Changes in Job Quality and Trends in Labor Hours

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 882 October 2006 Changes in Job Quality and Trends in Labor Hours Brahima Coulibaly NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from Social Science Research Network electronic library at http://www.ssrn.com/.

Changes in Job Quality and Trends in Labor Hours Brahima Coulibaly* October 2006 Abstract: Many economic models featuring labor supply decisions, especially in macroeconomic analysis, assume away heterogeneity in the nature of work, or assume that the nature of work is irrelevant to the labor/leisure choice. This paper studies the macroeconomic implications of relaxing this assumption. Estimation from micro data using labor hours, wages, consumption, and nonpecuniary job characteristics suggests that labor supply responds to differences and to changes in the nature of work. Ceteris paribus, some job characteristics induce more labor hours than others do. Labeling the jobs that embed the labor-inducing characteristics as better quality jobs, the study estimates a Job Quality index for the aggregate U.S. economy from 1850 to 2000. The results suggest that over the same period, improvements in Job Quality accounted for at least 20.4 percent of growth in labor hours. Keywords: Job Quality, Labor Supply, Trends in Labor Hours JEL classifications: E24; J22; O47 ___________________________ * Mailing address: Division of International Finance, Board of Governors, Federal Reserve System, Mail Stop 24, Washington DC 20551,USA; email: brahima.coulibaly@frb.gov. Tel.: (202)-452-2609; fax: (202)-736-5638. The author thanks Matthew Shapiro, Miles Kimball, Robert Barsky, and Charles Brown for encouraging work on this topic and for constructive feedback. Helpful comments by Gary Solon, John Laitner, Linda Tesar, Kerwin Charles, and Herman Kamil are gratefully acknowledged. My gratitude extends to seminar participants at the University of Michigan, the University of Notre Dame, Clemson University, Drexel University, the Federal Reserve Banks of Boston and Philadelphia, and at the Board of Governors of the Federal Reserve System. I am responsible for any remaining errors and omissions. Work on this project started while I was a graduate student in the Department of Economics and a Research Associate at the Institute for Social Research at the University of Michigan. The views in this paper are solely the responsibility of the author and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or of any other person associated with the Federal Reserve System.

I Introduction A well documented regularity of time series variables of the aggregate United States economy is the lack of trend in per capita labor hours. For example, between 1950 and 2000 economic variables such as consumption per capita and GDP per capita nearly tripled. Per capita labor hours, on the other hand, remained relatively stable with no sign of a strong trend (see (cid:133)gure I). While this observation is widely accepted among economists nowadays, it was not always expected. In his 1930 essay on the (cid:147)Economic Possibilities for Our Grandchildren(cid:148), Keynes predicted an increase in leisure as income rises.1 In other words, with the rise in income, we should have observed a strong downward trend in work hours. This decrease was quite conceivable, but didn(cid:146)t happen. There are two leading explanations for the lack of trends in labor hours.2 One explanation is that the income and substitution e⁄ects o⁄set each other. As the real wage grows, two opposing e⁄ects determine a worker(cid:146)s labor supply decision. The substitution e⁄ect tends to motivate a worker to increase labor hours to take advantage of the higher real wage rate, whereas the income e⁄ect induces a reduction in hours since the worker is wealthier for a given amount of labor input. Since the two e⁄ects have opposite signs and are equal in magnitudes, labor hours do not change. Asecondexplanation-providedbythehomeproductionliterature(cid:150)isbasedonthenotionthat the quantity and productivity of hours allocated to home production (home sector) is important in understanding labor supplied in the market sector (Benhabib, Rogerson, and Wright [1991], and Greenwood and Hercowitz [1991]).3 Technological progress in both the market sector and in the home sector a⁄ect labor supply. In a scenario where technological progress in both sectors occur at roughly the same pace, there would be no substitution away from one sector in favor of the other, hence labor hours remain relatively constant. Reservations about this theory include its reliance on speci(cid:133)c technological progress in the two sectors, and on the size of the elasticity of substitution between the home and the market sector. There is, however, little independent empirical evidence on the estimation of home technological progress, or on the elasticity of substitution between the two sectors. This study o⁄ers an alternative explanation. Speci(cid:133)cally, it argues that the nature of work is relevant for the worker(cid:146)s labor supply decision. Some job characteristics induce more disutility for work than others do, and the quality of these jobs is lower. The lower the quality, the more likely a worker will reduce hours following an increase in the real wage, and as the economy grows, the aggregate Job Quality improves, thereby, causing a decline in disutility for work. The decline in disutility induces more labor. In other words, the income and substitution e⁄ects do not necessarily cancel. The observed cancellation is due to the Job Quality e⁄ect, which complements the substitution e⁄ect to counterbalance the income e⁄ect. An implication of this hypothesis is that 1

holding the Job Quality constant, we should observe a domination of the income e⁄ect, consistent with predictions in Keynes [1930]. Using a unique dataset on labor hours, wages, and nonpecuniary job characteristics from the Health and Retirement Study (HRS)4, this study provides empirical evidence suggesting that some job characteristics induce more labor hours than others do. In the aggregate economy, the share of jobs with these characteristics has increased over time, resulting in an improvement in the economy wide Job Quality index. Between 1850 and 2000, the increase in Job Quality contributed to at least 20.4 percent of growth in labor hours in the U.S. The remainder of the paper is organized as follows: Section II de(cid:133)nes Job Quality and introduces the theory. Section III provides some empirical evidence supporting the theory. Section IV is devoted to the analysis of Job Quality in the aggregate economy. Section V suggests some implications of the results. Section VI discusses some caveats, and Section VII concludes. II The Job Quality Theory The theory of job quality draws inspiration from observed changes in the nature of work over time and from the heterogeneity in hours by occupation. Table I displays the share of total employment by industry in the United States in 1850, 1920, and in 2000. In 1850, Agriculture made up approximately 58.0 percent of total employment. The percentage decreased to 26.2 percent in 1920, and to 2.9 percent in 2000. The shares of employment in the Manufacturing sector were 12.3 percent, 24.6 percent, and 14.9 percent in the corresponding years. For the retail sector, the corresponding percentages were 8.1 percent, 9.9 percent, and 17.9 percent. Table II, presents the share of employment by occupation in 1850, 1920, and in 2000. In 1850 3.3 percent of total employees were Professionals, the percentage increased to 5.5 percent in 1920 and to 21.1 percent in 2000. Farmers and Farm laborers made up 57.5 percent, 25.8 percent, and 1.7 percent of employment in 1850, 1920, and 2000, respectively.5 Furthermore, for some occupations, the nature the work has evolved over time. For example, some sales professions evolved from door-to-door to telemarketing. In the cross-section, there is heterogeneity in average work hours by occupation. Rones, Gardner, and IIg [1997] analyzed trends in hours of work in the United States between 1976 and 1993. The most noteworthy observation is the increase in the share of persons who are working very long hours, exceeding the average workweek by a full eight-hour day. They further (cid:133)nd that the long workweek itself is associated with certain types of occupations.6 In light of these signi(cid:133)cant changes in the nature of employment in the US in the last century and half, it is surprising that little research has been devoted to understanding how changes in 2

nature of work have a⁄ected workers(cid:146)labor supply decisions. This study investigates the issue. II.1 De(cid:133)ning Job Quality The fact that workers require some compensation in order to supply labor indicates that there is some distaste for working. The distaste for work results not only from the opportunity cost of going to work, but it also depends on the nature of the tasks involved. In this study, Job Quality is de(cid:133)ned as the degree of satisfaction, or the pleasantness associated with the process of working. Job Quality does not refer to any bene(cid:133)ts such as wages, employer-provided health care, fame, or status associated with or resulting from the job, nor does it refer to the quality of the output.7 Some illustrative examples are in order. A few decades ago, an economics professor would have had to draw charts and graphs manually on the board to illustrate an economic theory. Nowadays, he/she can perform the same task with less pain by means of an overhead projector. Similarly, the construction of a tunnel or highway would have required several manually intensive labor hours. The same task is performed more e¢ ciently and with far less discomfort by today(cid:146)s heavy machinery.8 Acomparisonofjobsinthecross-sectionillustratestheJobQualitytheoryaswell. Forexample, ceteris paribus, a garbage worker will most likely display a higher disutility for work compared to an o¢ ce worker for a given amount of labor hours. Collecting garbage is not as pleasant as sorting mail in a clean o¢ ce environment. Likewise, Job Quality can di⁄er for two workers in the same occupation. A computer programmer with a high-resolution computer screen will (cid:133)nd work more pleasant compared to an identical worker whose computer screen resolution is poor.9 II.2 Modeling To understand the Job Quality hypothesis, recall the following intratemporal equilibrium for a lifetime utility maximizing worker over consumption and leisure. W U t N (1) = P (cid:0)U t C Intuitively, equation (1) states that the real wage is equal to the marginal rate of substitution between consumption and labor. A worker cannot increase utility by giving up one unit of leisure, and spending the proceeds Wt on units of consumption valued at U per unit. Neither can he/she Pt C improve utility by giving up one unit of consumption, and spending the proceeds on leisure valued at U per unit. N Considerthesameworker,andassumeanimprovementinthequalityofhis/herwork,orassume he/sheswitchestoahigherqualityjobwithanequalwagewhilemaintainingthesameconsumption 3

