Main article

Warning View the most recent version.

Archived Content

Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.

  1. Introduction
  2. Literature review
  3. Data description
  4. Methods
  5. Results
  6. Conclusion

1   Introduction

The economic downturn of 2008–2009 resulted in large scale job losses across most of the industrialized world. In Canada, a net loss of almost 500,000 paid jobs was registered between October 2008 and October 2009 according to the Labour Force Survey (LFS). 1  This represented about 3% of the paid workforce. The downturn has renewed interest in the outcomes of displaced workers. There is a rich research literature documenting the negative outcomes experienced by workers displaced by layoffs or plant closures (e.g., Jacobson, LaLonde, and Sullivan 1993; Morissette, Zhang, and Frenette 2007; Hijzen, Upward, and Wright, 2010).

Faced with job loss, individuals have several options available to them. First, they can search for new employment. Little is known, however, about the effectiveness of job search on re-employment, although de Raaf, Dowie and Vincent (2009) are implementing a social experiment to study this question. Second, displaced workers may create their own job by becoming self-employed. The literature generally finds that some workers are 'pushed' into self-employment (Moore and Mueller 2002). Third, households experiencing a job displacement may respond by adjusting their labour supply. Morissette and Ostrovsky (2008) find that among couples in which husbands were displaced by layoff or plant closure, subsequent earnings gains of wives were 22% as large as the earnings losses of displaced husbands five years after job loss. Fourth, displaced workers—particularly those in older age groups—can withdraw from the labour force (Schirle 2009). And fifth, displaced workers can re-invest in their human capital. This last response is the focus of this study.

The Longitudinal Worker File (LWF), a large administrative database representing 10% of Canadian workers, is used to study the long-term impact of post-displacement training on earnings. Post-displacement training is defined as attendance at a post-secondary educational institution shortly after job displacement. The data follow workers for up to nine years following job displacement and, for comparison purposes, over the five years prior to displacement. The estimation strategy exploits the longitudinal nature of the data by implementing a simple difference-in-difference approach.

As well, the impact of job displacement on post-secondary education attendance is estimated using a two pronged approach. First, a difference-in-difference model is estimated, as in the case of earnings. However, because of data limitations, the time frame for this analysis is somewhat shorter than that for earnings. Specifically, workers are followed from four years before to four years after potential displacement. The possible selectivity of workers who are let go by firms is addressed by focusing on those displaced by mass layoffs and firm closures—events that are likely precipitated by a decline in market product demand rather than individual worker productivity.

The results suggest that, over the period spanning five years preceding and nine years following job loss, workers who took education and training following job loss saw their earnings increase by almost $7,000 more than displaced workers who did not take education or training. Significant benefits are found by sex, marital status, union coverage, and age group, with the exception of men aged 35 to 44 years. Despite the benefits of training, job displacement was found to be associated with only a modest increase in post-secondary education attendance for all groups examined.

In the following section, a review of the literature on the impact of training on earnings is provided. In Section 3, the data are described; in Section 4, the econometric approach used in the study is outlined. Results are presented in Section 5, and Section 6 concludes.

2   Literature review

The literature is rich in studies showing that displaced workers experience large earnings losses following job loss. Since the publication of Jacobson et al. (1993), a number of research papers from several different countries have used large administrative data bases to track displaced and non-displaced workers over relatively long periods of time. For example, Huttunen, Moen, and Salvanes (2006), Eliason and Storrie (2006), Ichino, Schwerdt, Winter-Ebmer, and Zweimüller (2007), Morissette et al. (2007), and Hijzen et al. (2010) produce estimates for Finland, Sweden, Austria, Canada, and the UK respectively. These studies strive to match displaced workers with comparable non-displaced workers on the basis of their observable characteristics, and to control for any unobservable differences by using difference-in-difference methods.

A number of stylized facts have emerged from this literature. Most importantly, income losses are large and long-lasting. Older workers and workers with longer tenure of employment experience particularly large losses. However, the source of the income loss varies across countries. In the United States, significant income losses persist even for those workers who return to employment, because displaced workers accept lower wages in their new jobs. In Europe, income losses are primarily the result of spells of non-employment; wages for those displaced workers who return to work are only slightly lower, on average, than pre-displacement wages.

