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

A. Claim Rates of Those Who Quit Voluntarily

In order to establish that the legislation had an impact on voluntary quitters' propensity to claim UI and therefore on their costs, we first examine the claim rates of that group. All four regressions are clearly autoregressive, but the specification criteria reject the annual lags. As can be seen from the Breusch-Godfrey tests, autocorrelation of the residuals is not a problem. Interestingly, there appears to be a decreasing trend in the claim rate independent of the policy changes for all but the prime-age male group. In the second policy period, all four groups experience a moderate but significant decrease in their UI take-up rates. In contrast, the decrease in the third period, relative to the first, is quite large, as might be expected, given the much more severe nature of the UI change. These latter decreases are quite large, given the levels seen in Table 2, but, as noted above, voluntary quitters are far from completely disentitled. Although the two older groups start from a higher claim rate in period 1, the coefficient on the period 3 dummy variables for the two 25- to 54-year age groups indicates a reduction of just over 7%, which is very large. Recall that these are only the short-run effects and that the long-run effects are much larger (equal to the period-specific coefficients divided by the sum of the coefficients on the lagged dependent variables).

B. Separation Rate Regressions by Reason for Separation

Tables 4 (for men) and 5 (for women) present the separation rate regressions. A more complex autoregressive pattern is required for these data than was required for the claim rates to remove the autocorrelation of the residuals. Four regressions are presented for each age-sex group. The first is the voluntary quit rate (VQ/E), which is central to the analysis. Next are the 2 groups in which relabelling is expected to occur. First is the combined "other and dismissed" category, which obviously is a problematic grouping, since it contains a disentitled label as well as the label into which we most expect any relabelling to occur. This combination is an unfortunate necessity since, as mentioned earlier, these groups only became distinct in 1990. Still, one effect might dominate, or relabelling might occur only from the voluntary quit and not the dismissed group; thus, the regression is included. More attention is given to this issue in the next subsection. The category of the short work (layoff) reason for separation follows in the next column; relabelling might also be expected for this group. Recall that both temporary and permanent layoffs are subsumed under this group, since they cannot be separately identified in our data. Finally, the total separation rate (S/E) regression is presented; it includes separations for all reasons.

Looking at the policy variable's coefficients, the 25- to 54-year-old men, in Table 4, exhibit no response to the legislative changes. The coefficients are not only insignificant, but the point estimates are quite small. For the 15- to 24-year-old males, the period 3 dummy in the voluntary quit regression is marginally significant (P = 0.75) and of a moderate size, suggesting that their quit propensity is reduced by the legislation. None of the policy coefficients in this group's other regressions are significant though. However, the point estimates in the total separation regression (S/E) are large and negative, as would be expected if there is a reduction in the quit rate with no relabelling. But the standard error is also quite large, so little can be said with confidence.

Both female age groups, in Table 5, have strongly significant period 3 dummy variable coefficients in the VQ regressions that are of an appreciable economic size. Further, the overall separation rate (S/E) for the young females has a negative period 3 coefficient that is economically large and statistically significant. This suggests that inhibition is very likely for this group.

For none of the three out of the four groups that exhibit a reduction in their quit rates is there any evidence of an increase in the layoff or the "other and dismissed" rates, as would be required for relabelling to be credible. In fact, the point estimates are not even positive, but small and insignificantly negative. Overall, the naive no-response model seems to work fairly well for the prime-age males, whereas both female age groups, as well as the younger males, appear to be inhibited from quitting by the total disentitlement.

C. "Dismissed" and "Other" Reasons for Separation

Prior to 1990, the "dismissal" category was included with the "other" one. Their division, unfortunately, coincides too closely with the introduction of the partial disentitlement for any meaningful analysis of it to be done for that policy change. A simple before-after comparison is, however, possible around the total disentitlement. The separation rates are shown in Table 6, and the claim rates are shown in Table 7. We have a strong desire to use all 15 months of data that are available after the change for the separation rates. Seasonal and cyclical factors are, however, quite important. Fortunately, business cycle conditions before and after the total disentitlement were remarkably stable. Seasonality is addressed by matching months, that is, ensuring that there are the same number from each month in the "before" and "after" groups. Two possible "before" samples are feasible; both will obviously contain the period April 1992 to March 1993. The question is how to match the second April-to-June sequence and hence the total 15 months required. One alternative (Before 1) is to use April to June in 1992; the other is to count those months in 1993 twice (Before 2). The former has the advantage of not introducing the same error terms twice, while the latter is a period with more business cycle conditions similar to those of the period after the change. Notice how similar the unemployment rates and HWI are across the periods. Whichever specification might be preferred, each gives very similar results for the "dismissed" and "other" groups. None of the differences in the separation levels across the periods are significant; in fact, there is almost no change at all for either reason-for-separation group.

Unlike the separation rates, there are only 9 months of claim data in period 3, so only the matching months prior to the Bill are used. Both age groups exhibit a reduction in the claim rate for those who are dismissed, as evidenced in Table 7. This suggests that the financial incentives to motivate a behavioural change exist, although 34 and 16% of the 25- to 54- and 15- to 24-year groups continue to receive benefits following total disentitlement. In contrast, the claim rate of the "other" group is remarkably constant across the periods 2 and 3. These tables provide evidence against relabelling away from the UI disentitled group "dismissed for cause," who clearly suffer a financial loss because of the UI legislation. Either the label has value in itself that outweighs the costs of relabelling or the institutional constraints are sufficient to prevent side deals. The stability of the "other" group further strengthens the conclusion, suggested in the previous subsection, that the voluntary quit group is not relabelling to the "other" category following the complete disentitlement of UI eligibility (i.e. in period 3) where significant declines in the quit rate are observed. This is a strong point, since if firms are concerned about possible penalties, the "other" rather than the "short work" category is arguably the less problematic category into which workers might be relabelled.