UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The labour market implications of job quality Vahey, Shaun Patrick 1995

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

Item Metadata

Download

Media
831-ubc_1995-983590.pdf [ 1.23MB ]
Metadata
JSON: 831-1.0088853.json
JSON-LD: 831-1.0088853-ld.json
RDF/XML (Pretty): 831-1.0088853-rdf.xml
RDF/JSON: 831-1.0088853-rdf.json
Turtle: 831-1.0088853-turtle.txt
N-Triples: 831-1.0088853-rdf-ntriples.txt
Original Record: 831-1.0088853-source.json
Full Text
831-1.0088853-fulltext.txt
Citation
831-1.0088853.ris

Full Text

THE LABOUR MARKET IMPLICATIONS OF JOB QUALITYbySHAUN PATRICK VAHEYB.A., University of Essex, 1986M.A., University of Essex, 1988A THESIS SUBMITTED IN PARTIAL FULFILMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinTHE FACULTY OF GRADUATE STUDIESDepartment of EconomicsWe accept this thesis as conformingto the required standardTHE UNIVERSiTY OF BRITISH COLUMBIAApril 1995© Shaun Patrick Vahey, 1995In presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.Department of COO,JCThe University of British ColumbiaVancouver, CanadaDate f(f(ç(Signature)DE-6 (2/88)AbstractThis thesis takes the form of three essays about the labour market implications ofjob quality.In the first essay, I demonstrate, by analysing a two-type, two-period example, thathigh introductory wage offers can signal the quality of experience jobs. In this game,one type of firm - the “good” type - offers higher expected quality jobs. If this typeis less likely to exit from the industry than the “bad’ type, it can increase expenditureon introductory wages without being mimicked, distinguishing it from its inferior. Thegame has many equilibria with these separating wages. In each, the introductorycompensating differentials have the opposite sign to the usual case: higher expectedquality jobs pay more, rather than less.In the second essay, I present Canadian evidence that tests and supports the theoryof compensating differentials for a variety of job characteristics. The data used arefrom the National Survey of Class Structure and Labour Process in Canada (NSCS).These self-report data are preferable to the more conventional occupational-trait data;they provide information on individual jobs rather than averages across broadoccupational categories and industries.In the third essay, I focus on the mismatch between the educational requirementsof jobs and the educational attainments of workers. Using NSCS data, I find that thereturns to over- and undereducation for males are sensitive to the level of requirededucation. There is evidence of positive returns to overeducation for jobs that requirea university bachelor’s degree; but, in general, the returns are insignificant.Undereducated workers are penalised in jobs with low educational requirements. Forfemales, I find that the returns to over- and undereducation are insignificant for alllevels of required education.11Table of ContentsAbstract iiTnble of Contents iiiList of Tables vList of Figures viAcknowledgement viiChapter 1 Introduction 1Chapter 2 Signalling Job Quality 52.1 Introduction 52.2 Related Literature 72.3 The Game 92.4 Strategies and Payoffs 112.4.1 The Firm 112.4.2 The Worker 122.5 Equilibria 122.5.1 Separating Introductory Wages 122.5.2 Pooling Introductory Wages 152.6 Refinement 172.7 Discussion 212.8 Conclusions 22Chapter 3 Compensating Differentials: Some Canadian Self-ReportEvidence 243.1 Introduction 243.2 Related Literature 263.3 Model and Data 283.4 Results 313.5 Conclusions 39Chnpter 4 The Great Canadian Training Robbery 404.1 Introduction 414.2 Related Literature 424.3 The Incidence of Educational Mismatch 431114.4 Empirical Model.454.5 Results 464.6 Conclusions 53Chapter 5 Conclusions 59References 61Appendix 1 69ivList of TablesTable 1: Chapter 3, Variable Definitions and Means 29Table 2: Chapter 3, Regression Equations 32Table 3: Chapter 3, Further Regression Equations 37Table 4: Incidence of Skill Mismatch 45Table 5: Chapter 4, Variable Definitions and Means 47Table 6: Chapter 4, Regression Equations 49Table 7: Chapter 4, Male-Female Earnings Decomposition 53Table 8: Chapter 4, Further Regression Equations 54Table A: Variable Means by Sex and Union Status 70Table B: Further Means and Definitions 72Table C: Job Quality and Union Status Equations 73Table D: Senior Workers Regression Equations 76VList of FiguresFigure 1: Time Line 11Figure 2: Equilibrium Wages, where a<(1 -p) 18Figure 3: Equilibrium Wages, where a(1 -p) 18Figure 4: Earnings Profile - Full Sample 56Figure 5: Earnings Profile - Males 57Figure 6: Earnings Profile - Females 58viAcknowledgementI would like to thank my thesis committee, Ron Giammarino, Ken Hendricks andCraig Riddell, for their helpful guidance and patience. I also benefitted greatly fromdiscussions with Paul Beaudry, John Cragg, Denise Doiron, Jeff Frank, Steve Jones,Michele Piccione, Mark Thompson and Liz Wakerly. I am indebted to my fellowgraduate students and my parents for their support and encouragement, without whichthis research would not have been possible.viiChapter 1IntroductionThis thesis takes the form of three related essays about the labour marketimplications of job quality.The first essay, Chapter 2, is a theoretical piece about signalling job quality.Economists have long recognised that the theory of compensating differentialsdoes not sufficiently explain the observed distribution of wages across job quality.The theory predicts that lower quality jobs pay higher wages than more desirablejobs. However, casual observation suggests that workers in more desirable jobsare often paid more, rather than—as the theory predicts—less.Following the terminology coined by Nelson [71] in the context of productquality, I draw a distinction between “search” and “experience” jobs. Manyunskilled jobs are of the former type: quality is immediately apparent. Incontrast, the quality of skilled jobs is often difficult to verify by inspection, andworkers learn through experience. For a firm offering this type of job, conveyingquality is difficult; a straightforward claim is unverifiable and can be costlesslycopied by firms with lower quality jobs. It is, however, in the interests of botha firm supplying high quality jobs, and workers searching for such jobs, thatinformation about quality is revealed.In this chapter, I demonstrate by analysing a two-type, two-period example,that high introductory wage offers can signal the quality of experience jobs. Inthis game, one type of firm, hereafter referred to as the “good” type, offershigher expected quality jobs. Assuming this type is less likely to exit from theindustry than the “bad” type, it can increase expenditure on introductory wageswithout being mimicked, distinguishing it from its inferior. The game has manyequilibria with these separating wages. In each, the introductory compensatingdifferentials have the opposite sign to the usual case: higher expected quality1jobs pay more, rather than less. There are also equilibria in which the twotypes offer the same (or pooling) introductory wage. In these equilibria, theintroductory wage offer is independent of job quality.The intuition captured by the model is very straightforward. Wage offersaffect inexperienced workers’ beliefs about job quality. Realising this, the firmresists cutting the wage to the market clearing level.In the second essay, Chapter 3, I present Canadian evidence that tests andsupports the theory of compensating differentials for a variety of job characteristics. Previous Canadian studies, such as those by Meng [64 and 65], Martinelloand Meng [62], and Cousineau et al [22] have found support for the theory forrisk of injury or death. But no previous research has found Canadian evidenceof compensating wage payments for non-hazardous job characteristics.The data used in this chapter come from the National Survey of Class Structure and Labour Process in Canada (NSCS). These unique data, which arecross-sectional and relate to 1981 incomes, provide detailed self-report information about the respondents’ job quality and personal characteristics. Meng’s[64] study of compensating differentials also used this data set. However, tomeasure job quality, Meng used occupational-trait data developed by StatisticsCanada and Employment and Immigration Canada, rather than the self-reportinformation contained in the Survey. Unfortunately, these occupational-traitdata provide information on averages across broad occupational categories andindustries, introducing error into the job quality variables. Using self-reportdata—which avoids this particular problem—I find stronger support for thetheory of compensating differentials than did Meng [64].After controlling for personal characteristics, I find evidence of compensating differentials for working with data, working with hands, bureaucratic procedures, and responsibility over other workers. I also find evidence of differentialsfor the control of hours and pace. Although these characteristics are generallyheld to be desirable, I find that they are associated with higher, rather thanlower, wages. That is, the coefficients have the “wrong” signs. I find no evidence of differentials for working with people, working with machines and the2freedom to design work. In short, my results support the theory of compensating differentials across a wider range of characteristics than previous Canadianstudies.The third essay, Chapter 4, takes a slightly different perspective on jobquality. I focus on the mismatch between the educational requirements of jobsand the educational attainments of workers. In his seminal work, “The GreatTraining Robbery”, Ivar Berg [14] argued that overeducated workers may be lessproductive than their less skilled counterparts because they become bored withtheir jobs and lose motivation. Despite the widespread concern over educationalmismatch, there have been few studies of its impact on earnings; data sets rarelycontain information on both the educational attainments of workers and therequirements of jobs. A small number of studies have used either US or Dutchdata; but there have been no previous Canadian studies. Duncan and Hoffman[28], Rumberger [84], Hersch [43] and Sicherman [87] have estimated earnings(or wage) equations including both the years of required schooling for the job,and the years of over- or undereducation. These researchers have found strongevidence that the earnings of overeducated workers are greater than those ofotherwise identical workers who are neither overeducated nor undereducated;and, that the earnings of undereducated workers are lower. Assuming earningsreflect marginal productivity, their results refute Berg’s [14] hypothesis.I use NSCS data to estimate the returns to educational mismatch in Canada.I find that the returns to over- and undereducation are sensitive to the level ofrequired education. There is evidence of positive returns to overeducation forjobs that require a university bachelor’s degree; but, in general, the returnsare insignificant. Unlike previous studies, I find little evidence of lower pay forundereducated workers; though they are penalised in jobs with low educationrequirements. I also estimate separate equations for male and female workers.Although I find that the results for the male sub-sample are similar to the fullsample, I find that the returns to over- and undereducation for females are insignificant for all levels of required education. Since, in general, overeducatedworkers have identical earnings to their less skilled counterparts with the re3quired level of schooling, the data do not support Berg’s [14] claim. But, in thesense that overeducated workers do not receive the full returns to their attainededucation, they are “robbed”.In the final chapter, I draw some conclusions from the thesis.4Chapter 2Signalling Job Quality2.1 IntroductionEconomists have long recognised that the theory of compensating differentials (or equalising differences, as it is also known) does not fully explain theobserved distribution of wages across job quality. The theory, originally due toAdam Smith [89] and formalised by Rosen [81], predicts that better jobs paylower wages. Yet, casual observation suggests that often workers in high qualityjobs are paid more, rather than—as the theory predicts—less.For example, in the early 1980s, ICI made particularly enticing offers tofreshly trained UK chemical engineers. Ex post, it is apparent that the firm hadhigh safety standards, offered good prospects for promotion and training, andinfrequently laid-off workers. Yet the pay offers exceeded those of its immediatecompetitors.’On the face of it, ICI gained little from offering high wages; except a reputation as a high paying employer. And the message, “We devote a vast sum toour wage bill” appears—at first glance—to be useless. In this chapter, I proposethat this message can inform newly qualified, inexperienced labour about jobquality.2Following the terminology coined by Nelson [71] in the context of productquality, I draw a distinction between “search” and “experience” jobs. Manyunskilled jobs are of the former type: the quality is immediately apparent. Incontrast, the quality of skilled jobs—whether it be repetition, stress, or risk‘In 1982, the basic weekly wage of the lowest-skilled ICI worker was over l5at Glaxo, itsbest-known domestic competitor (Smith [90]).