Chapter 3. Wage Elasticity of Labor Supply: A Survey-Based Experimental Approach *

Transcription

1 Chapter 3 Wage Elasticity of Labor Supply: A Survey-Based Experimental Approach * Fumio Ohtake, Osaka University Shinji Takenaka, Osaka University Kengo Yasui, Osaka University Abstract This research is primarily intended to measure elasticity directly through hypothetical answers to questions eliciting personal responses to wage and income shocks. The analyses clarify the difference in wage elasticity of labor supply between Japanese high-income earners and the whole labor force; in addition, this paper examines the same kind of disparity between Japanese and US workers. The principal findings are as follows. First, little significant difference can be found between the Marshallian and Hicksian elasticities of all workers and of high-income earners in Japan. Second, US workers of all income groups show larger Marshallian and Hicksian elasticities of work hours than do their Japanese counterparts. Both types of elasticities of retirement are, in contrast, estimated to be smaller in the US than in Japan. * The authors are indebted to Kiyoshi Ota, Hisahiro Naito and other participants of Labor Economics Conference in November 2005 in Kanagawa, Japan, for their comments on an earlier version of this paper. Institute of Social and Economic Research (ISER) Graduate School of Economics 28

2 1. Introduction Wage elasticity of labor supply is a critical parameter in the fields of public finance and macroeconomics as well as labor economics, where elasticity plays a pivotal role in explaining labor supply behavior. Public finance is concerned with how income taxation distorts labor supply in the first place. Some argue that progressive taxation inhibits the labor supply of high-income earners, assuming that the wage elasticity of the wealthy group is higher than that of lower income classes. Justification of the argument depends on practically measured Marshallian labor supply elasticity with regard to a permanent and unexpected change of wage. Another issue is welfare loss caused by a change in the income tax rate, discussed in public finance. In other words, a potential increase in Hicksian substitution elasticity, correlated positively with the marginal income tax rate, would result in a greater welfare loss to the high-income group. The implementation of progressive taxation then implies a greater welfare loss to taxpayers in the top bracket. As reviewed comprehensively by Blundell and MaCurdy (1999), a number of economists have contributed to an array of empirical studies measuring labor supply elasticity. Among those analyses, estimates of Marshallian elasticity have ranged from approximately to 0.25 for married men. Some of the estimates for married women are in the same range as those for married men, while others exceeded unity. However, no research has obtained any decisive result to prove that high-income earners show greater wage elasticity of labor supply than others 1. Moffit and Wilhelm s (2000) research is an exceptional study dealing with the difference in elasticity between those two groups. According to their findings, wealthy individuals increased their pretax income after a tax rate reduction, whereas their working hours remained unchanged. As an interpretation of the inelastic labor supply, they suggested that the affluent group, characterized by long work hours, has little room for adjusting the number of hours worked. 1 Showalter and Thurston (1997) researched only the wage elasticity of high-income physicians. From the cross-section data of the high-income, self-employed subjects, the elasticity was estimated to be significant at The estimate was insignificant for those working as employees, apparently having a comparative disadvantage in their ability to change their work hours. 29

3 This research is primarily intended to measure elasticity directly using the answers to hypothetical questions eliciting personal responses to common shocks to wage and income 2. The method has the advantage of securing the exogeneity of economic shocks 3. The following analyses then clarify the difference in wage elasticity of work hours between Japanese high-income earners and the whole labor force. In addition, this paper will examine the same kind of disparity between Japanese and US workers. Furthermore, the elasticity of retirement will be measured in both the Marshallian and the Hicksian sense. The workers can change not only their work hours but also the end of their labor market participation when they face unanticipated shocks. The lifetime labor supply may, in other words, be elastic through flexible decisions to retire. Similar to the viewpoints on work hours, disparities brought about by individual characteristics in the elasticity of retirement will be examined. This paper is organized as follows. In Section 2, we will provide a concise description of our questionnaire surveys. In Section 3, the measurement of each kind of elasticity will be explained. Differences in men s elasticity according to personal attributes will be examined in Section 4. Section 5 concludes the paper. 2. Questionnaire Surveys The analyses in this paper are based on data from three questionnaire surveys: (1) Osaka University 21st Century Center of Excellence Panel Survey [conducted in Japan] (OPSJ); (2) Japanese High-Income Earners Survey (JHIES); and (3) Osaka University 21st Century Center of Excellence Panel Survey [conducted in the USA] (OPSUS). A brief description of each survey follows. 2 Kimball and Shapiro (2003) also measured and analyzed the wealth elasticity of labor supply in the United States, using the responses to hypothetical questions. Referring to the statistics showing equality between the substitution effect and the income effect of a permanent wage rise on labor supply, they estimated each respondent s elasticity. Assuming a certain degree of substitution between leisure and consumption, the average estimates of the Frisch, Marshallian, and Hicksian elasticities were equal to 1.004, 0.793, and 0.327, respectively. 3 A major criticism on the contingent valuation method (CVM) like Diamond and Hausman (1994) cannot apply to this research, at least directly. Their argument against such a survey-based valuation was based on the dubious existence of the preference for unfamiliar public goods. However, the economic analyses on labor supply in general suppose the existence of an individual preference for leisure and consumption because the allocation of time between them should be one of the most familiar economic choices in individual life. 30

