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

3.1 COEP — Introduction

The 1996 Canadian Out of Employment Panel (COEP) data set is ideal for looking at many issues related to Employment Insurance (EI), and the Unemployment Insurance (UI)-EI transition. It is a quarterly survey of a sample from the population of all individuals experiencing job separations in Canada for which Records of Employment (ROE) were issued, and it combines survey and administrative data.¹⁰ Individuals are selected into the sample from the population of ROEs in the Human Resources Development Canada (HRDC) file for the appropriate period. The reference ROE is the one in which a person is selected into the sample. However, the respondent may have (frequently has) other ROEs (other job separations) that are relevant to the study in that they may be combined with the reference ROE to establish an EI claim. A warning is required in interpreting the COEP’s results: it is (approximately) a random sample of people who experience a separation in a given quarter of the calendar year, not a random sample of either EI claimants, the labour force, or the population. The key point is that individuals who experience frequent separations are over-represented in the sample relative to their frequency in the labour force. Perhaps more subtly, EI claimants who base their claims on multiple ROEs will be over-represented in the sample relative to all claimants.¹¹

At the time of writing there were nine cohorts of data from the COEP available. Four are from the four quarters (year) prior to the start of the phase-in of the UI-EI transition in July 1996, two are during the phase-in period, and three are following the completion of the phase-in in January 1997. Each cohort is interviewed twice (two waves for each cohort): about 12 and 20 months following their job separation. This study uses only first interviews, since very few second interviews are completed at the time of writing. Further, only cohorts 7, 8 and 9 are employed.

Although I use the phrase “random sample” above, the COEP is in fact not a random sample. It is a stratified and partially clustered survey, and requires correct weights to be applied to correctly represent the population.¹² For all tabulations and regression analysis, appropriate survey weights are applied.

3.2 COEP — The Tricky Question of Hours

Since the focus of the paper is on the impacts of the move from a weeks to an hours based system, an important preliminary issue concerns identifying which measure of hours is the most appropriate to use. There are two candidates: the hours reported on the version of the ROE form instituted after January 1997 which captures the information (of course the UI version of the ROE form does not contain information on hours), and the hours reported by respondents as part of the survey. Several related issues need to be noted. First, HRDC policy for dealing with weekly hours in the UI period for claims made in the EI period is to assume that all jobs were 35 hours per week (very close to the national average). This transitional policy will, obviously, be an advantage to low hours workers, but a disadvantage for high hours workers. Second, for jobs that span the UI and EI regimes, hours are supposed to be reported for the latter and weeks for the former portions; and the 35 hours per week assumption will be used for the weeks portion. Third, for EI purposes, hours include all paid hours, both regular and overtime, but not, for example, unpaid overtime.

Survey and ROE (administrative) average weekly hours are compared in Tables 15 and 16. In each table the continuous hours variables are categorized into four groups to facilitate comparisons. The first table includes those observations for whom working hours are missing for either variable. About 7.5 percent of the sample have one of the variables missing (a few observations with various other variables missing, in particular those with both hours variables missing, are removed since they would be removed from any analysis). Table 16 includes only those jobs that are entirely in the EI period so that there is no confusion around the regime switch. Most obviously, it is clear in both tables that ROE and survey reports of hours differ dramatically even for the wide groupings employed. Only 52 percent of the observations lie on the diagonal (i.e. the groupings match) in Table 15, and 56 percent do so in Table 16. Further, many of the mismatches are quite gross with ROEs indicating over 40 hours and individuals reporting less than 15, and vice versa. The number of individuals over-reporting hours relative to employer reports substantially outnumbers the percentage under-reporting. This level of disagreement between the two measures seems too gross to be credible, but the numbers are fairly easy to generate from the data so a computational error is unlikely. I intend to pursue this issue at a later date. As a result of this data issue, I use ROE hours wherever possible in the work that follows, and I only use survey hours for those jobs, or percentage of jobs, that are in the UI period where ROE hours are not available. I am able to use ROE hours a large percentage of the time. This also has the advantage that, regardless of which is “right”, ROE hours are what are actually used to run the EI system.

