UBC Faculty Research and Publications

Valuing productivity loss due to absenteeism: firm-level evidence from a Canadian linked employer-employee… Zhang, Wei; Sun, Huiying; Woodcock, Simon; Anis, Aslam H Jan 19, 2017

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

Item Metadata


52383-13561_2016_Article_138.pdf [ 593.76kB ]
JSON: 52383-1.0362029.json
JSON-LD: 52383-1.0362029-ld.json
RDF/XML (Pretty): 52383-1.0362029-rdf.xml
RDF/JSON: 52383-1.0362029-rdf.json
Turtle: 52383-1.0362029-turtle.txt
N-Triples: 52383-1.0362029-rdf-ntriples.txt
Original Record: 52383-1.0362029-source.json
Full Text

Full Text

RESEARCH Open AccessValuing productivity loss due toabsenteeism: firm-level evidence froma Canadian linked employer-employeesurveyWei Zhang1,2 , Huiying Sun1, Simon Woodcock3 and Aslam H. Anis1,2*AbstractIn health economic evaluation studies, to value productivity loss due to absenteeism, existing methods use wagesas a proxy value for marginal productivity. This study is the first to test the equality between wage and marginalproductivity losses due to absenteeism separately for team workers and non-team workers. Our estimates are basedon linked employer-employee data from Canada. Results indicate that team workers are more productive and earnhigher wages than non-team workers. However, the productivity gap between these two groups is considerablylarger than the wage gap. In small firms, employee absenteeism results in lower productivity and wages, and themarginal productivity loss due to team worker absenteeism is significantly higher than the wage loss. No similarwage-productivity gap exists for large firms. Our findings suggest that productivity loss or gain is most likely tobe underestimated when valued according to wages for team workers. The findings help to value the burden ofillness-related absenteeism. This is important for economic evaluations that seek to measure the productivity gainor loss of a health care technology or intervention, which in turn can impact policy makers’ funding decisions.Keywords: Productivity loss, Absenteeism, Marginal productivity, Wage, Teamwork, ValuationJEL codes: J31, D24, I12, I15IntroductionIt is still under debate whether we should take accountof productivity gains or losses from a health care inter-vention in economic evaluation studies [1, 2]. Cost-effectiveness studies, for example, are routinely used todetermine the eligibility of health technologies such aspharmaceuticals for coverage under national or provin-cial health plans. The inclusion of productivity losses insuch analyses would have a significant influence ondeterminations of cost-effectiveness, leading to differentresource allocation decisions. Krol et al. find thataccounting for productivity costs can either increase ordecrease the incremental cost-effectiveness ratio (ICER)between treatment arms [3, 4]. Thus, cost-effectivenessstudies that account for productivity losses are useful inidentifying interventions with a potentially broad impact,and do not necessarily lower the ICERs of anintervention.Despite robust arguments in favour of including prod-uctivity loss in evaluation studies [3–6], current methodsto value productivity loss are limited. Existing methodsusually quantify productivity loss using wages as a proxyfor marginal productivity [1, 7, 8]. However, wages maynot equal marginal productivity for many reasons, mak-ing it a poor proxy and reducing the accuracy of esti-mated productivity loss. In imperfect labour markets,wages may not equal marginal productivity due to in-equities, such as race or gender discrimination, wherebyan identifiable group routinely receives lower wages.More commonly, risk-averse workers might willingly* Correspondence: aslam.anis@ubc.ca1Centre for Health Evaluation and Outcome Sciences, St. Paul’s Hospital,588-1081 Burrard Street, Vancouver, BC V6Z1Y6, Canada2School of Population and Public Health, University of British Columbia, 2206East Mall, Vancouver, BC V6T1Z3, CanadaFull list of author information is available at the end of the article© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made.Zhang et al. Health Economics Review  (2017) 7:3 DOI 10.1186/s13561-016-0138-yaccept a wage below their marginal productivity inexchange for job security, e.g. allowances for sick days[9, 10].A wedge between a worker’s wage and marginal prod-uctivity may also arise if a job involves team productionor if the firm output is time-sensitive [9, 11]. Pauly et al.presented a general model demonstrating that whenthere is a team production and substantial team-specifichuman capital, the value of lost output to the firm froman absence will exceed the daily wage of the absentworker and could be as large as the total output of theteam [9]. Similarly, the cost of an absence will exceedthe wage when a firm incurs a penalty if it misses anoutput target due to the absence. In both situations, theproductivity loss could be reduced if replacements arefound who are either inexpensive or are close substitutesfor the absent worker.Although there are many reasons that wage maynot equal marginal productivity, there is still lack ofempirical evidence on their equality with regard toabsenteeism and team participation. This is the firststudy to empirically test the wage and marginalproductivity losses due to absenteeism and measurethe multiplicative effect of absenteeism for teamworkers. This study examines the theoretical implica-tions on the relationship between wages and product-ivity when a job is involved in team production. Itsfindings will help determine whether wages can beused as a precise proxy of marginal productivity inestimating productivity loss due to illness-relatedabsenteeism. In addition, we use a unique employer-employee data, the Workplace and Employee Survey(WES). The advantage of these data is that they con-tain information on a firm’s output, capital, materials,other expenditures, payroll, and industry as well asits workers’ age, sex, education, occupation, teamparticipation status and absenteeism. The availabilityof such data allows us to test the equality of wageand marginal productivity for groups of workers withdifferent characteristics. The WES is one of only afew linked employer-employee databases worldwideand the only one for Canada. Furthermore, we con-duct robustness checks using alternative specifica-tions and dropping some of the assumptions. Wefind that our estimates of wage and marginal prod-uctivity losses due to absenteeism appear relativelyrobust and reasonable. We also divide the full sampleinto small firm and large firms and examine whetherour estimates vary by firm size.The remainder of this paper is organized as follows.Section 2 contains the conceptual framework and ashort review of related studies. In section 3, we presentour empirical specification. Section 4 describes our dataand defines the main variables. In section 5, we presentour findings and parameter estimates. Section 6 summa-rizes our findings and their implications for economicevaluators.BackgroundConceptual frameworkA large literature has documented substantial wagedifferentials on the basis of firm size [12, 13], indus-try [14–16], group or non-group work [17, 18],union and non-union contracts [19, 20], business cycle[21, 22], competitiveness of the industry [23, 24], and gov-ernment regulation [25, 26]. These wage gaps are conven-tionally estimated from a wage regression usingindividual-level data. Without an independent measure ofworker productivity, however, it is difficult to determinewhether these estimated wage differentials reflect product-ivity differentials or other factors such as wage discrimin-ation [27, 28]. Hellerstein et al. have developed aframework to simultaneously estimate firm-level wageequations and production functions on population-baseddatasets that link employees’ input to their employers’ out-put [27, 28]. Their approach yields estimated marginalproductivity differentials and wage differentials forworkers with different characteristics, and a framework