Does Inconvenience Explain Low Take-up? Evidence from UI Claiming Procedures

Transcription

1 Preliminary and Incomplete. Please Do Note Cite. Does Inconvenience Explain Low Take-up? Evidence from UI Claiming Procedures Avraham Ebenstein 1 Kevin Stange 2 November 7, 2005 Abstract One popular explanation for why many individuals do not apply for and claim the social benefits to which they are entitled is that doing so is inconvenient. Applications take time and some individuals may decide that the financial benefits do not outweigh these costs. This paper investigates this explanation by documenting the take-up and compositional effects of recent administrative changes that have made it easier for individuals to file initial claims for Unemployment Insurance. We find that the adoption of remote forms of UI claiming (e.g. phone, mail, Internet) did not have an appreciable impact on one rough measure of UI take-up, the claimant to unemployed ratio. At face value, this finding is inconsistent with a time- and transaction-cost explanation for low take-up if remote UI claiming is indeed less time-intensive, as claimant surveys suggest. We do find weak evidence for compositional shifts towards wealthier, more able claimants, which is consistent with the theory. 1 Department of Economics, University of California, Berkeley. ebenstei@econ.berkeley.edu 2 Department of Economics, University of California, Berkeley. kstange@econ.berkeley.edu 1

2 I. INTRODUCTION It has been documented that take-up of social programs in the United States is less than complete. In her extensive survey of the empirical evidence, Currie (2004) finds that take-up rates for means tested programs range from very low (8-14% for State Children s Health Insurance Program) to relatively high (82-87% for the Earned Income Tax Credit and 60-90% for Temporary Assistance for Needy Families), with most other major social programs falling somewhere in between. Take-up of non means-tested programs is generally higher (nearly 100% for Medicare), but often far from complete (72-83% for unemployment insurance). The latter is particularly puzzling if one expects the stigma associated with participation in non means-tested programs to be lower than that associated with means-tested ones. While stigma and informational issues are important, Currie concludes that concrete transaction costs must also be a major determinant of participation in social programs among those eligible. However, there has been relatively little direct evidence on the importance of these costs in practice. The importance of non-monetary program features such as the complexity of application paperwork, the presence of default or automatic enrollment, mandated in-person interviews or the frequency of re-application have been largely ignored in studies of social program participation, despite their potential importance. This present study addresses this void by examining the take-up consequences of changes in unemployment insurance claiming procedures over the 1990 s. Between 1990 and 2003, many states abandoned the practice of requiring UI claimants to appear in person at local offices in favor of easier, remote filing methods such as phone, mail, and the Internet. If these changes reduced the transaction costs associated with receiving UI benefits, then a simple model of program participation would suggest that overall take-up should increase as a result. Furthermore, the marginal UI claimant should shift further up the income distribution if income is a proxy for the value of eligibles time. We exploit cross-state variation in the timing of UI claiming procedures in order to test these predictions. The experience of Colorado, the first state to receive initial UI claims by phone, is typical of our general findings. In 1990, nearly all initial UI claimants in Colorado filed in-person. By 1991, the share had dropped to fifty-four percent and by 1992, only three percent of claims were made in person. However, the number of claims per unemployed person our measure of take-up and the composition of claimants were virtually unchanged during that period. Figures 1 and 2 report the fraction of claims filed in-person, the number of initial claims filed per unemployed person, and some characteristics of claimants over time for Colorado. This same pattern emerges when we look across all 21 states that implemented sharp changes in UI claiming procedures and when we control for other factors that may influence take-up and claimant composition. Though preliminary, these findings suggest that in aggregate 2