‘Efficiency wage theories, associated with the work by Salop [85], Shapiro and Stiglitz[86] and Weiss [108] can also account for non-market clearing wages. My explanation iscomplementary.5of fatality—is often difficult to verify by inspection, and workers learn throughexperience. For a firm offering this type of job, conveying quality is troublesome;a straightforward claim is unverifiable and can be costlessly copied by inferiors.It is, however, in the interests of both a firm supplying high quality jobs, andworkers searching for such jobs, that information about quality is revealed.In this chapter, I demonstrate, by analysing a two-type, two-period example,that high introdnctory wage offers can signal the quality of experience jobs. Onetype of firm, hereafter referred to as the “good” type, offers higher expectedquality jobs. If this type is less likely to exit from the industry than the “bad”type, it can increase expenditure on introductory wages without being mimicked,distinguishing it from its inferior. The game has many equilibria with theseseparating wages. In each, the introductory compensating differentials have theopposite sign to the usual Smith/Rosen case: higher expected quality jobs paymore, rather than less.There are also equilibria in which the two types offer the same (or pooling)introductory wage. In these equilibria, the introductory wage offer is independent of job quality.The intuition captured by the model is very straightforward. Wage offersaffect inexperienced workers’ beliefs about job quality. Realising this, the firmresists cutting the wage to the market clearing level.The distinction between search and experience jobs has been made before by(amongst others) Adam Smith [89], Reynolds [79] and Johnson [46]. A series ofpapers by Viscusi [99, 100, 101, 102 and 103] and Viscusi and Moore [107] has explored the relationship between compensating differentials and worker learning.In the model common to these papers, inexperienced workers are compensatedfor undesirable jobs; but, because they are more optimistic about job quality, they receive smaller compensating differentials than their more experiencedcounterparts. However, the result relies on the assumption that workers’ beliefsare independent of the wage offers. In this chapter, I show that if this assumption is relaxed, inexperienced workers are generally uncompensated ex ante forundesirable jobs. In a separating equilibrium, it is the type of firm with higher6expected quality jobs that pays the differential.The remainder of this chapter is set out as follows. In the following section,I review the related literature. I analyse the signalling game in Section 2.3.The strategies and payoffs of the firm and the workers are considered in Section2.4. The (Bayesian) Nash equilibria to the game are characterised in Section2.5. Signalling games generally have a plethora of equilibria; and this one is noexception. In Section 2.6, I show that (a two-period version of) the Cho-KrepsIntuitive Criterion renders a unique separating equilibrium, where one exists.The implications of the game are discussed in Section 2.7, and some conclusionsare drawn in the final section.2.2 Related LiteratureAs a rule, researchers in the compensating differentials tradition have assumed that workers know the quality of the jobs available to them. The initialidea, developed by Adam Smith [89], was formalised by Rosen [811. A job isviewed as a tied transaction: workers simultaneously sell labour and buy jobcharacteristics. The equilibrium wage distribution clears the market so that theworker’s and firm’s preferences are matched. As a result of this sorting process,there is an equilibrium trade-off between each job characteristic and the wage.The standard textbook treatment is as follows (see Gunderson and Riddell[38]). Consider a job with a single observable characteristic. Suppose workersdislike this characteristic, but firms find it costly to eradicate. A decrease in theprovision of the characteristic can be thought of as an increase in job qualitycThe equilibrium locus is downward sloping in wage/job quality space. A positivecompensating differential is paid to workers who take on lower quality jobs.This equilibrium wage-job quality schedule can be expressed as:(1) w=w(x,z),where w, x and z denote the real wage, job quality, and other wage determinantsrespectively. Typically, this locus is assumed to be concave. The trade-off7captures the workers’ “willingness-to-pay” for increases in quality.3The idea that workers have difficulty observing job quality prior to employment has a considerable tradition. Adam Smith [89] noted that workers tendto underestimate the risk of injury and death, particulary when young. Oi [74],Diamond [23] and Rea [78] have argued that workers are systematically misinformed about safety levels. Carmichael [19] has shown that if workers take timeto learn about the risk of injury (but are fully informed in the steady state)then there is a moral hazard problem: firms may cut costs by offering riskierjobs.Outside the safety literature, Reynolds [79], Johnson [46], Jovanovic [48 and49] and Wilde [110] have argued that a variety of job characteristics are unobservable prior to employment. This leads to high mobility or “job shopping” asyoung workers experiment with different jobs.The connection between experience jobs and compensating differentials hasbeen made by Viscusi [99, 100, 101, 102 and 103] and Viscusi and Moore [107].In the two-period, two-type model in these papers, one type of firm offers higherexpected quality jobs. It is assumed that the workers have heterogeneous work-leisure preferences. They learn about expected job quality from experience,updating their beliefs using Bayes’ rule. The wage in each period is determinedby market clearing, so that the marginal worker in each period receives reservation utility. The period 2 marginal worker has had an unfavourable period 1job experience and, therefore, is less optimistic about job quality than the period 1 marginal worker. Consequently, inexperienced workers demand a smallercompensating differential. However, this result depends on the assumption thatworkers’ beliefs are independent of the introductory wage offers. In this chapter,I show that if this assumption is relaxed, inexperienced workers are generallyuncompensated ex ante for undesirable jobs.A number of other researchers have argued that prices signal firm-side privateinformation. Milgrom and Roberts [67], Bagwell [9], and Allen and Faulhaber[4] have examined models in which product quality is signalled through theof the compensating differential can be obtained from the regression coefficientin the hedonic wage equation (1).8introductory price. Frank [32], Beaudry [10, 11 and 12], Giammarino and Nosal[36], Aryan and Esfahani [8] and Kuhn [56] have shown that a variety of firmcharacteristics (such as the marginal product, the quality of management, andfirm-specific human capital) can be signalled through the wage. However, noneof these researchers have analysed the use of the wage to signal job quality orthe impact of this on compensating differentials. In the following sections thisanalysis is carried out.2.3 The GameThe game between the single firm and the workers is structured as follows.The firm can be one of two types: good or bad. Firm type, indexed by q,is good {q = G} with probability z, and bad {q = B} with probability (1 — z).The firm incurs a fixed cost of production in each period. I shall subsequentlyrestrict how this cost, denoted Gq, varies with type. The cost is never observedby the workers; they cannot learn type directly, but must infer it from the firm’sbehaviour.In both periods, the good type provides higher expected quality jobs. Let xdenote the job quality in period 1. For the good type, this is high {x = H} withprobability p, and low {x = L} with probability (1 —p). The bad type providesonly low quality jobs. Period 2 job quality is determined in an identical manner.Given its type, the firm must choose a period 1 wage offer. It must alsodecide whether to stay in the market in the subsequent period; and if it doesdecide to stay, it must choose a period 2 wage offer. At the beginning of thesecond period, the firm learns the value that can be achieved by relocating itscapital to the best alternative. This outside option, denoted r, is equal to 0 withprobability a, and some positive value, R, with probability (1 — a). I assumethat the realisation of this outside option is unobserved by the workers.The perfectly competitive output market price is normalised to one. Production occurs according to a strictly concave production function, lit = f(ni),where lit and itt represent output and the quantity of labour demanded in period t respectively. The amount of labour the firm would like to employ at wage9w is determined by its inverse labour demand schedule, a = g(w).4 With noloss of generality, I assume that the firm is never constrained by the size of theworkforce. The discount factor is set to one.The workforce is comprised of N identical, atomistic, risk-neutral workers.The (representative) worker has additively separable preferences over time, andin each period is endowed with an indivisible unit of labour. If the worker doesnot sell the unit of labour, it obtains the reservation utility level, which (again,with no loss of generality) I set to zero. If it sells the unit of labour at wage w,it obtains utility w if job quality is high; and w — L if job quality is low. HereL measures the disutility from a low quality job.Firm type is never directly observed by the worker, and job quality is onlyobserved at the end of each period. The worker’s initial beliefs are that jobquality is high with probability zp, and low with probability (1 — zp). Thesebeliefs are revised upon observing the period 1 wage offer according to Bayes’rule. Let b1 denote the worker’s posterior probability that job quality is low.The belief function lR —* [0, 1] maps the wage offer into a period 1 posteriorprobability. In the second period, if the firm does not exit, this probability isrevised based on the realisation of period 1 job quality, and the subsequent wageoffer. If the firm exits, no decision is required by the worker. The belief functionin period 2 can be defined as /32 [L, H] x —* [0, 1]. Let b2 denote a value of/32 for a particular history.In summary, the sequence of events is described by the time line shown inFigure 1. At the start of the game, the firm’s type is determined. This is privateinformation to the firm. The worker has prior belief (1— zp) that job quality inperiod 1 will be low. The firm offers a one period contract specifying the firstperiod wage. The worker forms a posterior belief that the period 1 job qualitywill be low, b1, and either accepts or rejects the offer. If it accepts, productionoccurs, and job quality in period 1 is realised.At the beginning of period 2, the value of the firm’s outside option is determined, and it decides whether to exit or stay. If it stays, it makes another4Note that since f(S) is type independent, so is g(.).10wage offer, accept orWqi reject productionPeriodlj Iwage offer,Wq2Period 2accept orreject productionEndwage offer. The worker forms a (revised) posterior belief that the job will below quality, b2; and again makes an accept/reject decision. If no job is offeredin period 2 or accepted in either period, the game ends.2.4 Strategies and Payoffs2.4.1 The FirmRecall the firm’s problem. First, the firm chooses a period 1 wage offer.Then, at the start of period 2, it decides whether to exit or stay; and if it stays,it chooses another wage offer. The good type’s period 2 decisions are conditionalon both the period 1 job quality and the outside option. The bad type’s period2 behaviour is conditional only on the outside option—it does not offer highquality jobs.For each type, a pure strategy is comprised of: a first period wage offer; anindicator function for the exit decision (which takes the value 1 if the firm exits,and 0 if it stays); and, if it stays, a period 2 wage function.5The optimal strategy for a type q firm maximizes expected two-period prof5For simplicity, attention is restricted to pure strategies. Mixed strategies could be permitted but, wbere separating equilibria exist, none survive tbe adopted refinement.firm type, q job outside job quality, a:option, rFigure 1: Time Line11its, given the strategy of the other type and the worker’s decision rule.2.4.2 The WorkerThe worker’s problem consists of deciding whether to accept or reject wageoffers, given beliefs about job quality. The first period decision is based solelyupon the period 1 wage offer. In the second period, the worker utilises theinformation revealed by the two wage offers, period 1 job quality and the firm’sdecision to stay.A pure strategy for the worker is a pair of indicator functions, one for eachperiod (which take the value 1 if the worker accepts, and 0 otherwise).6 Theoptimal strategy maximizes the expected utility over the two periods.2.5 EquilibriaA (Bayesian) Nash equilibrium for this game is a strategy combination suchthat both types behave optimally given each other’s strategy, and the worker’sdecision rule. There are two types of equilibria: those in which the introductorywages are pooling, and those in which they are separating. In the latter, firmtype is revealed by the period 1 wage offer; in the former, it is not.In the following sections, I characterise the wage offers in each type of equilibria.2.5.1 Separating Introductory WagesSuppose the two types choose different period 1 wage offers, that is w1w1. The worker can differentiate between the two types from the wage offeralone. Therefore, the worker’s beliefs about job quality are given by:(2) (w1) == /3(w1,w2L)=1.6Mixed strategies for the worker could be permitted but are never optimal.12Given these beliefs, consider the period 2 wage offers. Since there is onlyone source of demand for labour, the (atomistic) worker’s wage is bid down tothe reservation level—the worker’s participation constraint binds with equality.The period 2 wage offers are (1 —p)L for the good type; and L for the bad type.In the light of this behaviour, consider the firm’s exit decision at the startof the second period. If, for either type, period 2 profits are exceeded by thevalue of the outside option, then the firm exits. For the good type these profitsare given by:(3) ir((l — p)L, G) =f(g((l — p)L)) — ((1 — p)L) . g((l — p)L)— CG.And, for the bad type they are given by:(4) ir(L, B) = f(g(L)) — (L) . g(L)— GB.I assume that for both types, one-period profits are strictly positive for any nonzero quantity of labour. Hence, if r = 0, neither type exits. But if r = R, bothtypes may exit. I make two assumptions about the firm’s behaviour in thesecircumstances. First, I assume the good type stays, even if the worker believesthe firm is the bad type (in which case the wage is L). Second, I assume thebad type exits, even if the worker believes the firm is the good type (in whichcase the wage is (1 — p)L). That is, I assume that:(Al) 7r(L,G) > R> ir((l —p)L,B).A necessary (but not sufficient) condition for this to hold is that the bad type’sfixed costs exceed the good type’s, GB > CG. Assumption (Al) ensures thatthe probability of exit is 0 for the good type; and (1 — a) for the bad type.7Hence, the probability of exit is type dependent (even though the value of theoutside option is not); the bad type finds the outside option more attractivethan the good type.7Assumption (Al) is sufficient—but not necessary—for the existence of separating equilibria. The necessary and sufficient condition is that the probability of exit is lower for the goodtype. Else, there are only pooling equilibria. The role of this assumption is discussed furtherin Section 2.7.13Given this period 2 behaviour, consider the period 1 wage offers. The badtype’s offer just satisfies the worker participation constraint, L: a higher wagewould lower period 1 profits but would leave the worker’s beliefs (and thereforeperiod 2 profits) unaffected.The good type’s offer must ensure that the bad type’s behaviour is incentivecompatible. In equilibrium, the bad type’s (expected) payoff is:(5) ir(L, B) + air(L, B) + (1 — a)R.If, however, the bad type deviates by mimicking the good type’s period 1 offer,its payoff is:(6) ir(w1,B) + air((1 — p)L, B) + (1 — a)R.Let WB define the wage that equates conditions (5) and (6). If the good type’soffer is at least as large as WB, then the bad type has no incentive to deviate.Notice that since (1— p)L is less than L, WB is strictly greater than L: to avoidmimicry, the good type must exceed the bad type’s post-separation offer.The good type’s behaviour must also be incentive compatible. The goodtype’s equilibrium payoff is:(7) ir(w1,G) + ir((1 — p)L, G).Suppose the good type deviates by mimicking the bad type. In this case,its type may still be revealed if period 1 job quality is high. I assume that,should this occur, the worker correctly infers that the firm is good. That is,/3(W1,WG2H) = (1 —p). Then, the good type’s payoff from the deviation is:(8) 7r(L, G) + (1 — p)ir(L, G) +pir((l — p)L, G).Let WG define the wage that equates expressions (7) and (8). To ensure incentivecompatibility, the good type’s offer must be strictly less than WG. Notice thatlike WB, the wage offer WG is strictly greater than L.Clearly, the period 1 offers can be separating only if WQ > WB. Thiscondition is the two-period equivalent of the Mirrlees-Spence single crossing14property for this game. As the following lemma makes apparent, this holdsonly for a particular parameter space.Lemma 1 The introductory wage WG exceeds WB if, and only if, a < (l—p).Proof Recall that WB is defined by expressions (5) and (6); and that WG isdefined by (7) and (8). Rearranging these expressions gives:(9) 1r(WB, B) = (1 + a)ir(L, B) — air((1 — p)L, B),and,(10) Ir(WG, G) = (2 — p)ir(L, G) — (1— p)ir((l — p)L, G).Subtracting (10) from (9), and adding fixed costs to both sides gives:(11) lr(T’VB) — lr(WQ) = [a— (1 — p)][ir(L)— ir((1— p)L)j,where ir(w) = f(g(w)) — w g(w). The lemma follows. QED.The period 1 wage offer affects the firm’s subsequent behaviour in two cases:first, if the firm is bad and stays; and second, if the firm is good and period1 job quality is low. The good type can separate if, and only if, (conditionalon type) the probability of the latter event, (1—p), is greater than that of theformer, a.If this condition holds, there is a continuum of separating equilibria in whichthe good type’s period 1 wage offer is given by w1 e [WB, WG). These equilibria can be supported by the belief that any offer outside this interval comesfrom the bad type. Elimination of some of these equilibria is only possible byimposing further structure on the worker’s beliefs off the equilibrium path.Equilibria with pooling wages are considered in the following section.2.5.2 Pooling Introductory WagesSuppose that the two types choose the same period 1 wage offer, w,1 == w1. The worker cannot differentiate between the two types from the15first period wage offer alone. The worker’s posterior period 1 belief that jobquality is low is, therefore, /3j(w1)= (1 — zp).Recall that the worker’s beliefs in period 2 are dependent, in part, upon therealisation of period 1 job quality. If this was high, the workers know the firmis good. The period 2 wage offer is then (1 — p)L—just satisfying the worker’sparticipation constraint.If job quality in the previous period was low and the firm stays, then theworker knows that either the firm is bad and r = 0, or the firm is good.8The worker, therefore, updates the period 1 belief, (1 — zp), in the light ofthis information. Let this posterior belief be denoted (1 — z’p).9In order to satisfy the worker’s participation constraint, the period 2 wagemust be greater than, or equal to, (1— z’p)L. This is the most efficient poolingwage from the firm’s point of view. The wage must, however, be less thanL—given any beliefs, the worker will accept this offer. There is a continuumof pooling equilibrium wage offers between these two bounds. Equilibria withwage offers in this interval can be supported by the belief that any other offercomes from the bad type.Similarly, there is a continuum of pooling period 1 wage offers. The lowestintroductory wage at which the two types can pool is (1 — zp)L; any lower wagedoes not meet the worker’s participation constraint. The highest introductorywage that can be offered is dependent upon the period 2 offer. Suppose, theperiod 2 pooling wage is (1— z”p)L—the lowest possible pooling wage in thatperiod. Then the bad type’s equilibrium payoff is:(12) ir(w1,B) + air((1 — z’p)L,B) + (1 — a)R.If it deviates and the worker believes that the deviation comes from the badtype, its payoff is:(13) ir(L, B) + air(L, B) + (1 — a)R.8Recall that r is private information to the firm.9From Bayes’ rule, it can be shown that z’ = z(1— p)/(z(1—p) + (1 — z)a), such thatz’ > z if (1— p) > a.16Let W define the wage that equates expressions (12) and (13). By definition,this wage exceeds L, but is less than WB.The equilibrium payoff for the good type is:(14) Tr(w,1,G) + (1 — p)ir((l — z’p)L, G) + p7r((l — p)L, G).If it deviates and the worker believes that the deviation was made by the badtype, its payoff is:(15) ir(L, G) + (1 — p)ir(L, G) + pir((1 — p)L, G).Let W define the wage that equates expressions (14) and (15). It too exceedsL, but is less than WG.Lemma 2 The introductory wage W exceeds W if, and only if, a < (l—p).Proof The proof is similar to that of Lemma 1.The upper bound to the set of pooling equilibria is strictly less than theminimum of W and Pooling equilibria with introductory wages in theinterval w1 e [(1 — zp)L, [min(W, W)]), can be supported by the belief thatany offer outside this range comes from the bad type.2.6 RefinementThe multiple equilibria can be characterized according to the good type’speriod one wage offer (see Figures 2 and 3). In the parameter space a < (1 —p),the introductory wage is pooling if (1 — zp)L w1 < W; and type-revealingif WB w <WG. If a> (1—p), the introductory wage is pooling such that(1 — zp)L w1 <Wa.The multiplicity of equilibria stems from the indeterminacy of worker’s beliefs off the equffibrium path. Bayes’ rule restricts the worker’s beliefs along theequilibrium path but not off it. The Cho-Kreps Intuitive Criterion can reducethe number of equilibria by restricting off-equilibrium beliefs as follows. Consider a candidate equilibrium, and an out-of-equilibrium offer. Suppose one of17(1—zp)L W WB WGI I [\‘\\\\‘\\\\\\)o (1—p)L LFigure 2: Equilibrium Wages, where a < (1—p)(1—zp)L W WG WBI I Io (1—p)L LFigure 3: Equilibrium Wages, where a (1—p)18the two types obtains a smaller payoff from the deviation than from the candidate equilibrium, regardless of beliefs. Further suppose, that if the workerbelieves this deviation was made by the other type, then that type’s payofffrom the deviation is higher than from the candidate equilibrium. Then, theequilibrium does not satisfy the Intuitive Criterion.This refinement reduces the set of separating equilibria to a singleton. Tosee this, fix a candidate separating equilibrium such that the good type offers aperiod 1 wage greater than WB, and the bad type offers L. Suppose there is adeviation to a wage greater than WB, but less than the good type’s equilibriumoffer. Given any beliefs, this deviation yields the bad type a lower payoff thanits equilibrium strategy (this follows from the definition of WB). Therefore, theworker should believe that the deviation was made by the good type, and acceptthe offer. But, given these beliefs, the good type obtains a higher payoff bydeviating than from the conjectured equilibrium, overturning that equilibrium.By this reasoning, there are no equilibria in which the good type’s period 1 wageoffer exceeds WB.A two-period version of the refinement also eliminates some of the poolingequilibria. Fix a pooling equilibrium in which the two types offer w1, andconsider a deviation to a higher wage. The equilibrium payoff for the bad typeis:(16) 7r(w1,B) + air(w,2,B) + (1 — a)R.If it deviates and the worker believes that the deviation was made by the goodtype, its payoff exceeds the equilibrium payoff. I assume that, should this occur,the worker maintains this belief, regardless of both the realisation of period 1job quality and the period 2 wage offer.1° For these particular beliefs, the badtype’s payoff is:(17) ir(w, B) + air((1 — p)L, B) + (1 — a)R.‘°The Intuitive Criterion is usually applied to one-period games. By specifying beliefs inthis way, I have extended the one period reasoning to the two-period case. This extendedcriterion is equivalent to the requirement that equilibrium outcomes remain such after theremoval of strategies that are strictly inferior to an associated equilibrium strategy.19Let WB define the wage that equates expressions (16) and (17).The equilibrium payoff for the good type is:(18) ir(w,, G) + (1 — p)ir(w2,G) + p-((l — p)L, B).If it deviates, and the worker believes that the deviation was made by the goodtype (regardless of period 1 job quality or the period 2 offer), its payoff is:(19) ir(w, G) + ir((1 — p)L, G).Let WG define the wage that equates expressions (18) and (19).Regardless of worker beliefs, a deviation greater than Wq, yields a lowerpayoff to the type q firm than the conjectured equilibrium. If, and only if,a < (1—p), the wage WG exceeds WB (the proof is similar to that of Lemma1). If there is a deviation in the interval [WB, WG), the worker infers that thedeviation was made by the good type. Given these beliefs, the good type obtainsa higher payoff from deviation than from its equilibrium strategy, overturningthe conjectured pooling equilibrium. In this parameter space, no pooling equilibrium satisfies the (extended) Intuitive Criterion.”On the other hand, where a (1—p), the wage WB is weakly greater thanWG. Any deviation that yields the bad type a lower profit than the conjectured equilibrium, also yields a lower profit for the good type. In this case, theIntuitive Criterion does not reduce the set of equilibria.Proposition 11. If a < (1—p), and beliefs satisfy the (extended) Intuitive Criterion, thenthere exists a unique outcome such that w1 = WB > w, = L, andw2 = (1—p)L<w2=L.. If a > (1—p), there is a continuum of pooling introductory wage equilibriasuch that (1 — zp) < W. In the second period, if period 1 jobquality was low, the wages are (1 — z’p)L W2 < L; and if it was high,W2 = (1 —p)L.strategy equilibria can be ruled out in a similar fashion.202.7 DiscussionIn the separating equilibria, the introductory compensating differentials havethe opposite sign to the usual Smith/Rosen case; workers with lower expectedquality jobs are paid less—not more. In the first period, the market rewardsworkers with more desirable jobs.’2Separation is possible in this game because the good type is more likely tostay than the bad type. As noted earlier, this occurs only if the good type’sfixed costs are lower. Assumption (Al) simplifies the analysis by ensuring that(even in a pooling equilibrium) the bad type exits with probability (1 — a); but,the good type always stays. If I had assumed that the good type’s probabilityof exit was (weakly) greater than the bad type’s, all equilibria would have beenpooling. In these equilibria, workers cannot distinguish between the two typesprior to employment, and the introductory wages are independent of job quality.Notice that, typically the worker does not receive reservation utility, (1 — zp)L.In fact, the upper bound to the set of pooling equilibria exceeds the bad type’spost-separation offer, L.In both types of equilibria, the firm makes wage offers according to thelabour demand schedule. Hence, signalling also impacts on employment levels;any offer above the market clearing level causes introductory-period employmentto fall. For example, in the separating equilibrium of Proposition 1, the goodtype employs fewer workers than if its type were known: g(WB) <g((l — p)L).The intuition behind the model is straightforward. Introductory wage offersaffect inexperienced worker’s beliefs about job quality. Realising this, a firm mayresist cutting wage offers to market clearing levels, impacting on employment.The model can be generalised in a number of ways. For instance, the gamecan easily be extended to the multi-firm case. Suppose that each firm has apool of labour (perhaps defined by geographical location or by worker type) towhich it alone makes wage offers. Then, with respect to its own pool, each firm‘2Note that, in a separating equilibrium, wages at a good firm fall through time. I haveabstracted from human capital and monitoring considerations that usually ensure an upwardsloping wage profile.21behaves as described above. Jobs at good firms do not pay lower introductorywages.’3It is also straightforward to allow for more than two periods. As in the two-period game, separation can occur through the introductory wage offer. In anypost-separation period, the wage offered is (1 —p)L for the good type, and L forthe bad type. Alternatively, the equilibria may be pooling. Then, the workersupdate their beliefs in each post-introductory period exactly as iu period 2 ofthe above game.Recent work by Mester (1992) suggests another direction in which the modelcould be extended. In his multi-period, product-market game, the firm has private information which varies through time. This causes “perpetual signalling”—(through pricing) separation can occur in post-introductory periods. In futurework, I intend to address the analogous job quality signalling case.2.8 ConclusionsIn this chapter, I have demonstrated, by analysing a specific two-type, two-period example, that introductory wage offers can inform workers about thequality of experience jobs. In this game, the good type offers higher expectedquality jobs than the bad type. If the good type is less likely to exit, the gamehas many equilibria with separating wages. In each of these, the introductorycompensating differentials have the opposite sign to the usual Smith/Rosencase: higher expected quality jobs pay more, rather than less. I have alsoshown that there are pooling equilibria. In this case, the introductory wages areindependent of job quality. I have demonstrated that, for some parameter space,the (extended) Intuitive Criterion renders a unique separating equilibrium.The intuition captured by the model is straightforward. Wage offers affectinexperienced worker’s beliefs about job quality. Realising this, the firm resistscutting wage offers to market clearing levels.This paper has important implications for empirical work on compensating‘3However, allowing a firm to compete for workers in another firm’s pool considerablycomplicates the analysis.22differentials. Workers with experience jobs are not paid the Smith/Rosen compensating differentials associated with search jobs. Hence, it is unsurprising thatmany studies report mixed support for the theory (for example, Smith (1976),Brown (1980) and Meng (1989)).23Chapter 3Compensating Differentials: Some Canadian Self-Report Evidence3.1 IntroductionThe theory of compensating differentials, originally due to Adam Smith [89]and subsequently formalised by Rosen [81], predicts that less desirable jobs paya wage premium. Smith’s own example was the executioner: a distasteful jobthat paid more than comparable trades.In this chapter, I present Canadian evidence that tests and supports thetheory of compensating differentials for a variety of job characteristics. PreviousCanadian studies, such as those by Meng [64 and 65], Martinello and Meng [62],and Cousineau et al [22] have found support for the theory for risk of injury ordeath. But no previous research has found Canadian evidence of compensatingwage payments for non-hazardous job characteristics.The data used in this study come from the National Survey of Class Structureand Labour Process in Canada (NSCS). These unique data, which are cross-sectional and relate to 1981 incomes, provide detailed self-report informationabout the respondents’ job quality and personal characteristics.Meng’s [64] study of compensating differentials also used this data set. However, to measure job quality, Meng used occupational-trait data developed byStatistics Canada and Employment and Immigration Canada, rather than theself-report information contained in the Survey. The use of occupational-traitdata is quite common in the compensating differential literature; this approachhas also been used by (among others) Brown [17], Garen [35] and Biddle andZarkin [15]. Unfortunately, as Smith [92] has noted, occupational-trait dataprovide information on averages across broad occupational categories and industries, introducing error into the job quality variables. Using self-report dataavoids this problem.24After controlling for personal characteristics, I find evidence of compensating differentials for working with data, working with hands, bureaucratic procedures, and responsibility over other workers. I also find evidence of differentialsfor the control of hours and pace. Although these characteristics are generallyheld to be desirable, I find that they are associated with higher, rather thanlower, earnings. That is, the coefficients have the “wrong” signs. I find no evidence of differentials for working with people, working with machines and thefreedom to design work.Most researchers have used Ordinary Least Squares (OLS) to estimate anhedonic earnings or wage equation. However, Viscusi [98], Garen [35], Biddleand Zarkin [15] and Kostiuk [53] have argued that job quality is endogenous. Ifjob quality is a normal good, richer workers will choose higher quality jobs. Thiseffect causes the estimated coefficients on the job quality variables to be biaseddownwards. In studies with a small number of continuous job characteristics,endogeneity can be easily dealt with by using instrumental variables. However,in cases in which there are many, binary job quality variables—such as thisone—this is impossible. For this reason, I first construct and then instrumentfor an aggregate job quality index in the earnings equation. I also estimate amore general model, suggested by Garen [35], in which productivity is a functionof job quality. The results confirm the existence of compensating differentialsfor undesired job characteristics.The theory of compensating differentials is based upon the notion of competitive labour markets. The observed relationship between job quality and wagesis the result of sorting by both workers and firms. An alternative theory of wagedetermination is that the labour market is characterised by non-competitive behaviour, that prevents matching. This may be particularly true of the unionisedsector—one interpretation of union behaviour being that they reduce the variation in their members’ earnings—but may also hold for non-unionised workers.For both types of workers, firm’s pay evaluation schemes often dictate wages,rather than allowing competitive forces to prevail. An important finding of thischapter is that labour markets are sufficiently competitive for compensating dif25ferentials to be important determinants of earnings, regardless of the gender orunion status of the workers.The rest of the chapter is organised as follows. In the following section, Ireview the related literature. In Section 3.3, I set out the empirical model andprovide details of the data. I present my results in Section 3.4. In Section 3.5,I draw some conclusions, and make some suggestions for subsequent research.3.2 Related LiteratureA number of researchers have surveyed the empirical literature on compensating differentials. Smith [92], Rosen [82], Digby and Riddell [24], Jones-Lee[47], Moore and Viscusi [69] and Viscusi [104] all provide excellent reviews.Many studies have found a positive relationship between earnings and bothfatality and injury rates. Examples include the US studies by Viscusi [100], andOlson [75]; and the Canadian studies by Cousineau et al [22], and Martinelloand Meng [62].Outside the value of workplace safety literature, the support for compensating differentials in multi-characteristic studies is weaker. In particular, Smith[92] and Brown [17] have found little support on US data. In the only Canadianstudy (on non-hazardous characteristics), Meng [64] has found similar resultsfor Canada. On the other hand, the US study by Lucas [59] is broadly supportive, as is one by McNabb [63] on UK data. All of these studies have used maleworkers; appropriate data for females are very scarce.A number of studies have examined the impact of unionisation on compensating differentials. Since unions are in a better position to monitor job qualitythan individual workers, one might expect larger differentials for unionised workers. But unions also tend to reduce the variance in wages of their members—creating an offsetting effect. A number of US researchers, including Duncanand Stafford [30], Olson [75] and Falrris [31] have found that unionised workersdo receive larger compensating differentials. However, using British data Mannand Psacharopoulos [61] have found that they receive smaller premia; as hasMeng [64] using Canadian data.26Like this study, Meug [64] also used data from the NSCS. This survey provides detailed, self-report iuformation about the job quality and personal characteristics of the respondents. Meng used only the information on personalcharacteristics. To capture job quality, he used occupational-trait data developed by Statistics Canada and Employment and Immigration Canada. Theuse of occupational-trait data is quite common in the compensating differential literature; the approach has also been used by (among others) Brown [17],Garen [35] and Biddle and Zarkin [15]. Unfortunately, as Smith [92] has noted,occupational-trait data provide information on averages across broad occupational categories and industries, introducing error into the job quality variables.Within an occupational title, the tasks involved can be extremely varied. For example, the term “general labourer” can cover a variety of jobs. In some circumstances, a labourer may have to work with machines, in others (s)he may not.Job quality may also vary greatly with location; for example, the characteristicsof a police officer’s job depend heavily on the allotted “beat”. Quality may alsovary with industry. An engineer in the mining industry may experience differentworking conditions from one in the service sector. Using self-report data avoidsthese problems since the data are (by definition) job specific, but at the costof objectivity. This could be a problem if respondents with lower pay (falsely)report dissatisfaction with non-pecuniary job characteristics—introducing a spurious correlation between wages and job quality.Brown [17] and Duncan and Holmlund [29] have noted the omitted variable bias caused by unobserved worker ability. Ability is likely to be negativelycorrelated with desirable job characteristics, causing the estimates of the compensating differentials to be biased downwards. Estimating fixed effect wageequations mitigates this bias, but is impossible for the (cross-sectional) NSCSdata.Most researchers have used Ordinary Least Squares (OLS) to estimate anhedonic earnings or wage equation. Viscusi [98], Garen [35], Biddle and Zarkin[15] and Kostiuk [53] have argued that the job quality variables—which appearon the right-hand side—may be endogenous. If job quality is a normal good,27richer workers will choose higher quality jobs. This effect causes the estimatedcoefficients on the job quality variables to be biased downwards. With a smallnumber of job characteristics, a simple way to deal with this problem is to instrument for the endogenous variables in the earnings equation. Unfortunately,this is impossible if there are many, dichotomous job quality variables—as thereare in this case. (Hence, I instrument for an aggregate job quality index in theearnings equation.)Garen [35] has noted that a worker’s productivity may be a function of jobqnality. For example, some individuals may possess an unobserved characteristic(perhaps, dexterity) that makes them particularly productive when workingwith their hands. In this case, the instrumental variables technique yields biasedestimates. Garen [35] has shown that consistent estimates can be obtained byconstructing the predicted residnals from an OLS job quality equation, andincluding them, together with an interaction term, into the wage equation.3.3 Model and DataThis study is based upon the following earnings equation:rn z(20) lnY=ao+ZnjFCj+ Ei=1 j=m+1where Y denotes hourly earnings, PC a series of personal characteristics; anda series of job quality variables. A fnll list of these variables and their meansare given in Table 1.14All data are taken from the NSCS, a cross-sectional survey that contalnsinformation on approximately 3,000 respondents. The survey was carried outby Canada Facts, who conducted face-to-face interviews. I exclude workers over64 and under 18, and anyone with non-positive 1981 earnings or hours workedper week. After removing the self-employed, who were not asked many of thejob quality questions, and those who did not work year round, the final sampleis 993.‘4Means by sex and union status together with the means of the variables used to instrumentfor job quality are given in Appendix 1.28Table 1:Variable definitions and meansName Definition MeanIndividual characteristicsANY Annual incomeWY ANY per weekY WY per hour usually workedin Y Natural log of YEDUC Years of educationAGRADE Attained GRADE = 1; otherwise = 0ASOME Attained SOME = 1; otherwise = 0AHIGH Attained HIGH = 1; otherwise = 0ACOLL Attained COLL = 1; otherwise = 0ABACH Attained BACH = 1; otherwise = 0APOST Attained POST = 1; otherwise = 0EXP Experience in yearsEXP2 EXP squaredUNION Union member = 1; otherwise = 0TEN Years of tenure with present employerTEN2 TEN squaredBIL Bilingual = 1; otherwise = 0SEX Male = 1; otherwise = 0LocationATL Atlantic =1; otherwise = 0QUE Quebec 1; otherwise = 0ONT Ontario = 1; otherwise = 0PRA Prairies = 1; otherwise =0BC British Columbia = 1; otherwise = 0CITY Community > 100,000 = 1; otherwise = 0Job characteristicsCDESNCHRSCPACERESPBURHANDSPEOPLEDATAMACHINESQControl design of work = 0; otherwise = 1Control hours of work = 0; otherwise = 1Control pace of work = 0; otherwise = 1Responsibility over others =1; otherwise = 0Bureaucratic procedures = 1; otherwise = 0Work with hands = 0; otherwise = 1Work with people = 0; otherwise = 1Work with data = 1; otherwise = 0Work with machines = 1; otherwise = 0Job quality index$22,685$435.08$11.292.2712.910.090.170.180.360.130.0717.95481.680.467.78125.520.190.540.080.290.360.150.120.620.490.720.530.380.500.470.340.520.754.7029The hourly earnings variable is constructed as follows. Respondents to theNSCS were asked to estimate their personal income in 1981, and the numberof hours usually worked per week. Having removed from the sample those whodid not work year round, I calculate the earnings per hour usually worked.The human capital variables are years of education and six dummies reflecting the highest level of schooling achieved. The dummy variables are asfollows: grade school diploma or less (AGRADE), some high school (ASOME),completed high school (AffiGH), college/vocational school (ACOLL), bachelor’sdegree (ABACH), and postgraduate/professional degree (APOST).As mentioned earlier, the unique feature of these data is the quantity of jobquality information. The respondents were asked a series of questions abouttheir self-control in the job including whether they could: design their ownwork; decide their hours worked; and adjust their pace of work. They were alsoasked about their responsibility for other workers and the control others haveover them in the form of written bureaucratic procedures. In addition, therewere a battery of questions about “job complexity”: whether the job requiredworking with hands, people, information and machines. These last questionsare particularly interesting. Some researchers have looked at the compensatingdifferentials paid for working with machiues; but not for these other aspects ofjob complexity.The drawback to using this self-report data is that the information is basedupon individual assessments of job quality. They are, by definition, subjective.The sign of the coefficients on the job quality variables is a controversial(and much discussed) issue in the literature. The coefficients should reflectthe preferences of the marginal worker; if that worker finds the characteristicundesirable, the coefficient should be positive. The researcher may have somea priori beliefs about the marginal worker’s preferences—perhaps based uponhis/her own preferences—but these are very imprecise. A characteristic that Ifind desirable, such as bureaucratic procedures, the marginal worker may findattractive. The issue is further clouded by the possibility of omitted variablebias. Some of the job characteristics could be correlated with some unobserved30aspects of ability, biasing the coefficients. For example, in some firms’ jobevaluation plans, “autonomy” is associated with higher payments in order toreward more able workers. Hence, the coefficients on the self-control variablesmay be biased.I construct the job characterisitics dummies such that each variable takes thevalue one in what I expect to be (a priori) the undesirable state. The aggregatejob quality index is simply the sum of the job characteristic variables; the indexhas a maximum score of 9 and a minimum of zero. I also constructed a jobquality index using the coefficients from an OLS earnings equation as weights.However, the results are largely similar to those using the simple (unweighted)index, and so are not reported.3.4 ResultsThe results are presented in Tables 2 and 315The first column of Table 2 includes only the personal characteristics of therespondents. Column (2) includes these and the job quality variables. Thethird column includes the same personal characteristics as Column (2), butthe job quality variables are replaced with the aggregate index, Q. The fourthcolumn includes an instrumental variable for the job quality index, Q. The OLSjob quality regression used to construct the instrument is shown in Appendix 1,Table C. To capture the impact of job quality on marginal productivity (Garen’smodel), the fifth column includes the aggregate index Q, together with thepredicted residuals from the job quality equation, ñ and an interaction term,Q . ñ. The final column