Extended Benefits And The Duration Of Ui Spells: Evidence


Journal of Public Economics 107138 www​.else​vier​.nl /​locate /​econbase Extended beneï ts and the duration of UI spells: evidence from the New Jersey extended beneï t program a,c b,c , * David Card , Phillip Levine a University of California at Berkeley ‚ Berkeley, CA, USA b Department of Economics ‚ Wellesley College, Wellesley, USA c National Bureau of Economic Research ‚ Cambridge, MA, USA Abstract This paper examines the impact on the duration of unemployment insurance receipt of apolitically motivated program that offered 1996. Using state-​level data and individual administrative records from before, during andafter the program, we ïnd that it raised the fraction of claimants who exhausted theirregular beneïts by 13 percentage points. Had the program run long enough to affectclaimants from the ïrst day of their spell, the fraction exhausting would have risen by 7 percentage points, and the average recipient would have collected regular beneïts for oneextra week. © 2000 Elsevier Science S.A. All rights reserved. Keywords : Unemployment insurance; Spell duration; Extended beneïts JEL classiïcation : ; 1. Introduction One of the key factors that may explain some of the signiï cant gap between European and American unemployment rates is the relative generosity of un–employment beneï ts. Although beneït levels tend to be somewhat higher in Europe, there is amuch larger difference in the maximum duration of unemploy–ment beneïts. In the United States unemployment insurance ( E-​mail address : [email protected]​wellesley.​edu (P Levine 0047 – 2727⁄00 /​$ see front maer © 2000 Elsevier Science S.A. All rights reserved. PI I: S9 138 were more likely to have full information regarding the programs existence to seeif their response is larger than that observed from other workers. 4. Empirical strategy and description of the data The legislative history of the NJEB program makes it clear that the beneï t extension was unrelated to changes in business cycle conditions in the state. Its introduction and ending create conditions that are ideal for aquasi-​experimentalanalysis of the effect of maximum beneït durations on the behavior of UI claimants. In fac the short-​term nature of the policy provides two opportunities toexamine the impact of higher beneït durations: one as the NJEB program began; and another when the program ended. Any effect measured at the onset of the program should dissipate at its expiration. We use two different sources of data to evaluate the effects of the the District of Columbia from January 12 Department of Labor. These state-​level data contain information on the number of initial UI claims and ï rst payments, the fraction of claimants that exhaust theirregular beneïts, and the level of covered employment in each month over this period. These data allow us to determine whether the rate of regular beneït exhaustion (deïned as the number of exhaustions divided by the 6-​month lag of ïrstpayments) increased and then returned to its previous level, during and after the 13 period in which New Jerseys extended beneï ts were available. We test for the presence of such apaern in three ways: (1) by comparing exhaustion rates in New Jersey over time; (2) by comparing exhaustion rates in New Jersey with rates in neighboring Pennsylvania; and (3) by comparing exhaustion rates in New Jersey with those in the rest of the country. Our second source of data is administrative records from New Jerseys UI system for all initial claims ï led between January of Some 1.3 million claims were ïled over this period, with ïrst payments made to 12 At the time we obtained these data in the Spring of 1998, we could only create exhaustion rates for initial claims ï led to October of 1997. Claims ï led later than that had not hit their 6-​month potential maximum duration ye Therefore, we have no data for November of 1997 which would have been useful for comparison purposes with the 13 Another possible outcome from extending the maximum duration of beneï t receipt is that individuals who would not have applied for regular UI beneïts may choose to apply. Such behavior inresponse to New Jerseys beneï t extension, however, is unlikely because the extended beneïts wereonly available for about 6 months; few individuals could have ïled anew claim and then exhaustedtheir regular beneï ts before the extension expired. Nevertheless, in preliminary data analyses, we used the aregate, state-​level data to test whether initial claims as ashare of covered employment was affected by NJEB and found no statistically signiï cant relationship. NJEB intervention therefore affected different individuals differently, depending on how many weeksthey had been on UI at the announcement of the program. Such atime-​varying intervention is most easily modeled in the context of aconventional hazard model. Asecond use of the individual claims data is to examine the effect of the ts. To the extent that this spike reïects abehavioral response to theimpending cut-​off in beneïts, one might expect asmaller spike among claimants who were eligible for NJEB than among claimants in the comparison group. (Although, as noted earlier, Meyer (1997 for New Jersey, Pennsylvania, and the entire United States. One obvious difference across these geographic entities is that the average exhaustion rate is higher in rst payments hovers around for the country as awhole. Nevertheless, movements in exhaustion rates tend to 19 follow each other rather closely. Beginning in June of 1996, however, New Jerseys exhaustion rate began increasing slightly, while rates elsewhere drifted down. The New Jersey rate stood at about 1996 for theïrst time in over ayear. No such trend appears in Pennsylvania or in the national data. This relative upward trend is consistent with the expected effect of the NJEB program. In particular, one would expect the availability of le aclaim after the June NJEB became effective and still exhaust their regular beneï ts by November 24 of that year. For instance, an individual ïling aclaim onJune for whom this is true is very small. 18 We graph 3-​month backward-​looking moving averages because the month-​to-​month variation in exhaustion rates is considerable, possibly overshadowing other paerns. Use of amoving average means that any policy effect will not be observed as adiscrete break in the trend, but will be more gradual. 19 Several commentators have noted that the higher average rate of beneï t exhaustion in New Jersey than in Pennsylvania suests that the laer may not be agood control for analyzing the effect of NJEB. An obvious alternative is New York. However, there is anotable outlier in the exhaustion series for New York in July 1996 that makes it an unaractive choice. Other states with average exhaustion rates comparable to New Jersey are Washinon DC, Montana, North Dakota and Rhode Island. Rather than use these states, we decided to use all the US as an alternative control. D . Card, PLevine /​Journal of Public Economics ts in New Jersey, Pennsylvania and UI for longer. Such behavior would lead to agradual rise in the exhaustion rate, with aplateau after 26 weeks, as all those who became eligiblefor NJEB while on UI eventually exhaus Given the short timeframe of the NJEB program, one would therefore expect amonotonically rising effect throughout the June November 1996 period. In the months after the beneït extension ended,exhaustions fell considerably in New Jersey, although asmall decline is also observed in the US as awhole. Simple estimates of the impact of NJEB can be obtained by computing the change in exhaustion rates in New Jersey relative to the change in other states asNJEB beneïts turn on and turn off. Such estimates are reported in Table 1. Theïrst three columns of this table present exhaustion rates in New Jersey, Pennsylvania, and the US (excluding New Jersey) for the July November periods 1997. We use a July November window rather than a June 20 Mortensen (1977) uses asimple search model to derive the predicted effects of longer beneï t availability on search behavior of unemployed workers, and on the exit rates off UI. 21 We seasonally adjusted these data using the deviation of the monthly average exhaustion rate within each state over the 19851997 period from the overall average within the state. Nevertheless, we use the same 5-​month period in the preceding year because it is possible that the beneït extensionmay have affected spell lenhs earlier in 1996 in anticipation of the change as the policy was being debated. For consistency, we