The literature is thinner on studies that explore the factors that help workers respond to job displacement and the potential for training programs to reduce earnings losses. In a detailed survey of US government employment and training programs, Heckman, LaLonde, and Smith (1999, p.1868) conclude that: "For most groups of participants, the benefits are modest, and at worst participation in government programs is harmful." However, as Heckman, LaLonde, and Smith also note, these training programs are usually targeted at "disadvantaged" adults, particularly those with low skills and long spells of unemployment. The literature shows that displaced workers have characteristics quite different from those of unemployed or disadvantaged workers more generally (Jacobson, LaLonde, and Sullivan, 2005a). Displaced workers tend to be older and to have considerable work experience, although their skills may be firm- or industry-specific. They may also lack job search skills, particularly if they had been employed for an extended period prior to displacement.

What is known about the effectiveness of training programs for displaced workers? Almost all the evidence comes from the United States. A number of American studies conducted in the 1980s and 1990s suggest that displaced workers benefit from job search assistance, but that the additional benefits from training are negligible. For example, Leigh (1994) summarizes evidence on four government-sponsored programs targeted at displaced workers in the United States and finds no evidence that classroom training in vocational skills yielded any additional benefit. Decker and Corson (1995) conduct an evaluation of the Trade Adjustment Assistance program and find no evidence that training has a positive impact on trainees' earnings, at least in the first three years after their initial unemployment insurance claim.

Dar and Gill (1998) summarize studies of eleven different retraining programs designed to assist displaced workers (US, Sweden, Australia, Canada, Denmark, and France). They conclude that retraining is generally far more expensive than job search assistance, but does not necessarily yield better results. However, the studies summarized by Dar and Gill could not provide estimates of longer-term effects since none of them were longitudinal. Dar and Gill conclude that there is a lack of rigorous evidence on the costs and effectiveness of retraining programs for displaced workers.

In a recent American study, Heinrich, Mueser, and Troske (2008) evaluate the Adult and Dislocated Worker Programs using data from 12 states and nearly 160,000 participants. The authors find that recipients of training services who were not displaced have lower initial earnings than non-participants, but that their earnings catch up within 10 quarters, ultimately registering large total gains. Nonetheless, displaced workers experience significantly smaller returns to training than do non-displaced workers who trained.

There are several possible reasons why these studies find that government-sponsored programs generate little or no increase in earnings. First, the studies are usually short-term, often assessing outcomes within six months of program completion. If training improves workers' ability to progress in their careers (as opposed to simply raising their initial wage in the first job held following displacement), then a longer time horizon would be required to observe the benefits achieved. Second, the nature of the training is often very job-specific or, at the other extreme, of a general nature (e.g., basic life skills), as opposed to the more balanced training normally acquired in formal education. A third reason is that often individuals are selected for government-sponsored training programs on the basis of their specific characteristics. Aside from the usual identification challenges this may create, there may be a stigma effect associated with program participation. Finally, some of the earlier studies listed here are demonstration projects and hence are based on small sample sizes. Jacobson, LaLonde, and Sullivan (2005a) note that it is often difficult to measure the effect of the programs precisely with such data.

In their studies, Jacobson, LaLonde, and Sullivan (2005a; 2005b) link administrative earnings records with community college transcripts for workers displaced from their jobs in the early 1990s in Washington State and the city of Pittsburgh. They estimate returns to one year of college to be around 9% for men and around 13% for women. These estimates are much closer to the usual rate of return associated with one year of schooling found in more general studies of returns to education, and suggests that more rigorous college-level publicly-provided training may be effective. Jacobson, LaLonde, and Sullivan also show that it takes time for the benefits of training to be realized: earnings may actually be lower in the treatment group shortly after the completion of college-level training, but increase in subsequent years.

