Chapter 15 The distinctive patterns of work-to-retirement transition among the self-employed1
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.
With special assistance from Harpreet Kaur Randhawa
Growth of self-employment in Canada among older workers
Trend in the average age of retirement among the self-employed
The distinctive patterns of the transitions to retirement of the self-employed
General summary and policy relevance
Among several OECD countries, the aging of population, the growth in the ratio of retirees to workers and related changes in retirement patterns and longer life expectancy have raised concerns about the viability of social security programs. This has brought into focus the issue of securing adequate labor supply in future years (OECD 2001). Labor supply of older workers will be a key factor in the resolution of this issue. Improving labor supply of older workers will require a greater prevalence of gradual and flexible work-to-retirement transitions. A number of experts concur that this achievement requires attention to opportunities for self-employment (see ILO 1990, Parker and Rougier 2004, Morris and Mallier 2003 and Taylor 2002).
Although self-employment will be a prominent aspect of labor market participation among older workers in the years ahead (see Morris and Mallier 2003, Parker and Rougier 2004 and Taylor 2002), their labor supply decisions and the related patterns of transition to retirement are an area in which few research studies have been published. Among them are Fuchs 1982, Quinn and Kozy 1995, Bruce, Holtz-Eakin and Quinn 2000, Parker and Rougier 2004, and Karoly and Zissimopoulos 2004.
None of these studies deals with Canada, and we have found only one Canadian study that devotes attention to transitions to retirement among self-employed workers (Sunter 2001). Therefore, to support both policy deliberations and teaching about a major dimension of later-life transitions in Canadian society, it is worthwhile to devote this chapter to the distinctive patterns of work-to-retirement transitions among the self-employed in Canada.
The questions addressed in this chapter are as follows: What are the major or systematic differences between the patterns of transition to retirement of the self-employed and those of workers who were receiving wages and salaries as employees? (The latter are called "salaried employees" or simply "employees" below.) Are the main patterns in these differences evident after we hold constant a set of variables that could simultaneously determine both the probability of being self-employed and the pattern of transition to retirement, and what is the relative importance of self-employment as a variable in helping to predict how a cohort transiting to retirement will be distributed over alternative patterns of transition to retirement?
Before going on to outline how we developed answers to the questions cited above, we present, some contextual information about recent developments affecting older self-employed workers in Canada, in the next two sections.
The labor market experienced dramatic changes during the 90s. The corporate and government down-sizing in the mid-1990s resulted in a significant shift in the types of jobs created, and this led to a relatively rapid growth of self-employment among older workers. During the period 1990 to 1999, among those aged 45 and above the average annual growth rate for the self-employed was 6.4% while it was 3.4% for the wage and salary earners. The spurt in self-employment growth during this period was especially strong among women. Their average annual growth rate was 9.4% while it was 5.3% for men.
There was an abrupt downturn in the growth rate of self-employed from 2000 to 2003 (the average annual rate fell to 2.7% for women and to 2.4% for men). It seems likely, however, that with the massive future wave of retirements among baby boomers, there will be a further rise in the rate of self-employment. (For related discussion see Quinn 2000.)
There has been a systematic tendency for the self-employed to retire two or more years later (on average) than wage and salary earners. This pattern has been found by researchers in the USA, going back to work in the early 1980s (Quinn 1980 and Fuchs 1982).
Chart 15.1 shows that among the self-employed, since the mid 1970s the average retirement age has fluctuated within a band marked by 65 and 67. Statistics for the median age of retirement among the self-employed, show the same fluctuation just cited; except that for this measure the range is from 64 to 66 years (Statistics Canada 2004:81).
In contrast to the pattern just noted for the self-employed, the average retirement age for employees had a marked downward trend from 1976 to about 2000, going from nearly 65 to about 61. However, since 2000 the curve has flattened and is hinting at a modest rise; though in 2004 it was still below 62. Thus, among employees, the trend toward retirement well below the age of 65 remains in force in 2004, even though a decline in the curve seems to have stopped.
The net result of this contrast between the self-employed and wage-and-salary earners is that the gap between their average retirement ages has widened. It was about two years in the late 1970s, and now it is close to four years. This finding suggests that whether self-employment resumes strong growth among older workers will be a key factor in achieving a substantial rise in the trend toward more gradual retirement.
The data source for this analysis is Panel Two of the Survey of Labour and Income Dynamics (SLID). The target population for SLID is all persons living in Canada, excluding people in Yukon or Northwest territories, residents of institutions, persons living on reserves and full-time members of the Canadian Armed Forces living in barracks. The exclusions constitute about 3% of the population.
The members of Panel Two of SLID were repeatedly interviewed about 12 times, from 1996 to 2001.The data for the first two years have been used to apply our scale for determining which panel members had begun their transitions to retirement during 1996 to 1997. Focusing on these members, we then mapped transitional patterns from 1998 to 2001. (Additional information about SLID is given in Appendix A, see also Statistics Canada 2004.)
As already noted, Panel Two comprises the same sample of main respondents and their household members, starting in 1996 (initial interviews) and ending in 2001. Our plan to show results for Panel One (1993 to 1998) had to be abandoned because some key variables needed for this analysis began to be measured only in the 1996 Panel.
