Chapter 13 The probability of reaching the state of retirement - a longitudinal analysis of variations between men and women¹

View the most recent version.

The research question
Key concepts
Method
Research hypotheses
Probabilities of reaching retirement, and variations between men and women
Conclusion
Bibliography

The research question

Researchers generally agree that retirement is not so much an event as a process that may extend over a number of months, or even years. This process, which we shall label "the transition to retirement", ends when retiree status is permanently achieved. However, there is very little available information concerning the length of time this transition takes, or the kind of trajectory it follows.

Despite a labour market participation rate that has been rising strongly since the early 1970s, we still know very little about the dynamics of women's transition to retirement, and the factors that influence it. It has only quite recently become a subject for research. Until the mid-1990s, in fact, the vast majority of studies of retirement considered the situation of men only.

Nevertheless, growing concern about the situation of women during their retirement years has led a few researchers to conduct a more specific study of the transitions they go through when their working lives end. Some have suggested a more thorough study of the dynamics of the transition to retirement that would take their special characteristics into consideration (Atchley 1982, George, Fillenbaum and Palmore 1984, McBride 1988, McDaniel 1995, McDonald 1996, McDonald, Donahue and Moore 1997, McDonald 2003, Richardson 1999, Ware-Hargis 2001, Weaver 1994).

The same questions about transitions at the end of working life arise in most developed countries, but our study will be restricted to the Canadian context. The data are from the Survey of Labour and Income Dynamics (SLID) for individuals surveyed annually from 1996 to 2001. These individuals form the second panel of the survey, which covers six years of observation.

The general objective of our study will be to use longitudinal data to examine certain differences between men and women with respect to the transition to retirement. First, for those who have begun the transition to retirement, we shall analyse the differences relative to the probability of reaching retirement. That is to say, we shall focus on the probability of reaching retirement during the years following the year in which the respondent-man or woman-is deemed to have begun his or her transition to retirement. Where there are differences, we shall then try to see to what extent the variables related to work and family offer an explanation.

In the second part of the study, we shall describe the variations according to sex in the different types of trajectory followed in the transition to retirement. We chose to classify the trajectories observed by certain criteria, such as whether they were voluntary or not.

One of our main objectives is to determine how women behave in their transition to retirement. The question is highly pertinent since most of the studies on retirement of both sexes have noted specific differences between men and women. It emerges from these studies that retirement models, most of which still relate to the experience of men, should not be applied to the experience of women without taking women's special characteristics into account (Atchley 1982, George, Fillenbaum and Palmore 1984, McBride 1988, McDaniel 1995, McDonald 1996, Richardson 1999, Ware-Hargis 2001, Weaver 1994).

Key concepts

Retirement

In our study, we distinguish between the condition of retirement and the process of retiring: the transition to retirement. Here, retirement is considered a condition achieved when a person leaves the labour market for good, and receives retirement income (CPP/QPP, private pension, etc). Retirement is deemed to have been achieved when a person has spent at least a year out of the labour market, has received retirement income during that period, and does not return to the labour market before the survey ends. It is not possible to determine whether retirement is indeed permanent from the data we have. However, we believe that after at least a year spent outside the labour market, the probability of returning to work is much lower, and diminishes with age (Galarneau and Stratychuk 2001, Pyper and Giles 2002).

The transition to retirement

The literature offers few explicit definitions of the concept of the "transition to retirement". Most researchers have mainly considered a few particular aspects of the transition, such as partial retirement, the end of a career job, and so on. To that end, they have used certain indicators, including only one or two variables related to labour market activity, such as the drop in personal income,the drop in the number of hours worked, the end of a job held over many years etc. However, to determine the starting point of the transition to retirement, Stone and Nouroz (2004a) concluded it was necessary to consider a larger number of aspects within a single indicator (see Appendix A for more details).

The transition to retirement occupies a period of uncertain length during which workers take certain related decisions and steps and carries on linked activities. The transition may begin when a person begins to plan for retirement or when their spouse retires or when they reduce the hours they work or change jobs or, in a more involuntary way,or when they lose their job or fall ill.

