This action might not be possible to undo. Are you sure you want to continue?
Contents lists available at ScienceDirect
Journal of Public Economics
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j p u b e
Job search and unemployment insurance: New evidence from time use data
Alan B. Krueger a,b, Andreas Mueller c,d,
Woodrow Wilson School and Economics Department, Princeton University, Princeton, NJ 08540, United States NBER c Institute for International Economic Studies, Stockholm University, SE-106 91 Stockholm, Sweden d Economics Department, Princeton University, United States
a r t i c l e
i n f o
a b s t r a c t
This paper provides new evidence on job search intensity of the unemployed in the U.S., modeling job search intensity as time allocated to job search activities. The major ndings are: 1) the average U.S. unemployed worker devotes about 41 min to job search on weekdays, which is substantially more than their European counterparts; 2) workers who expect to be recalled by their previous employer search substantially less than the average unemployed worker; 3) across the 50 states and D.C., job search is inversely related to the generosity of unemployment bene ts, with an elasticity between −1.6 and −2.2; 4) job search intensity for those eligible for Unemployment Insurance (UI) increases prior to bene t exhaustion; and 5) time devoted to job search is fairly constant regardless of unemployment duration for those who are ineligible for UI. © 2009 Elsevier B.V. All rights reserved.
Article history: Received 26 March 2009 Received in revised form 2 December 2009 Accepted 11 December 2009 Available online 22 December 2009 JEL classi cation: J64 J65 Keywords: Unemployment Unemployment insurance Job search Time use Unemployment bene ts Inequality
1. Introduction It is well known that since the early 1980s the unemployment rate has been lower in the U.S. than in Europe. Our tabulations of international time use data (circa 1998–2007) also indicate that unemployed Americans tend to devote much more time to searching for a new job than their European counterparts (see Fig. 1). On weekdays, for example, the average unemployed worker spent 41 min a day searching for a job in the U.S., compared with just 12 min in the average European country with available data. One explanation for the comparatively low unemployment rate and high search time in the U.S. is the relatively modest level and short duration of Unemployment Insurance (UI) bene ts in most states in the U.S. In this paper we examine the effects of UI on the amount of time devoted to job search by unemployed workers in the U.S., using features of state UI laws for identi cation. A large and related literature examines the effects of UI on the duration of unemployment spells. For example, more generous UI bene ts have been found to be associated with longer spells of unemployment, with an elasticity of about 1.0 (see Krueger and Meyer (2002) for a survey). In addition, the job nding rate jumps up
around the time bene ts are exhausted (Mof tt, 1985; Katz and Meyer, 1990a; see Card et al., 2007 for a critical review). UI is expected to affect the duration of unemployment through its effect on the amount of effort devoted to searching for a job and the reservation wage of the unemployed, yet these variables have rarely been studied directly.1 We attempt to ll this void by modeling the amount of time that unemployed individuals devote to searching for a new job over the course of unemployment spells using data from the American Time Use Surveys (ATUS) from 2003 to 2007. Section 2 describes the ATUS data and presents summary statistics. In Section 3, we evaluate the predictions of Mortensen's (1977) canonical model of UI and job search.2 The Mortensen model predicts that for a newly laid-off worker, search effort is decreasing in the level of UI bene ts, whereas for those unemployed who are not eligible for UI or who have exhausted their UI bene ts, search effort is increasing in the bene t level. This latter implication is called the entitlement effect, as higher bene ts raise the value of being unemployed in the future and thus raise the value of obtaining a job.3 Furthermore, the
Corresponding author. Institute for International Economic Studies, Stockholm University, SE-106 91 Stockholm, Sweden. 0047-2727/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jpubeco.2009.12.001
1 An exception is Barron and Mellow (1979), who used the May 1976 CPS supplement on job search activities in the last month, and nd that the unemployed searched an average of 7 h a week. See Feldstein and Poterba (1984) for related evidence on self-reported reservation wages and unemployment in the U.S. based on the same CPS data. 2 Labor supply models such as, e.g., Mof tt and Nicholson (1982) yield similar predictions. 3 Levine (1993) provides some evidence on the entitlement effect.
After bene ts are exhausted. we infer UI eligibility from the type of unemployment and the workers' full-time/ part-time status on the previous job. The ATUS collects detailed information on the amount of time respondents devoted to various activities in the previous day. however. • Re.A. Note that we were able to identify voluntary job leavers only when they were already unemployed at the time of the CPS interview. we classify those who worked part-time as ineligible. Our sample consists of 2171 unemployed individuals. In states where part-time job seekers do not qualify for UI. those who report in the CPS that their temporary job has ended and those who are employed at the time of the CPS interview (and subsequently became unemployed). job search effort is predicted to remain constant.B. 6 In a companion paper (Krueger and Mueller.2%. compared with the CPS the proportion of the unemployed classi ed as job leavers is relatively low in our sample. . the unemployed spend around 32 min a day (including weekends) searching for a job. 5 See also Feldstein (1976) and the empirical work of Katz and Meyer (1990a. The sample unemployment rate is 5.6 Even if we restrict the sample to those model predicts that search effort is increasing in the mean wage offer and the dispersion of potential wage offers. In the ATUS data. whereas the employed and those classi ed as out of the labor force devote less than a minute a day to job search. regardless of whether they looked for work in the four weeks prior to the survey.or new entrants are de ned as those unemployed who indicate that they were re.or new entrants in the CPS. we nd that job search intensity is inversely related to UI bene t generosity for those who are eligible for UI. job interviewing. 2. A. new entrants and voluntary job leavers as ineligible. which is a nationally representative time use survey covering the whole civilian non-institutional population aged 15 and older. with a higher dispersion of potential wages. and re-/new entrants into the labor force. The Appendix Table provides a detailed list of activities that are identi ed as job search. 12% in our period). We classify job losers and those on temporary layoff as eligible for UI. Such misclassi cation errors are likely to lead us to underestimate the effects of UI in Sections 3 and 4 below. only contains a question on whether the unemployed expect to be recalled. de ned as the fraction of those with non-zero search time on the diary day.b) on recall and job nding hazards. 1986). We restrict our sample to the population aged 20–65 to abstract from issues related to youth unemployment and retirement. The ATUS questionnaire. The ATUS labor force recode de nes unemployment in the same way as the CPS (not working in the reference week. The model predicts that for an eligible unemployed. 