The evolution of job stability in Canada: trends and comparisons with U.S. results

Transcription

1 The evolution of job stability in Canada: trends and comparisons with U.S. results Andrew Heisz Business and Labour Market Analysis Division, Statistics Canada Abstract. Using data from the Canadian Labour Force Survey, we examine the stability of currently held jobs and find no period-long drop in job stability. However, job stability declined across the 1980s and rose across the 1990s for workers with less than one year of tenure. When 1987 and 1995 are compared, it can be seen that job stability was steady in Canada but fell slightly in the United States, with the difference concentrated among medium tenured workers. We suggest that this difference was due to a slower recovery in Canada in the 1990s, which caused Canadian workers to be less mobile. JEL classification: J63 L e volution de la stabilite des emplois au Canada: tendances et comparaisons avec les re sultats aux Etats-Unis. En utilisant les donne es de l Enqueˆte sur la population active pour la pe riode de 1976 a` 2001, l auteur examine la stabilite des emplois de tenus pour le moment et trouve qu il n y a pas eu de chute de la stabilité des emplois en longue pe riode. Cependant, la stabilite des emplois a chute dans les années 1980 et cruˆ dans les années 1990 pour les travailleurs qui de tenaient leur emploi depuis moins d un an. Si l on compare 1987 et 1995, la stabilite des emplois se maintient au Canada, mais chute un peu aux Etats-Unis: la diffe rence est concentre e dans le groupe de travailleurs qui ne de tenaient leur emploi que depuis une pe riode moyennement longue. On sugge` re que les diffe rences entre les deux pays sont attribuables à une reprise e conomique plus lente au Canada dans les années 1990 qui aurait rendu les travailleurs canadiens moins mobiles. An earlier draft of this paper, co-authored with Mark Walsh, was presented at the meetings of the Canadian Economics Association at the University of British Columbia in May Valuable comments were received from David Green. andrew.heisz@statcan.ca Canadian Journal of Economics / Revue canadienne d Economique, Vol. 38, No. 1 February / février Printed in Canada / Imprimé au Canada / 05 / / # Canadian Economics Association

2 106 A. Heisz Three important elements of job quality are wage, hours and duration. While there is substantial recent evidence on the Canadian wage and working hours distributions, there has been much less evidence produced on the stability of Canadian jobs. In the light of the large changes observed in the wage and hours distributions, including the declining wages of young men, rising relative wages of women, increasing polarization of weekly work hours, and increasing use of part time workers, one might expect to see a change in job stability. In fact, some have seen the contemporary Canadian labour market as being divided into good jobs with high wages, benefits, and security, and bad jobs with none of these. This reflects a belief held by many that job stability is on the decline in the industrialized world as globalization and new technologies fundamentally change the employer-employee relationship. 1 Job stability may change in the short run, owing to shifts in job opportunities for workers or increased layoffs associated with changes in the business cycle, but a long-run decline in job stability would signal an important development in the labour market. A structural (or long-run) decline in job stability implies that workers may experience less accumulation of job related skills; no wage increases through seniority; limited access to employer-based pension plans and training; and increased exposure to unemployment. Overall, a decline in job stability might signal an important decline in job quality. Recent evidence from the United States indicates that job stability in that country has been steady in recent decades. Farber (1995) showed that the distribution of job tenure did not change in any remarkable way in the United States from 1973 through 1993, with the notable exception that long-term jobs became more scarce for less educated men. Diebold, Neumark, and Polsky (1997) used a different methodology and concluded similarly. Neumark, Polsky, and Hansen (1999, 2000) updated the latter work, and found that aggregate job stability declined slightly in the first half of the 1990s and more so for long-tenured workers, but they concluded that there was no evidence that this was a permanent decline in job stability. 2 Some evidence on job stability exists for Canada, but up to now research has addressed different time periods and used methods that render them hard to compare and draw conclusions from. Christophides and McKenna (1993, 1995) found that job duration fell between the mid- and late 1980s. Green and Riddell (1997) found that there had been a hollowing out of the tenure 1 Riddell and Gunderson (2000) provide a recent survey of the Canadian job quality literature. 2 Further evidence from the United States includes Hall (1982), Ureta (1992), and Swinnerton and Wial (1995, 1996). While these authors focus on results from cross-sectional surveys, other studies have looked at U. S. job stability using panel data. Marcotte (1999) examines workers under age 45 from 1976 though 1992 and finds a mild decline in job stability over this period. Monks and Pizer (1998) and Bernhardt et al. (2000) examine the job stability of young men who were aged at the beginning of the panel and find a decline in job stability between the 1970s and 1980s. Jaeger and Huff-Stevens (2000) compare results from U.S. cross-sectional and panel data and find similar trends across datasets.