includes the same variables but uses White’s [109]correction for heteroskedasticity.From the first column of Table 2, it is apparent that the variables for thepersonal characteristics are generally significant at the 10% level, and have theexpected sign. However, the location dummies are insignificant, apart fromthose for living in British Columbia (BC) and living in a city (CITY), which‘I have used sample weights that reflect the population by region and household size. Theresults from unweighted regressions are similar.31Table 2:Regression equations; dependent variable in YVariable (1) (2) (3) (4) (5) (6)EDUC 0.038 0.030 0.038 0.041 0.042 0.036(5.298) (4.246) (5.302) (5.648) (5.718) (4.958)AGRADE -0.241 -0.160 -0.241 -0.228 -0.227 -0.256(-3.460) (-2.321) (-3.453) (-3.278) (-3.271) (-3.877)ASOME -0.166 -0.141 -0.167 -0.169 -0.170 -0.185(-3.208) (-2.791) (-3.208) (-3.276) (-3.295) (-3.485)ACOLL 0.015 -0.218 0.138 -0.018 -0.016 -0.012(0.332) (-0.498) (0.312) (-0.403) (-0.361) (-0.261)ABACH 0.119 0.045 0.121 0.187 0.193 0.234(1.919) (0.721) (1.937) (2.829) (2.926) (3.331)APOST 0.261 0.167 0.263 0.330 0.323 0.311(3.184) (2.060) (3.196) (3.874) (3.802) (3.807)EXP 0.023 0.021 0.023 0.023 0.023 0.018(5.032) (4.643) (5.027) (4.960) (4.922) (3.706)EXP2 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000(-3.297) (-3.236) (-3.290) (-3.124) (-3.130) (-2.325)UNION 0.112 0.182 0.109 0.006 0.005 -0.041(3.702) (5.464) (3.457) (0.126) (0.108) (-0.885)TEN 0.023 0.022 0.022 0.022 0.007 0.015(4.009) (3.912) (4.000) (3.763) (0.884) (1.630)TEN2 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001(-3.518) (-3.527) (-3.509) (-3.285) (-3.203) (-3.226)BIL 0.058 0.038 0.058 0.073 0.077 0.082(1.352) (0.911) (1.359) (1.694) (1.804) (1.862)SEX 0.307 0.267 0.307 0.313 0.315 0.332(9.980) (8.569) (9.979) (10.19) (10.29) (10.39)32Table 2 cont.Variable (1) (2) (3) (4) (5) (6)ATL -0.071 -0.052 -0.071 -0.085 -0.072 -0.052(-1.198) (-0.896) (-1.204) (-1.441) (-1.221) (-0.839)QUE 0.001 0.012 0.001 -0.002 -0.006 0.006(0.014) (0.294) (0.012) (-0.055) (-0.147) (0.145)PItA 0.053 0.059 0.053 0.060 0.059 0.030(1.147) (1.317) (1.151) (1.302) (1.272) (0.605)BC 0.093 0.100 0.092 0.089 0.093 0.133(1.853) (2.048) (1.850) (1.784) (1.878) (2.598)CITY 0.072 0.076 0.072 0.079 0.083 0.063(2.275) (2.465) (2.280) (2.513) (2.639) (1.957)CDESN-0.030(-0.877)CHRS-0.083(-2.177)CPACE-0.076(-2.380)RESP 0.104(3.179)BUR 0.052(1.675)HANDS 0.084(2.569)PEOPLE-0.580(-0.166)DATA 0.124(3.757)MACHINES 0.016(0.447)33Table 2 cont.Variable (1) (2) (3) (4) (5) (6)Q 0.003 0.107 0.122(0.293) (2.373) (2.428)Q 0.127(2.849)i-0.067 -0.048(-1.237) (-0.803)Q• -0.015 -0.017(-2.298) (-2.387)Q . TEN 0.003 0.002(2.431) (1.182)CONSTANT 1.147 1.232 1.131 0.556 0.665 0.690(10.21) (9.902) (9.154) (2.361) (2.797) (2.714)N 993 993 993 993 993 993li2 0.350 0.393 0.349 0.355 0.361 0.344Note: t-statistics in bracketshave positive coefficients. Meng [64] found broadly similar results for thesevariables.The inclusion of the job quality variables, Column (2), reveals some support for the theory of compensating differentials, but not for all job attributes.Support for the self-control variables is particularly weak. Control of hours(CHRS) and control of pace (CPACE) are significant at the 10% level, but thecoefficients are incorrectly signed. The other self-control variable, for designingwork (CDESN), is insignificant and incorrectly signed. One interpretation ofthese results is that the marginal worker does not require compensation for self-control characteristics. Alternatively, these variables may be correlated withsome omitted variable, such as ability, biasing the estimated coefficients.Support is stronger for the responsibility (RESP) and bureaucratic procedures (BUR) variables: the coefficients on both of these have the anticipated(positive) sign. Both are significant at the 10% level.The dummy variables for working with hands (HANDS), and for workingwith data (DATA) are also significant at the 10% level. As expected, the coefficients on these variables are positive. There appear, however, to be no dif34ferentials for working with people (PEOPLE) or working with machines (MACHINES).Unfortunately, including the job quality index instead of the individual characteristics, Column (3), contradicts this support for the theory of compensatingdifferentials. The index variable is insignificant, and the coefficient rather small.However, the instrumental variable, included in Column (4) is significant, witha correctly signed coefficient. This suggests that the OLS estimates were biasedby endogeneity and that job quality is a normal good: richer workers choosehigher quality jobs.The specification proposed by Garen [35] is shown in Column (5). Thisallows for the fact that productivity is likely to be a function of job quality. Theestimated job quality coefficient is smaller than in the instrumental variablecase, but much larger than the OLS estimate. The negative sign on the ñ termconfirms that job quality is a normal good: richer workers choose better jobs. (Ifthe coefficient had been positive, lower job quality would have been associatedwith higher earnings—suggesting that job quality was an inferior good.) Thenegative sign on the ñ term indicates that workers with unobserved returnsto low quality jobs choose more desirable jobs. The interaction term involvingjob quality and tenure, Q . TEN, has a positive coefficient: more senior workersreceive larger compensating payments for undesirable characteristics.The Breusch-Pagan test indicates that the null hypothesis of homoskedasticity is rejected at the 5% level. The heteroskedasticity-corrected regression,Column (6), has a slightly larger job quality coefficient, but gives broadly similarresults.The four columns of Table 3 give the (heteroskedasticity-corrected) resultsfor the male, female, union and non-union sub-samples respectively. The results in Columns (1) and (2) reveal substantial differences between the earningsequations for male and female workers.’6 In common with many other studies,I find that the returns to years of education are lower for females; however,16An F-test of the hypothesis that there is no difference between the male and femalecoefficients is rejected at the 5% significance level.35the coefficients on the edncation level dummies are larger in absolnte value.’7Unlike for females, job quality is insignificant for males. However, in both cases,job quality has important interaction effects—even if job quality does not havedirect effects on earnings, it has indirect effects.Columns (3) and (4) reveal considerable differences between the earningsequations for union and nonunion workers.18 Since workers self-select into unioncoverage, I include a selectivity term (A) in the earnings equations. This isconstructed from a probit estimate of union status (shown in Appendix 1). Ifind that for the union sector the regression constant is larger, and that thereturns to years of education (EDUC) and tenure (TEN) are smaller, as is thewage premium for being male (SEX). Hence, there is some support for thenotion that unions reduce the variance in their members’ earnings. Job qualityis insignificant for union workers, but significant for non-union workers. Agaln,job quality has important interaction effects.It seems that the theory of compensating differentials—based upon the notion of competitive labour markets—is an important determinant of earnings formales, females, union and non-union workers. Even though some labour markets may be non-competitive, the sorting process betweeen workers and firms issufficiently strong for the theory to be meaningful.In the previous chapter, I have argued that the Smith/Rosen theory of compensating differentials is sensitive to the assumed information structure. Ifworkers have difficulty learning job quality, firms may signal this private information through wage offers. In these circumstances, more desirable jobs maypay higher, rather than lower, wages. In an attempt to analyse this issue, I estimate compensating differentials for sub-samples of senior workers. The resultsfor workers with greater than 1, 3 and 5 years of tenure are shown in Appendix1, Table D, Columns (1), (2) and (3) respectively. Because more senior workersare better informed about job quality, their wages may be less influenced bysignalling. However, I find little evidence of this; the estimated compensating“See Gunderson and Riddell [38] for a review of the literature on male/female earningsdifferentials.15A test of the hypothesis that there is no difference between the union and nonunioncoefficients is rejected at the 5% significance level.36Table 3:Regression equations; dependent variable ln YVariable (1) (2) (3) (4)EDUC 0.040 0.030 0.023 0.048(4.214) (2.539) (2.990) (3.299)AGRADE -0.244 -0.269 -0.099 -0.376(-3.321) (-1.971) (-1.241) (-3.454)ASOME -0.126 -0.216 -0.146 -0.200(-1.894) (-2.351) (-1.938) (-2.716)ACOLL 0.011 0.001 0.032 -0.027(0.178) (0.011) (0.532) (-0.398)ABACH 0.147 0.284 0.234 0.164(1.626) (2.767) (2.784) (1.532)APOST 0.212 0.454 0.300 0.255(2.162) (3.016) (3.124) (2.083)EXP 0.029 0.008 0.010 0.023(4.799) (0.973) (1.727) (3.067)EXP2 -0.000 -0.000 -0.000 -0.000(-3.245) (-0.664) (-1.164) (-1.847)UNION -0.056 0.082(-0.961) (1.332)TEN 0.008 0.034 -0.005 0.026(0.827) (1.731) (-0.536) (1.960)TEN2 -0.001 -0.001 -0.000 -0.001(-2.366) (-2.747) (-0.345) (-3.789)BIL 0.138 -0.017 0.128 0.049(2.576) (-0.230) (2.298) (0.759)SEX 0.241 0.387(5.985) (7.649)37Table 3 cont.Variable (1) (2) (3) (4)ATL -0.043 -0.076 -0.067 -0.032(-0.661) (-0.659) (-1.100) (-0.332)QUE -0.034 0.067 -0.010 -0.038(-0.697) (0.878) (-0.188) (-0.567)PRA -0.009 0.035 0.000 0.02 1(-0.015) (0.454) (0.003) (0.309)BC 0.241 -0.037 0.144 0.110(3.559) (-0.477) (1.972) (1.511)CITY 0.026 0.094 0.033 0.087(0.702) (1.592) (0.835) (1.862)Q 0.054 0.109 0.047 0.071(0.927) (1.778) (1.007) (1.775)0.011 0.011 -0.066 0.059(0.163) (0.127) (-1.094) (0.790)Q -0.015 -0.026 -0.000 -0.032(-2.112) (-1.894) (-0.035) (-3.017)Q . TEN 0.002 0.001 0.003 0.003(1.266) (0.212) (2.374) (1.181)A 0.052 -0.034(1.183) (-0.530)CONSTANT 1.215 0.851 1.358 0.659(4.024) (2.639) (5.048) (1.666)N 569 424 449 5440.301 0.230 0.272 0.377Note: t-statistics in brackets38differentials for cut-off points less than 5 years ar.e similar to those for the fullsample. For more senior workers, the job quality terms are insignificant.3.5 ConclusionsIn this chapter, I have presented Canadian evidence that tests and supportsthe theory of compensating differentials for a variety of job characteristics. Previous Canadian studies, such as those by Meng [64 and 651, Martinello andMeng [62], and Cousineau et al L22] have found support for the theory for riskof injury or death. This is the first study, however, to have found Canadianevidence of compensating wage payments for non-hazardous characteristics.By exploiting the self-report information contained in the NSCS, I haveavoided using the occupational-trait data used by Meng [64]. Occupational-trait data provide information on averages across broad occupational categoriesand industries, introducing error into the job quality variables.After controlling for personal characteristics, I have found evidence of compensating payments for working with data, working with hands, bureaucraticprocedures, and responsibility over other workers. I have also found evidence ofdifferentials for the control of hours and pace, though the coefficients are incorrectly signed. No evidence of differentials was found for working with people,working with machines or for the freedom to design work. I have also foundevidence supporting the theory using a job quality index constructed from thevarious job characteristics.39Chapter 4The Great Canadian Training Robbery4.1 IntroductionThe mismatch between the skill requirements of jobs and the educational attainments of workers has long concerned social scientists. In his seminal work,“The Great Training Robbery”, Ivar Berg [14] argued that overeducated workersmay be less productive than their less skilled counterparts because they becomebored with their jobs and lose motivation. A similar view was expressed by Freeman [33 and 34] who coined the phrase “the overeducated American”. Someresearchers, (among others) Kuttner [57], Picot et al [77] and Bluestone andHarrison [16] argued that the incidence of skill or educational mismatch (hereafter, I use the terms interchangeably) is increasing.’9 Industrial restructuringhas caused a number of traditional, medium- to high-skill jobs to disappear—the so-called “declining middle”—forcing many skilled workers into low-skill,service-sector jobs.Despite the widespread concern over educational mismatch, there have beenfew studies of its impact on earnings; data sets rarely contain information onboth the educational attainments of workers and the requirements of jobs. Asmall number of studies have used either US or Dutch data; but there havebeen no previous Canadian studies. Duncan and Hoffman [28], Rumberger [84],Hersch [43] and Sicherman [87] have estimated earnings (or wage) equationsincluding both the years of required schooling for the job, and the years of over-or undereducation. These researchers found strong evidence that the earningsof overeducated workers are greater than their counterparts with exactly the19The term “skill mismatch” is sometimes used to refer to differences between the kind ofskills required by firms and those attained by unemployed workers. In this chapter, however,I use the term to refer to differences between required and attained levels of education.40required level of schooling; and, that the earnings of undereducated workersare lower. (Hereafter, I shall refer to workers who are not mismatched, butare in jobs with the same educational requirements as “otherwise identical”.)Assuming earnings reflect marginal productivity, their results so not supportBerg’s [14] hypothesis: overeducated workers earn more—not less—than otherwise identical workers.In this chapter, I use Canadian data from the National Survey of ClassStructure and Labour Process in Canada (NSCS) to estimate the returns toeducational mismatch. I measure both education and educational mismatchin terms of discrete levels of achievement, rather than years of schooling. Ifind that the returns to over- and undereducation are sensitive to the level ofrequired education. There is evidence of positive returns to overeducation forjobs that require a university bachelor’s degree; but, in general, the returns areinsignificant. I find evidence of lower pay for undereducated workers with loweducation requirements. I also estimate separate equations for male and femaleworkers. Although I find that the results for the male sub-sample are similar tothe full sample, I find that the returns to over- and undereducation for femalesare insignificant for all levels of required education.For Canadian policy makers concerned with the returns to education, theseresults are a mix of good and bad news. On the one