2. The intuition for the latter is that. Average number of minutes devoted to job search per day on weekdays by unemployed workers in various countries. We undoubtedly have some classi cation errors when it comes to assigning UI eligibility in our sample. we nd a striking contrast in the pro les of job search activity across those with different durations of unemployment: search activity increases as week 26 (bene t exhaustion) approaches for the UI eligible. Thus. Job search activities include contacting a potential employer. we use information from the nal CPS interview to classify individuals into the other three groups. etc. of which 344 were on temporary layoff. calling 4 See also Stigler (1962) for a seminal discussion of how wage dispersion affects the payoff from search effort. and Ljungqvist and Sargent (1995) for how progressive taxation affects job search effort through after-tax wage compression. Because the ATUS lacks information on UI receipt. Respondents are interviewed within 2–5 months of their last CPS interview. The CPS/ATUS de nition of unemployed also includes those on temporary layoff with an expectation of recall to their previous employer. Descriptive statistics of job search activities Table 1a reports descriptive statistics on the average number of minutes devoted to job search by labor force status. Several results are worth highlighting. and available for work in the reference week). Data and descriptive statistics We use data from ve consecutive years (2003–07) of the ATUS. within-state residual wage dispersion. Those who are classi ed as out of the labor force in the CPS but as unemployed in the ATUS are also included in this category.4 We also expect search effort to be lower for those unemployed who expect to be recalled to their previous job (see Katz. Krueger.1. and re-entrants. In Section 4. we evaluate the predictions of the Mortensen model regarding job search intensity and unemployment duration. Section 5 offers some concluding thoughts as to how our results relate to search theory and how time use data can be used to further study UI and job search behavior. We classify people who were employed in CPS and unemployed in ATUS as job losers because the share of voluntary job leavers among the unemployed in CPS is much lower than that of job losers (43% vs. First. or visiting an employment agency. Consequently. job seekers' predicted wages. actively looking for a job in the 4 weeks prior to the interview. Mueller / Journal of Public Economics 94 (2010) 298–307 299 Fig. while the pro le is fairly at for those who are ineligible for UI. 1. The sample is drawn from the 8th outgoing rotation group of the Current Population Survey (CPS). reading and replying to job advertisements. Speci cally: • Job losers are de ned as those on layoff in the CPS. • Voluntary job leavers are de ned as those who indicate in the CPS that they quit their job.5 We empirically test these predictions and estimate the effect on job search of the generosity of UI bene ts. which exactly matches the of cial unemployment rate over the same period. voluntary job leavers. We can disaggregate the unemployed into four groups: job losers. An unemployed individual who is ineligible for bene ts is predicted to devote a constant amount of time to job search because of the absence of learning and the assumption of stationarity in the Mortensen model. those expecting to be recalled to their previous employer. Most importantly. as the effects are expected to be of opposite sign for the UI eligible and ineligible. 2008a) we found similar evidence across 14 countries. on average. job search effort increases over the unemployment spell as bene ts are exhausted. there is a greater bene t from searching for a high paying job. Sample weights are used in all of our estimates. It also shows the participation rate in job search. recall expectations and other variables.
091 % of total Average job search.6% 3.7% (% of employed) 94.300 A.9% 14.9% 106.6% 0.2 185. compared with 6.7% 107.5 3.1 Participation in job search 0.4 Notes: Averages and participation rates are computed with survey weights.0 15. those classi ed as out of the labor force in ATUS searched for only 4.B.9% 3.5% 1.9 19.8% 3.7% 24.8 Participation in job search 0.9% 2. Universe: Civilian. job search is heavily concentrated on weekdays (see Table 1b).7% (% of employed) 94.2 16. Nearly a quarter of the unemployed engage in job search activities on any given weekday.4 49. Second.6% 27.5 1.0% (% of not in labor force) 10.6% 2.6 0. aged 20–65.1 0.8 76.6 115.0% 1. and a quarter of job searchers spent more than 240 min .1 130.3 33.6% 22.4% 3.6% 1.576 824 1534 42. # of respondents By labor force status Employed Unemployed Not in labor force By type of employed Working in CPS Unemployed in CPS Not in labor force in CPS By type of unemployed Jobloser On temporary layoff w/ recall expectation Jobleaver Re.0 123.9% 19.6 3.7 45.8 56.7% Average job search (participants).6 136.9% 33.7% 0.7% 17.141 395 755 21.2 13.5 5. in min per weekday 0.3% 0.6% 19.8 176.9% 86.1% 0. Krueger.8 0. by labor force status (weekdays and weekends). in min per weekday 99.5% 1.4% 4.4% 143.8% 167.3 163.8 0.6 181.6 32.9 23.7 41.7% on weekends.9 76.5% 7.6% (% of unemployed) 43.0 188. non-institutional population. The estimates are based on weekdays only.9 206.7 Notes: Averages and participation rates are computed with survey weights.1% 164.9 152.1% 53.9% (% of not in labor force) 10.5% 163.1 2.4 38.0 166.5% (% of unemployed) 45. aged 20–65. This suggests that the conventional labor force categories represent meaningfully different states.2% 0.or new entrant By UI eligibility status UI ineligible UI eligible By type of “not in labor force” Working in CPS Unemployed in CPS Not in labor force in CPS 572 159 4764 473 603 488 171 25 392 20.5 2.5% Average job search (participants).0% 56.7 0.8% 3. The average duration of job search among those who searched is 167 min. A. Fig.8 0. those who participate in job search on the diary day tend to devote a great deal of time to it.7% (% of unemployed) 46.6 30. by labor force status (weekdays only). 2 shows a kernel density diagram for the duration of job search conditional on searching on the diary day.9% 3.4% 15.6 25.4% 2. Third. Table 1b Descriptive statistics ATUS 2003–2007.2 52.6% 8.291 1076 5495 % of total Average job search. in min per day 0.7% 86.8 159.5 33.4% 3.7 170.3% 36.7 7 Corroborating evidence from job nding rates is in Flinn and Heckman (1983). who were classi ed as unemployed in the CPS interview (2–5 months prior to the ATUS interview).4% (% of unemployed) 44.0% 37.3% 134.1% 24.2 47.7 2.8 212. non-institutional population. in min per day 101.7 69.9% 169.1% 0.9% 19.5% 2. # of respondents By labor force status Employed Unemployed Not in labor force By type of employed Working in CPS Unemployed in CPS Not in labor force in CPS By type of unemployed Jobloser On temporary layoff w/ recall expectation Jobleaver Re.2 0.5 171.934 2171 11.4% 15.2 min.4 169.1 27.3 40. Mueller / Journal of Public Economics 94 (2010) 298–307 Table 1a Descriptive statistics ATUS 2003–2007.or new entrant By UI eligibility status UI ineligible UI eligible By type of “not in labor force” Working in CPS Unemployed in CPS Not in labor force in CPS 1181 305 9605 1000 1171 943 344 65 819 40.4 171.1 1. Universe: Civilian. Both weekdays and weekends are included in the sample.4 4.
10 See http://workforcesecurity. ranging from $210 in Mississippi to $575 in Massachusetts in 2007. for unemployed with dependents. however.9 The data for maximum weekly bene t amounts is taken from the U. We proceeded in two steps. . The replacement rate is typically around 50% to 60% of the wage earned on the previous job. This difference. UI recipients were able to receive up to 13 additional weeks of bene ts through the federal Temporary Extended Unemployment Compensation Act of 2002. Mueller / Journal of Public Economics 94 (2010) 298–307 301 As described below. on average. denoted log(wis). The maximum bene t varies with time only for unemployed with dependents in New Mexico. Zi includes dummies for each group of unemployed workers (job loser. Finally. the respondent's predicted wage.doleta. we estimated micro regressions in which the total amount of time allocated to job search on the diary day was the dependent variable and the explanatory variables included the maximum weekly UI bene t. married or cohabiting with a partner. is not very strict as most states rely on postal or phone reports to enforce these job search requirements (see Anderson. We rst estimated the predicted wage and residual wage dispersion facing each job seeker. the regression models we estimated are: logðwis Þ = a + bXi + ds + εis ð1Þ Fig.11 We predicted each ATUS respondent's expected log wage. the presence of children under age 18 in the household. subject to a maximum bene t. 2. Job losers search 32 min more than those who expect to be recalled to their previous job. we take the average of maximum weekly bene t amounts across the 5 years of the ATUS by state and number of dependents.2.813 workers from the CPS outgoing rotation group les for 2004 and 2005. In most states. Unemployment insurance To qualify for unemployment bene ts all states require a worker to have earned a certain amount of earnings during a reference period or to have worked for a certain period of time. sex. we report average minutes of job search by UI eligibility status. using the coef cients from the wage regression (1). and in some cases the spouse. NM. a measure of wage dispersion in the state. ME. and a dummy for the presence of children. Monitoring in the US. the maximum duration of bene ts is 26 weeks. marital status. full-time and part-time students and employed with hourly earnings of less than $1 or more than $200. and a dummy for whether the diary day was a weekend. which are included in the wage and job search equations.asp#Statelaw. CT. In 2003 in New Mexico. 11 The results of the wage regressions are reported in column 4 in Table 2. Relationship between unemployment bene ts and job search To evaluate the predictions of the models outlined in the introduction. Kernel density: Job search (conditional on non-zero search). The maximum weekly bene t amount varies with individual characteristics in the states where dependents' allowances are provided beyond the maximum weekly bene t of a single earner. Standard errors are robust to correlated residuals within states and heteroskedasticity. Fourth. however. devoting almost an hour to job search a day. such as a certain number of employer contacts per week. IL. Krueger. Xi is a set of controls such as education and sex. and bene ts were extended for 26 weeks in a small number of “high” unemployment states. interaction terms of partner and children with female. there are large differences in job search effort depending on the reason for unemployment. and then used these estimates as explanatory variables in the job search equation.A. We computed the standard deviation of residuals from the wage 8 According to Krueger and Meyer (2002) around 35% of the unemployed receive the maximum bene t. OH. The maximum duration of bene ts may be less than 26 weeks for UI claimants who had insuf cient earnings during the reference period. where wis is the hourly wage of worker i in state s. sist = α + β1 logðwbaist Þ + β2 logðwis Þ + β3 stdðresid: wÞs + γ1 Xi ð2Þ + γ2 Zi + dt + ist 2. Zi is a set of controls exclusively included in the search equation. education. Speci cally. to continue to qualify for UI bene ts. PA and RI.S. MA. on temporary layoff. falls to 6 min when we control for observable characteristics such as age. A. Montana (28 weeks) and Washington (30 weeks until 2007). we set the maximum weekly bene t to the maximum weekly bene t of a single earner. The main source of variation in maximum bene ts comes from variation across states as we take into account dependents' allowances only in those ten states that provide these allowances beyond the maximum bene t. which introduced dependents' allowances in 2004. We took differences across states in the de nition of the spouse as a dependent into account. IA. Most states in the US require active job search. dt a time effect (month and year) and ds a state effect. 3. searching for a job on the diary day. The number of dependents usually includes children aged 17 and younger. 2001).or new entrants. During 2003.B. and personal characteristics.8 Ten states provide dependents' allowances beyond the maximum bene t. Those eligible for UI search 13 min more on an average day than those who are not eligible.10 Except for New Mexico. and around 22 min more than re. job leaver and re-/new entrant). The wage equation was estimated using a sample of 319. According to Krueger and Meyer (2002) around half of the recipients qualify for the full 26 weeks. We exclude observations from 2003 when we examine job search behavior around 26 weeks of unemployment for the UI eligible because of complications caused by the extended bene ts program.gov/unemploy/statelaws. The hourly wage is adjusted for top coding and overtime earnings/tips. Job leavers also have a high intensity of search. The maximum weekly bene t varies widely across states. 9 These states are AK. although there are some exceptions: Massachusetts (30 weeks). Department of Labor's Comparison of State UI Laws. wbaist is the maximum weekly bene t amount. sist is time allocated to job search (in minutes per day) of individual i in state s and time t. We exclude from the sample self-employed and self-incorporated. our regression model exploits variation in the maximum weekly bene t amounts across states and number of dependents.
full-time and part-time students and employed with hourly earnings of less than $1 or more than $200.052 (0.479) 21. (2) for three separate samples. equation for each state (denoted std(resid. w)s denotes the standard deviation of the residuals from the wage equation by state.061 (0.682 (18. suggesting a different response to bene t generosity.347) −14.717 (17.507 (20.080) 9. aged 20–65.656 (5.247) 83.13 Wage equation-dependent variable: log(hourly wage) 2. and re-entrants. this is the only sample for which the coef cient on bene ts is statistically signi cant at the 10% level. we classify those who worked part-time as ineligible. To gauge the magnitude of this elasticity. (1) The base group consists of Job losers.751 (18. This is a contrast to Krueger and Mueller's (2008a) cross-country study.329) 0.950) −11.039) −14.2 (the elasticity is computed by dividing the coef cient estimate by the mean of the dependent variable).086) −16.857) −3.06 (26. in min per day Mean of dependent variable Log(maximum weekly bene t amount) Fitted log(hourly wage) Std(residual of wage equation) — by state On temporary layoff w/ recall expectation (1) Jobleaver Re.492)*** −71.693 (4.066 (48.16 0.296) −34.884 (4.2 (0.497 (12.75 0. A test of the equality of the bene t coef cients for those eligible and ineligible for UI rejects at the 10% level.341 (66. The CEPR version of the ORG contains hourly wage series that adjust for topcoding and overtime earnings/tips.555 (100.905) −11.282 (32.864) 105. new entrants and voluntary job leavers as ineligible.4 10.363) −5.1 −57.09 (8.053 (0.772) 274. Column 2 reports the results for the subsample of eligible without an expectation of recall to their previous employer and with a duration of unemployment of 26 weeks or less.096 (19.29 0.389) −21.099 (64. Mueller / Journal of Public Economics 94 (2010) 298–307 Mean (Std) Full sample (1) 32.002)*** 0.876 (16.914)** −53.002)*** 1.883 (3.302 Table 2 Results of linear regressions.649 (16.S.744) 7.991) −68. are not on temporary 12 The coef cient on the tted log hourly wage in our regression Tables 2 and 3 shows that the tted wage is a strong and signi cant predictor of job search.813 0.911) 7. As described in the text.605 (3.16 0. states than across countries.676)*** −169.816 (7.51 0. layoff.062)* x 2171 0. Time devoted to job search is predicted to decrease by 54 min a day.1 Subsample (2): eligible w/o recall expectation and unempl. with an elasticity in excess of 2.078 (0.786) −30. 2004 and 2005. For those not eligible for bene ts in column 3 the elasticity is positive but not signi cant.49 0. w)) as an indicator of the dispersion in the potential wage offer distribution.89 (0.113) 14.348 (37. and therefore conveys less signal than in the cross-country data. The results shown in columns 1 to 3 are based on the following regression equation: sist = α + β1log(wbaist) + β2loĝ(wis) + β3std(resid. (2) The base group consists of those with a high school degree or less.09 −6.797)*** −115.023) 0. ** signi cant at 5%.379 (196.925 (14.220) 2. Table 3 reports estimated coef cients of the Tobit model as well as an adjustment factor that allows one to compute the marginal effect of each variable. Dependent variable: time allocated to job search.615) −59. Column 1 shows the results for the full sample of unemployed individuals aged 20–65.407)* −6. which found that job search is higher in countries with higher wage dispersion.50 0. the elasticity for the maximum weekly bene t is −1. The results shown in column 4 (the wage equation) are based on the following regression equation: log(wis) = a + bXi + ds + εis.B. When we restrict the sample to those who appear eligible for UI bene ts. w)s + γ1Xi + γ2Zi + dt + µist where loĝ(wis) is the tted wage based on the coef cients estimated in the wage equation and std(resid. Columns 2 and 3 report the same regressions for UI eligible and ineligible.632) 39.966) 0.034) −13.663)* 174.29 Robust standard errors in parentheses. In states where part-time job seekers do not qualify for UI.991) −46.209 (0.456 (10.76 5. controlling for bene ts and other factors. *** signi cant at 1%.198) 0.732) −32.15 0.021 (13.634) 52. Krueger.167)** −26.001 (0.284 (14. Column 3 reports the results for the subsample of ineligible. * signi cant at 10%.001)*** −0.868 (101.585) −13.60 (0.543) −11.28 0. Notes: Regressions are weighted using survey weights.176 (8. dur.400) −7.5. consider the effect of changing the WBA from the state with the lowest to the highest bene t (for a person without dependents).573 (0. A.877 (28.971) 110.805 (33.872 (17.375 (128.691) 0.973)*** 12.280)** −5.231 (0.275 (30. We also estimated Tobit models for the same three samples to account for the mass of workers with 0 min of job search on the diary day.764 (72. The residual wage dispersion term is insigni cant but usually positive in most of the OLS and Tobit models.048 (120.or new entrant Age Age^2 Some college or associate degree (2) College degree (BA.138 (6.12 Table 2 reports the results of estimating Eq.113 (12.181)* x 1000 0.334 (16.703 (17.715)** 92.38 36.003)*** −0.013)*** Year dummy x 319.03 0.133 (12. MA or PhD) Female Female * partner Female * children Partner Children Weekend Constant Year and month dummies State dummies Observations R-squared A. The marginal effect of a Tobit model is dE(y|x)/ .089) Subsample (3): ineligible 25. Universe: Unemployed.161 (111. <= 26 49. One reason might be that residual wage dispersion is lower across the U.86 (11.362) 0.577) x 671 0. In the full sample the coef cient on the log of the maximum weekly bene t amount is negative but not statistically signi cant.652 (13.12 (3.30 0. we classify job losers and those on temporary layoff as eligible for UI. We exclude from the sample selfemployed and self-incorporated. and have been unemployed for 26 weeks or less (column 2).28 −6. Source for wage equation: CPS outgoing rotation group extract.490 (0.000)*** 0.016) 6. Errors are clustered at state-level.
715) −4.15 0.50 0.979) 303 Subsample (3): ineligible 25.28 5.780)*** −530. In the subsample of eligible unemployed with spells of 26 weeks or less (column 2).483) 30. Again.629)** 87. Table 4 reports the marginal effects of the log average weekly bene ts.329) 0.92).987) −218.036) −66.735 (334.asp.642)** 0.049 (23.054) −80. (2) We exclude the dummy for temporary layoff w/ expectation of recall for this regression.220) 2.216)* −119.342)* −111.256 −156.855 (53.023) 0. 14 The state average weekly bene t is de ned as bene ts paid for total unemployment divided by weeks compensated for total unemployment.329 (189.) is the standard normal cdf and.409 (57. Dependent variable: time allocated to job search.709)*** x 671 0.902) −175.329) −241.B.41 (20. Note that the reported elasticities are all calculated with respect to the legislated maximum weekly bene t amount.298)*** 10. Taking the IV estimates from column 2.845 (574. Krueger.645)** −223.408 (332. ** signi cant at 5%.S.653) −239.344 (42.8 (78. to make the Tobit estimates comparable to the linear regression models. See the notes in Table 2 for details about the estimated regression equation. because there are only 27 of them in the sample of ineligible (part-time workers in states were part-time workers are not eligible for UI) and they all have zero search on the diary day. *** signi cant at 1%.38 36.245 (0.gov/unemploy/content/data.1 0.421 (0.038 (37.601) 12.49 (0. Based on Krueger and Mueller 13 Note that the effect of dummy variables is different because of the non-linear nature of the Tobit model.28 0.825) −93.14 We instrument for the actual average bene t with the log maximum weekly bene t.895) 0.916)** −88.60 (0.419 (52.127)*** x 2171 0.642 (88.538 (82.572)*** −12.797 (503.06 −25.636 (73.212 (205.674)*** −24. we evaluate the adjustment factor at the mean values of x. Mueller / Journal of Public Economics 94 (2010) 298–307 Table 3 Tobit model regressions.685 (47.049)** 380.337 (95. . * signi cant at 10%.115 117.808 (572.881)*** x 1000 0.doleta. 1 that is predicted by the difference in bene t generosity and the bene t coef cients.874) −14.855 (31.259) −77.471) 201.230)** −456.04 Robust standard errors in parentheses.905) 0. A. Universe: Unemployed. <= 26 49.or new entrant Age Age^2 Some college or associate degree (3) College degree (BA. aged 20–65.087 (15.16 0. We take the average of the state average weekly bene t over the years 2003–07 from http://workforcesecurity.960)*** 261.04 Subsample (2): eligible w/o recall expectation and unempl. the implied elasticity is −2. Notes: Regressions are weighted using survey weights.506 (38.344 (46.579) 0.663 (40.857 (77.905)*** −1590.30 0.073) −38.953) −34.49 0.173) −53.A.18 (26. the coef cient on bene ts is signi cant at the 5% level and the implied elasticity is −0. in min per day Mean of dependent variable Adjustment factor for marginal effects Log(maximum weekly bene t amount) Fitted log(hourly wage) Std(residual of wage equation) — by state On temporary layoff w/ recall expectation (1) Jobleaver Re.851 (88.067) −437. dur.485 (60.917 (110.132 (113. To estimate the elasticity of job search with respect to actual UI bene ts.834 (24.146 (709.807) 548.167 (20.338 (59.173)** 652.1 0. (1) The base group consists of Job losers. and the 11 European countries shown in Fig.477) (2) 98.006 (46. dxi =βi Φ(xβ/σ) where Φ(.084)*** 264.653)*** −1062.484 (315.770) −44.787 (66. Errors are clustered at state-level.368 (81.892 (11.237 (15.571) 230. MA or PhD) Female Female * partner Female * children Partner Children Weekend Constant Sigma Year and month dummies Observations Pseudo R-squared 0.75 Mean (Std) Full sample (1) 32.03 0. To put our estimates in perspective. the contrast between the bene t effect for those eligible (column 2) and ineligible (column 3) is statistically signi cant.945 (25. the coef cient on bene ts is positive and not signi cant at conventional levels.29 0. which is not surprising given the high correlation between state average and state maximum bene t amounts (0.13 In the full sample in column 1. The difference between the OLS and IV estimates is small.271 (0.332 (52.277 (69.51 0. Column 2 reports the results for the subsample of eligible without an expectation of recall to their previous employer and with a duration of unemployment of 26 weeks or less.8.2 for the linear model and around −1.496 (648.859) 120. (3) The base group consists of those with a high school degree or less. Column 3 reports the results for the subsample of ineligible.283) 12.886) −269.194 (58. we estimated a linear and a Tobit model with the log of the state average weekly bene t in place of the maximum weekly bene t.153 24.89 (0.4 0.517)** 75.082) 801.6 for the Tobit model. we can calculate the differential search time between the U.902 (188.
bene ts are 0. this provides some indirect support for our identi cation strategy. the effect of bene ts on job search is predicted to be positive (the entitlement effect).S.2.. A. which is the after-tax value of UI bene ts. those with an expectation of recall hardly search at all. Universe: Unemployed.511 weekly bene t) (13. social assistance.312 (22.473)** 41. we used the number of job search methods used during the last 4 weeks as a dependent variable in our linear regressions of Table 2.4 12. (2008a). However. in the presence of borrowing constraints and. 3. Also. the regression results provide support for Mortensen's (1977) model to varying degrees. The coef cient on “on temporary layoff with recall expectation” in Tables 2 and 3 also shows that unemployed workers with an expectation of recall search signi cantly less than job losers. as the unemployed may take into account the option to retire already in their late 50s. in the absence of insurance markets for unemployment risk.16 Because of concern about simultaneous causation – a high unemployment rate could cause fewer people to search for a job and could be caused by low job search intensity – we excluded the unemployment rate from the models in Tables 2 and 3. ** signi cant at 5%.4 for the UI eligible in column 2 and 2. however. Differences across states in the level of bene ts have a negative relationship with job search in the subsample of UI eligible job seekers with unemployment duration of 26 weeks or less. *** signi cant at 1%.504) (58. Krueger. Mueller / Journal of Public Economics 94 (2010) 298–307 Table 4 Instrumental variables (IV) regressions. Although there are some obvious limitations of this calculation – such as the fact that we were not able to restrict the European sample to UI recipients – the results suggest that UI bene t generosity can potentially explain a nontrivial share of the difference in search behavior of the unemployed in the U.74 weekly bene t) (22. food stamps and housing bene ts relative to after-tax earnings. and Europe.. 1.489)** Observations 2171 671 28. however.612 −109. This highlights the utility of time use data for research on job search intensity.004) 1000 Robust standard errors in parentheses. we would like to mention that we control for state-level characteristics of the wage distribution and that our results are robust to the inclusion of the state-level unemployment rate.<=26 49. Indeed.1. * signi cant at 10%. providing too little variation to identify the effects of UI bene ts with any reasonable precision. e. Errors are clustered at statelevel.S. Unfortunately. We also excluded the duration of unemployment because it is endogenously determined with search time.731)** IV-2SLS (Instrument: log(maximum weekly bene t amount)) Log(state average −12. is warranted as our identi cation strategy relies on cross-state variation of maximum bene ts and omitted state-level covariates could lead to biases in our estimates. which had a negative coef cient but was not statistically signi cant. Those with access to a secondary income source are more likely to maintain consumption during a spell of unemployment 15 The bene t indicator they use is the net replacement rate. because Fig.562) −99.273)** 50. contrasting the coef cients on bene ts in columns 2 and 3 in Table 2 or 3). could therefore account for from 38% to 54% of the difference in search time.B. Notes: Regressions are weighted using survey weights. Overall. For these reasons. over the 5 years of the ATUS. Column 2 reports the results for the subsample of eligible without an expectation of recall to their previous employer and with a duration of unemployment of 26 weeks or less. we tested the robustness of the ndings in Tables 2 and 3 by including the state-level unemployment rate.304 A.S.05 methods for the ineligible in column 3. in min per day Mean of dependent variable OLS Log(state average weekly bene t) Full sample (1) 32. consistent with Katz's (1986) prediction. .15 The IV-Tobit estimate in column 2 of Table 4 therefore implies that job search time would be 9 min longer in the U.458 (11.44 methods used over the last 4 weeks. 16 See Shimer (2004) for an analysis of how search intensity varies with the business cycle.696 (42. The coef cients on the log weekly bene t remained of similar size and signi cance.008)** IV-Tobit (Instrument: log(maximum weekly bene t amount)) Log(state average 7.433)* Tobit Log(state average weekly bene t) 18. dur.004) 20.0 for the ineligible in column 3. compared to a decrease of 0. Column 3 reports the results for the subsample of ineligible.126) (39. The lower bene t levels in the U. 1 shows time spent on job search on weekdays only).649 (24. American job seekers search about 23 min more per day than European job seekers (this number is slightly lower than the differences shown in Fig.17 Both coef cient estimates. We also expect that the differential effect of UI bene ts on eligible and ineligible subjects is less likely due to state-level omitted variables.. However.S. It is nonetheless reassuring that none of the variables of interest had a qualitatively different effect if these variables were included. aged 20–65. marginal effect of log(average weekly bene t). a one log point increase in the weekly bene t is associated with a decrease of 0.1 Subsample (3): ineligible 25.564 (16. for the UI ineligible.004 (34.909 −77.583 (18. changes in maximum bene ts were small.109 (35. one might be concerned about endogeneity of our bene t variable as.e. UI also enables job seekers to smooth consumption and thus reduces the pressure for them to rush back to work. See the notes in Table 2 for details about the estimated regression equation. than in the 11 European countries over the rst 26 weeks of a spell of unemployment. We leave this task for the future when more years of the ATUS become available. Moreover. The average number of methods used over the last four weeks is 2. To evaluate the importance of such “liquidity effects” we follow Chetty (2008) and split the sample of UI eligible job seekers in column 2 of Table 2 into those with a working partner (married or unmarried) and those without. Dependent variable: time allocated to job search. we can reject at the 10% level the null hypothesis that the coef cient on bene ts is equal for the UI eligible and ineligible (i. The coef cient has the expected sign but is not signi cant. One word of caution. and the TwoStage Least Squares model predicts that it would be 13 min longer. Moral hazard versus liquidity effects of UI One way to interpret our ndings regarding the effects of UI bene t generosity is as a “moral hazard” effect: UI indirectly subsidizes leisure while unemployed and thus reduces the incentives to search for a new job and return to work. 17 The ATUS has the same categorical measures of job search as the CPS. The point estimates were consistent with our results above: For the UI eligible in column 2. we would prefer to identify the effects of UI bene ts on job search intensity from variation of bene ts across time rather than states. In results not presented here. states with high unemployment rates might enact more generous bene ts. We also probed the robustness of our results by excluding those older than 55.620)* −71.114 log points lower in the U.1 Subsample (2): eligible w/o recall expectation and unempl. Despite these limitations. were insigni cant with p-values in excess of 0. The average weekly bene t is de ned as bene ts paid for total unemployment divided by weeks compensated for total unemployment. more generally. Finally. other things equal.g.
with only 14% after 4 months or more. we derive unemployment duration by taking the unemployment duration reported in the last CPS interview and adding the number of weeks that elapsed between the CPS interview and the ATUS interview. Consequently. and then falls back to around 25 min. but the difference is not statistically signi cant at the 10% level. .29) for those with household income higher than $25. Krueger. as the imputed unemployment duration are quite noisy for those who become unemployed after their last CPS interview. a slightly smaller increase of time spent on job search before week 26 for the UI eligible (from 20 to 65 min). Lowess: job search by unemployment duration. To assess the validity of our assumption. The pro le of search time for the UI ineligible group was hardly affected in these simulations. We nd that the unemployed with low annual household income are more responsive to bene ts with an elasticity of −2. 19 About one third of our sample of unemployed individuals (excluding those on temporary layoff) has an unemployment duration of 14 weeks or more.8% of these individuals were employed in wave 2 and/or wave 3. Moreover. To nonparametrically estimate the unemployment duration–job search pro le we utilize LOWESS to plot the tted values of a locally weighted regression of minutes spent in job search on unemployment duration at the time of the ATUS. One problem with our measure of unemployment duration is that it does not take into account the possibility of job spells between the CPS and the ATUS interview. we impute duration of unemployment by taking half the number of weeks between the CPS and the ATUS interviews. We nd support for this hypothesis as the coef cient on bene ts for those with a working partner is positive and statistically insigni cant whereas the elasticity for those without a working partner is −2. The general patterns in the duration–search pro les are fairly 18 Note that STATA does not allow the use of survey weights for LOWESS. We do not show the weekly LOWESS plot for 13 weeks or less. We also would like to estimate the elasticity of job search with respect to increases in cash on hand. A. 4 provides LOWESS plots of the residuals. Relationship between unemployment duration and job search The standard search model makes strong predictions regarding the amount of time spent searching for a job by duration of unemployment. job search intensity should increase as bene ts approach the exhaustion date.B. job search increases sharply between weeks 15 and 26 of unemployment. Although it would be preferable to examine these relationships with longitudinal data. In particular.98). Mueller / Journal of Public Economics 94 (2010) 298–307 305 Fig. For each individual with a simulated interim job spell we subtracted 15 weeks from unemployment duration in the ATUS. we can use ATUS data to examine the cross-sectional patterns of job search across those with different durations of unemployment at the time of the survey.18 We exclude those who have an expectation of recall to their previous employer. 3 by removing the effects of age. consistent with standard search models. due to severance payments. and other characteristics (i. we performed simulations with our ATUS sample in which we randomly assigned job spells to 11.1 and signi cant at the 5% level (t-ratio 2.20 The unemployment duration–search pro le for the UI ineligible group is fairly at. 4.e.19 Fig.000. however.A. The large majority of the ATUS interviews were conducted 3 months after the last CPS interview. Fig. there is no such information currently available in the ATUS. the ATUS interview does not collect information on unemployment duration. for those eligible for bene ts. sex. and then used the residuals in the LOWESS analysis.. This generates a dataset representative of the population. 20 Note that we exclude observations on eligible individuals from 2003 because the federal extended bene ts program was in effect that year.8 (t-ratio 1.02).8% of individuals who were unemployed in the CPS as well as in the ATUS. the difference between the bene t coef cients in the two samples is statistically signi cant at the 10% level (t-ratio 1. To assess how this source of mismeasurement could affect our LOWESS plots. We iterated this procedure 400 times and found. 4.7 (t-ratio 1. on average. such as. We nd that 11. but simply report the average time allocated to search. For the UI eligible. search intensity by the ineligible should remain constant throughout the unemployment spell. the explanatory variables in column 1 of Table 2). We also split the UI eligible sample into those with annual household income below and above $25. For this reason. as their search behavior is different and affected by the recall strategy of the employer.000. Unfortunately. As a further robustness check. we duplicate each observation x number of times where x corresponds to the survey weight (with the “expand” command in STATA). these results suggest that liquidity constraints have a potentially important impact on many job seekers. For those who were not unemployed at the time of the CPS interview. from less than 20 min to greater than 70 min. and thus should be less responsive to unemployment bene ts.78) compared to −0. Fig. for example. Although not de nitive. 3 shows the LOWESS plot separately for those eligible and ineligible for UI bene ts. Future research with time use data might be able to distinguish the liquidity effect from the moral hazard effect. 3. as the search intensity of those who have less access to nancial resources appears to respond more strongly to UI bene ts. By contrast. Lowess: job search by unemployment duration. we probed the robustness of the pro les in Fig. we matched the CPS waves 1 to 4 over the years 2003 to 2007 and looked at individuals who were unemployed and without an expectation of recall both in wave 1 and wave 4 three months later. Unfortunately.
g.: Contacting employer Making phone calls to prospective employer Sending out resumes Asking former employers to provide references Auditioning for acting role (non-volunteer) Auditioning for band/symphony (non-volunteer) Placing/answering ads Researching details about a job Filling out job application Asking about job openings Reading ads in paper/on Internet Checking vacancies Researching an employer Submitting applications Writing/updating resume Meeting with headhunter/temp agency Picking up job application Interviewing (050403). to job search/interviewing (050405) Travel related to job search (180504) Job search activities. Phil Levine. although the increase in time devoted to job search between weeks 15 and 26 for the UI eligible sample is somewhat smaller after removing the effects of the explanatory variables. if future ATUS surveys collect data on unemployment duration. however. and seminar participants at Princeton.g.. is that search effort appears to decline after week 26..B. and rationally feel they have little to gain from maintaining the same level of search effort over the next few months. we could not reject either null hypothesis). One possible explanation is that the unemployed become discouraged if they fail to nd a job despite substantially increasing their search effort before UI bene ts run out at 26 weeks. data on severance payments and asset positions of the unemployed could allow one to determine the relative importance of moral hazard and liquidity effects of unemployment bene ts. One explanation for the decline after week 26 is a potential selection issue due to unobserved heterogeneity in the propensity to search for a job: job seekers who devote a lot of effort to searching for a job are more likely to nd one and exit the sample. because they do not have a working spouse) tend to respond more to UI bene ts than do those with greater nancial wherewithal. Katz and Meyer. e. Summary and conclusion This paper provides new evidence on job search intensity and Unemployment Insurance. These ndings highlight the utility of simple search models for understanding job search behavior and UI. e.b) and the reservation wage (e. however. Mueller / Journal of Public Economics 94 (2010) 298–307 similar to those in Fig. Moreover. A similar issue affects studies of the effect of UI on exit rates (e. e.22 A nding that is inconsistent with Mortensen's (1977) search model. In particular. Alan Krueger was the Leon Levy member of the Institute for Advanced Study at Princeton when the paper was written. job seekers who likely have less access to nancial resources (e. The increase in job search in the weeks prior to bene t exhaustion for the UI eligible sample and the fairly constant amount of time devoted to job search for the UI ineligible are both consistent with Mortensen's (1977) search model. not elsewhere speci ed (050499) . 22 See also the nonparametric Monte Carlo technique in the working paper version (Krueger and Mueller. the linear and quadratic terms were jointly signi cant at the 10% level and the predicted patterns of job search by unemployment duration looked similar to the LOWESS in Fig.2. consistent with theoretical models. Andreas Mueller gratefully acknowledges nancial support from the Handelsbanken's Research Foundations. For the UI ineligible. This creates a possible “lengthbased sampling” bias that would tend to cause the search pro les to slope down with unemployment duration. in line with Mortensen's (1977) model. Our ndings suggest that time use data offer a fruitful approach for research on job search intensity. whereas those with a low proclivity to search remain in the sample. 1990a.g. 1984). Krueger. consistent with a role for liquidity constraints. 4: job search increases by 24 min between weeks 14 and 26 and then strongly decreases by 66 min as week 39 approaches.g. Appendix A 5.6 to −2. The fact that the relationship between spell duration and job search is fairly at for the UI ineligible sample is an indication that bias due to length-based sampling is probably small.: Interviewing by phone or in person Scheduling/canceling interview (for self) Preparing for interview 21 In order to be consistent with the sample used for the LOWESS.S.: Waiting associated with job search interview (050404) Security procedures rel.21 For the UI eligible.306 A. This nding deserves further attention. 3.g. Job search activities (050401). with an elasticity of −1. one could further investigate the link between unemployment duration and job search. we nd that job search increases sharply in the weeks prior to bene t exhaustion. Longitudinal time use data would help to control for length-based sampling and individual heterogeneity in job search activity. A. Other activities related to job search. Furthermore. Finally. which is large enough to account for much of the gap in job search time between the U. However. We use data from the American Time Use Survey and model job search intensity as time allocated to job search activities. which suggests that the relationship between job search effort and the duration of unemployment for a cross-section of job seekers is only slightly biased by length-based sampling.e. Acknowledgements We have bene ted from helpful discussions with Larry Katz. Feldstein and Poterba. for both the UI eligible and ineligible. Moreover. and we couldn't statistically distinguish their pattern of job search from a constant nor from the pattern of the UI eligible (i. we introduced quadratic polynomials for duration of unemployment with breaks at weeks 14 and 39 into the linear regression model in column 1 of Table 2. as this group would also be subject to length-based sample bias if workers have heterogeneous commitments to job search. the standard errors on the coef cients for the linear and quadratic terms were large. Table 1 De nition and examples of job search activities in ATUS 2006. the NBER. rather than remain constant. the decline in job search after week 26 is unexpected. We nd that time allocated to job search is inversely related to the maximum weekly bene t amount for UI eligible workers. which analyze one unemployment spell per person or reservation wages for a cross-section of job seekers. and Bruce Meyer. A related explanation is that the unemployed may feel that they have explored all of their plausible job opportunities after they sharply raised their search effort in the weeks leading up to the exhaustion of their UI bene ts. and Europe. and consequently stop searching. when bene ts run out. we excluded those on temporary layoff with an expectation of recall and observations on UI eligible individuals from 2003. the University of Lausanne and the 2nd Nordic Summer Symposium in Macroeconomics. 2008b). as the model predicts that workers allocate a constant amount of time to job search after bene ts are exhausted..g.. Per Krusell.
Krueger. Industrial and Labor Relations Review 47 (1). Unemployment insurance and reservation wages.. The Review of Economics and Statistics 64 (1).B. 937–957. R. Journal of Political Economy 116 (2). Meyer.F. Temporary layoffs in the theory of unemployment. M.J.).. Information in the labor market. Krueger. Weber. Unemployment insurance and job search decisions.B. Cambridge. 45–72. 143–160.D. Mortensen. 173–234.B. 2008a.. 1990b... Krueger. Feldstein. D. (Eds. Journal of Political Economy 70 (5. Flinn. 1976. 2001. Sargent.. National Bureau of Economic Research. NBER Working Paper 1825.J. Amsterdam. Nicholson. Institute for the Study of Labor (IZA).. B. Bonn. 1979. A. 1990a.. A. Feldstein.T. Katz. Krueger. Quarterly Journal of Economics 105 (4). 73–86.. A. 389–404.S. Chicago. mimeo.F.. Meyer. R. recall and the duration of unemployment... The effect of unemployment insurance on unemployment: the case of federal supplemental bene ts. Mueller. OECD.B. J. W. 2004. A. 2007. Search effort in the labor market.D. R. P. Handbook of Public Economics. Card. R...S. 505–517. Paris. 4.. NorthHolland. 1993. A.. Mueller. and unemployment outcomes.. Mueller / Journal of Public Economics 94 (2010) 298–307 307 References Anderson. Unemployment insurance and the distribution of unemployment spells. 1962. Moral hazard versus liquidity and optimal unemployment insurance. Ljungqvist. Unemployment insurance.. University of Chicago. Germany. IZA Discussion Paper 3490.. Katz. Job search and unemployment insurance: new evidence from time use data. Welfare states and unemployment. 2327–2392.D. 2002. 2008b. M. D. 1983. Mof tt. Chetty. Spillover effects between the insured and uninsured unemployed. Poterba. C.. Stigler. Mof tt. 113–118. Meyer. Heckman. . B.. 85–101.. IZA Discussion Paper 3667.. Mellow.. Journal of Public Economics 23 (1–2). W. Germany. In: Alan Auerbach. vol. Economic Theory 6 (1). 1–11. A. 1977. 1995. Labor supply effects of social insurance. Monitoring and assisting active job search.. Journal of Public Economics 41 (1).F.J.. L. The impact of the potential duration of unemployment bene ts on the duration of unemployment.A. 28–42. J. 1984. J.J. Industrial and Labor Relations Review 30 (4). P. 1982. pp. L. Levine.. Edition 1. Bonn. Chetty. 94–105. 1986. T. 2008. A.M. The lot of the unemployed: a time use perspective. IL.M. The spike at bene t exhaustion: leaving the unemployment system or starting a new job? American Economic Review: Papers and Proceedings 97 (2). Shimer.. R. B. L. OECD Proceedings: Labour Market Policies and the Public Employment Service. M. Institute for the Study of Labor (IZA). Journal of Human Resources 14 (3).. Layoffs. MA. Journal of Econometrics 28 (1). Are unemployment and out of the labor force behaviorally distinct states? Journal of Labor Economics 1 (1). Barron. Katz. recall expectations. 1985. Feldstein. 141–167.S... The Journal of Political Economy 84 (5)..B. 973–1002. Part 2). G.J. A. L.. Search Intensity..
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.