III. Data and concepts

View the most recent version.

We use the Longitudinal Administrative Databank (LAD) constructed by Statistics Canada along with records from the Employment Insurance Administrative Database (EIAD) constructed by Human Resources and Social Development Canada (HRSDC). The LAD is a 20% random sample of Canadian tax filers. It contains numerous income and demographic variables such as employment income, self-employment income, Employment Insurance (EI) benefits, registered pension plan contributions, age, marital status, gender and family composition.⁵

The LAD has numerous strengths. First, because it draws information from personal income tax returns, it provides fairly accurate measures of the annual earnings of husbands, wives and other family members for the 1982-to-2004 period. Second, it contains information about government transfers and after-tax income and thus allows us to assess the stabilization role played by EI benefits and the tax system. Third, because it covers a relatively long time period, it allows us to analyse the earnings of husbands, wives and teenagers several years before and after husbands' layoffs, thereby satisfying an important requirement of any reliable estimation method of the impact of programs or events (Jacobson, LaLonde and Sullivan 1993).

Like most administrative datasets, the LAD has limited information about family demographics. While it includes data on individuals' age, sex, marital status and province of residence, it has no information about a person's work hours, educational attainment or occupation. To minimize concerns regarding unobserved heterogeneity in workers' age-earnings profiles (across, say, education levels), we use fixed effects models that allow for the presence of worker-specific intercepts.

To identify layoffs, we link the LAD to the EI record files containing selected variables from the EIAD developed by HRSDC. The EI files contain data for all EI applicants (rather than only for EI recipients) from 1987 to 2001 and provide the breakdown of specific EI benefit types as well as reasons for separation taken from the Record of Employment (ROE).

Our comparison of LAD-EIAD with data from the Longitudinal Worker File (LWF) of Statistics Canada shows that LAD-EIAD captures between 91% and 100% of the layoffs registered by LWF (Table A.1), thereby confirming that LAD-EIAD provides a fairly good coverage of layoffs that occurred during the 1990s. Unlike LAD-EIAD, the LWF also contains information that allows us to distinguish permanent layoffs from temporary layoffs.⁶ The LWF shows that during the 1990s, roughly 55% of the layoffs observed were permanent while the remaining portion consisted of temporary layoffs. Hence, the earnings losses and income losses documented in this study using LAD-EIAD will result from a mixture of permanent layoffs and temporary layoffs. As will be shown below, however, we will also conduct separate analyses for a subset of permanent layoffs: those that are associated with a loss in pension coverage.

Our first sample consists of families where husbands are aged from 25 to 40 in 1987. This age restriction is implemented to ensure that husbands are aged at most 54 in 2001, the last year used in our multivariate analyses. To have at least three usable observations, we require that couples be present in the sample for at least three years (1987, 1988 and 1989). We do not have layoff information prior to 1987, so our estimation results are based on the 1987-to-2001 period. Nevertheless, we exclude those husbands who received any EI benefits from 1982 to 1986 as the best possible way of avoiding the 'contamination' from previous layoffs.

To focus on families that primarily rely on earnings from paid work, we exclude couples with combined self-employment income over $500 (in 1992 dollars, or over $595 in 2002 dollars) in absolute terms, in any year from 1982 to 2001. Similarly, we exclude families with combined total income over $200,000 (in 1992 dollars, or over $238,000 in 2002 dollars) and those who lived outside the 10 Canadian provinces. Couples had to be married during the period from 1987 until and including the year in which the husband was laid off (or until 1989 if the husband was laid off before 1989).

Our layoff variable is based on the first layoff experienced by a husband during the 1987-to-2001 period. After the first layoff occurs, couples are kept in the sample for as long as they are present in LAD as a married couple but for a maximum of five years. If the first layoff occurred after 1987, we also require couples to have had positive earnings prior to and including the layoff year. For instance, if a husband was laid off in 1992, he must have had positive earnings and to have been married during the 1987-to-1992 period. Couples are then followed for another five years (until 1997) or until their break-up ifit happened before 1997, regardless of whether the husband experienced other layoffs or had positive earnings after 1992.

Those families in which husbands were laid off in any year from 1987 to 2001 constitute our 'treatment' group. Our 'control' group consists of families in which husbands had positive earnings in the 1987-to-2001 period, remained married during this period and did not experience a layoff in any year from 1987 to 2001.

The sample of unattached individuals (single men and single women) has been selected in a similar fashion. The restrictions on total income, self-employment income and EI receipts from 1982 to 2001 were identical to those for the married men.

Our final sample of married couples consists of approximately 60,500 families and 806,800 family-year observations. The sample of single men consists of 4,700 men and 52,100 person- year observations. The sample of single women consists of 6,100 women and 76,200 person-year observations.

Finally, we build on the idea that the loss of pension coverage coinciding with a layoff is likely to indicate that the lay off is permanent. Our 'pension loss' indicator variable ( M_i) takes the value of 0 if both criteria hold: (1) husband's contribution to a registered pension plan (RPP) was positive in the year before the layoff and (2) it equals zero in the year after the layoff; otherwise the laid- off worker is not a 'pension loser' ( M_i=1). Such a definition allows us to use 'pension losers' as a bench-mark category in our multivariate analyses.⁷

Throughout this paper, we use the term husbands (wives) to refer to men (women) who are either married or live in common-law relationships. All estimates of earnings and income losses are expressed in 2002 dollars, using the Consumer Price Index as a deflator.

 

⁵ Filers are attached to their spouses (legal and common law) by the spouse's Social Insurance Number (SIN) listed on the tax form or by matching on name, address, age, sex, and marital status. Once selected, individuals are in the sample whenever they appear on the annual T1 Family File (T1FF). The Longitudinal Administrative Databank is representative on a cross-sectional basis, i.e., each year's sample is a representative sample of the population of all persons in T1FF who have a SIN. To maintain a representative sample, a part of each year's sample consists of a selection of those individuals who appear on the T1FF for the first time since 1982.

^6. Permanent layoffs occur when workers separate from their employer in year t and do not return to that employer in year t or year t+1. Otherwise, layoffs are classified as temporary.

^7. See Table A.2 for the sample sizes of various groups of families and unattached individuals.