less cheap sources of nutrients and more expensive sources of nutrients. However, the changes in food habits can be at the expense of nutrient adequacy. As the changes in food patterns are complex, a simple inference about diet quality and nutritional value is impossible to make from the above results. We need to examine empirically whether structural changes in food preferences actually lead to substantial changes in nutritional status. Even if we find strong evidence of changing nutritional intake in India, it is difficult to relate them without proper analysis. While the limitation of data prohibits us from undertaking such an analysis to identify the causal relation between changes in preferences and changes in nutritional status, we may, however, examine some of the changes in the nutritional intake of the Indian population. In the next section we analyse the nutrition and determinants of diet quality in India. 4.5 Nutrition The nutrient values of food items consumed by households are calculated by applying conversion factors reported in Gopalan et al. (1974) for each disaggregated food item. The estimates of total calories and nutrients of different food groups are thus derived by aggregation over all food items in that food group. Some of the food items are reported in value (Rs.) only and some items have ambiguous quantity units. The nutrient conversion for these items is done based on nutrient per constant rupee unit, as suggested in NSSO (2007). It is important to note that the nutrient consumption so derived may not reflect the true estimate of nutrient intake. Households may serve food to guests and other non-members and, similarly, household members may eat food 108 from outside. Therefore, I use the procedure suggested by Minhas (1991) to calculate adjusted nutritional intake at the household level 31. The total quantity of food reported as consumed by the household may be divided into three components: 1) number of meals served to household members (Mh), 2) number of meals served to guests (Mg) and 3) number of meals served to employees (Me). The number of free meals taken by household members from outside as guest or employee (Mf ) is not reflected in reported quantity. All this information is collected for the recall period. Following Minhas (1991), the adjusted nutrient intake is given by Na = N Mh +Mf Mh +Mg +Me (4.22) where Na is the adjusted nutrient intake and N is the calculated nutrient intake from the reported quantity. There have been declining trends in per capita calorie intake in both the rural and urban sectors. Per capita per day calorie consumption in rural India dropped from 2,266 kcal in 1972-73 to 2,047 kcal in 2004-05 -a drop of almost 10% (Table B.10). In general, the urban population consumes fewer calories than their rural counterparts. However, Figure 4.12(a) shows that calorie consumption in the rural sector has been converging towards that of the urban sector since 1972-73. Protein intake in the rural sector shows a declining trend from 1983 (Figure 4.12(b)). whereas, in the urban sector, protein intake increased from 1972-73 to 1983 and then stays stable. In 2004-05, per capita per day protein intake is the same in the rural and urban sectors. Contrary to the downward and converging trend in calorie and protein intake, we observe an upward and slightly diverging trend in fat intake. Figure 4.12(c) shows that the urban population consumes more fat compared to the rural population and while per capita per day fat consumption in the urban sector has been rising sharply since 1983, the rural sector shows a steady upward rising trend in fat consumption since 1972-73. During the period 1972-73 to 2004-05 per capita fat consumption in the rural and urban sectors increased by 48% and 32%, respectively, whereas protein consumption has dropped by 8% in the rural sector and increased by only 1.8% in the urban sector during this period. It is expected that per capita per day calorie consumption will be different across income groups. Calorie Engel curves are estimated using the locally weighted regres- 31The same methodology is used by the NSSO in its report NSSO (2007). Therefore, my results will be easily comparable with those of the NSSO. 109       K ca l Per capita per diem intake of Calorie  (in Kcal)          K ca l (a) Calorie      gm Per capita per diem intake of Protein  (in gm)           gm (b) Protein       gm Per capita per diem intake of Fat  (in gm)          gm (c) Fat Source: NSSO (2007). Figure 4.12: Trend in energy and macronutrients intake in India sion method (Figures 4.13(a) and 4.13(b)). The logarithm of real MPCE is measured along the horizontal axis and the logarithm of per capita per day calorie intake in measured along the vertical axis. The dotted lines show a 95% confidence band using bootstrapping. Per capita calorie consumption has a positive relationship with monthly per capita consumption expenditure. There are significant changes in this relationship between 1983 and 2004-05. The extremely poor households in the urban sector consume more calories and richer households consume fewer calories in 2004-05 compared to 20 years ago. The drop in per capita per day calories in the rural sector is observed across all expenditure groups, with a higher proportion for richer households. Changes in per capita calorie consumption due to changes in total income (expendi- ture) may differ across expenditure groups. It is plausible that poorer households, with insufficient income to buy a subsistence basket of foods, may have higher elasticity of calorie consumption with respect to total expenditure. Non-parametric calorie elastic- ity curves are drawn in Figures 4.14(a) and 4.14(b) for the rural and urban sectors, respectively. The lines give the slope of the calorie Engel curve in Figure 4.13. Here we observe more clearly that elasticity decreases with an increase in MPCE. It is quite clear 110 PL6. 5 7 7. 5 8 8. 5 Ca lo rie  in ta ke  (l og ) 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Rural Percaipta perday calorie (a) Rural PL 7 7. 5 8 Ca lo rie  in ta ke  (l og ) 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Urban Percaipta perday calorie (b) Urban Figure 4.13: Calorie Engel curve that the relationship between per capita calorie consumption and total expenditure has changed between 1983 and 2004-05. The measured elasticity is lower in 2004-05 for all expenditure groups. The drop in elasticity for richer households is bigger in the rural sector and it is smaller in the urban sector. Calorie elasticity for rural households with MPCE close to the poverty line is around 0.37 in 2004-05. The corresponding figure for urban households is around 0.22. These figures give some estimate of the extent to which nutrition responds to income. It is widely believed that malnutrition will be eliminated by economic prosperity. However, declining calorie elasticity, specifically for those below the poverty line, poses a serious concern. As calorie elasticities are lower now compared to 25 years ago, improvement in the economic condition of those below the poverty line will not lead to faster improvement in nutritional intake. PL. 2 . 4 . 6 . 8 el as tic ity  o f c al or ie 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Rural Elasticity of percaipta perday calorie (a) Rural PL . 2 . 4 . 6 . 8 el as tic ity  o f c al or ie 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Urban Elasticity of percaipta perday calorie (b) Urban Figure 4.14: Calorie elasticity curve 111 Nutrient elasticity with respect to total expenditure is closely related to total ex- penditure elasticity of food demand. In Section 4.4 I have shown the change in the response of demand for different food groups to a change in total expenditure. Let us look more closely at how changes in preference could affect nutrient elasticity. Changes in the total expenditure elasticity of nutrient intake can be split into two parts: 1) changes in expenditure elasticities of food groups demand and 2) changes in the shares of nutrients obtained from food groups in total nutrient intake. Let n be the amount of total nutrient. Therefore, n = ∑ i kiqi where ki is the nutrient conversion factor for food group i and qi is the demand for food group i. Differentiating with respect to total expenditure E and assuming ki do not change with total expenditure 32, ∂n ∂E = ∑ i ki ∂qi ∂E Expressing in elasticities, ζnE = ∑ i νiζqiE where ζxy is the elasticity of x with respect to y and νi is the share of nutrient obtained from food group i in total nutrient intake. Change in nutrient elasticity is expressed as ∆ζnE = ∑ i ∆νiζqiE + ∑ i νi∆ζqiE (4.23) Therefore, both changes in expenditure elasticities of food groups demand and shares of nutrient obtained from food groups affect the response of nutritional intake to a change in income status. In the previous section we have already seen how structural changes in the food demand pattern affect expenditure elasticities. For example, the average expenditure elasticity of cereals in the rural sector has dropped by 0.16 and that for dairy products has increased by 0.27. The corresponding nutrient shares in total nutrient consumption will determine the aggregate effect of changes in expenditure elasticities 32I assume this for simplicity in the exposition of the link between elasticity of food demand and nu- trient elasticity with respect to total expenditure. This assumption is not valid if households substitute food items within groups. As households become richer, they will substitute away from coarse food items to more quality products within a group. This is reflected in differences in prices per nutrient obtained from each group for the top and bottom quartiles of MPCE. This derivation is not used for any calculation. 