totest their equality.Hellerstein and Neumark use Israeli labour marketdata to test whether the wage gap between men andwomen exceeds the gap between them (if any) inmarginal productivity [27]. Hellerstein et al. use USpopulation data to estimate wage and marginal prod-uctivity differentials for worker groups with differentage, sex, and race characteristics [28]. Many recentstudies have applied the Hellerstein et al. framework.For example, Haegeland and Klette analyze wage andproductivity gaps among Norwegian workers groupedby sex, education and work experience [29]; vanOurs and Stoeldraijer identify 13 studies on age,wage and productivity using linked employer-employee data [30].Our theoretical framework is based on Pauly et al. [9].They develop a general model to examine the magnitudeand incidence of costs associated with absenteeismunder alternative assumptions about firm size, the pro-duction function, the nature of the firm’s product, andthe competitiveness of the labor market. We test twokey theoretical predictions of their model using the Hel-lerstein et al. [27] and Hellerstein and Neumark [28]framework.The first prediction is that the productivity loss associ-ated with a worker’s absence will be larger than the wagein firms with team production. If a team worker is ab-sent, the output of the entire team may be affected.Hence the impact on firm output exceeds the wage thatwould have been paid to the absent team worker. WeZhang et al. Health Economics Review  (2017) 7:3 Page 2 of 14test the hypothesis that the absence of team workers hasa larger effect on firm-level production than wages (i.e.,a significant difference between productivity effects andwage effects). In contrast, we hypothesize that theabsence of non-team workers has a similar effect onproduction and wages.The second prediction is that the difference betweenthe wage and the productivity loss due to absence willbe larger in small firms than large firms. While largefirms can hire extra employees to ensure that a givenoutput level can be maintained if a team worker isabsent, small firms may not be able to afford this ex-pense. We test whether the difference between product-ivity effects and wage effects is larger in small firms thanlarge firms.Previous literature on the impact of absenteeism andteam on wages and productionA related literature seeks to uncover factors that de-termine or affect worker absence by modeling absence[17, 31–35] or focuses on the association betweenhealth conditions and absenteeism [36–39]. Few stud-ies have estimated the impact of absenteeism onwages or production, and none have examinedwhether their impact varies by team work status andfirm size.Allen estimates the trade-off between wages andexpected absence via a hedonic wage equation usingindividual worker level data in 1970s, and the effectof absenteeism on output per man-hour via a plant-level production function for manufacturing [40]. Hefinds a small difference between the wage effect andthe productivity effect. However, he uses differentdata for the effects and does not estimate the twoequations simultaneously. Thus, the absence-rate co-efficients from the two equations might not becomparable.Several studies have estimated the impact of absentee-ism on productivity using plant-level data. In the pro-duction function of Allen [40], the elasticity of theabsence rate is −0.015, meaning an increase in theabsence rate from 0.1 to 0.2 reduces the output perman-hour by 1%. In addition, Mefford examines theeffect of unions on productivity in 31 plants of a largemultinational firm from 1975 to 82 [41]. He also in-cludes the absence rate into the production function andfinds that the elasticity of the absence rate is −0.033, im-plying if the absence rate increases from 0.1 to 0.2, prod-uctivity will decrease by 2.3%. The direction of theestimated effect in our study is consistent with theseprevious studies yet the magnitude of the effect size isgreater.Coles et al. introduced the idea of the shadow cost ofabsenteeism: the relatively high wage paid by firmsrequiring a low level of absenteeism, to compensateworkers for attending work reliably [17]. They use just-in-time as an indicator of an assembly line productionprocess. Using individual worker level data, they find anassociation between higher wages and lower absencerates; however, the relationship is almost twice as steepin just-in-time firms contrasted to non-just-in-timefirms.Measure of compensationWage rate versus the impact of absenteeism on aggregatewagesIn the absenteeism literature, the measure of opportun-ity cost of absenteeism is usually proxied by the worker’swage rate (wage per unit time) taken from firm data. Inthis paper, however, the wage cost of absenteeism comesfrom an estimate of the impact of worker absenteeismon aggregate wages for workers at a firm. It may dif-fer from a direct measure of the wage rate becausethe equilibrium wage incorporates any effects of ab-senteeism as a compensating differential. For example,the observed wage per day may vary much less be-tween a firm where (for some exogenous reasons) ab-senteeism is common and one where it is rare thandoes the estimate from our wage regression. Most im-portantly, with only an aggregate measure of outputavailable, we prefer to use the aggregate wages at thefirm level in order to obtain the most comparable es-timates. As Hellerstein et al. pointed out, by jointlyestimating the firm-level production function andwage equation, we can conduct straightforward statis-tical tests of the equality of wages and marginal prod-uctivity [27]. Furthermore, the biases from someunobservables are more likely to affect the estimatedabsenteeism impacts on productivity and wages simi-larly when both are estimated at the firm level. Theirimpact on the tests of the equality of marginal prod-uctivity and wages is therefore diminished.Payroll and non-wage benefitsIn our main analysis, we use payroll as a measure ofcompensation. Payroll or wage is only part of the totalemployee compensation. Non-wage benefits are alsoavailable to employees, e.g., health related benefits (e.g.dental care, life insurance), pay related benefits (e.g.severance allowances), or pension related benefits. As arobustness test, we also use the total compensation (pay-roll plus non-wage benefits) as the outcome in our wageequation.Measure of absenteeismBecause we are primarily interested in estimating theproductivity loss due to illness for applications inhealth care economic evaluation studies, an idealZhang et al. Health Economics Review  (2017) 7:3 Page 3 of 14measure of absenteeism would reflect illness-relatedabsences only. However, data limitations dictate thatwe rely on a broader measure of absenteeism. TheWES data used in this study only measure absencesdue to paid sick leave, but not unpaid sick leave. Fol-lowing the definition of Dionne and Dostie [32], ourmeasure of absenteeism includes the number of daysof paid sick leave; other paid leave encompassing edu-cation leave, disability leave, bereavement, marriage,jury duty, and union business; and unpaid leave. Itdoes not include paid vacations, paid paternity/mater-nity leave, or absence due to strikes or lock-outs. Al-though our measure of absenteeism is broader than apure measure of illness-related absenteeism, our find-ings are still useful to determine whether wages are areasonable proxy of the productivity loss due toillness-related absenteeism under the assumption thatillness-related absenteeism and other forms of paidand unpaid leave have a similar impact on wages andoutput.MethodsOur empirical analysis is based on two firm-levelequations which we specify and estimate jointly: aproduction function and a wage equation. The pro-duction function is used to capture productivity ef-fects related to absenteeism and team work at thefirm level, and the wage equation is to capture thecorresponding wage effects. By simultaneously esti-mating the two equations, we can compare the prod-uctivity effects with wage effects to