report the same period in 1997, although our data only runs to October of that year as indicated earlier. D . Card, PLevine /​Journal of Public Economics 4 presents the actual distribution of weeks of regular UI receipt for those recipients eligible for 26 weeks of beneïts in each of the three sample periods. Theproportion of recipients who exhausted their regular beneïts is somewhat smallerhere than reported in Table includes recipients eligible for fewer than 26 weeks of beneïts, who are morelikely to exhaus Column ( the averages for 1.5 percentage point increase in the share of spells that exhausted their regularbeneïts in share of recipients ïnding jobs in weeks to 1995 and 1997, indicating that some individuals may have shifted their jobïnding behavior to take advantage of the extended beneï ts. Again the relative Table 4 Distribution of regular ts, by potential date of aexhaustion No. of weeks July 199621995⁄1997 average ts. 26 weeks, while in many European countries themaximum duration of unemployment beneïts is measured in years. Conventional economic models suest that the availability of longer UI beneïts providesincentives for individuals to remain unemployed longer, contributing to the 1 problems of high unemployment and long-​term joblessness. In fac existing research in the United States ï nds astrong positive relationship between the maximum duration of beneï ts and the lenh of an individuals spell 2 of unemployment beneïts. Empirical identiïcation in this body of work is provided by differences in the maximum duration that occur across states and overtime. Apotential difï culty with this identiïcation strategy is that states may decide to offer longer UI beneï t durations during recessions, in response to low 3 rates of job-ï nding that cause more individuals to exhaust their beneïts. Suchendogenous policy formation may lead to an overstatement of the effect of longerUI beneïts on the duration of UI spells. In this paper we use the experiences generated by aunique legislative episode in the state of New Jersey that led to the adoption of extended unemployment 4 beneï ts for a had been using funds from its Unemployment Insurance Trust Fund to ïnance theindigent care costs of hospitals in the state. In the Spring of 1996, opponents ofthis ïnancing method blocked its re-​authorization, precipitating alegislative crisis. In adeal struck to gain the support of labor organizations, alaw passed in May of 1996 included the provision of up to 13 weeks of extended beneïts for workerswho exhausted their regular UI beneïts hose to which they are normallyentitled typically 26 weeks These beneïts were available retrospectively toclaimants whose beneï ts had expired as long ago as December 1995, and prospectively to claimants who exhausted their regular 1996. This policy change provides two important advantages for astudy of the effect of maximum beneï t durations on the lenh of unemployment spells. Firs itslegislative history makes the beneït extension unrelated to changes in the condition of the states labor marke New Jerseys economy remained robust 1999) for discussions of long-​term unemployment in Europe and the contribution of unemployment beneï ts to this phenomenon. 2 See, for example, Mofï and Nicholson (1982 Mofï ( nance question regarding the impact of maximum beneïtduration on compensated unemployment spell lenhs, not on total unemploymen due to data availability. For similar reasons, the research presented in this paper also focuses on this narrower question. If one had access to similarly complete micro-​data on unemployment spells, an analysis of these hypothetical data would beer address differences in total unemployment rates across countries. 3 Indeed, the federally funded extended beneï t program is automatically triered when insured unemployment rates reach acertain threshold. The fact that beneïts are typically extended during arecession would not bias the results if the econometric model controlled for the determinants of the extension (like the insured unemployment rate 4 Meyer (1992) undertakes asimilar case-​study approach to examine the impact of an increase in UI beneï t levels. D . Card, PLevine /​Journal of Public Economics th week, this gap has risen to 4 percentagepoints. In interpreting this apparent twist in the 1997 rates, it is important to keep in mind that most claim spells in our 1996 sample were in progress when the NJEB program was announced. Indeed, among the subset of the 1996 sample eligible for 26 weeks of regular beneïts, the medianclaimant would have been in his /​her 13th week of not left search intensity, one would expect to see agradual downward shift in the average 1996 hazard from earlier claim weeks (which mostly occurred before the NJEB program was announced) to later claim weeks (which were increasingly likely to have occurred after the program was announced Evidence from the hazard models presented below suests that this is indeed areasonable description of the programs effec Before turning to the hazard models, however, we present avariety of simpler probit and censored normal regression (Tobit-​style) models for the determinants of the lenh of completed regular models, which are presented in Table 5, can be interpreted as models for the latent duration of UI claim spells. Speciï cally, let ydenote the amount of time an i D . Card , P . Levine /​Journal ofPublic Economics 100, standard errors in parentheses) Recipients eligible for 26 weeks of beneï ts, All recipients: All recipients: probit models for spell lasting at least: probit for exhausting censored regression regular ts 010) Other individual characteristics YY YY YY YMonth of initial claim ïxed effects YY YY YY YMajor industry ïxed effects YY YY YY YNumber of observations 308 a Other individual characteristics include age, race, gender education, union status, citizenship, weekly wage prior to job loss and weeks worked for former employer. See Card and Levine (1998) for parameter estimates on these variables. Sample restricted to those with potential exhaustion dates between July 1 and November ts. D . Card, PLevine /​Journal of Public Economics UI leaving behavior will be measured bythe time-​varying post-​NJEB coefïcien Our hazard model estimates are presented in Table scheduled exhaustion dates from July This sample of nal payment weeks (weeks in which claimants exhaust their UI 35 entitlemen which are treated as right-​censored observations. The risk set for our hazard analysis therefore contains paern of the hazards shown in Fi 4, we include avariety of controls for thebaseline exit probabilities: dummies for the ïrst cubic in the number of elapsed weeks of regular t exhaustion. We also experimented with avariety of other controls, including linear and quadratic terms for the number of weeks remaining until exhaustion. The addition of such terms had essentially no effect on the estimates of the NJEB program impacts nor of theeffects of the other covariates. The speciïcation in column (1) includes asingle dummy variable for 1996 claims, along with the same set of individual covariates used in the models in Table 5. The estimate of the 1996 effect is negative, but small, and statisticallyinsigniïcan The effects of the control variables are typically signiïcan and consistent with the signs of the coefïcients of the models in Table 5. The speciï cation in column ( indicating a sample of indicating a place. The paern of these estimates provides asimple interpretation of the average hazards graphed in Fi the positive coefïcient for UI-​leaving rates in 1996 were slightly higher than those in the 1995⁄1997 sample. On average, the earlier weeks in the and the overall hazard rate was above the average 1995⁄1997 rate (as shown in Fi 4 leading to somewhat fewer spells lasting longer than shown by the probit models in Table adrop in UI leaving rates, causing agradual drop in the average hazard among 35 For individuals who contributed two or more claims to our sample, we included only the ï rst claim. This eliminated about 2% of all claims. 36 We also estimated models on the subset of individuals eligible for 26 weeks of regular beneï ts that included dummies for each individual week of regular UI receip Such aspeciïcation would becomparable to the semi-​parametric proportional hazards model estimated by Meyer (1990 Again, theestimates of the key coefïcients are similar. D . Card, PLevine /​Journal of Public Economics cation in column (2) with an interaction between thepost-​NJEB dummy and aquadratic in the elapsed spell duration. The resultinginteractions are at best marginally signiïcan and show only asmall increase in the NJEB effect with elapsed duration. We also tried an ad hoc re-​weighting scheme to evaluate the average effect of NJEB if the distribution of weeks at risk for the NJEB treatment was representative of the overall distribution of weeks atrisk to exit UI. Speciïcally, for each person-​week at risk to leave UI in the post-​NJEB 1996 sample, we weighted the observation by the ratio of the relative number of person-​weeks of the same elapsed duration in the 1995⁄1997 com– parison sample to the relative number in the post-​NJEB 1996 sample. We then ïhe duration model by weighted logi The resulting estimate of the post-​NJEBcoefïcient was results we conclude that any effects of heterogeneity in the NJEB effect are small. As indicated earlier, another issue regarding the NJEB program is that some UI recipients may have been unaware of their eligibility for the program. To address this, we replicated our analysis on two subgroups of workers that we suspect were relatively well-​informed about the program: union members (whose leaders lobbied for the extension and workers in the construction industry (who are much more likely to be repeat users of Estimation results for these groups are shown in columns ( Interestingly, the estimated effects of NJEB for these subgroups are quite similar to those obtained for the overall sample. In particular, the announcement of NJEB seems to have lowered exit rates by about 20% for both groups, with lile indication of any effect on the size of the pre-​exhaustion spike. 6. Discussion and conclusions Taken as awhole the results of our analysis provide two alternative views of the effect of the 1996 beneï t extension in New Jersey. Overall, the NJEB program appears to have had avery modest impact on UI claim behavior in the state. Thefraction of recipients who exhausted regular beneïts increased by about percentage points and the average spell lenh was largely unchanged [Table 5, columns ( affected [Table program impact was due to the short-​term nature of the program. Many NJEB– eligible recipients spent alarge share of their unemployment spell looking for work before that in our comparison sample of the impact of NJEB on weeks of claim recipiency after the program wasimplemented, we ïnd that UI-​leaving rates declined signiï cantly. Our estimates D . Card, PLevine /​Journal of Public Economics 109 throughout the period, with overall unemployment rates drifting down at about thesame rate as in nearby states. Second, the short-​term nature of the New JerseyExtended Beneït (NJEB) program allows us to compare unemployment spell durations and other outcomes during the program period with comparable data from immediately before and immediately after the NJEB interval. We use two complementary sources of data for our analysis. We begin by studying aregated monthly data for New Jersey and other states on the fraction of techniques provide two estimates of the effect of the NJEB program: one effect when the program turned on; and asecond when the program turned off. Our second data source is administrative claim records from the state of New Jersey from these records to compare regular UI spell durations in the program period to spell durations before and after. An important feature of the NJEB program is thatalmost all potential recipients of extended beneïts had begun their UI spells beforethe beneït extension was announced. Standard hazard-​modeling techniques allow us to compare rates of leaving UI before and after the announcement of the NJEBprogram among these ongoing spells. Our ïndings suest that the NJEB program, as enacted, had avery modest effect on overall UI claim characteristics. Our aregate and micro-​level estimates indicate a13 percentage point increase in the fraction of claimants who exhausted their regular UI eligibility. The impact of the policy, however, appears to have been substantially moderated by its short-​term nature. Many recipients were well into their unemployment spell at the time the extension was im– plemented and had lile opportunity to alter their behavior. Our hazard models suest that the regular UI-​leaving rate declined substantially (by about 15 following the programs introduction. Simulations of the long-​term effect of abeneït extension similar to the NJEB program indicate that the availability of 13extra weeks of beneïts would raise the fraction of claimants who exhaust regularUI beneïts by 7 percentage points, and would raise the average duration of regular 2. Review of the literature Previous research on the effect of maximum UI eligibility on claim durations has used data from the United States (cf. Mofï and Nicholson, 1995 and Austria (Winter-​Ebmer, 1998 These studies have generally found thatan increase in the maximum duration of beneïts leads to an increase in average UI spell durations. As in other policy evaluation research, an important issue in all of these studies is the potential endogeneity of maximum beneït durations tounobserved conditions in the labor market that also contribute to longer (or shorter) UI spells. D . Card, PLevine /​Journal of Public Economics estimates of the parison between estimate (from the comparison between average of these estimates is comparing New Jersey in comparable difference for Pennsylvania. Finally, athird alternative is to compare New Jersey to all other states in the estimate when NJEB program may have raised exhaustion rates in the state in the July November 1996 period by something like 14 percentage points, although none of the estimates is statistically different from zero. Interestingly, there is no indication from Table 1 that asimple within New Jersey comparison [as in column ( program effect than adifference of differences comparison with either Penn– sylvania or the rest of the US. Unfortunately, however, the standard errors for the estimated impacts are so large that we cannot rule out an effect of 0, or one as large as 68 percentage points. In an effort to improve the precision of the impact estimates in Table 1, we ï t aseries of regression models using monthly exhaustion rates for July November for all the states from 1985 to 1997. These models include afull set of state and yearïxed effects that absorb permanent differences in exhaustion rates across states, as 22 well as any aregate shocks that affect all states in agiven year. Five of the models are reported in Table 2. The ïrst speciï cation includes only asingledummy for New Jersey observations from 1996. This model provides avalid impact estimate under the assumption that exhaustion rates in New Jersey would move in parallel with the average changes in other states in the absence of NJEB. The estimated impac differences estimate for New Jersey relative to the Column (2) includes asecond dummy variable for New Jersey data in 19951997. With this dummy included, the from the average of the averaged difference-​in-​difference estimate in Table 1. This change in spe–ciïcation has lile effec Columns (35) present models that include the state unemployment rate as acontrol variable for cyclical conditions in the labor marke This variable is strongly correlated with exhaustion rates, and its addition signiïcantly reduces thestandard error of the regression models, albeit at the cost of some potential endogeneity bias. Controlling for state unemploymen the estimated impact of 22 We use seasonally adjusted exhaustion rates, so we do not include month effects. 2 a Estimated models for monthly state exhaustion rates (July November only) (3 a Standard errors in parentheses. Estimated on sample of unavailable The dependent variable is the seasonally adjusted monthly state exhaustion rate (in percentages Models include unrestricted state and year effects. Monthly trend variable is normalized to have mean 1996 New Jersey dummy) is somewhat sensitive to the inclusion of the 19951997 dummy, although the estimates are still quite imprecise. Finally, in column (5) we include amonthly trend variable that increases linearly over the July November 1996 period. or ease of interpretation this trend variable has a mean of 0 This term allows us to test for any systematic trend in the relative New Jersey exhaustion rate during estimated trend is positive, although very imprecisely estimated. Overall, the results in Tables 1 and 2 suest that there was amodest increase in exhaustion rates in New Jersey during the period that UI claimants were eligiblefor extended beneïts of the order of 13 percentage points. However, given the rather large month-​to-​month variability in state-​level exhaustion rates, we cannot rule out an effect of 2. Analysis of administrative records We turn now to amore detailed analysis of individual throughout this analysis is that claimants who were scheduled to exhaust in the July November period in 1996 were comparable to those claimants in apooled 1995⁄1997 sample from the same months with the exception of NJEB eligibility. Weak evidence in favor of this hypothesis is provided by the similarity of the impact estimate in Table 1 that uses only New Jersey data [i.e. the estimate in the boom row of column (1 to estimates that use either Pennsylvania or other US states as acomparison group. Some further evidence on the validity of the 1995⁄1997 pooled sample as a variety of descriptive statistics for for the hypothesis that the rst row presents county level unemployment rates (at the ï rst payment date for each claim By this measure, economic conditions were fairly 1997. Other than this change, the characteristics of New Jersey UI claimants were fairly stable over our sample period. Nevertheless, the large samples provide very precise estimates, so many ofthese small differences are statistically signiïcan as indicated by the t-​statistics in 25 the ï fth column of the table. The boom panel of Table 3 displays UI claim characteristics over the three periods. Just over two-​thirds of recipients are eligible for the full 26 weeks ofregular beneïts in each year. The percentage of recipients that exhausted theirregular beneïts in each period is very similar to the aregate exhaustion rates reported in Table 1, indicating that the bias introduced in the aregated data by using apotentially mis-​measured denominator is small. Acomparison of the 1996 rate to the average of 1995 and 1997 indicates that the percentage of respondentsthat exhausted their regular beneïts climbed difference may be aributable to the availability of extended beneïts or,alternatively, to the relatively higher rate of unemployment in 1996 compared to the 1995⁄1997 average. Almost one-​third of respondents in the ts. 23 Ninety-​four percent of claims that were scheduled to exhaust in the period June November, 1996, were ï led before June composition of the claims sample was directly affected by NJEB. 24 The average county unemployment rates are higher than the state averages in Fi 1 because the administrative records sample over-​weights counties with higher unemploymen Similar weighting issues may also explain the fact that the state average unemployment rate is slightly higher in 1995 than 1996, but no difference is observed here. Although one would expect the sample sizes in 1995 and 1996 to be larger than that for 1997 based on the unemployment rates across years, the number of observations available for 1995 is diminished somewhat by the sample design. Some individuals ïlingan initial claim towards the end of 1994 might not receive aï rst payment until sometime in 1995 if there was some question regarding his /​her eligibility for beneïts and could have potential exhaustiondates between July and November of that year. Because our sample consists of initial claims ïled on orafter January 25 The standard errors reported here, and in the remainder of our analysis, are subject to some understatement because they do not incorporate the correlation that may exist across observations within aparticular geographic region at apoint in time. In our analysis of state-​level data, we estimated comparable models to those reported in Table 2 but controlled for the correlation across monthly observations within states and years and found no substantive changes in statistical inference. For instance, using the cluster option in Stata to estimate the model in Table areduction in the estimated standard from micro-​data later in the paper. NJEB were eligible by virtue of having been temporarily disqualiïed for beneïts during their unemployment spell, thus extending their exhaustion date beyond the potential exhaustion date we have calculated. 138 frequencies are precisely estimated, and many of the differences are statisticallysigniïcan An alternative way to organize the same data is to construct the hazard rates out of ts before, durin and after 5. As in other administrative data bases, the New Jersey sample shows anotable spike in UI-​leaving rates just prior to regular beneï t 28 29 exhaustion. Somewhat to our surprise, however, afairly similar spike is also apparent in 1996, when NJEB was in effec The traditional interpretation of the rapid rise in 30 until the last minute to begin anew job (or begin searching for anew job On this basis, one might expect to see amuch smaller spike at were available. The presence of such astrong spike in our 1996 sample suests that the rise in UI-​leaving rates at week due in part to factors other than the strategic timing of job starting dates. Aclose examination of the hazard rates in Fi the 13th week. Similarly, although the survivor function for 1996 claims is parallel to the function for 1997 for the ï rst 1012 weeks, after that point the two functions begin to diverge. After 27 An alternative approach that would utilize all spells would be to create hazard rates by weeks until exhaustion, which should show aspike within the few weeks prior to the regular beneït cut-​off. Weexperimented with these hazards and found similar paerns to those reported here. We chose to report hazard rates by weeks unemployed among asample of recipients eligible for auniform 26 weeks because it is easier to interpre 28 Because of data limitations, the actual spike is probably somewhat more muted than that presented here. All the statistics reported in this paper refer to full weeks of beneït receip Yet the administrativedata from which our statistics are derived enumerate calendar weeks, in which any beneït receivedduring the week is counted. Although we can largely correct for this distinction, we cannot identifythose recipients whose ïrst calendar week of beneï ts was apartial week. Therefore, for some recipientsour count of full weeks of beneït receipt is overstated by one. If, for example, aclaimant became unemployed in the middle of aweek and started anew job on a Monday, the measured number ofcalendar weeks of beneïts received will be one higher than the number of full weeks and we are unable to correct for this. Individuals who are coded in our data as leaving UI in the week just prior to beneïtexhaustion may have actually collected only 24 weeks of full regular beneïts. This problem leads tosome overstatement of the pre-​exhaustion exit spike. We are unsure whether asimilar issue may be present in earlier data sets. ects the fact that recipients are paid the weekly beneï t for the initial waiting week if their spell extends to 4 weeks or longer. Therefore, the marginal beneït ofremaining unemployed from the third to the fourth week is really 2 weeks worth of beneïts, not 1.Standard search models would predict such an incentive would lower the hazard rate in the third weekand this prediction appears to be veriïed in the data. 1977 an optimal job search strategy in the presence of limited duration beneï ts will lead to arising exit rate as exhaustion draws near. le shifted down by about 17% in each weekfollowing the onset of the extended beneït program [Table 6, column (2 We used this estimate to simulate the long-​run effect of a13-​week beneï t extension on apool of unemployed workers who were eligible for 26 weeks ofregular beneïts and knew from the start of their spell that they could receiveextended beneïts. Starting with the sample of population, we calculated claim survivor functions assuming that the weekly hazard rates were 16.6% lower than the observed rates. The results of the simulation suest that the long run effect of a and aroughly 1 week increase in the average number of weeks of regular UI collected by claimants. The laer estimate of the sensitivity of weeks of regular UIreceipt to average beneït duration is lower than the estimate reported by Katz and Meyer ( t extension should UI are signiïcantly affected by abeneït extension, there is noindication that the availability of extended beneïts has much affect on the rise in UI-​leaving rates in the weeks just prior to the exhaustion of regular beneïts. This ïnding raises an important question regarding the cause of the pre-​exhaustion spike in exit rates. It is still possible that this spike is caused by the existence of a UI system that typically offers regular beneï ts for 26 weeks. For instance, Topel (1983) argues that employers enter into implicit contracts with workers and cycle them through spells of unemployment to extract the surplus created by imperfectexperience-​rating in the ïnancing of UI beneïts. If the terms of the agreementinclude a26-​week spell of unemploymen then changing these contractual arrangements in response to ashort-​term policy may be impractical. Alternatively, workers may have been conditioned to become less selective regarding possible job opportunities around the time that UI typically expires. Again, alonger-​term policy might be expected to have abier effect on the size of the pre-​exhaustion spike than ashort-​term policy like NJEB. Other explanations may be available, but regardless, the evidence indicates that at least ashort-​term beneït extension haslile or no impact on that spike. These considerations also suest that even our long-​term estimates of the effect of a13-​week extended beneï t program may be understated. If the program was in 39 One additional explanation for the discrepancy between our estimates and those obtained in past work (besides the potential endogeneity bias discussed earlier) is that there may be an interaction between the unemployment rate and the responsiveness of abeneït extension. For instance, durinood times, the opportunity cost of cuing back on job search activities may be too great for many andwould limit the extent to which the beneït extension would lenhen their unemployment spellecause our study is the ïrst that we know of that examines the impact of an increase in maximumbeneït duration during aperiod of economic expansion, such an interaction could explain the smaller response compared to earlier work. We have no evidence regarding the validity of this hypothesis. 1996. Unemployment insurance rules, joblessness, and part-​time work. Econometrica 1990. Unemployment insurance and unemployment spells. Econometrica 1992. Quasi-​experimental evidence on the effects of unemployment insurance from New York State. Unpublished manuscrip February 27. Meyer, DRosenbaum, D.T 1996. Repeat use of unemployment insurance. Working Paper No. 5423, January, National Bureau of Economic Research. Mofï R1985. Unemployment insurance and the distribution of unemployment spells. Journal of Econometrics RNicholson, W1982. The effect of unemployment insurance on unemployment: The case of federal supplemental compensation. Review of Economics and Statistics 1977. Unemployment insurance and job search outcomes. Industrial and Labor Relations Review 1986. Job search and labor market analysis. In: Ashenfelter, OLayard, R. (Eds Handbook of Labor Economics, Vol. 1999. Labor market institutions and economic performance. In: Ashenfelter, OCard, D. (Eds Handbook of Labor Economics, Vol. 4. North Holland, Amsterdam. Vroman, W1991. The decline in unemployment insurance claims activity in the Administration. Winter-​Ebmer, R1998. Potential unemployment beneï t duration and spell lenh: Lessons from aquasi-​experiment in Austria. Oxford Bulletin of Economics and Statistics ts. In: OLeary, JWandner, S.A. (Eds Unemployment Insurance in the United States: Analysis of Policy Issues. Upjohn Institute, Kalamazoo, MI. 138 extension for the large majority of recipients who were eligible for 26 weeks ofregular beneïts. The extension was available to all recipients who exhausted theirregular applied retrospectively to the set of claimants whose beneïts expired as far back asDecember 2 of 1995, which we subsequently refer to as the reachback group. To collect these beneï ts, an exhaustee needed to return to the UI ofïce to ïle aseparate claim for the extension. Formal notiïcation leers were sent to in–dividuals in the reachback group who had exhausted their beneïts soon after thelegislation was enacted. Claimants currently receiving UI, and those who started a new claim after June 2, were not individually notiïed of the beneï t extension untilthey received their ïnal regular engaged in avariety of outreach activities, including press releases, meetings withunion ofïcials, and the like. In addition, the notiïcation of the reachback grouppresumably generated word-​of-​mouth dissemination, particularly among frequent users of the ts among eligible UI recipients by the month of exhaustion of regular UI beneïts. The fraction of those in thereachback group who ï led a exhausted regular NJEB program by date of potential exhaustion of regular UI. D . Card, PLevine /​Journal of Public Economics ts just prior to the laws enactmen The take-​up rate for later claimants (i.e. those who exhausted after the effective date of the law) remainsfairly steady at about 70% arate similar to estimates of the take-​up rate forregular UI beneïts among eligible job losers (see, for example, Mc Call, NJEB raises an issue regarding the interpretation of the results presented here. If eligibleindividuals did not take up beneïts because they did not know about their availability, then estimates of the behavioral response to the program would be biased downwards, relative to the expected responses from amore widely advertised program. We believe that this bias is unlikely to be severe, however. Firs the program dissemination and application processes were similar to thoseused in other beneït extensions in New Jersey, including Emergency Unemploy– ment Compensation in the early 1990s. Although that program, in particular, did generate higher take-​up rates than NJEB according to New Jersey ofïcials, it wasalso offered during arecession. Evidence indicates that the most common reasonfor not applying for regular beneïts among those who are eligible is theexpectation of ïnding ajob soon (Vroman, Such expectations are no doubt more likely when economic conditions arefavorable and probably affect the decision to ïle for extended beneïts as well.Second, representatives of the New Jersey Department of Labor reported to us that within afew weeks of the programs inception, recipients were well-​informedabout the beneït extension. In fac in the few weeks following the expiration of the program on November complaints that they could not get i In our empirical analysis, we investigate this potential source of bias further by separately examining groups of workers who 6 months earlier is potentially surprisin and suests that many of these individuals had not found work even after 12 months of joblessness. 11 The slight drop-​off towards the end of the NJEB window is most likely related to mistakes made in determining arespondents actual exhaustion date. For an individual eligible for 26 weeks of regularbeneïts, we determined his /​her potential exhaustion date by moving 26 weeks forward from his /​her date of initial claim. This procedure may be inaccurate for two reasons. Firs some recipients becomedisqualiï ed for beneïts for ashort period during their spell because of, say, insufïcient work search.These individuals may subsequently collect beneïts within the same claim, but their actual date ofexhaustion will be delayed, potentially beyond the date that NJEB expired. Second, some recipientsïnd temporary employment before exhausting regular beneï ts and then reapply for UI. These individuals are eligible for the remaining number of weeks of eligibility from the initial claim, but their actual exhaustion date will be pushed back beyond that which we predicted. In both cases, the likelihood of incorrectly categorizing an potential exhaustion date is to the expiration date of NJEB. D . Card, PLevine /​Journal of Public Economics 077 claimants. We restrict our aention to the subsample of claimants who received aïrst paymen whose ïles include complete demographic and industryinformation, and who received no more than one week of partial UI beneïts. Thelaer restriction is adopted to eliminate the small fraction of claimants who recipients. The estimation results reported in this paper are based on the subsample of 1996 includes most of the claims that were prospectively eligible for NJEB, allowing aone-​month lag for information about the program to disseminate among 16 claimants. We use data for those claimants scheduled to exhaust their regular beneï ts from the same months of constant the seasonal differences that exist in the composition of UI claimants andin job-ïnding behavior. For those scheduled to exhaust regular beneïts within these three annual windows, job-ïnding activity in each week of their unemploy– ment spell is analyzed, not just those weeks within the July November period. It is important to note that our micro-​sample is limited to New Jersey UI claims. We can only use this sample to make comparisons within New Jersey over time. Thus, an assumption in most of our micro-​analysis is that claims from 1995 and factors such as unemployment rates We provide some limited evidence on the validity of this assumption below. The individual claims micro-​data can be used to reï né our analysis of aregate exhaustion rates for example, by taking into account differences acrossclaimants in the maximum duration of regular beneïts. The more important use of the micro-​data, however, is to estimate weekly hazard rates for ending aUI claim spell, and to measure the effect of NJEB eligibility on these hazard rates. Because 14 Individuals who ïle an initial claim but do not receive aïrst payment include those who ï nd ajob during the waiting week, who are deemed ineligible for beneïts after ïling aclaim, or who ïnd ajobduring the period in which eligibility is being determined among those for whom eligibility has been questioned. 