Adjusting to the Optimal Claim Learning Effects for UI

Adjusting to the Optimal Claim: Learning Effects for UI Benefits in Canada David Gray Ted McDonald June, 2007 We gratefully acknowledge funding from the Audit and Evaluation Branch of Human Resources and Social development Canada. The views expressed in this piece are attributable only to the authors, and do not necessarily represent the perspective of HRSDC. We have benefited from input from Harold Henson, Tom Crossley, and Saul Schwartz. Thomas Lang provided indispensable assistance in processing the data set. I. Introduction The topic of frequent or repeat use of Canada’s unemployment insurance (UI) system has generated much attention from policy analysts in Canada since the late 1970s. 1 While its origins were centered on insuring workers against temporary loss of income stemming from an unforeseen layoff, the passage of the Unemployment Insurance Act in 1971 greatly loosened eligibility requirements such that the program assumed the added function of providing income maintenance for seasonal, part-year workers during their off-seasons. The phenomenon of workers drawing UI benefits frequently – typically defined as filing 3 or more claims within a 5 year window – has been quite prevalent since the late 1970s.2 A series of empirical studies, policy discussion papers (e.g. Nakamura and Diewert (1997,2000)), and governmental commission reports (e.g. Green Paper (HRDC (1994)) have ensued that deal with some aspect of UI benefits.3 It is widely believed that the regime’s benefits and parameters generate economic incentives and disincentives militating toward dependence on the UI regime. These concerns are shared by the Organization for Economic Cooperation and Development (OECD (1999)). Research for the United States indicates that the frequent use of UI benefits is not solely a Canadian phenomenon (McCall (2000), Meyer and Rosenbaum (1996)). The phenomenon of repeat use of UI benefits invokes a number of economic problems, two of which are elaborated in Lemieux and Macleod (2000). First, there is the issue of how workers and firms adjust their economic choices to changes in the incentive structure. One expects for the process by which workers and firms modify their behaviour in order to attain their optimal Within Canada the official title of the UI regime is “Employment Insurance”, which is abbreviated EI. Throughout this paper we refer to it by the generic term UI. 2 Between 2000 and 2004, approximately one-third of claimants met this definition, and they accounted for about 40 % of all regular benefits paid (EI Monitoring and Assessment Reports, EI Commission, Human Resources and Social Development Canada). 3 A non-exhaustive list features Corak (1993a, 1993b), Lemieux and McLeod (2000), Green and Riddell (1995), Baker and Rae (1995), Green and Sargent (1998), Gray (2001), Gray and Sweetman (2001), Schwartz et al. (2001), and Corak and Chen (2003) 1 1 response to take some time, and this would have repercussions for how observed outcomes are statistically correlated either across time periods or across UI claims. Second, there is a more specific yet related issue within the context of repeated trials of whether and how actors learn information relevant to their choices. Given a rather elaborate set of program regulations and labour market features, one might expect that prior experience in accessing the UI regime would influence current use patterns. This adjustment process over a multiple UI claim horizon is the subject of this paper. Most of the existing literature that deals with the ‘take-up’, or the incidence, of UI benefits, such as Blank and Card (1991), Anderson and Meyer (1997), McCall (2000), and Lemieux and Macleod (2000) models the discrete, binary outcome - usually at an annual frequency - of whether or not an individual receives any UI benefits over a given time interval. In all of these papers, however, there is typically no use of information on how the regime is used, nor on whether the use patterns evolve with experience. A UI claim has several different attributes, such as the length of the benefit period, the length of the qualifying period, the replacement ratio, and the calculation of the earnings base from which the benefit amount is determined. Canada’s UI system is quite intricate and contains a number of arcane provisions that are primarily and deliberately targeted at frequent users. Given these complexities, optimal use of the UI regime could involve a trial-anderror process spanning several claims cycles, as individuals learn not only of the existence of these provisions, but also how to incorporate them into their working activities such that they benefit from the provisions as much as possible. To our knowledge this is the first paper to observe and analyse how specific facets of the UI claim evolve with repeated experience with the UI regime. We employ statistical and econometric analysis of how UI claimants adjust certain features of their claims as they gain 2 experience through repeated interactions with it. The empirical work draws on an administrative data base that contains longitudinal information on individuals’ UI claims histories between 1980 and 2003. We find strong empirical patterns suggesting that over the span of the first three or claims that a worker has filed, several attributes of these claims evolve and appear to approach certain values. We also devise empirical strategies for potentially discerning both social learning effects and individual learning effects, and uncover some evidence that is consistent with their existence. II. Survey of the literature In the spirit of Lemieux and MacLeod (2000), our focus is the worker’s inter-temporal patterns within his/her claims profile. Those authors model empirically the likelihood of an individual drawing UI benefits during a given interval conditional on his/her past incidence of UI use. They are primarily interested in the pattern of occurrence dependence, which refers to the degree to which use of UI during any prior time period is correlated with current or future use.4 Subsequent to any major change in the provisions of a social insurance system, such as the major change in program parameters implemented in Canada in 1971, one might expect actors to exhibit lagged adjustment patterns in finding their optimal response to new incentives. In the near term, they observed a minor response for UI take-up to the large increase in the benefits that occurred, but in subsequent years the incidence of UI receipt gradually increased despite some cutbacks in the generosity of UI benefits in the 1980s. They attribute that observed phenomenon to an occurrence dependence process. The absence of any correlation would imply a Markov process. Another possible time-series pattern is the one of inertia or persistence, which refers to the degree to which use of UI in the immediately preceding time period(s) is (positively) correlated with current use. In that case, the lagged values of UI use have an influence on the current likelihood of use, but whereas persistence effects are expected to dissipate over time, occurrence dependence effects endure. 4 3 The behavioural effect that Lemieux and Macleod suggest that could give rise to the statistical pattern of occurrence dependence is individual learning effects. In the event that the worker is initially exposed to the UI system through a layoff that was infrequent and unanticipated, he/she learns about how to access the system upon filing a claim for the first time and gains an appreciation of its provisions.5 They posit that this tends to elicit a supply-side response on the part of these first-time UI claimants such that they are ‘conditioned’ toward a higher propensity to claim in subsequent periods. They also mention that employers likely play a role in determining UI use patterns, and thus the learning process plays a role on the demand side as well. Note that there is a distinction between individual learning effects, in which an individual learns in a somewhat autonomous fashion through trial and error, and social learning effects, which are centered on interactions with peers and are associated with mores or observational learning. The two processes are not mutually exclusive. Social learning effects are an element of the burgeoning social capital literature. The social capital approach is often applied to situations in which norms and social networks facilitate cooperative action among groups of individuals designed to develop attributes that increase an individual’s earnings. Its central tenet is the observed propensity of an agent to behave in some way that varies positively with the prevalence of this behaviour in the group, as a contagion effect within a neighbourhood or some spatial unit (Manski [2000]).6 The approach has been applied to outcomes associated with crime, fertility, health care choices, and educational attainment. Most important for this paper, it has been applied to social insurance programs such as social assistance (Borjas and Hilton [1996], Bertrand et al These authors address the possibility that the observed adjustment process could have been driven by other economic effects, such as addiction behaviour, scarring effects among displaced workers, or positively auto-correlated separations giving rise to a series of UI claims. They reject those alternative explanations in favour of the learning effects conjecture. 6 The antithesis of a social capital transaction is a totally impersonal and anonymous exchange in an environment of risk, uncertainty, and possible distrust for which the only safeguard is the rule of law. 5 4 [2000]), but infrequently in the context of UI programs. Gray [2001] finds empirical associations between the frequency of UI use and indicators for social learning. The two basic elements of the generalised social capital approach set out in Bertrand et al. [2000] can be applied to a UI regime. First, there is contact availability, which refers to how much exposure a subject might have to the social network of influential peers. In the context of UI, the coherent social group or community consists of the players in the local labour market where UI use is widespread.7 Second, there is network quality, which refers to the usefulness of the information that is transmitted through the network. This refers to information pertaining to the employment patterns, UI program regulations, the job finding contacts and referrals that are critical in qualifying for benefits, and the social mores related to program participation. In this application we treat learning not only through initial exposure to the UI regime, but through repeated use and interactions – a potential ‘learning by doing’ process. Given the arcane complexities of Canada’s UI system, it is quite possible that claimants will not learn all there is to know about the system upon their first claim. There are a number of UI provisions and regulations that tend to be utilized primarily by frequent users; occasional users tend to avail themselves of them less frequently. One of the primary points raised by de Raaf, Motte, and Vincent (2004) in their treatise on the UI system is that some of the regulations are designed expressly for the benefit of frequent users. Lesson #1: Those who claim EI the most know how to benefit the most from its rules…claimants are often required to navigate through a myriad of rules to determine their eligibility and entitlement and consider what impact any work they chose to accept whether on claim or off – will have on their paid benefits. (page 7) A joint and collective effort aiming at obtaining UI benefits from the federal government develops. Within the confines of the community, the benefits are non-rivalous and inclusive, and there are few internal costs as the financing of the UI benefits is external to the local community. 7 5 Optimal use of the intricate apparatus of provisions and regulations, such as the ‘matrix of EI entitlement levels’, the ‘divisor’ rule, the ‘small weeks initiative’, the ‘clawback rule’, the ‘allowable earnings’ provision (described below), and the ‘renewal’ provision (described below) is probably an acquired skill. The central thrust of this paper is that experience in accessing the UI regime is required in order for a claimant to fully adapt his/her UI claim – characterized by several attributes - along with the concomitant employment pattern to attain the optimal claim that maximizes his/her benefits. This adjustment process can reflect pure learning effects, such as awareness of all of the provisions and regulations, as well as adapting their employment activity (i.e. their start-dates and end-dates, their weekly hours, etc.) in order to benefit fully from them. The adjustment process can also reflect learning of similar information by employers who, because of their interest in shifting some of their workers’ remuneration to the UI regime, share the workers’ objectives. In addition, social service agents of the provincial governments have an incentive to encourage unemployed workers to benefit as much as possible from the UI system due to the division of responsibility for social insurance between the two levels of government.8 They therefore might be expected to participate in the learning process by transmitting this information to workers. III. Description of relevant features of Canada’s UI system It is well-known that the lattice of parameters and the provisions of Canada’s UI system engender a wide margin for strategically ‘gaming the system’. This phenomenon typically consists of workers, often in conjunction with employers, tailoring hires and separations to the incentive structure in order to gain eligibility and maximize receipt of benefits. The shorter qualifying As the social assistance benefits are funded from provincial revenues, there is an incentive to ‘upload’ those costs to the federally-funded EI system. This economic phenomenon, which is sometimes labelled ‘vertical fiscal competition’ or ‘cost-shifting’, is described in OECD (1999). It is more likely to occur in decentralised federations such as Canada. 8 6 periods and longer benefit entitlement periods that prevail in high unemployment areas - a salient feature of the regime - are expressly designed to accommodate workers engaged in seasonal, fragmented, and part-year employment patterns. It has been demonstrated empirically that the regional variations in the entry requirement and benefit duration periods shape employment patterns in certain segments of the labour markets.9 Based on both theoretical and empirical analysis, Green and Riddell [1995,1997] and Green and Sargent [1998] investigate the behavioural implications that stem from these two provisions. They apply three labour market paradigms to the Canadian UI system, namely the implicit contract model, the search model, and the static, conventional labour supply model. In the case of frequent and/or seasonal workers employed in a framework with a one-year planning horizon, they find empirical support for the implicit contract approach. For this type of employment pattern, they assert that there are two implications. First, for seasonal workers with brief employment spells, there is a very strong incentive to work long enough to meet the minimum qualification requirements for UI benefits, and a lesser – but still positive – incentive to work marginal weeks beyond that point. Second, many seasonal workers have a strong incentive to ‘max out the year’, which means to work a sufficient number of weeks in order to cover the entire calendar year with either working weeks or benefit weeks. After that point has been reached, however, the incentive for working more weeks is greatly diminished because the period over which benefits can be collected is limited by the start of the next working season. Any additional week of work would reduce the number of usable UI benefit weeks one-for-one, after which UI benefits have little value. Since the shadow price of more work is quite high after max- Anecdotal evidence going back to the 1970s abounds. For recent and rigorous scientific evidence based on fully representative samples, see Kuhn and Riddell (2006). 9 7 year point has been reached, one does not expect to observe a large number of work weeks after that threshold. A less well-known, yet qualitatively and quantitatively important, phenomenon is the effect of two other program provisions, namely the ‘allowable earnings’ and the ‘renewal’ clauses, on the working patterns of claimants while they are on claim. As explained in Gray and de Raaf [2002] and Gray [2006a], it is a very common practice for claimants – particularly frequent users – to accept brief, interim work (between the end of the normal working season and the beginning of the next one) while on active claim status. They may earn up to the level of ‘allowable earnings’ (currently $ 150 per week) without having any of their benefit clawed back Claimants who are offered these very short-term jobs typically suspend their EI claim entirely. When the brief employment spell ends, the ‘renewal provision’ allows them to resume the UI claim without losing any of their entitlement. They can redeem the benefits to which they would have been entitled for those weeks of work at a later date (within a year of the start date of the claim.) These two provisions give a high degree of flexibility to workers with sporadic working patterns, as the traditional distinction between working activity and benefit receipt is blurred. Any working activity occurring during the claim counts as ‘insurable earnings’ towards qualification for a subsequent claim. Furthermore, the foregone benefit weeks can be deferred to the end of the original benefit entitlement period, thus prolonging the date of exhaustion (perhaps until the start of the next working season.) By combining working activity with the receipt of UI benefits, the annual income received from such an employment pattern can be increased substantially compared to the case of an employment cycle characterized by one continual working period followed by one uninterrupted claim period. The arrangement can also be very beneficial for employers, who have an on-call, contingent labour force available that they can lay off and recall with great flexibility. 8 For this purposes of this paper, three propositions are gleaned from the literature cited above. First, the UI system provides very strong incentives for individuals who work in high unemployment regions to work the minimum number of weeks that are required to qualify for UI. After that point has been reached, the rate of return to working additional weeks diminishes. Second, in many cases seasonal layoffs and quits do not occur as soon as the worker has qualified for benefits, but rather at the point in the one-year working cycle where the number of benefit weeks is sufficient to fill the remainder of the 52 week period. After that point has been reached, the rate of return to working additional weeks diminishes. Third, for claimants who experience frequent interruptions to their employment spells, there is a strong incentive to suspend ongoing claims and increase number of weeks of deferred benefits. IV. Description of the Data Set The longitudinal feature of the data set allows for the observation of repeated claims for an individual. There is also quantitative information on the attributes of these claims that allow us to trace the evolution of these values across a sequence of claims for those individuals filing multiple claims. The data set is derived from the administrative data bases of HRSDC. From the Status Vector Data File (STVC), which is a longitudinal data set of weekly frequency containing an individual’s EI claim history, we randomly selected a sample of 100,000 individuals who experienced at least one claim over the period from 1985 to 2003. The claim upon which selection is based must involve at least $1 paid out in regular benefits or fishing benefits. No attempt was made to balance the composition of the individuals selected over the 18 annual periods that make up the period spanned by this study. All UI claims filed between 1980 and 2003 by the selected workers are selected into the working sample, which generates a total of 426,082 claims. All of 9 these claims are for the payment of regular benefits or fishing benefits, but not for sickness or maternity/parental benefits. Each record for a claim contained in the STVC file is linked to the record(s) containing the employment history that was the basis for qualifying the individual for the claim, which is located in the Record of Employment Data File (ROE). The estimating sample is restricted to those aged between 18 and 65. The resulting data set comprises 96,413 individuals with a total of 330,995 claims.10 Our data set has the advantage of covering at least two complete business cycles, but the time interval spanned is somewhat shorter than those in the earlier works of Corak (1993a) and Lemieux and Macleod (1995, 2000). Since we observe EI claims histories for all claimants going back to 1980, for all years after 1984, we can determine the repeat-user status. For our sample, the earliest year for which a person could be selected for sample inclusion is 1985, by which point in time a five-year claims history could be calculated. Taking the unit of analysis to be the individual claim, the distribution of the observations according to the year in which the claim was initiated over the interval 1980-2002 is fairly evenly spread, as can be seen in appendix table A1. Approximately 2.25 % of the claims were initiated in 1981, while the frequencies for all subsequent years range from 3.7 % to 5.4 %. The distribution of individuals according to the year in which they are observed to have filed their first claim since 1980 (but not necessarily their first claim ever) was also calculated. Because 72 % of the individuals have multiple claims over this interval, this distribution is concentrated much more heavily toward the earlier years relative to the distribution of all claims. Approximately 57 % of A number of observations and individuals were dropped from the estimating sample as a result of the merger. For 2,333 individuals and 60,481 claims, no corresponding ROE records could be found despite that fact that a non-null claim was filed. These omissions are probably due to administrative coding errors; the missing information might be reported in another time period. Due to other coding errors, such as a misclassified administrative region or week in which the claim was initiated, 10 more individuals and 553 claims were dropped. In the case of 31,108 claims, no benefits were ever paid; thus these observations were dropped. 10 10 the individuals filed their first claim during the decade of the 1980s, while only 38.7 % of the all claims occurred during that period. Approximately 38 % of the individuals filed their first claim after 1990. The share of men in the sample is 55 % (accounting for 59 % of the observations for claims), while the share of women is 44 % (accounting for 40 % of the observations).11 Information regarding the covariate of age is listed in appendix table A2. The average age of the claimants at the time of their first claim since 1980 is 31.6 years, and the median is 27 years; 25 % of these individuals are aged 22 or below, and 75 % are younger than 39. When the observations are weighted by the number of claims, the average age of the claimants at the time of their claims is 35.5 years, the median is 33; 25 % of these individuals are age 25 or below, and 75 % are younger than 44. Thus UI users tend to be younger than the labour force as a whole, as other authors have also reported. On average, each individual filed 3.43 claims between 1980 and 2003. Of the 96,413 individuals, 27 % filed only 1 claim over that 18 year period, while 8.3 % had 2 claims, 13.2 % had 3 claims, 9.1 % had 4 claims, and 32.4 % filed 5 or more claims. Among the 330,995 claims in our sample, approximately 30.3 % of them were filed by individuals who were repeat users at the time of observation. A somewhat lower share of claims (23.2 %) is accounted for by individuals who were seasonal users at the time, and approximately 51 % of claims were filed by individuals who at one point of time over the course of the interval were seasonal claimants.12 For approximately 1 % of individuals, program administrators did not record the gender. This reporting anomaly occurred during a brief period in the 1990s. 12 With the aid of an algorithm, we identify seasonal claimants according to their observed claims patterns. Using the current claim as a reference point, if the previous 2 claims filed within a 5-year window occurred at the same time of the year (give or take 4 weeks on either side of the initiation point), the individual is considered to be a seasonal user. He/she might not necessarily remain so for the duration of our interval. 11 11 V. Empirical Analysis V.1 Outcome measures and Cross-tabulations The phenomenon of interest is the ‘sophistication of UI use’, meaning the degree to which a claimant has fully benefited from the provisions of the program. Given this perspective, the empirical objective is to investigate the extent to which UI use patterns evolve as the individual files more and more claims. We deal with three outcomes pertaining to UI claims that are associated primarily with repeat use of the UI regime and for which a thorough knowledge of the regulations and provisions of the program would be beneficial. The basic empirical approach is to link indicators for these particular attributes to the extent of the individual’s prior exposure to the UI regime. It is fairly straightforward to investigate the extent to which some adjustment process is occurring. It is more challenging to investigate whether learning effects – either individual or social - underlie an observed adjustment process. The three outcome variables for facets of UI claims are the following. 1. The number of weeks of insurable earnings beyond the minimum entrance requirement. (WKSDIFF1) As mentioned above, many frequent claimants may have an interest in minimizing the number of weeks of work provided that they have a sufficient number to qualify. 2. The number of work weeks combined with benefit weeks beyond the minimum number required to cover the complete working cycle (WKSDIFF2) This measure refers to the slack in benefit weeks in excess of the number of calendar weeks remaining before the start of the next working season. A zero value means that the individual has exactly the number of EI benefit weeks that are required to cover his/her lost earnings before the start of his/her next working season, so the entire cycle is covered by either EI benefits or paid 12 employment. As mentioned above, claimants have an interest in minimizing this variable. 3. The number of active-claim weeks of zero benefit. (WEEKSZERO) As explained above, frequent claimants have an interest in maximizing this variable for several reasons. This indicator does not include the obligatory two-week waiting period. Some descriptive statistics for these indicators are presented in table 1 below. The distributions for most of them are skewed, highlighting the well-known fact that UI use patterns tend to be very heterogeneous. In table 2, means that are cross-tabulated according to seasonal user status are presented in order to give some idea of the extent to which a frequent or occasional user might resort to the provisions that are associated with those outcome measures. Given the large sample size, these discrepancies are statistically significant. As expected, the values for WKSDIFF1 and WKSDIFF2 are lower for seasonal users than for non-seasonal users, while the reverse is true for WEEKSZERO. We calculate the cross-tabulated means of these three indicators according to the number of claims that the individual had filed in the past. The measure of prior exposure to the UI regime is called the claim sequence number (CSN). It assumes a value of unity for the first claim, and ranges from 1 to 10 within our sample.13 The results are presented in table 3. For WKSDIFF1, the results are as expected: there is a monotonic, decreasing relationship with CSN, implying that as the claimant gains experience with the UI regime, there is a tendency to reduce the number of work weeks exceeding the minimum entry requirements. Although the results are presented only for the entire sample, the patterns for men and for women examined Our sample contains observations for CSN with values as high as 23. Some of these very high values represent reporting anomalies stemming from administrative coding, as normally it is not possible to register more than 1 claim per year. For that reason as well as for presentational convenience, we recode the variable CSN by assigning a value of 10 for any observation greater than 10, which comprises about 7 % of all observations. This simplification of the claim sequence number variable should preserve the rank ordering of individuals, and thus serve as a suitable indicator for the frequency of EI use. 13 13 separately are similar. These finding could be spurious, however, for two reasons. First, due to the nature of their employment patterns, repeat claimants or seasonal workers tend to have much shorter employment spells. The fact that lower values for WKSDIFF1 are observed for those with more claims could partially reflect a mechanical effect as opposed to a behavioural effect. Second, there is another compositional issue that we mention presently. For WKSDIFF2, the results are also as expected: there is a monotonic, decreasing relationship with CSN, implying that as the claimant gains experience with the EI regime, there is a tendency to reduce the number of slack weeks of UI benefits. Although the results are presented only for the entire sample, the pattern for men is similar to that for women. As in the case for WKSDIFF1, there are some confounding compositional issues. In graphical analysis that we do not show, for seasonal workers we discern significant spikes for WKSDIFF1 at the point of minimal qualification requirements and for WKSDIFF2 at the point of ‘maxing out the year’. These finding are consistent with those of Green and Sargent [1998], which are based on data drawn from the Labour Market Activity Survey. For WEEKSZERO, the results are as expected: there is a monotonic, increasing relationship with CSN, implying that as the claimant gains experience with the UI regime, there is a tendency to increase the number of weeks of active claim status during which no benefits are received. Although the results are presented only for the entire sample, the pattern for men is similar to that for women. While the figures that are presented in Table 3 indicate the existence of strong patterns associated with the CSN indicator, it is likely that they also reflect compositional effects. By definition any individual that is observed to have a high value for CSN has had considerable exposure to the UI system and is thus a relatively heavy user, while many of the individuals with 14 values of 1 or 2 are relatively light users of the UI system. The current sample thus contains users with quite heterogeneous employment and benefit use patterns. The sample upon which the figures for the first few rows of Table 3 are based reflect a mix of one-time users, occasional users, and frequent users. The sample corresponding to the last few rows consists mostly of frequent users and is much more homogenous. Thus, as one compares numbers down a column, one is confounding two effects; the difference in behaviour across claims for a given individual, which is the effect of interest for this paper, and the difference between frequent and infrequent users. To address this compositional issue, the analysis presented in Table 3 is carried out with a restricted sample, namely those individuals having 10 or more claims over the entire period. By construction, this working sample is constant as one stratifies across claim sequence numbers, at least up to the eleventh claim. Although this procedure generates a selected sample, any empirical patterns that are discerned can be attributed to behavioural effects of some kind as the composition of the group is totally fixed, while the degree of exposure to the UI system varies. These figures, presented in Table 4, have the same general pattern for WEEKSZERO. However, for the measures of WKSDIFF1 and WKSDIFF2, there appears to be an immediate levelling off after the first claim, instead of the gradually decreasing relationship that was found for the full sample presented in Table 3. We carried out the same calculations based on a lower threshold for the number of claims (8 or more) and obtained similar results. V.2 The Regression Model Three regression equations are estimated using the three measures of UI use (WKSDIFF1,WKSDIFF2,WEEKSZERO) as the dependent variables. We focus on the estimated coefficients of the claim sequence number (CSN) taking into account the independent effects of numerous other influences and control variables. The first proposition that we 15 investigate is whether UI claimants display a tendency to approach their optimal UI claim as they file additional claims, with the various facets of that claim converging toward desired levels. All equations were estimated separately for men and women, but since the results were virtually qualitatively identical and quantitatively similar, only the pooled male-female results are presented and discussed here. Pooling the two genders together also yields degrees of freedom that are useful in identifying certain estimated coefficients.14 The primary set of regressions take the form of linear equations, and are estimated using OLS. The estimated standard errors for all equations are adjusted for the clustering of disturbance terms for each of these individuals. Because all three of the dependent variables are limited dependent variables denoting counts, they are characterized by significant clusters at zero weeks and other points. To take account of the resultant heteroskedasticity, the equations are also estimated using the negative binomial count specification. The regression results are quite robust to this change in specification. The control variables include indicators for age, province, UI program generosity, broad sector of the economy, calendar year, and the local unemployment rate. The effect of age is captured in semi-parametric fashion by a series of 8 categorical variables.15 Each province has its own categorical variable, with Ontario serving as the omitted category.16 The 12 broad industrial sectors of the economy are agriculture, forestry, fishing & trapping, mining & energy, construction, transportation & communications & utilities, manufacturing, trade, finance & insurance & real estate, business & personal service, public administration, and unspecified. The omitted category is manufacturing. Year-specific effects are modelled as binary variables for each Although the overall sample size is large, there are a few instances where the number of degrees of freedom might cause concern. For instance, we estimate effects that are specific to filing the first claim, the second claim, the third claim, etc., and these effects are subsequently interacted with the time period before and the time period after 1993. This parametrization can result in some small cell sizes. 15 The omitted category is the youngest workers aged 18-24. The included groups are: 25-29 years, 30-34 years, 3539 years, 40-44 years, 45-49 years, 50-54 years, 55-59 years, and 60-65 years, 16 All 3 of the territories are combined into one group. 14 16 year from 1980 until 2003, with 1997 serving as the omitted category. These are designed to capture global effects such as changes in aggregate labour market conditions. The program generosity effects are specified as a series of 12 categorical variables that are derived from the official unemployment rate for each UI administrative region that prevailed at the time.17 The UI program eligibility parameters are determined by the administrative document (the so-called ‘entitlement matrix’) that specifies, for each of the brackets of regional unemployment rates, the number of benefit weeks that are awarded for a given number of hours of qualifying employment. In this fashion the two eligibility parameters, namely the maximum benefit duration and the minimum entry requirement, are determined by the respective bracket containing the regional unemployment rates. The omitted category is the bracket with the highest unemployment rates, which is 16 % and above. Relative to all other regions, the program’s provisions are more generous for the omitted category. Note that we also include the regional unemployment rate variable in continuous form as well in an attempt to capture the role of local labour market conditions. Although by construction the eligibility variables are quite collinear with the continuous unemployment rate variable, there is some margin for variation in the continuous local unemployment rate variable that is independent of the generosity parameters.18 The key exogenous variable CSN is parametrized by a set of binary variables for the first, second, third, fourth, and fifth claim, with the last one serving as the omitted category (CLAIM1, CLAIM2, CLAIM3, CLAIM4, CLAIM5). This flexible specification allows for non-linear patterns. One might expect the adjustment process to occur primarily over the first 4 claims, and 17 Specifically, they are 0-6 %, 6-7 %, 7-8 %, 8-9 %, 9-10 %, 10-11 %, 11-12 %, 12-13 %, 13-14 %, 14-15 %, 15-16 %, and above 16 %. 