Notes

^1.

As mentioned above, these numbers include both temporary and permanent layoffs. As will be shown below, annual estimates of the total number of temporary and permanent layoffs obtained from the LFS are, during the 1978-to-2007 period, fairly similar to those obtained from the LWF. Since estimates of the number of permanent layoffs from the LWF currently end in 2007, they cannot be used to quantify permanent layoff rates during the most recent downturn.

^2.

In absolute terms, the peak-to-trough employment decline amounted to roughly 610,000 in the early 1980s, compared to roughly 440,000 in the early 1990s and roughly 430,000 in the last recession (CANSIM Table 282-0087).

^3.

The LFS sample is drawn from an area frame and is based on a stratified, multi-stage design that uses probability sampling. In addition to adjustments for the probability of being surveyed and for non-response, stratification and clustering in the sample design are also required in order to produce proper standard error of the estimates. The authors thank Emmanuel Benhin and Scott Meyer, both of Statistics Canada, for useful information related to this adjustment. More details are provided in Appendix III.

^4.

Without an individual identifier in the Labour Force Survey Master File, a (pseudo) individual identifier is created by combining the household identifier (HHLDID), the rotation number identifier (ROTATION), a sequential number assigned to every dwelling within a cluster (LISTLINE), a code used for structures that have more than one dwelling (MULT), a sequential number that uniquely identifies a person within a household (LINE), and a family identifier that uniquely assigned an economic family to a household (FAMID). Given that each individual should be followed for, at most, six consecutive months, a handful of observations were dropped when the particular individual identifier was linked to more than six records in the data (0.01% in the 1981-to-1984 data, 0.03% in the 1990-to-1994 data, and no such observations in the 2008-to-2011 data). Among the individual identifiers that were linked to six records or less in the data, reported individual characteristics were checked to ensure that they were recorded consistently across the sampling months. This was done in order to ascertain that the linked records belonged to the same individual. Specifically, individuals with change in reported sex, individuals with an increase in reported age of more than one year, and individuals with change in the highest education attainment to a lower level within the six survey months were dropped. Such occurrences were rare, however (0.30% among the 1981-to-1984 data, 0.25% among the 1990-to-1994 data, and 0.78% among the 2008-to-2011 data).

^5.

From CANSIM Table 282-0089, the seasonally-adjusted total employment number is as follows: 11,375,600 in June 1981 and 11,386,700 in October 1984; 13,140,800 in April 1990 and 13,176,600 in September 1994; and 17,175,100 in October 2008 and 17,214,500 in January 2011.

^6.

Non-employment includes both the unemployed and those who dropped out of the labour force.

^7.

In the LWF, a temporary (permanent) layoff occurs when a laid-off worker does (does not) return to his/her employer during the same year or during the year following the layoff.

^8.

Since one of the goals of the study is to compare layoff rates across downturns and since peak-to-peak employment periods cross calendar years, averages of month-specific layoff rates (rather than year-specific layoff rates) have to be used.

^9.

A worker is defined as being laid-off if he/she had been laid-off at least once between the second month and the fifth month of the survey. This includes individuals with multiple layoffs. Cases of multiple layoffs are relatively rare: they amount to about 4% of the sample of individuals who had been ever been laid-off between the second month and the fifth month.

^10.

This is done by selecting workers whose first month of interview is observed during a "peak-to-peak" period and by using the following sets of panels: 40 (six-month) panels from June 1981 to February 1985; 53 panels from April 1990 to January 1995; and 23 panels from October 2008 to January 2011. January 2011 is the most current LFS data that could be accessed at the time that the paper was written. The LFS data from October 2008 to January 2011 allow for calculating the "peak-to-peak" layoff rate for the 2008-to-2011 period. However, in order to follow workers for the full six-month panel for the December 2010 cohort, LFS data up to May 2011 are needed. The current paper therefore can follow the workers whose first month of interview took place between October 2008 and August 2010 in order to examine the short-term consequence following their layoffs for the most recent downturn.

^11.

As a result of the introduction of a new set of LFS questions measuring educational attainment in 1990, detailed information on education levels for years prior to 1990 is not comparable to that measured for subsequent years. As a result, only the disaggregation used in Table 1 can be presented.

^12.

Industries are aggregated into the following six categories based on the North American Industry Classification System (NAICS) 2002 codes: 1) "Primary industries and construction" include agriculture, forestry, fishing, mining, oil and gas, utilities, and construction; 2) "Manufacturing"; 3) "Retail trade, accommodation, and food services"; 4) "High-skill services" include finance, insurance, real estate, and leasing, as well as professional, scientific, and technical services, business, building, and other support services; 5) "Public services" include education services, health care and social assistance, and public administration; 6) "Other service-producing industries" include wholesale trade, transportation and warehousing, performing arts, heritage, and amusement.

^13.

Occupations are aggregated into the following five categories based on the National Occupational Classification for Statistics (NOC-S) 2001 codes: 1) Management; 2) Professionals; 3) Semi-professionals and technicians; 4) Clerical, sales, and service personnel; 5) Manual workers and trades personnel. See Appendix IV for details.

^14.

In the LFS, standard NOC-S 2001 occupational codes are available from 1987 to the present. Therefore, the occupation comparison can be done only between the 1990-to-1994 and 2008-to-2011 downturns.

^15.

One exception is retail trade, accommodation, and food services, which registered a 2-percentage-point drop in the share of laid-off workers, most of which would have taken place even in the absence of changes in the composition of employment by industry (Table 2, columns 3 and 4).

^16.

The difference between the 2008-to-2011 layoff rate and the layoff rates of previous periods is statistically significant at conventional levels.

^17.

Text Table 1 in Appendix II shows that LFS estimates of the total number of layoffs are fairly similar to those obtained from the LWF. During the period 1978-to-2007, the total number of (permanent and temporary) layoffs per year averaged 2.5 million with the LFS, compared to 2.7 million with the LWF. While the estimated number of layoffs from LWF changed little between 1996 and 1997, the estimated number of layoffs from the LFS fell by roughly 15% between these two years. This suggests that the 1997 redesign of the LFS led to a reduction in the estimated number of layoffs. Scaling up the 2008-to-2011 average monthly layoff rate of 2.0% by a factor of 1.15 yields a revised layoff rate of 2.3%, which is still lower than the average monthly layoff rates observed during the previous two downturns.

^18.

The relevant tabulations for expansionary periods are available upon request.

^19.

Calculations of probabilities of being laid-off for both groups of workers, conditional on time-invariant means of other explanatory variables, confirm this point.

^20.

This can be seen by comparing the (positive) marginal effect for the Atlantic Provinces to the most negative marginal effect observed among the remaining provinces/regions.

^21.

Gender differences are also observed, but they are not qualitatively stable over time. Women were slightly more likely than men to be laid-off during the first two downturns; this was not the case with the most recent one, however.

^22.

The difference between the re-employment rate of 2008-to-2011 and the re-employment rates of previous downturns is statistically significant at conventional levels. Adding the set of explanatory variables shown in Table 6 to a vector of period effects in a regression that pools data from all three downturns indicates that about 17% of the increase in employment rates between the early 1980s and the most recent downturn can be accounted for by changes in the composition of laid-off workers by age, education, seniority, region, and industry of employment. Adding only a binary indicator for being a university graduate (in a regression that pools data from all three downturns) suggests that about 4% of the increase in employment rates between the early 1980s and 2008-to-2011 can be accounted for by the growing proportion of university graduates observed over the last three decades.

^23.

The proportion of laid-off employees who became self-employed shortly after being laid-off amounted to 1% during the first two downturns and to 2% during the most recent one.

^24.

Table 6 presents regression results from a linear probability model of re-employment. Similar results were obtained by using logit models and probit models.

^25.

The use of contract workers in the public sector is one consideration. Of all workers laid-off from public services during the most recent downturn, about 30.1% were in temporary, term, or contract jobs. This is three times the rate of 9.9% observed for workers laid-off from other industries. However, adding a binary indicator measuring whether the end of a job is related to a temporary, term, or contract job reduces the coefficient estimate for public services only slightly in 2008-to-2011 (from 21.9 percentage points to 21.3 percentage points). Since this indicator is not available during earlier periods, this alternative specification cannot be replicated for the early 1980s and the early 1990s.

^26.

Percentage changes are obtained by taking the antilog of the numbers shown in Table 10, minus 1. Thus, hourly wage losses of 16% equal e^-0.18-1.

^27.

Together, these three groups account for 25% of re-employed laid-off workers in 2008-to-2011.

^28.

Together, these two groups account for 16.5% of re-employed laid-off workers in 2008-to-2011.

^29.

Changes in log weekly hours are also regressed on the aforementioned variables, as can be seen in the fifth column of Table 11. On average, workers who were laid-off during the most recent downturn and found a paid job shortly after being laid-off saw their average weekly work hours drop by 0.6 hours, from 34.5 before the layoffs to 33.9 after the layoffs. The corresponding decreases are -0.7 hours during the early 1980s and -1.2 hours during the early 1990s, respectively.

^30.

This gender difference does not hold for the weekly wage changes in the 2006-to-2008 multivariate analysis (see Text Table 5).

^31.

The percentage decline in wages is obtained by taking the antilog of the numbers shown in the first two columns of Table 12, minus 1. The large wage losses resulting from transitions across industries and occupations likely reflect the loss of a combination of skills: firm-specific skills, industry-specific skills (Neal 1995), and occupation-specific skills (Poletaev and Robinson 2008).