15 See Mc Call (1996) for discussion of the set-​aside provisions that allow UI recipients to work part-​time and collect some beneï ts. We include claimants who collect one week of partial payments because recipients frequently obtain employment in the middle of the week and their last payment is a partial one. As discussed in more detail below, alimitation of the data available to us is that we cannotidentify those respondents whose ïrst weekly payment was apartial one. 16 Restricting the sample to those whose regular beneï ts were scheduled to expire after July 1 of each year provides another advantage in that many claimants whose regular beneïts would expire in June of1995 would have ï led their claim at the end of 1994 since most are eligible for 26 weeks of beneïts.None of these claims are available in our data and their omission could affect the comparability across years. Nevertheless, some recipients whose regular beneï ts did not expire until after July cation during the spell models for the event that yexceeds agiven threshold (6) describes the event that yexceeds the individuals maximum weeks of ieligibility (M and is ï t over the entire sample of claimants with potential iexhaustion dates between July and November of model in column (7) is acensored normal regression model for y, taking into iaccount that y# M. The laer model is interesting in part because similar models ii have been ï t in the previous literature, allowing us to draw comparisons between the New Jersey claimant sample and earlier samples. As determinants of latent UI spell durations we include adummy for observations from on UI for their full entitlement period the unemployment rate in the individuals county at the start of the claim, alinear trend variable (measuring months since January 1995) to capture any generic changes in the job-ïnding environment inNew Jersey over time, aset of dummy variables for the month the claim started, a set of individual characteristics, including age, gender, education, union status, and citizenship status, the individuals average weekly wage (in the period before the claim started) and UI replacement rate, the number of weeks worked for the 138 previous employer, and aset of major industry ïxed effects. In the models incolumns (6) and (7) we also include the individuals maximum weeks of cient estimates for the NJEB-​eligible dummy in Table 5 suest that although UI claims with scheduled exhaustion dates after July were somewhat less likely to survive ndings mirror the paern of the unadjusted survivor functions in Fi somewhat below an average of the survivor functions for that 1996 spells were less likely to survive than spells in the pooled comparison group of above the average for to last over ecting the fact that 1996 spells were more likely to end quickly, but also more likely to exhaus the estimates of the censored normal regression model in column (7) imply that on balance the number of weeks of regular beneïts receivedby 1996 claimants was not too different from the average in Several other aspects of the estimates from this model are also worth notin Forexample, the estimated coefïcient of the replacement rate variable implies that a 10 percentage point increase in the replacement rate (e from increase the average duration of UI spells by about one week. This ïnding iscomparable to estimates in the previous literature (see, for example, Mortensen, 1986; Meyer, 1990 The signs of the coefïcient estimates for the censored normalmodel are consistent with those of the probit model for exhaustion, and themagnitudes of the coefïcient estimates in the two models are also roughly consisten suesting that the normality assumption used in these models, although surely incorrec does not affect the qualitative inferences from the 32 As indicated earlier, the standard errors reported here are subject to some understatement because they do not incorporate the correlation that may exist across observations at apoint in time. This bias is probably larger in models that exclude the county-​level unemployment rate, since the laer presumably accounts for some of the correlation across individuals. We re-​estimated some models excluding the county unemployment rate and found that the standard errors on our key variables were only raised by34% he coefïcient estimates are also not much affected suesting that local shocks are small relative to other variance components. In light of the relatively small standard errors for the estimates of the key parameters in our models, any bias caused by common local shocks would have to be quite large to affect our inferences; nevertheless, readers should be aware of this potential problem. 33 In principle the probit coefï cients in the exhaustion model should equal the coefï cients in the censored regression model, divided by the estimated standard deviation of spells ( cients. D . Card, PLevine /​Journal of Public Economics 131 As we noted in the discussion of the hazard rates and survival functions in Figs. the NJEB program was announced. For this reason, it is likely that the estimates in Table 1996 sample were more likely to end early than those in apooled sample of 1995 and NJEB program, it seems implausible that UI-​leaving behavior in these weeks was affected by NJEB. Rather, we conjecture that economic conditions in early 1996 may have been somewhat beer than the average conditions in leading to asomewhat higher exit rate from UI and an increase in the fractions of claims ending in 1995⁄1997 comparison sample. If true, this suests that the estimates in Table 5 (and those in our aregate analysis in Tables of the 1996, we turn to ahazard modeling framework forreïning our estimates of the impact of the program. Speciïcally, we ï t discreteime hazard models for the probability l(i that individual iexits UI in week conditional on having remained on both conventional proportional hazard models and asimple logit functional form, 34 and found very similar estimates from the two alternatives. For simplicity, we report only the estimates from the logit speciïcations here. Since the probability ofexiting 24 the logit coefïcient estimates showthe approximate percentage change in the exit probability per unit change in the associated covariate. Akey advantage of the hazard framework is that it allows us to measure the effect of covariates whose values change over time, including the unemployment rate and most importantly the presence of the NJEB program. We therefore include in our hazard models two dummy variables: one indicating spells from our 1996 sample, and asecond indicating whether the current week is after July The former measures any differences in and those in the comparison sample of these spells. The laer measures any differential change in UI leaving rates for the information about the program to disseminate In this speciïcation, any un–observed factors that happened to shift average rate in cation is l(i cation is lo l(i cations iare nearly equivalent when the hazard probability is low (as it is in our application We chose to estimate logit models because of computational expedience. 132 D. Card , P . Levine /​Journal ofPublic Economics 6 a Hazard models of exit from unemployment insurance receipt (logit coefï cients multiplied by 73) Other individual characteristics YY YY YNumber of observations (weeks at-​risk) cient estimates for probability of leaving UI in agiven week, conditional on remaining on UI up to the previous week. Other individual characteristics include age, race, gender education, union status and citizenship, weekly wage prior to job loss, and weeks worked for former employer. See Card and Levine (1998) for parameter estimates on these variables. Sample includes claimants with potential exhaustion dates between July cations include ïxed effects for season and major industry. The baseline hazard is parameterized by separate dummy variables for each of theïrst three weeks on t exhaustion, as well as acubic in the number of elapsed weeks on UI. 1996 (which were more and more likely to have occurred afterJune and leading to an increase in the fraction of spells that exhausted (as shown by the probit models for exhaustion The model in column (3) adds three additional variables, representing interac– tions of the post-​NJEB dummy with the dummies for periods 3 weeks justprior to regular beneït exhaustion. The estimated coefïcients on these interactionterms are small and individually and jointly insigniïcan suesting that availabil–ity of NJEB led to only small changes in the size of the spike in exit rates prior 37 to exhaustion. The models in Table 6 all ignore the presence of unobserved individual-speciï c 38 heterogeneity. To get some sense of the possible implications of this omission, we performed anumber of checks. Firs we estimated the models without anyindividual-speciïc covariates, to gauge the sensitivity of our estimates to observ– able heterogeneity. This yielded estimates of the remaining baseline and NJEBcoefïcients very similar to the ones from the richer speciïcations reported in thetable. For example, with no individual-speciï c controls, the estimate of the 1996dummy in aspeciïcation similar to the one in column ( controls while the estimate of the post-​NJEB dummy is all controls These results suest that our estimates are not very sensitive to controlling for observed heterogeneity and, therefore, may not be terribly sensitive to unobservable heterogeneity either. Second, we compared the observable characteristics of individuals at risk to leave UI after various numbers of weeks. These comparisons show surprisingly lile systematic trend with time on UI. For example, average education is 12.4 years one week prior to exhaustion of regular beneïts. Similarly, the meanlog average weekly wage (for the old job) is 6.13 one week prior to exhaustion. Based on these results for the observable covariates, we think it is unlikely that unobserved characteristics lead to much bias in our estimates of the impact of NJEB on later weeks of long spells (sincethe average beneït week at risk in the post-​NJEB period of 1996 is about week 15 This is only aproblem, of course, to the extent that the post-​NJEB effect varies with spell duration, or varies across spells by the duration of the completed spell. To assess the possible magnitude of this type of heterogeneity, we 37 In this speciï cation, the indicator variables for the weeks immediately preceding exhaustion do not also need to be interacted with apost-​implementation dummy variable because all those eligible for eligibility after June 1990) considers aproportional hazards model with an unobserved individual-speciï c component that is assumed to follow agamma-​distribution in the claimant population. The presence of unobserved heterogeneity may lead to under-​stated standard errors in the models in Table 6, and also to bias in the estimated parameters. D . Card, PLevine /​Journal of Public Economics 137 effect for alonger time, implicit contractual arrangements could be modiïed,workers could be reconditioned to incorporate the longer availability of regularbeneïts etc. The quasi-​experiment created by the New Jersey Extended Beneïtprogram provides no information on such longer-​term adaptive behavior. Acknowledgements We thank Bruce Meyer, Roger Gordon, and two anonymous referees for comments on an earlier draf Tom Stengle at the US Department of Labor for providing the monthly, state-​level data employed in this analysis and Jean Behrens and Jim Phillips at the New Jersey Department of Labor for their assistance in using the administrative records data from New Jersey. We also thank Jean Behrens for many detailed comments that substantially improved the paper. All of the views expressed in this paper are our own and not those of the United States or New Jersey Departments of Labor or the National Bureau of Economic Research. References Allison, P.D 1982. Discrete-​time methods for the analysis of event histories. In: Leinhard S. (Ed Sociological Methodology. Josey-​Bass, San Francisco. Anderson, PMeyer, D1997. Unemployment insurance take up rates and the after-​tax value of beneï ts. Quarterly Journal of Economics 1991. Recent trends in insured and uninsured unemployment: Is there an explanation? Quarterly Journal of Economics 1993. Unemployment Insurance in the United States: The First Half Century. W.E. Upjohn Institute for Employment Research, Kalamazoo, ts and the duration of UI spells: Evidence from the New Jersey Extended Beneï ts Program, Working Paper No. 6714, Augus National Bureau of Economic Research. Ham, JRea, Jr. S.A 1987. Unemployment insurance and male unemployment duration in Canada. Journal of Labor Economics 1995. The effect of unemployment compensation on unemployment duration in Germany. Journal of Labor Economics 1990a. The impact of the potential duration of unemployment beneï ts on the duration of unemploymen Journal of Public Economics, February. Katz, LMeyer, D1990b. Unemployment insurance, recall expectations, and unemployment outcomes. Quarterly Journal of Economics, November, pp. 9731002. Gustafson, KLevine, P1998. Less-​skilled workers, welfare reform, and the unemployment insurance system. Working Paper No. 6489, March, National Bureau of Economic Research. Machin, SMannin A1999. The causes and consequences of long-​term unemployment in Europe. In: Ashenfelter, OCard, D. (Eds Handbook of Labor Economics, Vol. 4. North Holland, Amsterdam. Mc Call, P1995. The impact of unemployment insurance beneï t levels on recipiency. Journal of Business and Economic Statistics 12, 189198. UI spell durations, neither of which necessarily provides exogenous changes in the duration of beneïts. At thearegate level, policy changes (enacted by federal or state governments) alter theduration of beneïts for all claimants. The problem with these changes is that they are almost always triered by slackness in the labor market that has lead to high unemployment rates, leading to apotential reversal of causality. At the individual level, differences in past labor market histories create differences in the maximum amount of time that different individuals can receive UI. The formula that converts differences in labor market histories into different entitlement periods varies across states, providing some geographic variation in maximum beneït lenh. Howevero the extent that differences in UI entitlement are correlated with (or caused by) unobserved individual characteristics that also affect UI-​leaving rates, variation inindividual-speciïc UI beneï t durations is problematic. Perhaps the most convincing evidence that job-ïnding behavior is inïuenced by the maximum duration of beneïts comes from an examination of the rate ofleaving the UI roles in the weeks before beneï t exhaustion (cf. Meyer, 1990; Katz and Meyer, 1990b The available data clearly indicate that the probability of leaving UI he hazard rate) rises sharply in the last few weeks of beneïteligibility. Although this evidence is strongly suestive that some individualssearch harder to ïnd ajob (or return to pre-​arranged jobs) just prior to beneïtexhaustion, it does not directly address the policy question of the impact of abeneït extension on exit rates from that individuals who were already collecting UI at the time of abeneït extensionalso have aspike in their UI-​leaving rate prior to the time their beneïts werepreviously scheduled to exhaus even though they were eligible for longerbeneïts. These concerns underscore the potential usefulness of studying the effect of alegislative change in maximum beneït durations that came about at atime ofstable macroeconomic conditions, such as that generated by the NJEB program. We therefore turn to adetailed discussion of this program and its origins. 3. The New Jersey beneï t extension UI system in the United States is administered by the individual states under aset of national guidelines established by the federal governmen Regular 5 UI beneïts are ïnanced through apayroll tax that is mainly levied on ïrms. Eachstate operates aUI Trust Fund that accumulates funds during expansionary years 5 Most states levy the tax exclusively on ï rms while some, including New Jersey, levy part of the tax on workers. 138 arrangements. We detail some of this turbulent history here because it illustrateshow the 1996 beneït extension came about as ashort-​run solution to apolitical dilemma. In its original formulation the New Jersey Charity Care program was funded by the Uncompensated Care Trust Fund, which collected a19% surcharge on the hospital bills of paying patients. Soon after its introduction the surcharge cameunder ïre for driving up hospital rates and insurance premiums, and lowering the number of individuals covered by insurance. Legislative extensions of the program became hotly contested and the program even expired brieï y in only to be revived shortly thereafter. In 1992, alawsuit successfully challenged thesurcharge tax, ending this method of ïnancin To replace the revenues from the surcharge, state legislators agreed to ï nance Charity Care by diverting some of the surplus available in New Jerseys Unemployment Insurance Trust Fund. This plan was very unpopular among both labor and business groups. Labor groups worried that using funds from the Trusund would reduce the beneïts available to unemployed workers in the future. Business groups viewed the plan as ahidden payroll tax. Despite these concerns, the Charity Care program was funded in this manner from 1993 to the end of 1995, when opposition grew strong enough to block an extension. However, none of the alternatives proposed at the time, including apayroll tax, atax on health insurance premiums, atax on revenues from video poker games, and arise in the tobacco tax, could garner enough support to be enacted. The resulting legislative gridlock led the Charity Care program to expire at the end of nd ways to reinstate the program. One proposal to break the deadlock was to continue drawing funds from the UI trust fund, bu in agesture to organized labor, to authorize a short-​term extension in the maximum duration of UI beneïts. The ï rst referencewe have found to this proposal appears in asingle sentence near the end of a New York Times article (March 3) on the ïnancing crisis. Support for the proposal grewas the crisis continued; hospitals received their last payment for indigent care in February and were warning of layoffs and possible hospital closures if the issue was not resolved quickly. In the middle of May, legislation was enacted thaamong other things, traded abeneït extension for the continued use of the eliminates the reliance on the cigarees and appropriating general revenues to cover the remainder of the cos An examination of paerns in labor market activity by state demonstrates that the NJEB program was unrelated to changes in business cycle conditions. Fi 1 displays unemployment rates in New Jersey, Pennsylvania and for the entire 8 Acut in the UI tax for both employers and workers was also included in the package. UI exhaustion rates in New Jersey, Pennsylvania and the United States Period New Penn. 4) a Standard errors in parentheses. Exhaustion rate represents the number of claims exhausting in amonth divided by the number of ï rst payments 6 months earlier. The averages reported forJuly November are weighted averages of the respective months, using as weights the number of claims (laed respective averages. The entries in the last row of the table represent the difference between the 1996 average and the simple average of the unavailable. starting date to allow for information lags during the ï rst few weeks of the NJEB program. Columns (4) and (5) report the differences in exhaustion rates in New Jersey relative to the two comparison groups. As noted in Fi 3, average exhaustion rates are higher in New Jersey than in Pennsylvania, and also higher than in the rest of the from differences-​in-​differences in exhaustion rates between New Jersey and either comparison group as NJEB started and ended. Finally, the last row of the table shows the difference in average exhaustion rate for July November, 1996, relative to the average for the same months in 1995 and 1997. Anumber of alternative estimates of the effect of the NJEB program on New Jersey exhaustion rates can be drawn from Table 1. For example, suppose that average exhaustion rates would have followed alinear trend in New Jersey from exhaustion rates is avalid counterfactual for raised exhaustion rates by about 2 percentage points. An alternative is to assume that exhaustion rates would have paralleled those in D . Card , P . Levine /​Journal ofPublic Economics 199621995⁄1997 average 1997 Difference t-​statistic Unemployment rate (county) Percentage eligible for 2 a Samples include valid claims of individuals between ages 18 and 65 and excludes those with missing data on age, wages, industry or UI claim characteristics. D . Card, PLevine /​Journal of Public Economics nance higher expenditures in economic downturns. UI taxes arepartially experience rated: ïrms whose previous employees have drawn morebeneïts are taxed at higher rates, subject to (often bindin minimum and maximum rates. Unemployed individuals are eligible to collect UI beneïts if they have asufïcient work history and if they remain able, available, and actively seekingwork. Weekly UI beneïts are paid out according to an individuals earnings history prior to job loss, subject to aminimum and maximum beneïThe maximum beneït rate varies tremendously across the states, ranging from $ among the most generous states, providing amaximum beneït of $362 per weekin 1996. In contrast to the interstate variation in beneït levels, almost all states,including New Jersey, specify amaximum entitlement period of ts are usually available for up to 26 weeks, the maximum duration of beneïts is sometimes extended in cyclical downturns. Infac since 1970 there has been afederal program that provides 13 weeks ofextended beneïts when astates insured unemployment rate he number of current UI claimants divided by the number of employed workers covered by thesystem) exceeds aspeciïc threshold. Changes in the UI system over time, however, have made the trier virtually unaainable lank and Card, 1991 and over the past two decades federal emergency legislation has been enacted on an adhoc basis to provide extended beneïts during recessions laustein et al 1993; Woodbury and Murray, 1997 In addition, individual states can (and sometime do)raise the maximum duration of beneïts. To the best of our knowledge, such increases have occurred exclusively during periods of adverse labor market conditions. t extension In contrast to the traditional paern of linking UI beneï t extensions to changes in labor market conditions, the NJEB program emerged from apolitical com– promise around the states Charity Care program for indigent hospital patients. Since its inception in 1987, the ïnancing of this program was controversial, andover its 10-​year history state legislators struled to devise alternative ïnancing 6 These beneït levels are exclusive of dependents allowances which are available in some states. The additional payments made for each child is small in each of the handful of states which offers them. 7 The maximum duration of regular beneï ts in most states is lower for workers with limited work histories. Gustafson and Levine (1998) report that the average maximum duration of beneïts amongyounger workers in New Jersey is between 24 and 25 weeks. D . Card, PLevine /​Journal of Public Economics November of 1996. Unemployment held roughly constant in New Jersey and much of the rest of the country in economy appeared to grow more quickly than the US as awhole over this period, but no noticeable break from trend is apparent within New Jersey or between New Jersey and other states around the period in which NJEB was in effec As one might expect based on the legislative history, no obvious relationship exists between changes in business cycle activity and the timing of the NJEB The speciïc provisions of the beneït extension included a50% increase in the number of weeks for which beneï ts could be received, equivalent to a13-​week 9 The unemployment rate is not aperfect measure to use for this analysis because changes in UI policy may also affect the unemployment rate. Nevertheless, UI recipients represent aminority of the unemployed, and unless the impact of NJEB on spell lenhs was very large its effect on the aregate unemployment likely will be imperceptible. We have also conducted acomparison using the rates of growth in employment covered by the UI system and reached similar conclusions.

    Details

    • Title: PII
    • Producer: Acrobat Distiller 3.02
    • CreationDate: Tue Aug 8 10
    • ModDate: Tue Aug 8 11
    • Tagged: no
    • Pages: 32
    • Encrypted: no
    • Page size: 470 x 684 pts
    • File size: 257836 bytes
    • Optimized: yes
    • PDF version: 1.2