This paper provides the first large-scale, long-term evidence on the utilization and effectiveness of retraining for displaced workers in Canada. 2  An administrative database representing 10% of Canadian workers is used. The database allows workers to be followed from five years before to nine years after job displacement, and allows post-secondary education to be identified. The earnings consequences of training following displacement are examined over a period of nine years. The findings from this study also contribute to the literature by estimating the extent to which job displacement is associated with post-secondary training. 3  In this case, a shorter time frame is used (from four years prior to potential displacement to four years after) because of data limitations.

3   Data description

The LWF is used in this study. 4  The LWF, created by Statistics Canada, is a 10% random sample of all Canadian paid workers, constructed from four linked administrative data sources: the Record of Employment (ROE) files of Human Resources and Skills Development Canada; the T1 (T1 General - Income Tax and Benefit Return) and the T4 (T4 slip, Statement of Remuneration Paid) files of the Canada Revenue Agency; and the Longitudinal Employment Analysis Program (LEAP) of Statistics Canada. The data are longitudinal and span the period 1983 to 2007.

Canada's Employment Insurance Act and Employment Insurance Regulations require every employer to issue an ROE when an employee working in insurable employment has an interruption in earnings.  5  The information contained on the ROE is used to determine whether a person qualifies for Employment Insurance (EI) benefits, the benefit rate, and the duration of the claim. The ROE must be issued at the point of job separation regardless of the intent of the employee to file a claim for EI benefits. More important, the ROE indicates the reason for the work interruption or separation. 6  The ROE can thus be used to identify workers who are laid off, workers who quit and workers who separate from their employer for other reasons. For the purposes of this study, a displaced worker is a worker who was laid off in the 1998 calendar year.

It is possible to distinguish between workers who are temporarily or permanently laid off. Permanently laid off workers are those who do not return to the same firm during the 12 months following lay-off. The definition of job displacement used in this analysis encompasses all permanently displaced workers. However, for analyzing the impact of displacement on the decision to train, it is important to compare displaced workers whose characteristics are similar to those of a control group of non-displaced workers. For this reason, a sample of displaced workers who were displaced as a result of firm closures or mass layoffs—events which are arguably exogenous with respect to the characteristics of individual workers—is used.

Identification of firm closure is made possible by the fact that the LWF includes LEAP, a longitudinal file that tracks all Canadian companies. Since LEAP identifies firms' births and deaths, the linkage between LEAP and LWF makes it possible to identify layoffs that occur as a result of firm closures. 7  However, the LWF contains no information on establishment closures. Because many large firms consist of multiple establishments, plant closures may occur and cause mass layoffs without inducing firm closures. Thus, firm closures will not capture these job losses. To take these into account, a broader definition of displacement, one that includes not only workers who lose their jobs through firm closures, but also those who lose their jobs through mass layoffs, is used. This approach follows Morissette et al. (2007), with a mass layoff in year t identified as a situation in which 30% or more of a firm's workforce is laid off between years t - 4 and t + 1. A firm closure in year t is defined as a situation in which a firm went from positive employment in year t to no employment in year t + 1.

LEAP also contains information related to the industry in which the firm operates. Specifically, the 2002 version of the North American Industry Classification System is used.

The Canadian Revenue Agency requires employers to issue a T4 slip to any employee with paid earnings of more than $500 over the course of the year. This information is used as a measure of paid earnings over the course of the calendar year. 8  Firm size can be approximated by dividing the total payroll for a firm by average wages within province and industry cells. 9 

Finally, the LWF includes information from the T1 General – Income Tax and Benefit Return. The T1 file contains information on several personal characteristics used as control variables in the study, including sex, age, province of residence, marital status, and union dues paid (to identify union membership).

The LWF does not contain information on the stock of education at a point in time. Nevertheless, participation in post-secondary schooling, which is central to this study, can be identified. Specifically, the T1 file includes information on tuition credits and education deductions claimed for courses taken at a post-secondary education institution in Canada. And while a student may transfer these credits or deductions to a parent for tax purposes, it is possible for the years 1999 onwards to identify both the student and the claimant (if they are different). In short, training information is available from 1999 onwards.

Measuring the intensity, or quantity, of post-displacement training is not straightforward. Similarly, training may begin prior to layoff (perhaps in anticipation of such an event), or it may continue for several years following job loss. The incidence of training following displacement, which is the focus of this study, is simpler to define conceptually and far easier to measure in practice. Also for reasons of policy relevance and measurability, the scope of the study is limited to the estimation of the effect of post-secondary training attendance on post-displacement earnings.

4   Methods

4.1  The impact of training on earnings

The first part of the analysis consists of estimating the impact of post-displacement education (i.e., attendance at a post-secondary education institution) on paid earnings, among a sample of displaced workers. Specifically, workers who were displaced in 1998 following five consecutive years of positive paid earnings are considered. The workers were between 25 and 44 years old in 1997 (and thus no older than 54 years old in 2007) and therefore less likely than older individuals to consider retiring after displacement. The measure of post-displacement training is a simple binary indicator of post-secondary education attendance in 1999, the calendar year following job displacement. 10  Figure 1 summarizes the sample construction.

Figure 1

The annual ordinary least squares (OLS) estimates of Equation (1) for workers who were displaced in 1998 are calculated separately by year and by sex.


Earnings of individual i in year t (yit) are regressed on a dummy variable Si indicating post-secondary participation in the year following displacement and on a vector xi containing several earnings determinants (measured in 1997, the year prior to displacement), including age, age squared, a vector of province of residence dummies, a vector of industry dummies, and a vector of firm size dummies.

Selection into training is a concern in this specification, leading to a correlation between Image and S. Displaced workers may selectively choose training on the basis of unobserved characteristics. For example, highly productive workers may be more likely than less productive workers to train given differences in expected returns. Alternatively, workers with the lowest opportunity costs (i.e., those who are least likely to find employment) may be most likely to train. The former case would generate a positive bias of the effects of training on earnings, while the latter would generate a negative bias. 11 

If the unobserved characteristics which cause selection into training do not vary over time for each individual, these fixed influences can be removed by estimating first-differences, comparing the change in earnings for those who attend post-displacement education with the change in earnings for those who do not. This is a simple difference-in-difference estimator. 12 


Selection may also be based on time-varying unobserved characteristics, in which case estimates from Equation (2) will still be biased and inconsistent. For example, displaced workers may choose to train if they know that their skills (unobserved to the researcher) have been deteriorating in the years leading up to displacement. As there is five years of pre-displacement information, one can examine whether the treatment and control groups have similar pre-displacement trends in earnings.

4.2  The impact of displacement on educational participation

In the second part of the paper, the impact of job displacement on post-secondary education attendance is estimated. Since the dependent variable in this case is post-secondary attendance, information for each year pre- and post-displacement is required. Consequently, the analysis is limited to the 1999 to 2007 period (when post-secondary attendance is reliably available on the LWF). The cohort of workers who were potentially displaced in 2003 is selected, yielding pre- and post-displacement periods of four years each. Other sample criteria are similar to those used in the earnings analysis. Specifically, workers who had positive earnings in each of the pre-displacement years and were between the ages of 25 and 44 in the year prior to displacement are selected. The sample design in this case is shown in Figure 2.

Figure 2

Equation (3) summarizes the regressions that are estimated. Separate OLS models are estimated by year and by sex, as before. As before, Si,t is a dummy variable indicating attendance at a post-secondary institution in year t, Di is a dummy variable indicating displacement in the year 2003, and xi is a vector of characteristics (measured in 2002, the year prior to potential displacement), which includes age, age squared, paid earnings, a vector of province of residence dummies, a vector of seven industry dummies, and a vector of four firm size dummies.


As in the previous section, it is assumed that there are unobservable characteristics which may be correlated with the decision to enter post-secondary education and with the probability of being displaced. This is dealt with in two ways. First, if the unobservable characteristics are not time-varying, fixed-effects methods can be used. This compares the change in training participation between the treatment and control groups between 1999 and 2004.