4 1. OPSJ is a panel study, begun in February 2004, as part of the Osaka University 21st Century Center of Excellence Program. The survey has been conducted annually since 2004 using a random sample drawn from 6,000 individuals by a placement (self-administered) method. There were 4,224 and 2,987 respondents in 2004 and 2005, respectively. This research will use only the 2005 survey data because the principal questions were included only in the latest questionnaire. The questions yielding our critical information are reproduced in Appendix JHIES was a cross-sectional mail survey conducted by Fumio Ohtake at Osaka University in February It randomly sampled 10,000 high-income taxpayers among those taxed more than 25 million yen (about US$232,040.1) of their income in 2004, recorded in published lists of high-income taxpayers. There were 921 respondents. The analyses here used the same questions as those in the OPSJ. 3. OPSUS was also administered as part of the Osaka University 21st Century Center of Excellence Program. This mail survey was conducted with 12,338 individuals in January and February The number of respondents was 4,979. All the questions were the same as those in the OPSJ. The analyses below are based on samples of male respondents up to 64 years of age and working for pay. Tables 1 and 2 show the characteristics of those samples. Among the figures, those of annual earned income of respondents and their households are calculated from the expected values of each income range designated in the questionnaire instead of specific amounts of earnings. The calculation procedure is described in Appendix 2 4. As the respondents could choose to answer questions regarding their earned income in terms of hourly wage, monthly income, or annual income, the hourly wage and the annual income in Table 1 were calculated as follows. First, the exact number given was used for the respondents who provided their hourly wage. Second, for those who did not give their hourly pay but stated their monthly income, their hourly wages were calculated from their monthly income and work hours for a typical week. Third, if a respondent gave only their annual income, then the hourly wage was obtained from the annual income and weekly work hours. 4 The maximum hours worked for Japanese respondents equal 99 because the space for answering their work hours contained two digits only. An answer of 99 work hours was then assumed to be in the range of 99 hours or more. Assuming the frequency of work hours followed the log-normal distribution, the expected value of the work-hours range was calculated and used for the following analyses. The detailed procedure for calculating the expected value is given in Appendix 2. 31

5 Table 1: Descriptive Statistics 1 Sample size Mean Std. Dev. Min. Max. Work hours Japan Overall (per week) High-income U.S. Overall Hourly wage Japan Overall High-income U.S. Overall Annual individual Japan Overall Income High-income U.S. Overall Annual household Japan Overall Income High-income U.S. Overall Age Japan Overall High-income U.S. Overall Risk tolerance Japan Overall coefficient High-income U.S. Overall Notes: The amounts of wage, individual income and household income are expressed in ten thousand yen. The exchange rate for the above tabulation is yen to the US dollar. That is the average of the Tokyo market s end-of-month spot rates in 2004, tabulated in a Bank of Japan s web page: Foreign Exchange Rates ( 32

6 Table 2: Descriptive Statistics 2 Countries Japan U.S. Samples Overall High-income Overall Marital status Unmarried 16.1% 1.3% 63.1% Married 80.1% 96.0% 27.8% Divorced/Separated 2.6% 1.6% 8.0% Widowed 1.0% 1.1% 0.8% No answer 0.1% 0.0% 0.3% Educational Junior high 12.0% 1.6% 4.0% attainment High 46.2% 11.1% 41.2% Vocational 0.0% 1.6% 0.0% Two-year college 7.4% 1.3% 10.9% Four-year college 28.9% 66.8% 27.8% Graduate school 2.8% 16.6% 16.2% No answer 2.6% 1.1% 0.0% Types of Employee 70.0% 4.7% 73.0% employment Businessman/Director 8.1% 57.8% 3.8% Self-employed 12.6% 34.8% 10.9% Family business employee 3.4% 0.5% 2.0% No answer 5.9% 2.1% 10.3% Sample size The calculation of annual income would be, in addition, based on their answers about hourly wage and work hours. The coefficient of relative risk tolerance, included in Table 1, is defined as the inverse of the coefficient of constant relative risk aversion (CRRA). Its estimation procedure is provided in Appendix Measurement of Elasticity Marshallian elasticity represents the percentage change in the worker s optimal labor supply for a one percent rise in his wage. Hicksian elasticity, in contrast, measures the change in the worker s optimal labor supply for a one percent wage rise, when he keeps minimizing the total cost of consumption and leisure to keep a certain level of utility. The latter can be called substitution elasticity and be expressed as the difference between the Marshallian elasticity and the income effect of a wage change. Those two elasticities can be then mathematically expressed as follows, according to the Slutsky equation (Cahuc and Zylberberg, 2004): 33