112 of food groups demand. Households allocate their budgets to different sources of nutrients depending on their income status. Thus, the distributions of nutrients over food groups vary across income groups and over time. Tables B.11 and B.12 show the distribution of calories and two major macronutrients over six food groups in the rural and urban sectors, respectively. Generally, the biggest source of energy is staples. The bottom quartile in the rural sector consumed 83% of total calories from cereals in 1983 which dropped to 77% in 2004-05, whereas the top quartile consumed 63.7% of the total calories from cereals in 2004-05 compared to 70% in 1983. On average, we observe a 6.7 percentage point drop in the share of calories from cereals during this period. There is a similar drop in the share of calories from cereals in the urban sector. We have already noted in Section 4.3 3 that households substituted away from staples. Therefore, a drop in food share in staples is reflected in a drop in calorie share from staples. The main sources of proteins are pulses, meat, eggs and fish. The tables show a decline in the share of protein from pulses and an increase in meat, eggs and fish in both sectors. The drop in share of proteins from cereals and pulses in being compensated by dairy products, animal products, fruits and vegetables. It is interesting to note that the drop in share of fat from cereals and pulses is completely compensated by a rise in the share from the ’other’ food group which includes beverages, refreshment and processed foods. Changes in the distribution of fat share from different food groups is striking in the rural sector. The poorest quartile in the rural sector consumed 43.6% and 40.6% of total fat from cereals and the ’other’ food group, respectively, in 1983. The corresponding figures changed to 20.6% and 62.6%, respectively, in 2004-05. In fact, the share of fat from the ’other’ food group for the bottom quartile has surpassed that for the top quartile in 2004-05. As households substitute between food groups, the cost of nutrients change. Some foods are rich in calories and are cheap, some contain few calories and are expensive. Therefore, substitution between food groups drives the changes in the cost of nutrients and elasticities of calories with respect to total food expenditure. The gap between expenditure share and calorie share of cereals indicates that these are the cheapest sources of energy. Tables B.13 and B.14 show the cost of calories, protein and fat in the rural and urban sectors, respectively. Cereals are the cheapest source of calories, whereas meat, eggs and fish are the most expensive calorie sources. It required Rs. 0.82 to get 1,000 kcal from cereals in the rural sector in 1983, whereas meat, eggs and 113 fish cost more than Rs. 15 in 1983. The cost of calories from meat, eggs and fish has increased by Rs. 25.86 from 1983 to 2004-05. Meat, eggs and fish account for less than one percent of total calories, around 6% of the total food budget and cost around Rs. 37 for 1,000 Kcal in 2004-05. It is important to note that the ’other’ food group is a cheaper source of energy compared to dairy products, fruits and vegetables and it also accounts for more than 60% percent of fat intake. When households become richer, they substitute coarse quality foods with finer quality and that results in the difference in price per calorie between the rich and the poor. The cheapest source of proteins was pulses and pulse products in 1983. However, cereals became the cheapest protein source in 2004-05. The price per 100 gm of fat is lowest for the ’other’ food group in both the rural and urban sectors. The results in these tables suggest that the cheapest sources of calories, protein and fat are substituted by more expensive sources. Recommended daily allowances (RDA) of nutrients vary across age/sex groups. I use the RDA of nutrients and energy by age and sex groups published by ICMR (2009) to calculate household-level RDA. Aggregate RDA at the household level is derived by using the number of household members in 12 age/sex groups and their corresponding RDA. Table 4.6 reports the proportion of households above the recommended allowances in calorie, protein and fat intake. Only 47% of rural households met the recommended Table 4.6: Proportion of households meeting RDA RURAL 1983 2004-05 Nutrient Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Calorie 46.57 15.32 78.32 30.70 11.21 53.09 Protein 68.15 48.45 86.85 59.67 41.57 77.40 Fat 26.56 4.63 59.27 44.40 14.42 77.96 URBAN 1983 2004-05 Nutrient Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Calorie 30.34 8.53 58.00 20.15 7.52 37.68 Protein 58.50 36.48 80.47 51.79 36.03 68.84 Fat 49.01 12.45 90.10 69.34 32.34 97.36 energy demand per day in 1983 and that figure dropped to 30% in 2004-05. The 114 corresponding figures for the urban sector are 30% and 20%. This drop in proportion of households meeting the RDA of calories is mostly driven by a drop in carbohydrate and protein intake. During this period the distributions of the ratio of nutrients to RDA have shifted left except for fat (Figure B.3). A nutrient/RDA ratio lower than one means that the household falls short of recommended daily allowances. The proportion of households that meet RDA in fat intake has increased by 67% in rural areas and 41% in urban areas. Almost 69% of urban households meet the recommended daily fat intake while only 52% of them meet the protein intake. This disproportionate increase in fat intake is more clear in Figure B.3. It is quite clear that more households are closer to their recommended daily protein and carbohydrate allowances in 2004- 05 compared to 1983. While the absolute numbers of households that do not meet the RDA in calorie and fat intake (Table 4.6) show a disappointing nutrition status in India, changes is distributions of protein/RDA and carbohydrate/RDA clearly have an uplifting message. The thicker positive ‘tail’ of fat/RDA distribution can possibly explain the higher incidence of obesity, specifically in urban India. How does the nutrients/RDA ratio respond to changes in MPCE distribution? I estimate the relationship between log(realMPCE) and nutrient/RDA ratio in 1983 and 2004-05 using non-parametric estimation explained in the previous sections. The plots are given in Figure B.4. There is no change in nutrient/RDA ratio for households below the poverty line and there is a significant downward shift for households in the top quantiles of MPCE distribution in the rural sector. On the other hand, the urban poor experience an upward rise and the urban rich experience a downward fall in the nutrient/RDA ratio. A balanced diet consists of four major sources of energy - carbohydrates, proteins, fats and fibre - as well as different vitamins, minerals and salts. Using Thiele et al. (2004), I calculate the diet quality index of households. Twelve important nutrients are used to calculate the Deficiency Index (DI). The deficiency score for each nutrient is given by di = 100 ∗min ( 1, xi RDAi ) (4.24) where xi is the amount of nutrient i consumed and RDAi is the recommended daily allowance of nutrient i. This score is bounded between 0 and 100 (perfect score). A higher score implies that it is closer to the recommendation. I consider the following macro and micro nutrients: carbohydrates, proteins, fat, fibre, calcium, phosphorous, 115 iron, carotene, thiamine, riboflavin, niacin and Vitamin C. The household-level Defi- ciency Index (DI) is derived by summing the deficiency score, di over all 12 nutrients. Therefore, the Deficiency Index takes a value between 0 and 1200, where a higher score implies better diet quality. There are several nutrients which are bad if taken in excess. Thiele et al. (2004) define a similar index, called Excess Index (EI), for these nutrients. I calculate the EI of fat using their methodology. Excess score is defined as ei = 100− 100 ∗ ( max ( 1,min ( 2, xi RDAi )) − 1 ) (4.25) Therefore, ei is bounded between 0 and 100. Households with fat intake below RDA get a perfect score of 100 and households with fat intake more than double of RDA get a minimum score 0. Since I have taken only fat for EI calculation, EI also takes a value between 0 and 100. It is worth noting that a higher index (both EI and DI) implies better diet quality. Table 4.7 shows that Deficiency Index improved by merely 4% and 3.86% in the rural and urban sectors, respectively, whereas the Excess Index worsened by larger degree in both the rural and urban sectors at 8.45% and 15.9%, respectively. Table 4.7: Average diet quality index Rural 1983 2004-05 Change(%) Mean SD Mean SD Mean DI 877.20 139.93 912.28 116.28 4.00 EI 87.87 26.80 80.44 31.05 -8.45 Urban 1983 2004-05 Change(%) Mean SD Mean SD Mean DI 899.27 151.59 934 117.03 3.86 EI 72.74 37.37 61.18 38.80 -15.90 Diet quality depends on income as well as other socio-demographic characteristics of households. For example, better education may help them choose a better diet over expensive and nutritionally poor foods. It also depends on household composition and the community they belong to. I estimate a simple model of diet quality. It is evident that diet quality index is bounded above and below and we need to use a censored 116 regression model. Let y∗ be the latent diet quality index of the household and y∗ = β0 + β1log(realMPCE) + β2(log(realMPCE)) 2 + β3children + β4adult male+ β5adult female+ β6muslim+ β7SCST + β8head female + β9head age+ β10urban+ 4∑ i=2 γihead edui + 31∑ i=2 δistatei + 4∑ i=2 θisub roundi + u (4.26) where SCST is a dummy for scheduled castes and scheduled tribes and head edui is the education dummy for head’s education. Five education categories are defined, viz., illiterate, below primary, primary, middle school and secondary or above. I also control for geographical effect using state dummies and seasonal effects using sub-round dummies. Muslim dummy and SC/ST dummies are used to see whether social or religious affinity affect diet pattern. Children is defined as age below 15 years. A normal error structure is assumed. The observed diet quality index (calculated index) y is censored by upper (ul) and lower bounds (ll), ie. y =   y∗ if ll ≤ y∗ ≤ ul, ll if y∗ < ll, ul if y∗ > ul. I estimate Equation 4.26 using censored regression for Deficiency Index (DI) and Excess Index (EI) in 1983 and 2004-05 separately. Table 4.8 shows the log-likelihood estima- tion results. The first two columns give the results for the Deficiency Index and the last two columns give the results for Excess Index. Both indices are non-linear with log(realMPCE). Figure 4.15 shows the predicted values for the Deficiency Index and the Efficiency Index from the estimated censored regression model. Several features of these plots are worth mentioning. 1) The deficiency index of diet quality is positively associated with income status. This means that richer households face less deficiency in the 12 nutrients considered here. On the other hand, excess index is negatively as- sociated with affluence level, ie., richer households have worse diet quality in terms of excess index. Note that this index captures only excess of fat intake. 2) Over time the entire schedule of DI has moved down slightly, though it is significant. Reduction in the deficiency index is higher for the richer quantiles of expenditure distribution. 