There may also be time-varying unobserved characteristics which affect both S and D. For example, some workers may have been laid off because their employer saw their motivation levels deteriorate over time. The solution to this second identification issue is to focus on workers who were displaced for reasons not likely related to their individual productivity. Specifically, all workers who were displaced for reasons other than a firm closure or a mass layoff are dropped. Mass layoffs and firm closures are more likely to be associated with declining market demand for the firm's output rather than an individual's productivity. Equation (3) is thus re-estimated, where Di is redefined to mean displacement in 2003 due to a mass layoff or firm closure (excluding workers displaced for other reasons).

5   Results

5.1  The impact of post-secondary education attendance on earnings

In this section, the link between post-displacement training and earnings is examined. Chart 1 shows the estimates of αSt from Equation (1). Equation (1) provides an estimate of the gap in earnings between those who attended post-secondary education in 1999 and those who did not. At this stage of the analysis, no covariates are included.

In the years leading up to displacement, females who subsequently attended post-secondary schooling earned about $2,000 more than those who did not subsequently attend. This suggests positive selection into the treatment group. In contrast, the male treatment group earned about $3,000 less than the control group before displacement. The pre-treatment difference in earnings is highly significant but quite stable up to and including 1997. In 1998 (the year of displacement), one observes a fall in the earnings of the male treatment group compared to the male control group. There are a number of possible explanations for this. It might be that displaced workers in the treatment group are displaced earlier in the calendar year. Perhaps more likely is the possibility that workers who found work soon after being displaced (possibly before the end of 1998) were then much less likely to enter training programs in 1999.

For males, participation in training is associated with a large temporary loss of earnings in 1999, the year in which training credits are received. 13  However, by 2002, the earnings of men in the treatment group and control group are equal, and from 2004 onwards the earnings of those in the treatment group are significantly higher despite having been significantly lower before displacement and training took place. By 2007, displaced men who attended post-secondary education institutions earned $6,719 more than non-trainees. This benefit alone (in a single year) is approximately equal to the loss in earnings experienced in 1999.

For females, participation in training is not associated with a large temporary fall in earnings relative to the control group, a result which suggests that women attended shorter courses or were more likely to be earning at the same time as participating in education. From 2000 onwards, their earnings relative to those of female non-trainees increases in a pattern similar to that observed among men. By 2007, displaced women who had trained earned $11,541 more than those who had not.

These raw estimates imply a large earnings premium to post-displacement training. A crude difference-in-difference calculation which takes into account pre-displacement differences suggests increases around $9,000 for both men and women.

To begin addressing the selectivity of training, observable characteristics by training status are examined; results are shown in Table 1. Individuals who attended a post-secondary education institution are younger, more likely to reside in British Columbia, and less likely to be married. There are also some differences in their pre-displacement firm characteristics: trainees are more likely to be employed in public services, less likely to be employed in manufacturing, and more likely to be employed in large firms.

To the extent that these characteristics are associated with the earnings trajectory of displaced workers, the results should take differences in these factors into account. Chart 2 provides estimates of αSt estimated from Equation (1) with a full set of control variables xi included (for the year 1997). The chart shows that the pre-displacement difference in earnings between trainees and non-trainees is almost entirely explained by differences in observable characteristics. Estimates of αSt for males are insignificantly different from zero for 1993–1997, although there is still a noticeable decline in male trainees' earnings relative to those of male non-trainees in the year of displacement (1998). For females, the inclusion of covariates reduces the positive pre-treatment gap in wages only slightly. For both males and females, the inclusion of covariates reduces the implied impact of training on earnings to about $6,500.

In Table 2, estimates of the fixed-effects model from Equation (2) are reported. Estimates of the size of the training effect in 2007 are very similar to the informal difference-in-difference calculations from Chart 2. By 2007, the treatment effect is $6,551 for males and $6,672 for females.

The size of the sample permits the estimation of Equation (2) separately by various characteristics. There are significant effects by age, marital status, and union membership for both men and women, with the exception of 35-to-44-year-old men. There are also statistically significant differences in the effects by each of the characteristics for men only. Specifically, men who were older, married, or part of a union prior to being displaced benefit less from training (compared to their respective comparison groups). For women, no significant differences are observed.