3 The evolution of job stability in Canada 107 distribution between 1979 and 1991 such that by 1991 there were more shorttenure and more long-tenure jobs and a tendency for shorter jobs for the youngest and least educated. Heisz (1999) reported a fall in expected job duration for older and less educated workers from 1981 to Picot, Heisz, and Nakamura (2001) found that job duration rose substantially in the 1990s, offsetting declines in the 1980s and rising to its highest level since The picture that emerges from these studies is that jobs have been remarkably stable in Canada over the past two decades, although stability in Canada declined across the 1980s, particularly for the less educated, and rose in the 1990s. However, this picture is incomplete and leaves several important questions unanswered. 3 The motivation for this paper is to advance our understanding of trends in Canadian job stability using up-to-date data and employing a widely used and accepted methodology. It is the first Canadian study in which job stability over two complete business cycles is examined. Previous studies have focused on a relatively short period of time, or on changes across only one business cycle. Our approach allows us not only to describe movements in job stability over the business cycle, but also to abstract from changes in job stability that are related to the business cycle and search for evidence of a structural decline in job stability. We also examine whether the apparent stability in Canadian jobs found in earlier studies was due to compositional effects, as workers age and move into more stable tenure categories, or if job stability is constant within groups. To examine changes in job stability, we replicate the method used in recent U.S. studies, which, as we discuss below, is superior to that used in some previous Canadian studies. In particular, we replicate the approaches used in Diebold, Neumark, and Polsky (1997) and Neumark, Polsky, and Hansen (1999, 2000) (hereafter referred to as DNP and NPH, respectively). Upon completing indicators for Canada, we compare these with published results from NPH (2000). This comparison is rendered especially important given the U.S. evidence on stable jobs discussed above. If job stability has declined in Canada over the period, then this may reflect a relative decline in job quality in Canada. 1. Data and measurement issues We obtain job tenure information from the Canadian Labour Force Survey (LFS), which has been conducted monthly with few important changes since 1976 and is a representative sample administered on approximately 60,000 3 Hasan and de Broucker (1985) represents an early study of job stability in Canada. Canadian research has also focused on permanent layoff and quit rates. Picot, Lin, and Pyper (1998) examine permanent layoff rates in Canada over 1978 to 1994 and find that layoffs became more common during recession years, but were not higher in the 1990s than the 1980s. Picot, Heisz, and Nakamura (2001) find the quit rate to be low in the 1990s compared with that in the 1980s.

4 108 A. Heisz households. The LFS is available monthly from 1976 to (at the time of writing) 2001, permitting an examination of job stability across two complete business cycles. The LFS is similar in content to the American Current Population Survey (CPS), which provides the data used in the American studies we wish to compare our results with. However, there are some important differences between the two surveys that make the LFS much better suited to studying job tenure. The LFS collects information on job tenure in each month, while the CPS collects job tenure information in a supplementary survey that has been conducted at irregular intervals of at least two years. The regular frequency of the LFS allows one to compare retention rates across short intervals and choose points near business cycle peaks. Additionally, because the CPS asks the tenure questions in a supplement, there is important non-response that does not occur in the LFS. Also, the tenure question in the LFS has been consistent since 1976, while in the CPS the data series is broken by a change in the question between the 1981 and 1983 tenure supplements. As noted in many other studies, small differences in surveys can have important effects on measures of job stability. Even though the CPS and LFS are similar surveys, one should be cautious when comparing levels of job stability between the two countries. However, differences in changes over time in the two countries estimates should be more comparable, and it is these that we focus on in the inter-country analysis. 4 For our analysis of Canadian trends, our sample includes workers aged 15 and over who were paid employees or self-employed owners of incorporated firms between 1976 and We exclude unincorporated self-employed workers because the tenure question asked of these workers is different. When years close to cyclical peaks are compared, the fraction of workers covered in this sample is 90.7% in 1980, 90.6% in 1989, and 90.1% in For the intercountry comparison we choose a more restricted sample based on the criteria of NPH. This sample includes only non-agricultural workers aged 16 and over who were paid employees or self-employed owners of incorporated firms. In the latter sample we have data only from 1976 to This is because the LFS moved from the SICC80SE classification of industries to the NAICS classification between 1998 and 1999, which makes it impossible to exclude the exact same classes of agricultural workers after Unlike CPS data, Canadian data also show little evidence of response heaping, so we do not adjust for it in the Canadian data. More discussion regarding the survey questions and response heaping is available in Heisz (2002). Longitudinal data, such as the Labour Market Activity Surveys of and or the Survey of Labour and Income Dynamics of , could be used to examine job stability (as in Christophedes and McKenna (1993, 1995, 1996), for example). While these surveys benefit from true longitudinal data and an expanded set of covariates, they suffer from short time horizons. Neither of these surveys includes data from a complete business cycle, which is essential for evaluating trends in job stability. In addition, establishing comparability between these longitudinal datasets and longitudinal datasets available from the United States would be difficult.