3) Diet 117 PLr PLu40 0 60 0 80 0 10 00 12 00 Pr ed ic te d D I 3 4 5 6 7 ln(realMPCE) 1983 2004−05 Predicted value of Deficiency Index (a) Deficiency Index PLr PLu − 20 0 0 20 0 40 0 Pr ed ic te d EI 3 4 5 6 7 ln(realMPCE) 1983 2004−05 Predicted value of Excess Index (b) Excess Index Note: Predicted over 50 specified values of ln(realMPCE) at mean values of other covariates. Dotted lines are twice standard error confidence band using bootstrapping. Figure 4.15: MPCE and diet quality index quality in terms of excess index is lower now compared to 20 years ago at any given real expenditure level for households below the poverty line. This is driven by the fact that even the poorer sections of society change their preference from low-cost nutrient-rich foods to oily and fatty products. How does diet quality respond to the other socio-demographic factors? Table 4.8 shows that the number of children positively affects diet quality, whereas adult males and females affect excess index in the opposite way. Having more female members reduces diet quality and having more male members improves diet quality in terms of EI. On the other hand, more male members can reduce diet quality in terms of the deficiency index. Households belonging to the Muslim community have a significantly low DI and a significantly high EI. The same is true for scheduled castes and scheduled tribes. It should be noted that these effects are significant after controlling for affluence level [log(realMPCE)], household head’s education, geographical effects (state dummies) and seasonal effects (sub-round dummies). The urban sector has better diet quality in DI and worse diet quality in EI. The better diet quality in urban sector is seemingly contradictory to the result depicted in figure 4.12. In figure 4.12 the rural sector has higher per capita calorie and protein intake. However, it should be noted that the diet quality index is a unweighted sum of all twelve nutrients - not just the major sources of energy. Therefore, the rural sector may have higher energy intake, but at the same time urban sector may have better diet quality since average urban diet is rich in other nutrients. Female-headed households have a significantly lower diet quality. 118 Table 4.8: Censored regression model of diet quality index DI EI 1983 2004-05 1983 2004-05 lmpce 588.4250*** 636.3060*** -160.8603*** -208.7728*** (15.4769) (10.7479) (22.7309) (9.6416) lmpce2 -44.1364*** -49.5816*** 4.8424** 12.7633*** (1.5933) (1.0124) (2.3102) (0.9364) children 1.4513*** 3.2767*** 2.2921*** 3.0137*** (0.2767) (0.2682) (0.1987) (0.1750) adult male -7.7513*** -7.1857*** 2.8104*** 2.8790*** (0.5081) (0.4643) (0.3317) (0.2631) adult female -1.2172** 0.2619 -2.5853*** -1.5950*** (0.6019) (0.4998) (0.3860) (0.2901) muslim -10.2394*** -10.4078*** 8.6318*** 5.6520*** (1.3511) (1.3561) (0.9821) (0.7744) scst -9.1544*** -10.3878*** 9.3207*** 8.1495*** (1.1100) (0.9474) (0.8679) (0.5415) head female -4.5044*** -2.5978* -1.9849* -2.6165*** (1.6568) (1.4611) (1.0542) (0.7856) head age -0.1604*** -0.0337 0.0611** -0.0288 (0.0352) (0.0353) (0.0245) (0.0197) urban 10.8423*** 14.2298*** -25.9077*** -19.1534*** (1.0661) (1.0051) (0.6877) (0.5479) head edu 2 -9.1032*** 0.3748 0.7982 -1.7147*** (1.2991) (0.9594) (0.9146) (0.5525) head edu 3 -8.1110*** -0.8799 -1.2662 -3.7118*** (1.2895) (1.5337) (0.8650) (0.8286) head edu 4 -4.4269*** -4.3022** -5.0952*** -3.8413*** (1.5089) (1.8031) (0.9686) (1.0440) head edu 5 -5.1697*** -4.6760*** -8.4309*** -4.4210*** (1.7279) (1.6846) (0.9703) (1.0300) N 101719 124486 101719 124486 Pseudo R2 0.0558 0.0524 0.1839 0.1479 F (47, N-47) 1155.58 1005.3 483.3 683.96 Note: Robust standard errors are in parenthesis. ∗p < 0.05, ∗ ∗ p < 0.01, ∗ ∗ ∗p < 0.001. Only selected coefficients are reported. Detail result is available upon request. 119 It is interesting to note that higher education of the head of household does not help improve diet quality. The apparent unjustifiable change in diet pattern in India is not unique though. There are two striking observations from the above analysis. The calorie intake is falling over time and the composition of energy sources itself is changing over time. Indian households are taking more fat based calorie instead of other cheap and healthy sources. These changes are irrespective of income levels. There are evidences of similar pattern elsewhere in the developing world. In contemporary China, per capita calorie consumption declined in the 1980s and 1990s. Du et al. (2002) observe that there are several nutritional transition period in modern China. Since 1985 total energy intake and energy expenditure decreased due to falling cereal consumption. Vegetables and fruits consumption also decreased during 1989 to 1997. It should be noted that this falling nutrients intake was preceded by rapid economic growth since 1979 reforms. The composition of calorie sources also changed during this period. The percentage of protein source in total energy has not changed, but that of fat sources has increased by almost 100%. Clark et al. (1995) find that there is a similar historical episode of stagnated or even declined consumption of food in spite of increased real wages during 1775 to 1850 in Britain. This apparent puzzle of food demand in Britain is partially resolved by the evidences of a sharp decline in food demand elasticity with income as income increases and decline in food demand as a result of urbanisation and change in occupational structure. Deaton and Dreze (2009) point out that there is no tight link between income and calorie consumption and also there is not tight link between calorie consumption and nutritional health status. Infact, during the period of decreased calorie intake in the 1980s and 1990s in China, there was rapid improvement in nutritional health indicators such as height and weight. However, the situation in India is not so favourable in terms of anthropometric measures. The height and weight indices for both adult and children are among the worst in the world. The rate of growth of these indicators are slower than what might be expected given the faster economic growth and poverty reduction in recent decades. Virtually there is no change in the proportions of underweight chindren from 1998 (47%) to 2005-06 (46%) according to the waves of National Family Health Surveys. Prevalence of stunting in children aged below 5 in India is even higher than the median of Sub-Saharan Africa (Poel et al., 2008). Recent research provides some evidences that lead us to assert that the rapid onset of overweight, obesity and diet-related non-communicable diseases (DRNCDs) in 120 developing countries are linked to the changing diet pattern (Popkin, 2004). 4.6 Conclusion There has been constant academic interest in the potential implications of the recent favourable changes in India. My study adds to this literature by analysing the changes in food preferences between 1983 and 2004-05, which covers the pre-reform and post- reform periods, and the nutritional implications of these changes. I found that during this period food share in total expenditure has decreased across all real expenditure groups. In other words, for any real monthly expenditure, households spend more on non-food items now compared to 25 years back. Moreover, it is empirically shown that households below the poverty line have a downward sloping Engel curve at a given time point. Social and economic transformations have led to changes in tastes and preferences, significantly influencing the composition of food demand and diet quality. These changes are visible in both the rural and urban sectors due to developments in communications and media, rapid urbanisation and the irresistible demonstration effect. The drop in expenditure share on food in general for all income groups is accompanied by a shift in food composition. While some parts of these changes could be attributed to visible changes in economic affluence and market conditions, taste and preference changes appear to be significant determinants of this shift. The income inelastic nature of cereals and pulses help explain the increase in food variety in diets. Moreover, expenditure elasticities of cereals and pulses decreased substantially in all sectors during this period. In the urban sector, while expenditure elasticities have dropped for almost all food groups, it has increased for fat-rich dairy products and remains almost the same for beverages, processed foods and other fatty foods. It is observed that the changes in food habits is at the expense of nutrient adequacy. There have been declining trends in per capita calorie intake in all sectors. Per capita protein intake in the rural sector decreased substantially and that in the urban sector shows no improvement in recent times. It is also observed that calorie and fat intake in the rural sector have been converging to that of the urban sector. On the other hand, per capita fat intake has an increasing trend in both sectors and there is no sign of convergence. The declining trend in per capita calorie consumption is accompanied by a downward shift in expenditure elasticity of calorie intake across all 121 income groups. Declining calorie elasticity poses a serious concern. It is widely believed that improvement in poverty will eliminate the malnutrition problem in India. However, lower calorie elasticity now compared to 25 years ago makes it harder to eliminate malnutrition with improving affluence levels. The diet quality in terms of deficiency of 12 nutrients has improved by only 4% from 1983 to 2004-05. However, diet quality in terms of excess fat intake has deteriorated by 8% and 16% in the rural and urban sectors, respectively. Diet quality depends on income as well as other socio-demographic characteristics. 