3.3 COEP — Changes in Eligibility

Eligibility measures whether a person is able to obtain any EI benefits at all. Thus, to lose eligibility (to have the discrete variable go from one to zero), is to move from being able to obtain some benefits to not being able to obtain any. Whether or not the person actually takes up the benefits that they are eligible for is a separate issue and is the subject of a different evaluation report. Further, I use the word “entitlement” to indicate the number of weeks of benefits to which a person is entitled and I address this issue in the next section.

A cross tabulation comparing eligibility under the UI and EI rules is presented in Table 17. This table is very similar in format to those that follow. The upper number in each cell is the weighted number of observations, and the one below is the row percent in Table 17 (in all subsequent tables, percentages sum up to 100 for each column). Although the number of observations represented by each cell is generally not a whole number (because of the survey sampling weights), the total for the table equals the overall number of observations. Three categories are defined for each regime: those who are ineligible, those who can claim based only on their reference ROE (i.e. using only the reference ROE, even though they may also have others; any additional ROEs would only affect the individual’s entitlement but not eligibility), and those who are eligible but need to combine multiple ROEs to claim. I use the simple rule that only ROEs in the 52 weeks (the UI and EI qualifying period) prior to the reference ROE are considered. Of course, in practice, the reference ROE may be combined with subsequent ROEs rather than previous ROEs to establish a claim. But given that any subsequent ROEs may not be in the data file yet, I use the retrospective rule for the purposes of my counterfactual believing it to be an interesting representation of the impact of the UI-EI transition. This technique has been employed in previous HRDC evaluation studies and does not appear to distort measures of the change in eligibility or entitlement, which is the focus of this study. However, those individuals who claim EI benefits may be able, on average, to increase their benefits relative to those predicted here by optimally combining ROEs. Thus, these calculations can be thought of as a (close) underestimate. Additionally, only a fraction of those who are eligible actually take up benefits (see Storer and Van Audenrode 1998).

Table 17 indicates that the regime change did not have a uniform impact on the population. Some workers became eligible while others became ineligible, although substantially more became eligible. About (9.1 + 14.5=) 23.6 percent of those who were not eligible under UI can claim under EI. In contrast, about (3.3 + 4.5=) 7.8 percent of those who could claim under UI cannot under EI. These changes follow directly from the move to the hours based system. High wage (e.g. summing to above $150 per week) but very low hours individuals are no longer eligible, while those with lower wages, but slightly more hours, are newly eligible. Overall, the vast majority of separators, about 88 percent, do not change categories as I’ve defined them, and only 7.9 percent change their eligibility status (become newly eligible or ineligible).

In the series of Tables 18 through 24, differences in the nine UI-EI transition categories outlined in Table 17 are explored according to a variety of worker characteristics (in the order of the tables): province, sex, age, education, marital status, household type, and ethnic affiliation.¹³ Looking first at the provincial data in Table 18, one interesting feature is that the Atlantic provinces have between 3 to 7 percent of workers who move from being ineligible to being eligible based on a single job, in contrast to about 2 percent nationally. They also have a slightly lower percentage that lose eligibility based on a single job (and a higher percentage with missing information).

In Table 19, comparing females and males, women can be seen to be slightly more likely to have both increased and decreased eligibility under EI, and they are less likely to maintain eligibility based on multiple jobs. Comparing across three broad age categories: youth (age 25 or less), prime age (age 26–55), older worker (age 56 or greater), it is clear in Table 20 that the distribution is very different for youth. About 33.6 percent of them are ineligible in both the pre- and post-EI periods, compared to 12.7 percent for the prime age and 8.2 percent for the older workers. As a group, the “youth” category contains a higher percentage of individuals who are more likely to change eligibility status, becoming either eligible, or ineligible, based on one job. Table 21 looks across education categories and shows few substantive differences, except that the some or completed university category is more likely to be ineligible both before and after the change — this might largely be caused by university students who are in the “some university” category. Differences across marital status are described in Table 22. The single group, which tends to be younger, is much more likely to be ineligible under both regimes, and less likely to have a single job that is eligible for both UI or EI benefits. There is very little difference between the married group and those who are widowed separated, or divorced.