This chapter focuses on those members of Panel Two whom we judge to have begun their transitions to retirement between 1996 and 1997, and were aged 45 to 69 in 1996.2Appendix A describes how a sub-group within this age range is identified to have started their transitions in the 1996 to 1997 period.
Several concepts that require special definitions are used in this chapter. The next few paragraphs present definitions and discuss measurement for the special concepts that are unique to this chapter. They are "self-employed" and "salaried employee" (the more accurate term for the latter is "wage and salary earner").
Generally, a self-employed person is an independent business owner who is active in the operation of the business, often as a sole proprietor or partner. This person may or may not have employees. When there are no employees, he/she is called an "own-account self-employed person".
The salaried employee, is defined as one who, provides paid labor services in an organization that is owned by another person (or a group of persons that exclude this employee).
A notable issue here is that many people both own (and help to operate) a business and are employees in another organization. The classification of a worker as being self-employed (or not) is usually based on her/his main job.
The categories explained above are used in this study to define a two-year class-of-worker variable, for the period 1996 to 1997. Two categories of this variable are the focus of our attention: (1) those who were self-employed in both 1996 and 1997 as their main job and had no wage and salary earnings, and (2) those who were salaried in both years and had no self-employment.
Table 15.1 shows the breakdown of those who started their transitions during 1996 to 1997 according to the categories of the said two-year class-of-worker variable, by sex and age in 1996. A shortage of sample observations for important sub-groups is indicated by this table. However, the number of self-employed respondents is similar to that found in some other studies that have focused on this group of older workers (see Quinn and Kozy 1995, and the samples of self-employed in Parker and Rougier 2004, Fuchs 1982, Quinn 1980, and Zissimopoulos and Karoly 2003).
To compare patterns of transition to retirement between employees and the self-employed, one must define the phrase "transition to retirement", and devise a scale to identify who began transiting to retirement within a specified time period.
In this chapter, "transition to retirement" refers to a process that often involves multiple movements among designated positions toward the state of being retired. ("Retirement" here means the state of being retired, which is often marked by a long-term departure from the labor market, accompanied by the reception of some kind of retirement-related income.) The specific sequence of positions that comprises one person's transition to retirement is called her "trajectory of transition to retirement". (See Appendix A for details.)
Trajectories have properties that we can define and measure. And we compare patterns of transition to retirement between the self-employed and the salaried in terms of these properties of trajectories of transition to retirement.
This chapter deals with will four properties of trajectories of transition to retirement. These properties are speed of closure, presence of flexibility in the work-to-retirement transition, exposure to events that threaten to reduce standard of living in retirement and propensity to return to the labor market after leaving it. The following paragraphs will offer brief remarks about each of these properties, to allow the reader to understand the research findings that follow without having to spend a lot of time studying the more detailed information given in Appendix A.
Speed of closure. A person (identified as being in transition during 1996 to 1997 is said to have closed her trajectory when (a) she has left the labor market and has been in receipt of some form of retirement-related income for at least six-consecutive months, and (b) following those six months she did not return to the labor market up to the end of 2001. The sooner the person begins this period of uninterrupted departure from the labor force (while receiving some form of retirement-related income) the faster is her speed of closure.
Closure can begin in any quarter from the first of 1998 to the third of 2001, i.e., 15 quarters. A sixteenth category contains those whose trajectories were unclosed as of the end of December 2001 (for details see Appendix A.) Although this implies 16 levels of speed of closure, based on the quarter as the unit of time, they are grouped into four broad speeds, when comparing the patterns of employees with those of the self-employed.
Flexibility of the work-to-retirement transition. In this chapter, flexibility in the transition to retirement is measured by means of counting certain movements that form part of a person's trajectory of transition to retirement. The movements in question are those that indirectly point to certain voluntary actions - those that reflect the use of available options or choices in how one's retirement process unfolds.
Exposure to events that increase vulnerability. Whereas the index of flexibility deals with movements that suggest voluntary changes undertaken by the SLID respondent, the index of vulnerability focuses on job loss and involuntary job change. In this text "vulnerability" means risk of loss or setback. Here we are referring to risk of setback to whatever plans or arrangements the person has made concerning standard of living in retirement.
Propensity to return to the labor market after leaving it during 1996-97. Finally, we devised a special scale of propensity to return to the labor market, after leaving it at some time during the 1996 and 1997. The measure of the propensity to return to the labor market rises with the number of returns (starting in the first quarter of 1998) and the length of stay in the market for each return. All of these returns form part of a person's trajectory of transition to retirement.
The next few paragraphs outline the general strategy of analysis, using the property "speed of closure" as an example to make the discussion more concrete.
Speed of closure is represented by an index with four rank-ordered categories: rapid closure, moderately rapid closure, slow closure and very slow closure (this last being the set of unclosed trajectories). (Defining four speeds involves collapsing the 16 categories initially used in Appendix A when trajectories were introduced.) At the stage of multivariate analysis, the model will predict the probability distribution of trajectories over these four categories, and one of the predictor variables will be based upon the special two-year class-of-worker variable defined above.