122 5. Conclusion In this thesis, I have attempted to address some of the key distributional issues of the recent development experience in India. The major findings are summarised as follows. First, the tariff reforms have a pro-poor distributional impact in the rural sector and a pro-rich distributional impact in the urban sector. These differentiated effects are linked to the fact that labour market adjustment in the rural sector has resulted in relatively higher gains for the poor. I also find that all income groups have a positive and statistically significant welfare gain. This is contrary to the findings in Topalova (2007). The contradictory results in the literature demand further research on this question. Second, the period of rapid economic transformation since the early1980s has wit- nessed dramatic changes for historically disadvantaged groups. SC/STs have systemat- ically reduced the gap with non-SC/STs in education attainment levels and have been changing their occupation and industry of employment at increasingly faster rates. Moreover, the wage gap between SC/STs and non-SC/STs has narrowed sharply dur- ing this period. We have also found that most of the wage gap is accounted for by differences in education whose contribution has been rising over time. The caste effect on wages appears to have almost disappeared. Crucially, we find that these trends are the sharpest among the younger cohort and in urban areas. The last two features are especially uplifting since they are indicative of the types of changes one may expect in the future since India has been becoming increasingly urbanised and younger over time. Third, the shift in tastes and preferences has been a significant factor affecting the nutritional intake of the Indian population, particularly for the households below the poverty line. I found that from 1983 to 2004-05, food share in total expenditure has decreased significantly across all income groups. In other words, for any given real income level, households spend more on non-food items now compared to 25 years back. Social and economic transformations have led to changes in tastes and preferences, 123 significantly influencing the composition of food demand and diet quality. These changes are visible in both the rural and urban sectors due to developments in communications and media, rapid urbanisation and the irresistible demonstration effect. It is also found that there have been declining trends in per capita calorie intake in all sectors. Per capita protein intake in the rural sector decreased substantially and that in the urban sector shows no improvement in recent times. On the other hand, per capita fat intake has an increasing trend in both sectors. The diet quality index, measured as deficiency in 12 nutrients, has shown only a 4% improvement during this period, whereas the diet quality index, measured as excess of fat intake, has deteriorated by 8% and 16% in the rural and urban sectors, respectively. While positive gains from tariff reforms and the reduced gap between higher and lower castes are positive indications of human development, the falling nutritional status is a serious concern. Apart from deteriorating nutritional status, two important issues should be addressed by policymakers to avoid any increasing social gaps between the rich and the poor. 1) The urban poor do not have sufficient welfare premium from labour market adjustment compared to their rural counterparts and 2) education is the driving force to bridge the gap between social groups. 124 Bibliography Ahmed, N., Brzozowski, M., and Crossley, T. (2006). Measurement errors in recall food consumption data. IFS Working Papers W06/21, Institute for Fiscal Studies. Banerjee, B. and Knight, J. (1985). Caste discrimination in the Indian urban labour market. Journal of Development Economics, 17(3):277–307. Banks, J., Blundell, R., and Lewbel, A. (1997). Quadratic engel curves and consumer demand. The Review of Economics and Statistics, 79(4):527–539. Barten, A. P. (1964). Family composition, prices and expenditure patterns. In Hart, P. E., Mills, G., and Whittaker, J. K., editors, Econometric Analysis for National Economic Planning. Butterworth. Becker, G. and Tomes, N. (1986). Human capital and the rise and fall of families. Journal of Labor Economics, 4(3):1–39. Behrman, J. and Taubman, P. (1985). Intergenerational earnings mobility in the United States: some estimates and a test of Becker’s intergenerational endowments model. The Review of Economics and Statistics, 67(1):144–151. Behrman, J. R. and Deolalikar, A. B. (1987). Will developing country nutrition improve with income? a case study for rural south India. The Journal of Political Economy, 95(3):pp. 492–507. Bhagwati, J. N. and Srinivasan, T. N. (1975). Foreign Trade Regimes and Economic Development: India. National Bureau of Economic Research, Inc. Blundell, R., Pashardes, P., and Weber, G. (1993). What do we learn about consumer demand patterns from micro data? American Economic Review, 83(3):570–97. Borooah, V. K. (2005). Caste, inequality, and poverty in India. Review of Development Economics, 9(3):399–414. 125 Chattopadhyay, N., Majumder, A., and Coondoo, D. (2009). Demand threshold, zero expenditure and hierarchical model of consumer demand. Metroeconomica, 60(1):91– 118. Clark, G., Huberman, M., and Lindert, P. H. (1995). A british food puzzle, 1770-1850. The Economic History Review, 48(2):pp. 215–237. Coondoo, D. and Majumder, A. (1987). A system of demand equations based on price independent generalized linearity. International Economic Review, 28(1):213–28. Davidson, R. and MacKinnon, J. G. (1993). Estimation and Inference in Econometrics. Oxford University Press, New York. Davis, D. R. (1996). Trade liberalization and income distribution. Working Paper 5693, National Bureau of Economic Research. Deaton, A. (1987). Estimation of own- and cross-price elasticities from household survey data. Journal of Econometrics, 36(1-2):7–30. Deaton, A. (1997). The Analysis of Household Surveys – A Microeconometric Approach to Development Policy. World Bank. Deaton, A. and Dreze, J. (2009). Food and nutrition in India: Facts and interpretations. Economic and Political Weekly, 44(7):p. 42–65. Deaton, A. S. and Muellbauer, J. (1980). An almost ideal demand system. American Economic Review, 70(3):312–26. Dixit, A. and Norman, V. (1980). Theory of International Trade: A Dual, General Equi- librium Approach (Cambridge Economic Handbooks). Cambridge University Press. Du, S., Lu, B., Zhai, F., and Popkin, B. M. (2002). A new stage of the nutrition transition in China. Public Health Nutrition, 5(1a):169–174. Fan, J. (1992). Design-adaptive nonparametric regression. Journal of the American Statistical Association, 87(420):998–1004. Fortin, N. M. (2006). Greed, altruism, and the gender wage gap. Working papers, University of British Columbia. Goldberg, P. K. and Pavcnik, N. (2007). Distributional effects of globalization in devel- oping countries. Journal of Economic Literature, 45(1):pp. 39–82. 126 Gopalan, C., Sastri, B. R., and Balasubramanian., S. (1974). Nutritive value of Indian foods. National Inst. of Nutrition, Indian Council of Medical Research, Hyderabad, India. Gorman, W. M. (1981). Some engel curves. In Deaton, A., editor, The Theory and Measurement of Consumer Behaviour. Cambridge University Press, Cambridge, UK. Haider, S. and Solon, G. (2006). Life-cycle variation in the association between current and lifetime earnings. American Economic Review, 96(4):1308–1320. Hnatkovska, V. V., Lahiri, A., and Paul, S. B. (2010). Castes and Labor Mobility. SSRN eLibrary. ICMR (2009). Nutrient requirements and recommended dietary allowances for Indi- ans. A report of the expert group of the Indian council of medical research 2009, National Institute of Nutrition, Indian Council of Medical Research, Jamai-Osmania PO, Hyderabad, India. IIPS (2007). National family health survey (NFHS-3), 2005-06, India: Key findings. Technical report, International Institute for Population Sciences (IIPS) and Macro International, Mumbai. Ito, T. (2009). Caste discrimination and transaction costs in the labor market: Evidence from rural north India. Journal of Development Economics, 88(2):292–300. Jalan, J. and Murgai, R. (2009). Intergenerational mobility in education in India. Working papers, Indian Statistical Institute, Delhi. Jorgenson, D. W., Lau, L. J., and Stoker, T. M. (1982). The transcendental logarithmic model of aggregate consumer behavior. In Basmann, R. and Rhodes, G., editors, Advances in Econometrics. JAI Press, Greenwich, Greenwich. Kumar, U. and Mishra, P. (2008). Trade Liberalization and Wage Inequality: Evidence from India. Review of Development Economics, 12(2):291–311. Lee, C. and Solon, G. (2009). Trends in intergenerational income mobility. The Review of Economics and Statistics, 91(4):766–772. Madheswaran, S. and Attewell, P. (2007). Caste discrimination in the Indian urban labour market: Evidence from the national sample survey. Economic and Political Weekly, 42(41):4146. 127 Maitra, P. and Sharma, A. (2009). Parents and children: Education across generations in India. Working papers, Monash University, Department of Economics. Majumder, A. (1992). Measuring income responses: A log-quadratic demand model for consumers in India. Empirical Economics, 17(2):315–21. Meenakshi, J. V. and Ray, R. (1999). Regional differences in India’s food expenditure pattern: a complete demand systems approach. Journal of International Develop- ment, 11(1):47–74. Minhas, B. S. (1991). On estimating the inadequacy of energy intakes: Revealed food consumption behaviour versus nutritional norms (nutritional status of Indian people in 1983). Journal of Development Studies, 28(1):1–38. Muellbauer, J. (1976). Community preferences and the representative consumer. Econo- metrica, 44(5):979–99. Munshi, K. (2010). Strength in numbers: Networks as a solution to occupational traps. Review of Economic Studies, forthcoming. Munshi, K. and Rosenzweig, M. (2006). Traditional institutions meet the modern world: Caste, gender, and schooling choice in a globalizing economy. American Economic Review, 96(4):1225–1252. Munshi, K. and Rosenzweig, M. (2009). Why is mobility in India so low? social insurance, inequality, and growth. Working papers, Brown University, Department of Economics. NSSO (2007). Nutritional intake in India 2004-05. NSS report 513(61/1/0/6), National Sample Survey Organisation, Ministry of Statistics & Programme Implementation, Government of India, New Delhi, India. Pal, P. and Ghosh, J. (2007). Inequality in India: A survey of recent trends. DESA Working Paper 45, Department of Economic and Social Affairs, United Nations. Panagariya, A. (2004). India’s Trade Reform: Progress, Impact and Future Strategy . Unpublished working paper, Columbia University. Poel, E. V. d., Hosseinpoor, A. R., Speybroeck, N., Ourti, T. V., and Vega, J. (2008). Socioeconomic inequality in malnutrition in developing countries. Bulletin of the World Health Organization, 86(4):pp. 241–320. 128 Poi, B. P. (2002). From the help desk: Demand system estimation. Stata Journal, 2(4):403–410. Poi, B. P. (2008). Demand-system estimation: Update. Stata Journal, 8(4):554–556(3). Popkin, B. M. (2004). The nutrition transition: An overview of world patterns of change. Nutrition Reviews, 62(7):pp. S140S143. Porto, G. G. (2006). Using survey data to assess the distributional effects of trade policy. Journal of International Economics, 70(1):140–160. Prakash, N. (2009). The impact of employment quotas on the economic lives of dis- advantaged minorities in India. Working papers, Dartmouth College, Department of Economics. Ray, R. (1980). Analysis of a time series of household expenditure surveys for India. The Review of Economics and Statistics, 62(4):pp. 595–602. Ray, R. (1985). A dynamic analysis of expenditure patterns in rural India. Journal of Development Economics, 19(3):283–297. Solon, G. (1992). Intergenerational income mobility in the United States. The American Economic Review, 82(3):393–408. Solon, G. (2002). Cross-country differences in intergenerational earnings mobility. The Journal of Economic Perspectives, 16(3):59–66. Subramanian, S. and Deaton, A. (1996). The demand for food and calories. The Journal of Political Economy, 104(1):pp. 133–162. Thiele, S., Mensink, G. B., and Beitz, R. (2004). Determinants of diet quality. Public Health Nutrition, 7(1):29–37. Topalova, P. (2007). Trade liberalization, poverty and inequality: Evidence from Indian districts. In Harrison, A., editor, Globalization and Poverty. National Bureau of Economic Research, Inc. United Nations Conference on Trade and Development(UNCTAD) (2009). TRade Anal- ysis and INformations System. World Integrated Trade Solution (WITS) software. published by United Nations. 129 Woodland, A. D. (1982). International Trade and Resource Allocation (Advanced Text- books in Economics). Elsevier Science Ltd. 130 Appendices Appendix A: Appendix to chapter 3 Data appendix National Sample Survey (NSS) The National Sample Survey Organization (NSSO), set up by the Government of India, conducts rounds of sample surveys to collect socioeconomic data. Each round is ear- marked for particular subject coverage. We use the latest five large quinquennial rounds – 38(Jan-Dec 1983), 43(July 1987-June 1988), 50(July 1993-June 1994), 55(July 1999- June 2000) and 61(July 2004-June 2005) on Employment and Unemployment (Schedule 10). The survey covers the whole country except for a few remote and inaccessible pock- ets. The NSS follows multi-stage stratified sampling with villages or urban blocks as first stage units (FSU) and households as ultimate stage units. The field work in each round is conducted in several sub-rounds throughout the year so that seasonality is min- imized. The sampling frame for the first stage unit is the list of villages (rural sector) or the NSS Urban Frame Survey blocks (urban sector) from the latest available survey. We describe the broad outline of sample design – stratification, allocation and selection of sample units - with a caveat that the details have changed from round to round. The whole country is divided politically into states and union territories, and each state is further divided into districts for administrative purpose. The NSSO also con- structs regions by grouping contiguous districts within a state which are similar in population density and crop pattern for the sampling purpose. Two different stratifica- tion methods are used for rural and urban sector in each state. In the rural sector, each district is generally counted as a separate stratum (populous districts are split into two or more strata) whereas in the urban sector, strata are formed within the NSS region based on population size of cities. For example, all towns with population less than 50,000 in a region will form stratum 1 and so on. In the 61st round, the stratification method was changed substantially. For this round, each district is divided into two basic strata – rural and urban. Then the rural and urban strata are further divided 131 into sub-strata. The total sample size of first stage unit (villages/urban blocks) is allocated to the states and union territories in proportion to population. The subsequent allocations to rural and urban sector and at stratum level within a state are based on population size as well. In rural sectors, sample FSUs are selected with probability proportional to population from each stratum (sub-stratum for 61st round). In urban sectors, they are selected by simple random sampling without replacement in 38th and 61st round and circular systematic sampling with equal probability in the 43rd, 50th and 55th round. Within each stratum (sub-stratum for 61st round), samples are drawn in the form of two independent sub-samples for both rural and urban sectors. Once the FSUs are randomly drawn, the large FSUs are subdivided into certain number of parts (hamlet-group/sub- block) with approximately equal population and one of them selected randomly for listing of households. Complex second stage stratification based on “means of liveli- hood class” is implemented to select households randomly from the sample frame of households in each FSU (or hamlet-group/sub-block). As the sample design changes over the rounds, estimation without considering the complex design may be misleading. The NSSO supplies household level multipliers with the unit record data for each round to help minimize estimation errors on the part of researchers. The questionnaire collects demographic details like age, sex, marital status, education, etc. and information about occupation, industry, activity, time disposition in reference week, wage, etc. of household members. It also collects monthly total household expenditure along with other household level characteristics. The data are given in fixed format text files with a list of variable names and byte positions. We have checked the validity of our data extraction process by comparing the statistics on a number of the variables with numbers reported in published works by other authors. However, there is some miscoding which is typical for any survey data and we tried our best to clean it. Other notable changes over the rounds are formation of new states, deletion of the social group called “Neo-Buddhist” and formation of new social group called “Other Backward Class” or “OBC” (see below), and changes in coding for education, enrolment in educational institution, activity status and industry. We recoded all these changes to make it uniform and consistent over the time. Sample selection We drop all households for which we have no information on social group or whose social group is miscoded (3/ 120706 households in 38th round, 43/ 129060 households in 43rd 132 round, none for 50th and 55th rounds (115409 and 120386 households, respectively), and 86/124680 households for 61st round are dropped). The classification of Scheduled Castes (SC) and Scheduled Tribes (ST) groups remain unchanged over the rounds. However, there is a new classification of “Other Backward Classes” (OBC) from the 55th round while the “Neo-Buddhist” classification was discontinued from the 50th round. We club these groups with non-SC/ST so that the scheduled caste and scheduled tribe groups (SC/ST) remain uniform throughout the period. In our data work, we only consider individuals that report their 3-digit occupation code and education attainment level. Occupation codes are drawn from the National Classification of Occupation (NCO) – 1968. We use the “usual” occupation code re- ported by an individual for the usual principal activity over the previous year (relative to the survey year). The dataset does not contain information on the years of schooling for the individuals. Instead it includes information on general education categories given as (i) not literate -01, literate without formal schooling: EGS/ NFEC/ AEC -02, TLC -03, others -04; (ii) literate: below primary -05, primary -06, middle -07, secondary -08, higher secondary -10, diploma/certificate course -11, graduate -12, postgraduate and above -13. We aggregate those into five similarly sized groups as discussed in the main text. We are also interested in studying the patterns of industry employment for dif- ferent social groups. We employ 5-digit National Industry Classification (NIC) – 1998 industry code that is reported for each individual over the previous year (relative to the survey year). In our analysis we dedicate a lot of attention to studying wage dynamics. NSS only reports wages from activities undertaken by an individual over the previous week (relative to the survey week). Household members can undertake more than one activity in the reference week. For each activity we know the “weekly” occupation code, number of days spent working in that activity, and wage received from it. We identify the main activity for the individual as the one in which he spent maximum number of days in a week. If there are more than one activities with equal days worked, we consider the one with paid employment (wage is not zero or missing). Workers sometimes change the occupation due to seasonality or for other reasons. To minimize the effect of transitory occupations, we only consider wages for which the weekly occupation code coincides with usual occupation (one year reference). We calculate the daily wage by dividing total wage paid in that activity over the past week by days spent in that activity. Lastly, we identify full time workers in our dataset. We assume that an individual is a full time worker if he is employed (based on daily status code) for at least two and 133 half days combined in all activities during the reference week.33 We drop observations if total number of days worked in the reference week is more than seven. To summarize, our working sample imposes the following restrictions on the data: 1) The overall sub-sample includes all households with a male head of household in the 16-65 age group with at least one other directly related male member of a younger generation (son or grandson) also in the 16-65 age group, where neither is enrolled in an educational institution, both have education and occupation information and are working full-time. Within included households, we only consider the head of the household and his direct male descendants. 