5 The evolution of job stability in Canada In-progress job length There are a number of different meaningful ways to measure job stability. A common approach is to examine the average tenure of currently employed individuals. We refer to this as the average in-progress job duration. It does not reflect the completed tenure of jobs, but rather the length of jobs at the time of the survey. From table 1 it can be seen that the average in-progress tenure of workers rose substantially over the 1976 to 2001 period. However, the in-progress tenure distribution is inappropriate for examining changes in job stability over time. To illustrate this point, consider the following formula for the average in-progress job length: average in-progress job length c ¼ P n t ¼ 1 P n t ¼ 1 N 0, c t S t, c t t N 0, c t S t, c t : (1) N 0,c t is the number of workers starting jobs in period c t and S t, c t is the survival rate, or the probability that a job that begins at time c t will last at least t periods, where t ranges from 1 to n, n representing the largest tenure value observed in the data. For evidence on changes in job stability, one would want to look directly at changes in the survival rates. However, the average in-progress job length in period c is affected by the level of inflows in all previous periods in which someone currently with a job became employed and all the survival rates in those periods (as are all points in the in-progress tenure distribution). Hence, changes in these statistics over time tend to be difficult to interpret. 5 Nevertheless, the in-progress job tenure distribution remains interesting to examine. It reveals the present state of the labour force and may be most easily related to other phenomena, such as changes in wages or employment. Furthermore, if the average in-progress job length rises, as it did in table 1, and jobs are more stable when they are in this advanced stage, then the average worker will appear to have more job stability. Likewise, if the in-progress distribution becomes more polarized, with more workers having short in-progress job tenure and more with long in-progress tenure, then larger shares of workers will be found in less stable and more stable phases of their jobs, even though job stability itself may not have changed The retention rate A better tool for measuring job stability is the retention rate which is the conditional probability that a job will continue for some specified period of 5 It is also important to note that the in-progress tenure distribution is a biased distribution of spell lengths. First, the spells are sampled in progress. They may end the next day or they may end far in the future. Second, in a point-in-time survey, the probability of sampling a spell is proportionate to its length, making the distribution of in-progress jobs heavily weighted by long spells (Salant 1977).

6 110 A. Heisz time, given that it has reached a certain initial level of tenure. Denoting tenure as t, time as c, and the retention rate R t,c, the retention rate can be derived using two consecutive cross sectional surveys as R t, c ¼ N t, c =N t i, c i : (2) This is simply the number of respondents reporting tenure of t in the present survey divided by the number of respondents reporting tenure of t i in a previous survey. 6 R t,c is one minus the hazard rate, and a full set of retention rates defines a survival function. In standard survival model terminology, N t, c is the group surviving and N t i, c i is the group at risk. The symbol i refers to the interval width, or the spacing of the surveys that is measured in the same units as t, while the value t i is referred to as the initial tenure. The computation of the retention rate using cross-sectional data is an application of the synthetic cohort approach, wherein representative individuals from the same job entry cohort are sampled for the numerator and denominator. Retention rates can also be computed for any demographic subgroup defined by the data. 7 The retention rate does not suffer from the biases that affect the distribution of in-progress tenure. Since it measures changes for a single job entry cohort, it is not biased by the rate of inflow into jobs. Furthermore, it measures single points in the survivor function, so changes in retention rates across a fixed period of time can be associated with economic events that occurred between those dates. Retention rates may be defined over any interval permitted by the data. The LFS has collected tenure information monthly since 1976, so retention rates could be computed for intervals as short as one month. In the American literature, retention rates are computed for four- and eight-year intervals, owing to the irregular frequency of the CPS tenure supplements. In principle, retention rates calculated across shorter intervals are preferable in that they allow one to more effectively tie changes in retention rates to economic events, such as turning points in the business cycle. In what follows we compute oneyear retention rates for an analysis of Canadian trends and four-year retention rates for an analysis of inter-country differences. Shorter intervals of less than one year tended to be unstable, reflecting sampling error introduced into the process by the synthetic cohort approach. Once a full set of retention rates are computed, a single average retention rate can be derived by R c ¼ 0 R 1c þ 1 R 2c þ 2 R 3c þ..., (3) 6 Hall (1982), computed retention rates using a single cross section of data. Ureta (1992) demonstrates that retention rates calculated from a single survey are biased. 7 When we compute retention rates for age categories, it is necessary to choose the surviving group such that the age of the surviving group is equal to the age of the at-risk group plus the retention rate interval.

7 The evolution of job stability in Canada 111 where i is a weight that represents the fraction of the at-risk population in the tenure category i. Likewise, subgroups of retention rates can be combined to reduce the number of rates to analyse. We compute 30 retention rates for initial tenure values of 0 to 29 years and analyse the following groups of retention rates by initial tenure: initial tenure <1 year ¼ R 0 <1 ¼ R 1 initial tenure from 1 to <2 years ¼ R 1 <2 ¼ R 2 initial tenure from 2 to <9 years ¼ R 2 <9 ¼ 2 R 3, c þ 3 R 4, c þ 4 R 5, c þ...þ 8 R 9, c initial tenure from 9 to <15 years ¼ R 9 <15 ¼ 9 R 10, c þ 10 R 11, c þ 11 R 12, c þ...þ 14 R 15, c initial tenure of 15 years or more ¼ R 15þ ¼ 15 R 16, c þ 16 R 17, c þ 17 R 18, c þ...þ 29 R 30, c : (4) Note that t is indexed in years such that R 1,c ¼ N months, c /N 1 12 months, c 12 months. Weights are defined such that the sum of s equals one for whichever summary retention rate is being derived. Four-year rates are defined similarly with R 1,c ¼ N months, c /N 1 12 months, c 48, and for comparisons to U.S. results we group R 1 and R 2. The convention used is to define retention rates according to the time period identified in the numerator. Thus, using 1976 to 2001 data, we can compute monthly one-year retention rates for and four-year rates for Retention rates and their summarized values reflect the average experience of the currently employed, and examining job stability conditional upon initial tenure is one way to account for rising in-progress tenure among workers. Alternatively, one can generate a single indicator of job stability that is not affected by changes in the in-progress tenure distribution by setting initial tenure at its value observed in a single year. In the notation of equation (3) we hold i values at period start values, yielding a job stability indicator for a representative group of workers with fixed initial tenure Standard errors We compute standard errors of retention rate estimates as outlined in NPH (2000). This method models the retention rate as a binomial random variable, where the retention rate is the proportion of successes, and the variance is appropriately adjusted upward to account for the fact that we are using synthetic cohort data rather than actual longitudinal data. Unweighted cell 8 We also examine fixed initial tenure indicators for the other grouped retention rates R 2 <9, R 9 <15, and R 15þ. Since changes in these composition constant retention rates tend not to be statistically different from changes in unadjusted retention rates, we do not report them.

8 112 A. Heisz counts were used to compute standard errors. From equation (1) in NPH (2000, 75) we define the standard error as rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi nsample p (1 p) s ¼ nrisk 2, where p ¼ nsurv nsample, (5) where nsurv is the unweighted count of the surviving group in period c, nrisk is the unweighted count of the at-risk group in period c i, and nsample is the unweighted count of all workers in period c. As an alternative to this, we also tested for significant changes across time periods using a weighted least squares regression where the dependent variable was the log of the retention rate and the weights used were the counts of observations observed to be at-risk. This method provided similar standard errors and did not change our results in any important way. The LFS is conducted using a rotational design that has households rotating into the sample for six months at a time, and one-sixth of the sample is replaced each month. The job tenure question is asked in the first interview and then validated in subsequent interviews. 9 When computing descriptive statistics and charts, we use all 12 months of data; however, for the purposes of computing tables that contain standard errors we use only the March and September surveys. These surveys are seven months apart and represent two independent samples that we can use to compute retention rates and their standard errors in each year. We sum the at-risk group in March and September and the corresponding surviving group in March and September of the next year before computing retention rates. This tends to add additional stability to retention rates when they are computed for small sub-groups of data. 10 Choosing other pairs of months does not affect the results, as one would expect given that the rotational design of the survey determines that only a minority of the sample changes from month to month. For comparisons with U.S. results, we use only the January LFS, since January is the CPS tenure supplement reference month. 2. The in-progress job tenure distribution We first turn our attention to conducting a review of developments in Canadian job tenure. Table 1 shows mean in-progress tenure for all workers 9 Researchers working with U.S. panel data have noted that individuals often give tenure values that are inconsistent across interviews. This inconsistency can be corrected if the individual s employer is collected (Bernhardt et al. 2000). Given that the CPS tenure supplements do not involve repeated sampling, no such consistency check can be done. In the LFS, the employer name is collected in the first interview and validated, in subsequent interviews, along with the job start date, which likely minimizes inconsistencies in responses across subsequent surveys. 10 For data in tables that are presented as three-year averages, we sum at-risk and survival groups across six independent samples.