7 wh η c = η u η R0, (1) R 0 where η u, η c, and η R denote the Marshallian, Hicksian, and income elasticities, respectively. 0 w, h, and R 0 signify the real hourly wage, work hours, and the potential income (i.e., the sum of nonlabor income and the product of real wage and a worker s time endowment), respectively. Hicksian elasticity of labor supply is theoretically expected to be positive, and the income elasticity negative. Their absolute sizes are then supposed to decide the sign of Marshallian elasticity, as is obvious in equation (1). A few hypothetical environments were set for the measurement of η c and η R in the 0 questionnaire. The Marshallian term was, for instance, directly measured by asking how many hours a week each respondent would increase or decrease work hours if his hourly pay were permanently doubled (Question 40). Income elasticity was derived through a question about how many hours per week he would change work hours by, following a lottery win sufficient to pay the previous year s annual household income every year as long as either spouse was alive (Question 39). Employing those two measured values, Hicksian elasticity was then obtained with equation (1). To exclude potential corner solutions, any observation was then excluded from a sample if either of its Marshallian or income elasticity equaled minus one, implying no work after a change. It should also be noted that the following two points are assumed when calculating the potential income R 0. First, the nonlabor income of a respondent equals the part of household income that excludes his own earned income. Second, their time endowment for paid work was fixed at 40 hours a week, the legal working hours, or 80 hours a week by assumption. In addition, the elasticity of retirement was measured in both the Marshallian and the Hicksian senses. The workers could change not only their work hours but also the end of their labor market participation when they experienced an unanticipated wage shock. The lifetime labor supply may be, in other words, elastic through a flexible choice to retire. As for work hours, the same two kinds of elasticity will be examined concerning the projected age of retirement. The preshock projected age of retirement is addressed in the question about the expected age of retirement (Question 38). The Marshallian elasticity of retirement was then measured by asking the expected age of retirement if the respondent s wage was doubled 34

8 permanently (Question 40a). The question identified the change in the respondent s remaining term of labor supply after the hypothetical change. In a similar way as the Marshallian, income elasticity was based on answers to a question about the expected age of retirement after a lottery win sufficient to pay the previous year s annual household income every year for the rest of both spouses lives. As with the counterpart of hours worked, Hicksian elasticity of retirement was also obtained from equation (1), Marshallian and income elasticity 5. The maximum number of work hours per week was fixed at 40 and 80 hours, as in the analysis of work hours. 4. Comparison of Elasticity by Attributes 4.1 Descriptive Statistics This subsection will identify the features of elasticity measured from each sample, Japanese of all income groups (OPSJ), Japanese high-income earners (JHIES), and US workers of all income groups (OPSUS), referring to their mean values and histograms. The mean values of Marshallian elasticity of labor hours are shown in Table 3 and by histogram in Figures The means equal , , and for Japanese workers of all incomes 6, Japanese high-income workers, and US workers of all incomes, respectively. The US mean is positive although quite small and nearly zero, but it is much greater than those from the two Japanese samples. That is, the Japanese workers are likely to reduce their labor supply following a wage rise, whereas the US workers are far from being elastic. In addition, the difference in means between the entire labor force and high-income earners is very small. The median values from those three samples are all zero and the distributions of measured elasticity look very similar. The medians of income elasticity equal zero for all the samples. Their distributions, depicted in Figures , look very similar. Concerning the means, the value of the US 5 Annual work hours are assessed by the answers to questions concerning the hours usually spent on paid work by respondents. Even if the annual work hours are identified as the hours after the hypothetical wage-doubling shock, the statistical features of estimated Hicksian elasticity remained almost unchanged. The analyses in Subsection 4.2 and later would thus yield quite similar results, whichever way we identified work hours. 6 Bessho and Hayashi (2006) analyzed the micro data of male aged in Employment Status Survey. Their estimates of Marshallian elasticity range from to with a linear labor supply function, while those estimates are bounded by and with the CES utility function. 35

9 sample of all income groups shows the highest elasticity ( 0.177), while those of the Japanese overall and high-income samples are and , respectively. Their negative values are consistent with the assumption that leisure is a normal good. Assuming 40 hours to be the maximum number of work hours, the mean value of Hicksian elasticity for all US workers (0.151) was greater than that for all Japanese workers (0.0366) and high-income earners (0.0198) 7, as shown in Table 3. Referring to the histograms of Figures , a greater portion of US elasticity is distributed over the positive range than the Japanese counterparts are. The above tendency can be also confirmed if the maximal work hours are assumed to be 80 hours (see the bottom row of Table 3). Table 3: Means of Elasticity of Work Hours Japan (Overall) Japan (High-income) U.S. (Overall) Marshallian (0.224) [849] (0.185) [376] (0.303) [1195] Income (0.199) [722] (0.173) [344] (0.309) [977] Hicksian (40 hours max.) (0.199) [575] (0.171) [314] (0.300) [918] Hicksian (80 hours max.) (0.189) [575] (0.166) [314] (0.280) [918] Notes: Standard deviations in parentheses; Sample sizes in brackets 7 Bessho and Hayashi (2006) estimate the Hicksian elasticity at the range from to with a linear labor supply function, while those estimates are bounded by 0.9 and 1.0 with the CES utility function. 36

10 Figure 1.1: Marshallian Elasticity of Work Hours (Japan, Overall) Density kdensity marshall Figure 1.2: Marshallian Elasticity of Work Hours (Japan, High-Income) Density kdensity marshall Figure 1.3: Marshallian Elasticity of Work Hours (U.S., Overall) Density kdensity marshall 37

11 Figure 2.1: Income Elasticity of Work Hours (Japan, Overall) Density kdensity q39_change_2 Figure 2.2: Income Elasticity of Work Hours (Japan, High-Income) Density kdensity q39_change_2 Figure 2.3: Income Elasticity of Work Hours (U.S., Overall) Density kdensity q39_change_2 38

12 Figure 3.1: Hicksian Elasticity of Work Hours (40 work hours max., Japan, Overall) Density kdensity hicks40 Figure 3.2: Hicksian Elasticity of Work Hours (40 work hours max., Japan, High-Income) Density kdensity hicks40 Figure 3.3: Hicksian Elasticity of Work Hours (40 work hours max., U.S., Overall) Density kdensity hicks40 39