Types of trajectory taken in the transition to retirement

The concept of the "trajectory" with reference to the transition to retirement, means a sequence of movements among a set of positions, or conditions, through which a person passes (voluntary or involuntary job change, job loss, reduction in employment income, and so on) before reaching retirement. Appendix A presents a detailed discussion.

In this chapter, we shall classify the various types of transition to retirement according to the voluntary or involuntary aspect of the movements mentioned above. The end of gainful participation in the labour market is a significant step for all workers, and whether it is voluntary or involuntary can have a great influence on their behaviour and their feelings about their situation. Note, however, that this is but one method of classification among many others (see Abott 1995, Han and Moen 1999).

We first divided these trajectories into two main groups: completed trajectories, and incomplete trajectories. A trajectory is completed when the respondent has achieved retirement before the end of the period of observation; otherwise, it is considered incomplete. Next, within these two groups, we determined three subcategories: voluntary trajectories, involuntary trajectories and unknown trajectories.

A voluntary trajectory means that the respondent did not undergo an involuntary job change or experience a period of unemployment between the year in which the transition to retirement began and the year in which he/she achieved retirement or the year in which the survey ended. An involuntary trajectory means that the respondent underwent an involuntary job change or experienced one or more periods of unemployment between the year in which their transition to retirement began and the year in which he /she achieved retirement or the year in which the survey ended.

"Unknown trajectories" means a residual category for respondents who do not fall into any of the above mentioned categories. For example, workers in transition to retirement might find themselves in this type of trajectory if at certain points in their transition to retirement they left the labour market. After further examination, we concluded this could involve people who were away from their job for a certain period (sick leave, sabbatical and so on) or it could involve a period between two work contracts, or unpaid workers within a family, like homemakers.

In all, we used six types of trajectories to classify those transiting to retirement, as Figure 13.1 shows.

Method

First, it must be determined if and when a SLID respondent began the process of transition to retirement. Our procedure for this is similar in many ways to that discussed in Appendix A. However, there are some slight differences. The objective was to develop an annual indicator (Appendix A discusses an indicator for the period 1996-1997) that could identify the events, or rather the combinations of events, likely to mark the beginning of the transition to retirement.

We recall briefly, that this determination procedure looks for various behaviour identifiers (receiving retirement income, having ceased working, caring for a loved one, having reduced their income) that often mark the beginning of the transition.

Using this measure, we analysed duration by means of a logistic regression in which we call the year in which the respondent began the transition "year zero". Note that in our study, the unit of time is a year, the maximum time a respondent can remain in transition is four years and the minimum is one. We then try to see whether this respondent has achieved retirement (by our definition, which requires a complete year out of the labour market, and the receipt of retirement income) before the survey ends. We thus obtain the dependent variable for our analysis of duration. This dependent variable is thus treated in one of two ways for each year spent in transition (value 1 means that the person achieved retirement at duration "i"; value 0 means that the person had not achieved retirement at duration "i").

We thus obtain a measure of the time spent in transition to retirement with which it is possible to make a direct comparison between men and women. Moreover, this dependent variable serves to estimate the probability of reaching retirement, based on certain social and economic attributes such as employment factors, and the family situation as it applies to men and women.

To obtain these probabilities using logistic regression, the dependent variable must be altered slightly: consider a group of respondents for whom "P" represents the probability of being retired, and "1 - P" as the probability of not being retired.

The natural logarithm of P/(1 - P), called "logit", constitutes the dependent variable of the model, the value of which fluctuates according to the predictive variables and their estimated coefficients.

To facilitate our interpretation of the results, we converted the estimated coefficients into predicted probabilities of reaching retirement. These probabilities were calculated as follows:

We chose to estimate two distinct models, one for women and another for men. We would thus be able to isolate and compare the differing effects of the same independent variable on the process of transition to retirement by sex.

The idea of separating women and men into two models is consistent with what is found in the literature, and with our own regression tests, which led us to conclude that two separate models better reflect the different occupational patterns of men and women. As a result of these different patterns, certain independent variables will have a different impact according to sex on the probability of reaching retirement.