determine theequality of marginal productivity and wages. The trad-itional approach of estimating the wage equationalone to measure the impact of absenteeism does notfully capture productivity differentials associated withdifferent levels of absenteeism.We think it is useful to baseline our results with an es-timate of economy-wide aggregate effects. Thus webegin by estimating a baseline model that restricts theeffect of absenteeism to be the same for team workersand non-team workers and in small and large firms. Wesubsequently relax these restrictions by assuming thatabsenteeism affects team workers and non-team workersdifferently, and then by estimating our model separatelyfor small and large firms.Production functionOur baseline specification of the production function is anextension of the standard Cobb-Douglas [27, 28, 42, 43].See Additional file 1: Appendix B for its complete devi-ation. Because the Cobb-Douglas form is restrictive, weassess the robustness of our estimates to more generalalternatives described in Section 3.4.For each workplace, we start with a simple Cobb-Douglas production function:InQj ¼ α InLAj þ β InKj þ ηFj þ μj ð1Þwhere Qj is output, measured as value added by firm j;LAj is an aggregate labour input defined below, Kj is thecapital stock, Fj is a matrix of various firm characteris-tics, α, β are the elasticity of output with respect tolabour and capital, respectively, η is a vector of parame-ters for firm characteristics and μj is the error term.We divide the labour input into different worker types,that is, workers with different characteristics such asage, sex, education, occupation and team participation.If the total number of characteristics is I and workersare divided into Vi categories by each characteristic i,then the total number of worker types will beYi¼1I V i .Our aggregate labour input LAj can be simplified aftermaking several assumptions: First, we assume perfectsubstitutability among all types of workers and differentmarginal productivity for each worker type [27, 28]. Sec-ond, we assume that the proportion or distribution ofone type of worker defined by one characteristic is con-stant across all other characteristic groups, which is re-ferred to as the equi-proportionate restriction [27, 28].1Third, we assume the relative marginal productivity oftwo types of workers within one characteristic group isequal to those within another characteristic group,which is referred to as the equal relative productivity re-striction [27, 28].2 Fourth, attendance rates have thesame marginal impact on productivity for differentworker types.The aggregate labour input can then be written as(equation 8 from Additional file 1: Appendix B):LAj ¼ 1−aj θλ0;ILj 1þ γG−1 PGj YI−1i¼11þXVi−1v¼1γiv−1 Pivj ! ð2Þwhere aj is the absence rate in firm j, Lj is the number ofall workers in the firm j, PGj is the proportion of teamworkers among all workers in the firm j, i = 1, 2, …, I-1indicates worker characteristics other than team partici-pation, vi = 1, 2, …, Vi-1 represents worker categoriesdivided according to the worker characteristic i, Pivj ¼ LivjLjis the proportion of the worker type iv among all workersin the firm j, θ is the parameter of (1-absence rate), i.e.,the attendance impact on the marginal productivity forany worker type, λ0,I is the marginal productivity for thereference group when work force is divided by I character-istics and absence rate = 0, γG is the relative marginalproductivity of team workers compared to non-teamworkers, and γiv ¼ λivλio is the relative marginal productivityZhang et al. Health Economics Review  (2017) 7:3 Page 4 of 14of one worker type iv to the worker type i0 for each char-acteristic i.By substituting LAj into the simple production function,equation 1, we obtain our baseline specification (equa-tions 9 and 10 from Additional file 1: Appendix B), i.e., a“restricted model” as follows:InQj ¼ β0 þ β InKj þ α InLj þ αθ In 1−αj þα In 1þ γG−1 PGj þ αEj þ ηFjþ μjð3ÞWhereEj ¼XI−1i¼1In 1þXVi−1v¼1γ iv−1 Pivj !ð4ÞEj refers to workforce characteristics other than teamparticipation, and β0 is a constant term that incorporatesa In λ0,I.In addition, we relax the fourth assumption for team-work participation, that is, the attendance impact on themarginal productivity for team workers (θG) is differentfrom that for non-team workers (θN). A relatively“complete model” (equations 12 and 13 from Additionalfile 1: Appendix B) is therefore presented as:LAj ¼ λ0;I 1−aj θN Lj 1þ γG 1−aj θG−θN−1 PGj YI−1i¼11þXV i−1v¼1γiv−1 Pivj !ð5Þand,lnQj ¼ β0 þ β lnKj þ α lnLjþ αθN ln 1−aj þ α ln 1þ γG 1−aj θG−θN−1 PGj þ αEj þ ηFj þ μjð6ÞWage equationApplying the same approach as above, wage effects canbe estimated through the relationship between payrolland average absence rate and share of workers partici-pating in a team at the firm level. We write the aggre-gate wage wj as the sum of wage for each worker type.Applying the same assumptions in the production func-tion, the aggregate wage can be simplified as:wj ¼ w0;I 1−aj ζLj 1þ ϕG−1ð ÞPGj YI−1i¼11þXV i−1v¼1ϕiv−1ð ÞPivj ! ð7Þwhere wj is the annual payroll of firm j, w0,I is the wagefor the reference group when work force is divided by Icharacteristics, ζ is the parameter of attendance rate, i.e.,the attendance impact on wages for any worker type, ϕGis the relative wage of team workers to non-teamworkers, ϕiv ¼ wivwi0 is the relative wage of one worker typeiv to the worker type i0 for each characteristic i otherthan team participation.After log transforming equation 7, the “restrictedmodel” for wage equation is written as:lnwj ¼ βw0 þ βw lnKj þ αw lnLj þ ζ ln 1−aj þ ln 1þ ϕG−1ð ÞPGj þ Ewj þ ηwFj þ μw;j ð8Þwhere,Ewj ¼XI−1i¼1ln 1þXv¼1V i−1ϕiv−1ð ÞPivj !ð9Þβw0 is a constant term incorporating w0,I, αw, βw arethe elasticity of wage with respect to labour and capital,respectively, ηw is a vector of parameters for firm char-acteristics and μw,j is the error term.Correspondingly, we assume the attendance impact onwages differ by team participation and thus the relatively“complete model” becomes:wj ¼ w0;I 1−aj ζN Lj 1þ ϕG 1−aj ζG−ζN−1 PGj YI−1i¼11þXV i−1v¼1ϕiv−1ð ÞPivj !ð10Þandlnwj ¼ βw0 þ βw lnKj þ αw lnLjþ ζN ln 1−aj þ ln 1þ ϕG 1−aj ζG−ζN−1 PGj þ Ewj þ ηwFj þ μw;j ð11Þwhere ζN is the impact of attendance rate for non-teamworkers and ζG is the impact of attendance rate for teamworkers.EstimationWe estimate the production function and wage equa-tion simultaneously via nonlinear least squares (NLS)[27, 28]., under the assumption that errors are corre-lated across equations (nonlinear seemingly unrelatedregression).3 All observations are weighted usinglinked weights provided by Statistics Canada. Allstandard errors are computed as Statistics Canada’srecommended procedure [44] using 100 sets of pro-vided bootstrap sample weights.Our null hypothesis of primary interest is that the at-tendance coefficient in the production function equalsthe coefficient in the wage equation. In the restrictedmodel, the equality of marginal productivity and wage isZhang et al. Health Economics Review  (2017) 7:3 Page 5 of 14tested by comparing the attendance coefficients, θ and ζ.In the complete model, we compare the two coefficientsfor team workers, θG and ζG, and those for non-teamworkers, θN and ζN, respectively. We also test the equal-ity of relative productivity of team workers to non-teamworkers and their relative wage by comparing (λG − 1)and (ϕG − 1).In order to examine whether parameter estimates varyby firm size, we conduct our analyses separately on twosub-samples: small firms with less than 20 employeesand large firms (the remainder).RobustnessWe undertake further analyses to assess the robustnessof our estimates. First, we relax restrictions on the func-tional form of our production function by estimating aspecification using the much more flexible translogform. Second, we re-estimate our model using totalcompensation (payroll plus non-wage benefits) insteadof payroll as the outcome of the wage equation.Third, a key issue in the estimation of productionfunctions is the potential correlation between inputlevels and unobserved firm-specific productivity shocks.Firms that have a large positive productivity shock mayrespond by using more inputs, giving rise to an endo-geneity issue [45]. Following Hellerstein et al. [27], weaddress this issue by using value-added as the measureof output in the production function to avoid estimatinga coefficient on materials. We also attempt to correct forthe potential bias by estimating the model on first differ-ences, which eliminates the effect of any time-invariantunobserved heterogeneity that jointly affects productivityand wages. We also apply Levinsohn and Petrin’s ap-proach [46] using intermediate inputs (expenses on ma-terials which are subtracted out in our value-addedproduction function) to address the simultaneity prob-lem. Specifically, we estimate parameters of our value-added production function using NLS by adding a third-order or a fourth-order polynomial approximation incapital and material inputs [47].Finally, we conduct sensitivity analyses to examine theimpacts of some of the assumptions embodied in ourbaseline specification. We relax the equi-proportionaterestriction between occupation, age, sex, education (>university bachelor versus bachelor and below) and teamparticipation, respectively.4 That restriction also impliesthat the firm-average absence rate is common to allworker types. To test the impact of this assumption, weallow the average absence rate to differ for team workersand non-team workers in each firm. That is, the firm-average absence rate in the complete model is replacedwith the firm-average absence rate of team workers andthe absence rate of non-team workers, correspondingly,as follows.LAj ¼ 1−aGj θGλG;0;I−1 LGjYI−1i¼11þXV i−1v¼1γ iv−1 Pivj !þ 1−aNj θNλN ;0;I−1LNjYI−1i¼11þXV i−1v¼1γiv−1 Pivj !¼ λ0;I 1−aNj θN Lj 1þ γG1−aGj θG1−aNj θN −1 !PGj !YI−1i¼11þXV i−1v¼1γiv−1 Pivj !ð12ÞandlnQj ¼ β0 þ β lnKjþα lnLj þ αθN ln 1−aNj þα ln 1þ γG1−aGj θG1−aNj θN −1 !PGj !þαEj þ ηFj þ μjð14ÞDataThe WES is a survey of Canadian employers and em-ployees conducted by Statistics Canada over the period1999–2006 [48].5 These data have been used to estimateage-based wage and productivity differentials [49] and tocompare wages and marginal productivity for workerswith different levels of education and technology use[50, 51].The sampling frame for the WES includes all Canad-ian workplaces6 in the Statistics Canada Business Regis-try that had paid employees in March of the survey year.The sampling frame for employees comprises all em-ployees working at or on paid leave from the targetedworkplaces in March. In each year between 1999 and2006, Statistics Canada surveyed a representative sampleof approximately 6000 workplaces. The initial sample ofworkplaces was refreshed in odd-number years (2001,2003, and 2005) to reflect attrition and firm births. In1999–2005, Statistics Canada randomly sampled ap-proximately 20,000 employees of sampled firms. Thenumber of employees sampled from a firm was propor-tional to size, up to a maximum of 24. In workplaceswith fewer than 4 employees, all employees were sam-pled. Sampled workers were surveyed for two years, anda new sample of workers was drawn in the next odd-numbered year.Ethical approval for this study is not required becauseit was based exclusively on the WES conducted by Sta-tistics Canada and we did not directly approach thestudy subjects. Our analysis is based on the pooled data1999, 2001, 2003, and 2005 cross-sections.7 We furtherrestrict the sample to workplaces with at least one em-ployee interviewed, operating for profit, and withZhang et al. Health Economics Review  (2017) 7:3 Page 6 of 14positive output. Our sample includes 18,381 observa-tions on 7766 unique workplaces. There are 7784 obser-vations for small firms and 10,597 for large firms.Table 1 illustrates the transition from the gross work-place sample to our final sample in detail.Outcome variablesOur outcome variable in the wage equation is the firm’stotal annual payroll. Our outcomes variables in the pro-duction function is the firm’s output. Following Turcotteand Rennison [50, 51], we define output as value added,where value added is measured as annual gross operat-ing revenues minus expenses on materials.8 Expenses onmaterials equal annual gross operating expendituresminus total gross payroll and expenditures on non-wagebenefits and training.Independent variables of interestOur measure of absenteeism is the absence rate of thefirm’s employees. This is defined as the number of daysof total leave taken by employees, including paid sickleave, other paid leave (e.g., education leave, disabilityleave, bereavement, marriage, jury duty, union business)and unpaid leave [32] in the past 12 months or since theemployee started his/her current job (if less than 12months), divided by the total number of ‘usual work-days’9 over the same time period. The absence rate for afirm is the average absence rate for the employees sur-veyed at that firm. We define the firm’s attendance rateas one minus the absence rate.We identify workers as being a member of a teambased on their reported participation in “a self-directedwork group (semi-autonomous work group or mini-enterprise group) that has a high level of responsibilityfor a particular product or service area” [48].10 In ouranalysis, team workers are those who report participat-ing in such a group ‘frequently’ or ‘always’ and non-teamworkers are those who report participating in such agroup ‘occasionally’ or ‘never’.The Lj in our baseline specification is measured by thenumber of total employees employed by each workplace.Estimation of our production function also requires ameasure of the firm’s capital stock. Unfortunately, thereis no such measure in the WES. We therefore imputethe firm’s capital stock following the approach of Dostie[49] and Turcotte and Rennison [50, 51]. Our imputedcapital measure equals the five-year average capital stockin the firm’s industry, divided by the number of firms ineach industry represented by the WES. The industrycapital stock measure is the geometric (infinite) end-yearnet stock of non-residential capital reported in CANSIMTable 031–0002, obtained from Statistics Canada.11Control variables in our empirical specification includeother characteristics of the firm’s workforce (firm-aver-age proportion of employees grouped by age, sex, educa-tion, occupation, race, immigration status, andmembership in union or collective bargaining agree-ment, separately, included in Ej), workplace characteris-tics (an indicator for selling into an internationalmarket, an indicator for foreign country ownership, re-gion, and industry included in Fj), and calendar yeardummies. More details on the definition of all variableswe used in the study can be found in Additional file 1:Appendix A.Table 2 provides descriptive statistics for variablesused in our analysis. At the workplace level, the averageabsence rate is low (0.02), of which 65% is unpaid leave,19% is paid sick leave and 16% is other paid leave. Theshare of workers in teamwork is 8%. The average age is40 years old and the share of female workers is 54%.Only 38% of workplaces have at least 5 employees sur-veyed. The average number of employees per firm is 15and most firms (85%) fall in the category of 1–19 em-ployees. There are more large firms sampled in the WESsurvey than small firms (Table 1). However, the smallfirms are assigned higher sampling weights than largefirms to represent their much greater number in theCanadian economy.ResultsTable 3 presents parameter estimates for our baselinemodel, which provides an estimate of the economy-wideaggregate effect of absenteeism. With the full set of con-trols, our estimate of the overall effect of attendance onmarginal productivity (0.46) is almost identical to its es-timated effect on wages (0.47). We cannot reject the hy-pothesis that the two coefficients are the same atconventional significance levels. These coefficients canbe interpreted as elasticities: a 1% decline in the attend-ance rate reduces productivity by 0.95*0.46% = 0.44%12and wages by 0.47%.In Table 4, we relax our baseline specification byallowing the coefficient on the attendance rate to differfor team workers and non-team workers. The impact ofattendance is much larger for team workers: coefficientsTable 1 Transition from the gross sample to the final sampleObservations WorkplacesGross sample 43832 9372At least one employee without attritiona 36579 8875For profit 31786 7931Value added >0 30416 7812Odd years 18381 7766Small firms 7784 3870Large firms 10597 4385aIn even survey years, employees who had a different employer or left hisemployer and did not have a new employer were considered as attritionZhang et al. Health Economics Review  (2017) 7:3 Page 7 of 14are 2.38 in the production function and 1.43 in the wageequation. In this specification, the total effect of attend-ance (or absenteeism) on wages and productivity de-pends on both these coefficients and the proportion ofemployees that work in a team. Fig. 1. plots the rate atwhich productivity and wages decline when the absencerate increases by 0.1, at various levels of the firm’s ab-sence rate and proportion of team workers. For example,at a firm where all employees work in teams, an increasein the absence rate from 0.1 to 0.2 reduces output by23.4% and wages by 15.5%. At a firm where 20% of em-ployees work in teams, output would only decline byTable 2 Descriptive statistics at workplace levelVariables WeightedmeanStandarddeviationValue added (,000) 1393.333 38.705Log value added 12.526 0.026Total wage (,000) 524.346 10.281Log wage 11.892 0.021Employment 14.982 0.242Capital stock (,000) 1254.673 59.224Absence rate 0.019 0.001Proportion of workers participatingin a team0.079 0.003Other workforce characteristicsAge 40.472 0.175Proportion of workers by ageAge <35 0.353 0.00635≤ Age < 55 0.525 0.00755≤ Age 0.123 0.005Proportion of female workers 0.542 0.007Proportion of workers by level of education< High school 0.130 0.005High school graduate only 0.203 0.007Under university bachelor (completed/some college or university)0.539 0.007University bachelor 0.092 0.003> University bachelor 0.035 0.002Proportion of workers by occupationManagers/professionals 0.269 0.005Technical/trades/marking/sales/clerical/administrative0.463 0.007Production workers 0.200 0.006Others 0.068 0.004Proportion of ethnic minorities 0.187 0.006Proportion of immigrants 0.179 0.006Proportion of employees withbargaining agreement0.046 0.002Workplace characteristics %Establishment size1–19 employees 84.720–99 employees 13.5100–499 employees 1.6500 employees or more 0.2Number of employees surveyeda1 12.32 16.83 22.9Table 2 Descriptive statistics at workplace level (Continued)4 9.9> =5 38.0International market 5.1Foreign country owned 3.3IndustryForestry, mining, oil, and gas extraction 1.5Labour intensive tertiary manufacturing 3.3Primary product manufacturing 1.2Secondary product manufacturing 2.0Capital intensive tertiary manufacturing 2.6Construction 8.2Transportation, warehousing, wholesale 12.1Communication and other utilities 1.3Retail trade and consumer services 33.7Finance and insurance 5.3Real estate, rental and leasing operations 4.2Business services 13.2Education and health services 9.7Information and cultural industries 1.7RegionAtlantic 8.3Quebec 21.0Ontario 37.2Alberta 11.7British Columbia 14.9Manitoba 3.0Saskatchewan 3.8Yeara1999 25.22001 24.22003 24.22005 26.3Employer weight is used for workplace characteristics; linked weight is usedfor workforce characteristicsaunweighted estimatesZhang et al. Health Economics Review  (2017) 7:3 Page 8 of 148.6% and wages by 7.2%. Correspondingly, the differencebetween the attendance impact on marginal productivityand the impact on wage for team workers is also largerthan that for non-team workers (0.95 versus −0.02)(Table 4). However, the gap is not statistically significant.In Table 5, we further relax our baseline restrictionsby estimating the model separately on sub-samples ofsmall and large firms. The impact of non-team workers’attendance on output and wages is smaller for smallfirms than for large firms: coefficients are 0.47 versus1.32 in the production function and 0.44 versus 1.08 inthe wage equation. As hypothesized, the difference be-tween the two effects are not significantly different fromzero in small firms (0.03) or large firms (0.24). In con-trast, the impact of team workers’ attendance is muchlarger for small firms than for large firms. The product-ivity coefficients are 4.97 versus −0.76, and the wage co-efficients are 2.25 versus −0.33, for small and large firmsrespectively. The difference between the attendance im-pact on output and that on wages is much larger insmall firms (2.72) than in large firms (−0.43). The resultssuggest that in a large firm where all employees work inteams, absenteeism do not have any substantial impacton output or wages. On the other hand, absenteeism sig-nificantly reduces output and wages in small firms whereall employees work in teams. The reduction in output issignificantly higher than the reduction in wages at the10% significance level. The results are consistent withour hypothesis that the absence of team workers has alarger effect on firm-level production than wages insmall firms.Our estimates of the relative productivity and the rela-tive wage of team workers versus non-team workersimply that team workers are more productive and earnmore than non-team workers in the full sample (Tables 3and 4). This difference is statistically significant at the5% level in the specification including all controls. Thedifference between relative productivity and relativewage is larger in small firms but smaller in large firms(Table 5). This implies that on average, the higher wagespaid to team workers are considerably less than theirproductivity differential relative to non-team workers.In Additional file 1: Appendix C, we present parameterestimates for all covariates that are included in themodels of Table 3 to Table 5, as well as the results ofvarious robustness checks. These include estimatesbased on a translog production function (estimated onthe full sample) and using total compensation (payrollplus non-wage benefits) as the outcome of the wageequation. The estimates from these alternative specifica-tions are similar to what we have obtained above. Whenwe consider different absence rates for team workersand non-team workers, the coefficients do not changemuch, which suggests our main analyses are robust.When the equi-proportionate restriction is dropped foroccupation, age, sex and education with team participa-tion, the estimated coefficients change only slightly.13Nevertheless, the qualitative nature of the results staythe same after relaxing these assumptions.We have also re-estimated the model by excluding thecapital stock and the attendance rate coefficients remainvirtually identical. Therefore, we believe that our param-eter estimates are robust to our (imperfect) measure ofthe capital stock.We address the potential endogeneity of absenteeismand team work status in several ways. First, we haveTable 3 Parameter estimates for the restricted modelProduction P value Wage P valueBaseline controlsaLog (total no. of employees) 0.94 (0.02)*** <0.001 1.04 (0.01)*** <0.001Log (capital) 0.04 (0.01)*** <0.001 0.05 (0.01)*** <0.001Attendance rate 0.42 (0.12)*** <0.001 0.41 (0.07)*** <0.001Team 0.66 (0.19)*** <0.001 0.40 (0.08)*** <0.001Difference in attendance rate coefficients 0.01 (0.10) 0.958Difference in team coefficients 0.26 (0.14)* 0.056All controlsbLog (total no. of employees) 0.95 (0.02)*** <0.001 1.08 (0.01)*** <0.001Log (capital) 0.00 (0.01) 0.931 −0.03 (0.01)*** 0.002Attendance rate 0.46 (0.13)*** <0.001 0.47 (0.07)*** <0.001Team 0.26 (0.11)** 0.021 0.08 (0.05) 0.110Difference in attendance rate coefficients −0.01 (0.10) 0.953Difference in team coefficients 0.18 (0.09)** 0.037aModel adjusted for employment, capital stock, and years; bAdjusted for employment, capital stock, occupation, age, sex, education, race, immigrant, bargainingagreement, international market, foreign owned, region, industry and year; Standard error in the bracket; ***p ≤ 0.01; **0.01 < p ≤ 0.05; *0.05 < p ≤ 0.1Zhang et al. Health Economics Review  (2017) 7:3 Page 9 of 14estimated the equations in first differences to removeany time invariant components of the model as a sensi-tivity analysis. The first differences estimates reported inAdditional file 1: Appendix C are similar to the NLS es-timates. Differencing does not eliminate the effect ofcorrelated transitory shocks, however, and these areanother potential source of bias. For example, a chem-ical spill accident may instigate sick leave and a reduc-tion in output. Employee work attendance decisions alsodepend on the slope of the wage-absence tradeoff, whichmay introduced simultaneity problems [40]. In the pres-ence of correlated transitory shocks or simultaneity, anTable 4 Parameter estimates for the complete modelProduction P value Wage P valueBaseline controlsaLog (total no. of employees) 0.94 (0.02)*** <0.001 1.04 (0.01)*** <0.001Log (capital) 0.04 (0.01)*** <0.001 0.05 (0.01)*** <0.001Attendance rate, non-team workers 0.37 (0.12)*** 0.002 0.38 (0.07)*** <0.001Attendance rate, team workers 2.78 (1.44)* 0.054 1.83 (0.84)** 0.029Team 0.75 (0.17)*** <0.001 0.45 (0.08)*** <0.001Difference in attendance coefficients, non-team workers −0.01 (0.10) 0.876Difference in attendance coefficients, team workers 0.95 (0.95) 0.318Difference in team coefficients 0.30 (0.12)** 0.011All controlsbLog (total no. of employees) 0.95 (0.02)*** <0.001 1.08 (0.01)*** <0.001Log (capital) 0.00 (0.01) 0.935 −0.03 (0.01)*** 0.002Attendance rate, non-team workers 0.43 (0.13)*** <0.001 0.45 (0.07)*** <0.001Attendance rate, team workers 2.38 (1.40)* 0.090 1.43 (0.75)* 0.058Team 0.32 (0.12)** 0.012 0.10 (0.05)** 0.041Difference in attendance coefficients, non-team workers −0.02 (0.10) 0.816Difference in attendance coefficients, team workers 0.95 (1.00) 0.341Difference in team coefficients 0.21 (0.10)** 0.030aModel adjusted for employment, capital stock, and yearsbAdjusted for employment, capital stock, occupation, age, sex, education, race, immigrant, bargaining agreement, international market, foreign owned, region,industry and year; Standard error in the bracket; ***p ≤ 0.01; **0.01 < p ≤ 0.05; *0.05 < p ≤ 0.1Fig. 1 Rate at which output and wages decline for a 0.1 increase in the absence rate, at various levels of the firm’s absence rate and proportionof team workersZhang et al. Health Economics Review  (2017) 7:3 Page 10 of 14instrumental variable (IV) approach [30, 52, 53] can beused to consistently estimate parameters. We have esti-mated IV specifications of our model using the laggedattendance rate as an instrument. However this instru-ment turns out to be weak (F-statistic < 10), and we wereunable to identify other valid instruments in theWES. We therefore adopt the Levinsohn and Petrinapproach [46] and obtain estimates similar to ourmain findings. Overall, we find our estimates to bestable across different specifications, and this providesstrong evidence in support of our main conclusionsthat wages underestimate the productivity loss due toabsenteeism in the presence of team production,especially in small firms.Discussion and conclusionsThis study is the first to test the equality of the esti-mated absenteeism impacts on marginal productivityand wages using linked employer-employee data. Ourfindings support the theoretical predictions of Pauly etal. [9, 11] and provide compelling evidence that theproductivity loss due to worker absence exceeds thewage for team workers, especially in small firms.Our findings highlight that the productivity loss due toabsenteeism among team workers substantially exceedsthe wage in small firms. Interestingly, such a wage-productivity gap is absent in large firms. This may reflectdifferences in compensation policy between large andsmall firms, or differences in substitution possibilities.While team workers are more productive and earnhigher wages than non-team workers, our findings fur-ther imply that their higher marginal productivity ex-ceeds the wage premium they receive. Moreover,although we find that wages underestimate the product-ivity loss due to absenteeism for team workers, our esti-mates indicate that wages are reasonable estimate of theproductivity loss due to absenteeism for non-teamworkers.It is worth noticing that this study is an aggregate orecologic study that has focused on the effect of teamwork at the firm level rather than at individual workerlevel due to a lack of individual-level output data. Thus,it might be subject to ecological bias. According toGreenland and Morgenstern [54], ecological bias canoccur if confounders or other factors affecting output orwages are differentially distributed across firms (i.e.,Table 5 Parameter estimates for the complete model by firm sizeSmall firms Large firmsProduction P value Wage P value Production P value Wage P valueBaseline controlsaLog (total no. of employees) 0.87 (0.03)*** <0.001 1.04 (0.02)*** <0.001 1.07 (0.02)*** <0.001 1.01 (0.02)*** <0.001Log (capital) 0.03 (0.01)*** 0.005 0.04 (0.01)*** <0.001 0.09 (0.01)*** <0.001 0.08 (0.01)*** <0.001Attendance rate, non-team workers 0.39 (0.14)*** 0.005 0.36 (0.08)*** <0.001 1.95 (0.80)** 0.015 1.66 (0.58)*** 0.004Attendance rate, team workers 6.34 (2.25)*** 0.005 3.01 (1.03)*** 0.004 −0.57 (0.76) 0.449 −0.02 (0.70) 0.974Team 0.75 (0.27)*** 0.005 0.35 (0.10)*** <0.001 0.71 (0.15)*** <0.001 0.63 (0.12)*** <0.001Difference in attendance coefficients,non-team workers0.04 (0.11) 0.745 0.29 (0.36) 0.429Difference in attendance coefficients,team workers3.33 (1.59)** 0.036 −0.55 (0.70) 0.431Difference in team coefficients 0.40 (0.21)* 0.056 0.08 (0.10) 0.433All controlsbLog (total no. of employees) 0.88 (0.03)*** <0.001 1.07 (0.02)*** <0.001 1.10 (0.02)*** <0.001 1.03 (0.02)*** <0.001Log (capital) 0.00 (0.02) 0.939 −0.03 (0.01)*** 0.006 0.00 (0.01) 0.879 −0.01 (0.01) 0.263Attendance rate, non-team workers 0.47 (0.14)*** 0.001 0.44 (0.06)*** <0.001 1.32 (0.70)* 0.061 1.08 (0.47)** 0.021Attendance rate, team workers 4.97 (1.87)*** 0.008 2.25 (0.95)** 0.018 −0.76 (0.73) 0.300 −0.33 (0.64) 0.609Team 0.33 (0.18)* 0.073 0.06 (0.06) 0.260 0.19 (0.10)* 0.054 0.09 (0.07) 0.213Difference in attendance coefficients,non-team workers0.03 (0.12) 0.811 0.24 (0.37) 0.511Difference in attendance coefficients,team workers2.72 (1.49)* 0.068 −0.43 (0.72) 0.549Difference in team coefficients 0.27 (0.16)* 0.091 0.10 (0.07) 0.157Small firms are those with less than 20 employees; large firms are the remainderaModel adjusted for employment, capital stock, and yearsbAdjusted for employment, capital stock, occupation, age, sex, education, race, immigrant, bargaining agreement, international market, foreign owned, region,industry and year; Standard error in the bracket; ***p ≤ 0.01; **0.01 < p ≤ 0.05; *0.05 < p ≤ 0.1Zhang et al. Health Economics Review  (2017) 7:3 Page 11 of 14confounding by firms) or when the effects of absentee-ism and team work on output and wages vary acrossfirms (i.e., effect modification by firms). To minimize thebias, in our regression models, we have adjusted forfirms’ workforce characteristics that potentially affectoutput and wages, which were derived from individual-level worker data. Furthermore, we are more interestedin the equality of the effects of absenteeism and teamwork in the two equations: production equation andwage equation. By jointly estimating the two equationsat the firm level, the bias is more likely to affect the esti-mated effects on output and wages similarly [27] andthus the impact of bias on the tests of the equality ofmarginal productivity and wages might be diminished.Collectively, our findings help to value the burden ofillness-related absenteeism, by establishing situationswhere the wage can be used as a reasonable proxy forlost productivity, and situations where it will underesti-mate the loss. This is important for economic evalua-tions that seek to measure the productivity gain or lossof a health care technology/intervention, which in turncan impact policy makers’ funding decisions. Other re-searchers have proposed a multiplier to adjust wages toestimate the productivity burden of illness or the prod-uctivity gain from a health care intervention [9, 11, 55].Our study provides a justification for such a multiplier.In practice, the productivity loss can be estimated by cal-culating the measured number of absent workdays dueto health problems, multiplied by the daily wage and themultiplier.Finally, we have deliberately avoided being prescriptivewith respect to the method that should be employed inmeasuring productivity losses in economic evaluations.We believe that the appropriate measurement approach(which we focus on above) has many dimensions and inthis study our intention was to highlight the welfare eco-nomic implications of under/over estimating productiv-ity impacts due to absenteeism. We hope that the debateon the inclusion or exclusion of productivity losses ineconomic evaluations will be informed by this work overand above the normative aspects of the controversy.Endnotes1For example, older workers are assumed to be equallyrepresented among team workers and non-teamworkers; the distribution of absence rate is the sameacross different worker types.2For instance, the relative marginal productivity ofolder workers versus younger workers among teamworkers is assumed to be the same as those among non-team workers.3We have also estimated the equations in first differ-ences to remove the firm-level fixed effects. The esti-mates were similar to the NLS estimates but veryimprecise due to the large number of implied firm ef-fects relative to the sample size. The results are includedin Appendix C.4For example, when dropping the restriction betweensex and team participation, we allow the proportion ofteam workers to differ in female and male employees.The new specification includes the proportion of femaleteam workers, proportion of male team workers andproportion of female non-team workers as the independ-ent variables.5Only employers were surveyed in 2006.6Employers in Yukon, Nunavut, and the NorthwestTerritories are excluded from the survey, as are thoseoperating in crop production, animal production, fishing,hunting and trapping, private households, religious orga-nizations and public administration.7We do not use data from even-numbered years fortwo reasons. First, employee attrition is high in their sec-ond survey year and is likely nonrandom [56]. Second,many sampled workers change employers between sur-vey years and only limited information is collected abouttheir new employer.8Using value added as an output measure helps ad-dress the potential endogeneity of materials by avoidingestimation of a coefficient on materials [27, 50, 51]. An-other advantage of a value-added specification is that itimproves comparability of data across industries andacross workplaces within industries when their degree ofvertical integration differs [27].9The total number of usual workdays equals to thenumber of days per week that employees usually workmultiplied by the number of weeks per year they usuallywork.10More information on self-directed work group wasprovided in the question, i.e., “In such systems, part ofyour pay is normally related to group performance. Self-directed work groups: 1) Are responsible for productionof a fixed product or service, and have a high degree ofautonomy in how they organize themselves to producethat product or service. 2) Act almost as ‘businesseswithin businesses’. 3) Often have incentives related toproductivity, timeliness and quality. 4) While most havea designated leader, other members also contribute tothe organization of the group’s activities.”11Although firms in the WES are classified into indus-tries according to 6-digit North American Industry Clas-sification System (NAICS) (a total of 837 uniqueindustries), the capital stock information provided byStatistics Canada is only available for 247 industries atvarying levels of NAICS detail (2–6 digits, depending onindustry). The 247 industries are not exclusive becauseboth higher level and lower level of their NACIS are in-cluded for some industries. Eventually, a total of exclu-sive 201 NACISs are used: 2 in 2 digits, 70 in 3 digits,Zhang et al. Health Economics Review  (2017) 7:3 Page 12 of 14107 in 4 digits, 20 in 5 digits and 2 in 6 digits. Hence, toimpute a net stock estimate, we had to impute somefirm’s capital stock using the average value in a higher-level aggregate of the firm’s industry.12Note the output elasticity of labour is 0.95.13Results are not presented but will be available uponrequest.Additional fileAdditional file 1: Appendix A. Definition of variables. Appendix B.Equations. Appendix C. Additional results. (DOCX 89 kb)FundingThis study was supported by a Canadian Institutes of Health Research (CIHR)operating grant (#231571). Wei Zhang was funded by the CIHR DoctoralResearch Award in the Area of Public Health Research and is supported bythe Michael Smith Foundation for Health Research Postdoctoral FellowshipAward.Availability of data and materialsThe Workplace and Employee Survey data are held by Statistics Canada.The data access can be applied for through Statistics Canada.Authors’ contributionsWZ designed the study, applied for the access to the data, performed thestatistical analysis, interpreted the analysis results, and wrote the manuscript.HS participated in the development of the econometric models, theinterpretation of the analysis results, and the finalization of the manuscript.SW and AHA participated in the design of the study and the interpretationof the data, and wrote the final manuscript. All authors read and approvedthe final manuscript.Competing interestsThe authors declare that they have no competing interests.Ethics approval and consent to participateThis is a secondary use of the survey data held by Statistics Canada and isexempted from an ethical review. 1) The survey data have been collectedthrough the provisions of the Statistics Act, respondents are informed thatthe survey is voluntary and that all information collected remainsconfidential and is solely used for statistical research purposes. 2) Theindividual data records are anonymous. 3) Access to the survey data isprovided through legislation and regulation. Statistics Canada has acomprehensive regime of policies and procedures to protect theconfidentiality of respondents, and to prosecute violations of legislation anddisciplinary procedures for violations of regulations to protect respondentconfidentiality (Statistics Canada, 2015. Mitigation of risk to respondents ofStatistics Canada’s surveys. URL http://www.statcan.gc.ca/eng/rdc/mitigation).Author details1Centre for Health Evaluation and Outcome Sciences, St. Paul’s Hospital,588-1081 Burrard Street, Vancouver, BC V6Z1Y6, Canada. 2School ofPopulation and Public Health, University of British Columbia, 2206 East Mall,Vancouver, BC V6T1Z3, Canada. 3Department of Economics, Simon FraserUniversity, 8888 University Drive, Burnaby, BC V5A 1S6, Canada.Received: 29 August 2016 Accepted: 9 December 2016References1. Drummond MF. Methods for the economic evaluation of health careprogrammes. 3rd ed. Oxford: Oxford University Press; 2005.2. Gold MR, Siegel JE, Russell LB, Weinstein MC, editors. Cost-Effectiveness inHealth and Medicine. 1st ed. New York: Oxford University Press; 1996.3. Krol M, Papenburg J, Koopmanschap M, Brouwer W. Do productivity costsmatter?: the impact of including productivity costs on the incremental costsof interventions targeted at depressive disorders. Pharmacoeconomics.2011;29:601–19.4. Krol M, Papenburg J, Tan SS, Brouwer W, Hakkaart L. A noticeabledifference? Productivity costs related to paid and unpaid work in economicevaluations on expensive drugs. Eur J Health Econ. 2016;17:391–402.5. Johannesson M, Jönsson B, Jönsson L, Kobelt G, Zethraeus N. Why shouldeconomic evaluations of medical innovations have a societal perspective?London: Office of Health Economics; 2009. Report No.: No. 51.6. Jönsson B. Ten arguments for a societal perspective in the economicevaluation of medical innovations. Eur J Health Econ. 2009;10:357–9.7. Berger ML, Murray JF, Xu J, Pauly M. Alternative valuations of work lossand productivity. J Occup Environ Med. 2001;43:18–24.8. Johannesson M. The willingness to pay for health changes, the human-capital approach and the external costs. Health Policy. 1996;36:231–44.9. Pauly MV, Nicholson S, Xu J, Polsky D, Danzon PM, Murray JF, et al. Ageneral model of the impact of absenteeism on employers and employees.Health Econ. 2002;11:221–31.10. Zhang W, Bansback N, Anis AH. Measuring and valuing productivity lossdue to poor health: A critical review. Soc Sci Med. 2011;72:185–92.11. Pauly MV, Nicholson S, Polsky D, Berger ML, Sharda C. Valuing reductionsin on-the-job illness: “presenteeism” from managerial and economicperspectives. Health Econ. 2008;17:469–85.12. Brown C, Medoff J. The Employer Size-Wage Effect. J Polit Econ. 1989;97:1027–59.13. Fox JT. Firm‐Size Wage Gaps, Job Responsibility, and Hierarchical Matching.J Labor Econ. 2009;27:83–126.14. Goux D, Maurin E. Persistence of Interindustry Wage Differentials:A Reexamination Using Matched Worker‐Firm Panel Data. J Labor Econ.1999;17:492–533.15. Groshen EL. Sources of Intra-Industry Wage Dispersion: How Much DoEmployers Matter? Q J Econ. 1991;106:869–84.16. Krueger AB, Summers LH. Efficiency Wages and the Inter-Industry WageStructure. Econometrica. 1988;56:259–93.17. Coles M, Lanfranchi J, Skalli A, Treble J. Pay, Technology, and the Cost ofWorker Absence. Econ Inq. 2007;45:268–85.18. Hamilton BH, Nickerson JA, Owan H. Team Incentives and WorkerHeterogeneity: An Empirical Analysis of the Impact of Teams onProductivity and Participation. J Polit Econ. 2003;111:465–97.19. Card D, Lemieux T, Riddell WC. Unions and wage inequality. J Labor Res.2004;25:519–59.20. Freeman RB, Medoff JL, Feeman RB. What Do Unions Do? New York:HarperCollins Canada / Basic Books; 1984.21. Hoynes H. The Employment, Earnings, and Income of Less Skilled WorkersOver the Business Cycle [Internet]. National Bureau of Economic Research;1999 Jun. Report No.: 7188. Available from: http://www.nber.org/papers/w718822. Dustmann C, Glitz A, Vogel T. Employment, wages, and the economic cycle:Differences between immigrants and natives. Eur Econ Rev. 2010;54:1–17.23. Borjas GJ, Ramey VA. Foreign Competition, Market Power, and WageInequality. Q J Econ. 1995;110:1075–110.24. Guadalupe M. Product Market Competition, Returns to Skill, and WageInequality. J Labor Econ. 2007;25:439–74.25. Fortin NM, Lemieux T. Institutional Changes and Rising Wage Inequality: Isthere a Linkage? J Econ Perspect. 1997;11:75–96.26. Koeniger W, Leonardi M, Nunziata L. Labor Market Institutions and WageInequality. Ind Labor Relat Rev. 2007;60:340–56.27. Hellerstein JK, Neumark D, Troske KR. Wages, Productivity, and WorkerCharacteristics: Evidence from Plant‐Level Production Functions and WageEquations. J Labor Econ. 1999;17:409–46.28. Hellerstein JK, Neumark D. Sex, Wages, and Productivity: An EmpiricalAnalysis of Israeli Firm-Level Data. Int Econ Rev. 1999;40:95–123.29. Hægeland T, Klette TJ. Do higher wages reflect higher productivity?Education, gender and experience premiums in a matched plant-workerdata set. In: Haltiwanger JC, Lane JI, Spletzer JR, Theeuwes JJM, Troske KR,editors. The creation and analysis of employer-employee matched data.New York: Elsevier; 1999. p. 231–59.30. van Ours JC, Stoeldraijer L. Age, wage and productivity [Internet]. IZA; 2010.Report No.: 4765. Available from: http://econpapers.repec.org/paper/izaizadps/dp4765.htm.31. Brown S, Sessions JG. The Economics of Absence: Theory and Evidence.J Econ Surv. 1996;10:23–53.32. Dionne G, Dostie B. New Evidence on the Determinants of AbsenteeismUsing Linked Employer-Employee Data. Ind Labor Relat Rev. 2007;61:108–20.Zhang et al. Health Economics Review  (2017) 7:3 Page 13 of 1433. Heywood JS, Jirjahn U. Teams, Teamwork and Absence. Scand J Econ. 2004;106:765–82.34. Markussen S, Røed K, Røgeberg OJ, Gaure S. The anatomy of absenteeism.J Health Econ. 2011;30:277–92.35. Ose SO. Working conditions, compensation and absenteeism. J Health Econ.2005;24:161–88.36. Darr W, Johns G. Work strain, health, and absenteeism: a meta-analysis.J Occup Health Psychol. 2008;13:293–318.37. Dewa CS, Loong D, Bonato S, Hees H. Incidence rates of sickness absencerelated to mental disorders: a systematic literature review. BMC PublicHealth. 2014;14:205.38. Keech M, Beardsworth P. The impact of influenza on working days lost:a review of the literature. Pharmacoeconomics. 2008;26:911–24.39. Neovius K, Johansson K, Kark M, Neovius M. Obesity status and sick leave:a systematic review. Obes Rev. 2009;10:17–27.40. Allen SG. How Much Does Absenteeism Cost? J Hum Resour. 1983;18:379–93.41. Mefford RN. The Effect of Unions on Productivity in a MultinationalManufacturing Firm. Ind Labor Relat Rev. 1986;40:105–14.42. Cobb CW, Douglas PH. A theory of production. Am Econ Rev. 1928;18:139–65.43. Zhang W, Sun H, Woodcock S, Anis A. Illness related wage and productivitylosses: Valuing “presenteeism.”. Soc Sci Med. 2015;147:62–71.44. Statistics Canada. Guide to the Analysis of the Workplace and EmployeeSurvey [Internet], Report No.: Catalogue no. 71–221–GIE. Ottawa: StatisticsCanada; 2004. Available from: http://www.statcan.gc.ca/pub/71-221-g/71-221-g2007001-eng.pdf.45. Griliches Z, Mairesse J. Production Functions: The Search for Identification[Internet]. National Bureau of Economic Research; 1995. Report No.: 5067.Available from: http://www.nber.org/papers/w506746. Levinsohn J, Petrin A. Estimating Production Functions Using Inputs toControl for Unobservables. Rev Econ Stud. 2003;70:317–41.47. Petrin A, Poi BP, Levinsohn J. Production Function Estimation in STATAusing Inputs to Control for Unobservables. Stata J. 2004;4:113–23.48. Statistics Canada. Workplace and Employee Survey (WES) [Internet]. [cited2014 Feb 25]. Available from: http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=261549. Dostie B. Wage, productivity and aging [Internet]. IZA; 2006. Report No.:2496. Availabe from: http://econpapers.repec.org/paper/izaizadps/dp2496.htm.50. Turcotte J, Rennison LW. The Link between Technology Use, Human Capital,Productivity and Wages: Firm-level Evidence. Int Productivity Monit. 2004;9:25–36.51. Turcotte J, Rennison LW. Productivity and wages: Measuring the effect ofhuman capital and technology use from linked employer- employee data[Internet]. Department of Finance; 2004 [cited 2013 May 25]. Available from:http://www.fin.gc.ca/pub/pdfs/wp2004-01e.pdf52. Aubert P, Crépon B. La productivité des salariés âgés: une tentatived’estimation. Économie et Statistique. 2003;368:95–119.53. Crépon B, Deniau N, Pérez-Duarte S. Wages, Productivity and WorkerCharacteristics : A French Perspective [Internet]. Centre de Rechercheen Economie et Statistique; 2003. Report No.: 2003–04. Available from:https://ideas.repec.org/p/crs/wpaper/2003-04.html54. Greenland S, Morgenstern H. Ecological Bias, Confounding, and EffectModification. Int J Epidemiol. 1989;18:269–74.55. Zhang W, Bansback N, Boonen A, Severens JL, Anis AH. Development of acomposite questionnaire, the valuation of lost productivity, to valueproductivity losses: application in rheumatoid arthritis. Value Health. 2012;15:46–54.56. Pendakur K, Woodcock S. Glass ceilings or glass doors? Wage disparitywithin and between Firms. J Bus Econ Stat. 2010;28:181–9.Submit your manuscript to a journal and benefi t from:7 Convenient online submission7 Rigorous peer review7 Immediate publication on acceptance7 Open access: articles freely available online7 High visibility within the fi eld7 Retaining the copyright to your article    Submit your next manuscript at 7 springeropen.comZhang et al. Health Economics Review  (2017) 7:3 Page 14 of 14


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items