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

The statistical methodology used in this paper follows that used in a previous analysis of re-employment outcomes by Storer and Van Audenrode (1995) and the studies of links between UI and search outcomes by Crémieux et al (1995a and 1995b).¹ These methods are applied with an expanded set of criteria to measure long-term outcomes.

The first part of the statistical analysis of this paper is an examination of (i) staying in the labour force, (ii) finding a job, (iii) obtaining full-time versus part-time work and (iv) obtaining non-wage job attributes such as work in the unionized sector, medical benefits and a pension plan. This is accomplished through the use of limited dependent variable techniques. Taking the example of unionization, a dichotomous variable is defined that equals 1 for a unionized job and 0 for a non-unionized job. The probability that a new job is unionized is then obtained from:

The function F( X ) is chosen so that probabilities between zero and one are obtained for any and all values of X . Two common choices are the logistic function which gives rise to a logit regression and the standard normal distribution function which yields a probit regression. Given that there is no reason to prefer one over the other, we adopt the probit approach here.

For this study, it would also be useful to examine long-term outcomes with regard to the sector and nature of new jobs obtained, particularly for persons previously employed in low attachment career profiles such as seasonal/temporary occupations. The probability of transitions from low attachment to high attachment jobs might be thought to increase due to C-17 which makes seasonal cycling less profitable. This is done by looking at transition probabilities to and from seasonal jobs and through the statistical analysis of the (self-reported) expected duration of a new job.

Wages earned in new jobs are analysed as in previous studies. Here, it is possible to use OLS regression techniques to compare wages earned on new job for persons with various characteristics. In particular, job losers under the pre- and post-C-17 regimes can be compared with this regard. For this comparison it is useful to adopt the framework used by Addison and Portugal (1989). Addison and Portugal model the wage of individual i prior to losing job j-1 with the following equation:

(1)

In this specification, observable characteristics of individual i have been partitioned into those specific to the individual X ^I and those representing interactions between the individual and the job X ^IE. In the context of the COEP data, the vector X ^I includes variables such as the age, sex, marital status, educational level, and region of residence of an individual. The vector of individual-firm characteristics X ^IE is composed of variable such as tenure, the union-status of a job, the industry and occupation of the worker at a particular job, wages earned and hours worked in that job.

Building on this framework, it is possible to specify an equation for the wage obtained in the job j found after a period of unemployment. Addison and Portugal adopt the following specification in this case:

The principal modification between (1) and (2) is the introduction of the variable dur_i,j which measures the amount of time that individual i spends without a job between jobs j-1 and j. This duration effect is intended to capture the possibility that levels of human capital depreciate during a period of unemployment although in a non-structural framework it may also capture the effect of the degree of patience of the unemployed. Workers who are willing to be more selective will have longer spells of unemployment but will also find higher new wages as a result.

Estimation in this study proceeds through the use of a hybrid version of equation (2) in which the previous wage, the dependent variable of (1), is also added as an explanatory variable, giving rise to equation (3):

The effect of this modification is to incorporate into the new wage equation all of the information of equation (1), including the unobserved error term u_{i, j -1} that may account for unobserved individual-level heterogeneity. The coefficient d₄ of the old wage will be less than one to the extent that the old wage was determined by either non-transferable individual heterogeneity or previous productivity specific to that worker-employer match. Coefficients on other variables in the equation capture new-wage effects only since their effect on the old wage is already included in the equation.

This study seeks to determine whether changes to the unemployment insurance system introduced by Bill C-17 have altered the determinants of the wageobtained after a period of unemployment. There are two ways that C-17 could have such effects. Suppose that we determine that the following relationship holds between new wages and UI benefit entitlements (b) and other variables (X):
 
 

Bill C-17 changed the rules relating insurable weeks and regional unemployment rates to benefit entitlement periods so that persons may have faced very different benefit entitlements under pre and post-C-17 regimes. The relationship above indicates how these policy changes would translate into wage effects. It is also possible, though, that C-17 would induce changes in behaviour so that the effect of a given level of benefit entitlement upon post-unemployment wages was itself modified by C-17. In terms of the equation above, this would involve a change in the nature of the h(b, X) function.

Any such changes will be detected in this report by investigating the effects of C-17 upon new wages in several steps. In a first step, the determinants of re-employment wages are examined using separate samples of individuals from the 1993 (pre-C-17) and 1995 (post-C-17) Canadian Out of Employment (COEP) samples. An informal comparison of the coefficients for the two periods is undertaken. Next, the estimated d values for the 1993 sample are used to determine how changes in benefit entitlements induced by C-17 would have translated into changes in wage outcomes given the 1993 behaviour. In a similar way it is possible to calculate the change in wages implied by the modification of the d coefficients assuming that C-17 did not change benefit entitlements. While each of these calculations gives only part of the total effect of C-17, the breakdown into benefit entitlement and behavioural effects is useful information for policy evaluation.

In a second approach, the total effect of C-17 can be calculated by using a pooled 1993 and 1995 regression in which binary ("dummy") variables are entered interactively with key UI policy variables such as regional unemployment rates and insurable weeks worked. These interactive dummy variables allow the effects of policy variables to differ before and after C-17. Tests of statistical significance of the dummy variables yield a formal econometric test of the constancy of the d parameters for UI related variables across the pre- and post-C-17 regimes. The sign and magnitude of these dummy variables indicate how persons with given numbers of insurable weeks were affected by C-17. This effect incorporates both changes in benefit entitlement given insurable weeks and changes in the effect of a given entitlement.

Footnotes

1	A more detailed description of the NESS data is found in Crémieux et al (1995a).