hand, the hypothesisthat overeducated workers have identical earnings to otherwise identical workers cannot (in general) be rejected, so the Canadian evidence does not supportBerg’s [14] clalm. In Canada, overeducated workers do not receive lower earnings than otherwise identical workers. On the other hand, those workers witheducational attainments in excess of requirements do not receive the full returnsto their attained education: their earnings would have been higher in a job withrequirements that matched their attalnments. This finding will concern policy makers: rising educational attalnments are insufficient to guarantee higherearnings for workers.Evidently, job requirements or “pigeonholes” are important determinants ofearnings—and particularly so for females. This chapter, therefore, offers some41support for Thurow’s [94] claim that marginal productivity resides in the job,rather than in the individual characteristics of the worker.The rest of the chapter is organised as follows. In the following section, Ireview the related literature. In Section 4.3, I discuss the incidence of educational mismatch. I set out the empirical model in Section 4.4; and present theresults in Section 4.5. I draw some conclusions in the final section.4.2 Related LiteratureAs already noted, the literature on skill mismatch can be traced back toIvar Berg [14]. He argued that if highly trained workers perform low-skill jobs,they become bored, and their productivity falls below that of their less-skilledcounterparts.Economists in the US and Canada became concerned about this issue in the1970s, when increases in educational attainments coincided with declines in themonetary returns to education. Freeman [33 and 34], coining the phrase “theovereducated American”, argued that the entry of the “baby-boom” generationinto the labour force caused an increased supply of highly educated workers. Atthe same time, the demand for these workers fell, forcing many of them into jobswith lower educational requirements. Dooley [27] identified similar demographicand demand-side changes in Canada.There have been a number of studies concerned with recent industrial restructuring which suggest educational mismatch has been increasing. Kuttner[57], Picot et al [77] and Bluestone and Harrison [16] have argued that manytraditional, high-skill jobs are disappearing—the so-called “declining middle”—displacing some skilled labour into low-pay, low-skill jobs in the personal servicesand retail trades (see Gunderson and Riddell [38]). As a result, there has beensome increase in wage polarisation.Despite the widespread concern over educational mismatch, there have beenvery few studies of its impact on earnings; data sets rarely contain information on both the educational attainments of workers, and the requirements ofjobs. A small number of studies have used either US or Dutch data; but, there42have been no previous Canadian studies. Duncan and Hoffman [28], Rumberger [84], Hersch [43] and Sicherman [87] have estimated earnings (or wage)equations including both the years of required schooling, and the years of over-or undereducation. These researchers found strong evidence that the earningsof overeducated workers are higher than the earnings of otherwise identical,non-mismatched workers; and, that the earnings of undereducated workers arelower. Assuming that earnings reflect marginal productivity, their studies donot support Berg’s [14] proposition.Most researchers have measured skill mismatch by years of schooling. Usinga (small) Dutch data set, Hartog [41] has estimated a model in which educationalattainments and requirements are measured by discrete “levels” of difficulty—such as graduating from high school or completion of an undergraduate degree.In the labour market, education is usually measured in this way—job advertisements usually specify education levels rather than years of schooling. Hartogfound that the returns to educational mismatch varied with required education;and that over- (under-) educated workers generally earned more (less) thanotherwise identical workers.Only two previous studies have examined male-female differences in the returns to skill mismatch. Using US data, Duncan and Hoffman [28] found positive, significant returns to overeducation for both sexes; and, negative, significant returns to undereducation for males. In a subsequent study, Hartog andOosterbeek [42] found similar results using a small sample of Dutch workers; but,in addition, found negative, significant returns to undereducation for females.In this chapter, using Canadian data, I find that for males the returns tounder- and overeducation vary with the level of required education. For females,I find that the returns to both under- and overeducation are insignificant for alllevels of required education.4.3 The Incidence of Educational MismatchThe data are taken from the NSCS, a cross-sectional survey that containsinformation on approximately 3,000 respondents. This survey was carried out by43Canada Facts, who conducted face to face interviews. I have excluded workersover 64 and under 18, and anyone with non-positive 1981 earnings or hoursworked per week. After removing the self-employed, and those who did notwork year round, the final sample is 993, of which 424 are female.These Canadian data are unique in that respondents were asked about boththeir attained education, and the educational requirements for the job. The following question was asked about educational attainments: “What is the highestlevel of education you have completed?”. The answers were categorised into sixclasses: grade school diploma or less (GRADE), some high school (SOME),completed high school (HIGH), college/vocational school (COLL), bachelor’sdegree (BACH), and postgraduate/professional degree (POST). The questionasked about required education was: “What type of formal schooling is nownormally required for people who do your type of work?”. Individuals weredefined as over- (under-) educated if their attained schooling was greater (less)than their required education. Since the second question inquired about schooling “now normally required”, the resulting variable arguably understates (overstates) the extent of overeducation (undereducation)—education requirementshave generally increased with time.This self-report approach is preferable to Rumberger’s [84] in which requiredschooling is measured by occupational means. Educational requirements canvary greatly within occupations. For example, the schooling requirements for apost as an economist can vary from an undergraduate degree for some privatesector jobs, to a Ph.D. for academic jobs. Furthermore, Rumberger’s measurewas calculated from the Dictionary of Occupational Titles (DOT), which reportsthree distinct measures of “General Educational Development”. Unfortunately,there is little consensus on how to aggregate these measures (see Cain andTreiman [18]).The incidence of educational mismatch in the NSCS is described in Table4. There are a number of striking features about these data. First, educationalmismatch is a common phenomenon; but, the incidence of overeducation (males30%, females 32%) is greater than the incidence of undereducation (males 24%,44Table 4:Incidence of Skill MismatchAttained RequiredGRADE SOME HIGH COLL BACH POST TOTALMales n=569GRADE 34 15 16 3 0 0 68SOME 19 32 39 13 3 0 106HIGH 8 23 47 10 11 4 103COLL 12 15 56 72 13 4 172BACH 0 2 4 10 51 5 72POST 0 0 2 0 18 28 48TOTAL 73 87 164 108 96 41 569Females n=J4GRADE 20 6 2 0 0 0 28SOME 23 26 23 6 2 0 80HIGH 8 7 54 9 3 0 81COLL 2 11 58 71 16 2 160BACH 0 3 7 8 40 1 59POST 0 0 3 1 6 6 16TOTAL 53 53 147 95 67 9 424females 17%). Second, attained schooling is generally within one educationlevel of required schooling; the incidence of skill mismatch outside this intervalis small. Third, for both sexes, the peak in required schooling is at the HIGHlevel, but the peak in attained education is at COLL. Fourth, the distributions of attained and required education are flatter for males; the job market isparticularly thin for females in the upper tail.4.4 Empirical ModelConsider the following earnings equation:(21) lnY=8’ PC+cV REQ+r’OVER+6’UNDER+e,where Y denotes hourly earnings, and PC a vector of personal characteristics(including a constant). The vector REQ contains one dummy variable for eachrequired education level. The vectors OVER and UNDER contain dummy45variables for over- and undereducation respectively; each variable correspondsto a specific required schooling level. It is, of course, possible to allow a dummyfor each combination of attained and required education. Recall from Table 4,however, that required education is rarely more than one education level fromattained education. Hence, such a model yields little additional insight.A full list of the variables and their means are given in Table 5. The hourlyearnings variable is constructed as follows. Respondents to the NSCS were askedto estimate their personal income in 1981, and the number of hours worked perweek. After removing those who did not work year round, the earnings per hourusually worked is calculated.If Berg’s proposition is correct, the coefficients on the overeducation dummiesshould be negative.4.5 ResultsThe results from the OLS regressions are presented in Table 6.20 The firstcolumn includes only the personal characteristics of the respondents and thecontrol variables. Column (2) includes these and the education variables fromEquation (21). The third column includes the same variables as Column (2),but the sample is restricted to male workers; and the final column includes onlyfemales.From the first column, it is apparent that the personal characteristic variables are generally significant at the 10% level (t-statistics in brackets), andhave the expected sign. The occupational dummies are all significant; as are allthe industry dummies, except public services (PUB). The dummy for retail andother services (RET) has a particularly strong (negative) impact on earnings.2’The location dummies are insignificant, apart from those for living in BritishColumbia (BC) or living in a community with a population greater than 100,000(CITY), which both have positive coefficients.2OJ have used sample weights that reflect the population by region and household size. Theresults from unweighted regressions are similar.21Using job characteristics (see the previous chapter) rather than occupational and industrydummies yields similar results.46Table 5:Variable definitions and meansName DefinitionPersonal characteristicsANY Annual incomeWY ANY per weekY WY per hour usually workedin Y Natural log of YEXP Experience in yearsEXP2 EXP squaredUNION Union member = 1; otherwise = 0SEX Male = 1; female = 0TEN Years of tenure with present employerTEN2 TEN squaredBIL Bilingual = 1; otherwise = 0IndustryEXTRMANUFDISTPUBINFORETOccupationPROFSEMISUPERSKILLSEMUNLocationATLQUEONTPRABCCITYProfessional = 1; otherwise = 0Semi-professional = 1; otherwise = 0Supervisory = 1; otherwise = 0Skilled trade = 1; otherwise = 0Semi-skilled and unskilled = 1; otherwise = 0Atlantic 1; otherwise = 0Quebec = 1; otherwise = 0Ontario = 1; otherwise = 0Prairies = 1; otherwise = 0British Columbia = 1; otherwise = 0Community > 100,000 = 1; otherwise = 0Males FemalesExtraction and construction = 1; otherwise 0Manufacturing = 1; otherwise = 0Distribution = 1; otherwise = 0Public services = 1; otherwise = 0Information services = 1; otherwise = 0Retail and other services = 1; otherwise = 0$27,556$528.49$13.082.4519.94561.840.511.009.33162.970.230.080.320.170.240.060.130.190.160.070.250.340.080.330.360.120.110.58$16,866$323.48$9.152.0615.56385.920.390.005.9380.790.150.020.100.100.400.160.210.140.170.030.240.410.090.250.350.180.120.6547Table 5 cont.Name Definition Males FemalesEducationEA Attained education in levels 3.44 3.49ER Required education in levels 3.32 3.30AGRADE Attained GRADE = 1; otherwise = 0 0.12 0.07ASOME Attained SOME = 1; otherwise = 0 0.17 0.17AHIGH Attained HIGH = 1; otherwise = 0 0.17 0.18ACOLL Attained COLL = 1; otherwise = 0 0.32 0.39ABACH Attained BACH = 1; otherwise = 0 0.13 0.14APOST Attained POST = 1; otherwise = 0 0.09 0.04REQ vectorRGRADE Required GRADE = 1; otherwise = 0 0.14 0.10RSOME Required SOME = 1; otherwise = 0 0.15 0.12RHIGH Required HIGH = 1; otherwise = 0 0.28 0.35RCOLL Required COLL = 1; otherwise = 0 0.18 0.23RBACH Required BACH = 1; otherwise = 0 0.17 0.17RPOST Required POST = 1; otherwise = 0 0.07 0.02OVER vectorOGRADE Overed. & required GRADE = 1; otherwise = 0 0.07 0.06OSOME Overed. & required SOME = 1; otherwise = 0 0.08 0.05OHIGH Overed. & required HIGH = 1; otherwise = 0 0.13 0.15OCOLL Overed. & required COLL = 1; otherwise = 0 0.02 0.02OBACH Overed. & required BACH = 1; otherwise = 0 0.03 0.02UNDER vectorUSOME Undered. & required SOME 1; otherwise = 0 0.02 0.02UHIGH Undered. & required HIGH = 1; otherwise = 0 0.09 0.06UCOLL Undered. & required COLL = 1; otherwise = 0 0.04 0.03UBACH Undered. & required BACH = 1; otherwise = 0 0.05 0.05UPOST Undered. & required POST = 1; otherwise = 0 0.02 0.0448Table 6:Regression equations; dependent variable in YVariable (1) (2) (3) (4)EXP 0.020 0.020 0.022 0.017(4.488) (4.505) (3.871) (2.330)EXP2 -0.000 -0.000 -0.000 -0.000(-4.214) (-3.390) (-3.017) (-1.637)UNION 0.095 0.107 0.069 0.140(2.840) (3.280) (1.707) (2.479)SEX 0.303 0.285(9.389) (9.088)TEN 0.018 0.017 0.017 0.016(3.241) (3.238) (2.540) (1.735)TEN2 -0.001 -0.001 -0.001 -0.000(-2.646) (-3.235) (-2.723) (-1.274)BIL 0.081 0.050 0.121 -0.066(1.946) (1.253) (2.463) (-0.949)EXTR 0.162 0.137 0.110 0.213(2.320) (2.041) (1.579) (1.264)DIST 0.083 0.035 -0.013 0.086(1.687) (0.731) (-0.259) (0.812)PUB 0.036 -0.056 -0.012 -0.140(0.791) (-1.268) (-0.238) (-1.546)INFO 0.132 0.042 0.105 -0.027(2.262) (0.735) (1.289) (-0.268)RET -0.204 -0.230 -0.156 -0.298(-4.081) (-4.799) (-2.620) (-3.306)49Table 6 cont.Variable (1) (2) (3) (4)PROF 0.511 0.158 0.179 0.114(10.65) (2.508) (2.212) (1.153)SEMI 0.487 0.285 0.242 0.311(10.72) (5.442) (3.664) (3.600)SUPER 0.285 0.167 0.172 0.082(4.140) (2.453) (2.169) (0.632)SKILL 0.233 0.142 0.162 0.092(6.125) (3.620) (3.220) (1.434)ATL -0.075 -0.073 -0.107 -0.016(-1.300) (-1.333) (-1.531) (-0.175)QUE -0.025 -0.007 -0.063 0.070(-0.633) (-0.170) (-1.288) (1.098)PRA 0.035 0.032 -0.001 0.040(0.775) (0.723) (-0.013) (0.596)BC 0.110 0.089 0.223 -0.054(2.230) (1.870) (3.676) (-0.708)CITY 0.068 0.051 0.012 0.124(2.202) (1.690) (0.343) (2.374)RGRADE -0.314 -0.312 -0.322(-3.944) (-3.151) (-2.323)RSOME -0.155 -0.081 -0.219(-2.046) (-0.813) (-1.861)RCOLL 0.051 -0.001 0.110(0.846) (-0.006) (1.244)RBACH 0.251 0.209 0.330(3.353) (2.027) (2.942)50Table 6 cont.Variable (1) (2) (3) (4)RPOST 0.514 0.370 0.886(5.170) (3.070) (4.470)OGRADE-0.070 0.000 -0.156(-0.852) (0.005) (-1.098)OSOME 0.030 -0.058 0.188(0.376) (-0.591) (1.374)OHIGH 0.018 0.039 0.010(0.306) (0.467) (0.115)OCOLL -0.049 -0.180 0.089(-0.460) (-1.285) (0.527)OBACH 0.208 0.172 0.161(2.188) (1.573) (0.866)USOME-0.120 -0.264 0.088(-1.020) (-1.904) (0.420)UHIGH-0.121 -0.162 -0.019(-1.774) (-1.833) (-0.165)UCOLL-0.033 -0.053 0.035(-0.388) (-0.524) (0.237)UBACH-0.052 0.072 -0.165(-0.657) (0.691) (-1.349)UPOST-0.151 -0.094 -0.231(-1.088) (-0.654) (-0.611)CONSTANT 1.507 1.672 1.979 1.673(24.32) (22.73) (20.14) (12.96)N 993 993 569 4240.383 0.440 0.393 0.364Note: t-statistics in brackets51The inclusion of the education variables, Column (2), reveals that the returns to over- and undereducation vary with the required level of education.Generally, the educational mismatch dummies are insignificant at the 10% level.However, the overeducation dummy at the BACH level of required educationand the undereducation dummy at the HIGH level are significant. In these cases,overeducation is associated with higher earnings, and undereducation with lowerearnings. Nevertheless, in general, workers are neither penalised nor rewardedfor having educational attainments that differ from requirements.22Restricting the sample to males and females in turn, Columns (3) and (4)respectively, reveals some startling differences between the sexes.23 First, thecoefficients for the required schooling variables at the BACH and POST levelsare much larger for females. Second, although the impacts of the educationalmismatch variables for males are similar to those for the full sample, all theseterms are insignificant for females.Table 7 shows the contributions of the explanatory variables to the male-female earnings gap. For each variable, I use Doiron and Riddell’s [26] gendergap decomposition—a variant of that used by Oaxaca [73]. This decomposes thegap into the difference in the sample means multiplied by the estimated malecoefficient (see Column 1) and the difference in the coefficients multiplied bythe female mean, (see Column 2). The first component shows the difference dueto male-female characteristics; the second, the return to these characteristics.The latter is sometimes attributed to discrimination. The largest proportionof the gap is explained by the returns to personal characteristics. The over-(under-) education variables have relatively small impacts; and their net effectis negative. The majority of the earnings gap accounted for by the educationalmismatch variables is due to the characteristics themselves, rather than thereturns to the characteristics.For comparison, Table 8, Columns (1), (2) and (3) include dummies forattained—rather than required—education for the full, male and female samples22F-tests of the hypotheses that the coefficients on the over- undereducation dummies areidentical cannot he rejected at the 5% level.23An F-test of the hypothesis that there is no difference between the male and femalecoefficients was rejected at the 5% significance level.52Table 7:Male-female earnings decompositioncharacteristics returnsPC 0.080 0.320REQ 0.008 -0.040OVER 0.001 -0.004UNDER -0.020 0.007TOTAL 0.069 0.283respectively. The returns to attained education are much lower than the returnsto required education. The earnings profile is considerably flatter for the fullsample, males and females. In Figures 4, 5 and 6 the coefficients on the requiredand attained dummies for each sample are plotted. The returns to attainededucation conditional on having a job with matching educational requirementsare considerably higher than the unconditional returns.4.6 ConclusionsIn his seminal work, “The Great Training Robbery”, Ivar Berg [14] arguedthat overeducated workers may be less productive than their less skilled counterparts because they become bored with their jobs and lose motivation. EarlierUS and Dutch studies do not support Berg’s [14] proposition; they found thatovereducated workers earn more than their counterparts.In this chapter, I have used canadian data from the NSCS to estimate thereturns to educational mismatch; and have shown that the returns to overeducation are sensitive to both the educational requirements of the job and thesex of the workers. For males, there is weak evidence of positive returns toovereducation if the job requires a university bachelor degree; but, the returnsare insignificant for other required education levels. Unlike previous studies, Ihave found little evidence of lower pay for undereducated males: they are onlypenalised in jobs with low education requirements. Remarkably, I have foundno evidence of returns to either over- or undereducation for females.530.020 0.023 0.015(4.476) (4.148) (2.068)EXP2 -0.000 -0.000 -0.000(-3.384) (-3.159) (-1.452)UNION 0.096 0.047 0.156(2.921) (1.176) (2.724)SEX 0.287(9.022)TEN 0.019 0.176 0.020(3.401) (2.657) (2.054)TEN2 -0.001 -0.001 -0.000(-2.999) (-2.627) (-1.434)BIL 0.065 0.119 -0.003(1.588) (2.400) (-0.050)EXTR. 0.149 0.135 0.134(2.182) (1.902) (0.775)DIST 0.065 0.003 0.123(1.342) (0.053) (1.155)PUB -0.013 0.022 -0.112(-0.300) (0.414) (-1.222)INFO 0.079 0.116 0.024(1.366) (1.380) (0.249)RET -0.216 -0.158 -0.294(-4.416) (-2.601) (-3.248)Hence, the hypothesis that overeducated workers have identical earnings tootherwise identical workers cannot (in general) be rejected, so the Canadian evidence does not support Berg’s [14] claim. In Canada, overeducated workers donot receive lower earnings than otherwise identical workers. But, those workerswith educational attainments in excess of requirements do not receive the fullreturns to their attained education: their earnings would have been higher in ajob with requirements that matched their attainments. This finding will concernpolicy makers: rising educational attainments are insufficient to ensure higherearnings for workers.Table 8:Regression equations; dependent variable ln YVariable (1) (2) (3)EXP54Table 8 cont.Variable (1) (2) (3)PROF 0.320 0.329 0.297(5.663) (4.535) (3.268)SEMI 0.392 0.353 0.410(8.186) (5.860) (5.094)SUPER 0.226 0.217 0.195(3.318) (2.762) (1.503)SKILL 0.195 0.211 0.148(5.106) (4.412) (2.329)ATL -0.072 -0.089 -0.054(-1.285) (-1.249) (-0.582)QUE -0.018 -0.060 0.031(-0.450) (-1.227) (0.490)PRA 0.030 -0.001 0.030(0.670) (-0.020) (0.441)BC 0.089 0.229 -0.049(1.847) (3.683) (-0.642)CITY 0.052 0.019 0.114(1.704) (0.510) (2.171)AGRADE -0.252 -0.244 -0.220(-3.988) (-3.290) (-1.877)ASOME -0.151 -0.074 -0.211(-3.082) (-1.204) (-2.545)ACOLL -0.004 0.032 -0.000(-0.095) (0.591) (-0.001)ABACH 0.116 0.119 0.144(2.016) (1.625) (1.556)APOST 0.322 0.282 0.412(4.404) (3.175) (3.116)CONSTANT 1.597 1.857 1.647(23.51) (22.56) (13.60)N 993 569 4240.412 0.357 0.327Note: t-statistics in brackets55Figure 4: Earnings Profile- Full Sample0.6 - -0.5 - -0.4 - -0.3 - -0.2 - -0.1 - -0- IGr de Some High Coll Bach Post-0.1 - --0.2 - -IEducationAttained Required56Figure 5: Earnings Profile - Males0.4 -0.3 - -0.2 - -0.1 - -0- I IGr dc Some High Coil Bach Post-0.1 - --0.2 - -EducationAttained Required57Figure 6: Earnings Profile - Females0.80.60.40.20-0.2-0.4EducationAttained Required1High Coll Bach Post58Chapter 5ConclusionsThis thesis has taken the form of three related essays about the labourmarket implications of job quality.In the first essay, Chapter 2, I have demonstrated, by analysing a specifictwo-type, two-period example, that introductory wage offers can inform workersabout the expected quality of experience jobs. In this game, the good type offirm offers higher expected quality jobs than the bad type. If the good typeis less likely to exit, the game has many equilibria with separating wages. Ineach of these, the introductory compensating differentials have the opposite signto the usual Smith/Rosen case: higher expected quality jobs pay more, ratherthan less. I have also shown that there are pooling equilibria. In this case, theintroductory wages are independent of job quality. I have demonstrated that,for some parameter space, the (extended) Intuitive Criterion renders a uniqueseparating equilibrium.In the second essay, Chapter 3, I have presented Canadian evidence thattests and supports the theory of compensating differentials for a variety of jobcharacteristics. Previous Canadian studies, such as those by Meng [64 and 65],Martinello and Meng [62), and Cousineau et al [221 have found support for thetheory for risk of injury or death. This is the first study, however, to havefound Canadian evidence of compensating wage payments for non-hazardouscharacteristics.By exploiting the self-report information contained in the NSCS, I haveavoided using the occupational-trait data used by Meng [64]. Occupationaltrait data provide information on averages across broad occupational categoriesand industries, introducing error into the job quality variables.After controlling for personal characteristics, I have found evidence of compensating payments for working with data, working with hands, bureaucratic59procedures, and responsibility over other workers. I have also found evidence ofdifferentials for the control of hours and pace, though the coefficients are incorrectly signed. I have found no evidence of differentials for working with people,working with machines or the freedom to design work.The third essay, Chapter 4, takes a slightly different perspective on jobquality. I have focused on the mismatch between the educational requirementsof jobs and the educational attainments of workers. In his seminal work, “TheGreat Training Robbery”, Ivar Berg [141 argued that overeducated workers maybe less productive than their less skilled counterparts because they becomebored with their jobs and lose motivation. Earlier US and Dutch studies donot support Berg’s [14] proposition; they found that overeducated workers earnmore than otherwise identical workers (with just the required level of schooling).I have used Canadian data from the NSCS to estimate the returns to educational mismatch; and have shown that the returns to overeducation are sensitiveto both the educational requirements of the job and the sex of the workers. Formales, there is evidence of positive returns to overeducation if the job requires auniversity bachelor degree; but, the returns are insignificant for other requirededucation levels. I have also found little evidence of lower pay for undereducatedmales; though they are penalised in jobs with low education requirements. Remarkably, I have found no evidence of returns to either over- or undereducationfor females.Since, in general, the hypothesis that overeducated workers have the sameearnings as otherwise identical workers (with the required level of schooling)cannot be rejected, the evidence does not support Berg’s [14] claim. However,job requirements or “pigeonholes” are important determinants of earnings—and particularly so for females. This chapter, therefore, offers some support forThurow’s [94] notion that marginal productivity resides in the job, rather thanin the individual characteristics of the worker.60References[1] Abowd J.M. and 0. Ashenfelter (1981): “Anticipated Unemployment,Temporary Layoffs, and Compensating Wage Differentials”, in S. Rosen,ed., Studies in Labor Markets, University of Chicago Press.[2] Akerlof, G. and J.L. Yellen (1986): Efficiency Wage Models of the LaborMarket, New York, Cambridge University Press.[3] Allen, F. and G.R. Faulhaber (1988): “Optimism Invites Deception”,Quarterly Journal of Economics, 103, 397-407.[4] Allen, F. and G.R. Faulhaber (1991): “Rational Rationing”, Economica,58(230), May, 189-198.[5] Altonji, J.G. and C.H. Paxson (1988): “Labour Supply Preferences, HoursConstraints, and Hours-Wage Trade-offs”, Journal of Labor Economics,6(2), April, 254-276.[6] Antos, J. (1982): “Union Impacts on White Collar Compensation”, Officeof Research and Evaluation, U.S. Bureau of Labor Statistics.[7] Arnould, R. and L. Nichols (1983): “Wage-risk Premiums and Workers’Compensation: a Refinement of Estimates of Compensating Wage Differentials”, Journal of Political Economy, 91, 332-340.[8] Aryan, L. and S. Esfahani (1993): “A Model of Efficiency Wages as aSignal of Firm Value”, International Economic Review, 34, 3, 503-524.[9] Bagwell, K. (1987): “Introductory Price as a Signal of Cost in a Model ofRepeat Purchases”, Review of Economic Studies, 54, 365-384.[10] Beaudry, P. (1989a): “Entry Wages Signalling Future Wages”, Universityof Montreal Discussion Paper 8905.[11] Beaudry, P. (1989b): “Job Rationing with Complete Contracts: An Informed Principal Approach”, University of Montreal Discussion Paper8906.[12] Beaudry, P. (1994): “Why an Informed Principal May Leave Rents to anAgent”, International Economic Review, 35, 4, 821-832.[13] Becker, G.S. (1964): Human Capital, National Bureau of Economic Research, New York.[14] Berg, I. (1970): Education and Jobs: The Great Training Robbery, BeaconPress, Boston.61[15] Biddle, J.E., G. Zarkin, (1988): “Worker Preferences and Market Compensation for Job Risk”, Review of Economics and Statistics, 70, 660-667.[16] Bluestone, B. and B. Harrison (1988): The Great U-Turn: CorporateRestructuring and the Polarization of America, Basic Books, New York.[17] Brown, C. (1980): “Equalizing Differences in the Labor Market”, Quarterly Journal of Economics, 94, 113-134.[18] Cain, P.S. and D.J. Treiman (1981): “The Dictionary of OccupationalTitles as a Source of Occupational Data”, American Sociological Review,no 46 (June), 253-278.[19] Carmichael, H.L. (1986): “Reputations for Safety: Market Performanceand Policy Remedies”, Journal of Labor Economics, 4(4), 458-472.[20] Cho, I.K. and D. M. Kreps (1986): “Signalling Games and Stable Equilibria”, Quarterly Journal of Economics, 102, 2, 179-221.[21] Cousineau, J.-M., R. Lacroix and A.- M.Girard (1988a): “An Evaluationof Wage-risk Premiums”, University of Montreal, unpublished manuscript,May.[22] Cousineau, J.-M., R. Lacroix and A.- M.Girard (1992): “OccupationalHazard and Wage Compensating Differentials”, Review of Economics andStatistics, 74, 166-169.[23] Diamond, P.A. (1977): “Insurance Theoretic Aspects of Workers’ Compensation” in Diamond, P.A. ed., Natural Resources, Uncertainty, andGeneral Equilibrium Systems, New York: Academic Press, 67-89.[24] Digby, C. and W.C. Riddell (1986): “Occupational Health Safety inCanada”, in W.C.Riddell, ed., Canadian Labour Relations, University ofToronto Press, Toronto.[25] Dillingham, A. (1985): “The Influence of Risk Variable Definition on Valueof Life Estimates”, Economic Inquiry, 25, 277-294.[26] Doiron, D.J. and W.C. Riddell (1992): “The Impact of Unionization onMale-Female Earnings Differences in Canada”, UBC Working Paper No.92-30, December.[27] Dooley M.D. (1986): “The Overeducated Canadian?”, Canadian Journalof Economics, 19 (1), 142-159.[28] Duncan, G. and S.D. Hoffman (1981): “The Incidence and Wage Effectsof Overeducation”, Economics of Education Review, 1, no 1, 75-86.62[29] Duncan, G. and B. Hoimlund (1983): “Was Adam Smith Right After All?Another Test of The Theory of Compensating Differentials”, Journal ofLabor Economics, 2, 366-377.[30] Duncan, G. and F. Stafford (1980): “Do Union Members Receive Compensating Wage Differentials?”, American Economic Review, 70, 355-371.[31] Fairris, D. (1992): “Compensating Payments and Hazardous Work inUnion and Nonunion Setting”, Journal of Labor Research, 13(2), 205-221.