Before going to the multivariate analysis, differences in speed-of-closure pattern between the self-employed and the salaried will be portrayed in the context of bivariate association. We use the model to re-estimate the pattern of these differences in terms of a comparison of conditional odds ratios for the two categories, as well as of the computed marginal effects (on the probability of any category of speed of closure) of shifting (analytically) from one category to the other. The terms "conditional" and "marginal" used here are meant to point to the fact that several other determinants of speed of closure are held constant statistically when making the computations.
Speed of closure is just one of the properties for which models were developed. The models for the other properties are presented in Appendix C The results presented in this chapter deal exclusively with the role of self-employment in the models. (Methodological issues concerning the models are discussed in Appendix C.)
Our central research question is the following: what are the major or systematic differences between the patterns of transition to retirement of the self-employed and those of workers who were receiving wages and salaries as employees? Related to this is a key subsidiary question: what is the relative importance of class of worker (of which self-employment forms one category) among the variables that are useful in a model that generates probabilities for different classes of transition from work to retirement? The remainder of this chapter provides our partial answers to these questions, dealing with the each of the properties of trajectories.
Speed of closure of trajectories
Bivariate analysis. On average, the speed of closure of the trajectories of the predominantly wage and salary workers, who will be called "employees" or "the salaried" from here onwards, is noticeably faster than that of the self-employed. Over 30% of the salaried closed their trajectories before the end of the third quarter of 1998, as opposed to slightly less than 15% of those who were self-employed in both years (Chart 15.2).
Distribution according to speed of closure of trajectories, for self-employed and salaried employees, cohort aged 45 to 69 in 1996, by sex, Canada, 1998 to 2001
Across all three slower speeds of closure shown in Chart 15.2, the percentages for the self-employed are higher than those for the salaried. At the last quarter of 2001, just over 50% of the employees had unclosed trajectories, while a slightly higher percentage of the self-employed had unclosed trajectories. Thus, for the population in transition to retirement, the speed of closure of trajectories was systematically slower for those self-employed in both years (1996 and 1997) than for the group of employees.3
When those in transition to retirement as of 1996-97 are classified into sub-groups according to gender, age in 1996, household income in 1996 and education, there is a tendency toward repetition of the pattern reported above - the trajectories for persons who were self-employed in 1996 and 1997 showed systematically slower speeds of closure than those of employees. The next few paragraphs will summarize the key patterns among selected sub-groups.4
Chart 15.2 shows that, among those who started their work-to-retirement transitions during 1996 to 1997, men and women had similar patterns as that for both sexes taken together. But there is a slight variation - in the case of women the percentage with unclosed trajectories among the employees is slightly higher than that of the self-employed. This divergence for women is confirmed when the sample is enlarged by lowering the TRANSCOR boundary to 2.1; but is not confirmed among all women aged 55 or more. 5 In the latter category of women the direction of difference between self-employed and salaried is the same as that for men.
Focus upon the 55 to 59 and 60 to 64 age groups is warranted. In Canada, opportunities to retire early without major penalties in pension entitlements begin to become available for many employees at age 55.and without penalities after age 60. Thus, across the 45 to 69 age range the propensity to have begun transitions during 1996 to 1997 was at its highest in these age groups, as Appendix A shows. Within each of these age groups the above-mentioned basic pattern of differentials in speed of closure of trajectories between the salaried and the self-employed recurs.
Moreover, the difference in the speed of closure between the self-employed and employees is sharpened, by comparison with what one sees when all the age groups are lumped together into one sample (Chart 15.2). For example, in Chart 15.2 the gap between the self-employed and employees as regards the percentage that began closing their trajectories in or before the third quarter of 1998 is roughly 15 percentage points, with the higher figure belonging to the salaried. However, in Chart 15.3 the corresponding gap jumps to nearly 30 percentage points for those aged 55 to 59 in 1996 and to over 20 percentage points among those aged 60 to 64 in 1996. Because the two groups have very similar percentages at the next two speeds of closure, that gap necessarily sets up a large corresponding gap in the category of unclosed trajectories. This substantial increase in the size of the gap between self-employed and salaried, when one focuses on these two age groups is confirmed with the two larger samples cited above.
Speed of closure of trajectories of transition to retirementby self-employment status, cohorts aged 55 to 59 and 60 to 64 in 1996, Canada, 1998 to 2001
In short, the gap between the self-employed and the salaried as regards the delay of retirement is greatest in the age groups where the rate of beginning transitions to retirement is highest.
Wide differences in income do not change the pattern being discussed here. Chart 15.4 allows one to make a comparison between those in the lowest and the highest quartiles of household after-tax income in 1996. Despite this major difference in household income, the pattern of variation in speed of closure of trajectories between the self-employed and employees is very similar to that which we noticed for the whole population aged 45 to 69 in 1996 (Chart 15.2).
Speed of closure of trajectories of transition to retirement by self-employment status, cohorts aged 45 to 69 in 1996, by household income in 1996, Canada, 1998 to 2001
When workers with less than high school graduation are compared with those that had university degrees, among those who began their transitions during 1996 to 1997, a notable divergence from the patterns for the whole sample emerges (see Chart 15.5). The former group follows the pattern of the whole sample (Chart 15.2), while those with university degrees diverge. The gap between employees and self-employed as regards rapid closure of trajectories nearly disappears (both having close to 20%). The same is true in the category of unclosed trajectories. This divergence is also seen in the larger sample that comprises all persons aged 55 or more in 1996. An analysis designed to confirm and explain this divergence, using another dataset where much larger samples are available, could provide some useful lessons concerning the influence of higher education on the propensity to delay retirement among employees, and the reasons for that influence.6
Speed of closure of trajectories of transition to retirement by self-employment status, cohorts aged 45 to 69 in 1996, by education level in 1996, Canada, 1998 to 2001
In summary, as regards patterns in speed of closure, among those that started their work-to-retirement transitions between 1996 and 1997, and for key sub-groups as well as the sample as a whole, fast closure of trajectories is much more likely among the salaried than among the self-employed. At slower speeds of closure, the self-employed take the lead. However, by the time the six-year observation ended (in 2001), and unclosed trajectories are tallied, the lead of the self-employed is slight in the whole sample of persons aged 45 to 69 in 1996. However, in the sub-sample of those aged 55 or more, where the proportions transiting to retirement become high, the gap between the self-employed and the salaried remains marked in the category of unclosed trajectories. In this sub-population with high probability of being in transition to retirement, the gap between the two groups is in the order of 20 percentage points or more (at each end of the spectrum of speeds of closure).
Multivariate analysis. To what extent is this type of pattern supported by data that arise from a multivariate analysis where several pertinent variables are held constant?
To answer the question just posed, the text that follows presents the results of fitting a multivariate model that generates predicted conditional probabilities for the different speeds of closure. The model makes use of several predictor variables, and for each combination of values for these variables a separate probability distribution of speed of closure is predicted. Among them are variables that could have important separate associations with the propensity to be self-employed and the probabilities of specific speeds of closure - this would lead us to regard the association between the latter two as spurious (i.e., lacking in any causal significance, regardless of any indicated statistical significance). To check whether the pattern of association described above is repeated among these distributions, we hold constant at selected values all variables except those that pertain to class-of-worker status (self-employed versus salaried). By examining how the distributions change when we shift analytically from the self-employed to the salaried (this is the so-called "marginal effect"), we test both the direction and the strength of association of class-of-worker status with speed of closure.7
The selection of variables to be held constant is based on our theoretical considerations about the processes involved in determining the gradualness of retirement, or on findings from other research projects. The variables held constant statistically include sex, age, cultural background, marital status in 1996, whether marital status changed in the year before closure of trajectory began, an index of whether responsibility for providing family care increased in the year before closure of trajectory began, whether health status declined in the year before closure of trajectory began, education level in 1996, occupation group in 1996, an index for irregularity of work history, whether there was another economic family member who began receiving retirement-related income in the year before closure of the trajectory started, an index of wealth ranking and an index of increase in wealth in the year before commencement of the closure of the trajectory. (Appendix C presents the model, brief definitions of the variables, discussing its overall performance and the contributions of important variables that are not part of the story-line of this chapter. Appendix B presents detailed technical definitions of several complex variables defined specially for this study and included in the model.)
Table 15.2 supports the information provided above based on bivariate analysis. The average predicted conditional probability of closure in or before the first three quarters of 1998 is 0.09 for the self-employed and 0.21 for employees - twice as high for employees as for the self-employed. At each of the other (slower) speeds, the self-employed have higher probabilities; but by the time unclosed trajectories are tallied the gap falls to 0.05 (five percentage points - 0.63 for the self-employed and 0.58 for the salaried).
Marginal effects, on speed of closure of trajectory, of the difference between (a) being self-employed in both 1996 and 1997, and (b) being salaried in both years, Canada, 1998 to 2001
The marginal effects shown in Table 15.2 are derived from coefficients of the kind shown in the top line of Table 15.3 - the parameter estimates and their associated Wald chi-squares. These data merely repeat, in a different, (but more comprehensive way, the message of Table 15.2. (Appendix C provides assistance in interpreting the data presented by Table 15.3.)
The heading above the row descriptions in Table 15.3 does not refer to self-employment in both years because, in the model fitted, the variable used to represent self-employment includes persons self-employed in both years as well as those who switched from being self-employed in 1996 to being employed in 1997. This was done to increase the reliability (from the viewpoint of sampling) of the parameter estimate for self-employment. Having fitted the model, when generating its predicted probabilities (to show marginal effects) we reverted to comparing self-employed in both years with being salaried in both years (Table 15.2).