Table 23 looks at eligibility and household type. Those who are single (living alone), or economically “single” but living with others, are more likely to not be eligible for either UI or EI. Many of these individuals are youth; they are also more likely to change eligibility status (although some are students and are thus rendered ineligible). There do not appear to be any startling differences across the household types.

Eligibility across self reported ethnic groups (non-visible minority, visible minority, and Aboriginal) can be observed in Table 24.¹⁴ Visible minorities appear to be less likely to combine multiple jobs, and Aboriginals appear to be somewhat less likely to be eligible based on a single job. Non visible-minorities appear to be slightly more likely than the population average to lose eligibility based on a single job, whereas visible minorities are slightly more likely to lose eligibility based on multiple jobs.

Results of differences across demographic groups are summarized in a parsimonious way that controls for confounding interactions between the observed variables in Table 25 where the results from two probit, and two logit regressions are presented.¹⁵ The coefficients are difficult to interpret, so the probability of eligibility changing as the dummy variable (all right hand side variables are dummy variables), evaluated at the mean of the sample, makes the discrete change from zero to one is also presented.¹⁶ These are in two columns with “dF/dx” as the heading and are associated with the probit regression. A “dF/dx” value of, for example 0.017, implies that the average person experiences a 1.7 percent increase. These regression results should be viewed simply as descriptive tools to help summarize the data. The first regression has a dependent variable that is equal to one, if the observation becomes eligible and zero otherwise, while the dependent variable in the second is set to one if the individual becomes ineligible. This approach has the advantage of providing some sense of the statistical significance of any effects.

One notable impact of the legislation observed in Table 25 is that workers in the Atlantic provinces are both more likely to become eligible, and less likely to become ineligible, than other Canadians. Other impacts of significance are that women and youth are more likely to both become newly eligible and become newly ineligible than male and prime age workers, respectively. This arises because these populations have a much higher percentage of jobs near 15 hours per week, and they are affected in both directions by the legislation. There appears to be no difference in the impact of the legislation by education and visible minority status. While the coefficients are statistically significant, the probabilities they represent are everywhere quite small in magnitude. For example, for the probits for becoming eligible in the first column, a female’s probability of being eligible increases by just under 2 percent, while that for youth increases by about 3.4 percent. Newfoundland’s increases by about 4 percent, while the Territories’ decreases by about the same percentage.

3.4 COEP — Changes in Entitlement

Benefit entitlement is calculated for cohorts 7, 8 and 9 using first the UI and then the EI rules, and the results are presented in Table 26 using the same 9 categories as in the eligibility analysis. Of course the categories have a slightly different interpretation now since some individuals who are eligible for EI benefits using only the reference ROE, in fact have multiple ROEs, and all of these are used in the entitlement calculation.¹⁷ The individuals in the top row, representing 15 percent of the sample, are not eligible under either regime and, therefore, trivially have no change in eligibility. Those who become eligible based on a single job have, on average, almost 24 weeks of entitlement. Their minimum entitlement is 14 weeks, and their maximum is 45; in other words, they span the full range of possible benefit durations. Most, however, are close to the “mean” since the 25 and 75 percentiles are 19 and 26 weeks, respectively. Those who become eligible (move from being ineligible to being eligible) based on multiple jobs, increase their benefit entitlement by 29.3 weeks on average. Those who become ineligible, face a reduction of about 30 weeks if their claims were based on a single job, and 23 weeks if they were based on multiple jobs.

Since there are more individuals who become eligible, overall the number of weeks of entitlement in the system increases. But those who maintain their eligibility also have entitlement changes. Under UI, individuals who combined ROEs to establish claims, on average, increased entitlement from the move to EI. Meanwhile, those who established a claim based on a single job have theirs reduced — much of the latter decrease in entitlement arises from the reduction in the maximum weeks of benefits from 50 to 45 which is discussed in more detail below. Despite the large changes for some individuals, across the entire system there is little change in entitlement since the average reduction is only about 0.3 weeks.

A histogram of the distribution of entitlement losses is plotted in Figure 2 with a vertical line drawn at zero. The distribution is bimodal with most people having a small change in their weeks of entitlement. There is a fairly large spike at five weeks of lost entitlement reflecting the change in the maximum. There are, however, also long but thin right and left tails to the distribution reflecting the large changes experienced by some workers who are strongly affected by the new hours provisions. Some, who under UI had a (second) job that was ineligible for benefits have a large increase, while others who have a 16–18 hour a week job that was previously treated for a full week for benefit purposes, now find it only counts for about half a week.

Table 27 extends the analysis of Table 26 by asking the question: what would entitlement have been had the EI hours changes taken place, but the move from a maximum entitlement of 50 weeks not been reduced to 45? (This is done by converting the benefit eligibility table under UI to a new benefit eligibility table using 35 hours per week as a conversion factor; this is the same approach employed by HRDC in moving from weeks to hours). That is, Table 27 isolates the hours change by, counterfactually, removing the reduction to the maximum. For almost all groups excluding those who lose eligibility there is a small increase in the “mean” weeks of entitlement. Overall, instead of a decrease of about 0.3 weeks, the hours changes alone would have actually increased entitlement by about 0.4 weeks on average. And recall that eligibility has also changed so that there are more separators who are eligible to claim. The move to hours makes the EI system slightly more generous than the UI one on average. Only the reduction in the maximum, which affects the long tenured worker, causes the EI system to be less (potentially) generous in total.

Figure 3 plots the change in entitlement associated with Table 27 (eliminating the 50 to 45 week maximum entitlement effect). The large mass to the left of zero is effectively removed and there is an enormous spike at zero (notice that the y-axis scale has changed relative to Figure 2).

To identify the characteristics of those who lose or gain entitlement, Ordinary Least Squares (OLS) and as a specification test ordered probit regressions are run using the change in entitlement of those who are eligible under either UI or EI as a dependent variable. The sample is all those who are eligible under either policy regime. These results are presented in Table 28, and suggest that women face a reduction of about 2 weeks more than men, and older workers’ entitlement is reduced about 1.7 weeks more than that of prime age workers. There are very few differences across geographic regions, with the exception that people in the Territories lose about 3 weeks more than those in Ontario. Further, those with greater than a high school education also lose slightly more than those with less education.

Footnotes

10	An ROE is issued by the employer at the termination of the employee-employer relationship; thus, the self-employed are not normally included in the sample frame. An ROE is required to initiate an EI claim.
11	In what follows, all ROE reasons are used in the calculation of eligibility and entitlement even though voluntary quitters, for example, are less likely to claim than those who are laid off for “short work”. A more refined version might weigh each observation by the probability of take-up for its ROE reason for separation group.
12	In addition to the usual issues of stratification and non-response, part of the survey design includes certain communities being over sampled.
13	Note that the sample sizes differ slightly across these tables since each one excludes those for whom we do not know the relevant variable.
14	Visible minority and Aboriginal identification is self-reported in the COEP.
15	Since the percentage of observations with a dependent variable set equal to one is not very large, both logit and probit regressions were run to ascertain the sensitivity of the results to the distributional assumption.
16	These changes in probability measures have interpretations very much like ordinary least squares regression coefficients.
17	Therefore, some individuals are in one of the “one job” categories since they are eligible for benefits based on a single job, but their entitlement is in fact calculated using all of the jobs they have in the relevant window.