pro(cid:133)le so that U improves. Holding real wages and consumption constant, the marginal utility N for work improves, causing the rate of substitution between labor and consumption to exceed the real wage. The worker exploits the improvement in Job Quality to increase utility by providing more labor until the marginal rate of substitution between labor and consumption matches the real wage. AnotheraspectoftheJobQualityhypothesisisthroughtheresponseofaworkertoapermanent increase in the real wage rate. The Job Quality hypothesis suggests that following the increase in the real wage rate, a worker in a high-quality job will decrease labor hours by less, compared to a worker in a low quality job. In other words, the income e⁄ect tends to be higher in lower-quality jobs. Over time, as income grows, the share of employment in lower-quality jobs decreases causing an overall decline in the income e⁄ect. Job Quality and the Consumer Problem In most macroeconomic models, the utility function contains two choice variables, labor or leisure, and consumption. Any utility function with only consumption and labor (or leisure) implicitly assumes homogeneity in the nature of work, that is, the nature of work is irrelevant to the worker(cid:146)s labor supply decision. Allowing for heterogeneity in jobs, consists of including a third variable in the utility function that captures di⁄erences in Job Quality. Suppose a job consists of several nonpecuniary characteristics that are relevant to the worker(cid:146)s labor supply decision. The worker chooses leisure L and consumption C given a bundle of nonpecuniary job characteristics J ;:::;J and a vector of demographic variables, X. The model assumes the wage rate and the 1 I characteristics are exogenous for simpli(cid:133)cation.10 (2) Max U (C;L;J ;:::;J ) 1 I C;L The utility function is maximized subject to the following budget constraint: PC = A+wN: The variable A represents non-labor income, N is total labor annual hours and L = T N, where (cid:0) T is the total time endowment. X represents the set of demographics such as age, gender etc. C is a composite of market consumption goods purchased with the labor income wN. w is the prevailing wage rate. To simplify the model, let(cid:146)s omit X, and let Q represent a composite of job characteristics. We can rewrite the maximization problem as follows: Max U (C;N;Q) C; N 4

S:t: PC = A+wN where Q = (cid:9)(J ;:::;J ) is a composite index of the nonpecuniary job characteristics. In addition 1 I to the standard assumptions @U > 0, @U < 0, let(cid:146)s assume that agents prefer higher-quality jobs, @C @N @U > 0, and that @2U > 0. The last assumption states that job quality and labor hours are @Q @N@Q interdependent. The higher Q is, the higher the marginal utility for work holding labor hours constant. Let(cid:146)s further assume that consumption and quality are independent(U = U = 0). QC CQ With this modi(cid:133)cation of the utility function, the (cid:133)rst order conditions of utility maximization can be re-written as follows: (3) U (C;N;Q) = (cid:21)P C (4) U (C;N;Q) = (cid:21)w N (cid:0) Equation (4) captures labor supply for given levels of C and Q. Combining equations (3) and (4) results in equation (5) below: w U (C;N;Q) N (5) = (cid:0) P U (C;N;Q) C II.3 Econometric Model Assume the utility function takes the following functional form: 1+1 U (C;N;Q) = C t 1 (cid:0) (cid:13) Q (cid:16) t N t (cid:17) 1 (cid:13) (cid:0) (cid:16) 1+(cid:17)1 (cid:0) (cid:17) where, (cid:13) denotes the risk aversion parameter, (cid:17) the real wage labor supply elasticity, and (cid:16) the Job Quality-labor hours elasticity. Q is the Job quality variable, C consumption, and N the observed labor hours. Normalize the price for goods to unity and re-write equation (5) as: (6) Q t (cid:17) (cid:16)+(cid:16) N t (cid:17) 1 = C t(cid:0) (cid:13) W t 5

Taking logarithms of (6) and re-arranging terms results in the following testable cross-section and di⁄erence equations: (7) ln(N ) = (cid:17)ln(W ) (cid:13)(cid:17)ln(C ) (cid:16)(1+(cid:17))ln(Q ) t t t t (cid:0) (cid:0) (8) (cid:1)ln(N ) = (cid:17)(cid:1)ln(W ) (cid:13)(cid:17)(cid:1)ln(C ) (cid:16)(1+(cid:17))(cid:1)ln(Q ) t t t t (cid:0) (cid:0) where (cid:1) denotes the di⁄erence operator. (cid:1)x = x x . Equations (7) and (8) provide testable t t t 1 (cid:0) (cid:0) relationships between labor hours, real wages, consumption, and Job Quality in the cross-section and between two periods. Labor hours and real wages are available in the data. The Job Quality variable is not. It is derived in the following section. III Measuring Job Quality and its E⁄ect on Labor Supply Measuring the e⁄ect of Job Quality on labor supply is a di¢ cult exercise for several reasons. Job Quality appears to be an abstract concept, and as such, di¢ cult to measure. Even with a good measure of Job Quality, how to assess its e⁄ect on labor supply is not obvious. After all, the set of job characteristics that generates disutility for one worker may be attractive to another worker. Workers may display di⁄erent levels of disutility with respect to a given quality job. For example, a worker physically (cid:133)t to lift twenty-(cid:133)ve-pound boxes would have a lower disutility performing this task compared to another worker not well suited for the same task. Just as consumers have di⁄erenttastesandpreferencesforproducts,workershavedi⁄erenttastesandpreferences(forjobs), re(cid:135)ectingacombinationoftheirindividualinterestsandabilities. Justasconsumerspurchasegoods and services that maximize their welfare, workers may sort into jobs that minimize their respective disutility for work. In an ideal scenario, where workers match perfectly to the jobs they desire, the e⁄ect of Job Quality on labor supply will be less signi(cid:133)cant in a cross-section analysis. Despite these di¢ culties, the paper attempts the exercise using a unique data set from the Health and Retirement Study. The measurement of the e⁄ect of Job Quality on labor hours begins withananalysisoflaborsupplyandjobcharacteristics. Ithenbuildontheresultsfromtheanalysis to obtain a measure of Job Quality in the cross-section. The cross-sectional measure provides the basis for the aggregate Job Quality index for the US economy. 6

III.1 Measurement of Job Quality in Micro Data The Health and Retirement Study collects data on labor hours, compensations, and nonpecuniary job characteristics. The responses to the job characteristics questions provide the basis for the quality measure. III.1.1 Job Characteristics Data The Health and Retirement Study asks the following seventeen questions about job characteristics. (cid:147)...Thinking of your job, please tell how often these statements are true(cid:148) (1=All or almost all of the time, 2=Most of the time, 3=Some of the time, 4=None or almost none of the time) o My job requires lots of physical e⁄ort (PHYSICAL) o My job requires lifting heavy loads (LIFTING) o My job requires stooping, kneeling, or crouching (BENDING) o My job requires good eyesight (EYESIGHT) o My job requires intense concentration or attention (ATTENTION) o My job requires skill in dealing with other people (PEOPLE) o My job requires me to work with computers (COMPUTERS) o My job requires me to analyze data or information (DATA) o My job requires me to keep up with the pace set by others (PACE) o My job requires me to do the same things over and over (REPETITION) o My job requires that I learn new things (LEARN) o I have a lot of freedom to decide how I do my own work (FREEDOM) o People I work with are helpful and friendly (COWORKERS) In addition to these questions, the study asks respondents the following: (cid:147)...thinking of your job, this time please indicate how much you agree or disagree with each statement(cid:148). (1=Strongly agree, 2=Agree, 3=Disagree, 4=Strongly disagree) o I could do my job a lot better if I received training to update my job skills (TRAINING) o My job requires me to do more di¢ cult things than it used to (DIFFICULT) o My job requires a very good memory (MEMORY) o My job involves a lot of stress (STRESS) Each of the above statements captures a speci(cid:133)c job characteristic best described by the word in parentheses.11 One can argue that these questions do not cover every aspect of all jobs, but they 7

do capture the main characteristics of most jobs. The following section constructs the Job Quality index by applying the common factor analysis technique to the answers to these questions.12 Results Before performing the factor analysis, it is worth looking at the correlation matrix of the job characteristics in Table III. It provides insights in the correlation structure of the data. Some job characteristics correlate with others and some do not. For example, PHYSICAL, LIFTING, BENDING are strongly correlated with each other, and COMPUTERS is highly correlated with DATA. This indicates that the variables capture jobs with similar characteristics. On the other hand, there is little correlation between EYESIGHT and LIFTING or between ATTENTION and LIFTING. These attributes most likely re(cid:135)ect jobs that are not similar. The results from the common factor analysis presented in Table IV suggest seven common factors for the jobs in our data. The (cid:133)rst factor correlates positively with the following variables (in order of signi(cid:133)cance): DATA, COMPUTERS, LEARN, MEMORY, ATTENTION, DIFFICULT, STRESS,PEOPLE,PACE,andcorrelatesnegativelywithPHYSICAL,LIFTING,BENDING,and REPETITION. The second factor correlates positively with such variables as PHYSICAL, LIFT- ING, BENDING, and correlates negatively with COMPUTERS. The last (cid:133)ve factors appear to re(cid:135)ect some variation of the (cid:133)rst two. For example, the third factor is not correlated with PHYS- ICAL, LIFTING, nor BENDING, but correlates positively with FREEDOM and COWORKERS, and negatively with TRAINING and DIFFICULT. Based on the signi(cid:133)cance of the contribution of the factors to the variability of the data, the (cid:133)rst two factors are retained as our measure of job characteristics. The last (cid:133)ve factors are not as desirable as the (cid:133)rst two. They do not correlate highly with the observed variables as evidenced by their low factor loadings and small eigenvalues. Together the (cid:133)rst two factors explain approximately 76 percent of the variation in the observed data. The remainder of the study focuses on these two factors. Overall, the (cid:133)rst factor re(cid:135)ects jobs that are non-physical in nature, whereas the second factor re(cid:135)ects physical jobs. To understand the factors, Tables V and VI present their means by industry andoccupation. Theresultsareinlinewithexpectations. Forexample,wewouldexpectjobsinthe constructionandminingindustriestobephysicallyintensive. ThemeanofthePHYSICALfactoris positive for this industry whereas the mean of the MENTAL factor is negative. Similarly, the mean for the MENTAL factor is positive for the Finance and Management industries. The means for the PHYSICAL factor are negative for the same industries. For the transportation industry, both means are positive, indicating that jobs in the transportation industry contain both mental and physicalcomponents. ForoccupationssuchasManagerial, SpecialtyOperation,TechnicalSupport, or Sales, the mean of the MENTAL factor is positive. For the same occupations, the mean of the 8