2) The wage sub-sample includes only those households from the overall sample for which wage data for head and at least one of his descendants are non-missing and non-zero. The working sample is further subdivided into two generational groups – children and parents. Only household heads are considered as parents in our analysis. Any members from younger generations are considered as children (therefore it includes grandchildren). Occupation categories Table A.1 summarizes the one-digit occupation categories in our dataset and presents our grouping of these categories into the Occ 1 - “white collar”, Occ 2 - “blue collar” and Occ 3 - “agriculture” groups that we used in the text. Table A.1: Occupation categories Occupation code Occupation description Group 0-1 Professional, technical and related workers Occ 1 2 Administrative, executive and managerial workers Occ 1 3 Clerical and related workers Occ 1 4 Sales workers Occ 2 5 Service workers Occ 2 6 Farmers, fishermen, hunters, loggers and related workers Occ 3 7-8-9 Production and related workers, transport equipment operators and labourers Occ 2 Table A.2 summarizes one-digit industry codes in our dataset. In the presentation in the text we group these codes further into three broad industry categories: Ind 1 refers to Agriculture, Hunting, Forestry and Fishing; Ind 2 collects all tradable industries; 33Based on daily status code we can classify all individuals into employed, unemployed and not in labor force. 134 while Ind 3 refers to all non-tradable industries. These groupings are detailed in Table A.2. Table A.2: Industry categories Industry code Industry description Group A Agriculture, Hunting and Forestry Ind 1 B Fishing Ind 1 C Mining and Quarrying Ind 2 D Manufacturing Ind 2 E Electricity, Gas and Water Supply Ind 3 F Construction Ind 3 G Wholesale and Retail Trade; Repair of Motor Vehicles, Ind 2 motorcycles and personal and household goods H Hotels and Restaurants Ind 2 I Transport, Storage and Communications Ind 2 J Financial Intermediation Ind 3 K Real Estate, Renting and Business Activities Ind 3 L Public Administration and Defence; Compulsory Social Security Ind 3 M Education Ind 3 N Health and Social Work Ind 3 O Other Community, Social and Personal Service Activities Ind 3 P Private Households with Employed Persons Ind 3 Q Extra Territorial Organizations and Bodies Ind 3 Intergenerational education mobility Table A.3 presents average conditional probabilities of education improvements (panel (a)) and education reductions (panel (b)) for the overall sample and separately for non- SC/STs and SC/STs over different survey rounds. These probabilities were estimated following the procedure we used to obtain average conditional probabilities of education switches, which is described in details in the main text. 135 Table A.3: Intergenerational education improvements and reductions (a) education improvements (b) education reductions overall non-SC/STs SC/STs overall non-SC/STs SC/STs 1983 0.4557 0.4874 0.3552 0.0863 0.0915 0.0700 (0.0008) (0.0008) (0.0011) (0.0003) (0.0003) (0.0004) 1987-88 0.4684 0.5014 0.3676 0.0915 0.0949 0.0811 (0.0007) (0.0007) (0.001) (0.0002) (0.0003) (0.0004) 1993-94 0.5234 0.5449 0.4621 0.0827 0.0885 0.0664 (0.0006) (0.0006) (0.0008) (0.0002) (0.0003) (0.0004) 1999-00 0.5363 0.5488 0.5035 0.0900 0.0951 0.0767 (0.0006) (0.0007) (0.0012) (0.0003) (0.0003) (0.0004) 2004-05 0.5806 0.5779 0.5880 0.0884 0.0921 0.0785 (0.0006) (0.0007) (0.0011) (0.0003) (0.0004) (0.0005) Notes: This table presents average probabilities of education improvements (Panel (a)) and education reductions (Panel (b)) for the overall sample and separately for SC/STs and non-SC/STs. These probabilities were estimated using equation (3.1), except we used a binary variable denoting education improvements or education re- ductions as the left-hand-side variable. Standard errors are in parenthesis. 136 Appendix B: Appendix to chapter 4 Table B.1: Frequency distribution (%) of households classified by food share 1983 2004-05 Rural Urban Rural Urban Share Freq. Cum. Freq. Freq. Cum. Freq. Freq. Cum. Freq. Freq. Cum. Freq. 0.05 0.01 0.01 0.06 0.06 0.03 0.03 0.08 0.08 0.10 0.02 0.03 0.07 0.13 0.09 0.12 0.27 0.35 0.15 0.02 0.06 0.07 0.20 0.20 0.32 0.49 0.84 0.20 0.04 0.09 0.15 0.35 0.35 0.67 1.27 2.11 0.25 0.07 0.16 0.18 0.53 0.67 1.34 2.63 4.74 0.30 0.13 0.29 0.46 0.99 1.40 2.74 5.33 10.07 0.35 0.25 0.54 1.03 2.03 2.56 5.30 8.32 18.39 0.40 0.47 1.01 2.00 4.03 4.73 10.03 11.51 29.91 0.45 0.90 1.91 3.53 7.55 8.30 18.33 14.00 43.91 0.50 1.96 3.87 6.43 13.98 13.07 31.40 15.28 59.19 0.55 4.15 8.03 9.96 23.94 17.71 49.11 14.83 74.01 0.60 8.45 16.47 14.45 38.39 19.56 68.67 11.91 85.92 0.65 13.97 30.44 18.03 56.42 16.29 84.96 7.78 93.70 0.70 18.91 49.35 17.68 74.10 9.86 94.83 3.94 97.64 0.75 20.37 69.72 13.31 87.41 3.92 98.74 1.51 99.15 0.80 17.14 86.86 7.75 95.16 0.95 99.69 0.48 99.64 0.85 9.51 96.37 3.28 98.44 0.21 99.90 0.23 99.87 0.90 3.06 99.43 1.13 99.57 0.07 99.97 0.10 99.97 0.95 0.49 99.92 0.27 99.83 0.02 99.99 0.03 100.00 1.00 0.08 100.00 0.17 100.00 0.01 100.00 0.00 100.00 137 Table B.2: Food share by selected states in top and bottom quartile - rural Rural 1983 2004-05 State Overall Poorest 25% Richest 25% Overall Poorest 25% Richest 25% Jammu & Kashmir 0.71 0.72 0.68 0.56 0.61 0.50 Himachal Pradesh 0.66 0.67 0.63 0.52 0.57 0.43 Punjab 0.64 0.64 0.63 0.49 0.56 0.39 Haryana 0.65 0.65 0.66 0.51 0.55 0.44 Rajasthan 0.67 0.68 0.65 0.55 0.57 0.51 Uttar Pradesh 0.68 0.69 0.66 0.55 0.59 0.49 Bihar 0.75 0.75 0.75 0.61 0.64 0.58 Manipur 0.75 0.78 0.72 0.53 0.61 0.44 Mizoram 0.67 0.70 0.63 0.54 0.58 0.46 Meghalaya 0.72 0.81 0.61 0.53 0.55 0.50 Assam 0.74 0.75 0.70 0.63 0.67 0.59 West Bengal 0.76 0.77 0.72 0.60 0.65 0.52 Orissa 0.76 0.76 0.73 0.60 0.64 0.52 Madhya Pradesh 0.70 0.72 0.67 0.53 0.57 0.48 Gujarat 0.67 0.70 0.63 0.56 0.61 0.49 Maharashtra 0.65 0.66 0.62 0.52 0.57 0.45 Andhra Pradesh 0.67 0.71 0.62 0.55 0.61 0.47 Karnataka 0.67 0.68 0.64 0.53 0.57 0.47 Goaa 0.64 0.72 0.57 0.49 0.57 0.41 Kerala 0.68 0.70 0.64 0.49 0.55 0.39 Tamil Nadu 0.69 0.71 0.65 0.55 0.58 0.48 Pondicherry 0.67 0.70 0.65 0.48 0.55 0.42 A & N Islands 0.71 0.74 0.66 0.53 0.59 0.46 aGoa, Daman & Diu 138 Table B.3: Food share by selected states in top and bottom quartile - urban Urban 1983 2004-05 State Overall Poorest 25% Richest 25% Overall Poorest 25% Richest 25% Jammu & Kashmir 0.63 0.66 0.58 0.50 0.58 0.41 Himachal Pradesh 0.58 0.62 0.50 0.44 0.53 0.33 Punjab 0.58 0.61 0.53 0.42 0.52 0.32 Chandigarh 0.60 0.68 0.52 0.39 0.54 0.27 Haryana 0.59 0.63 0.51 0.44 0.51 0.35 Delhi 0.58 0.64 0.50 0.44 0.54 0.32 Rajasthan 0.63 0.66 0.59 0.47 0.55 0.36 Uttar Pradesh 0.64 0.68 0.57 0.49 0.58 0.39 Bihar 0.70 0.74 0.63 0.53 0.63 0.41 Sikkim 0.62 0.66 0.54 0.41 0.46 0.36 Arunachal Pradesh 0.74 0.74 0.68 0.50 0.54 0.46 Nagaland 0.67 0.72 0.65 0.45 0.53 0.37 Manipur 0.74 0.76 0.69 0.48 0.55 0.39 Mizoram 0.59 0.63 0.56 0.45 0.52 0.38 Tripura 0.68 0.73 0.60 0.53 0.63 0.41 Meghalaya 0.59 0.65 0.54 0.40 0.46 0.33 Assam 0.68 0.75 0.60 0.50 0.60 0.41 West Bengal 0.64 0.70 0.56 0.48 0.61 0.34 Orissa 0.68 0.69 0.63 0.52 0.63 0.42 Madhya Pradesh 0.63 0.67 0.56 0.46 0.55 0.34 Gujarat 0.62 0.67 0.56 0.46 0.55 0.35 Maharashtra 0.60 0.64 0.54 0.43 0.54 0.31 Andhra Pradesh 0.62 0.69 0.53 0.43 0.53 0.31 Karnataka 0.61 0.67 0.54 0.43 0.54 0.31 Goaa 0.62 0.76 0.50 0.45 0.52 0.35 Lakshdweep 0.56 0.67 0.59 0.53 0.60 0.46 Kerala 0.66 0.69 0.58 0.45 0.54 0.33 Tamil Nadu 0.64 0.70 0.54 0.45 0.55 0.33 Pondicherry 0.67 0.72 0.60 0.46 0.54 0.36 A & N Islands 0.62 0.67 0.54 0.43 0.50 0.32 aGoa, Daman & Diu 139 Table B.4: Mean and SD of food group shares 1983 2004-05 Absolute Change Relative Change(%) Rural Urban Rural Urban Rural Urban Rural Urban Food groups Mean SD Mean SD Mean SD Mean SD Mean Mean Mean Mean Cereals 0.530 0.181 0.368 0.167 0.361 0.129 0.274 0.107 -0.169 -0.095 -31.840 -25.705 Cereal substitutes 0.003 0.019 0.001 0.007 0.001 0.007 0.001 0.006 -0.002 0.000 -64.202 -13.408 Pulses 0.059 0.047 0.059 0.035 0.061 0.031 0.058 0.026 0.002 -0.002 4.149 -2.564 Dairy 0.093 0.120 0.137 0.114 0.133 0.131 0.175 0.114 0.040 0.038 43.106 27.562 Edible oil 0.060 0.037 0.081 0.044 0.087 0.034 0.088 0.034 0.028 0.007 46.600 8.220 Meat, egg and fish 0.041 0.056 0.059 0.068 0.055 0.066 0.063 0.071 0.014 0.003 35.132 5.764 Vegetables 0.074 0.043 0.086 0.047 0.117 0.045 0.112 0.044 0.043 0.026 58.476 30.019 Fresh fruits 0.014 0.029 0.025 0.033 0.023 0.029 0.035 0.032 0.009 0.010 61.334 38.815 Dry fruits 0.003 0.011 0.005 0.014 0.006 0.013 0.010 0.017 0.003 0.005 86.946 89.263 Sugar 0.040 0.036 0.042 0.027 0.044 0.028 0.039 0.021 0.003 -0.003 8.658 -6.282 Salt 0.003 0.004 0.002 0.002 0.004 0.003 0.004 0.002 0.001 0.001 30.779 61.292 Spices 0.038 0.024 0.038 0.026 0.044 0.020 0.040 0.018 0.006 0.002 15.017 6.178 Beverages 0.042 0.074 0.096 0.134 0.064 0.068 0.103 0.115 0.022 0.007 51.344 7.727 140 Table B.5: Multivariate model of food share 1983 2004-05 Rural Urban All Rural Urban All lmpce 0.255*** 0.361*** 0.288*** 0.324*** -0.012 0.278*** (12.72) (8.51) (16.72) (14.28) (-0.60) (17.33) lmpce2 -0.031*** -0.045*** -0.036*** -0.040*** -0.010*** -0.038*** (-14.52) (-10.85) (-20.34) (-19.41) (-5.81) (-26.09) adult males -0.003 -0.018 -0.011* 0.015* 0.024** 0.013* (-0.52) (-1.55) (-2.02) (2.20) (2.94) (2.51) adult females 0.003 0.027 0.012 0.021** 0.021** 0.023*** (0.41) (1.62) (1.73) (2.84) (2.69) (4.14) children 0.019*** 0.034*** 0.012*** 0.031*** 0.012* 0.018*** (4.97) (4.63) (3.52) (6.82) (2.08) (4.85) adult male*lmpce 0.000 0.003 0.002 -0.003* -0.006*** -0.003** (0.04) (1.36) (1.49) (-2.23) (-3.77) (-2.79) adult female*lmpce 0.001 -0.007* -0.001 -0.004** -0.006*** -0.005*** (0.36) (-2.13) (-1.00) (-2.72) (-4.52) (-4.31) children*lmpce -0.003*** -0.008*** -0.002* -0.007*** -0.004*** -0.004*** (-3.69) (-5.56) (-2.50) (-7.14) (-3.92) (-5.13) adult male2 0.001** -0.000 0.000 0.000 0.001 0.000* (2.92) (-0.08) (1.63) (1.91) (1.31) (2.49) adult female2 -0.001 0.000 -0.001** 0.000 0.001** -0.000 (-1.43) (0.19) (-2.84) (0.93) (2.93) (-0.77) children2 -0.000 0.000 -0.000 0.000*** 0.000 0.000** (-0.30) (1.20) (-0.92) (3.68) (0.82) (2.87) adult male*children -0.001* 0.001 -0.000 -0.000 0.001 -0.000 (-2.44) (1.26) (-0.62) (-1.81) (1.44) (-1.32) adult female*children -0.000 0.001 0.000 -0.001* 0.001 0.000 (-0.13) (0.99) (1.86) (-2.26) (1.62) (0.01) Sub-round dummy Yes Yes Yes Yes Yes Yes State dummy Yes Yes Yes Yes Yes Yes R-sqr 0.204 0.271 0.223 0.337 0.555 0.407 RMSE 0.088 0.095 0.092 0.084 0.080 0.090 N 72404 29785 102189 79253 45288 124541 Note: t-values are in parenthesis. ∗p < 0.05, ∗ ∗ p < 0.01, ∗ ∗ ∗p < 0.001 141 Table B.6: Multivariate model of food share with region dummies 1983 2004-05 Rural Urban All Rural Urban All lmpce 0.251*** 0.349*** 0.282*** 0.309*** -0.020 0.274*** (12.58) (8.43) (16.29) (13.50) (-0.95) (17.14) lmpce2 -0.030*** -0.044*** -0.035*** -0.039*** -0.010*** -0.037*** (-14.34) (-10.88) (-19.91) (-18.61) (-5.29) (-25.73) adult males -0.002 -0.019 -0.010 0.014* 0.023** 0.012* (-0.34) (-1.65) (-1.79) (2.14) (2.89) (2.37) adult females 0.003 0.026 0.011 0.021** 0.022** 0.024*** (0.35) (1.56) (1.70) (2.84) (2.91) (4.32) children 0.019*** 0.034*** 0.012*** 0.030*** 0.011 0.018*** (5.21) (4.66) (3.80) (6.53) (1.90) (5.04) adult male*lmpce 0.000 0.003 0.002 -0.003* -0.005*** -0.003** (0.12) (1.46) (1.53) (-2.16) (-3.69) (-2.62) adult female*lmpce 0.000 -0.007* -0.001 -0.004** -0.007*** -0.005*** (0.31) (-2.06) (-1.05) (-2.81) (-4.81) (-4.63) children*lmpce -0.003*** -0.009*** -0.002** -0.006*** -0.004*** -0.004*** (-3.74) (-5.71) (-2.63) (-6.91) (-3.80) (-5.35) adult male2 0.001* 0.000 0.000 0.000 0.001 0.000* (2.14) (0.01) (1.04) (1.76) (1.34) (2.25) adult female2 -0.001 0.000 -0.001** 0.000 0.001** 0.000 (-1.24) (0.09) (-2.65) (1.35) (3.08) (0.02) children2 -0.000 0.000 -0.000 0.000*** 0.000 0.000** (-0.38) (1.23) (-0.79) (3.52) (0.87) (2.69) adult male*children -0.001* 0.001 -0.000 -0.000 0.001 -0.000 (-2.10) (1.24) (-0.65) (-1.44) (1.45) (-0.83) adult female*children -0.000 0.001 0.000 -0.001* 0.001 -0.000 (-0.97) (1.26) (1.14) (-2.48) (1.64) (-0.61) Sub-round dummy Yes Yes Yes Yes Yes Yes Region dummy Yes Yes Yes Yes Yes Yes R-sqr 0.244 0.290 0.256 0.355 0.567 0.425 RMSE 0.086 0.094 0.091 0.083 0.079 0.088 N 72404 29785 102189 79253 45288 124541 Note: t-values are in parenthesis. ∗p < 0.05, ∗ ∗ p < 0.01, ∗ ∗ ∗p < 0.001 142 . 4 . 5 . 6 . 7 . 8 Fo od ha re 3 4 5 6 7 ln(realMPCE) Household size 1 Household size 2 Household size 3 Household size 4 Household size 5 Household size 6 Household size 7 Household size 8 Rural − 1983 Foodshare and MPCE for households of different sizes (a) Rural . 2 . 4 . 6 . 8 Fo od ha re 3 4 5 6 7 ln(realMPCE) Household size 1 Household size 2 Household size 3 Household size 4 Household size 5 Household size 6 Household size 7 Household size 8 Rural − 2004−05 Foodshare and MPCE for households of different sizes (b) Rural . 3 . 4 . 5 . 6 . 7 Fo od ha re 3 4 5 6 7 ln(realMPCE) Household size 1 Household size 2 Household size 3 Household size 4 Household size 5 Household size 6 Household size 7 Household size 8 Urban − 1983 Foodshare and MPCE for households of different sizes (c) Urban . 2 . 4 . 6 . 8 Fo od ha re 3 4 5 6 7 ln(realMPCE) Household size 1 Household size 2 Household size 3 Household size 4 Household size 5 Household size 6 Household size 7 Household size 8 Urban − 2004−05 Foodshare and MPCE for households of different sizes (d) Urban . 4 . 5 . 6 . 7 . 8 Fo od ha re 3 4 5 6 7 ln(realMPCE) Children 0 Children 1 Children 2 Children 3 Children 4 Rural − 1983 Foodshare and MPCE for households of different number of children (e) Rural 0 . 2 . 4 . 6 . 8 Fo od ha re 3 4 5 6 7 ln(realMPCE) Children 0 Children 1 Children 2 Children 3 Children 4 Rural − 2004−05 Foodshare and MPCE for households of different number of children (f) Rural 0 . 2 . 4 . 6 . 8 Fo od ha re 3 4 5 6 7 ln(realMPCE) Children 0 Children 1 Children 2 Children 3 Children 4 Urban − 1983 Foodshare and MPCE for households of different number of children (g) Urban . 2 . 3 . 4 . 5 . 6 . 7 Fo od ha re 3 4 5 6 7 ln(realMPCE) Children 0 Children 1 Children 2 Children 3 Children 4 Urban − 2004−05 Foodshare and MPCE for households of different number of children (h) Urban Figure B.1: Local regression estimate of Engel curve for varying household size compo- sition. 143 Table B.7: Mean and SD of food group quantity (per capita, per month) 1983 2004-05 Rural Urban Rural Urban Food group Mean SD Mean SD Mean SD Mean SD Cereal (kg) 15.10 6.88 11.25 4.42 12.02 3.60 10.15 2.94 Pulse (Kg) 1.01 0.96 1.06 0.69 0.72 0.54 0.83 0.43 Milk (Kg) 3.53 4.14 4.28 3.96 3.49 3.79 4.61 3.82 Meat and fish (Kg) 0.68 0.90 0.79 1.06 0.73 2.50 0.79 1.05 Egg (no.) 2.70 4.96 4.28 7.65 3.23 2.64 4.10 3.74 Vegetables (Kg) 4.20 3.75 4.59 3.76 5.50 5.76 5.86 8.65 Table B.8: Number of households with non-zero share Number of households 1983 2004-05 Food group Items Rural Urban Rural Urban cereal all Cereal, Cereal substitute 71716 28131 78758 43574 pulse Pulses and Pulse products 66384 27480 77223 43168 milk Milk and milk products 45659 24277 61019 38712 mef Meat, egg and fish 39990 17929 50671 28439 vegfruit Fruits and vegetables 71649 28561 78966 44543 other Beverages, refreshment and processed foods, spices, salt, sugar, pan(betel leaf), edi- ble oil, tobacco and intoxi- cants 72368 29760 79223 45255 144 0 . 1 . 2 . 3 m ilk  sh are 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Rural milk share and total expenditure in food (a) Dairy - Rural . 05 . 1 . 15 . 2 . 25 m ilk  sh are 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Urban milk share and total expenditure in food (b) Dairy - Urban 0 . 05 . 1 . 15 m ef sh are 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Rural mef share and total expenditure in food (c) Meat, Egg and Fish - Rural 0 . 05 . 1 . 15 m ef sh are 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Urban mef share and total expenditure in food (d) Meat, Egg and Fish - Urban . 08 . 1 . 12 . 14 . 16 . 18 ve gf ru it sh are 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Rural vegfruit share and total expenditure in food (e) Fruits and Vegetables - Rural . 1 . 12 . 14 . 16 . 18 ve gf ru it sh are 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Urban vegfruit share and total expenditure in food (f) Fruits and Vegetables - Urban . 1 . 2 . 3 . 4 . 5 ot he r s ha re 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Rural other share and total expenditure in food (g) Others - Rural . 2 . 3 . 4 . 5 . 6 . 7 ot he r s ha re 3 4 5 6 ln(real percapita food expd.) 1983 2004−05 Urban other share and total expenditure in food (h) Others - Urban Figure B.2: Non-parametric and quadratic Engel curves for food groups 145 Table B.9: Estimation of QAIDS model Rural Urban Parameters 1983 2004-05 Difference 1983 2004-05 Difference α1 0.437*** 0.197*** -0.240*** 0.229*** 0.0877*** -0.142*** (98.51) (51.97) (-41.03) (27.99) (17.79) (-14.82) α2 0.0281*** 0.0250*** -0.00310 0.0370*** 0.0282*** -0.00876*** (22.83) (21.32) (-1.83) (27.12) (25.11) (-4.96) α3 0.158*** 0.294*** 0.136*** 0.197*** 0.285*** 0.0887*** (57.45) (75.44) (28.57) (43.85) (59.30) (13.49) α4 0.0625*** 0.0740*** 0.0114*** 0.0783*** 0.0769*** -0.00139 (38.26) (45.38) (4.96) (33.19) (27.22) (-0.38) α5 0.0920*** 0.131*** 0.0394*** 0.132*** 0.162*** 0.0298*** (76.29) (95.89) (21.56) (48.64) (68.84) (8.30) α6 0.223*** 0.278*** 0.0555*** 0.326*** 0.360*** 0.0334*** (99.39) (98.61) (15.42) (72.51) (69.82) (4.88) β1 -0.175*** -0.202*** -0.0267*** -0.169*** -0.179*** -0.0105 (-46.91) (-53.75) (-5.04) (-13.38) (-39.48) (-0.78) β2 0.00129 -0.0169*** -0.0182*** -0.00758*** -0.0225*** -0.0149*** (1.64) (-20.96) (-16.15) (-3.91) (-23.47) (-6.88) β3 0.117*** 0.220*** 0.104*** 0.0965*** 0.144*** 0.0477*** (49.47) (46.59) (19.61) (12.96) (30.43) (5.41) β4 0.0168*** 0.0131*** -0.00374 0.0336*** 0.0257*** -0.00789 (12.66) (9.03) (-1.90) (10.72) (9.11) (-1.87) β5 0.00388*** -0.0146*** -0.0184*** 0.0237*** 0.0000473 -0.0236*** (3.63) (-7.99) (-8.73) (5.44) (0.02) (-4.59) β6 0.0364*** -0.0000663 -0.0364*** 0.0224*** 0.0315*** 0.00914 (19.07) (-0.02) (-10.87) (5.08) (7.33) (1.48) γ11 0.171*** 0.125*** -0.0458*** 0.158*** 0.0806*** -0.0771*** (60.40) (34.90) (-10.04) (27.45) (19.10) (-10.82) γ12 -0.0359*** -0.0152*** 0.0206*** -0.0207*** -0.00177* 0.0190*** (-44.76) (-20.22) (18.77) (-14.43) (-2.08) (11.34) γ13 -0.0770*** -0.0786*** -0.00161 -0.0802*** -0.0643*** 0.0159*** (-63.83) (-34.54) (-0.62) (-31.65) (-27.15) (4.58) γ14 0.00241*** -0.00409*** -0.00650*** -0.00705*** -0.00652*** 0.000532 (3.81) (-5.74) (-6.82) (-4.50) (-5.95) (0.28) γ15 -0.0202*** -0.0281*** -0.00795*** -0.0436*** -0.0345*** 0.00914*** (-31.49) (-29.73) (-6.96) (-28.88) (-24.85) (4.46) γ16 -0.0399*** 0.00126 0.0412*** -0.00608 0.0265*** 0.0326*** (-31.78) (0.76) (19.82) (-1.81) (9.02) (7.30) Continued on next page 146 Table B.9 – continued from previous page Rural Urban Parameters 1983 2004-05 Difference 1983 2004-05 Difference γ21 -0.0359*** -0.0152*** 0.0206*** -0.0207*** -0.00177* 0.0190*** (-44.76) (-20.22) (18.77) (-14.43) (-2.08) (11.34) γ22 0.00647*** 0.00881*** 0.00234* 0.0106*** 0.0116*** 0.00104 (7.12) (11.80) (1.99) (7.19) (13.04) (0.60) γ23 0.00182*** -0.00922*** -0.0110*** -0.00458*** -0.0127*** -0.00817*** (5.97) (-25.36) (-23.26) (-12.31) (-32.47) (-15.11) γ24 0.00875*** 0.00108*** -0.00768*** 0.000693 -0.000792** -0.00148** (26.47) (4.59) (-18.93) (1.52) (-2.89) (-2.79) γ25 0.00468*** 0.00400*** -0.000678 0.00271*** -0.00103** -0.00374*** (11.29) (10.98) (-1.23) (4.93) (-2.78) (-5.64) γ26 0.0142*** 0.0106*** -0.00359*** 0.0113*** 0.00472*** -0.00660*** (33.72) (19.70) (-5.27) (16.98) (6.49) (-6.69) γ31 -0.0770*** -0.0786*** -0.00161 -0.0802*** -0.0643*** 0.0159*** (-63.83) (-34.54) (-0.62) (-31.65) (-27.15) (4.58) γ32 0.00182*** -0.00922*** -0.0110*** -0.00458*** -0.0127*** -0.00817*** (5.97) (-25.36) (-23.26) (-12.31) (-32.47) (-15.11) γ33 0.0703*** 0.0741*** 0.00378 0.0571*** 0.0596*** 0.00248 (72.11) (29.88) (1.42) (31.29) (25.83) (0.84) γ34 -0.00567*** 0.0163*** 0.0220*** 0.000621 0.00552*** 0.00490*** (-18.28) (29.74) (34.89) (0.81) (6.20) (4.16) γ35 -0.00336*** -0.000667 0.00270*** 0.00734*** 0.00479*** -0.00255* (-11.90) (-1.14) (4.14) (9.52) (6.35) (-2.37) γ36 0.0139*** -0.00187* -0.0158*** 0.0197*** 0.00717*** -0.0125*** (27.80) (-2.28) (-16.41) (15.68) (5.74) (-7.08) γ41 0.00241*** -0.00409*** -0.00650*** -0.00705*** -0.00652*** 0.000532 (3.81) (-5.74) (-6.82) (-4.50) (-5.95) (0.28) γ42 0.00875*** 0.00108*** -0.00768*** 0.000693 -0.000792** -0.00148** (26.47) (4.59) (-18.93) (1.52) (-2.89) (-2.79) γ43 -0.00567*** 0.0163*** 0.0220*** 0.000621 0.00552*** 0.00490*** (-18.28) (29.74) (34.89) (0.81) (6.20) (4.16) γ44 -0.00291*** -0.00000465 0.00291*** 0.00305*** 0.00884*** 0.00578*** (-7.31) (-0.01) (5.48) (3.77) (16.34) (5.93) γ45 -0.00365*** -0.0105*** -0.00689*** -0.00593*** -0.00731*** -0.00138 (-10.94) (-30.94) (-14.45) (-8.95) (-13.47) (-1.61) γ46 0.00106** -0.00275*** -0.00381*** 0.00861*** 0.000256 -0.00836*** (2.76) (-5.74) (-6.21) (10.02) (0.30) (-6.96) γ51 -0.0202*** -0.0281*** -0.00795*** -0.0436*** -0.0345*** 0.00914*** Continued on next page 147 Table B.9 – continued from previous page Rural Urban Parameters 1983 2004-05 Difference 1983 2004-05 Difference (-31.49) (-29.73) (-6.96) (-28.88) (-24.85) (4.46) γ52 0.00468*** 0.00400*** -0.000678 0.00271*** -0.00103** -0.00374*** (11.29) (10.98) (-1.23) (4.93) (-2.78) (-5.64) γ53 -0.00336*** -0.000667 0.00270*** 0.00734*** 0.00479*** -0.00255* (-11.90) (-1.14) (4.14) (9.52) (6.35) (-2.37) γ54 -0.00365*** -0.0105*** -0.00689*** -0.00593*** -0.00731*** -0.00138 (-10.94) (-30.94) (-14.45) (-8.95) (-13.47) (-1.61) γ55 0.00960*** 0.0187*** 0.00907*** 0.0227*** 0.0267*** 0.00400** (18.85) (25.67) (10.22) (23.64) (30.69) (3.09) γ56 0.0129*** 0.0166*** 0.00374*** 0.0168*** 0.0113*** -0.00547** (26.63) (24.57) (4.50) (14.97) (8.87) (-3.21) γ61 -0.0399*** 0.00126 0.0412*** -0.00608 0.0265*** 0.0326*** (-31.78) (0.76) (19.82) (-1.81) (9.02) (7.30) γ62 0.0142*** 0.0106*** -0.00359*** 0.0113*** 0.00472*** -0.00660*** (33.72) (19.70) (-5.27) (16.98) (6.49) (-6.69) γ63 0.0139*** -0.00187* -0.0158*** 0.0197*** 0.00717*** -0.0125*** (27.80) (-2.28) (-16.41) (15.68) (5.74) (-7.08) γ64 0.00106** -0.00275*** -0.00381*** 0.00861*** 0.000256 -0.00836*** (2.76) (-5.74) (-6.21) (10.02) (0.30) (-6.96) γ65 0.0129*** 0.0166*** 0.00374*** 0.0168*** 0.0113*** -0.00547** (26.63) (24.57) (4.50) (14.97) (8.87) (-3.21) γ66 -0.00214 -0.0238*** -0.0217*** -0.0504*** -0.0500*** 0.000396 (-1.90) (-12.34) (-9.71) (-13.01) (-11.38) (0.07) λ1 -0.0475*** -0.0512*** -0.00364 -0.0411*** -0.0468*** -0.00574 (-19.31) (-22.21) (-1.08) (-5.10) (-26.46) (-0.70) λ2 0.00113 -0.00420*** -0.00533*** -0.000498 -0.00691*** -0.00641* (1.82) (-12.12) (-7.51) (-0.18) (-20.05) (-2.34) λ3 0.0195*** 0.0401*** 0.0205*** 0.00737 0.0182*** 0.0109* (15.34) (16.46) (7.47) (1.60) (12.73) (2.25) λ4 0.00143 -0.00205*** -0.00348*** 0.00360* 0.00296*** -0.000643 (1.87) (-3.45) (-3.59) (2.08) (3.38) (-0.33) λ5 0.00618*** 0.00345 -0.00273 0.0121** 0.00190 -0.0102* (7.08) (1.94) (-1.38) (3.11) (1.06) (-2.39) λ6 0.0192*** 0.0139*** -0.00536* 0.0185*** 0.0306*** 0.0121* (14.86) (6.49) (-2.14) (4.21) (13.95) (2.47) θ1 0.0207*** 0.0238*** 0.00309** 0.0239*** 0.0246*** 0.000758 (25.09) (39.47) (3.03) (16.86) (28.74) (0.46) Continued on next page 148 Table B.9 – continued from previous page Rural Urban Parameters 1983 2004-05 Difference 1983 2004-05 Difference θ2 0.000322 0.00199*** 0.00167*** 0.000945*** 0.00279*** 0.00185*** (1.73) (13.68) (7.06) (4.89) (16.84) (7.25) θ3 -0.0116*** -0.0222*** -0.0106*** -0.0112*** -0.0142*** -0.00293* (-22.18) (-35.86) (-13.13) (-14.07) (-17.37) (-2.56) θ4 -0.00246*** -0.00141*** 0.00105** -0.00313*** -0.00278*** 0.000347 (-8.14) (-5.25) (2.60) (-7.10) (-6.08) (0.55) θ5 -0.000549** 0.000659** 0.00121*** -0.00185*** -0.00152*** 0.000332 (-2.79) (2.87) (3.99) (-4.14) (-4.15) (0.58) θ6 -0.00644*** -0.00283*** 0.00360*** -0.00858*** -0.00893*** -0.000353 (-17.63) (-8.16) (7.15) (-13.02) (-13.30) (-0.38) η1 0.0215*** 0.0269*** 0.00543*** 0.0258*** 0.0257*** -0.0000630 (33.89) (53.77) (6.72) (25.68) (34.73) (-0.05) η2 0.0000460 0.00111*** 0.00106*** 0.00101*** 0.00195*** 0.000940*** (0.28) (8.66) (5.08) (5.40) (12.03) (3.81) η3 -0.00917*** -0.0172*** -0.00799*** -0.00906*** -0.0135*** -0.00444*** (-24.19) (-34.69) (-12.82) (-15.98) (-20.93) (-5.17) η4 -0.00277*** -0.00482*** -0.00204*** -0.00294*** -0.00356*** -0.000622 (-16.27) (-20.83) (-7.11) (-7.84) (-8.95) (-1.14) η5 -0.00192*** -0.00180*** 0.000124 -0.00446*** -0.00343*** 0.00104* (-11.74) (-9.81) (0.51) (-14.29) (-10.05) (2.24) η6 -0.00767*** -0.00426*** 0.00341*** -0.0103*** -0.00719*** 0.00314*** (-25.53) (-14.68) (8.17) (-17.49) (-12.82) (3.86) N 151657 151657 151657 75073 75073 75073 Equation RMSE R-sq N RMSE R-sq N 1 cereal all 0.123595 0.9292* 151657 0.106779 0.8967* 75073 2 pulse 0.036493 0.7383* 151657 0.027795 0.8183* 75073 3 milk 0.103738 0.6423* 151657 0.101685 0.7414* 75073 4 mef 0.060821 0.4155* 151657 0.069217 0.4505* 75073 5 vegfruit 0.049156 0.8692* 151657 0.051868 0.8882* 75073 Note: t-values are in parenthesis. ∗p < 0.05, ∗ ∗ p < 0.01, ∗ ∗ ∗p < 0.001. * Uncentered R-sq. 149 0 . 5 1 D en sit y 0 .5 1 1.5 2 2.5 Fat/RDA 1983: Rural 2004−05: Rural Ratio of fat intake to Recommended Daily Allowance (RDA) Distribution of fat/RDA in rural sector (a) Fat/RDA - Rural 0 . 2 . 4 . 6 . 8 D en sit y 0 .5 1 1.5 2 2.5 Fat/RDA 1983: Urban 2004−05: Urban Ratio of fat intake to Recommended Daily Allowance (RDA) Distribution of at/RDA in urban sector (b) Fat/RDA - Urban 0 . 5 1 1. 5 D en sit y 0 .5 1 1.5 2 2.5 Protein/RDA 1983: Rural 2004−05: Rural Ratio of protein intake to Recommended Daily Allowance (RDA) Distribution of protein/RDA in rural sector (c) Protein/RDA - Rural 0 . 5 1 1. 5 D en sit y 0 .5 1 1.5 2 2.5 Protein/RDA 1983: Urban 2004−05: Urban Ratio of protein intake to Recommended Daily Allowance (RDA) Distribution of protein/RDA in urban sector (d) Protein/RDA - Urban 0 . 5 1 1. 5 D en sit y 0 .5 1 1.5 2 2.5 Carbohydrate/RDA 1983: Rural 2004−05: Rural Ratio of carbohydrate intake to Recommended Daily Allowance (RDA) Distribution of carbohydrate/RDA in rural sector (e) Carbohydrate/RDA - Rural 0 . 5 1 1. 5 2 D en sit y 0 .5 1 1.5 2 2.5 Carbohydrate/RDA 1983: Urban 2004−05: Urban Ratio of carbohydrate intake to Recommended Daily Allowance (RDA) Distribution of carbohydrate/RDA in urban sector (f) Carbohydrate/RDA - Urban Figure B.3: Distribution of households by ratio of nutrients intake to RDA 150 PL0 1 2 3 Fa t/R D A 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Rural Fat intake as proportion of RDA and MPCE (a) Fat/RDA - Rural PL.5 1 1. 5 2 2. 5 3 Fa t/R D A 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Urban Fat intake as proportion of RDA and MPCE (b) Fat/RDA - Urban PL0 . 5 1 1. 5 2 Pr ot ei n/ RD A 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Rural Protein intake as proportion of RDA and MPCE (c) Protein/RDA - Rural PL0 . 5 1 1. 5 Pr ot ei n/ RD A 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Urban Protein intake as proportion of RDA and MPCE (d) Protein/RDA - Urban PL . 05 . 1 . 15 . 2 . 25 Ca rb oh yd ra te /R D A 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Rural Carbohydrate intake as proportion of RDA and MPCE (e) Carbohydrate/RDA - Rural PL . 05 . 1 . 15 . 2 . 25 Ca rb oh yd ra te /R D A 3.5 4 4.5 5 5.5 6 ln(MPCE) 1983 2004−05 Urban Carbohydrate intake as proportion of RDA and MPCE (f) Carbohydrate/RDA - Urban Figure B.4: MPCE and nutrients intake as proportion to RDA 151 Table B.10: Per capita per diem intake Calorie(Kcal) 1972-73 1983 1993-94 1999-00 2004-05 Rural 2266 2221 2153 2149 2047 Urban 2107 2089 2071 2156 2020 Protein (gm) 1972-73 1983 1993-94 1999-00 2004-05 Rural 62 62 60.2 59.1 57 Urban 56 57 57.2 58.5 57 Fat (gm) 1972-73 1983 1993-94 1999-00 2004-05 Rural 24 27 31.4 36.1 35.5 Urban 36 37 42 49.6 47.5 Source: NSSO (2007) 152 Table B.11: Nutrient share from different food groups - rural RURAL Calorie 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 77.73 83.18 70.29 71.00 77.11 63.69 -6.73 pulse 4.73 3.89 5.84 4.08 3.56 4.66 -0.65 milk 2.92 1.30 5.03 4.48 2.18 7.16 1.56 mef 0.56 0.44 0.75 0.98 0.79 1.22 0.42 vegfruit 3.39 2.87 4.17 5.04 4.29 5.94 1.65 other 10.68 8.32 13.92 14.43 12.07 17.34 3.75 Protein 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 75.14 81.00 67.46 68.16 74.93 60.19 -6.98 pulse 11.29 8.99 14.02 10.39 9.32 11.56 -0.90 milk 4.21 2.00 7.02 7.36 3.70 11.56 3.16 mef 3.34 2.62 4.31 6.29 5.21 7.49 2.95 vegfruit 3.58 3.29 4.11 5.46 4.90 6.08 1.89 other 2.44 2.11 3.08 2.33 1.94 3.13 -0.11 Fat 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 31.77 43.64 20.49 15.75 20.63 11.04 -16.02 pulse 2.94 2.79 3.15 1.72 1.68 1.72 -1.22 milk 14.19 6.95 22.26 15.75 8.94 22.72 1.56 mef 1.62 1.30 1.99 1.89 1.74 2.04 0.26 vegfruit 5.28 4.71 5.92 4.86 4.42 5.42 -0.43 other 44.19 40.61 46.19 60.03 62.59 57.05 15.85 153 Table B.12: Nutrient share from different food groups - urban URBAN Calorie 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 64.85 73.57 52.74 60.31 69.30 49.59 -4.54 pulse 5.91 5.05 6.68 4.97 4.35 5.54 -0.94 milk 5.20 2.73 8.47 6.77 3.71 10.56 1.57 mef 0.95 0.72 1.26 1.30 1.07 1.52 0.35 vegfruit 4.73 3.69 6.20 6.10 4.96 7.44 1.37 other 18.37 14.23 24.65 20.55 16.62 25.35 2.18 Protein 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 63.12 71.88 51.30 58.94 68.04 48.46 -4.18 pulse 14.39 12.21 16.31 12.34 10.99 13.49 -2.05 milk 7.91 4.42 12.28 10.77 6.34 15.95 2.85 mef 5.07 4.13 6.12 7.65 6.69 8.27 2.58 vegfruit 4.77 4.00 5.84 6.29 5.47 7.22 1.52 other 4.74 3.36 8.15 4.01 2.48 6.62 -0.73 Fat 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 15.47 24.00 8.05 9.63 14.05 5.78 -5.84 pulse 2.49 2.59 2.19 1.65 1.72 1.52 -0.85 milk 18.11 11.70 24.89 19.65 13.08 26.55 1.54 mef 2.14 1.79 2.52 2.04 2.03 1.90 -0.10 vegfruit 5.52 5.20 5.69 5.25 5.03 5.35 -0.27 other 56.27 54.71 56.66 61.78 64.10 58.89 5.51 154 Table B.13: Prices per nutrient from different food groups - rural RURAL Calorie (Price (Rs.) per 1000 KCalorie) 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 0.82 0.65 0.93 2.62 2.27 2.96 1.81 pulse 1.55 1.44 1.64 7.79 7.35 8.14 6.24 milk 4.68 4.33 4.92 18.81 17.57 19.62 14.13 mef 15.62 14.15 17.10 41.48 37.41 44.52 25.86 vegfruit 3.96 3.79 4.12 16.06 14.87 17.17 12.10 other 2.31 2.13 2.47 9.15 8.31 10.00 6.84 Protein (Price (Rs.) per 100 gm Protein) 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 3.70 2.79 4.20 11.85 10.16 13.46 8.15 pulse 2.32 2.16 2.45 11.64 10.94 12.19 9.32 milk 11.83 10.10 13.47 41.42 38.43 43.47 29.59 mef 12.68 11.72 13.90 25.05 22.69 26.92 12.37 vegfruit 14.85 12.99 16.42 58.44 51.87 65.05 43.59 other 58.99 47.45 67.44 283.65 267.01 293.47 224.66 Fat (Price (Rs.) per 100 gm Fat) 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 43.46 29.27 51.35 147.65 124.33 169.82 104.19 pulse 34.24 31.29 35.08 167.37 167.30 169.02 133.13 milk 6.98 6.69 7.20 30.11 28.17 31.29 23.13 mef 49.83 46.63 52.93 196.62 193.59 196.74 146.79 vegfruit 55.60 54.93 57.65 227.38 227.19 222.04 171.78 other 6.03 6.05 6.26 15.59 14.13 17.36 9.56 155 Table B.14: Prices per nutrient from different food groups - urban URBAN Calorie (Price (Rs.) per 1000 KCalorie) 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 0.90 0.76 1.05 3.28 2.67 4.07 2.38 pulse 1.68 1.58 1.78 8.45 8.04 8.87 6.78 milk 5.38 5.10 5.64 22.41 21.73 23.78 17.03 mef 13.49 11.82 14.66 45.27 43.01 47.51 31.77 vegfruit 4.56 4.09 5.11 19.89 17.61 23.47 15.32 other 2.51 2.26 2.82 10.30 9.04 12.29 7.79 Protein (Price (Rs.) per 100 gm Protein) 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 3.84 3.16 4.51 13.90 11.10 17.13 10.06 pulse 2.53 2.36 2.69 12.74 12.10 13.41 10.22 milk 12.74 11.55 14.30 51.22 47.35 58.59 38.48 mef 9.74 7.95 11.13 27.82 25.94 29.88 18.08 vegfruit 17.42 14.42 21.15 73.97 61.65 91.28 56.55 other 74.65 66.18 80.97 327.81 312.90 344.71 253.16 Fat (Price (Rs.) per 100 gm Fat) 1983 2004-05 Change Groups Mean Poorest 25% Richest 25% Mean Poorest 25% Richest 25% Mean cereal all 43.91 34.61 52.54 163.01 127.51 199.35 119.10 pulse 33.64 33.52 33.79 160.85 158.84 160.56 127.22 milk 7.99 7.67 8.26 35.15 34.41 36.46 27.16 mef 43.35 43.38 41.82 190.95 190.47 188.23 147.60 vegfruit 59.94 61.25 61.12 244.52 237.95 257.61 184.58 other 5.37 5.88 5.51 21.36 15.43 39.76 15.99 156

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0072200/manifest

Comment

Related Items