Adjusted patterns of association of self-employment with speed of closure of trajectory of work-to-retirement transition during 1998 to 2001, for three nested models, Canada
In interpreting the data in Table 15.3 focus should be placed on the key sub-model - that in column C (the concept of sub-model is introduced in Appendix C). The positive parameter estimate in column C indicates that, relative to the reference category of class of worker (primarily salaried persons), being self-employed tends to increase the probability of having an unclosed trajectory. The odds ratio of 3.4 means that the odds favoring the self-employed, with regard to having unclosed trajectories, far exceed (are three times as high as) those favoring the salaried, under the key sub-model. Finally, although age is hugely dominant among the predictor variables concerning speed of closure, being self-employed contributes roughly 8% of the model's goodness of fit. The self-employment variable is also ranked number three (out of 16 variables), third highest, in the set of predictor variables, in terms of relative statistical importance in the model's goodness of fit.8
It is worth noting that the results for the first two sub-models (see columns A and B) are subject to much higher variability due to sampling than those of the third sub-model (see the values of the Wald chi-squares, and footnote 4 to the table). However, they too show a broadly similar pattern. Being self-employed boosts the probability of slower speed of closer.
In summary, the bivariate and the multivariate analyses concur as regards the association of class-of-worker status (self-employment versus salaried employment) with speed of closure. In the bivariate analysis, we estimated a minimum 20-percentage-point gap between the two groups in terms of concentration at either end of the spectrum of speed of closure, for those persons aged 55 or more in 1996. In the whole sample aged 45 to 69, when several related covariates are held constant, the gap falls to about10 percentage points when the speed of closure is fast, and to five percentage points at the opposite extreme of non-closure.
Presence of flexibility in the work-to-retirement transition
With regard to trajectories, flexibility in the transition to retirement involves the presence of changes in position that indirectly point to voluntary actions that reflect the use of available options or choices in how one's retirement process unfolds. (See Appendix A for elaboration of this idea.) The existing literature concerning retirement among the self-employed repeatedly emphasizes the advantage in flexibility that the self-employed enjoy by comparison with the salaried. However, other researchers have rarely devised a measure of this flexibility and used it to gauge the size of the advantage enjoyed by the self-employed.
It is precisely this kind of information that we are offering based on studying the trajectories that were presented in Appendix A. Our focus is not on establishing an already well-repeated finding by other researchers; but upon adding value by measuring the degree of advantage that the self-employed enjoy relative to the salaried.
Chart 15.6 confirms the expected substantial difference between self-employed and salaried status as regards flexibility of the work-to-retirement transition. Some 86% of the salaried had trajectories with Low levels on the index (14% with Medium or High levels). The corresponding figure for self-employed is at least 10 percentage points lower. Among the Medium and High levels of the index, the gap is most pronounced at the Medium level.
Index of flexibility in the work-to-retirement transition, for self-employed and salaried employees, cohort aged 45 to 69 in 1996, Canada, 1998 to 2001
This prominent difference between the two class-of-worker statuses is seen within categories of other important variables such as sex, age and education, as Chart 15.7 shows. This chart deals with the percentages at Medium or Higher levels of the index of flexibility of the work-to-retirement transition. While the details of Chart 15.6 are missing, the message is the same - within key categories of sex, age and education, the self-employed are much more likely to have trajectories that are rated at Medium or Higher levels of the index of flexibility of the work-to-retirement transition, and the difference is usually in the order of 10 and above percentage points.
Percentage of persons at Medium or High levels of the index of flexibility in the work-to-retirement transition, for self-employed and salaried employees, for categories of sex, age, and education, cohort aged 45 to 69 in 1996, Canada, 1998 to 2001
Is this pattern seen even when we hold constant several variables that may simultaneously affect both class-of-worker status and level of the index of flexibility of the transition from work to retirement? This question is answered by building a model similar to that just discussed concerning speed of closure. However, in this case we are predicting probabilities for the different levels of the index of flexibility, based on combinations of values of several predictor variables. The variables used are presented in Appendices B and C, and they include self-employment in 1996 as a dummy variable. The reference category for this variable comprises primarily persons who were employees.
The model predicts that self-employment is associated with greater than average levels of flexibility in the work-to-retirement transition. This is shown in the positive parameter value for the key sub-model, the one, shown in column A of Table 15.4. This model predicts the probability that the index value will be at the Medium or High levels.
Adjusted patterns of association of self-employment with flexibility of the trajectory of work-to-retirment transition during 1998 to 2001, for two nested models, Canada
However, there is no surprise or insight provided by the result just reported; because it is widely discussed in the literature. What we contribute here is a measure of the degree to which self-employment provides the 'flexibility advantage' just cited. This information is provided by the odds ratio in Table 15.4. The odds ratio of 1.9 means that, according to the fitted model, the odds of having Medium or High levels on the index of flexibility among the self-employed are roughly twice as high as those of the reference population, which is dominated by wage and salary earners.
The third line in Table 15.4 shows what the percentage contribution of the self-employment dummy variable to the model's overall performance would be if the predictor variables were mutually independent. The notation "1/13" in the fourth line of column A indicates that the self-employment variable is nevertheless the largest contributor to the model's overall goodness of fit. Over 20% of the model's goodness of fit can be attributed to this variable (see the third figure in column A). Unfortunately, however, the predictors are not mutually independent, and as a result these results comprise a rough indicator only.