[32] Frank, J. (1986): “A Signalling Approach to Wage Rigidity and Layoffs”,European Economic Review, 31, 1385-1405.[33] Freeman, R.B. (1976): The Overeducated American, Academic Press, NewYork.[34] Freeman, R.B. (1980): “The Facts About the Declining Value of College”,Journal of Human Resources, 15, 124-142.[35] Garen, J. (1988): “Compensating Wage Differentials and the Endogeneityof Job Riskiness”, Review of Economics and Statistics, 70(1), 9-16.[36] Giammarino, R.M. and E. Nosal (1990): “Wage Smoothing as a Signal ofQuality”, Canadian Journal of Economics, 90, 159-174.[37] Greenwood, M.J., G.L. Hunt, D.S. Rickman, and G.I. Treyz (1991): “Migration, Regional Equilibrium, and the Estimation of Compensating Differentials”, American Economic Review, 81(5), 1383-1390.[38] Gunderson, M. and W.C. Riddell (1993): Labour Market Economics,McGraw-Hill Ryerson Limited, Toronto.[39] Hammermesh, D.S. (1977): “Economic Aspects of Job Satisfaction”, in0. Ashenfelter and W. Oates, eds., Essays in Labor Market Analysis, NewYork.[40] Hammermesh, D.S. and J.R. Wolfe (1990): “Compensating Wage Differentials and the Duration of Wage Loss”, Journal of Labor Economics,8(1), Part 2, S175-197.[41] Hartog, J. (1986): “Allocation and the Earnings Function”, EmpiricalEconomics, 11, no 2, 97-110.[42] Hartog, J. and H. Oosterbeek (1988): “Education, Allocation and Earnings in the Netherlands”, Economics of Education Review, Vol 7, No 2,185-194.[43] Hersch, J. (1991): “Education Match and Job Match”, Review of Economics and Statistics, 73 (1), February, 140-144.63[44] Heywood, J.S. (1989): “Do Union Members Receive Compensating Differrentials? The Case of Employment Security”, Journal of Labor Research,10, 3, 271-283.[45] loannides, Y.M. and C.A. Pissarides (1985): “Monopsony and the LifetimeRelation between Wages and Productivity”, Journal of Labor Economics,3(1), 91-100.[46] Johnson, W. (1978): “A Theory of Job Shopping”, Quarterly Journal ofEconomics, 92, 261-277.[47] Jones-Lee, M. (1989): The Economics of Safety and Physical Risk, BasilBlackwell, Oxford.[48] Jovanovic, B. (1979a): “Job Matching and the Theory of Turnover”, Journal of Political Economy, 87, 972-990.[49] Jovanovic, B. (1979b): “Firm Specific Capital and Turnover”, Journal ofPolitical Economy, 87, 1246-1260.[50] Kahn, S. (1987): “Occupational Safety and Worker Preferences: Is Therea Marginal Worker?”, Review of Economics and Statistics, 69(2), 262-268.[51] Kahn, S. and K. Lang, (1988): “Efficient Estimation of Structural HedonicSystems”, International Economic Review, 29, 161-169.[52] Killingsworth, M.R. (1987): “Heterogeneous Preferences, CompensatingWage Differentials, and Comparable Worth”, Quarterly Journal of Economics, 102(4), 727-742.[53] Kostiuk, P.F. (1990): “Compensating Differentials for Shift Work”, Journal of Political Economy, 98(5), 1054-1075.[54] Kreps, D. and R. Wilson (1982): “Sequential Equilibria”, Econometrica,50, 863-894.[55] Krueger, A.B. (1993): “How Computers have Changed the Wage Structure: Evidence from Microdata, 1984-1989”, Quarterly Journal of Economics, 108(2), 33-60.[56] Kuhn, P. (1994): “Nonrevelation in Employment Contracts”, International Economic Review, 35(2), 261-283.[57] Kuttner, B. (1983): “The Declining Middle”, The Atlantic Monthly, July,60-72.[58] Leigh, J.P. (1989): “Compensating Wages for Job-Related Deaths: TheOpposing Arguments”, Journal of Economic Issues, Vol 23(3), September.64[59] Lucas, R.E.B. (1977): “Hedonic Wage Equations and the Psychic ReturnsTo Schooling”, American Economic Review, 67, 549-558.[60] Ma, C. and A.M. Weiss (1990): “A Signalling Theory of Unemployment”,European Economic Review, 37. 1, 135-158.[61] Mann, A. and G. Psacharopoulos (1982): “The Reward for Risk in theLabour Market: Evidence from the U.K. and a Reconcilation with OtherStudies”, Journal of Political Economy, 90, 827-853.[62] Martinello, R. and R. Meng (1992): “Workplace Risks and the Value ofHazard Avoidance”, Canadian Journal of Economics, 25(2), May, 333-345.[63] McNabb, R. (1989): “Compensating Wage Differentials: Some Evidencefor Britain” Oxford Economic Papers, 41, 327-338.[64] Meng, R. (1989): “Compensating Differentials in the Labour Market”,Canadian Journal of Economics, 89, 413-424.[65] Meng, R. (1991): “Compensating Wages For Long-Term Job Hazards InCanadian Industry”, Economics Letters, 36, 331-336.[66] Mester, L. (1992): “Perpetual Signalling with Imperfectly CorrelatedCosts”, Rand Journal of Economics, 23, 4, 548-563.[67] Milgrom, P. and J. Roberts (1986): “Price and Advertising as Signals ofProduct Quality”, Journal of Political Economy, 94(4), 796-821.[68] Mincer, J. (1974): Schooling, Experience, and Earnings, National Bureauof Economic Research, New York.[69] Moore, M.J. and W.K. Viscusi, (1990): Compensation Mechanisms forJob Risks, Princeton University Press, Princeton, New Jersey.[70] Myles, J. and G. Fawcett (1990): “Job Skills and the Service Economy”,Economic Council of Canada Working Paper, no 4.[71] Nelson, P. (1970): “Information and Consumer Behaviour”, Journal ofPolitical Economy, 78, 311-329.[72] Noldeke, G. and E. van Damme (1990): “Switching Away from ProbabilityOne Beliefs”, unpublished manuscript.[73] Oaxaca, R. (1973): “Male-female Wage Differentials in Urban Labor Markets”, International Economic Review, 14, 3, 693-709.65[74] Oi, W. (1973): “ Workmen’s Compensation and Industrial Safety”, inSupplemental Studies for the National Commission, on State Workmen’sCompensation Laws, 1, U.S. Government Printing Office, Washington,D.C..[75] Olson, C.A. (1981): “An Analysis of Wage Differentials Received by Workers on Dangerous Jobs”, Journal of Human Resources, 16, 167-185.[76] Palme, M.O. and R.E. Wright, (1992): “Gender Discrimination and Compensating Differentials” Applied Economics, 24, 751-759.[77] Picot, G., J. Myles and T. Wannell (1990): “Good Jobs! Bad Jobs andthe Declining Middle, 1967-1986”, Statistics Canada Analytical StudiesBranch, Ottawa.[78] Rea, S. A. (1981): “Workmen’s Compensation and Occupational Safetyunder Imperfect Information”, American Economic Review, 71, 1, 80-93.[79] Reynolds, (1951): The Structure of Labor Markets, New York, Harper.[80] Riley, J. (1979): “Informational Equilibrium”, Econometrica, 47, 331-360.[81] Rosen, S. (1974): “Hedonic Prices and Implicit Markets; Product Differentiation in Pure Competitive Markets”, Journal of Political Economy,82, 34-55.[82] Rosen, S. (1986): “The Theory of Equalizing Differences”, Handbook ofLabour Economics, Vol.1. 0. Ashenfelter and R. Layard (eds.), New York:Elsevier.[83] Rumberger, R.W. (1981): Overeducation in the US Labor Market, Praeger,New York.[84] Rumberger, R.W. (1987): “The Impact of Surplus Schooling on Productivity and Earnings”, Journal of Human Resources, 22, no 1, 1-50.[85] Salop, S.C. (1979): “A Model of the Natural Rate of Unemployment”American Economic Review, 69, 117-125.[86] Shapiro, C. and J. Stiglitz, (1984): “Equilibrium Unemployment as aTorker Discipline Device” American Economic Review, 74, 433-444.[87] Sicherman, N. (1991): “Overeducation in the Labor Market”, Journal ofLabor Economics, vol 9, no 2, 101-122.[88] Sicherman, N. and 0. Galor (1990): “A Theory of Career Mobility”, Journal of Political Economy, 98, no 1, 169-192.66[89] Smith, A. (1947): An Inquiry into the Nature and Causes of the Wealthof Nations, Reprint, Modern Library Edition, New York.[90] Smith, J. (1994): Wage Bargaining in the Chemical Industry, unpublishedmaunscript, University of Cambridge.[91] Smith, R.S. (1976): The Occupational Safety and Health Act: Its Goalsand Its Achievements, The American Enterprise Institute for Public PolicyResearch.[92] Smith, R.S. (1979): “Compensating Wage Differentials and Public Policy:A Review”, Industrial and Labor Relations Review, 32, 229-352.[93] Spence, M. (1974): Market Signalling, Cambridge, Mas., Harvard University Press.[941 Thurow, L.C. (1975): Generating Inequality, Basic, New York.[95] Tirole, J. (1988): The Theory Of Industrial Organisation, The M.I.T.Press, Cambridge, Mass.[96] Tsang, M.C. and H.M. Levin (1985): “The Economics of Overeducation”,Economics of Education Review, Vol 4, No 2, 93-104.[97] van Ophem, H. et al (1993): “Job Complexity and Wages”, InternationalEconomic Review, 34, no 4, 853-872.[98] Viscusi, W.K. (1978): “Wealth Effects and Earnings Premiums for JobHazards”, Review of Economics and Statistics, 60, 408-416.[99] Viscusi, W.K. (1979): “Job Hazards and Worker Quit Rates”, International Economic Review, 20, 29-58.[100] Viscusi, W.K. (1979): Employment Hazards: An Investigation of MarketPerformance, Cambridge, Mass.: Havard University Press.[101] Viscusi, W.K. (1980): “Unions, Labor Market Structure, and the WelfareImplications of the Quality of Work”, Journal of Labor Research, 1, 175-192.[102] Viscusi, W.K. (1980): “A Theory of Job Shopping: A Bayesian Perspective”, Quarterly Journal of Economics, 94, 604-614.[103] Viscusi, W.K. (1980): “Self-Selection, Learning, Induced Quits, and theOptimal Wage Structure”, International Economic Review, 21(3), 529-546.[104] Viscusi, W.K. (1993): “The Value of Risks to Life and Health”, Journalof Economic Literature, 31(4), 1912-1946.67[105] Viscusi, W.K. and M.J. Moore (1987): “Workers’ Compensation: WageEffects, Benefit Inadequencies, and the Value of Health Losses”, Reviewof Economics and Statistics, 69(2), 249-261.[106] Viscusi, W.K. and M.J. Moore (1990): “Models for Estimating DiscountRates for Long-term Health Risks using Labor Market Data”, Journal ofRisk and Uncertainty, 3(4), December, 381-401.[107] Viscusi, W.K. and M.J. Moore (1991): “Worker Learning and Compensating Differentials”, Industrial and Labor Relations Review, 45(1), 80-96.[108] Weiss, A. (1980): “Job Queues and Layoffs in Labor Markets with FlexibleWages”, Journal of Political Economy, 88, 526-538.[109] White, H. (1980): “A Heteroskedasticity-Consistent Covariance MatrixEstimator and a Direct Test for Heteroskedasticity.”, Econometrica, 48,817-838.[110] Wilde, L.L. (1979): “An Information-Theoretic Approach to Job Quits”,in Studies in the Economics of Search, S.A. Lippman and J.J. McCall(eds), New York: North-Holland.6869ix1puddyTable A:Variable means by sex and union statusName Male Female Union NonunionIndividual characteristicsANY $27,576 $16,866 $23,317 $22,155WY $528.49 $323.48 $447.20 $424.91Y $13.08 $9.15 $11.67 $10.97in Y 2.45 2.06 2.37 2.19EDUC 13.00 12.79 12.92 12.89AGRADE 0.12 0.07 0.10 0.09ASOME 0.17 0.17 0.17 0.17AuG11 0.17 0.18 0.17 0.18ACOLL 0.32 0.39 0.36 0.35ABACH 0.13 0.14 0.13 0.13APOST 0.09 0.04 0.06 0.07EXP 19.94 15.56 18.64 17.36EXP2 561.84 385.92 502.04 464.60UNION 0.51 0.39 1.00 0.00TEN 9.33 5.94 9.10 6.68TEN2 162.97 80.79 152.40 102.97BIL 0.23 0.15 0.21 0.18SEX 1.00 0.00 0.61 0.49LocationATL 0.08 0.09 0.09 0.07QUE 0.33 0.25 0.33 0.26ONT 0.36 0.35 0.32 0.39PRA 0.12 0.18 0.13 0.17BC 0.11 0.13 0.13 0.11CITY 0.58 0.65 0.61 0.62Job characteristicsCDESN 0.42 0.58 0.59 0.42CHRS 0.67 0.77 0.86 0.60CPACE 0.49 0.58 0.65 0.44RESP 0.43 0.31 0.28 0.46BUR 0.51 0.48 0.67 0.36HANDS 0.47 0.45 0.37 0.55PEOPLE 0.40 0.27 0.44 0.25DATA 0.55 0.49 0.51 0.53MACHINES 0.75 0.74 0.80 0.71Q 4.70 4.70 5.16 4.3170Table A cont.Name Male Female Union NonunionMAR 0.76 0.52 0.65 0.64HOME 0.70 0.56 0.63 0.65APP 0.25 0.11 0.18 0.19UNEM 1.29 0.63 1.30 0.73IndustryEXTR 0.08 0.02 0.07 0.04MANUF 0.32 0.10 0.25 0.20DIST 0.17 0.10 0.17 0.11PUB 0.24 0.40 0.45 0.20INFO 0.06 0.16 0.01 0.19RET 0.13 0.21 0.05 0.26OccupationPROF 0.19 0.14 0.13 0.20SEMI 0.15 0.17 0.18 0.15SUPER 0.07 0.03 0.03 0.07SKILL 0.25 0.24 0.27 0.23SEMUN 0.34 0.41 0.39 0.3671Table B:Further means and definitions, full sampleName Definition MeanMAR Married = 1; otherwise = 0 0.65HOME Own home = 1; otherwise = 0 0.64APP Apprenticeship = 1; otherwise 0 0.18UNEM Number of times unemployed 0.99IndustryEXTR Extraction and construction = 1; otherwise 0 0.05MANUF Manufacturing = 1; otherwise = 0 0.22DIST Distribution 1; otherwise = 0 0.14PUB Public services = 1; otherwise = 0 0.31INFO Information services = 1; otherwise = 0 0.11RET Retail and other services = 1; otherwise = 0 0.17OccupationPROF Professional = 1; otherwise=0 0.17SEMI Semi-professional = 1; otherwise =0 0.16SUPER Supervisory = 1; otherwise = 0 0.05SKILL Skilled trade = 1; otherwise 0 0.25SEMUN Semi-skilled and unskilled = 1; otherwise =0 0.3772Table C:Job quaiity and union status equationsVariable Q UnionEDUC -0.02 1 0.007(-1.015) (0.304)AGRADE -0.115 0.081(-0.564) (0.377)ASOME -0.013 -0.175(-0.084) (-1.082)ACOLL 0.267 -0.065(2.069) (-0.453)ABACH -0.420 0.085(-2.221) (0.408)APOST -0.327 0.125(-1.307) (0.452)EXP -0.000 -0.010(-0.015) (-0.686)EXP2-0.000 0.000(-0.292) (0.043)UNION 0.609(6.130)TEN 0.011 0.068(0.684) (3.963)TEN2 -0.000 -0.001(-0.561) (-2.513)BIL -0.104 -0.007(-0.848) (-0.050)SEX -0.148 0.267(-1.465) (2.378)73Table C cont.Variable Q UnionATL 0.167 0.070(0.981) (0.406)QUE 0.080 0.269(0.673) (2.134)PRA-0.021 -0.148(-0.157) (-1.012)BC 0.085 0.481(0.584) (2.847)CITY -0.040 0.206(-0.435) (2.137)MAR 0.029 0.078(0.293) (0.665)HOME -0.015 -0.137-0.146 (-1.188)APP 0.097 0.0400.861 (0.330)UNEM 0.045 0.0732.018 (3.087)EXTR -0.257 0.271-1.246 (1.373)DIST 0.099 0.2260.677 (1.546)PUB -0.120 0.859(-0.886) (6.102)INFO -0.170 -1.499(-0.978) (-5.864)RET -0.858 -1.068(-5.849) (-6.359)74Table C coat.Variable Q UnionPROF -0.445 -0.957(-2.563) (-4.813)SEMI -0.181 -0.408(-1.217) (-2.512)SUPER -0.383 -0.912(-1.217) (-4.018)SKILL -0.321 -0.217(-2.738) (-1.724)CONSTANT 5.078 -0.465(14.77) (-1.255)N 993 9930.170 0.291Note: t-statistics in brackets75Table D:Senior workers, regression eqnations; dependent variable in YVariable (1) (2) (3)EDUC 0.035 0.036 0.032(4.574) (4.074) (3.607)AGRADE -0.248 -0.221 -0.257(-3.445) (-2.726) (-3.071)ASOME -0.247 -0.207 -0.211(-4.139) (-3.156) (-3.082)ACOLL -0.039 -0.038 -0.015(-0.735) (-0.670) (-0.245)ABACH 0.322 0.237 0.246(3.822) (2.424) (2.494)APOST 0.284 0.211 0.179(3.253) (2.111) (1.756)EXP 0.022 0.024 0.027(3.458) (2.973) (2.528)EXP2 -0.000 -0.000 -0.000(-2.623) (-2.471) (-2.321)UNION -0.088 -0.076 -0.049(-1.780) (-1.425) (-0.961)TEN 0.002 0.012 0.003(0.176) (0.816) (0.174)TEN2 -0.000 -0.000 -0.000(-0.632) (-1.467) (-1.518)BIL 0.108 0.093 0.101(2.271) (1.866) (1.953)SEX 0.306 0.312 0.337(8.731) (8.512) (8.249)76Table D cont.Variable (1) (2) (3)ATL-0.025 -0.011 0.007(-0.398) (-0.140) (0.093)QUE-0.003 -0.010 -0.018(-0.069) (-0.228) (-0.373)PRA 0.058 0.069 0.042(1.040) (1.104) (0.627)BC 0.169 0.157 0.118(2.734) (2.074) (1.519)CITY 0.071 0.060 0.001(2.002) (1.491) (0.016)Q 0.155 0.111 0.017(2.623) (1.609) (0.249)-0.052-0.014 -0.009(-0.804) (-0.203) (-0.130)-0.018 -0.016 -0.006(-2.261) (-1.894) (-0.684)Q. TEN 0.001 0.001 0.003(0.481) (0.419) (1.189)CONSTANT 0.671 0.772 1.235(2.584) (2.610) (4.492)N 781 612 5110.353 0.335 0.344Note: t-statistics in brackets77

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}]}"
                            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:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0088853/manifest

Comment

Related Items