3 individuals eligible for UI did not respond to dramatic improvements in filing ease, casting doubt on inconvenience as a major determinant of program participation, at least in the case of UI. This paper is organized as follows. The next section provides a brief theoretical motivation for understanding how claiming convenience may influence individual s take-up decisions. Section III provides background on UI claiming procedures in the US since 1988 with particular focus on why and how states implemented changes. Section IV details our empirical approach and introduces our data. Results are reported in Section V and Section VI concludes. II. THEORETICAL MOTIVATION [In Process] III. BACKGROUND ON UNEMPLOYMENT INSURANCE CLAIMING PROCEDURES IN U.S. One of the biggest administration changes in state UI programs over the past decade has been the introduction of remote methods for filing an initial claim. In 1990, virtually all claimants were required to file their initial claim for Unemployment Insurance at a local state UI office. 3 Beginning in 1991, however, claimants in the State of Colorado were required to file their initial claim by telephone. A single call center was established in Denver where claims takers solicited caller s information and filed their claim over the telephone. The state UI agency no longer maintained staff in local offices to collect initial claims. This change was primarily in response to cost pressure and a desire to improve customer service. Since 1991, almost all other states have followed or plan to follow Colorado s lead by implementing procedures to receive initial UI claims over the telephone. More recently, many states have introduced Internet claims filing as an additional remote filing alternative. Figure 3 plots the fraction of all initial UI claims that were filed using various methods from 1988 to In the late 1990 s, the fraction of claims filed in person dropped dramatically as more and more states introduced remote filing. By 2003, only one-quarter of all initial UI claims were filed in person. There has been considerable variation in the motivation for, details of, and timing of states implementation of telephone (and now Internet) claiming methods. The U.S. Department of Labor commissioned an evaluation study, completed in 2000, of the implementation and impact of telephone initial claims filing in seven states [Needles et. al., 2000], which we summarize here and in Table 1. State UI administrators reported two common motivations for their switch from in-person filing at local UI offices to telephone claiming: administrative cost savings and improved customer service. It was believed that centralized call centers would allow states to realize economies in staff, office space, and training, 3 Perhaps due to the remote nature of many of its communities, Alaska has long permitted individuals to file UI benefits through the mail. 3

4 while making claiming easier for clients. The authors conclude that some states experienced a reduction in costs, while others did not. In some states, reductions in personnel and office space rental costs were offset by increases in communication and in equipment-related costs. States also vary in their implementation details. The primary implementation decisions facing states are how quickly to implement, whether to close local UI offices after implementation, the number of call centers to open, whether to offer a toll-free number, and whether to use a Interactive Voice Response (IVR) system. Since cost-savings was a primary motivation, most states closed local UI offices (eliminating the in-person filing option) after the introduction of telephone claiming. However, many states offer claimants dedicated on-site telephones at local UI offices and One-Stop Centers (which provide other UI-related and job-search activities) to use to file initial claims. Most states make use of IVR technologies to automate part of the initial claims process and four of the seven profiled states offer toll-free numbers. Important for our approach is whether phone claiming is preferred to and more convenient than in-person claiming since most states eliminated the in-person filing option following the introduction of telephone claiming. As reported in Needles et. al. (2000), customer satisfaction surveys in Maine, Massachusetts, Missouri, and Wisconsin suggest that claimants overwhelmingly prefer telephone to inperson claiming. Eighty-six to ninety-six percent of respondents in these states, half of whom are former in-person filers, report that telephone claiming is easier, more convenient, or faster. In a 1998 nationwide survey of 2,773 claimants, respondents estimated that it took 11 minutes to file an initial claim by telephone and 61 minutes to file in-person [Marcus and Frees, 1998)]. A similar picture of considerable timesavings from telephone claiming was also found in surveys of claimants in San Diego (2:20 hours inperson to 14 minutes via phone) and Colorado (3.4 hours to 1.7 hours). The case of Massachusetts is also illustrative. Despite being given the option to file in-person, only 11% of individuals chose to do so in Though far from conclusive, these anecdotes suggest that remote claiming methods are overwhelmingly preferred to in-person methods. States also differed in the precise timing of their remote claiming implementation. Table 1 also lists the dates each state introduced telephone initial claiming and the fraction of all claims filed in person in each state from 1988 to Two features of this data are important for our empirical approach. First, states introduced remote claiming at different times and during different macroeconomic environments. There is actually considerable more timing variation than shown in Table 1, as there are many states that implemented telephone claiming after 1997 or not at all by Second, the shift from in-person to remote claiming is remarkably rapid in many states. Most states that have nearly universal remote initial claiming saw this change happen over the span of a few short years. As discussed in the next section, the 4

5 cross-state variation in timing of rapid remote claiming adoption is used to identify the effect of these changes on take-up and claimant composition. IV. EMPIRICAL APPROACH We utilize cross-state variation in the timing of changes to UI claiming procedures to assess the importance of convenience to UI program participation. As discussed in the previous section, states implement remote UI claiming at different times or often not at all. This permits us to identify treatment effects separately from aggregate year effects and unobserved state characteristics, both of which may also influence take-up and claimant composition. A. THE DATA We construct a panel dataset of states for the years 1988 to 2003 from several different sources. Information on UI claiming procedures is from the Benefit Accuracy Measurement (BAM) program, administered by the U.S. Department of Labor. BAM is designed to measure the accuracy of paid and denied UI claims and determine the source of any inaccuracies, in order to improve UI administrative processes. BAM samples approximately 400 UI claimants per year in each state. Most important for our purposes is that claiming method (in person, phone, mail, Internet, , through employer) is recorded for each person in the sample. From this data, we estimate the fraction of UI claimants using each method in each state for every year from 1988 to As described in the next section, these fractions are used to construct policy event indicator variables and identify states to be used as a control group. Administrative data on the number of initial claimants, average duration on UI, number of weeks compensated and claimed, average weekly benefit amount, and several other measures of UI utilization were obtained from the Department of Labor Employment and Training Administration, quarterly by state. The unemployment rate and the number of individuals unemployed, employed, and in the labor force was obtained from the Bureau of Labor Statistics. We use the ratio of initial claims per unemployed person as our measure of UI program take-up. Finally, from the March CPS we obtain characteristics of all unemployed individuals at a point in time as well as characteristics of individuals who claimed to have received unemployment compensation in the previous year. The characteristics of the unemployed population in a given year are used as controls in our regression analysis, while characteristics of retrospective UI recipients in a year are used as a dependent variable. All nominal values for earnings, wages, and benefit amounts are converted to 2003 Dollars using the CPI-U. B. DEFINITION OF EVENT AND CONTROL STATES We define a remote claiming event to be a sudden decline in the fraction of individuals that file UI claims in-person. We operationalize this definition by identifying the first year in which more than 5

6 20% of a state s UI claims were filed remotely and defining this year as the event year. While this threshold is inherently arbitrary, it aligns reasonably well with the actual timing of phone claiming implementation for the seven states depicted in Table 1. We also restrict the sample to states that experienced a sharp event: dropping from at least 80% of claims in-person to at least 80% remote over a two-year span. Finally, to permit at least two post-event observations, we further restrict the sample to states whose remote claiming event occurred before Overall, twenty-one states passed these criteria and were used as event states in the analysis. To estimate aggregate-level yearly trends in the absence of remote claiming technology, we also sought to develop a group of control states whose claiming methods did not change appreciable over the period. Any state whose fraction of claims filed remotely did not change by at least thirty percentage points over the period was used as a control state. This procedure generates 12 control states (including the District of Columbia). C. ESTIMATION MODEL Throughout, we estimate variants of Equation (1) using least squares (1) yit = γ 0 + γ R Rit + γ R Rit + γ R Rit + γ RRit + γ RRit + γ RRit + γ RRit + γ X X it + γ i + γ t + ε it Our outcome measure y it is the number of UI claimants per person unemployed in state i in year t. This is one commonly used, though rough, measure of UI take-up. Our policy indicator variable R is an indicator for time relative to the first year state i permitted remote UI claiming. k it equals 1 if more than 20% of UI claims were filed remotely in state i for the first time k years earlier, and zero otherwise. Thus 3 R it, claiming event took place. 2 R it, and 1 R it are indicators for time periods before the remote X it is a vector of time-varying covariates such as the unemployment rate and the characteristics of the unemployed population in state i during year t. γ i and γ t are fixed state and year effects, while ε it is an error term. The parameters of interest are{ γ k R}, which can be interpreted as the effect of having adopted remote UI claiming k periods earlier on UI take-up in the 0 current period. γ R is the UI take-up change in the first year remote claiming was permitted, relative to not adopting the policy. We assume these effects to be constant across states and over calendar time. This method can be thought of as a generalized difference-in-difference, where a treatment-control difference is calculated at several points before and after treatment. When a full set of state and year effects is included, { γ k R} is estimated on policy changes within states over time, net of any aggregate yearly changes in UI take-up common to all states. Fixed differences in the levels of UI take-up k R it 6

7 across states will be absorbed into state fixed effects. In Equation (1), the unobserved counterfactual is implicitly estimated from individual state level effects (identified by pre-event observations) and aggregate time trends (identified by control states). This specification has the additional benefit of possessing a built-in test of our identification assumptions: if the policy event is sharp and not correlated with pre-existing trends in UI take-up within states, pre-event coefficients { 3, 2, γ γ γ 1 } should be zero. We can foresee at least two challenges to our identification strategy. First, states may have implemented other reforms that may also affect take-up (e.g. expanded outreach) concurrent with changes to claiming methods. Depending on the nature of the reform, omitted variable bias may over- or understate the true causal effect. A second problem is policy endogeneity. State administrators may adopt more automated claiming methods in response to higher anticipated demand placed on program resources due to higher anticipated take-up. If so, our estimates will be biased upwards. R R R V. RESULTS A. TAKE-UP Figure 4 provides a visual summary of our main findings on take-up. The figure plots the number of UI claimants per unemployed person from 1988 to 2003 for each event state along with the fraction of initial claimants that filed in-person. Vertical lines indicate the year a state first received more than 20% of UI claims remotely. From Figure 4, there does not appear to be a noticeable increase in take-up following the adoption of remote claiming procedures. To formalize this, Table 2 provides results from least squares estimation of Equation (1) with and without controls. Column (1) presents results from estimation with state and year fixed effects only. The change in take-up in the years leading up to and following remote claiming adoption are not significantly different from zero. Controlling for the unemployment rate and several characteristics of the unemployed (columns (2) and (3)) changes the point estimates very little and has no bearing on their statistical significance. Figure 5 plots the coefficients and 95% confidence intervals from the model with all controls. Though we find no statistically significant effects, we cannot rule out moderate effects of a few percentage point increase in take-up. B. COMPOSITIONAL EFFECTS An inconvenience explanation for low take-up also predicts that as the time cost of take-up is reduced, individuals with a higher time valuation or for whom benefits are less valuable will become more likely to participate. Micro data would be ideal for examining such heterogeneous effects, but is not readily available. Instead we note that heterogeneous effects would also shift the composition of UI claimants towards more wealthy (lower marginal utility of income) individuals. A wealthier claimant 7

8 population may have lower unemployment durations, higher weekly benefits, higher annual earnings, and be more likely to have graduated from college or be in an executive, professional, or specialized technical occupation. We examine each of these possibilities below. Figure 6 plots the average duration on unemployment insurance from 1988 to 2003 for each event state along with the fraction of initial claimants that filed in-person. Again, vertical lines indicate the year a state first received more than 20% of UI claims remotely. Average duration does decrease following the adoption of remote claiming procedures in many states, but not all. To formalize this, Table 3 provides results from least squares estimation of Equation (1) with average duration on UI as the dependent variable. In column (1), which controls for state and year effects only, coefficients on the post-event variables are all negative and statistically different from zero. Controlling for the unemployment rate and several characteristics of the unemployed (columns (2) and (3)) changes the point estimates very little, but does reduce their significance. One possible interpretation of the finding that average duration falls by about one half of a week when remote claiming is made available is that workers with short expected unemployment spells decide to claim unemployment insurance when it becomes easier to do so. Figure 7 plots the average weekly benefit amount (in 2003 Dollars) from 1988 to 2003 for each event state along with the fraction of initial claimants that filed in-person. Again, vertical lines indicate the year a state first received more than 20% of UI claims remotely. A clear visual pattern does not emerge from these graphs. Some states such as California, Maine, and Wisconsin saw a drop in average weekly benefits following their introduction of remote claiming methods, while others such as Kansas, Minnesota, and Vermont saw average benefit levels rise. To formalize this, Table 4 provides results from least squares estimation of Equation (1) with real average weekly benefit amount as the dependent variable. Column (1) controls for state and year effects only, while columns (2) and (3) also include controls for the unemployment rate and characteristics of unemployed workers, respectively. All three specifications result in negative and highly significant coefficients on all seven remote claiming variables. The significant coefficients on the pre-event indicator variables suggest the presence of an unobserved time-variant factor that is correlated with both timing of remote claiming implementation and average weekly benefit amount. We have not yet examined this anomalous result further. Our final approach to examine compositional shifts is to look at the characteristics of people who claimed to have received unemployment compensation at some time in the previous calendar year, from the March CPS. Figure 8 plots the share of UI recipients (as self-reported in the CPS) that have a college degree and that are typically in an executive, professional, or specialized technical occupation from 1988 to 2003 for each event state (using the left axis, denoted with triangles and circles respectively). On the same graph, we also plot the average real annual wage and salary earnings for the same group of individuals (right hand axis, denoted with squares). Again, vertical lines indicate the year a state first 8

9 received more than 20% of UI claims remotely. All three series are quite noisy due to limited sample size, but several patterns do emerge. The fraction of UI recipients with college degrees or in highly skilled occupations does appear to have grown over time in many states, but this does not appear to be associated implementation of remote claiming technology. Average real wages have also grown, but this does appear to coincide with the adoption of remote claiming. To formalize this, Table 5 provides results from least squares estimation of Equation (1) with these three characteristics of UI recipients as dependent variables. All three specifications include fixed state and year effects and controls for the unemployment rate and characteristics of unemployed workers. Economic changes that shift the pool of unemployed workers who could potentially claim UI are controlled for. We find no evidence of a shift in UI claimants towards the college-educated or workers in highly skilled occupations coinciding with remote claiming. We do, however, observe a statistically significant $1,600 increase in the average annual earnings of UI recipients. This is suggestive of an increased take-up rate for wealthier individuals, as easier and less time-intensive UI filing options became available. VI. CONCLUSIONS This preliminary analysis suggests that the adoption of remote forms of UI claiming (e.g. phone, mail, Internet) did not have an appreciable impact on one rough measure of UI take-up, the claimant to unemployed ratio. At face value, this finding is inconsistent with a time- and transaction-cost explanation for low take-up if remote UI claiming is indeed less time-intensive, as claimant surveys suggest. However, our estimates are sufficiently imprecise such that we cannot rule out moderate effects of a few percentage point increase in take-up rates. Large to moderate effects for a small population of marginal claimants may not show up in state aggregate take-up rates. This possibility underscores the importance of testing for heterogeneous effects. The evidence on whether claimant characteristics changed with the advent of remote claiming is weak. We find that the average UI benefit duration of claimants fell while their average real earnings increased, but the share of college graduates or workers in highly skilled occupations did not change. We could not identify the effect on average weekly benefit amount because our identification assumption does not appear to hold. Future analysis should test for heterogeneous effects directly using individual-level micro data with imputed UI eligibility. 9

10 REFERENCES Anderson, Patricia M. and Bruce D. Meyer, Unemployment Insurance Takeup Rates and the After-Tax Value of Benefits. The Quarterly Journal of Economics, August Blank, Rebecca M. and David E. Card, Recent Trends in Insured and Uninsured Unemployment: Is There an Explanation? The Quarterly Journal of Economics, November Currie, Janet, The Take-Up of Social Benefits. NBER Working Paper National Bureau of Economic Research, May Marcus, Steven S., and Joseph W. Frees, US Department of Labor Unemployment Insurance Claimant Satisfaction Study. Prepared for the US Department of Labor, Unemployment Insurance Service. Portland, OR: Bardsley and Neidhart, Inc. September Needles, Karen, Walter Corson, Tim Meier, Ira Harley, and Karen Blass, Evaluation of the Impact of Telephone Initial Claims Filing. Report submitted to the US Department of Labor and the National UI Information Technology Support Center. Princeton, NJ: Mathematica Policy Research, Inc. March Vroman, Wayne, Low Benefit Recipiency in State Unemployment Insurance Programs. Prepared for the US Department of Labor. Washington, DC: Urban Institute. June Storer, Paul and Marc A. Van Audenrode, Unemployment insurance take-up rates in Canada: facts, determinants, and implications. Canadian Journal of Economics, November

11 Figure 1: UI Take-Up and Initial Claiming Method In Colorado Initial UI Claims per Unemployed Person Fraction Filing in-person 11

12 Figure 2: Characteristics of UI Claimants In Colorado Real Wage and Salary Earnings Exec, Prof, Tech Occupation Have College Degree 12 Fraction of UI Recipients 2003 Dollars

13 Figure 3: Method Used to File Initial UI Claims Figure 2: Method Used To Claim Unemployment Insurance Person Phone Other Method Internet Year 13 Fraction of All Claims

14 Figure 4: UI Take-up and Fraction of Claims Filed In-Person 14

15 Figure 5: Regression-Adjusted Effect of Remote Claiming on UI Take-up 95% CI Upper Bound Coefficient on Event Year Indicator 95% CI Lower Bound Years Since Remote Claiming Event Note: This graph plots the coefficients and 95% confidence intervals from Table 2, specification (3) 15

16 Figure 6: UI Duration and Fraction of Claims Filed In-Person 16

17 Figure 7: Average Weekly Benefit Amount and Fraction of Claims Filed In-Person 17

18 Figure 8: Selected Characteristics of UI Recipients 18

19 TABLE 1: SELECTED CHARACTERISTICS OF EARLY PHONE CLAIMING IMPLEMENTATION Dates of Implementation Colorado Maine Massachusetts Missouri Montana Utah Wisconsin A: Selected Characteristics of Implementation April 1991 April 1997 July 1997 February 1996 November 1996 November 1996 October 1997 October 1996 October 1997 April 1997 August 1997 May 1995 January 1996 Main Motivation Reduce costs Reduce costs Reduce costs, response to privatized employment service Reduce costs, relieve office space pressure Improve training, reduce costs Improve customer service, reduce costs Reduce costs, improve customer service Closed local offices? Yes Yes No Yes Yes Yes Yes Toll-free number? Yes No No Yes No Yes Yes Use Interactive Voice Minimal Extensive Extensive Extensive Minimal Extensive Extensive Response System? B: Fraction of Claims Filed in Person Denotes year that state first received more than twenty percent of its initial UI claims remotely. This is the event indicator used in the analysis that follows. Source: Panel A is adapted from Needles et. al., (2000), Table 2-1. Panel B is from author s calculations of data from the U.S. Department of Labor Benefit Accuracy Management program. 19

20 TABLE 2: TAKE-UP Dependent Variable: Initial UI Claims per Unemployed Person (1) (2) (3) Time relative to event year 3 years before (0.020) (0.020) (0.020) 2 years before (0.018) (0.017) (0.017) 1 year before (0.020) (0.019) (0.020) Same year (0.023) (0.021) (0.021) 1 year after (0.026) (0.021) (0.022) 2 years after (0.031) (0.027) (0.024) 3 years after (0.029) (0.026) (0.026) Unemployment rate (0.612)** (0.606)** Characteristics of the unemployed Average age (0.001) Percent female (0.058)* Percent non-white (0.066) Percent with college degree (0.109) Average weeks looking for work (0.001) Percent in manufacturing or construction industry (0.058) Percent in executive, professional, or technical occupations (0.110) Average hourly wage (0.002) Average earnings in previous year (0.000) Average weeks worked in previous year (0.002)* State dummy Yes Yes Yes Year dummy Yes Yes Yes Observations R-squared Notes: All specifications are estimated with 21 Event states and 12 Control states (including the District of Columbia). Specifications (1) and (2) are estimated over the years 1988 through 2003, while specification (3) is estimated over the years 1989 through Robust standard errors in parentheses * significant at 5%; ** significant at 1% 20

21 TABLE 3: AVERAGE DURATION ON UI Dependent Variable: Average Weeks Receiving UI (1) (2) (3) Time relative to event year 3 years before (0.161) (0.150) (0.154) 2 years before (0.148) (0.134) (0.122)* 1 year before (0.174) (0.155) (0.157) Same year (0.199) (0.168) (0.170) 1 year after (0.273)* (0.250) (0.240) 2 years after (0.245)** (0.220)** (0.207)** 3 years after (0.206)* (0.177)* (0.189) Unemployment rate (4.611)** (5.240)** Characteristics of the unemployed Average age (0.011) Percent female (0.559) Percent non-white (0.519) Percent with college degree (0.866) Average weeks looking for work (0.011)** Percent in manufacturing or construction industry (0.526)* Percent in executive, professional, or technical occupations (0.774) Average hourly wage (0.006) Average earnings in previous year (0.000) Average weeks worked in previous year (0.014) State dummy Yes Yes Yes Year dummy Yes Yes Yes Observations R-squared Notes: All specifications are estimated with 21 Event states and 12 Control states (including the District of Columbia). Specifications (1) and (2) are estimated over the years 1988 through 2003, while specification (3) is estimated over the years 1989 through Robust standard errors in parentheses * significant at 5%; ** significant at 1% 21

22 TABLE 4: REAL AVERAGE WEEKLY BENEFITS Dependent Variable: Real Average Weekly Benefits, in 2003 Dollars (1) (2) (3) Time relative to event year 3 years before (2.220)** (2.284)** (2.320)** 2 years before (2.528)** (2.591)** (2.520)** 1 year before (2.653)** (2.660)** (2.628)** Same year (2.826)** (2.816)** (2.895)** 1 year after (3.643)** (3.658)** (3.614)** 2 years after (3.466)** (3.394)** (3.418)** 3 years after (3.644)** (3.537)** (3.683)** Unemployment rate (87.221)** (88.779)** Characteristics of the unemployed Average age (0.183) Percent female (8.198) Percent non-white (8.849) Percent with college degree (13.786) Average weeks looking for work (0.172) Percent in manufacturing or construction industry (9.587) Percent in executive, professional, or technical occupations (15.163) Average hourly wage (0.138) Average earnings in previous year (0.000) Average weeks worked in previous year (0.210) State dummy Yes Yes Yes Year dummy Yes Yes Yes Observations R-squared Notes: All specifications are estimated with 21 Event states and 12 Control states (including the District of Columbia). Specifications (1) and (2) are estimated over the years 1988 through 2003, while specification (3) is estimated over the years 1989 through Robust standard errors in parentheses * significant at 5%; ** significant at 1% 22

23 TABLE 5: CHARACTERISTICS OF UI RECIPIENTS Time relative to event year 3 years before (0.014) 2 years before (0.016) 1 year before (0.015) Same year (0.013) 1 year after (0.017) 2 years after (0.018) 3 years after (0.016) Dependent Variable: Characteristics of individuals receiving UI at any time during year Percent with college degree Percent in exec, prof, tech occ Average earnings during year UI was received (1) (2) (3) (0.021) (0.021) (0.014) (0.013) (0.015) (0.013) (0.015) 896 (984) 1,292 (699) -58 (755) 10 (684) 1,674 (850)* 683 (844) 1,669 (831)* Unemployment rate (0.492) (0.417) -62,361 (20,018)** Characteristics of the unemployed Average age (0.001) (0.001) (52.6) Percent female (0.039) (0.042) 1,877 (1,994) Percent non-white (0.048) (0.051) -4,564 (2,599) Percent with college degree (0.078) (0.080) 4,248 (3,663) Average weeks looking for work (0.001) (0.001) 79.4 (51.1) Percent in manufacturing or construction industry (0.050) (0.047) -2,374 (2,257) Percent in executive, professional, or technical occupations (0.076) (0.072) -12,685 (3,874)** Average hourly wage (0.001) (0.001) 15.6 (26.6) Average earnings in previous year (0.000) (0.000) (0.080) Average weeks worked in previous year (0.001) (0.002) (70.1) State dummy Yes Yes Yes Year dummy Yes Yes Yes Observations R-squared Notes: All specifications are estimated with 21 Event states and 12 Control states (including the District of Columbia). Specifications (1) and (2) are estimated over the years 1988 through 2003, while specification (3) is estimated over the years 1989 through Robust standard errors in parentheses * significant at 5%; ** significant at 1% 23

24 24