While the parameter estimate in the key sub-model (column A) is estimated with good reliability (statistically significant at the 5% level), that in column B is estimated with very low reliability - it is not statistically significant even at the 15% level. However, both columns of this table tell the same message - a positive partial association of self-employment with the probability of having Medium or High levels of the index of flexibility in the work-to-retirement transition. Also, the self-employed have much greater odds of having either of these levels than are the salaried - nearly 100% greater odds of being Medium or High in the whole sample, and 20% greater odds of being High in the sub-sample where all members have at least the Medium level. These odds ratios measure the estimated degrees of advantage that self-employment conveys with regard to the index of flexibility in the work-to-retirement transition.
Exposure to events that threaten to reduce standard of living in retirement
Like the index of flexibility in the transition from work to retirement, that which indicates exposure to events that increase risk of reduction in standard of living in retirement (also called "index of vulnerability" below) relies upon counting certain kinds of movement within a trajectory. Whereas the former index deals with movements that suggest voluntary changes undertaken by the SLID respondent, the latter index focuses on job loss or involuntary job change - negative events. This section examines the extent to which there are systematic differences in levels of the index of vulnerability between the self-employed and wage-and-salary earners.
For the properties of trajectories other than the speed of closure and flexibility, there is almost no literature to guide us in developing expectations as regards the directions of systematic differences between the two classes of worker just named. On one hand, it is evident that the self-employed are not subject to being 'fired by the boss'; but on the other hand they are constantly exposed to the well-known risks of business failure. We proceed, therefore, to display their differences with employees without any strong notion of what those differences will be like.
Wage and salary earners are more likely than the self-employed to have trajectories with Medium or High levels on the index of exposure to events that increase risk of reduction in standard of living in retirement (the "index of vulnerability"). More than 10% of the salaried who began their transitions during 1996 to 1997 and who were employed throughout the last quarter of 1997 had trajectories with Medium or High levels on this index. This is in contrast to less than 10% among their counterparts who were self-employed (See Chart 15.8). Although the margin of the difference between the two classes of worker is not large, the pattern just cited is confirmed in the two larger samples (described earlier) where parallel computations were done.
Index of exposure to events that increase risk of reduced standard of living in retirement, for self-employed and salaried employees, cohort aged 45 to 69 in 1996, Canada, 1998 to 2001
Although key sub-groups of sex, age and education repeat the pattern of variation just cited, the larger sample comprising all persons aged 55 or more (and who were employed throughout the last quarter of 1997) confirms this repetition for separate age groups only. Chart 15.8 shows that the differential between the self-employed and the salaried is especially marked at ages 55 or more.
To what extent is the pattern just cited repeated when we hold constant statistically several variables that may simultaneously explain both class-of-worker status and level of the index of exposure to events that increase risk of reduction in standard of living in retirement? Once again, to answer this question, we have created a model that predicts probabilities for the different levels of the index, based on combinations of values of several predictor variables. The model and variables used are presented in Appendices B and C, and they include self-employed in 1996 as a dummy variable. The reference category for this variable would very largely comprise persons who were employees in 1996 (other class-of-worker statuses are possible; but this one would be the largest share of the sub-sample's reference category.)
The data in Table 15.5 are consistent with the above mentioned pattern of association of self-employment with the index of vulnerability. Being self-employed tends to lower the probability of at least Medium values on the index, relative to that for the reference group (which is dominated by employees). This is shown by the negative parameter in the first column and the first line of Table 15.5 (the first column of this table is that for the key sub-model, which deals with the odds of having at least Medium levels relative to those of having a Low level of the index - see Appendix B for related details). Moreover, the odds for the self-employed are just over one-half as great as those for the reference group.
djusted patterns of association of self-employment with the index of vulnerability arising from the trajectory of work-to-retirement transition during 1998 to 2001, for two nested models
However, class-of-worker is not a major driver in the goodness of fit of the model. Among the 13 predictor variables, it ranks sixth in terms of contribution to the goodness of fit of the model. It is notable that none of the parameters in Table 15.5 is estimated with a high level of reliability, probably due to small sample size. However, inter-correlations among the predictor variables are also a factor.
Propensity to return to the labor market after departure
Whether the self-employed, who leave the labor market, during the stage of transition to retirement, are more likely to return than their counterparts who are wage earners is an issue that has notable policy implications. The final property of the trajectory to be considered here was designed to allow this issue to be addressed.
Return to the labor market may be a high probability for those who find their incomes in retirement otherwise unacceptable, and who have marketable skills. Since uncertainties in the flow of income in retirement may be unusually high among the self-employed, compared to the salaried, it is expected that they will be more likely to return to the labor market after leaving it.
This is precisely what Chart 15.9 shows for those who stated their transitions between 1996 and 1997, and who left the labor market at some time during those years. However, when the computation was done again for all persons aged 55 or more who left the labor market at some time during 1996 to 1997, the pattern is confirmed only if the scale values are grouped so that they have only two levels - zero and non-zero. Since sample size is a serious issue with this chart, the data for the larger sample are also shown. The larger sample indicates a much smaller difference between the two classes of worker than does the sample of "transiters".
Index of propensity to return to the labour market after leaving it during 1996 to 1997, for self-employed and salaried employees, Canada, 1998 to 2001
Unfortunately, the underlying sample size is too small to allow the display of charts that show breakdowns into categories of variables such as sex, age and education. So we used the multivariate analysis in search of results that may help in clarifying how the gap between the two classes of worker came about. While the results of this analysis confirm that being self-employed tends to boost the probability of return to the labor market, again data are not shown due to unduly small sample sizes. Also the multivariate analysis suggests that the difference between being self-employed and being salaried is a negligible factor in predicting the probability of return to the labor market after leaving it.
Opportunities for self-employment form an important dimension of efforts to achieve greater prevalence for gradual and flexible work-to-retirement transitions. In recent years, the gap between the average retirement age of the self-employed and that of wage and salary earners has widened in Canada. It was about two years in the late 1970s, and now it is close to four years - the average for the former has fluctuated around a flat trend while that of the latter declined steadily up to 2001. This difference suggests that a substantial rise in the trend toward more gradual retirement could be stimulated if there is a resumption of growth in the rate of self-employment. However, the labor supply decisions of the self-employed and their related patterns of transition to retirement are an area in which few research studies have been published.
As regards patterns in speed of closure, among those that started their work-to-retirement transitions between 1996 and 1997, and for key sub-groups as well as the sample as a whole, fast closure of trajectories is much more likely among the salaried than among the self-employed. At slower speeds of closure, the self-employed take the lead. In the bivariate analysis, we estimated a minimum 20-percentage-point gap between the two groups in terms of concentration at either end of the spectrum of speed of closure, where one limits observations to persons aged 55 to 69 in 1996. In the whole sample aged 45 to 69, when several related covariates are held constant statistically, the gap falls to near 10 percentage points at the extreme of fast closure and to five percentage points at the opposite extreme of non-closure.
Thus, while the general tendency of the self-employed to retire later is well known, we have provided for Canada, a measure of the degree of difference between the self-employed and the salaried in connection with the probabilities of closing trajectories soon after beginning the transition to retirement, and another concerning the probability of having unclosed trajectories four years after the transition began.
Flexibility in the transition to retirement is measured here in terms of changes in position (a trajectory is a sequence of positions occupied) that indirectly point to voluntary actions that reflect the use of available options or choices in how one's retirement process unfolds. We confirm a main finding of the existing literature - the self-employed seem to have an advantage in flexibility by comparison with the salaried. However, other researchers have rarely devised a measure of this flexibility and used it to gauge the size of the advantage enjoyed by the self-employed.
The present study offers this kind of information based on studying the trajectories of transition from work to retirement. While close to 75% of the self-employed had trajectories with Low levels on the index, the figure for employees is in the neighborhood of 85%. This wide margin of difference is confirmed in the larger sample that comprises all persons aged 55 or more in 1996. The pattern of differences between the two classes of workers recurs in the multivariate analysis.
The index of vulnerability focuses on job loss or involuntary job change. Wage and salary earners are more likely than the self-employed to have trajectories with Medium or High levels on the index of exposure to events that increase risk of reduction in standard of living in retirement (the "index of vulnerability"). The margin of the difference is substantial. However, key sub-groups of sex, age and education repeat the pattern of difference just cited, and the pattern is seen again in the context of multivariate analysis. However, class-of-worker is not a major driver in the goodness of fit of the model; because only 7% of the measure of model performance can be attributed to this variable.
Since uncertainties in the flow of income in retirement may be unusually high among the self-employed, compared to the salaried, it is expected that they will be more likely to return to the labor market after leaving it. This is precisely what our data show. However, the sub-sample of those that left the labor market at some time during 1996 and 1997 is too small to provide a reliable measure of the magnitude of the difference between the two classes of worker.
Due to an expected slowing of overall labor force growth along with an increasing percentage for its older members in future years, labor supply from older workers will be a matter of growing importance in connection with national economic output and competitiveness. In this context, the tendency for the self-employed to show distinctly greater flexibility in work-to-retirement transitions, compared to employees, suggests that special attention to policy issues concerning the environment for older entrepreneurs is warranted. ILO(1990:28) highlights the relevance of policy interest in this topic as follows: "The overall policy framework and institutions in an economy exert a decisive influence on the extent of self-employment and its prospects for growth."
Blau (1987:464) also finds evidence of the influence of the policy environment on the stimulation of growth in self-employment. A key aspect of this influence is whether the policy environment facilitates older workers' access to venture capital, according to Bruce, Holtz-Eakin and Quinn (2000).
Morris and Mallier (2003:15), point out that the evolving labor market environment extends opportunities to choose between alternative types of employment, including self-employment. In this connection, ILO (1990) cites an expansion of demand for labor inputs provided by self-employed persons among large organizations in some OECD countries, including Canada and the USA, and credits this expansion to new "information technology which gives the home-based self-employed worker immediate access to materials needed and allows outputs to be transmitted instantaneously". (See also Taylor 2002.)
Parker and Rougier (2004) suggest that schemes to promote self-employment among older workers might have two objectives: (1) increasing overall Labor Force Participation Rates (LFPRs), and (2) promoting a more competitive and entrepreneurial economy. Indirect support for this assertion concerning labor force participation rates is provided by Taylor (2002:6), who cites a key finding from research concerning the comparatively high LFPR found among older persons in Japan. A high proportion of self-employed or "family workers" is one factor that helps to explain their relatively high LFPR. A general review of the relevant field of policy concerns and of policy options pertaining to self-employment is provided by ILO (1990).
Blau, D. M. 1987. "A time-series analysis of self-employment in the United States." Journal of Political Economy. 95, 3.
Bruce, D., D. Holtz-Eakin and J. Quinn. 2000. Self-Employment and Labor Market Transitions at Older Ages. Working Paper 2000-13. Massacheuttus. Center for Retirement Research at Boston College.
Council of Europe, Parliamentary Assembly. 2003. Challenges of Social Policy in Europe's Ageing Society. Strasbourg, France.
Fuchs, V.R. 1982. "Self-employment and labour force participation of older males." Journal of Human Resources. 17: 339 to 357.
International Labour Organization (ILO). 1990. The Promotion of Self-Employment. Report VII. Geneva.
Karoly, L.A. and Julie Zissimopoulos. 2004. "Self-employment among older U.S. workers." Monthly Labor Review. 127, 7, July.
Morris, D. and T. Mallier. 2003. "Employment of older people in the European Union." Labour. 17, 4: 623 to 648.
Organisation for Economic Co-operation and Development (OECD). 2001. Ageing and Income: Financial Resources and Retirement in Nine OECD Countries. 12, 1: 1 to 180.
Organisation for Economic Co-operation and Development (OECD). 1995a. "The labour market and older workers." Social Policy Studies. 17: 1 to 316.
Parker, S.C. and J. Rougier. 2004. The Retirement Behaviour of the Self-Employed in Britain. Working Paper in Economics and Finance no. 04/08. Durham. University of Durham.
Policy Research Initiative (PRI). 2004a. "Population aging: From problem to opportunity." Horizons. 6, 2. Ottawa.
Policy Research Initiative (PRI). 2004b. Population Ageing and Life-Course Flexibility: The Pivotal Role of Later Retirement. Draft Discussion Paper. Ottawa.
Quinn, J.F. 1980. "Labor-force participation patterns of older self-employed worker." Social Security Bulletin. 43, 4: 17 to 28.
Quinn, J.F. and Michael Kozy. 1995. The Roles of Part-time Work and Self-employment in the Retirement Transition: A Preliminary View from the HRS. Paper presented at the HRS Early Results Workshop on Minority Perspectives. Ann Arbor. May 25 to 27.
Statistics Canada. 2004. Survey of Labour and Income Dynamics. Ottawa.
Sunter, D. 2001. "Demography and the labour market." Perspectives on Labour and Income. 13, 1: 28 to 39. Statistics Canada Catalogue no. 75-001-XPE.
Taylor, P. 2002. New Policies for Older Workers. University of Bristol. UK. The Policy Press in association with Joseph Rowntree Foundation.
United Nations (UN). 2002. Report on the Second World Assembly on Ageing. New York.
United Nations Economic Commission for Europe (UNECE). 2002. Berlin Ministerial Declaration: A Society for all Ages in the UNECE Region. 11 to 13 September.
Zissimopoulos, J. and L.A. Karoly. 2003. Self-Employment Trends and Patterns among Older U.S. Workers. Working Paper no. 136. Santa Monica, CA. RAND Labor and Population.
- The authors thank the peer reviewers for their contributions to improving earlier drafts of this chapter. The comments of Don Bruce, Kevin Cahill, and Hervé Gauthier were especially helpful. All opinions and errors herein are our sole responsibility.
- One reviewer suggested that there was no need to identify a group that started their transitions within a specified time period. However, without that step we will not be able to estimate the period covered by a person's work-to-retirement transition.
- The sample sizes shown in Table 15.1 are a basis for concern about statistical significance of differences cited or implied in the text here and below. This issue is discussed in Appendix C.
- The recomputation of the pattern among multiple subgroups has two purposes. For some of these subgroups there is substantial policy or scientific interest in examining their situations - men versus women, for example. The second purpose is to check for seemingly random appearances and disappearances of the pattern, since these would point to unreliable data.
- An impulse to attribute the pattern shown for women in the chart to an unduly small sample of self-employed women should be balanced against the consideration that women entrepreneurs may tend, much more than their spouses, to follow those spouses into retirement. Disrupting business careers to do so probably is less of a source of financial problems for self-employed than for salaried women whose retirment incomes are more subject to rules governing access to employee pensions.
- In the model used to generate the multivariate analysis results reported below, there was a direct association of the odds ratio in favor of unclosed trajectories with rising education. Also the Wald statistic for the parameter estimates for eductation is statistically significant at the 5% level. However, this was for a sample that included both employees and the self-employed.
- The concerns expressed above concerning small sub-sample sizes apply as well to the multivariate analysis. Special care is taken here in making claims about statistical significance. However, it is worth noting tht multivariate analyses with small samples are very common in research on inter-individual variations in the social sciences and especially in the health sciences.
- The data in the last two lines of Table 15.3 are rough approximations, since their computation in the SAS PROC LOGISTIC procedure assumes that the predictor variables are mutually independent.
- Date modified: