Directorates
·	Director General's Office
·	Data Development
·	Strategic Evaluations
·	Partnership and Continuous Evaluations
Publications
·	Evaluation Reports
·	Evaluation Briefs
·	Labour Market Development Agreement Series
·	Lessons Learned Series
·	Evaluation Tool Kit Series
·	Evaluation Forum Newsletter
·	From Theory To Practice

1.1 Changes in the UI/EI System

We will begin with a brief description of the Unemployment Insurance (UI) system and the changes it has gone through since 1990. Given our interest in entrance requirements, we will focus on changes in rules related to eligibility and entitlement. Between November 1990 (the effective date for Bill C-21) and April 1994 (the enactment date for Bill C-17), both the entrance requirement and entitlement calculations were based on a combination of the number of UI eligible weeks worked by the applicant in the qualifying period and the regional unemployment rate. The qualifying period was the lesser of the 52 weeks directly preceding the application or the time since the end of the individual’s last UI claim. The first column of Table 1 shows how the number of weeks an individual needed to work in the qualifying period in order to be eligible for benefits varied with the regional unemployment rate.¹ There is clearly a substantial range in the eligibility requirement, with individuals in very high unemployment regions needing only 10 weeks of work to qualify while individuals in regions with unemployment rates under 6 percent needed 20 weeks. In addition, there were special longer entrance requirements for individuals who were deemed new or re-entrants to the labour market by virtue of their having worked only a small number of weeks in the 52 week period preceding the qualifying period.

The entitlement calculation under Bill C-21 was broken down into three phases. Under phase 1, individuals gained entitlement to one week of benefits for each week worked, up to a maximum of 25 weeks. Under phase 2, individuals gained an additional week of benefits for every two weeks worked over and above 25 weeks, up to a maximum of 13 additional weeks. Under phase 3, the regional extended benefits phase, individuals were awarded a certain number of weeks of benefit regardless of how many weeks they worked, assuming they had worked enough weeks to satisfy the entrance requirement. The regional extended benefits were based on the unemployment rate in the individual’s region, with two weeks of benefits added for each 0.5 percent increment in the unemployment rate in excess of 4 percent. The maximum weeks that could be added under phase 3 was 32. The rationale was that it would take longer to find a job in those regions and hence a longer period of benefits was required. The total entitlement for an individual could not exceed 50 weeks and was the sum of the entitlement calculated under each of the three phases.

Under Bill C-17, both the eligibility and the entitlement schedules were changed in ways that made it significantly harder to collect benefits. As shown in column 2 of Table 1, the entrance requirement was increased by up to two weeks in regions with unemployment rates over 14 percent. More importantly, the functions generating entitlement were changed dramatically. First, phases 1 and 2 were reversed. After April 1994, an individual earned 1 week of benefits for every 2 weeks worked for the first 40 weeks of work in the qualifying period and then could earn 1 week of benefits for each added week of work, up to a maximum of an added 12 weeks of benefits. Regional extended benefits were maintained but now 2 weeks of benefits were added for each 1 percent increment in the unemployment rate over 4 percent, with a maximum possible 26 weeks being collectable under this component. The sum of the three components could still total at most 50 weeks of receipt. These changes drastically reduced the entitlement of many part-year workers as we will show in Section 3.

A further, large set of changes to the system were enacted under the Employment Insurance Act. The main change, for our purposes, was a shift from a weeks-based system to an hours-based system. At the outset, all this has meant is taking the same entrance and entitlement numbers that were relevant under Bill C-27 and multiplying them by 35, though the system opens the door for more changes in the future. Thus, column 3 of Table 1 (the EI entrance requirement) is just 35 times column 2 of Table 1 (the Bill C-27 entrance requirement). This is very important to keep in mind when we come to the discussion of changes in entitlement and eligibility. The only other change in this part of the system is that the maximum weeks of entitlement is now 45 instead of 50.

One interesting alteration in the system related to the entrance requirement is found in the calculation of the divisor used in deriving weekly benefit levels. To calculate weekly earnings, total earnings in the preceding 26 weeks are divided by the greater of: weeks actually worked and a specific divisor differing according to the unemployment rate in the region. The latter divisor is generally the entrance requirement number of hours plus 70. Consider an individual working 12 weeks at 35 hours per week in the highest unemployment rate region. This individual will just meet the entrance requirement of 420 hours but his earnings will be divided by 490 rather than 420 in calculating his weekly earnings measure used in deriving his weekly benefits. This has the effect of dividing his earnings by 14 weeks instead of by 12 weeks, establishing a lower basis for calculating weekly benefits. If, instead, the individual works two more weeks, the divisor is still 490 hours and, since his total earnings will be higher by two weeks worth of pay, the weekly earnings base calculated by the EI system will be higher. This wrinkle in the system directly discourages stopping jobs at the entrance requirement or the entrance requirement plus 1 week.

1.2 Incentives

The standard approach to discussing incentives associated with the UI entrance requirement (ER) is to make use of a static labour supply model (see, e.g., Moffitt and Nicholson (1982), Phipps (1990), Green and Riddell (1997)). In this model, individuals choose a number of weeks to work and a level of consumption in a defined period (typically a year) subject to a budget constraint. The shape of the budget constraint is affected by the attributes of the UI system and one can infer the incentives generated by the system by examining the shape of the constraint. Specifically, the budget constraint jumps up at the ER week as the individual’s total non-asset income moves from equalling his labour income alone (at week ER-1) to his labour income plus his benefit receipt under UI. In the weeks directly following the ER, the budget constraint becomes steeper as each added week of work not only increases total income by the weekly wage but also by increasing the weeks of UI entitlement, and hence the total benefits received. At some point, however, the sum of weeks worked and weeks of UI entitlement will total 52. We will call this the MAXYR point since it corresponds to the point at which the individual can collect the maximum possible UI benefits in the one year period. For each week worked past the MAXYR point, the total UI receipt must decline since there is one fewer week in which to collect. The incentives, discussed extensively elsewhere (see the references above, for example), suggest a likely bunching of weeks of work at both the ER and MAXYR points.

One troubling component of this model is that its predictions depend on the time horizon used by the individual. Thus, if the individual’s time horizon were two years rather than one, the MAXYR point would shift higher because it would take longer for the individual to reach the point at which he or she was running out of time to collect their benefits. This is particularly troubling because we cannot observe individual planning horizons. However, Green and Sargent (1998) argue that a one-year planning horizon is a reasonable assumption when considering seasonal workers. They look for and find evidence of a bunching of job endings at the MAXYR point among seasonal jobs in 1989. They argue, further, that a search theoretic framework may be more useful for considering outcomes for non-seasonal workers. They derive theoretical predictions from a search model that include a decline in the hazard rate as the weeks on a job approach the ER from below, a bunching of job endings at the ER point, and a higher level of job terminations after the ER has been passed than before. They also argue that comparing the number of job terminations at the ER point with the number in the weeks before the ER is not a good measure of the impact of the ER on the job duration distribution. This is so because, as seen in the search model, job terminations will be artificially low in the weeks just before the ER as terminations are “saved” for the point at which they will generate eligibility for UI. Thus, such a comparison would yield an overestimate of the actual impact of the ER. However, the extent to which job endings have been “saved” for the ER can be measured by comparing the number of job terminations at ER with the number in the following weeks.

Finally, an implicit contracting framework has also been advanced as a fruitful way to think about the impact of UI systems on employment outcomes. Such models help in making the point that, in the absence of complete experience rating, UI systems can serve to subsidize unstable work pattern industries as well as provide incentives to individual workers (see, e.g., Feldstein (1978), Anderson and Meyer (1992), Card and Levine (1994)). This framework has been suggested as particularly relevant for the Canadian situation since there is some evidence that estimated UI impacts on the job duration distribution stem at least in part from co-operation between employers and employees (e.g., Green and Riddell (1997)). In cases where the firm is assumed to view workers and weeks worked per worker as perfect substitutes in production, the firm is indifferent between adjusting to demand shocks using weeks or changes in employment. However, workers are not indifferent and, in particular, want to acquire enough weeks of work to qualify for UI. Firms will then have an incentive to follow worker preferences in terms of work schedules because workers will demand a lower wage when their preferences are met. Thus, the predictions will be the same as those under the static labour supply model that reflects worker incentives alone.

When considering the switch to an hours-based system, one possibility is that the same analytical frameworks can be used but with hours substituted for the weeks dimension. This would certainly be true if work weeks were of fixed length. However, with flexibility in setting hours per week, such a simple translation may not work well. In particular, Storer and Van Audenrode (1998) find evidence that workers and firms may have responded to the tightening of the UI system in 1994 by cutting hours worked per week in order to share out total hours in a way that generated UI eligibility for more people. If this is true, then a switch to an hours-based system will eliminate this mode of adjustment. One might then find that the number who qualify for UI will fall. As shown in Section 3, this conjecture is not borne out in the data.

A conclusion that the 1994 changes would likely lead to declines in hours worked per week and that the switch to an hours-based system would then lead to declines in eligibility seems plausible if firms and workers are focused on the ER. However, as discussed, the main impact of the 1994 changes was on entitlement, not eligibility. In a standard implicit contract model, UI benefits subsidize seasonal firms because these firms do not need to pay as large a compensating differential for their short job duration when UI benefits are available. What the 1994 changes did was reduce the amount of benefits a seasonal worker could expect to collect based on a given amount of work. This could imply a need to increase wages to keep workers in the industry. Firms might respond by reducing hours per worker in order to generate more weeks of work and hence increased UI entitlement for their workers. However, since benefits are based on weekly earnings, reducing hours would also have the effect of reducing UI based income for workers and lead to upward pressure on wages. Firms might, alternatively, keep hours of work relatively steady, reduce the number employed, and increase weeks worked among those employed to generate extra benefits for them. Which route firms adopt depends on a whole host of factors including the technology being used, how easy it is to train new workers, worker preferences concerning consumption of leisure through more hours off in a week versus more weeks off in a year, outside employment opportunities for workers, and the rate of conversion of weeks worked into UI entitlement. The result, including whether the switch to an hours-based system will lead to disentitlement, is an empirical matter.

1.3 Previous Studies

Several previous studies have examined the impact of the UI system on the distribution of job durations. Baker and Rea (1997) and Green and Riddell (1997) both examine the effects of an increase in the ER that occurred in 1990 when all regions in the country reverted to a 14 week entrance requirement because of an unrelated legislative battle between the House of Commons and Senate. Green and Riddell (1997) focus on the highest unemployment rate regions where the change caused an increase in the entrance requirement from 10 to 14 weeks. They find that the probability that a job ended in weeks 10 through 13 decreased but the probability that a job ended at the new ER (14 weeks) increased only modestly. Instead, under the new rules, more job terminations were observed in the 15 to 20 week range. This suggests that the ER does have an impact on employment outcomes but that the impact does not fit easily with a simple supply-side model in which workers unilaterally terminate jobs when they reach the ER. Baker and Rea (1997) find larger impacts directly at the ER point, perhaps because of differences in sample selection relative to Green and Riddell (1997).²

In a further study of the impact of the pre-1990 UI system on employment durations, Green and Sargent (1998) examine 1989 data on employment durations for evidence of UI impacts. As discussed above, they argue that breaking spells into seasonal and non-seasonal spells is important in such an examination and look for effects both at the ER and MAXYR points for seasonal workers. They find statistically significant but economically small effects in the form of an excessive bunching of job terminations at the ER for non-seasonal workers. For seasonal workers, they find that there is again an ER effect but that the MAXYR effect is larger. The effects for seasonal workers are larger than those for non-seasonal workers but are still not huge. They conclude that there is evidence of UI impacts but, apart from seasonal workers in high unemployment regions, these effects are not large. They use their estimates to generate predictions of effects from the 1994 UI changes. These predictions suggest that the decline in entitlement would cause increases in average employment durations primarily by increasing the MAXYR point for seasonal workers.

Two studies directly examine the impacts of the 1994 changes on related employment outcomes. Kuhn and Sweetman (1998) use administrative data to calculate predicted entitlement and eligibility for samples of workers and job terminations before and after the 1994 changes. They find that, using both pre- and post-1994 rules on a pre-1994 sample, the 1994 changes reduced weeks of benefit entitlement by an average of 10 weeks but that the impacts were far from uniform: the biggest reductions were in more seasonal industries and provinces. However, once behavioural adaptation to the new rules is included, by examining a post-1994 sample of workers, the losses are much more uniform: workers and firms in the high unemployment regions found ways to adjust.³ The authors also found that impacts in terms of workers failing to gain eligibility at all were small (less than a 5 percent reduction).

Finally, while the authors find evidence of excessive bunching of job terminations at the ER point, they argue that these effects are small in relation to the system as a whole: only 3.4 percent of all UI claims in the country were established with the ER number of weeks in high unemployment regions before 1994 (10 weeks), and, of course, not all those are necessarily terminations generated to adapt to the UI system. Storer and Van Audenrode (1998) use survey data linked to administrative records to investigate whether reductions in entitlement stemming from the 1994 changes caused workers to find lower paying jobs afterward. An interesting finding, mentioned above, is that hours per week on jobs appear to decline after 1994.

Footnotes

1	The regions used in these calculations are UI regions, which (with the exception of PEI) were sub-provincial constructs. Under Bill C-21, there were 48 such regions.
2	Christofides and McKenna (1996) also examine the impact of the ER on employment durations and find quite large effects. Unfortunately, Christofides and McKenna define their ER variable as equal to 1 for those jobs ending at the relevant ER rather than defining it as equal to 1 in the week a job passes the ER point, regardless of whether the job ends in that week. The latter is the ER variable used in all other studies of this effect. The fact that the Christofides and McKenna variable is inappropriate can be seen in the fact that they use it in their regression of log duration on various covariates. In that context, it is straightforward to see that a negative estimated coefficient on their ER variable indicates that jobs that end at the ER are shorter than the conditional average for spells that do not end at the ER. The estimated coefficient indicates nothing about how many spells ended at the ER and thus it is not possible to learn anything about the potential impact of the ER on the distribution of durations.
3	As an example, they show that in the absence of behavioural adaptation the loss in entitlement due to the1994 changes would have been 16 weeks on average in Newfoundland. In fact, the eventual loss was 7.6 weeks, which was approximately the national average after behavioural adaptation was allowed.