In order to discuss the probability of reaching retirement for each year spent in transition, the choice of independent variables was influenced mainly by the relevant literature on the subject. These variables fall into two broad groups: those related to work and the individual, and those related to the family and the household.

The variables related to work and the individual include level of education, number of years of work experience, category of worker (union, non union, self-employed) and state of health. These are variables often found in studies on the retirement of men. However, now that women are participating in greater numbers in the labour market, these characteristics are also likely to affect their probability of reaching retirement.

The other group, those related to the family and the household, are often used in the literature about women. They are: marital status, the presence in the home of a parent or a child, the presence in the home of a person (other than the respondent) with a health problem and household income. Retirement of the spouse was also mentioned often as a variable relevant to the model, but as such information is not available in the SLID, we replaced it with the presence in the home of at least one person receiving retirement income² With this variable, we would track the effect of this additional household income on the probability of reaching retirement, rather than consider the specific status of the person receiving it (spouse or parent).

The family-related variables are less of a factor in the studies of retirement in men. However, since retirement is regarded less and less as a single event, it will be equally interesting to see what effect these variables will have on men's probability of reaching retirement.

Two control variables used in the study are: time spent in transition (in years), and age group; these will be defined in more detail below. Appendix D gives the definitions of the special variables we used to develop the model outlined above. For technical details of these variables, see Deschênes (2005), Appendix B.

Before we proceed to the next section, below is a summary table (Table 13.1) of the breakdown of our sample by sex, according to the independent variables applied at year zero, that is the year in which we estimate that the respondents' transition to retirement began.

Research hypotheses

We assumed that the probability of achieving retiree status would be higher in women than in men, for each year of observation following the year in which the transition began, and that this difference would be statistically significant. This hypothesis is based largely on the relevant observations of other researchers (see McDonald (1996), and her remarks in Chapter 10 of this book) and on studies published by Statistics Canada showing that on average, women reach retirement sooner than men (Statistics Canada 2003a).

Our second research hypothesis was that a higher proportion of men who experienced an involuntary transition did not complete their trajectories before the end of the observation period³ , compared with women who experienced the same type of transition. This assumption is based on the idea that men will stay in the labour market longer than women, even if they suffer a job loss or an involuntary job change (Galarneau and Stratychuk 2001, Pyper and Giles 2002).

Our third hypothesis concerns the impact of the independent variables in the model. On the whole, it is expected that the variables related to work and the person have more significant effects on men's probability of reaching retirement. According to the results found in the literature, it is considered that the number of years of work experience, personal income, the worker's situation in the labour market and state of health are important determinants in the probability of a man's reaching retirement (Blau 1994, Fleury 2003, George, Fillenbaum and Palmore 1984, Han and Moen 1999, Levine, Mitchelle and Phillips 1999, McBride 1988).

These variables are also expected to have impacts on women's probability of reaching retirement, but according to the literature on retirement for women, family and household characteristics are expected to be have a greater impact (Bess 1999, Dailey 1996, Hatch and Thompson 1992, King 1995, McBride 1988, McDonald 1996, McDonald, Donahue and Moore 1997, McDonald 2003, Richardson 1999, Vinick and Ekerdt 1989, Ware-Hargis 2001).

Differences by sex in the probability of reaching retirement

According to our results, almost 74% of women and 61% of men identified to be in transition to retirement achieved retiree status before the end of the observation period. The average age for men was 63 years, and 61 years for women.

Chart 13.1 summarizes the differences by sex of the probability of reaching retirement for each year spent in transition. The probabilities are the result of the logistic regressions run separately for men and women, where the independent variable is limited to time spent in transition.

Note that in our study, the maximum time during which a respondent could be observed in transition was four years, and the minimum was one.

First of all, we can see that the probability of reaching retirement is much higher during the first year spent in transition (about 53% for men and 60% for women). Second, the probability falls much more quickly for men, creating a gap between them and the women, who maintain substantially higher probabilities each year.

These curves suggest that the transition to retirement is relatively brief. Most of the men and women who begin the transition in a given year reach retirement the following year. The tendency is more marked in women than in men.

The extinction rate is in fact much quicker among women. For 1,000 women identified as having begun their transition to retirement, 596 are expected to reach retirement after only one year spent in transition, compared with 528 men. In the second and third years, the gap widens. Finally, after four years spent in transition, only 152 women out of the original 1,000 have not yet reached retirement, compared with 314 men, which is more than double.

The results for the probabilities of reaching retirement for each year spent in transition, according to age group, are presented by sex in Table 13.2. We used logistic regression in which the only independent variables are age group and time spent in transition.

For all age groups, women have clearly higher probabilities than men of reaching retirement for each year spent in transition. Note that for men, those 65 or older have quite different probabilities than the other age groups, while for women, there are narrower differences between age groups.

To better illustrate the point, we shall now look at the curves for probabilities by sex, according to time spent in transition and age group (Chart 13.2).

For men (Chart 13.2), during the first year of observation, the 55 to 59, 60 to 64, and 65 or older age groups are clearly distinguished from the youngest group, 50 to 54, with much higher probabilities of reaching retirement, while for the following years, only those 65 or older are distinguished by higher probabilities. After the first year of observation, the probabilities for the 55 to 59 and 60 to 64 groups fall, moving closer to the level of probabilities for the 50 to 54 group. Thus for men, we see a stronger association between advanced age and a high probability of reaching retirement.

For women (Chart 13.2), during the first year of observation, we see that the 50 to 54 group is distinguished from the others by having the lowest probability of reaching retirement. This probability is nevertheless 50% higher than that for men in the same age group (34% and 21%, respectively). Note also that the probabilities for the 55 to 59 and 60 to 64 groups are much higher than for men. For women, however, those 60 to 64 have probabilities that come quite close to those for women aged 65 or older, whereas for men, there is a clearer distinction between the two groups.

Differences by sex in the type of trajectory followed

According to our results, almost 52% of all trajectories were voluntary and were completed before the end of the observation period. Thus, among persons in transition to retirement, most reach it without experiencing an involuntary job change or a period of unemployment. Only 14% of trajectories were voluntary and incomplete. Some 8% of those involved completed an involuntary trajectory, compared with almost 13% who followed an involuntary trajectory that was not completed before the end of the observation period.

Table 13.3 summarizes the breakdown by sex of those concerned, according to the type of trajectory followed. Overall, about the same proportion of men and women in transition completed a voluntary trajectory (50% and 53%, respectively). The resemblance appears to end there, however.

With respect to voluntary trajectories not completed, note that almost 20% of men, compared with only 7% of women, took such a trajectory. Thus, it would seem that even if the trajectory followed is voluntary, proportionally more men remain longer in the labour market, compared with women.

Among women, nearly 20% followed an involuntary trajectory: 10% had completed trajectories and about 9% of the trajectories were not completed. Thus it would seem that in the case of women, a period of unemployment or an involuntary job change does not have an explicit impact on whether or not the trajectory is completed. Among men, nearly 22% followed this type of trajectory, but only 6% of men completed it, compared with 16% who did not. Overall, then, the same proportion of men and women followed an involuntary trajectory. However, men were almost three times more likely than women to not have reached retirement before the end of the observation period. This seemingly tends to confirm one of our initial hypotheses: that men are more likely to remain in the labour market, even after experiencing events that have disrupted their employment activities.

Probabilities of reaching retirement, and variations between men and women

In the context of the two models presented above (one for each sex), we sought to examine two sets of independent variables. The first consists of those related to work and the individual, such as highest level of education attained, number of years of work experience, category of worker and whether a health problem was reported. The second consists of those related to the family and the household, such as being a couple, the presence in the home of a parent or child, the presence in the home of a person (other than the respondent) reporting a health problem and receiving retirement income, and total household income. These are variables generally found in explanatory models for retirement among women.

Calculation of the probabilities in a logistic regression model requires that we hold constant the values for all predictive variables for a "reference population". Note that the reference population includes anyone with the following characteristics: being in the first year of transition, aged 50 to 54, without a secondary school diploma, being in the first quartile for number of years of work experience/personal income/household income, non-union, without health problems, not one of a couple, not reporting children or parents in the home, or persons with health problems, or other persons receiving retirement income.

The separate analyses by sex use a reference person with the same characteristics, the difference being that the variables divided into quartiles (years of work experience, personal and household income) do not have the same values for both sexes. These values are summarized in Table 13.4.

Table 13.5 shows the estimated probabilities obtained by logistic regression, including all the variables listed above.

The effectiveness of the model

In order to determine how well the model fits the observed data, we used Wald's chi-square goodness-of-fit test.⁴ This tests the zero hypothesis: that the coefficients are equal to zero for all the terms in the model. Thus, in order to assess whether a model is meaningful overall, we have to obtain a chi-square that rejects the zero hypothesis that all its coefficients are equal to zero, and this difference must be statistically significant at an acceptable threshold. In both cases, our models yielded probabilities below 0.01, as can be seen in the bottom lines of Table 13.5.

First, we noted that the chi-square for men was substantially higher than those for women, suggesting that the variables chosen are less successful in explaining what was observed in women, compared with what was observed in men.

This could thus imply that more associations are found between variables related to work and family among women, which is perhaps not surprising given that-as we saw in the literature-among women, work and the family seem to be more interdependent than was observed among men, and as a result, it may be more difficult to isolate their respective effects.

Variations between women and men in the behaviour of the most important independent variables

In the model, the probability of reaching retirement during the first year spent in transition for the reference population is about 11% for men, as against 29% for women. Thus, keeping the values for all other variables in the model constant, women were three times more likely than men to reach retirement during the first year.

However, these probabilities are substantially lower than those we obtained in the model in which the only independent variable was the year the transition began (see Chart 13.1). This means that what was attributed to the passage of time was in fact due to other factors included in the model.

Compared with men, women have probabilities that are often quite different for a number of independent variables. These results are reassuring as to our choice of separate regression models for men and women.

The presence in the home of another person receiving retirement income has the effect of increasing the probability of reaching retirement for both women and men. For women, the probability rises from 29% (in cases where no one else was receiving retirement income) to only 34% in cases where there was at least one person in the home receiving retirement income. For men, it goes from 11% to 20%, and the difference is statistically significant at the 5% threshold. Thus, whether or not there is at least one other person in the home receiving retirement income seems to influence men rather more than it does for women.

This result may seem strange at first glance, but it should be pointed out that 55% of the women selected in the sample based on TRANSCOR2 (respondents who had begun their transition to retirement) were already living with at least one other person receiving retirement income, compared with 32% of the men. It is thus not so surprising that this variable had a lesser effect on women's probability of reaching retirement. Moreover, this result is consistent in part with the study by Gower (1998) on the move into retirement of working couples. This study shows that couples tend to take retirement at least a year apart (two-thirds of the couples in the study) and that the woman will likely cease working before her spouse does. This might explain why this variable appears significant in the case of men, since they seem to leave the labour market soon after their spouses do.

Concerning the other independent variables that reflect family factors (such as the couple, the presence in the home of a parent or a child, the presence in the home of a dependant), their impact on the probability of being retired is also very different according to sex. However, these variables did not show statistically significant effects on the probability of reaching retirement for women and men.

As to the "category of worker" variable, note that being unionized sharply increases the probability of reaching retirement for both sexes. It rises to 60% for men and 66% for women. Thus, we find that being a union member greatly increases the probability of reaching retirement for both men and women, and the difference is statistically significant at the 1% threshold in both cases.

The indicator we used to identify respondents who had begun their transition to retirement (TRANSCOR2) has already effectively identified a number of major characteristics that lead workers to go into retirement. This would explain why we observe many respondents who reach retirement during the first year spent in transition. It is also one of the reasons why in general, the variables we selected have few statistically significant effects on the probability of reaching retirement, because the probability is first and foremost related to the fact of having been identified as being in transition. Nevertheless, it is interesting to note that in spite of this, our results show that for men, the time spent in transition, age group, number of years of work experience, being a union member, and the presence in the home of another person receiving retirement income lead to statistically significant variations in the probability of reaching retirement, whereas for women, we find that time spent in transition, age group and being a union member are the only factors that have a statistically significant effect.

Conclusion

We stated three main hypotheses.

• We assumed that the probability of reaching retirement would be higher for women for each year of observation following the year in which the transition began, and that the difference would be statistically significant.
• It was expected that a higher proportion of men who have experienced an involuntary transition would not complete the trajectory before the end of the observation period⁵, compared with women who have also experienced such a transition.
• It was expected that the independent variables related to work and the individual would have more significant effects on the probability of reaching retirement for men. It was felt that these variables would also have impacts on the probability of reaching retirement for women, but it was expected that the characteristics of the family and the household would be more important factors affecting transition toretirementt.

Generally, our results support the first hypothesis: that the probability of reaching retirement would be higher for women for each year of observation following the year in which the transition began.

As to the types of trajectory followed, we note that a similar proportion of men and women completed a voluntary trajectory (50.14% and 53.38%, respectively). We also noted that about the same proportion of men and women followed an involuntary trajectory (22% and 20%, respectively), but a larger proportion of men than women did not complete the trajectory before the end of the observation period. This tends to confirm one of our initial hypotheses: that men are more likely to remain in the labour market, even after experiencing events that may have disrupted their activities in the labour market.

Women were also 2.5 times more likely than men to follow an unknown trajectory in their transition to retirement. Thus, women reach retirement in a greater proportion than men do, but they also seem more likely than men to follow different types of trajectory in their transition to retirement. In fact, when men do reach retirement, they seem more likely to have taken one type of trajectory in particular: the voluntary type.

We expected the characteristics of the family and the household to be more of a determining factor for women. Support for this aspect of the third hypothesis is somewhat weak, since family and household characteristics are not the main factors affecting women's retirement: no variable is significant.

However, the results of the regression analysis show that retirement t for women is different from that of men. We note this in the comparison of the structure of estimated parameters of the models for both sexes, where we encounter major variations. We are thus of the view that in order to understand the dynamics of retirement for women better, we should not include both sexes in a single model. The development of two distinct models gives us an opportunity to perceive the differential impact of certain explanatory variables on women and men.

Bibliography

Abbot, A. 1995. "Sequence analysis." Annual Review of Sociology. 21: 93 to 113.

Atchley, R.C. 1982. "The process of retirement: comparing women and men." Pp. 153 to 168 in Women's Retirement: Policy Implications of Recent Research. M. Szinovacz (ed.). Beverly Hills, CA. Sage.

Bess, I. 1999. "Widows living alone." Canadian Social Trends. Summer: 2 to 6.

Blau, D.M. 1994. "Labor force dynamics of older men." Econometrica. 62, 1: 117 to 56.

Dailey, N. 1996. Women & Retirement: The Retirement Experience of "Baby Boom" Women. Doctoral Thesis. Washington, D.C. American University.

Deschênes, N. 2005. La transition vers la retraite : une analyse longitudinale des variations entre hommes et femmes : Annexe B. Unpublished master's thesis. Montreal, Québec. University of Montréal. Department of demography, Faculty of Arts and Sciences.

Fleury, C. 2003. « La santé, un facteur déterminant de la retraite anticipée? » Cahiers québécois de démographie. 32, 2, autumn: 201 to 221.

Galarneau, D. and L.M. Stratychuk. 2001. "After the lay off." Perspectives on Labour and Income. 2, 10, October: 5 to 15. Statistics Canada Catalogue no. 75-001-XPE.

George, L.K., G.G. Fillenbaum and E. Palmore. 1984. "Sex differences in the antecedents and consequences or retirement." Journal of Gerontology. 39, 3: 364 to 371.

Gower, D. 1998. "Retirement patterns of working couples." Perspectives on Labour and Income. 10, 3, autumn: 27 to 30. Statistics Canada Catalogue no. 75-001-XPE.