13 Next, the elasticity of retirement will be characterized. The projected retirement ages, on average, equal 64.1, 68.0, and 63.0 years for Japanese workers of all incomes, Japanese high-income workers, and US workers of all incomes, respectively. Although the difference between the overall sample from Japan and that of the US is small, the average retirement of the wealthy Japanese is clearly later than the others (Table 4). Based on the originally projected retirement age, the expected age of retirement with a doubled wage, and the age of retirement after winning a hypothetical lottery, the elasticity of retirement was estimated for each respondent (see Table 4). Table 4: Means of Expected Age of Retirement and Means of Elasticity of Retirement Japan (Overall) Japan (High-income) U.S. (Overall) Age of retirement (original) 64.1 (5.61) [844] 68.0 (8.03) [238] 63.0 (6.96) [1130] Expected age of retirement with doubled wages 62.9 (5.71) [832] 67.2 (7.99) [365] 58.9 (8.31) [1147] Expected age of retirement in winning the lottery 61.8 (6.16) [750] 67.0 (8.32) [357] 56.0 (9.69) [1037] Marshallian elasticity (1.203) [794] (0.422) [346] (0.360) [1087] Income elasticity (1.227) [712] (0.650) [337] (0.294) [946] Hicksian (40 hours max.) (0.554) [571] (0. 232) [308] (0. 359) [890] Hicksian (80 hours max.) (0.665) [571] (0. 302) [308] (0. 337) [890] Notes: Standard deviations in parentheses; Sample sizes in brackets The mean values of its Marshallian elasticity are , , and for Japanese workers of all incomes, Japanese high-income workers, and US workers of all incomes, respectively. In other words, the US workers are expected to advance their retirement quite elastically on average, whereas the wealthy Japanese, and Japanese overall, are thought to plan on delaying retirement. Note that the positive average Marshallian elasticity of Japanese workers as a whole is severely affected by a small number of observations with extremely large elasticity. The maximum totals The median differs little between the two Japanese samples. Moreover, the two distributions of elasticity are similar (Figures ). The distribution of the US sample, in contrast, shows a clear distinction, demonstrating a greater tendency toward elasticity than the Japanese groups. In addition, the US median value equals The value 26 cannot be regarded as an outlier. If someone postponed his retirement by 25 years upon having his hourly wage doubled, a large elasticity could be derived. 40

14 Figure 4.1: Elasticity of Retirement with Doubled Wage (Japan, Overall) Density kdensity retire_40 Figure 4.2: Elasticity of Retirement with Doubled Wage (Japan, High-Income) Density kdensity retire_40 Figure 4.3: Elasticity of Retirement with Doubled Wage (U.S., Overall) Density kdensity retire_40 41

15 The average income elasticities of retirement are , , and for Japanese workers of all incomes, Japanese high-income workers, and US workers of all incomes, respectively. All of them take negative values, as expected. The highly elastic advance of retirement is again observed for the US workers, also confirmed in the distribution of elasticity (Figures ). Comparing the Hicksian elasticity values under the 40 maximum work hours among the three samples, the median values are all zero and their distributions look similar (Figures ). The means are equal to , , and for Japanese workers of all incomes, Japanese high-income workers, and US workers of all incomes, respectively. That is, in the comparison of substitution elasticity with the income effect excluded, the Japanese workers as a whole have the most elastic tendency to postpone their retirement. The same point will be found by comparing Hicksian elasticity under 80 maximum work hours 9. The next subsection will examine whether those various elasticities have statistically significant differences among the three samples. 9 Annual work hours are assessed using the answers to questions concerning hours usually spent on paid work by respondents. Even if the annual work hours are identified with the hours after the hypothetical wage-doubling shock, the same finding about Hicksian elasticity can be obtained. 42

16 Figure 5.1: Elasticity of Retirement in Winning the Lottery (Japan, Overall) Density kdensity retire_39 Figure 5.2: Elasticity of Retirement in Winning the Lottery (Japan, High-Income) Density kdensity retire_39 Figure 5.3: Elasticity of Retirement in Winning the Lottery (U.S., Overall) Density kdensity retire_39 43

17 Figure 6.1: Hicksian Elasticity of Retirement (40 work hours max., Japan, Overall) Density kdensity retire_hicks_40_1 Figure 6.2: Hicksian Elasticity of Retirement (40 work hours max., Japan, High-Income) Density kdensity retire_hicks_40_1 Figure 6.3: Hicksian Elasticity of Retirement (40 work hours max., U.S., Overall) Density kdensity retire_hicks_40_1 44

18 4.2 Comparison between All Japanese Workers and Japanese High-Income Earners This subsection reports the results of regression analyses to test for statistical differences in the Marshallian and Hicksian elasticities with regard to work hours and retirement between the two Japanese samples. Table 5 provides estimation results for the Marshallian elasticity of work hours. Looking at column (J1), the high-income dummy has no significant effect; thus, no significant difference between the elasticities of the two samples can be found. Further, the regression in column (J2) added the following explanatory variables: weekly work hours; squared weekly work hours; marital status dummies (with the base group of singles); age; educational attainment dummies (with junior high school graduates as the base group); the dummies for types of employment (with the base group of company employees, organization staff, or government employees); the coefficient of relative risk tolerance; and the interaction between the high-income earner dummy and each of the regressands above. The regression in column (J3) furthermore included hourly pay and its interaction with the high-income earner dummy. The regression in column (J4) included an annual earnings variable and its interaction with the high-income earner dummy rather than the variables composed of hourly wage. The inclusion of a high-income earner dummy and its interactions with the other variables was intended to examine the feasibility of pooling the two samples of distinct income levels because the coefficient of every explanatory variable could take a very different estimate. The estimate for the high-income earner dummy is significant at the 10% significance level in (J2); however, the significance vanishes when the income variables in (J3) and (J4) are controlled. It can therefore be concluded that no significant difference exists in this elasticity between the two Japanese samples. It is also worthwhile noting the variables of weekly work hours, age, and the dummies for divorce and completion of postgraduate study. Longer hours worked, aging, and the completion of graduate school significantly decrease the Marshallian elasticity of work hours. Meanwhile, a worker who has experienced divorce has a significantly greater elasticity because the divorce dummy coefficient is positive and significant at the 10% significance level It is difficult to interpret this positive estimate from the regression results alone. The evidence is inadequate to identify the causality concerning the dummy variable under the following two possibilities. First, workers may become more elastic after their divorce as they need to earn alimony. Second, the more elastic their labor supply is, the more likely they are to divorce. 45

19 Table 5: Comparison between the Entire Male Workers and the Male High-Income Earners in Japan (Elasticity of Work Hours, 40 work hours max. fixed for calculating the Hicksian) Marshallian Hicksian (J1) (J2) (J3) (J4) (J5) (J6) (J7) (J8) Constant *** *** *** *** *** *** *** *** (0.0073) (0.0671) (0.0832) (0.0739) (0.0079) (0.0849) (0.0919) (0.0852) High-Income * (0.0132) (0.1819) (0.1916) (0.1860) (0.0133) (0.2036) (0.2087) (0.2037) Work hours *** ** *** *** ** *** (per week) (0.0021) (0.0027) (0.0023) (0.0025) (0.0028) (0.0026) Work hours ** *** ** *** High-Income (0.0030) (0.0035) (0.0032) (0.0034) (0.0037) (0.0034) Squared work hours ** ** * ** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Squared work hours * * High-Income (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Married (0.0247) (0.0261) (0.0262) (0.0279) (0.0282) (0.0282) Divorced/Separated * * *** *** *** (0.0526) (0.0557) (0.0557) (0.0613) (0.0614) (0.0614) Widowed (0.0808) (0.0815) (0.0814) (0.0845) (0.0845) (0.0844) Married High-Income (0.1054) (0.1063) (0.1062) (0.1175) (0.1176) (0.1176) Divorced High-Income (0.1416) (0.1480) (0.1479) (0.1563) (0.1564) (0.1564) Widowed * * * High-Income (0.1657) (0.1768) (0.1765) (0.1994) (0.1998) (0.1994) Age ** ** ** (0.0009) (0.0009) (0.0010) (0.0010) (0.0010) (0.0010) Age High-Income ** * ** (0.0019) (0.0019) (0.0019) (0.0020) (0.0020) (0.0020) High school graduate (0.0272) (0.0293) (0.0295) (0.0321) (0.0323) (0.0324) Vocational School graduate (0.1233) (0.1239) (0.1238) (0.1322) (0.0000) (0.1322) Two-year college graduate (0.0368) (0.0396) (0.0399) (0.0423) (0.0427) (0.0430) Four-year college graduate (0.0289) (0.0314) (0.0321) (0.0341) (0.0350) (0.0354) Graduate School ** ** ** finish (0.0491) (0.0524) (0.0530) (0.0588) (0.0598) (0.0603) 46

20 Table 5 (Continued) (J1) (J2) (J3) (J4) (J5) (J6) (J7) (J8) High school High-Income (0.0954) (0.0966) (0.0966) (0.1009) (0.1011) (0.1010) Vocational school High-Income (0.0000) (0.0000) (0.0000) (0.0000) (0.1323) (0.0000) Two-year college High-Income (0.1296) (0.1311) (0.1309) (0.1403) (0.1407) (0.1404) Four-year college High-Income (0.0900) (0.0916) (0.0915) (0.0967) (0.0975) (0.0973) Graduate school High-Income (0.1019) (0.1044) (0.1044) (0.1117) (0.1128) (0.1126) Businessman/Director (0.0303) (0.0343) (0.0350) (0.0374) (0.0384) (0.0388) Self-Employed (0.0245) (0.0266) (0.0266) (0.0284) (0.0284) (0.0285) Family business Employee (0.0409) (0.0446) (0.0446) (0.0515) (0.0515) (0.0515) Businessman/Director High-Income (0.0621) (0.0647) (0.0650) (0.0686) (0.0692) (0.0694) Self-employed High-Income (0.0601) (0.0614) (0.0614) (0.0651) (0.0652) (0.0651) Family business employee High-Income (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Risk tolerance (0.0107) (0.0112) (0.0112) (0.0121) (0.0121) (0.0121) Risk tolerance High-Income (0.0144) (0.0150) (0.0150) (0.0160) (0.0160) (0.0160) Hourly wage (0.0468) (0.0543) Hourly wage High-Income (0.0468) (0.0543) Annual income (0.0000) (0.0000) Annual income High-Income (0.0000) (0.0000) Sample size R Notes: Standard deviations appear in parentheses. ***Significant at 1% level; **significant at 5% level; *significant at 10% level 47

21 The regression outcomes of the Hicksian elasticity of work hours under the 40 maximal hours per week are shown in Table The sets of explanatory variables in columns (J5), (J6), (J7), and (J8) are the same as those in (J1), (J2), (J3), and (J4), respectively. No significant difference between the Japanese overall and wealthy samples can be confirmed with regard to Hicksian elasticity. None of the estimates for the high-income earner dummies was significant in any column from (J5) to (J8). The estimates of the coefficient for weekly work hours and the divorce dummy deserve consideration. The coefficient of weekly work hours was, for instance, estimated to be significantly negative for each specification; that of its square takes a significantly positive estimate. In other words, the extension of work hours leads to the decline of Hicksian elasticity, though its marginal effect diminishes gradually. On the other hand, the above effects of weekly work hours and squared weekly work hours will be cancelled out by their respective interaction terms with the dummy for high-income earners. It can therefore be said that their effects cannot be found for the high-income workers. Another interesting finding here is the more significant and larger positive effect of divorce on Hicksian elasticity of work hours than that on the Marshallian described above. That is, the substitution elasticity between leisure and consumption, with the income effect excluded, is greater for those who have experienced divorce. It seems to confirm that increasing probability of divorce is caused by more elastic labor supply rather than by the effect of alimony payments on extending one s work hours. The estimation results for the Marshallian elasticity of retirement appear in Table 6. None of the estimates for high-income earner dummies from (RJ1) through (RJ4) is significant, nor can a disparity in elasticity between the two Japanese samples be observed statistically. 11 When fixing the maximum work time endowment at 80 hours, the corresponding results are so similar that they were omitted. 48

22 Table 6: Comparison between the Entire Male Workers and the Male High-Income Earners in Japan (Marshallian Elasticity of Retirement) (RJ1) (RJ2) (RJ3) (RJ4) Constant (0.0366) (0.1626) (0.1993) (0.1743) High-Income (0.0664) (0.4472) (0.4606) (0.4462) Work hours (per week) ** ** ** (0.0051) (0.0063) (0.0053) Work hours High-Income (0.0074) (0.0085) (0.0077) Squared work hours * * (0.0001) (0.0001) (0.0001) Squared work hours High-Income (0.0001) (0.0001) (0.0001) Married (0.0607) (0.0621) (0.0626) Divorced/ Separated (0.1294) (0.1339) (0.1341) Widowed (0.2080) (0.2043) (0.2045) Married High-Income (0.2528) (0.2484) (0.2487) Divorced High-Income (0.3406) (0.3467) (0.3471) Widowed High-Income (0.4044) (0.4196) (0.4197) Age *** *** *** (0.0022) (0.0023) (0.0023) Age High-Income (0.0046) (0.0047) (0.0047) High school graduate (0.0672) (0.0707) (0.0711) Vocational school graduate (0.3096) (0.3030) (0.3035) Two-year college graduate (0.0898) (0.0943) (0.0953) Four-year college graduate (0.0713) (0.0757) (0.0773) Graduate school finish (0.1185) (0.1234) (0.1251) High school High-Income (0.2508) (0.2472) (0.2472) Vocational school High-Income (0.0000) (0.0000) (0.0000) Two-year college High-Income (0.3246) (0.3198) (0.3199) 49

23 Table 6 (Continued) (RJ1) (RJ2) (RJ3) (RJ4) Four-year college High-Income (0.2353) (0.2329) (0.2330) Graduate school High-Income (0.2616) (0.2607) (0.2610) Businessman/ Director (0.0761) (0.0850) (0.0872) Self-Employed *** *** *** (0.0598) (0.0632) (0.0634) Family business employee ** ** ** (0.0980) (0.1041) (0.1043) Businessman/Director High-Income (0.1514) (0.1545) (0.1560) Self-employed High-Income (0.1450) (0.1443) (0.1446) Family business employee High-Income (0.0000) (0.0000) (0.0000) Risk tolerance * (0.0261) (0.0268) (0.0268) Risk tolerance High-Income (0.0353) (0.0360) (0.0360) Hourly wage (0.1117) Hourly wage High-Income (0.1118) Annual income (0.0001) Annual income High-Income (0.0001) Sample size R Notes: Standard deviations appear in parentheses. ***Significant at 1% level; **significant at 5% level; *significant at 10% level The estimates for weekly work hours, age, and the dummies for types of employment also require attention, in the light of these regression outcomes. Facing an unexpected wage rise, the habit of working for longer hours or being self-employed significantly delays retirement; on the other hand, aging and being a family business employee significantly advances retirement. Regression results for Hicksian elasticity of retirement are shown in Table 7. The coefficient of the high-income earner dummy has a significantly negative estimate only in column (RJ5), which fixes the weekly time endowment at 40 hours and includes only the 50

24 constant term and that dummy as explanatory variables. The estimate for the same variable did not achieve significance in any column from (RJ6) to (RJ12). The estimation was, in addition, conducted by identifying the annual work hours with the hours following the hypothetical wage-doubling shock, although none of the outcomes are shown in this paper. No corresponding estimates were ever found to be significant under assumptions of either 40 or 80 maximum weekly work hours. Table 7: Comparison between the Entire Male Workers and the Male High-Income Earners in Japan (Hicksian Elasticity of Retirement, 40 or 80 work hours max. fixed for the calculation) 40 work hours maximum 80 work hours maximum (RJ5) (RJ6) (RJ7) (RJ8) (RJ9) (RJ10) (RJ11) (RJ12) Constant *** *** (0.0196) (0.1304) (0.1453) (0.1310) (0.0236) (0.1495) (0.1666) (0.1502) High-Income ** (0.0330) (0.3127) (0.3225) (0.3133) (0.0399) (0.3585) (0.3697) (0.3592) Work hours * (per week) (0.0039) (0.0046) (0.0039) (0.0045) (0.0052) (0.0045) Work hours High-Income (0.0054) (0.0060) (0.0054) (0.0062) (0.0069) (0.0062) Squared work hours (0.0000) (0.0000) (0.0000) (0.0000) (0.0001) (0.0000) Squared work hours High-Income (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Married (0.0457) (0.0462) (0.0462) (0.0524) (0.0530) (0.0530) Divorced/ Separated (0.0977) (0.0979) (0.0979) (0.1120) (0.1122) (0.1123) Widowed (0.1511) (0.1513) (0.1513) (0.1733) (0.1734) (0.1735) Married High-Income (0.1692) (0.1696) (0.1698) (0.1940) (0.1944) (0.1946) Divorced High-Income (0.2376) (0.2380) (0.2382) (0.2724) (0.2728) (0.2731) Widowed High-Income (0.2909) (0.2912) (0.2913) (0.3335) (0.3338) (0.3339) Age *** *** *** *** *** *** (0.0016) (0.0017) (0.0017) (0.0019) (0.0019) (0.0019) Age High-Income (0.0033) (0.0034) (0.0034) (0.0038) (0.0038) (0.0039) High school graduation (0.0519) (0.0524) (0.0525) (0.0595) (0.0601) (0.0602) Vocational school graduation (0.2052) (0.0000) (0.2057) (0.2353) (0.0000) (0.2358) 51

25 Table 7 (Continued) (R9) (R10) (R11) (R12) (R13) (R14) (R15) (R16) Two-year college graduation (0.0687) (0.0694) (0.0701) (0.0788) (0.0796) (0.0804) Four-year college graduation (0.0553) (0.0569) (0.0576) (0.0634) (0.0652) (0.0661) Graduate School finish (0.0959) (0.0972) (0.0982) (0.1099) (0.1114) (0.1126) High school High-Income (0.1684) (0.1690) (0.1689) (0.1931) (0.1937) (0.1937) Vocational school High-Income (0.0000) (0.2054) (0.0000) (0.0000) (0.2354) (0.0000) Two-year college High-Income (0.2304) (0.2310) (0.2311) (0.2642) (0.2648) (0.2650) Four-year college High-Income (0.1587) (0.1598) (0.1599) (0.1820) (0.1832) (0.1833) Graduate school High-Income (0.1836) (0.1850) (0.1851) (0.2105) (0.2121) (0.2122) Businessman/ Director ** ** ** * (0.0632) (0.0652) (0.0668) (0.0725) (0.0748) (0.0766) Self-Employed ** ** ** ** ** ** (0.0464) (0.0465) (0.0466) (0.0532) (0.0533) (0.0534) Family business employee (0.0798) (0.0799) (0.0800) (0.0914) (0.0916) (0.0918) Businessman/Director High-Income (0.1142) (0.1155) (0.1165) (0.1310) (0.1324) (0.1335) Self-employed High-Income (0.1066) (0.1067) (0.1068) (0.1223) (0.1223) (0.1225) Family business employee High-Income (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Risk tolerance (0.0204) (0.0204) (0.0204) (0.0233) (0.0234) (0.0234) Risk tolerance High-Income (0.0267) (0.0267) (0.0267) (0.0306) (0.0306) (0.0306) Hourly wage (0.0891) (0.1021) Hourly wage High-Income (0.0891) (0.1022) Annual income (0.0000) (0.0001) Annual income High-Income (0.0000) (0.0001) Sample size R Notes: Standard deviations appear in parentheses. ***Significant at 1% level; **significant at 5% level; *significant at 10% level 52

26 Aging causes a significant increase in the Hicksian whenever it is included (R10 12, 14 16); in contrast, the state of being self-employed leads to a significant decrease. The coefficient of the dummy for being a businessperson, a director, or self-employed also takes significantly negative estimates in many cases (R10 12, R14). To summarize the comparison between the elasticity of overall and high-income Japanese samples, first, there was no solid evidence to prove any significant difference between the two groups concerning the Marshallian and Hicksian elasticities of work hours and the Marshallian of retirement. On the Hicksian elasticity of retirement, there was a case where a significant difference was observed, depending on how it was measured. However, the average elasticity of high-income Japanese workers never exceeded that of the entire labor force whether or not the control variables were included in the regressions. Second, longer work hours and aging tend to decrease elasticity; aging raises the elasticity of retirement. The retirement of self-employed workers is significantly more elastic and is brought forward if they have an unexpected increase in their wages. In addition, as a person s Hicksian elasticity gets larger, his probability of experiencing a divorce becomes higher. 4.3 Comparison between the Japanese and US Workers as a Whole This subsection will examine the difference between workers in Japan and the US. The regression results related to the elasticity of work hours are shown in Table 8. In all the regressions of Marshallian elasticity (columns (1) (4)), the dummy for the US workers is significantly positive. The result in column (1) means that the difference between the average elasticity of the Japanese ( ) and their US counterparts ( ) is statistically significant. In other words, Japanese workers would decrease their labor supply in a somewhat elastic way if there were a wage rise. The US labor supply would, in contrast, change little in the same situation. Considering the other explanatory variables, first, the extension of work hours has a greater, significant effect to decrease this labor-supply elasticity in the US. Second, the dummy variable of graduating from high school and that of completing two years in college take significant and negative estimates of their coefficients. 53

27 Table8: Comparison between the Entire Male Workers in Japan and Those in the U.S. (Elasticity of Work Hours, 40 work hours max. fixed for calculating the Hicksian) Marshallian Hicksian (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** ** *** *** *** ** *** (0.0094) (0.0827) (0.1021) (0.0911) (0.0111) (0.1136) (0.1229) (0.1141) U.S. sample *** *** *** *** *** *** *** *** (0.0123) (0.1304) (0.1455) (0.1391) (0.0141) (0.1617) (0.1684) (0.1632) Work hours *** ** ** ** * ** (per week) (0.0026) (0.0033) (0.0028) (0.0034) (0.0038) (0.0034) Work hours *** *** *** *** *** *** U.S. (0.0043) (0.0048) (0.0045) (0.0052) (0.0054) (0.0052) Squared work hours * (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Squared work hours *** *** *** *** *** *** U.S. (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) Married (0.0305) (0.0320) (0.0323) (0.0373) (0.0378) (0.0378) Divorced/ Separated ** ** ** (0.0648) (0.0683) (0.0687) (0.0820) (0.0822) (0.0822) Widowed (0.0995) (0.1000) (0.1004) (0.1130) (0.1130) (0.1131) Married U.S (0.0381) (0.0397) (0.0400) (0.0461) (0.0465) (0.0465) Divorced U.S (0.0749) (0.0782) (0.0786) (0.0947) (0.0948) (0.0949) Widowed U.S (0.1509) (0.1514) (0.1520) (0.1745) (0.1747) (0.1747) Age * * (0.0011) (0.0012) (0.0012) (0.0014) (0.0014) (0.0014) Age U.S (0.0014) (0.0015) (0.0015) (0.0017) (0.0018) (0.0018) High school graduation (0.0335) (0.0360) (0.0363) (0.0429) (0.0432) (0.0434) Vocational school graduation (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Two-year college graduation (0.0453) (0.0485) (0.0492) (0.0566) (0.0570) (0.0576) Four-year college graduation (0.0355) (0.0385) (0.0395) (0.0457) (0.0468) (0.0474) Graduate school * * finish (0.0605) (0.0643) (0.0654) (0.0788) (0.0799) (0.0808) High school * * * U.S. (0.0650) (0.0676) (0.0679) (0.0765) (0.0769) (0.0770) Vocational school U.S. (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Two-year college ** * * U.S. (0.0759) (0.0790) (0.0796) (0.0900) (0.0904) (0.0908) 54

28 Table 8 (Continued) (1) (2) (3) (4) (5) (6) (7) (8) Four-year college * U.S. (0.0664) (0.0697) (0.0704) (0.0784) (0.0797) (0.0800) Graduate school U.S. (0.0856) (0.0908) (0.0921) (0.1046) (0.1073) (0.1083) Businessman/Director (0.0373) (0.0421) (0.0432) (0.0500) (0.0513) (0.0520) Self-Employed (0.0302) (0.0326) (0.0329) (0.0380) (0.0380) (0.0381) Family business employee (0.0503) (0.0547) (0.0550) (0.0689) (0.0689) (0.0690) Businessman/Director High-Income (0.0591) (0.0629) (0.0639) (0.0731) (0.0741) (0.0747) Self-employed U.S. (0.0426) (0.0452) (0.0455) (0.0519) (0.0523) (0.0524) Family business employee * * ** ** ** U.S. (0.0825) (0.0870) (0.0874) (0.1124) (0.1124) (0.1125) Risk tolerance (0.0132) (0.0138) (0.0138) (0.0162) (0.0162) (0.0162) Risk tolerance U.S. (0.0190) (0.0196) (0.0197) (0.0231) (0.0231) (0.0231) Hourly wage (0.0574) (0.0726) Hourly wage ** U.S. (0.0756) (0.0916) Annual income (0.0000) (0.0000) Annual income * U.S. (0.0000) (0.0000) Sample size R Notes: Standard deviations appear in parentheses. ***Significant at 1% level; **significant at 5% level; *significant at 10% level With the Hicksian elasticity of work hours as the dependent variable, the effect of the US sample dummy is significantly positive ((5) (8)). The coefficient of the interaction between the weekly work hours and the US sample dummy was estimated to be significantly negative in any case of its inclusion ((6) (8)), as in the previous regressions. The dummy of experiencing divorce has a positive estimate in columns (6) (8), although its interaction with the US sample dummy shows no significant influence. Regardless of the geographical difference, higher elasticity may cause a man to be more likely to divorce. Also, the coefficient of the interaction term between the state of being a family business employee and the US sample dummy is significant and positive. 55