18 For instance, the local unemployment rate varies between 0 and 6 % for the least generous program category, and it varies between 16 % and all unemployment rates above 16 % for the most generous category. Furthermore, the EI eligibility parameters are constant within each of these categories. This means that when the local unemployment rate varies within any of the other 14 brackets, EI eligibility parameters are invariant. 17 to diminish thereafter. In other words, we anticipate that the marginal returns to filing a second and a third claim would be positive, but diminishing. Furthermore, it seems reasonable to expect an uneven but generally monotonic step function as claimants converge to their optimal UI claim configuration. The claim sequence variables pose some endogeneity issues because by construction, it differentiates between the frequent users and the occasional users. That distinction is relevant for the dependent variables, as frequent users have a greater inherent disposition to make use of the relevant provisions given their seasonal employment patterns. The effect of the claims sequence number variable captures not only effects within individuals, but also compositional effects, as some individuals become frequent users while others do not, thereby leading to over-estimation of the adjustment effects. We considered several strategies for dealing with this issue. It is not appropriate to divide the estimating sample according to frequent versus non-frequent users or seasonal versus nonseasonal users, as that is a choice variable that is associated with endogenous variables. It is difficult, particularly with administrative data, to find instruments that would help to determine repeat use status but would be independent of the dependent variables. Nevertheless, we do include exogenous variables that should at least partially control for whether one is a frequent or an occasional user. First, there is the complete set of measures to control for the program eligibility parameters, which in turn have a strong impact on the frequency of claims. There is also the set of industrial indicators, in particular construction, agriculture, forestry, and fishing, in which very proportionate numbers of repeat UI claimants work. We also estimate a version of these equations that includes a fixed effect for each individual worker. This specification has the advantage of conditioning out all forms of individual heterogeneity, such as the type of inherent UI 18 user the worker is, cohort effects, the effect of age at first claim, or the effect of idiosyncratic employment patterns. The disadvantage is that there are few degrees of freedom available to estimate individual fixed effects for light users of UI. Furthermore, this estimator often soaks up much of the variation that might contribute to identifying the coefficient of the variable of interest. V.3 Learning Effects If the empirical profile of UI claims do exhibit a monotonic convergence pattern, that would constitute evidence in favour of the proposition that workers and perhaps firms are adjusting the attributes of a succession of UI claims such that the benefits they receive are rising. One cannot conclude much, however, in regards to the underlying causes to such a convergence pattern, which could be due to omitted factors that might have been influencing all claimants over the estimating interval and contributing to the observed convergence patterns. We search for evidence of individual learning effects consistent with the phenomenon of claimants learning by doing through experience in filing UI claims and thus gathering relevant information regarding the program parameters and the set of available jobs. We adopt a difference-in-difference strategy that compares the claims profiles before and after the occurrence of a major event during the estimating interval that could have had an influence on the learning processes. Major reforms to the UI regime were implemented in 1994 (bill C-17) and 1996 (bill C-12). The former piece of legislation significantly reduced the generosity of the regime by lengthening qualifying periods and shortening maximum benefit periods were shortened. The latter piece of legislation changed the qualifying requirements such that they were based on hours worked rather than weeks worked. While that change did not render it harder to qualify for benefits, evaluative research has indicated that it incited major changes to workers’ employment 19 patterns.19 The year 1994 and thereafter is considered to be the post-reform period, as the changes that were implemented prior to 1994 were relatively minor.20 We conjecture that these two changes ‘pulled the rug out from under’ the constellation of employment patterns and their concomitant UI claim patterns that had developed gradually since the last major changes that had been implemented in 1971.21 Much of the information upon which the underlying labour supply and demand choices were based changed in a fairly significant fashion, which necessitated a fair amount of learning activity in order to adapt these choices to the new regime’s parameters and provisions so that the claims could be optimized. Given this demarcation into the pre-reform and post-reform periods, we estimated the claims profile during each period. During both periods, we do discern what appear to be adjustment profiles that converge to certain values. The question is whether the adjustment process decelerated after the reforms were implemented, perhaps because all users had to become informed about the new parameters of the program and adjust their employment activities accordingly. We calculate differences between the two periods in the adjustment profiles as the claims sequence unrolls. The demarcation indicator assumes a value of 0 from 1980 until 1993, and a value of unity thereafter. The coefficient of this regressor would capture an intercept-shift effect reflecting the average level of one of the three outcome variables during the earlier period compared to the average level during the latter period. Statistically it was not possible to identify this effect due to collinearity with the set of binary variables reflecting the year-specific effects. It is feasible, however, to interact the post-reform dummy variable with the claim sequence e.g. Friesen and Maki (2000), Green and Riddell (20000, Jones (2000), and LaCroix and Van Audenrode (2000) There were actually three major changes to UI/EI legislation over the early to mid 1990s. Bill C-113 was passed in 1993; the major change was that quitters were totally disentitled. In addition to the change from the weeks-based system to the hours-based system, Bill C-12 included a total of about 10 other changes to qualifying conditions and benefit calculation periods, which are summarized in Gray (2003). Most of the changes that have been implemented since 1996 are ‘pilot projects’ that do not amount to significant reform (Gray (2006b)). 21 We implicitly assume the political and economic forces which brought about the major change in policy are exogenous to the generation of the outcome variables. 20 19 20 indicators described above. These interactions generate five additional binary variables for the event of filing a claim after the reforms had been implemented (CLM1REF, CLM2REF, CLM3REF, CLM4REF, CLM5REF). For example, CLM1REF assumes a value of unity if an individual filed his/her first claim after 1993, and zero otherwise (which implies that the individual filed the first claim during the earlier period). CLM2REF assumes a value of unity if an individual filed his/her second claim after 1993, and zero otherwise (which implies that the individual filed the second claim during the earlier period.)22 We employ this parametrization in order to search for discrepancies in the claims profiles between the two periods. For each of the three specific UI outcomes that are described above, we focus on the gradient of the path as the claimant files his/her first claim, second claim, third claim, etc., which in our specifications takes the form of a step function. If differences in these differences – the step between claim #1 and claim #2, between claim #2 and claim #3, etc. - are discerned between the two periods, the next question is the sign of these discrepancies. The relearning scenario that was discussed above would constitute a setback that we would expect to decelerate the adjustment process to the optimal claim. One would thus expect the step function over the sequence of claims to be less steep after the reform period. To summarize the interpretation of the point estimates, the estimated coefficients of the claim sequence indicators (CLAIM1, CLAIM2, CLAIM3, CLAIM4) show the effect of filing one’s first claim before 1994 relative to the omitted category of claim number 5 (or higher). The focus is on the profile and its gradient as the individual files his/her sequence of claims. We 22 To give an idea of the number of degrees of freedom that are available in order to identify these deviation effects corresponding with the before/after reform period, CLM1REF has a non-zero value