Han, S-K and P. Moen. 1999. "Clocking out: Temporal patterning of retirement." American Journal of Sociology. 105, 1: 191 to 236.

Hatch, L.R. and A. Thompson. 1992. "Families responsibilities and women's retirement." Pp. 99 to 113 in Families and Retirment. Maximiliane Szinovacz, David J. Ekerdt and Barbara H. Vinick (eds.). Beverly Hills, CA. Sage.

King, B.A.S. 1995. Predictors of Retirement Timing for Women. Doctoral Thesis. Lincoln, Nebraska. University of Nebraska.

Levine, P.B., O.S. Mitchelle and J.W. Phillips. 1999. Worklife Determinants of Retirement Income Differentials Between Men and Women. Working paper 7243. Cambridge, MA. National Bureau of Economic Research.

McBride , T.D. 1988. Women's Retirement Behavior: Implications for Future Policy. Washington, D.C. Research Associate, The Urban Institute.

McDaniel , S.A. 1995. "Work, retirement in later life." Pp. 75 to 91 in Rethinking retirement. E.M. Gee and G.M. Gutman (eds.). Vancouver, B.C. Gerontology Research Center.

McDonald, L. 2003. Gendered Retirement: The Welfare of Women and the "New" Retirement. Paper presented at the Symposium on New Issues in Retirement. Statistics Canada, Ottawa. September 5 and 6.

McDonald, L. 1996. Transitions Into Retirement: A Time for Retirement. Toronto. University of Toronto. Centre for Applied Social Research, Faculty of Social Work.

McDonald, L., P. Donahue and B. Moore. 1997. Widowhood and Retirement: Women on the Margin. Toronto. Centre for Applied Research.

Pyper, W. and P. Giles. 2002. "Approaching retirement." Perspectives on Labour and Income. 14, 4, September: 5 to 12.

Richardson, V.E. 1999. "Women and retirement." Journal of Women and Aging. 11, 2/3: 49 to 66.

Stone, L. O. and H. Nouroz. 2004a. A New Comprehensive Measure to Identify Whether a Person is Engaged in a Transition to Retirement. Unpublished paper. Statistics Canada. Ottawa.

Statistics Canada. 2003a. "Fact-sheet on retirement." Perspectives on Labour and Income. 4, 9. Catalogue no. 75-001-XPE. Ottawa.

Statistics Canada. 2003b. 2002 General Social Survey cycle 16: Aging and Social Support - Tables. Catalogue no. 89-583-XWE. Housing, Family and Social Statistics Division. Ottawa.

Statistics Canada. 2003c. Canada's Retirement Income Programs, 1991 to 2001. Ottawa.

Statistics Canada. 2003d. Bootmac8.SAS v13.15 : User's Guide. Ottawa.

Vinick, B.H. and D.J. Ekerdt. 1989. "Retirement and the family." Generations: Journal of the American Society on Aging. Spring: 53 to 56.

Ware-Hargis, M. 2001. Women, Work, and Retirement: Issues and Questions for Policy and Practice. Doctoral Thesis. Knoxville, Tennessee. University of Tennessee.

Weaver D.A. 1994. "The work and retirement decisions of older women: a literature review." Social Security Bulletin. 57, 1, Spring: 3 to 24.

---

Notes

1. This text is a translation of the original French version.
2. Note that we refer to income normally associated with retirement, including CPP/QPP benefits. The latter begin only at age 60, and benefits received earlier come from a disability pension or survivor's allowance. Hoever, given the characteristics of our sample, in which a majority - 80% - of respondents were one of a couple, an about 70% did not report the presence in the home of a parent or a child, we can assume that CPP/QPP benefits were received as a retirement pension in most cases. Moreover, income associated with retirement included private retirement pensions and RRSP withdrawals, which may begin before age 60.
3. A completed trajectory means that the respondent achieved retiree sttus - reached retirement - before the end of the observation period.
4. Note that the Wald chi-square goodness-of-fit test was the only one available to us with which to assess the effectiveness of the models using the bootstrap weights (Statistics Canada 2003d:223).
5. A completed trajectory means that the respondent reached retirement before the end of the observation period.