PHYSICAL factor is negative. Other occupations such as Household Services and Protection have negative means for both factors. To further understand the di⁄erences in the two categories of jobs, Table VII displays the correlation between the factors and various job characteristics such as bene(cid:133)ts, compensation, and job (cid:135)exibility. The results show that the MENTAL factor correlates positively with Salary, and Net Wealth. The correlation coe¢ cient for the PHYSICAL factor is negative for these variables. Workers in MENTAL jobs also tend to have a higher education. They tend to be white, male, and with the (cid:135)exibility to increase hours, but they cannot decrease them. Furthermore, they tend to have better bene(cid:133)t packages such as higher number of paid vacation weeks per year, higher number of paid sick days, and an employer-provided retirement plan. They tend to make Pay and Promotion decisions for other employees, and they are healthier. Workers in PHYSICAL jobs, on the other hand, do not have such generous bene(cid:133)t packages. They tend to be members of an employee union, and do not work as much overtime as their counterparts in MENTAL jobs. Job insecurity and injury rates tend to be higher for workers in the PHYSICAL jobs. III.2 Job Quality and Labor supply This section examines the relationship between job characteristics and labor supply by estimating the labor supply equations derived in section II.3 while controlling for other factors that could in(cid:135)uence estimation. The (cid:133)rst model estimates the cross-sectional equation (7). The second model estimates the corresponding di⁄erence equation (8). The benchmark cross-section and di⁄erence models are estimated for employed respondents only. Furthermore, the models focus on the hours margin. Labor economists have recognized the importance of participation margin in labor supply. Focusing on the hour(cid:146)s margin allows extrapolation of the results to workers of all ages, for whom the hours choice is relevant but for whom the participation decision faced by respondents in this sample, namely retirement, is not relevant. III.2.1 Regression Analysis: Cross-section To understand how the factors in(cid:135)uence labor supply after controlling for all other covariates, the (cid:133)rst model regresses logarithmic labor hours on the vector of factors scores, real wage, some demographics, and pecuniary work variables.13 All the data are from the (cid:133)rst wave (1992) of the HRS. (9) ln(Hours i ) = (cid:11) 0 +(cid:23)0ln(wage i )+(cid:12)0X i +(cid:15) i 9

The results from the O.L.S. estimation are presented in Table VIII. The coe¢ cient of PHYSICAL factor variable is negative and insigni(cid:133)cant, whereas the MENTAL factor has a positive, signi(cid:133)cant coe¢ cient. Workers in jobs that are mentally intensive tend to work more hours compared to workers in physical jobs. The results imply that a unit improvement in the MENTAL factor accounts for approximately 3.4 percent of growth in annual labor hours. III.2.2 Regression Analysis: Di⁄erence The regression in this section uses data available in the (cid:133)rst and second waves of the HRS. The data in both waves are identical except the second wave(cid:146)s data were collected two years after those of the (cid:133)rst. The second wave contains questions on job characteristics for respondents who switched jobs between the two waves. The factor scores for these respondents are constructed as described above. No job characteristics data were collected for respondents who remained in the same job. For these respondents, the factor scores estimated in the (cid:133)rst wave are used. In this estimationasinthecross-section, consumptionisomitted. Intheend, wehavedataonlaborhours, factor scores, time-varying demographics, and pecuniary job characteristics for two waves of data collectedapproximatelytwoyearsapartforthesamerespondents. Thesedataallowsustotesthow changes in hours between 1992 and 1994 responded to changes in the factor scores, by estimating the following econometric model. (10) (cid:1)ln(Hours i;t ) = (cid:11) 0 +(cid:23)0(cid:1)ln(wage i;t )+(cid:12)0(cid:1)X i;t +(cid:1)(cid:15) i;t where (cid:1) denotes the di⁄erence between the value of variable in 1994 and its value in 1992, (cid:1)x = i;t x x . From the regression results in Table IX, we note again that the coe¢ cient on the i;1994 i;1992 (cid:0) change in the MENTAL factor is positive and statistically signi(cid:133)cant, whereas for the PHYSICAL factor, the coe¢ cient is negative and insigni(cid:133)cant. The results from this model imply that a oneunit increase in the MENTAL factor accounts for approximately 5.5 percent of growth in annual labor hours. The results from the two regression models indicate that, of the two main job characteristics, onlytheMENTALfactorin(cid:135)uenceslaborhourspositively. Thecoe¢ cientforthePHYSICALfactor isnotstatisticallydi⁄erentfromzero. Intheremainderoftheanalysis,IretaintheMENTALfactor as the Job Quality measure. 10

III.2.3 Robustness and Sensitivity Analysis This section conducts several sensitivity and robustness tests. The results are reported for the Job Quality coe¢ cient in Tables X and XI. The (cid:133)rst sensitivity tests exclude the insigni(cid:133)cant PHYSICAL factor from both models. The benchmark results are preserved. To better compare theresultsobtainedfrombothmodels,thenextsensitivitiesestimatebothmodelsusinganidentical sample. Reconciling the samples does not signi(cid:133)cantly alter the results. Next, the sample in the Di⁄erence model is restricted to the set of respondents who switched jobs between 1992 and 1994. Excluding respondents who didn(cid:146)t change jobs allows more variation intheJobQualityvariableandprovidesa(cid:133)rmertestofwhetherchangesinJobQualityandchanges in labor hours co-vary. The estimated coe¢ cient for the Job Quality variable is 0.047. In the next set of sensitivity tests, the sample is split between males and females for both models. The split of the sample allows a test of whether the relationship between Job Quality and labor hours di⁄ers by gender. The estimates are identical for men and women. The results obtained from the cross-section regression model can be biased if the sample is incidentally truncated along the Job Quality variable. For example, if poor Job Quality lowers the labor force participation rate, the sample would consist of workers with Job Quality over a certain threshold. The truncation introduces an omitted-variable bias that could drive the observed results if the omitted variable is correlated with the Job Quality measure. The estimation of a two-stage Heckman model where participation is modeled as a function of age, education, health, marital status, net wealth, gender, and occupation in previous jobs (proxy for Job Quality in previous job) addresses the issue. The estimated coe¢ cient for Job Quality is identical to the estimate from the benchmark regression. The occupation dummy variables in the participation model are signi(cid:133)cant. So is the Job Quality variable in the hours equation. These results indicate that Job Quality matters for labor force participation as well as for labor supply. Controlling for the participation e⁄ect does not signi(cid:133)cantly change the results. Marital status may also in(cid:135)uence a respondent(cid:146)s labor supply, especially if there are strong interactions between the respondent(cid:146)s and the spouse(cid:146)s labor supply. To the extent that the unobserved factors due to marital status are correlated with the Job Quality measure, our results will be biased. When the analysis is conducted separately for married and single respondents, the results are preserved. The next sensitivity analysis is motivated by potential biases introduced by the choice of the HRS data. This issue is of concern because in the HRS most respondents were born between 1931 and 1941.14 For these reasons, the average age in the sample (55 year of age) exceeds the national average worker(cid:146)s age. To the extent that the Job Quality e⁄ect is di⁄erent for older and 11

younger workers, the coe¢ cient in our sample will not re(cid:135)ect the coe¢ cient obtained using the entire population data. The regression analysis for various age brackets reveals a higher coe¢ cient for younger respondents. For respondents below 55 years of age, the average coe¢ cient is 0.043 and 0.027 for respondents over 55. One possible explanation for these results is that older workers are more experienced and accustomed to their jobs, and hence, tolerate more than younger works. The di⁄erences in coe¢ cients for age groups are mostly insigni(cid:133)cant. To the extent that di⁄erences in coe¢ cients across age groups matter, they indicate a higher coe¢ cient for younger workers, which indicates that our estimate of the quality e⁄ect would be higher if the age distribution of workers in the HRS sample were identical to the age distribution at the national level. Another possible and related bias comes from sample selection. For example, if older workers tend to join speci(cid:133)c types of occupations and industries, these occupations will be over represented in our sample. Given the close link between occupation and Job Quality, our estimate would be biased. To address this issue, we construct weights such that the distribution of employment in the HRS by industry and occupation cells matches the national distribution. The estimated coe¢ cient increases slightly from 0.034 to 0.037. A common problem in estimation of labor supply models using micro data is the (cid:147)division bias(cid:148). Division bias is present if there are measurement errors in labor hours, and the wage rate is computed as salary divided by the hours. For example, an over-estimation of the hours increases the dependent variable and decreases the independent variable, resulting in an upward bias in the wage coe¢ cient. If the correlation between the wage rate and the Job Quality measure is high, the estimate quality coe¢ cient will be biased. To test whether our results are a⁄ected by the division bias, the sample is restricted to employees paid hourly. In other words, the wage rate for respondents in this sample is no longer computed as the ratio of salary to total hours. These respondents make up approximately 56 percent of the sample. The regression coe¢ cient (0.020) remains positive and statistically signi(cid:133)cant. Furthermore, the correlation between the Job Quality measure and the wage rate is weak so that even in the presence of a division bias, the main results are not likely to be a⁄ected. Insum,thesensitivityanalysisrevealsthattheJobQualityestimateobtainedinthebenchmark model remains positive and statistically signi(cid:133)cant, and that it is robust to various controls and cuts of the data. IV Aggregate Job Quality Index and Quality Controlled Hours To construct the aggregate Job Quality index over time, the HRS data is limited since it is only available from 1992. I use IPMUS decennial data on employment by occupation and industry 12

from 1850 to 2000.15 IPUMS reclassi(cid:133)es occupations and industries based on the census 1950 classi(cid:133)cation scheme. The reclassi(cid:133)cation makes the data comparable across years. I match these industry and occupation groups with those of the HRS. In the end, we obtain thirteen industries and eight occupations groups that are comparable between the IPMUS data and the HRS data (see Appendix A for the industry and occupation classi(cid:133)cation in both datasets). Using the average Job Quality measure constructed from the HRS job characteristics by industry and occupation cells, we can infer the overall quality of jobs in the aggregate economy from the composition of the workforce. For example, if the fraction of workers in the higher-quality industries and occupations increases proportionally more than the fraction of workers in lower-quality industries, the aggregate Job Quality in the economy improves.16 The aggregate Job Quality index would decline if the opposite were true. LetEmp denotethenumberofworkersinoccupationiandindustryj attimet,andEmp the ij;t t total number of workers at time t. Let Q be the improvement in aggregate Job Quality between b;t a base-year b and time t. I J Emp Emp ij;t ij;b (11) Q = q (cid:0) b;t ij Emp t i=1j=1 (cid:18) (cid:19) XX where I is the total number of industries in the economy, and J the total number of occupations. q is the average quality for industry i and occupation j calculated in section III.1. The Job ij Quality variable is multiplied by 0.034, the quality coe¢ cient estimated in Table 2.6a, so that one hundredth of a unit change in the quality measure implies a 1 percent corresponding change in labor hours. For equation (11) reduces to Q = 0. b;b Figure II plots the aggregate Job Quality index for the U.S. from 1850 and 2000. The index increased by approximately 0.204. The results imply that the improvement in Job Quality was responsible for at least 20.4 percent growth in average annual labor hours between 1850 and 2000, which corresponds to approximately 0.14 percent of labor hours growth (less than three hours) per year. Three features of the index are worth noting. From 1850 to 1880, the index did not change. Between 1880 and 1950, it improved almost exponentially, and after 1950, it continued to improve at a slower rate. The increase between 1880 and 1900 re(cid:135)ects the shift from Agriculture toward Manufacturing, Transportation, and Retail trade industries. For occupations, the signi(cid:133)cant shift occurred from Farmers to Machine operatives. Post 1950, the share in employment in Agriculture and Manufacturing declined drastically, and the share of Professional services industry rose. From 1950 on, employment in the Clerical and Professional occupations surged, while employment shares 13

of Farmers and Operative laborers declined. These three phases seem to mark the movement from Agriculture to Manufacturing between 1880 and 1900, and from Manufacturing to Services after 1950. TablesXIIandXIIIbreakdownthegrowthintheJobQualityindexbyindustryandoccupation. TableXIIshowsthegrowthinqualityattributedtoindustryonly. Thisisthegrowthduetoshiftsin industry employment ignoring shifts in occupation employment. Similarly, Table XIII displays the growth in quality due to shifts in occupation employment, ignoring shifts in industry employment. The growth attributed to shifts in industry is 16.1 percent while the growth due to occupation changes is 19.5 percent. The derivation of quality-controlled labor hours follows from the growth in the Job Quality index derived in the previous section. Let 1850 be the base year, quality-controlled labor hours can be derived using the following formula: N t (12) H = 1850;t (1+(Q Q )) t 1850 (cid:0) where H denotes the quality-controlled labor hours, N is actual annual per person observed labor t t hours, Q is the aggregate quality derived in the previous section. Given the improvement in t Q between 1850 and 2000, formula (12) implies that quality-controlled labor hours declined by t approximately 20.4 percent. These results support the Job Quality hypothesis that the average hours would have been working approximately 20.4 percent lower in 2000 had the nature of work remained constant since 1850. V Some Caveats From the observed changes in the nature of work, the study characterizes jobs in the United States and assesses the implications of changes in Job Quality on changes in labor supply. The measures provided are not without limitations. First, the aggregate Job Quality measure assumes a constant ranking of industry and occupation cells across time. Sector-biased technological changes could cause quality in an occupation to exceed the Job Quality in occupations previously ranked higher. The assumption appears, however, reasonable. A correlation of the rankings of industry and occupation cells by education (a good proxy for Job Quality) is 0.90. Second, the derived relationship between Job Quality and work hours assumes that a worker in 1850 is identical to a worker in 2000. If the characteristics of workers change over time, this could in(cid:135)uence our measured quality e⁄ect. Further, and perhaps 14

more importantly, the current Job Quality index fails to account for within-jobs quality improvements. The derived index is entirely based on the reallocation of employment across industries and occupations over time. From the micro-data standpoint, there are reservations about the HRS data. One unique feature of this data set is the over representation of older workers. The mean and median age of the respondents in the data are approximately 56. This average is much higher than the age of the national average worker. To the extent that the relationship between Job Quality and labor hours is sensitive to the age of the respondent, our estimates would be biased. The sensitivity analyses performed in section III.2 indicate some di⁄erences in the estimated Job Quality e⁄ect for various age brackets. The estimate is higher for respondents below 55 years of age. Any potential bias introduced by the over-representation of older workers would suggest that the actual estimate is higher than the one reported. This limitation and the inability to capture within-jobs quality improvements make our estimates of the growth in hours attributed to Job Quality, a lower bound of the true estimate. VI Conclusion This study presents evidence of the heterogeneity in jobs, and analyzes the implications of this heterogeneity for labor supply. The results indicate that, ceteris paribus, some jobs tend to induce more labor hours. These jobs are prevalent in the services sector and in public administration, and in occupations such as management, professional specialty, sales, administrative, and technical support. Workers in these occupations tend to enjoy higher wages and more generous employer provided bene(cid:133)ts packages. The share of employment in these industries and occupations has increased over time, inducing an overall increase in aggregate Job Quality. The increase in labor resulting from the improvement in Job Quality explains, in part, the lack of a strong downward trend in the average work hours in the United States. More precisely, the improvement in Job Quality between 1850 and 2000 accounts for at least 20.4 percent of growth in per capita labor hours. Once we control for changes in quality, we observe a decline in labor hours consistent with a domination of the income e⁄ect over the substitution e⁄ect. One important limitation of the study is the inability to account for within-jobs quality improvements. As such, the measure only provides a lower bound of the true measure. To fully appreciate the extent to which changes in Job Quality impact trends in labor hours, more research along this line is warranted. 15

A Appendix: IPUMS and HRS Industry and Occupation Classi(cid:133)cation IPUMS 1950 Industry Classi(cid:133)cation HRS Industry Classi(cid:133)cation Agriculture, forestry, (cid:133)shing Agriculture, forestry, (cid:133)shing Mining and construction Mining and construction Manufacturing: non-durable Manufacturing: non-durable Manufacturing: durable Manufacturing: durable Transportation Transportation Wholesale Wholesale Retail Retail Finance, insurance, and real estate Finance, insurance, and real estate Business and repair services Business and repair services Personal services Personal services Entertainment and recreation Entertainment and recreation Professional and related services Professional and related services Public administration Public administration IPUMS industry classi(cid:133)cation categories are idential to those in the Health and Retirement Survey. IPUMS 1950 Occupation Classi(cid:133)cation HRS Occupation Classi(cid:133)cation Professional, Technical Prof. Specialty oper. & tech support Managers, O¢ cials, and Proprietors Managerial specialty operation Sales workers Sales Clerical and Kindred Clerical, administrative support Service Workers (private household) Service: priv. hshld, clean, building serv. Service Workers (not private household) Service: protection Service: foodpreparation Health services Personal services Farmers, Farm Laborers Farming, forestry, (cid:133)shing Operatives, Laborers Operators: machine Operators: transport, etc. Operators: handlers, etc. Craftsmen Mechanicsandrepair Constructiontradeand extractors Precision production Member of armed forces 16

References [1] Barsky, Robert, Miles Kimball, Thomas Juster and Matthew Shapiro, (cid:147)Preference Parameters and Behavioral. Heterogeneity: An Experimental Approach in the Health and Retirement Study,(cid:148)Quarterly Journal of Economics, CXII (1997), 537-579. [2] Basu, Susanto and Kimball, Miles, (cid:147)Long-Run Labor Supply and the Elasticity of Intertemporal Substitution for Consumption,(cid:148)Working Paper, (2002), University of Michigan. [3] Benhabib, Jess, Richard Rogerson and Randall Wright, (cid:147)Homework in Macroeconomics: Household Production and Aggregate Fluctuations,(cid:148) Journal of Political Economy, XCIX (1991), 1166-1187. [4] Campbell, John Y., and Sydney Ludvigson, (cid:147)Elasticities of Substitution in Real Business Cycle Models with Home Production,(cid:148) electronic (cid:133)le available at www.ny.frb.org- /rmaghome/economist/ludvigson/ludvigson.html, (2000). [5] Costa, DoraL., (cid:147)The Wage and the Length of the WorkDay: From the 1890s to 1991,(cid:148)NBER Working Paper 6504 (1998). [6] Gardner, Jennifer M., Philip L.Rones, and Randy E. IIg, (cid:147)Trends in hours of work since the mid-1970s,(cid:148)BLS Monthly Labor Review, CXX (1997). [7] Greenwood,JeremyandZviHercowitz,(cid:147)TheAllocationofCapitalandTimeovertheBusiness Cycle,(cid:148)Journal of Political Economy, XCIX (1991), 1188-1214. [8] Greenwood, Jeremy, Richard Rogerson and Randall Wright, (cid:147)Household Production in Real Business Cycle Theory," in Thomas F. Cooley, eds., Frontiers of Real Business Cycle Theory, (Princeton, NY: Princeton University Press, 1995). [9] Hall, Robert E., (cid:147)Intertemporal Substitution in Consumption,(cid:148)Journal of Political Economy, XCVI (1988), 339-357. [10] (cid:151)(cid:151)(cid:151), (cid:147)Stochastic Implications of the Life Cycle-Permanent Income Hypothesis: Theory and Evidence,(cid:148)Journal of Political Economy, LXXXVI (1978), 971-987. [11] Hamermesh, D., , (cid:147)Consumption During Retirement: The Missing Link in the Life Cycle,(cid:148) Review of Economics and Statistics, LXVI (1984), 1-7. [12] Hansen, G.D., and R. Wright, (cid:147)The Labor Market in Real Business Cycle Theory,(cid:148)Federal Reserve Bank of Minneapolis Quarterly Review, Spring (1992), 1-12. [13] Heckman, James and Mark R. Killingsworth, (cid:147)Female Labor Supply: A Survey,(cid:148)In Orley C. Ashenfelter, and Richard Layard, eds., Handbook of Labor Economics, Volume 1 (Amsterdam: North Holland, 1986). [14] Hurd, Michael and Kathleen McGarry, (cid:147)The relationship between job characteristics and Retirement,(cid:148)NBER Working paper 4558 (1993). 17

[15] Juhn, Chinhui and Kevin M. Murphy, (cid:147)Wage Inequality and Family Labor Supply,(cid:148)NBER Working paper MMMMMCDLIX (1996). [16] King, Robert G., Charles I. Plosser, and Sergio T. Rebelo, (cid:147)Production, Growth and Business Cycles: I. The Basic Neoclassical Model,(cid:148)Journal of Monetary economics, XXI (1988),195- 232. [17] (cid:151)(cid:151), "Production, Growth and Business Cycles: II. New Directions,(cid:148)Journal of Monetary economics, XXI (1988), 309-341. [18] Keynes, John Maynard, (cid:147)Economic Possibilities for our Grandchildren,(cid:148) Nation and Athenaeum, October 11 and 18, 1930. (Reprinted in The Collected Writings of John Maynard Keynes, Volume IX, Essays in Persuasion, London, 1972). [19] Mariger, R., (cid:147)A Life-Cycle Consumption Model with Liquidity Constraints: Theory and Empirical Results(cid:148), Econometrica, LV ( 1987), 533-557. [20] McDonald, Roderick P., (cid:147)Factor Analysis and Related Methods,(cid:148)(Hillsdale, NJ: Lawrence Erlbaum Associates, 1985). [21] Nakamura, Alice and Susan, Houseman, (cid:147)Working Time in Comparative Perspective,(cid:148)(Kalamazoo, Michigan: W.E. Upjohn Institute for Employment Research Volume II, 2001). [22] Pencavel, John, (cid:147)Labor Supply of Men: A Survey,(cid:148)In Orley C. Ashenfelter, and Richard Layard, eds., Handbook of Labor Economics, Volume 1 (Amsterdam: North Holland,1986). [23] Rebelo, Sergio and Danyang Xie, (cid:147)Beyond Balanced Growth,(cid:148)NBER Working Paper 6159, (1997). [24] Rummel, R. J., (cid:147)Applied Factor Analysis,(cid:148)(Evanston, IL: Northwestern University Press, 1970). [25] Smith, James P. and Michael P. Ward, (cid:147)Trends in Women(cid:146)s work, Education, and Family Building,(cid:148)Journal of Labor Economics, III (1985), S59-S90. [26] Stock,JamesH.,MarkW.Watson,(cid:147)BusinessCycleFluctuationsinU.S.MacroeconomicTime Series,(cid:148)NBER working paper 6528 (1998). [27] Vissing-Jorgenson, (cid:147) Limited Asset Market Participation and the Elasticity of Intertemporal Substitution,(cid:148)NBER Working paper 8896, (2002). [28] Viscusi W, Kip, (cid:147)The Value of Risks to Life and Health,(cid:148)Journal of Economic Literature, Volume XXXI (1993 ), 1912-1946. 18

Notes 1Infact,Keynespredictedthattheallocationoftheextraleisuretimewillbeman(cid:146)srealandpermanentproblem. Noting that man needs to do some work if he needs to be contented, Keynes envisioned a widely shared shorter workweek resulting in three-hours shifts or a (cid:133)fteen-hour workweek. 2Another explanation for the lack of trends in labor hours is implicit the King-Plosser-Rebelo (KPR) utility function. Labor and consumption are complements, so that an increase in consumption makes work more pleasant and discourages workers from reducing hours. For some parameter restrictions on the utility function, the increase in consumption increases the marginal utility for work enough to justify a lack of trends in hours. 3See Greenwood, Rogerson and Wright [1995] for a survey. 4The Health and Retirement Study samples individuals born between the years 1931 and 1941 living in the United States. Respondents also include spouses ofage eligible respondents even ifthese spouses are not age eligible themselves. It is a nationally representative sample. The exceptions are blacks and Florida residents that are over sampled. 5ThechangesinthesectoralcompositionofemploymentandthenatureofoccupationsintheUSareconsistentwith a well-known international stylized fact, namely, that income levels tend to be correlated with the share of services in total employment. In low-income countries, agriculture tends to be the predominant employment sector. For middle-income countries, employment in manufacturing dominates, and for high-income countries, services makeup the biggest share of employment. The composition of employment by sector tends to change systematically with the income level. Most economies start with agriculture as the predominant employment sector, as the income level expands, employments shifts to manufacturing, and to the services sector. 6See Costa [1998] for more evidence on heterogeneity of hours between occupations and over time. 7Notethatthisde(cid:133)nitionisdi⁄erentfromthede(cid:133)nitionsinsomestudiesonjobsbytheCenterforNationalPolicy, the Bureau of Labor Statistics, and International Labor Organization. According the their de(cid:133)nitions, wages, fringe bene(cid:133)ts, and health bene(cid:133)ts de(cid:133)ne Job Quality. The de(cid:133)nition does not deal with the nature of work. For example, theAverageJobQualityindexconstructedbyHarvardeconomistMedo⁄andusedbytheBureauofLaborStatisticsis entirelyconstructedbasedoncompensationdi⁄erencesacrossjobs. TheInternationalLaborOrganizationinGeneva de(cid:133)nes Job Quality based on Remuneration levels, Job Security, Social Protection, Safety and Health concerns, Human resource development, Management and Organization, Freely chosen employment. The de(cid:133)nition of Job Quality in this paper focuses on the nature of the work itself and factors that in(cid:135)uence the comfort or pleasantness of the working process. 8The changes in nature of work was well predicted by Keynes when he said (cid:147)...in quite a few years-in our own lifetimes I mean- we may be able to perform all the operations of agriculture, mining, and manufacturing with a quarter of the human e⁄ort to which we have been accustomed(cid:148)Keynes [1930]. 9Basedontheexamplesprovided,itappearsthatJobQualityandtechnologycapturethesamenotion. Although, there could be a link between technological progress and improvements in Job Quality, the two concepts are not identical. Technological progress need not lead to improvements in work quality. An improvement in the miles-pergallon of trucks due to technological progress does not improve the job quality for truck drivers. 10The compensating di⁄erential literature recognizes the dependence of the wage rate on the job characteristics. In the macroeconomic context, the average job quality change is considered independent of the wage. In subsequent microanalysis, both wages and job quality will be included in the regression analysis. Given a weak correlation between the quality measure and the wage rate, the quality e⁄ect through the wage rate appears negligible. 11Words in parentheses are not part of the survey. They are borrowed from Hurd and McGarry [1993] to capture the main job characteristic. 12Common factor analysis is one method among others to perform factor analysis. Other techniques include Principal Components, Iterated Principal Factor, and Maximum likelihood. All these methods yield similar results. 13Ideally, the list of regressors would include consumption. The inclusion of consumption is however, problematic becauseitisendogeneous. Furthermore,thedatasetdoesnotcontaingoodmeasuresofhouseholdconsumption. The omissionofconsumption,couldpotentiallyintroducesabiasintheestimation. Giventheassumptionofindependence between Job quality and consumption by construction, I expect bias on the coe¢ cient of interest, namely the Job Quality coe¢ cient, to be small. 14The possible exception is the spouses of the respondents. Spouses were included in the study so long as their mates where age eligible (born between 1931 and 1941). 15IPUMS (Integrated Public Use Microdata Series) consists of twenty-(cid:133)ve high-precision samples of the American population drawn from thirteen federal censuses. This study uses the 1% random sample drawn from the census data. 16The measure of aggregate quality does not take into account the within job improvement. The quality improvement is obtained from shifts in the industry and occupation mix. The resulting measure is therefore a lower bound of the true improvement in Job Quality. Furthermore, the methodology assumes a constant ranking of industryoccupation cells for Job Quality across time. 19

Table I: Share of Employment by Industry in 1850, 1920, and 2000 (percent) Industry 1850 1920 2000 Agriculture, forestry, (cid:133)shing 58.0 26.2 2.9 Mining and construction 8.1 7.8 7.4 Manufacturing: non-durable 5.6 12.9 8.6 Manufacturing: durable 6.7 11.7 6.3 Transportation 2.9 9.2 4.9 Wholesale 0.1 2.1 3.3 Retail 8.1 9.9 17.9 Finance, insurance, and real estate 0.3 2.1 5.8 Business and repair services 2.3 2.8 6.4 Personal services 4.2 7.5 3.2 Entertainment and recreation 0.1 0.6 2.6 Professional and related services 3.0 4.8 24.8 Public administration 0.6 2.5 5.9 Source: Author(cid:146)scalculationsfrom1percentSampleoftheCensusBureaucollectedintheIntegrated PublicUseMicrodataSeriesbytheMinnesotaPopulationCenter. (http://www.ipums.org). Industry in all years is rede(cid:133)ned by IPUMS to be consistent with the 1950 classi(cid:133)cation. Table II: Share of Employment by Occupation in 1850, 1920, and 2000 (percent) Industry 1850 1920 2000 Professional 3.3 5.5 21.1 Managers, o¢ cials, Proprietors 5.1 6.5 10.3 Farmers / Farm laborers 57.5 25.8 1.7 Service 1.5 8.1 14.9 Clerical 0.2 8.1 18.8 Sales 2.5 4.9 6.5 Craftsmen 18.7 14.4 11.0 Operative /Non farm Laborers 11.2 26.7 15.9 Source: Author(cid:146)scalculationsfrom1percentSampleoftheCensusBureaucollectedintheIntegrated Public Use Microdata Series by the Minnesota Population Center. (http://www.ipums.org). Occupation in all years is rede(cid:133)ned by IPUMS to be consistent with the 1950 classi(cid:133)cation. 20

selbairaV scitsiretcarahC boJ yrainucepnoN fo xirtaM noitalerroC :III elbaT ssertS .meM .c ¢iD .niarT.krwoC .deerF nraeL .tepeR ecaP ataD .pmoC elpoeP.tnettA .eyE .dneB gnitfiL .syhP)310;8 = sbO( 00.1 lacisyhP 00.1 56.0 gnitfiL 00.1 46.0 06.0 gnidneB 00.1 50.0 20.0 60.0 thgiseyE 00.1 04.0 00.0- 00.0 40.0 noitnettA 00.1 13.0 81.0 30.0- 60.0- 40.0elpoeP 00.1 51.0 81.0 61.0 52.0- 82.0- 53.0sretupmoC 00.1 55.0 42.0 72.0 41.0 22.0- 32.0- 03.0ataD 00.1 72.0 22.0 12.0 62.0 41.0 20.0- 10.0 20.0 ecaP 00.1 71.0 41.0- 80.0- 10.0 01.0 11.0 81.0 61.0 22.0 noititepeR 00.1 60.0- 82.0 93.0 62.0 23.0 03.0 81.0 30.0- 50.0- 40.0nraeL 00.1 51.0 31.0- 21.0- 71.0 40.0 11.0 40.0 10.0 60.0- 01.0- 11.0modeerF 00.1 32.0 80.0 40.0 10.0- 10.0- 10.0- 11.0 50.0 80.0 10.0- 40.0- 20.0srekrowoC 00.1 50.0- 70.0- 12.0 60.0- 01.0 90.0 90.0 60.0 70.0 30.0 20.0 10.0 10.0 gniniarT 00.1 63.0 90.0- 10.0- 13.0 60.0- 71.0 62.0 22.0 21.0 81.0 90.0 10.0 10.0 10.0tluc ¢iD 00.1 62.0 41.0 40.0 11.0 33.0 10.0- 02.0 03.0 02.0 03.0 23.0 22.0 50.0- 50.0- 60.0yromeM 00.1 92.0 63.0 51.0 61.0- 30.0- 62.0 40.0 42.0 52.0 71.0 32.0 92.0 01.0 20.0- 10.0 20.0 ssertS ydutS tnemeriteR dna htlaeH eht fo I evaw morf snoitaluclac s(cid:146)rohtuA :ecruoS 21

ataD scitsiretcarahC boJ yrainucepnoN gnisu stluseR sisylanA rotcaF nommoC :VI elbaT srotcaF noitpircseD .euqinU 7 6 5 4 3 2 1 elbairaV .tro⁄e lacisyhp fo stol seriuqer boj yM 93.0 20.0- 00.0- 20.0 30.0 20.0 66.0 14.0lacisyhP .sdaol yvaeh gnitfil seriuqer boj yM 83.0 00.0- 40.0- 51.0 21.0 20.0- 56.0 04.0gnitfiL .gnihcuorc ro ,gnileenk ,gnipoots seriuqer boj yM 44.0 30.0 00.0- 31.0 41.0 30.0 26.0 73.0gnidneB .thgiseye doog seriuqer boj yM 27.0 12.0 00.0- 10.0- 81.0- 42.0 62.0 82.0 thgiseyE noitnetta ro noitartnecnoc esnetni seriuqer boj yM 95.0 01.0 70.0- 80.0- 81.0- 12.0 33.0 54.0 noitnettA elpoep rehto htiw gnilaed ni lliks seriuqer boj yM 27.0 11.0- 40.0- 21.0- 10.0 12.0 02.0 14.0 elpoeP .sretupmoc htiw krow ot em seriuqer boj yM 55.0 70.0 30.0 03.0 40.0- 60.0- 21.0- 85.0 sretupmoC .noitamrofni ro atad ezylana ot em seriuqer boj yM 54.0 20.0- 40.0- 82.0 01.0 00.0- 30.0- 86.0 ataD srehto yb tes ecap eht htiw pu peek ot em seriuqer boj yM 07.0 61.0- 21.0 80.0 42.0- 40.0- 62.0 73.0 ecaP .revo dna revo sgniht emas eht od ot em seriuqer boj yM 08.0 50.0- 11.0 10.0- 03.0- 80.0 72.0 11.0noititepeR .sgniht wen nrael I taht seriuqer boj yM 95.0 70.0- 90.0 20.0- 81.0 50.0 52.0 55.0 nraeL krow nwo ym od I woh ediced ot modeerf fo tol a evah I 67.0 20.0- 40.0- 20.0- 33.0 13.0 01.0- 51.0 modeerF .yldneirf dna lufpleh era htiw krow I elpoep ehT 08.0 10.0- 71.0 40.0- 31.0 93.0 30.0- 30.0 srekrowoC slliks boj ym etadpu ot gniniart deviecer I fi retteb tol a boj od dluoC 67.0 80.0 71.0 41.0- 31.0 03.0- 71.0 22.0 gniniarT .ot desu ti naht sgniht tluc ¢id erom od ot em seriuqer boj yM 16.0 70.0 50.0 90.0- 61.0 43.0- 52.0 14.0 tluc ¢iD .yromem doog yrev a seriuqer boj yM 96.0 10.0- 70.0- 31.0- 60.0 80.0 32.0 84.0 yromeM .sserts fo tol a sevlovni boj yM 66.0 80.0- 61.0- 11.0- 60.0- 22.0- 92.0 14.0 ssertS ecnairav tcP 3 3 6 01 41 93 75 yb denialpxe rotcaf tsriF .stluser sisylana rotcaf eht syalpsid elbat dnoces ehT .eulav-negie gnidnopserroc sti dna rotcaf hcae syalpsid elbat tsr(cid:133) ehT .71 fo tes elbissop eht fo tuo deniater srotcaf eht yalpsid 8-2 snmuloC .noisserger eht ni desu selbairav eht sniatnoc nmuloc eht ni ytilibairav eht ,si taht ,elbairav eht fo sseneuqinu eht serutpac 9 nmuloC .deniater era seulav-negie evitisop htiw srotcaf ylnO retaerg sgnidaol rotcaF .1 nmuloc ni selbairav eht fo noitpircsed liated sevig 01 nmuloC nialpxe nac rotcaf nommoc on taht elbairav ydutS tnemeriteR dna htlaeH eht fo I evaW morf snoitaluclac s(cid:146)rohtuA :ecruoS dlob ni 3.0 ot lauqe fo naht 22

Table V: Mean of "PHYSICAL" and "MENTAL" Factors by Industry Industry MENTAL PHYSICAL Nobs Agriculture, forestry, (cid:133)shing -0.71 0.37 278 Mining and construction -0.51 0.45 515 Manufacturing: non-durable -0.16 0.04 614 Manufacturing: durable -0.07 0.03 906 Transportation 0.01 0.09 574 Wholesale 0.06 -0.02 324 Retail -0.26 0.16 935 Finance, insurance, and real estate 0.56 -0.35 533 Business and repair services -0.04 -0.07 495 Personal services -0.72 -0.05 412 Entertainment and recreation -0.35 0.05 125 Professional and related services 0.20 0.05 2,255 Public administration 0.60 -0.21 409 Source: Author(cid:146)s calculations from wave I of the Health and Retirement Study. A positive high average for the MENTAL factor indicates a mentally intensive industry. A negative high average indicates a low mentally intensive industry. The same is true for the PHYSICAL factor. 23

Table VI: Mean of "PHYSICAL" and "MENTAL" Factors by Occupation Occupation MENTAL PHYSICAL Nobs Managerial specialty operation 0.55 -0.25 1,247 Prof. Specialty oper. & tech support 0.51 -0.06 1,316 Sales 0.10 -0.08 855 Clerical, administrative support 0.42 -0.23 1,344 Service: priv. hshld, clean, building serv. -1.38 -0.33 130 Service: protection -0.10 -0.20 142 Service: food preparation -0.71 0.50 253 Health services -0.43 0.64 197 Personal services -0.82 0.15 570 Farming, forestry, (cid:133)shing -0.84 0.42 265 Mechanics and repair -0.30 0.43 306 Construction trade and extractors -0.63 0.72 289 Precision production -0.22 0.16 285 Operators: machine -0.54 0.25 566 Operators: transport, etc. -0.50 0.31 424 Operators: handlers, etc. -0.93 0.49 212 Member of armed forces 0.97 -0.42 7 Source: Author(cid:146)s calculations from wave I of the Health and Retirement Study. A positive high average for the MENTAL factor indicates a mentally intensive industry. A negative high average indicates a low mentally intensive industry. The same is true for the PHYSICAL factor. 24

Table VII: Correlation of Mental and Physical Factors with Job Characteristics and other demographics (OBS = 2,822) Occupation Mental Factor Physical Factor Salary 0.19 -0.07 Education 0.44 -0.19 Marital Status 0.03 -0.03 Net Wealth 0.13 -0.12 Male 0.13 -0.02 White 0.10 -0.10 Num. Paid Vacation Weeks 0.18 -0.06 Num. Paid Sick Days 0.21 -0.07 Long-term Disability 0.13 -0.02 Can Decrease Hours -0.00 -0.03 Can Increase Hours 0.14 -0.02 Employer Retirement Plan 0.23 -0.02 Job Security 0.04 -0.03 Union Membership -0.10 0.14 Make Pay/Promotion Decision 0.28 -0.12 Better Health 0.18 -0.07 Overtime Weeks 0.03 -0.04 Number of injuries -0.03 0.11 Source: Author(cid:146)s calculations from wave I of the Health and Retirement Study. 25

selbairav cihpargomed dna ,citsiretcarahC boJ ,etar egaW goL no sruoH robaL launnA goL fo noissergeR .S.L.O :IIIV elbaT eulaV-P rorrE .tS tneic ¢eoC elbairaV 000.0 400.0 430.0 latneM 062.0 300.0 400.0lacisyhP 130.0 800.0 710.0etar egaw goL 419.0 80-e80.3 90-e23.3 emocni robal noN 086.0 90-e51.9 90-e77.3 htlaeW teN 000.0 700.0 221.0 elaM 690.0 700.0 110.0 etihW 600.0 000.0 100.0 boJ no sraeY 600.0 700.0 020.0sutatS latiraM 000.0 100.0 200.0egA 100.0 100.0 400.0noitacudE 073.0 300.0 200.0ezis dlohesuoH 253.0 300.0 300.0 sutatS htlaeH 810.0 100.0 200.0 58 ot ytilibaborp lavivruS 000.0 600.0 850.0pihsrebmeM noinU 831.0 800.0 110.0 nalP tnemeriteR reyolpmE 000.0 600.0 420.0 sruoH esaercnI naC 851.0 700.0 010.0 sruoH esaerceD naC 000.0 200.0 810.0 skeeW noitacaV diaP .muN 000.0 600.0 030.0 ytilibasiD mret-gnoL 100.0 900.0 820.0 )seirujnI .mun( ytefaS boJ 000.0 040.0 417.7 tnatsnoC 730,5 :noitavresbo fo rebmuN 71.0 :derauqs-R s(cid:146)eeyolpme eht si(cid:148)etar egaw(cid:147) .ydutS tnemeriteR dna htlaeH eht fo I evaw ni stnednopser yb detroper dekrow sruoh robal launna fo gol larutan eht si ni elbairav tnedneped ehT s(cid:146)tnednopserehtsiegA .bojnahtrehtosecruos morfemocniehtsi(cid:148)emocnirobalnon(cid:147) .emocnirobalehtybdedividyralaslatotehtsiti,stnednopserdeiralasroF .etaregawylruoh .S.U ni htrow ten(cid:146)sdlohesuoh si htrow teN .rooP=1 ,tnellecxE=5.elacs 5-1 a no sutats htlaeh detroper-fles eht si htlaeH .noitacude fo sraey fo rebmun eht si noitacudE(cid:147) ,ega (cid:148)0(cid:147) dna deirram si tnednopser fi(cid:148)1(cid:147) si(cid:148)sutatS latiraM(cid:147) .boj tnerruc eht ta sraey fo rebmun eht si(cid:148)boJ no sraeY(cid:147) .etihW si tnednopser eht rehtehw serutpac(cid:148)etihW(cid:147) .srallod rotacidni na si(cid:148)esaercnI/esaerceD naC(cid:147) .cte snalp derrefed-xat ,noisnep ,k104 hcus nalp tnemeriter fo epyt emos sedivorp reyolpme fi(cid:148)1(cid:147) si(cid:148)nalP tnemeriteR reyolpmE(cid:147) .ton fi sedivorp reyolpme fi(cid:148)1(cid:147) si(cid:148)ytilibasiD mret-gnoL(cid:147) .esiwrehto(cid:148)0(cid:147) dna sruoh esaercni/esaerced nac tnednopser fi(cid:148)1(cid:147) .boj eht no sruoh esaercni/esaerced ot eeyolpme fo ytiliba rof (cid:148)1(cid:147) si pihsrebmeM noinU .weivretni eht gnidecerp raey eht ni boj eht no tnednopser ot seirujni fo rebmun eht si(cid:148)ytefaS boJ(cid:147) .ton seod eh fi(cid:148)0(cid:147) dna nalp ytilibasid mret-gnol a ton seod tnednopser fi(cid:148)0(cid:147) si 58 ot ytilibaborp lavivruS .tnednopser sediseb srebmem dlohesuoh fo rebmun eht si ezis dlohesuoH .ton fi(cid:148)0(cid:147) dna noinu a ot sgnoleb tnednopser fi .niatrec yletulosbafi(cid:148)01(cid:147) dna,egafo 58 ot evilll(cid:146)ehs/eh eveileb ydutS tnemeriteR dna htlaeH eht morf snoitaluclac s(cid:146)rohtuA :ecruoS 26

,etar egaW ni egnahC goL no 4991 dna 2991 neewteb sruoH robaL launnA ni egnahC goL fo noissergeR .S.L.O :XI elbaT selbairav cihpargomed rehto ni egnahC dna ,citsiretcarahC boJ ni egnahC eulaV-P rorrE .tS tneic ¢eoC elbairaV 170.0 030.0 550.0 latneM ni egnahC 971.0 020.0 720.0lacisyhP ni egnahC 800.0 040.0 701.0etar egaw egnahc goL 489.0 300.0 000.0 htlaeW ten ni egnahC 281.0 000.0 000.0 emocni robal non ni egnahC 180.0 710.0 920.0- )seirujni fo rebmun( ytefas ni egnahC 662.0 200.0 200.0ezis dlohesuoh ni egnahC 700.0 100.0 300.0 skeew noitacav diap ni egnahC 968.0 320.0 400.0 sutats latiram ni egnahC 987.0 500.0 100.0 tnemevorpmI htlaeH 590.0 900.0 610.0 sruoh esaerced ot ytiliba ni egnahC 002.0 600.0 700.0 sruoh esaercni ot ytiliba ni egnahC 700.0 900.0 620.0pihsrebmem noinu ni egnahC 626.0 900.0 500.0pihsrebmem ytilibasid mret-gnol ni egnahC 798.0 700.0 100.0 pihsrebmem nalp noisnep ni egnahC 233.0 010.0 900.0tnatsnoC 910,3 :noitavresbo fo rebmuN 40.0 :derauqs-R seulavehtsunim4991nielbairavehtfoseulavehtetonedselbairavllanisegnahC .4991dna2991neewtebdekrowsruohroballaunnafogolniegnahcehtsielbairavtnednepedehT .2991 ni ydutS tnemeriteR dna htlaeH eht morf snoitaluclac s(cid:146)rohtuA :ecruoS 27

Table X: Sensitivity Analysis of E⁄ect of Job Quality on Log Labor Hours Model Sensitivity Sample size Job Quality Coe¢ cient Cross-section Benchmark 5,037 0.034** (0.004) Di⁄erence Benchmark 3,019 0.055* (0.030) Cross-section Exclude insigni(cid:133)cant PHYSICAL factor from the model 5,037 0.034** (0.004) Di⁄erence Exclude insigni(cid:133)cant PHYSICAL factor from the model 3,019 0.0601* (0.030) Cross-section Reconcilesamplestousesamerespondentsinbothmodels 2,880 0.030** (0.004) Di⁄erence Reconcilesamplestousesamerespondentsinbothmodels 2,880 0.059* (0.031) Di⁄erence Restrictsampletoresp. whoswitched jobbetweenwave 1 574 0.047* & 2 (0.031) Cross-section Female 2,630 0.034** (0.005) Cross-section Male 2,407 0.034** (0.005) Cross-section Control for Labor force participation in a two-stage Heck- All:8,402 0.034** man model. Participation is modeled as a function of cens:3,371 (0.004) previous occupation (proxy for quality) and other demo- unc.:5,031 graphicssuchasage,education,health,Maritalstatus,net wealth, and Gender. Cross-section For married workers 4,008 0.036** (0.004) Cross-section For single workers 1,029 0.028** (0.008) Cross-section Hourly wage workers only 2,784 0.020** (0.004) Source: Author(cid:146)s calculations from wave Iand IIofthe Health and Retirement Study. *Indicates signi(cid:133)cance at 5 percent level **Indicates signi(cid:133)cance at the 1% level 28

Table XI: Sensitivity Analysis of E⁄ect of Job Quality on Log Labor Hours Model Sensitivity Sample size Job Quality Coe¢ cient Cross-section Respondents with non-labor income to labor income ratio 3,334 0.037** below 0.10 (0.005) Cross-section brackets: 1,230 0.047** (0.007) Cross-section Restrict sample to respondents with age below 52 years 1,313 0.038** (0.008) Cross-section Restrict sample to workers with age between 53 and 55 1,100 0.021** (0.007) Cross-section Restrict sample to workers with age between 56 and 58 921 0.030** (0.008) Cross-section Restrict sample to workers with age between 59 and 61 473 0.028** (0.012) Cross-section Restrict sample to workers with age between 62 and over 15,128 0.037** (0.002) Cross-section Inclusion of frequency weigths in the benchmark model. 2,063 0.020** Weightsareassignedtomatchthedistributionofindustry (0.006) and occupation frequencies in HRS to frequencies in the U.S. 1990 Census. Cross-section Occupations with non Mental Jobs 2,974 0.050** (0.005) Cross-section Occupation with Mental Jobs 2,130 0.030** (0.005) Cross-section Industries with non Mental Jobs 2,907 0.050** (0.005) Cross-section Industries with Mental Jobs 2,907 0.050** (0.005) Source: Author(cid:146)s calculations from wave Iand IIofthe Health and Retirement Study. *Indicates signi(cid:133)cance at 5% level **Indicates signi(cid:133)cance at the 1% level 29

)tnecreP( 0002 dna 0581 neewteb ytilauQ boJ ni stnemevorpmI ot eud yrtsudnI yb sruoH robaL ni htworG :IIX elbaT 0002 0991 0891 0791 0691 0591 0491 0291 0191 0091 0881 0781 0681 0581 yrtsudnI 1.0 7.0 8.0 5.0 1.0 1.0 1.0- 1.0 0.0 1.0- 2.0- 0.0 1.0- 0.0 gnihsiF/tseroF/..ggA 5.4 2.6 6.5 4.4 8.3 2.2 1.0 8.0 6.0 3.0 1.0 4.0- 7.0- 0.0 noitcurtsnoC/gniniM 6.8 4.8 6.6 9.5 6.5 9.2 7.2 7.1 1.0 2.0- 4.1- 9.0- 5.0- 0.0 gnirutcafunaM elbaruD 1.5 6.4 9.2 9.1 3.1 3.0 3.0- 1.1- 6.1- 0.2- 0.2- 8.1- 7.0- 0.0 gnirutcafunaM elbarudnoN 4.01 8.01 4.01 6.9 2.9 9.6 8.6 1.6 4.4 3.3 8.2 0.1- 2.1- 0.0 noitatropsnarT 6.8- 0.8- 2.9- 1.01- 6.8- 0.01- 6.7- 7.3- 0.2- 1.0- 2.6 1.5 3.7 0.0 elaselohW 2.2 8.0 4.0 4.1- 1.1- 6.1- 6.1- 4.2- 8.1- 0.2- 2.4- 1.2- 0.2- 0.0 liateR 5.2- 1.3- 8.3- 9.4- 1.5- 0.8- 9.9- 6.3- 9.7- 2.5- 7.0 9.0- 3.1- 0.0 etatse laeR/ecnarusnI/niF 8.41 1.41 3.31 5.11 6.9 9.5 3.4 4.8 5.5 3.7 0.6 2.0 5.1 0.0 riapeR ssenisuB 2.5- 3.5- 4.7- 6.8- 5.9- 4.9- 9.01- 5.11- 0.21- 3.21- 9.21- 1.31- 0.31- 0.0 secivreS lanosreP 6.1- 6.0- 3.1- 0.2- 6.1- 5.2- 4.3- 2.2- 8.0- 2.0- 1.0 3.1 8.0 0.0 noitaerceR .tnE 2.9- 9.9- 5.01- 4.01- 2.01- 2.9- 5.7- 3.4- 4.5- 3.1- 5.0- 0.1- 5.0- 0.0 secivreS detaleR .forP 8.3 2.2 7.1 6.0- 0.1- 2.0- 4.2 3.1- 7.0 4.1- 3.3 8.0- 2.0- 0.0 noitartsinimdA .buP 1.61 8.51 8.51 8.41 2.31 4.01 6.8 2.7 8.4 5.3 0.1 2.1 7.1 0.0 latoT yrtsudnI 4.02 6.91 3.81 2.61 2.41 2.01 3.8 6.6 8.3 7.2 2.0 5.0- 3.0 0.0 latoT .atad ydutS tnemeriteR dna htlaeH eht morf detaluclac era sllec yrtsudni dna noitapucco yb atad ytilauQ boJ ehT :ecruoS .SMUPI ni atad tnemyolpme noitapuccO dna yrtsudnI morf detaluclac si ytilauQ ni tnemevorpmi etagerggA stnemevorpmi devresbo eht taht os dex(cid:133) dleh si noitapucco nehw ytilauq ni tnemevorpmi eht si wor(cid:148)latoT yrtsudnI(cid:147) ehT stfihsyrtsudnidnastfihsnoitapuccoytilauqnitnemevorpmillarevoehtsiwor(cid:148)latoT(cid:147)ehT .seirtsudninistfihsybdesuacyleritneera .denibmoc 30

)tnecreP( 0002 dna 0581 neewteb ytilauQ boJ ni stnemevorpmI ot eud noitapuccO yb sruoH robaL ni htworG :IIIX elbaT 0002 0991 0891 0791 0691 0591 0491 0291 0191 0091 0881 0781 0681 0581 noitapuccO 4.0 4.0 6.0 5.0 4.0 1.0- 5.0- 5.0- 9.0- 1.1- 0.1- 3.0- 0.0 0.0 lanoisseforP 8.4 6.4 4.4 6.4 8.3 6.2 8.2 9.2 5.1 6.2 0.4 4.1 2.1 0.0 roteirporP/slaic ¢O/.naM 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 sremraF 0.7 3.6 5.5 2.3 2.2 9.1 9.1- 2.4- 1.6- 6.7- 3.8- 0.9- 3.9- 0.0 ecivreS 6.0 4.0 0.1 0.1 0.1 0.1 7.0 5.0- 7.0- 6.1- 4.1- 0.0 5.1- 0.0 lacirelC 8.01 5.11 2.01 3.7 4.6 4.6 2.7 6.6 2.7 6.4 0.2 7.2 6.0 0.0 selaS 5.0- 1.1 6.1 1.2 2.2 6.1 2.1 3.2 0.1 6.0 1.0 0.0 5.0 0.0 nemstfarC 0.1- 2.1- 3.0- 2.0- 0.0 1.0 9.0- 5.0 2.0 1.0 7.0- 8.0 1.0 0.0 srerobaL /evitarepO 5.91 7.81 3.71 6.51 9.31 2.01 8.8 7.6 3.4 3.3 7.0 2.0- 3.0 0.0 latoT noitapuccO 4.02 6.91 3.81 2.61 2.41 2.01 3.8 6.6 8.3 7.2 2.0 5.0- 9.0 0.0 latoT .atad ydutS tnemeriteR dna htlaeH eht morf detaluclac era sllec yrtsudni dna noitapucco yb atad ytilauQ boJ ehT :ecruoS .SMUPI ni atad tnemyolpme noitapuccO dna yrtsudnI morf detaluclac si ytilauQ ni tnemevorpmi etagerggA yleritneerastnemevorpmidevresboehttahtosdex(cid:133)dlehsiyrtsudninehwytilauqnitnemevorpmiehtsiwor(cid:148)latoTnoitapuccO(cid:147)ehT .snoitapucco ni stfihs yb desuac .denibmoc stfihs yrtsudni dna stfihs noitapucco ytilauq ni tnemevorpmi llarevo eht si wor(cid:148)latoT(cid:147) ehT 31

Figure I: Per Capita Real Consumption, Income, Labor Hours Growth from 1950-2000 3.5 3.0 Real consumption Real Income Labor hours 2.5 2.0 1.5 1.0 0.5 0.0 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 Year Data source: Per capita consumption expenditure,and per capita disposable personalincome are from the Table 8.7 ofthe NationalIncome and Products Account (NIPA) ofthe Bureau ofEconomic Analysis. The originaldata are in chained 1996 U.S.dollars. Percapita laborhours are computed as totalhours offull-time and part-time employmentdivided by the number ofthe population 16 years ofage or over. The hours data are from Table 6.9B ofthe NIPA and he population data are from the Bureau ofLabor Statistics. Allthe series are divided by their respective values in 1950. 32

Figure II: Growth in Labor Hours Due to Improvements in Job Quality 25% 20% 15% 10% 5% 0% 1850 1860 1870 1880 1900 1910 1920 1940 1950 1960 1970 1980 1990 2000 -5% Year Data sources: The industry average Job Quality is obtained from calculations using data in the Health and Retirement Study. The aggregate data on employment is obtained from IPUMS. 33