9 TABLE 1 Average in-progress tenure (months) Age Education Age and Education Composition Constant Year All Men Women þ High school or less Some post-secondary University or higher Men Women

10 114 A. Heisz and is subdivided by sex, age, and education. Over the period, mean tenure increased considerably (from 78 months in 1976 to 92 months in 2001). Women, who closed the gap significantly with their male counterparts, fuelled this rise. This long-term change for women is related to the rising attachment of females to the workforce since the 1960s. That is, women became less likely to withdraw from the workforce, thereby increasing the probability that longer-tenured women would be drawn from the survey. 11 The increase in mean in-progress tenure is also related to the ageing of the Canadian workforce. Mean values within age groups changed little, indicating that rising aggregate tenure can be attributed to between group shifts to older workers with traditionally longer tenures. In-progress tenure also rose over the period for all education groups, but there was little difference in in-progress tenure across education groups by The final columns of table 1 show conditional average tenure of men and women, holding the age and education distributions constant. Trends in conditional tenure are muted, with changes in worker characteristics accounting for the entire rise in men s and two-thirds of the rise in women s in-progress tenure. The percentages of workers with tenure of less than two years and tenure of more than ten years are shown in table 2. The fraction of jobs with an in-progress length of less than two years declined over the period for men and women, indicating that fewer were in the early phase of jobs. Like the average, the fraction of jobs in these categories is affected by the composition of workers. The conditional distribution of in-progress job lengths, which adjusts for changes in the age and education composition of workers, is shown in the last columns of table 2. Adjusting for changes in the age and education composition of workers, we find that the fraction of jobs with an in-progress length of less than two years rose from 34.9% in the late 1970s to 38.7% in the late 1990s for men, while the rise was muted for women, from 44.5% to 45.8%. Among men, the increase in short in-progress tenure was observed among all age groups, and increased most for men aged across the 1980s when it rose from 70.7% to 76.4%. The fraction of jobs that were short tenure also rose for men with high school or less education over the period and young women. The fraction of jobs with more than ten years tenure rose across the 1980s and fell in the 1990s for men, and rose over the period for women. These results are consistent with those seen in the existing literature on Canadian job stability (for example, Green and Riddell 1997). For men, the conditional in-progress tenure distribution filled with more short in-progress jobs, indicating that more men were more likely to be in the least stable phase of their jobs (conditional on their age and education). For women, there was a steady rise in in-progress tenure as more were found in long-term stable jobs. These trends are also consistent with others seen in the labour market, 11 Women s employment rates rose from 42% to 55.6%, while men s dropped from 72.7% to 67% between 1976 and 2001.

11 The evolution of job stability in Canada 115 TABLE 2 Short and long in-progress tenure Age Education Age and All þ High school or less Some post secondary University or higher Men % jobs with less than two years in-progress tenure % jobs with more than 10 years in-progress tenure Women % jobs with less than two years in-progress tenure % jobs with more than 10 years in-progress tenure education composition constant including increasing earnings polarization for men and relative improvement in earnings of women compared with those of men. However, they do not necessarily reflect a change in job stability over this period, since, as shown in equation (1), these changes could be related to historical changes in job stability or to changes in inflow rates. In the next section we directly examine the job survival function by looking at retention rates. 3. One-year retention rates Figure 1 shows the average one-year retention rate for all jobs in-progress. The data are characterized foremost by a positive trend and broad cyclical swings. Cyclical movements in job stability are the net outcomes of changes in the quit rate and the permanent layoff rate. The changes observed in figure 1 suggest that the pro-cyclical quit rate dominates during boom periods, while the counter-cyclical permanent layoff rate dominates in the event of economic downturn. The retention rate rose during periods of labour market slack, for example, between 1982 and 1983 and 1991 and 1994, and fell during boom periods such as and , when opportunities to advance through changing jobs were most likely to be present. It also dropped in 1982 and was low in 1991, consistent with higher layoffs in those years. Interestingly,

12 116 A. Heisz year FIGURE 1 One-year retention rate one-year retention rates continued to rise into the late 1990s, even though the economy was in an expansionary phase, and dropped between 1999 and 2001, consistent with more opportunities for mobility in those years. 12 Data points in figure 1 represent the average retention rate faced by workers in each respective year. Underlying changes in average job stability are changes in job stability conditional on initial job length. Figure 2 shows retention rates for all workers according to their initial tenure. Job stability rose up to 9 <15 years of initial tenure, after which stability fell. However, unlike the average retention rate shown in figure 1, none of the retention rates in figure 2 shows a strong trend increase over the whole period. The most striking change is a rise after 1993 in the probability that a job with initial tenure of 1 12 months will continue for one more year. This statistic dropped from a 1977 level of 46% to 40% in 1993, and rose steadily in the 1990s to 54% by Changes in job stability for other initial tenure groups were less dramatic. The one-year retention rate for jobs between one and two years long rose in the early 1990s as workers who entered jobs in the 1990s recession held their jobs longer than in other years. The one-year retention rate for jobs initially 2 <9 years 12 Green and Riddell (1997) examine the five-year retention rate for the period for initial tenure of zero to five years. They find that this retention rate declined over the period, which is consistent with our results. OECD (1997) calculates five-year retention rates for several countries, including Canada for the years 1985, 1990, and 1995.

13 The evolution of job stability in Canada less than 1 year 1 to < 2 years 2 to < than 9 years 9 to < 15 years 15 or more years year FIGURE 2 One-year retention rate, by initial tenure long rose in the recoveries of the 1980s and 1990s. Long job stability did not change substantially over the period. Given the relative stability of retention rates for initial tenure groups, it must be that much of the trend rise in the average one-year retention rate was due to a rise in in-progress tenure. In figure 3 we show average one-year retention rates, holding the initial tenure distribution constant at period start values. Here, we see a much less dramatic rise in retention rates. The unadjusted retention rate rose 3.8 percentage points between 1977 and 2001 compared with 1.6 percentage points for the initial tenure distribution constant retention rate over the same period. Similar results are observed by sex. Figure 4 shows average one-year retention rates (both unadjusted and with the initial tenure distribution held constant) for men and women. Trend increases for men and women were also mostly the result of workers shifting into more stable tenure categories. Table 3 presents the results of a regression designed to identify trends and the influence of the business cycle on retention rates. The dependent variable is the log of the one-year retention rate computed for March and September in each year from 1977 to 2001, for each of the sex and initial tenure subgroups identified in the table. This is regressed on the log of the unemployment rate, where the unemployment rate is averaged across the 13 months comprising t to t 12, a quadratic trend, and monthly dummies. The regressions are conducted using weighted least squares, where the weights are given by the unweighted count of

14 118 A. Heisz 0.8 unadjusted initial tenure distribution constant in year FIGURE 3 One-year retention rate observations at risk. The trend variables are scaled such that the sum of the trend and trend-squared coefficients approximates the percentage change between 1977 and 2001, after holding the unemployment rate constant. The estimates indicate that the unadjusted average one-year retention rate rose by 4.8% across the period for men and women combined. For men the increase was 2.9% and for women it was 8.1%. The signs of the trend variables show that job stability fell significantly over the 1980s and rose over the 1990s for men and women (the turnaround year was 1993). After adjusting for changes in the in-progress tenure distribution the trend increase is 1.4% for all workers, 1.3% for men and 1.9% for women. Changes are concentrated mainly among workers with initial tenure of less than one year, as indicated by the insignificant trend estimates for the other tenure groups. After controlling for this non-linear trend, job stability is shown to be counter-cyclical rising during recessions and falling during recoveries. Most of this counter-cyclical movement appears to come from medium initial length jobs of 2 to 15 years long, and is stronger for women than men By demographic group In table 4 we examine trends in one-year retention rates for demographic subgroups. We abstract from changes in retention rates that are related to the business cycle by comparing rates in years close to business cycle peaks: , , and Labour market conditions were likely to be most similar in these years, so a change in retention rates between these periods

15 The evolution of job stability in Canada 119 Men Women unadjusted initial tenure distribution constant in 1976 unadjusted initial tenure distribution constant in year year FIGURE 4 One-year retention rate, by sex would indicate a change in job stability, other things equal. The first panels examine one-year retention rates for men. The unadjusted average one-year rate rose from to over the period, up 2.9%. Retention rose over the period for male workers with tenure of less than two years and fell for male workers with tenure between two and nine years, with most of the latter change occurring in the 1980s. Controlling for changes in the initial tenure distribution results in a more muted increase in the retention rate from to 0.789, a 1.3% increase. Stability fell for young men aged 15 24, but not for other men. This stability in retention rates for most age groups suggests that some of the remaining increase in the unadjusted average one-year retention rate was due to ageing of the workforce within initial tenure categories. Controlling for both the in-progress tenure distribution and the age distribution, we see that the oneyear retention rate was in the late 1970s and in the late 1990s, which was not a statistically significant increase. The decline in job stability across the 1980s remains significant after controlling for changes in the initial tenure distribution and age. This decline was noted for all tenure and age groups but was largest for men with tenure of less than one year and those aged The one-year retention rate also increased over the period for women, up from in the late 1970s to in the late 1990s, an increase of 7.2%. Retention rose for women with less than one year, 2 <9 years, and more than 15 years of initial job tenure, fell for women with 1 <2 years of tenure, and remained steady in the 9 <15 year tenure category across the three cyclical peaks. When the initial tenure distribution is held constant the increase in the one-year rate fell to 2.7%. Job stability fell for young women and rose for older women, but when the age and initial tenure distributions are held constant, the remaining increase in job stability

16 120 A. Heisz TABLE 3 Trends in one-year retention rates Trend Trend squared ln (UR) Men and women Unadjusted initial tenure *** *** *** 0 < *** *** < < *** 9 < þ Initial tenure distribution held constant *** *** *** Men Unadjusted initial tenure *** *** *** 0 < *** *** < < ** 9 < þ Initial tenure distribution held constant *** *** ** Women Unadjusted initial tenure *** *** *** 0 < *** *** < < *** 9 < ** 15þ Initial tenure distribution held constant *** *** *** NOTES: ***, **, and * indicate that the difference is significantly different from zero at the 1%, 5%, and 10% levels respectively. Weighted least squares regression of the log of the retention rate on a quadratic trend, the log of the unemployment rate, and monthly dummies. Weights are given by the unweighted count of observations at-risk. The trend coefficient is scaled so that the estimate represents the percentage change in the dependent variable over the period. The unemployment rate is averaged across the 13 months comprising t to t 12. The unweighted number of observations in each regression is 50, comprising March and September values (two independent samples) for the respective retention rates. was negligible, with the one-year retention rate at in both and As was the case for men, the decline in job stability across the 1980s was most concentrated among women with low tenure and those aged Table 4 also stratifies by educational attainment. An important change was introduced to the education question in the LFS, which resulted in some regrouping of responses by educational attainment beginning in January It is unclear what effect changes in the education question would have on retention rates. While the change affected the proportions of respondents reporting certain educational attainments, it may be that these changes cancel out in the numerator and denominator of the retention rate, leaving the rate unaffected, as long as the same question was applied in period c and c i. This approach prevents computing a one-year retention rate for Visual inspection of the retention rates seems to support this approach, with changes

17 The evolution of job stability in Canada 121 TABLE 4 One-year retention rates Period (1) (2) (3) Men Unadjusted *** *** By initial tenure 0 < *** *** 1 < * 2 < *** *** 9 < þ Initial tenure distribution held constant *** *** By age a *** ** ** þ ** Initial tenure distribution and age held constant *** By educational attainment High school or less *** Some post-secondary ** *** University Women Unadjusted *** *** By initial tenure 0 < *** *** 1 < * 2 < * 9 < þ * Initial tenure distribution held constant *** *** By Age a *** *** *** *** *** *** 55þ *** Initial tenure distribution and age held constant *** By educational attainment High school or less *** 0.76 *** Some post-secondary *** University *** Men and Women Unadjusted *** *** Initial tenure distribution held constant *** *** (Continued)

18 122 A. Heisz TABLE 4 (Concluded) across the survey designs being attributable to cyclical factors. The fall in retention rates across the 1980s was concentrated in the high school or less category for both men and women. While increases across the to period should be interpreted with caution, we find that retention rates rose for all education groups across this period, which is consistent with the patterns of results we saw with other groups. To sum up this section, job stability measured by retention rates declined across the 1980s and recovered in the 1990s. The final panel of table 4 shows job stability for men and women combined. After controlling for the initial tenure distribution and the age distribution, the conditional one-year retention rate fell from in to in and recovered to in , indicating no period-long decline. The decline across the 1980s was concentrated among low-tenured workers, young men and women, and men and women with high school or less education. Job stability rose for most groups across the 1990s. Job stability fell significantly between the late 1970s and the late 1990s only for men with 2 <9 years of job tenure, women with 1 <2 yearsofjob tenure, and men and women aged Given the small size of these changes, the trend towards increasing job stability across the 1990s for most workers, and the magnitude of variations over the business cycle observed in these data, we conclude that these changes are not likely indicative of long-term trends Canada - U.S. comparisons Period (1) (2) (3) Initial tenure distribution and age held constant *** ***, **,and * indicate that the retention rate is significantly different from the value at the 1%, 5%, and10% levels, respectively. Standard errors are computed as described in the text. a Age refers to the age of the at-risk group, or the denominator of the retention rate. For comparisons with U.S. data, we generate a four-year retention rate (figure 5). As noted above, we use January data and a slightly more restricted sample for these series in order to improve comparability with U.S. results. The four-year retention rate shows movements over the period similar to the one-year rate (although at a lower level). As with the one-year rate, the probability that all 13 Job stability of young workers may also be affected by school enrolment patterns. The fraction of the year old population that was enrolled full or part time in school rose from 40.9% in September 1976 to 50.7% in September 1989 and further to 58.9% by September The period-long decline in job stability for this low-education group may have been the result of increased school attendance, since those in school are less likely to be firmly attached to the labour market

19 The evolution of job stability in Canada 123 FIGURE 5 Four-year retention rates, Canada and United States NOTES: Canada-U.S. comparable sample includes non-agricultural workers aged 16 and over who were paid employees or self-employed owners of incorporated firms. Canada- full sample includes all workers aged 15 and over who were paid employees or self-employed owners of incorporated firms. *Source: Neumark, Polsky, and Hansen (2000) **Source: Swinnerton and Wial (1995) jobs would last an additional four years appears high in the late 1990s. However, cyclical movements overshadow any trend in the four-year rate. Given the magnitude of changes over the business cycle, it is clear that it is quite important which years one compares when looking for long term changes. The four-year retention rate appears to peak two to three years before the business cycle peak in 1987 and 1997 reflecting heightened job stability during the recovery years. Ideally, one would compare Canadian and U.S. results at years close to their respective business cycle peaks, but this is not possible given the limits of the U.S. data. The figure also shows results for Canada, including agricultural workers and 15-year- olds and indicates that these restrictions make little difference. U.S. values are also shown in figure 5. In most of this discussion we compare Canadian results with those obtained from NPH (2000) for the United States for 1987, 1991, and 1995, but in this graph we also add the four-year retention rate value estimates for 1983 from Swinnerton and Wial (1995), although this point is not strictly comparable with the NPH results. 14 Despite the differences in the 14 See Neumark, Polsky, and Hansen (2000) for a discussion of differences between their results and those of Swinnerton and Wial.

20 124 A. Heisz TABLE 5 Four-year retention rate comparisons: Canada and U.S. (selected years ) Canada a United States b change change change change All groups ** ** ** Initial tenure 0 < ** ** ** 2 < ** ** ** ** 9 < ** ** ** ** 15þ ** Age ** ** ** ** ** ** ** ** ** ** 55þ ** Sex Men ** ** ** ** Women ** ** ** ** Denotes change is significant at 5% level. Standard errors are computed as described in the text. a Canada data is based on January LFS data. b Source: Neumark, Polsky, and Hansen (2000), tables 3.2, 3.3, and 3.5. NPH reported upper and lower bounds for their four-year retention rate for For our inter-country comparison, we use the lower bound estimates from NPH (2000), which are identified as their preferred estimates. survey instruments, the four-year retention rates are quite similar in level and also show qualitatively similar changes across the four points. Four-year retention rates rose in both countries after the recession, fell across the later 1980s as the economy picked up steam in both countries, and rose again in the 1990s, although the magnitude of this increase, at least up to 1995, washigherincanada. Table 5 shows results for four-year retention rates for the United States and Canada. Over the period average retention rates fell in Canada by 0.3 percentage points (a statistically insignificant drop) and fell in the United States by 1 percentage point. In many ways, the similarity in the changes across countries is striking. Job stability rose substantially for low initial tenure jobs in both countries, with the bulk of the growth happening the 1990s. A decline in job stability for jobs longer than two years was noted in both countries, although the declines were larger in the United States, especially in the 9 <15 years tenure category. In the United States there were declines in job stability for each age group, while in Canada significant declines were only noted only for workers aged Women increased job stability and men

21 The evolution of job stability in Canada 125 lost job stability in both countries. Across most groups the direction of change in job stability was the same in both countries, with a relative shift towards more instability in the United States. 15 Why the relative decrease in job stability in the United States across the 1990s? One possible explanation relates to the slow economic recovery in Canada in that decade, a period that was particularly hard for Canadian workers. 16 It may be that in the face of poor alternative job prospects, Canadian workers tended to stay longer in their jobs than in previous recoveries and, in turn, longer than their American counterparts, who enjoyed the benefits of a quicker recovery and expansion. We saw above that job stability tends to move in a counter-cyclical manner and that most of the counter-cyclical movement comes from jobs with medium-length initial tenure. A relative rise in unemployment in Canada may have driven a relative increase in job stability for medium tenured workers seen in table 5 as a relative rise in retention rates for workers with 9 <15 years initial tenure. However, we are cautious in this interpretation, since it is based on only a few data points, particularly when it is shown using Canadian data that job stability moves substantially with changes in the business cycle. 5. Conclusion In this paper we attempt to fill a gap in the Canadian job quality literature by looking for evidence of a structural (that is, long-run) decline in job stability. Examining retention rates, we find that job stability was remarkably constant between the late 1970s and the late 1990s. Unadjusted for changes in the composition of workers, we see a rise in one-year retention rates of about 3% for men and 7 8% for women, but after we control for changes in the in-progress tenure distribution and the age composition of the workforce, the difference between the late 1970s and the late 1990s was negligible. We note that job stability fell across the 1980s, especially for workers with less than one year of tenure, for young workers and for workers with a high school or less education. However, it recovered in the 1990s. Comparing the late 1970s with the late 1990s, we also find that all age groups of men were 15 Comparison of job stability in the two countries is limited by differences in the way each country measures educational attainment and changes in classification by educational attainment within both countries. Up to 1992 the education question in the CPS asked for the number of years of education completed, after which it asked for the highest grade completed. The LFS always asked for the highest grade completed, but, as noted above, the way groups were classified changed in Available evidence suggests that job stability fell for the least educated in both countries across the 1980s. In DNP (1997) the four-year retention rate for workers with high school or less education fell from to between 1987 and 1991 (table 3B). Our computations show the four-year rate falling from in 1980 to in 1989 for men and from to for women. 16 See Banting, Sharpe, and St-Hilaire (2001) for a review of economic developments in Canada in the 1990s.