for 6.8 % of all observations. CLM2REF has a non-zero value for 5.4 % of all observations. CLM3REF has a non-zero value for 4.6 % of all observations. CL4REF has a non-zero value for 3.7 % of all observations, and CLM5REF has a non-zero value for 18.4 % of observations 21 generally expect to discern an adjustment profile of increments for the measure of WEEKSZERO, for which claimants have an incentive to attain higher values, and decrements for the measures of WKSDIFF1 and WKSDIFF2, for which the claimant has an incentive to attain lower values. The sums of each of the estimated coefficients of (CLAIM1, CLAIM2, CLAIM3, and CLAIM4) and their respective counterparts (CLM1REF, CLM2REF, CLM3REF, and CLM4REF) depict the same adjustment profile during the later period. The estimated coefficients of CLM1REF, CLM2REF, CLM3REF, and CLM4REF show the deviations in these effects between the two periods. If these estimated coefficients are statistically significant, the effect of filing one’s first, second, third, or fourth claim before 1994 differs from the corresponding effect observed after 1993. If the deviation is negative (positive), the average from 1980 to 1993 is lower (higher) than the average from 1994 to 2003. In the former case, the adjustment profile has shifted downwards during the later period, and the claimants tend to be closer to their optimal EI claim given a particular claim on the claim sequence profile. The learning effects, however, are reflected not in the position of these claim sequence profiles, but the evolutions between claims. After the implementation of the reforms, one might expect for the adjustment process to be slower: in effect, the returns to claims experience would diminish somewhat. V.4 Regression Results The regression results are presented in Tables 4 through 6. For each of the three measures of UI outcomes, the equations were first estimated including only the controls (not shown) and excluding the CSN indicators. For most of the control variables, the estimated coefficients are consistent with prior expectations and findings drawn from the existing literature. These expectations vary across the three attributes of UI claims. Although we do not delve into the details of each case, we mention a few generalizations. More generous UI eligibility parameters 22 tend to be associated with either higher values in those cases where the claimant has an incentive to increase the value of the UI claim attribute, and vice versa. The same pattern applies for the highly seasonal sectors of agriculture, forestry, fishing, mining, and construction, as well as transportation and public administration in the cases of WKSDIFF1 and WKSDIFF2. The estimated coefficients of the age variables tend to be significant, but overall the patterns do not reveal major differences between the age categories. The familiar east-west pattern emerges from the set of provincial indicators: Newfoundland, PEI, Nova Scotia, New Brunswick, and Quebec are associated with either higher (lower) than Ontario, Manitoba, Saskatchewan, Alberta, and BC for those EI outcomes for which higher (lower) values are desirable from the point of view of claimants. The results for the key variables for the claim sequence numbers are discussed in turn. The results for the outcome variable WKSDIFF1 are presented in Table 4. Claimants have some incentive to minimize this variable. Since claim number 5 is the omitted category, there is a monotonically decreasing effect as the individual files his/her first, second, third, fourth, and fifth claim, indicating a convergence effect during the pre-reform period. Successive claims are associated with lower values of WKSDIFF1. These point estimates are summarized in the second column of Table 7. The deviations between the values in the post-reform period from the prereform period are listed in column 4, and the net effect indicating the claim sequence profile during the post-reform period is listed in column 5. That profile also displays a decreasing trend. The deviations are significantly different from zero and negative, implying that this profile shifted downward, and hence closer to the desired claim attribute, after the reform period. The returns to marginal claims are listed in columns 3 (for the earlier period) and 6 (for the later period) of Table 7. In both cases diminishing returns to filing subsequent claims are discerned, as expected, but 23 these decrements are much larger in the pre-reform period, meaning that on average claimants in the earlier period made more rapid, larger graduations toward the optimal value of WKSDIFF1 than was the case in the post-reform period. The results for the outcome variable WKSDIFF2, which claimants have some incentive to minimize, are presented in Table 5. The empirical patterns are qualitatively identical and quantitatively similar to the case of WKSDIFF1. There is a monotonically decreasing effect as the individual files additional claims in both the pre-reform period (column 2 of Table 7) and the post-reform period (column 5 of Table 7). The deviations between the values in the post-reform period from the pre-reform period, which are all negative, are listed in column 4. The claim sequence profile shifted down after the pre-reform period. The returns to marginal claims are listed in columns 3 and 6 of Table 7. In both cases diminishing returns to filing subsequent claims are discerned, as expected, but these decrements are much larger in the pre-reform period, with the exception of the case of the fourth claim. The results for the outcome variable WEEKSZERO, which claimants have some incentive to maximize, are presented in table 6. The estimated coefficients of the claim sequence number indicators are all negative relative to the omitted category of claim number 5. The claim sequence number is positively associated with the number of zero benefit weeks, so successive claims indicate higher values for WEEKSZERO, with the exception of the transition between the first and the second claim during the post-reform period. The absolute value of the increments between claims is almost constant (approximately 0.8 weeks) during the earlier period, and much lower during the later period (approximately 0.3 weeks). This pattern of lower returns to filing subsequent claims is consistent with learning effects. In summary of the empirical results, in two different time periods occurring before and 24 after major reforms to the UI regime were enacted in the mid 1990s, there is evidence that UI beneficiaries gradually approach a desired value for a particular facet of their claim. In almost all cases that we analyze, a monotonic step function pattern is discerned, with each subsequent claim associated with either a decrement or an increment in the value of the respective attribute of the UI claim. There is also strong evidence to indicate that these adjustment profiles are situated more closely to the desired point during the period after the reforms. The estimated deviation effects are nearly always of the opposite sign than the point estimates from the pre-reform period, so the net effects are lower in magnitude during the post-reform period. For all three of these UI claim attributes, this adjustment profile shifts down in absolute value in the latter period. This indicates that UI claimants tend to start and end their adjustment processes closer to the desired level after 1993. The reasons for this finding are not apparent, but a possible interpretation is that it is capturing an average, global effect over the entire 1980-1993 period compared to the 1994-2003 period. A long-term gradual trend by which most claimants were making such adjustments is consistent with that finding. Finally, the empirical evidence reveals that in the cases of all three outcome variables, during the post-reform period the slope of these profiles is flatter. Claimants tend to approach them more slowly than before, as the increments or decrements between consecutive claims diminish in magnitude. VI. Summary and Conclusions The theme of this paper is the extent to which UI claimants change their use patterns as they gain experience in filing multiple claims. Given the multi-faceted nature of UI claims, we 25 focus on three outcome variables that represent specific attributes of UI claims, investigating empirical patterns within the claims profiles of repeat users. During the pre-reform period of 1980-1993 as well as the post-reform period of 1994-2003, there is evidence that beneficiaries approach a desired value for a particular facet of their UI claim as they file additional claims. Those strong empirical regularities suggest that for perhaps a variety of reasons, as recipients gain experience in drawing benefits from the UI regime, they are able to gradually increase the extent to which they benefit from it. There appears to be some process of habituation – in terms of adjusting UI claims and the concomitant employment patters - to the provisions and rules of the regime. Based on a comparison of the claims profiles between these two time periods, it was determined that the slope of these profiles became flatter after major reforms were enacted. This result is consistent with the presence of learning effects, as both firms and workers had to re-learn the provisions and regulations of the regime and adjust their labour market activity patterns accordingly, thus decelerating the adjustment process. This paper fits into a large empirical literature, based mostly on cross-sectional analysis, that shows that UI claimants and to some extent employers, react to UI program parameters and provisions by internalizing them into their employment patterns. The existing literature has demonstrated how the program distorts employment outcomes such as the lengths of job spells. Further research based on longitudinal data is warranted into how the UI regime affects a worker’s employment profile over long time horizons. A future challenge for the topic we address is to develop more effective techniques for treating unobserved heterogeneity in employment patterns such that one can fully account for the type of UI user – for which there is a fairly wide continuum - in the analysis of how claims evolve through repeat use. One possible research strategy would be to merge administrative data with employer-based survey data so that more detailed information on 26 the firm can be exploited. 27 References Anderson, P.M. and B.D. Meyer (1997) “Unemployment Insurance Take-up Rates and the AfterTax Value of Benefits” Quarterly Journal of Economics 112, pp. 913-938 Baker and Rae (1997) “Employment spells and Unemployment Insurance Eligibility” Review of Economics and Statistics 80, 1, pp. 80-94 Bertrand, M., Luttmer, E. and S. Mullainathan (2000) “Network Effects and Welfare Cultures” Quarterly Journal of Economics 115, p. 1019-1056 Borjas. G. and L. Hilton [1996] ”Immigration and the Welfare State: Immigrant Participation in Means Tested Entitlement Programs” Quarterly Journal of Economics pp. 575-604 Corak, M (1993a) “Is Unemployment Insurance Addictive? Evidence from the Benefit Durations of Repeat Users” Industrial and Labor Relations Review, 47, pp. 62-73 Corak, M (1993b) “Unemployment Insurance Once Again: The Incidence of Repeat Participation in the Canadian UI Program” Canadian Public Policy, 29, pp. 162-176 Corak, M. and Chen (2003) “Who Benefits from UI in Canada – Regions, Industries, Individual Firms?” Ottawa: Social Research and Demonstration Corporation WP # 0307 De Raaf, S., A. Motte, and C. Vincent (2004) Understanding Employment Insurance Claims Patterns Final Report of the Earnings Supplement Project Social Research and Demonstration Corporation De Raaf, S. and C. Vincent (2003) “Avons-nous le régime d’assurance chômage que nous voulons?” Policy Options, August, pp. 63-68 Friesen, J. and D. Maki (2000) “The Effect of Bill C-12 on Weekly hours of Work” Unemployment Insurance Evaluation series, Human Resources Development Canada Gray, D. (2001) “The Impact of Regional Ties and Workers’ Attitudes Toward Job search on EI Receipt” in Schwartz, S and A. Aydemir (eds.) Essays on the Repeat Use of Unemployment Insurance: The Earnings Supplement Project Ottawa: Social Research and Demonstration Corporation Gray (2003) “Employment Insurance: What Reform Delivered” CD Howe Backgrounder Gray, D (2006a) "Recent Changes to the Employment Insurance Regime: Encouraging Work Onclaim and Off-claim" Canadian Public Policy Journal 33, 1, pp. 99-110 Gray, D. (2006b) “"Has EI Reform Unraveled? Canada’s EI regime in the 2000s" CD Howe Backgrounder 28 Gray, D. and A. Sweetman (2001) “A Typology Analysis of the Users of Canada’s Unemployment Insurance System: Incidence and Seasonality Measures” in Schwartz, S and A. Aydemir (eds.) Essays on the Repeat Use of Unemployment Insurance: The Earnings Supplement Project Ottawa: Social Research and Demonstration Corporation Gray, D, and S. de Raaf (2002) “The Impact of the Allowable earnings provision on EI Dependency” Social research and Demonstration Corporation Working Paper series 02-05 Green, D. and C. Riddell (1995) “Unemployment Insurance and Employment Durations: Seasonal and Non-Seasonal Jobs” Unemployment Insurance Evaluation series, Human Resources Development Canada Green, D. and C. Riddell (1997) “Qualifying for Unemployment Insurance: An Empirical Analysis in Canada” Economic Journal 107, No. 440, pp. 67-84. Green, D. and C. Riddell (2000) “The Effects of the shift to an hours-based Entrance Requirement” Employment Insurance Evaluation series, Human Resources Development Canada Green, D. and T. Sargent (1998) “Unemployment Insurance and Job Durations: Seasonal and nonSeasonal Jobs” Canadian Journal of Economics 31, pp. 247-278 Human Resources and Development Canada (1994) Improving Social Security in Canada: A Discussion Paper Jones, S (2000) “EI Impacts on Unemployment durations and Benefit receipt” Unemployment Insurance Evaluation series, Human Resources Development Canada Kuhn, P. and C. Riddell (2006), “The Long-Term Effects of a Generous Income Support Program: Unemployment Insurance in New Brunswick and Maine, 1940-1991”. LaCroix and Van Audenrode (2000) “An Assessment of Various Components of Bill C-12 on the Duration of Unemployment Spells” Unemployment Insurance Evaluation series, Human Resources Development Canada Lemieux, T and B. Macleod (1995) “State Dependence and Unemployment Insurance” Human Resources and Development Canada, Unemployment Insurance Evaluation Series Lemieux, T and B. Macleod (2000) “Supply side Hysteresis: the Case of the Canadian Unemployment Insurance System” Journal of Public Economics 78, pp. 139-170 Manski, C. [2000] “Economic analysis of Social Interactions” Journal of Economic Perspectives 14, 3 pp. 115-35 McCall, B. (2000) “Repeat Use of Unemployment Insurance” in Bassi, L. and S. Woodbury (eds.) Research in Employment Policies, Vol. 2, JAI Press, pp. 25-47 29 Meyer, B. and D. Rosenbaum (1996) “Repeat Use of Unemployment Insurance” National Bureau of Economic Research WP # 5423 Nakamura, A. and W.E. Diewert (1997) “Unemployment Insurance in Canada: Problems and Recent Reforms” discussion paper, faculty of business, University of Alberta Nakamura, A.O. and W.E. Diewert (2000), “Insurance for the Unemployed: Canadian Reforms and their Relevance for the United States”, pp. 217-247 in Long-Term Unemployment and Reemployment Policies, L.J. Bassi and S.A. Woodbury (eds.), Stamford Connecticut: JAI Press. OECD (1999) The Battle Against Exclusion, Volume 3: Social Assistance in Canada and Switzerland Organisation for Economic Cooperation and Development Schwartz, S., W. Bancroft, D. Gyarmati, and C. Nicholson (2001) The Frequent Use of Unemployment Insurance in Canada: The Earnings Supplement Project Ottawa: The Social Research and Demonstration Corporation Wesa, L. (1995), “Seasonal Employment and Repeat Use of Unemployment Insurance”, HRDC Evaluation Brief #24 30 Table 1 Descriptive Statistics for the EI outcomes associated with learning effects EI outcome (mnemonic) WKSDIFF1 Description Number of insurable weeks beyond the minimum required for eligibility for benefits Descriptive Statistics Mean: 22.5 Median: 22 Std. dev.: 13.2 25th pct.:11 75th pct.: 35 Mean: 12.6 Median: 11 Std. dev.: 11.3 25th pct.: 0 75th pct.: 24 Mean: 6.44 Median: 2 Std. dev.: 9.4 25th pct.: 0 75th pct.: 9 WKSDIFF2 WEEKSZERO Number of weeks of EI benefit entitlement beyond the number required to cover the yearly cycle of a seasonal user Number of weeks during an active EI claim for which the payout is zero Table 2 Average Values for Indicators of UI/EI Use Patterns Average for Non-Seasonal WKSDIFF1 26.3 weeks WKSDIFF2 19.2 weeks WEEKSZERO 4.8 weeks Source: HRSDC Administrative Data. Indicator of UI/EI Outcome Average for Seasonal 18.8 weeks 14.3 weeks 8.0 weeks Notes: A seasonal use pattern is defined as follows: taking the current claim as a reference point, if the individual’s prior 2 claims filed within the past 5 calendar years occurred at approximately the same time of the year (i.e. plus or minus 4 weeks), the individual is classified as a seasonal user for that year. For the purposes of this table, the individual is a seasonal user is considered to be a seasonal user for the entire interval. 31 Table 3 Mean Values of UI/EI Measures by Claim Sequence Number CSN WKSDIFF1 WKSDIFF2 1 27.2 19.1 2 23.2 17.8 3 22.1 17.1 4 21.1 16.4 5 20.1 15.7 6 19.3 15.1 7 18.6 14.6 8 17.7 14.1 9 17.2 14.0 10+ 17.1 13.4 Source: HRSDC Administrative Data. WEEKSZERO 4.8 5.6 6.3 7.1 7.7 8.1 8.3 8.5 8.5 8.7 32 Table 4 Regression results for the UI/EI outcome of number of weeks in the qualifying period beyond the minimum entrance requirement (WKSDIFF1) Variable Claim History (reference category is 5+ claims) First claim Second claim Third claim Fourth claim First claim - current claim is in EI reform period Second claim - current claim is in EI reform period Third claim - current claim is in EI reform period Fourth claim - current claim is in EI reform period Number of claims during period Industry (reference category is Manufacturing) Agriculture Forestry Fishing Mining Construction Transportation Trade Financial Services Commercial Public Service Not Stated Labour Market Conditions Unemployment Rate (UE) UE intervals (reference category is UE rate > 13%) UE rate < 6% UE rate between 6 and 7% UE rate between 7 and 8% UE rate between 8 and 9% UE rate between 9 and 10% UE rate between 10 and 11% UE rate between 11 and 12% UE rate between 12 and 13% Age (omitted category is age between 18 and 24) Age between 25 and 29 Age between 30 and 34 Age between 35 and 39 Age between 40 and 44 Coefficient t-statistic P-value 8.782 3.960 2.502 1.580 -5.477 -2.589 -1.536 -1.316 -0.381 68.87 33.91 22.64 14.81 -38.75 -17.39 -10.04 -8.27 -35.22 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -8.355 -4.835 -6.630 0.512 -3.540 -1.300 -0.419 0.525 -2.091 -3.401 -1.942 -0.316 -40.38 -18.04 -17.19 2.34 -33.29 -9.14 -4.43 3.61 -24.68 -24.44 -12.66 -15.54 0.000 0.000 0.000 0.019 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -4.179 -3.351 -2.382 -1.291 -0.367 0.067 0.459 0.041 -17.85 -16.05 -12.74 -7.74 -2.46 0.49 3.74 0.35 0.000 0.000 0.000 0.000 0.014 0.622 0.000 0.727 3.368 4.279 4.593 4.969 44.14 50.19 50.36 50.83 0.000 0.000 0.000 0.000 33 Age between 45 and 49 Age between 50 and 54 Age between 55 and 59 Age between 60 and 64 Province of Residence (reference category is Ontario) Newfoundland and Labrador Prince Edward Island Nova Scotia New Brunswick Quebec Manitoba Saskatchewan Alberta British Columbia Territories (combined) Year Indicators (reference year is 1997) Year = 1980 Year = 1981 Year = 1982 Year = 1983 Year = 1984 Year = 1985 Year = 1986 Year = 1987 Year = 1988 Year = 1989 Year = 1990 Year = 1991 Year = 1992 Year = 1993 Year = 1994 Year = 1995 Year = 1996 Year = 1998 Year = 1999 Year = 2000 Year = 2001 Year = 2002 Year = 2003 Intercept 4.823 4.444 3.927 3.803 45.11 37.85 29.83 23.28 0.000 0.000 0.000 0.000 -4.950 -5.194 -2.233 -4.323 -1.464 0.058 -0.178 -0.012 -0.696 1.499 -22.04 -13.99 -12.24 -21.26 -17.33 0.36 -0.99 -0.12 -6.66 2.66 0.000 0.000 0.000 0.000 0.000 0.721 0.321 0.906 0.000 0.008 -14.116 -7.154 -4.319 -5.297 -4.174 -3.520 -2.476 -2.077 -1.401 -0.936 -0.373 -3.926 -3.985 -4.042 -2.522 -2.792 -1.640 0.635 0.515 0.594 0.960 0.356 -0.132 28.316 -14.61 -31.43 -21.98 -27.22 -22.43 -20.34 -14.74 -12.46 -8.51 -5.76 -2.36 -25.61 -26.13 -26.70 -18.11 -21.41 -12.59 4.89 3.81 4.25 7.07 2.59 -0.92 78.16 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.018 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.010 0.357 0.000 34 Table 5 Regression results for the UI/EI outcome of number of weeks of benefit entitlement beyond the number required to cover the non-working season (WKSDIFF2) Variable Claim History (reference category is 5+ claims) First claim Second claim Third claim Fourth claim First claim - current claim is in EI reform period Second claim - current claim is in EI reform period Third claim - current claim is in EI reform period Fourth claim - current claim is in EI reform period Number of claims during period Industry (reference category is Manufacturing) Agriculture Forestry Fishing Mining Construction Transportation Trade Financial Services Commercial Public Service Not Stated Labour Market Conditions Unemployment (UE) Rate UE rate intervals (reference category is UE rate > 16%) UE rate < 6% UE rate between 6 and 7% UE rate between 7 and 8% UE rate between 8 and 9% UE rate between 9 and 10% UE rate between 10 and 11% UE rate between 11 and 12% UE rate between 12 and 13% UE rate between 13 and 14% UE rate between 14 and 15% UE rate between 15 and 16% Age (omitted category is age between 18 and 24) Coefficient t-statistic P-value 5.805 3.428 2.427 1.605 -4.832 -2.816 -2.109 -1.676 -0.363 51.69 31.99 22.59 14.54 -40.00 -22.02 -15.47 -11.36 -43.19 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -5.411 -3.107 -4.757 0.208 -2.825 -1.143 -0.342 0.360 -1.943 -2.284 -1.186 -0.103 -26.56 -13.40 -10.79 1.15 -33.30 -10.14 -4.66 3.22 -30.00 -20.36 -9.23 -3.48 0.000 0.000 0.000 0.248 0.000 0.000 0.000 0.001 0.000 0.000 0.000 0.001 -4.308 -4.091 -3.655 -2.757 -2.497 -1.825 -0.754 -0.757 -0.064 -0.054 0.056 -10.39 -10.95 -10.62 -8.72 -8.62 -6.92 -3.15 -3.49 -0.33 -0.29 0.33 0.000 0.000 0.000 0.000 0.000 0.000 0.002 0.000 0.743 0.769 0.745 35 Age between 25 and 29 Age between 30 and 34 Age between 35 and 39 Age between 40 and 44 Age between 45 and 49 Age between 50 and 54 Age between 55 and 59 Age between 60 and 64 Province of Residence (reference category is Ontario) Newfoundland and Labrador Prince Edward Island Nova Scotia New Brunswick Quebec Manitoba Saskatchewan Alberta British Columbia Territories (combined) Year Indicators (reference year is 1997) Year = 1980 Year = 1981 Year = 1982 Year = 1983 Year = 1984 Year = 1985 Year = 1986 Year = 1987 Year = 1988 Year = 1989 Year = 1990 Year = 1991 Year = 1992 Year = 1993 Year = 1994 Year = 1995 Year = 1996 Year = 1998 Year = 1999 Year = 2000 Year = 2001 Year = 2002 Year = 2003 Intercept 2.697 3.254 3.410 3.700 3.459 3.244 3.158 3.039 39.91 44.72 44.37 45.60 39.21 34.01 29.80 23.32 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -1.346 -1.927 -1.022 -1.434 -0.894 0.017 -0.352 -0.411 -0.583 0.494 -7.14 -5.71 -7.15 -9.06 -13.34 0.14 -2.46 -4.98 -7.46 0.91 0.000 0.000 0.000 0.000 0.000 0.891 0.014 0.000 0.000 0.361 -6.781 -5.674 -5.250 -5.997 -5.400 -4.914 -3.944 -3.462 -2.834 -2.356 -1.477 0.582 0.588 0.431 0.394 -0.798 -0.377 0.225 -0.224 -0.298 0.085 -0.168 -0.471 21.074 -6.93 -29.48 -31.70 -35.56 -33.52 -32.59 -27.25 -24.21 -19.97 -16.80 -10.78 4.16 4.17 3.05 3.06 -6.84 -3.24 1.95 -1.90 -2.47 0.73 -1.42 -3.87 36.55 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.002 0.002 0.000 0.001 0.051 0.058 0.014 0.463 0.155 0.000 0.000 36 Table 6 Regression results for the UI/EI outcome of number of benefit weeks for which no benefits were collected (WEEKSZERO) Variable Claim History (reference category is 5+ claims) First claim Second claim Third claim Fourth claim First claim - current claim is in EI reform period Second claim - current claim is in EI reform period Third claim - current claim is in EI reform period Fourth claim - current claim is in EI reform period Number of claims during period Industry (reference category is Manufacturing) Agriculture Forestry Fishing Mining Construction Transportation Trade Financial Services Commercial Public Service Not Stated Labour Market Conditions Unemployment (UE) Rate UE rate intervals (reference category is UE rate > 16%) UE rate < 6% UE rate between 6 and 7% UE rate between 7 and 8% UE rate between 8 and 9% UE rate between 9 and 10% UE rate between 10 and 11% UE rate between 11 and 12% UE rate between 12 and 13% UE rate between 13 and 14% UE rate between 14 and 15% UE rate between 15 and 16% Age (omitted category is age between 18 and 24) Age between 25 and 29 Coefficient t-statistic P-value -1.647 -1.207 -0.736 -0.185 1.545 0.855 0.419 0.027 0.205 -15.04 -12.32 -8.16 -2.16 12.67 7.02 3.36 0.21 18.48 0.000 0.000 0.000 0.031 0.000 0.000 0.001 0.831 0.000 -4.340 -1.317 -3.308 -0.796 -0.373 -0.849 -2.226 -0.994 -2.046 -2.462 -0.786 -0.027 -28.23 -5.42 -8.22 -4.00 -3.40 -5.91 -27.47 -7.97 -25.63 -19.50 -6.41 -1.14 0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.253 -1.231 -0.870 -0.862 -0.509 -0.450 -0.274 -0.007 0.124 0.250 0.325 0.339 -3.83 -2.99 -3.20 -2.06 -1.99 -1.32 -0.04 0.74 1.68 2.40 2.68 0.000 0.003 0.001 0.039 0.047 0.186 0.970 0.461 0.093 0.016 0.007 0.169 3.14 0.002 37 Age between 30 and 34 Age between 35 and 39 Age between 40 and 44 Age between 45 and 49 Age between 50 and 54 Age between 55 and 59 Age between 60 and 64 Province of Residence (reference category is Ontario) Newfoundland and Labrador Prince Edward Island Nova Scotia New Brunswick Quebec Manitoba Saskatchewan Alberta British Columbia Territories (combined) Year Indicators (reference year is 1997) Year = 1980 Year = 1981 Year = 1982 Year = 1983 Year = 1984 Year = 1985 Year = 1986 Year = 1987 Year = 1988 Year = 1989 Year = 1990 Year = 1991 Year = 1992 Year = 1993 Year = 1994 Year = 1995 Year = 1996 Year = 1998 Year = 1999 Year = 2000 Year = 2001 Year = 2002 Year = 2003 Intercept 0.551 0.855 1.075 1.382 1.628 0.942 -0.015 8.33 11.42 12.90 14.55 14.87 8.06 -0.11 0.000 0.000 0.000 0.000 0.000 0.000 0.914 1.762 0.016 2.156 2.232 1.763 -1.284 -0.732 0.221 1.888 -1.309 8.82 0.06 12.20 11.60 21.50 -10.94 -5.06 2.37 19.12 -4.14 0.000 0.956 0.000 0.000 0.000 0.000 0.000 0.018 0.000 0.000 -0.927 -0.056 0.032 -0.041 0.345 0.345 0.599 0.436 0.175 0.124 -0.096 -0.205 0.020 0.591 0.360 0.262 0.253 -1.253 -1.852 -1.991 -2.160 -2.363 -4.039 6.274 -1.76 -0.32 0.19 -0.25 2.18 2.33 4.15 3.05 1.26 0.91 -0.73 -1.61 0.16 4.63 3.11 2.41 2.37 -12.46 -17.62 -18.31 -20.45 -22.12 -40.29 13.69 0.079 0.751 0.847 0.803 0.029 0.020 0.000 0.002 0.207 0.362 0.466 0.108 0.876 0.000 0.002 0.016 0.018 0.000 0.000 0.000 0.000 0.000 0.000 0.000 38 Table 7 Evidence of Learning – Summary of Point Estimates 1980 to 1993 (Base Period – pre-form) (Col. 2) Change From Previous Claim (Col. 3) 4.8 1.5 0.9 2.4 1.0 0.8 -0.4 -0.5 -0.5 Deviation between the two periods (Col. 4) -5.5 -2.6 -1.5 -1.3 -4.8 -2.8 -2.1 -1.7 1.5 0.9 0.4 0.0 After 1993 (postreform) (Col. 5) 3.3 1.4 1.0 0.3 1.0 0.6 0.3 -0.1 -0.1 -0.3 -0.3 -0.2 Change From Previous Claim (Col. 6) 1.9 0.4 0.7 0.4 0.3 0.4 0.2 0.0 -0.1 WKSDIFF1 Claim1 8.8 Claim2 4.0 Claim3 2.5 Claim4 1.6 WKSDIFF2 Claim1 5.8 Claim2 3.4 Claim3 2.4 Claim4 1.6 WEEKSZERO Claim1 -1.6 Claim2 -1.2 Claim3 -0.7 Claim4 -0.2 Source: HRSDC Administrative Data. 39 Table A1 Distribution of person-claim observations (n=330,995 claims) Year Percent Cumulative Percent 1980 0.06 0.06 1981 2.24 2.30 1982 3.75 6.06 1983 3.82 9.88 1984 4.29 14.16 1985 5.09 19.26 1986 5.21 24.46 1987 5.00 29.47 1988 5.05 34.52 1989 4.98 39.50 1990 5.41 44.92 1991 5.60 50.52 1992 5.51 56.03 1993 5.04 61.07 1994 4.05 65.12 1995 4.70 69.82 1996 4.34 74.16 1997 4.16 78.32 1998 3.75 82.07 1999 3.49 85.56 2000 3.47 89.03 2001 3.87 92.89 2002 3.72 96.61 2003 3.39 100.00 Source: HRSDC Administrative Data: STVC file 40