5.2  The impact of displacement on post-secondary education attendance

The analysis conducted so far suggests that job training following displacement may increase earnings. 14  Chart 3 displays the gap in post-secondary attendance rates between workers who were displaced and those who were not in 2003 (βDt from Equation (3), but without covariates).

Four years prior to the year of potential displacement, there is a significant positive gap in the post-secondary education attendance rates of those workers who are subsequently displaced and those who are not. This gap becomes larger the closer one gets to 2003; this finding suggests that workers may have anticipated the possibility of job loss in the future and prepared for this by investing in their human capital. 15  In 2003, the gap is 3 percentage points for males and nearly 5 percentage points for females by 2004. This represents a large proportional difference, since mean attendance rates for non-displaced workers in 2003 are only 8% (males) and 11% (females). However, the difference in participation rates peaks either in 2003 (males) or in 2004 (females), and displacement is not associated with a large increase in post-secondary education attendance overall.

Differences in observable characteristics between the displaced workers and the non-displaced workers are shown in Table 3. In general, displaced workers are slightly younger, earn much less prior to displacement, and are less likely to be married. Differences exist in terms of union coverage, but the gap runs in opposite directions for men and women. Some minor differences in province of residence also exist. In terms of pre-displacement firm characteristics, displaced workers are more likely to be employed in primary industries or construction, are less likely to be employed in public services, and are more likely to be employed in a small-to-medium sized firm (no more than 100 employees).

The differences between these displaced workers and those that were not displaced are taken into account in the OLS results shown in Chart 4. Despite the large differences in worker characteristics, the results do not change in any meaningful way. It is observed that the same pattern of a slow upward creep prior to displacement occurs, then a sudden spike during the year of displacement, followed by a downward trend. Again, the spike is larger for women. In general, the magnitudes of these trends are more or less the same as before, and the general conclusion remains the same: displacement does not lead to a large increase in post-secondary attendance.

As a result, differences in pre-displacement and post-displacement education attendance rates are not due to differences in observed characteristics. As before, one can control for fixed unobserved differences between displaced and non-displaced workers by estimating a fixed-effects (difference-in-difference model. Results of this estimation are reported in Table 4. When all displacements are considered (as has occurred so far), the results suggest a small effect on post-secondary attendance (1.6 percentage points for men and 3.3 percentage points for women, both statistically significant at 1%). However, when one estimates the effect of displacement resulting from a mass layoff or firm closure, the effects are much smaller: 0.6 percentage points for men (significant at 5%) and 1.3 percentage points for women (significant at 1%). These are the preferred set of results since they are less likely to be contaminated by differences in unobserved characteristics. Note that these effects are small in relation to the overall rates of post-secondary attendance, which in 1999 were between 13% and 14% for men and between 17% and 19% for women (Appendix Text table 3). The results are also disaggregated by age, marital status, and union coverage. Modest effects are generated for all groups of women. For men who are aged 35 to 44, married, or not unionized, there is no evidence whatsoever of an uptake in post-secondary education.

6   Conclusion

The economic downturn that began in late 2008 and the resultant job losses have renewed interest in the outcomes of displaced workers. The purpose of this study was to investigate the effectiveness of one response to mitigating earnings losses following job displacement: training in a post-secondary institution. Given the possibility of selection into the treatment, the approach used in this paper has been to estimate a difference-in-difference (fixed-effects) model.

The results suggest that, over the period spanning five years preceding and nine years following job loss, workers who attended post-secondary education shortly following displacement saw their earnings increase by almost $7,000 more than displaced workers who did not attend post-secondary education. Significant benefits are found by sex, age, marital status, and union coverage, with the exception of men aged 35 to 44. Despite the benefits of training, job displacement is found to be associated with only a modest increase in post-secondary attendance for all groups examined.

These results suggest a substantial benefit of post-secondary education for displaced workers. The key issue is whether one can interpret these earnings differences as the causal impact of education, or whether the different earnings paths of the treatment and control groups are the result of non-random selection into the training program. Further analysis will allow for worker-specific time-trends in earnings, because workers who choose training after displacement may have a different pre-displacement pattern of earnings growth.

Date modified: