Chapter 17 The relationship between career instability and health condition in older workers: A longitudinal analysis¹

View the most recent version.

Introduction
Literature review
Methodology
Analytic approach used
Results
Discussion
Bibliography

Introduction

This study addresses the association between career instability and long-term health conditions in older workers. The literature has identified that the nature of the transition from employment to retirement has been changing dramatically in industrialized societies such as Canada in the past few decades, with a decreasing proportion of one's working life being spent in stable career progression. Many individuals who 'retire' from long-service career jobs continue to seek paid employment in sporadic (pick-up) work before completely exiting the labor force. Increasingly common are transitions between full employment status, unemployment and withdrawal/expulsion from the full-time labor force prior to permanent retirement: a situation referred to as career instability. The relationship of such career instability in later work life to worker's health was felt to be not wholly elucidated, prompting this study.

Literature review

Career instability in later life

Commencing in the mid-1970s, there has been a significant number of workers aged 45 to 64 experiencing career instability transitioning between such job status situations as early retirement, sporadic work, unemployment and re-employment (Reimers and Honig 1989, Mutchler et al. 1997).

As an illustration of the early retirement situation, the 1991 Survey on Aging and Independence revealed that the actual Canadian average age of retirement had fallen to 62 years for men and 59 for women (McDonald 1994), notwithstanding the conventional retirement age having remained at 65. From the same 1991 survey, Schellenberg (1994a) found that a significant proportion (25% with an increasing trend) of early retirement was involuntary. Supporting this proposition, the Survey of Persons Not in the Labor Force (November 1992) found that of those persons who had retired earlier than they had expected, 13% attributed their retirement to their companies going out of business (McDonald 1994).

Schellenberg (1994b) reported that the unemployment rate among older Canadian workers (aged 45-64) increased during the recessions of the early 1980s and 1990s and that despite the 1986 to1989 economic recovery, this unemployment rate failed to return to pre-recession levels. Displaced older workers remained jobless for a longer duration: an average of 32 weeks for those aged 45 and over compared to 17 weeks for younger job seekers (age 15 to 24). As well, more older than younger job seekers were found to have become discouraged and to have dropped out of the job market, leading to a possible understatement of the true unemployment rate for older workers (Schellenberg 1994b).

Mutchler et al. (1997) using longitudinal data from the 1984 Survey of Income and Program Participation, found that one-quarter of a sample of men aged 55 to 74 experienced 'blurred transitions' or career instability (repeated exits, entrances, or unemployed spells).

In summary, the literature has indicated that a significant number of older workers experience career instability involving unemployment, repeated job loss, prolonged periods of unemployment, early labor force exit, etc. Career instability can be expected to impact health through increased levels of stress from dealing with career transitions, the associated uncertainty and the necessity of adapting to a new life situation and/or by creating problematic economic situations. Career instability could thus act as an exogenous stressor influencing endogenous job and life satisfaction variables, posited to then influence health.

Career instability and health

While there has been little research focussing on the health effects of career instability in later working life, limited research on such instability early in working life has shown that instability can be linked to subsequent degraded health.

Iwatsubo, Derriennic and Cassou (1991) conducted a cross-sectional study of 627 subjects living in the Paris area and found that career instability (measured by the number of changes of company and branch of activity during working life) was significantly associated with health impairment after retirement. Pavalko, Elder and Clipp (1993) used career patterns up to the year 1960 of American men born between 1900 and 1920 to predict mortality after 1960. They found that men who experienced several unrelated job changes in early working life had a mortality risk that was 57% higher than for those who did not have such experiences. This relationship is not adequately accounted for by the possible explanatory factors measured in the study: physical health, alcoholism, anxiety and depression measured in 1960, or variability in occupational status in 1959.

The above two studies explored career instability over the entire or early working life span without specifically segregating the later life component. They only sampled men and measured career instability only in terms of job-to-job changes. Consequently, it is suggested that more research be done looking particularly at the health effects of career instability in late working life measured in a more sophisticated way that takes into account frequent changes between employed and unemployed status.

Recently, Marshall, Clarke and Ballantyne (2001) investigated the relationship between health and instability in the move to permanent retirement in a sample of early retirees from a major Canadian telecommunications company. Instability was found to be associated with adverse health effects and with variability by gender and type of health measure. A similar study, using data from the 1994 General Social Survey (GSS) - Cycle 9, found that many aspects of career instability (being not employed/early retired, the experience of job loss/job interruption, etc.) were significantly related to poor self-rated health, after controlling for age, education, body mass index and activity limitation (He, Colantonio and Marshall 2003). It should be noted that because these studies used cross-sectional data, they were unable to adequately address the issue of direction of causation between career instability and health.

With unemployment being a major aspect of career instability, some useful inferences regarding career instability and health can be drawn from longitudinal studies of health effects of unemployment. A majority of such studies conducted since the 1980s have concluded that unemployment is associated with deterioration in health (Martikainen and Valkonen 1996). Mathers and Schofield's (1998) review of recent studies in different countries has suggested a strong, positive association between unemployment and many adverse health outcomes. From the many longitudinal studies using a range of designs, they concluded that there is reasonably good evidence that unemployment itself is detrimental to health, as shown by its impact on such health outcomes as increased mortality rates, increased physical and mental ill-health and greater use of health services.

Iversen et al. (1987), in their 10-year follow-up of a large cohort of unemployed Danish men and women, reported that both gender groups exhibited elevated mortality rates compared to a control sample of employed individuals. Moser et al. (1987), using longitudinal data covering a 1% sample of the population of England and Wales, found that the standardized mortality ratio in 1971 to 1981 of the 5,861 unemployed men was 136. The ratio was increased for all age groups and was over 200 in those aged 35 to 44. Many other studies have revealed similar findings (Morris, Cook and Shaper 1994).

In view of the fact that career instability is a more comprehensive condition than unemployment per se, the health impact of related career transitions among older workers bears further study.

The objective of this study was to exploit the longitudinal features of SLID data to assess the extent to which different career instability patterns lead to changes in health condition over time. More specifically, it was proposed to test the research hypothesis that career instability would be associated with long-term health conditions and indeed would be one cause of such conditions.

Methodology

Data file

This study employed a secondary analysis of a large national longitudinal data file, SLID, over four waves (1994, 1995, 1996 and 1997). This data file provided rich data on labor force transitions. While the health measurement in SLID is quite crude, it was felt that the ability to explore different labor force transition patterns compensated for this limitation.

The SLID was designed to capture changes in the economic well-being of individuals and families over time and to identify determinants of their well-being. Individuals originally selected for the survey were interviewed once or twice per year for six years to collect information on their labor market experiences, income and family circumstances. The target population of SLID was all persons living in Canada, excluding those living in the Yukon and Northwest Territories, residents of institutions, persons living on reserves, and full-time members of the Canadian Armed Forces living in barracks.

For this study, a sample consisting of 8,567 subjects, aged 45 to 64 in 1994 was selected from the SLID data file. Approximately 80% of subjects from 1994 to 1997 were successfully tracked throughout the three-year period under study.

Measures used in the analysis

The predictive variable, career instability, was comprehensively measured by the number of jobless spells, number of weeks unemployed and number of weeks not in the labor force. All the career instability variables were defined on the basis of number of weeks during the current year.

Being unemployed was defined as not being employed during the week and looking for work at some time during the week or being absent from a job due to layoff. A greater number of weeks unemployed was thought to reflect a higher degree of career instability.

A jobless spell was defined as a period in which the subject had no attachment to any employer during that week, though having an attachment at the start and end of the spell. The duration of the spell was derived from the start and end dates of jobs held during the year, that is, they corresponded to the 'gaps' between employment dates. Seasonal layoff or any other type of layoff where there was expectation of return to work with the employer was treated as unemployed. The jobless state is therefore distinct from being unemployed: jobless spells cannot be linked in any way to specific jobs. A higher number of jobless spells was regarded as a higher degree of career instability.

Not in the labor force was defined as being neither employed nor unemployed during the week. The number of weeks not in the labor force could be obtained by subtracting from the full year (52 weeks) the sum of the number of weeks employed, the number of weeks unemployed and the number of weeks spent in jobless spells. Being employed was defined as having worked at any time during the week or being absent from a job for a reason other than layoff.

The outcome variable, health, was measured by long-term (health) condition. This variable indicated whether or not a person suffered from any long-term physical or mental condition or any other health problem during the reference year.

Measurement of career instability and long-term condition in this study were based on self-reporting, raising a concern as to validity and reliability. Several studies investigating links between work environment exposure and disease have shown self-reporting of employment status, jobless spells etc. to be valid (Bourbonnais, Meyer and Theriault 1988, Brisson et al. 1991) and reliable (Brower and Attfield 1998, Rosenberg 1993). However, no literature was found on validity and reliability of self-reported long-term condition.

Five socio-demographic factors with potential to complicate any possible associations were identified: age, sex, marital status, income and education. With respect to income, the adjusted family income (per capita income) was derived by dividing family income by family size. There is evidence showing that income is strongly associated with health (Kingson 1982).

Education has also been shown to be strongly associated with health (20, 21). Educational level begins to affect health early in life and continues to affect health in adulthood through its effect on health habits, the use of health care and exposure to health-affecting life situations (Crimmins, Reynolds and Saito 1998).

It is well established that with increasing age, workers are more prone to negative health conditions, such as cardiovascular disease, pulmonary problems and deterioration of the musculo/skeletal system.

In the broad perspective, women's work situations are different from men's. Women have tended to be concentrated in industries where pension benefits are rare and are overly represented in part-time employment. Moreover, women have tended to experience an inordinate degree of interruption in their paid working lives due primarily to family responsibilities (Marshall and Clarke 1998).

Some research suggests that marital status is significantly associated with health, with married individuals having a lower risk of poor health (Smith, Shelley and Dennerstein 1994). Literature indicates that family (marriage), in the role of a nonemployment source of social support, is among the most important of social networks in assisting older workers cope with stress (Betancourt 1991). Marital status is a simple measure of social support and for this reason was included in the analysis as a control variable.

Analytic approach used

The Cox proportional hazards model was used to investigate whether or not career instability increases the likelihood of developing a long-term condition over time. This model is well suited to investigating the time dependent relationship for the probability of an event using multiple explanatory covariates (Kalbfleisch and Prentice 1980). In this analysis, we model newly developed a long-term condition over a three-year investigation period. Subjects who already had a long-term condition by the beginning of the study were excluded. Time was measured in years, 1, 2, or 3. Though this discrete measurement may be considered crude, it is the only data available. The event was defined as first occurrence of a long-term condition during the study period. The covariates of career instability change at multiple and regularly-spaced points in time. Therefore, our data sets were discrete-time data with time-dependent covariates, making Cox proportional regression model the best choice for the analysis (Allison 1995).

The possibility exists that a long-term condition affects career rather than vice versa: if someone had a long-term condition near the beginning of a particular year, the probability of having career instability is likely to increase during that year. In order to reduce ambiguity in the causal ordering, lagging of the predictive variable values was created. Besides predicting a long-term condition in a given year by career instability in that same year, career instability in the preceding year was also used. Another possibility is that the hazard of a long-term condition may depend on the cumulative career instability experience with a one-year lag rather than the single experience of career instability in the preceding year. In order to address this possibility, a cumulative variable for career instability was created to predict long-term conditions (Allison 1995).

Since 1,269 subjects who initially had a long-term condition were excluded from the survival analysis, 'biased sampling' is a limitation for this approach. To compensate for this limitation, we used the Generalized Estimating Equation (GEE) model, which is able to include all subjects in the model, as an alternative analysis.

The GEE model is an extension of Generalized Linear Model (GLM). A GLM consists of the following components:

1. Link function: a monotonic differentiable link function describes how the expected value of response variable is related to the linear predictor variables.
2. Variance function: the response variables are independent for different subjects and have a probability distribution from an exponential family. This implies that the variance of the response depends on the mean through a variance function.

In addition to specifying a link function and a variance function as in GLM, the GEE model requires a working correlation matrix for the observations from each subject for repeated measurement data when the responses are discrete and correlated.

In this study, the categorical response variable and the explanatory variables were measured at year 1994, 1995, 1996 and 1997. This scenario of longitudinal data with repeated categorical response variable fits the GEE model very well. The SAS GENMOD procedure was used to implement the GEE method. The GEE approach provides consistent estimators of the regression coefficients and of their variance under weak assumptions about the actual correlation among observations of a single subject (Diggle, Liang and Zegar 1994).

Separate and distinct Cox proportional hazards models and GEE models (including terms for the career instability variable, and for age, sex, marital status, education and income control variables) were fitted for each career instability variable in turn.

Results

In the sample under study, men and women were almost evenly distributed (49.7% male vs. 50.3% female). The age distribution was skewed toward the younger groups (32.4% between 45 and 49 years old compared to 20.5% between 60 and 64). In terms of education, it was found that approximately 12% had achieved bachelor's level and above, 39.6% had some post-secondary education, 14.5% had obtained a high school diploma, 15.5% had some high school and 16.9% had some or no elementary school (Table 17.1). In terms of income, the median annual after-tax family, personal and per capita incomes were $38,503, $17,757 and $14,927, respectively. In view of the skewed distribution of income, log transformation of the per capita income was used in later survival analysis.

In 1994, 14.8% (1,269) of subjects had a long-term health condition and in 1997 a total of 18.9% (1,623) of subjects had a long-term condition. During the years 1995, 1996 and 1997, there were 374, 355 and 276 subjects, respectively, who developed a new long-term condition, a total of 1,005 individuals during the three years of the study (Table 17.2). The 1997 total (1,623) was less than the 1994 to 1997 sum, 1,269 + 1,005, indicating that a certain percentage of subjects who developed a long-term condition in a particular year had recovered sufficiently to subsequently rejoin the group without a long-term condition.

As shown in Table 17.2, over the study period, approximately 40% of subjects experienced at least one jobless spell, the majority of them experiencing only a single such spell. The incidence of jobless spells increased from 38.4% in 1994 to 45% in 1997. With respect to the number of weeks unemployed, more than 15% were unemployed for more than a week in each year of the study, with no significant trend over the study period. However, an increasing trend for number of weeks not in the labor force was observed.

Bivariate analysis showed several factors to be associated with a long-term condition. These factors were: being female, being not married/common law, being older, having lower education, having a lower income, having a greater number of jobless spells and having a greater number of weeks unemployed or not in the labor force.

Using the number of jobless spells to predict the development of a long-term condition among those without such condition at baseline (1994), the Cox regression model showed that the hazard for those with a high number of jobless spells was significantly higher than for those with a lower number, even after controlling for age, sex, marital status, income and education. The hazard of a long-term condition was increased 21.3% for each increase of one jobless period (RR = 1.213, p = 0.0005). This model also showed that age, education and marital status were significantly associated with a long-term condition, while sex and income was not (Table 17.3).

The GEE, examining the determinants of change in a long-term condition over time, showed the number of jobless spells to be a significant predictor of a long-term condition after controlling for baseline long-term condition, age, sex, marital status, income and education. The GEE model showed there were significant effects of the number of jobless spells, baseline long-term condition, sex and education. The adjusted odds of having a long-term condition were EXP (0.2094) = 1.233 times higher for each increase of jobless spell (p = 0.0001). The adjusted odds ratios for baseline long-term conditions were approximately equal to EXP (3.1184) = 22.6. This implies that the odds of having a long-term condition were 22.6 times higher for those with a long-term condition at baseline than for those without a long-term condition (Table 17.6). The GEE and survival models yielded relatively consistent results in terms of a relationship of number of jobless spells and long-term conditions.

Lagging the predictive variable by one year was used to reduce causal ordering ambiguity. It was found that the hazard for those with a high number of jobless spells in the preceding year was significantly higher than for those with a lower number. The hazard of a long-term condition was increased 22% for each increase of jobless spell in the preceding year (p = 0.0006). Similar results were obtained for the cumulative count of the number of jobless spells. However, after controlling for age, sex, marital status, income and education, the number of jobless spells and cumulative number of jobless spells in the preceding year were no longer significantly related to a long-term condition (p = 0.2811, p = 0.3502, respectively) (Table 17.3).

Similar patterns of association were found for the other career instability variables: number of weeks not in the labor force (Table 17.4) and number of weeks unemployed (Table 17.5).

For example, it was found that the hazard of a long-term condition was increased 0.99% for each increase of one week of not in the labor force (p = 0.0001). The GEE model showed there were significant effects of the number of weeks not in the labor force, after controlling for baseline long-term condition, age, sex, marital status, income and education. The adjusted odds of having a long-term condition were EXP (0.0053) = 1.005 times higher for each increase of one week not in the labor force (p = 0.0001).

The hazard of long-term conditions was increased 0.76% for each week increase of not being in the labor force in the preceding year (p = 0.0001). Similar results were obtained for the cumulative count of number of weeks not in the labor force. However, after controlling for age, sex, marital status, income and education, the number of weeks not in the labor force and the cumulative number of weeks not in the labor force in the preceding year were no longer significantly related to long-term condition (p = 0.2186, p = 0.2081, respectively).

Discussion

This study has confirmed that many older Canadian workers have experienced career instability, as previously observed for Canada by He, Colantonio and Marshall (2003) and Marshall, Clarke and Ballantyne (2001); and for the U.S. by other investigators (Reimers and Honig 1989, Ruhm 1990, Mutchler et al. 1997). Furthermore, over the study period, there was a trend of increasing career instability, evidenced by increased frequency of subjects with at least one jobless spell. There was also an increased frequency of subjects having a higher number of weeks not in the labor force, which may be due to some subjects in this cohort (aged 45-64) having attained the conventional retirement age at or before the end of the study and having withdrawn from the labor force permanently.

Both the survival and the GEE models showed that career instability was significantly associated with a long-term condition, even after controlling for age, sex, marital status, income and education. These findings are consistent with the results of three studies based on different data files, conducted by Marshall, Clarke and Ballantyne (2001), He, Colantonio and Marshall (2003) and Mutchler et al. (1999).

If career instability was lagged by one year to predict long-term conditions, without controlling for previously noted socio-demographic factors, it was found that those who experienced career instability in the preceding year were significantly more likely to have a long-term condition. However, after controlling for the socio-demographic factors, the relationship became not significant. These results suggest that the health impact of career instability may be partially determined by other characteristics, such as marital status, education, income, etc. From this study, there is insufficient evidence to conclude that career instability can cause a long-term condition. It is more likely that poor health causes career instability or that poor health and career instability are closely intertwined: at any given time, both health status and work status being embedded within a series of interlocking transitions.

Although our hypothesized causal relationship between career instability and long-term conditions was not substantiated by this data, the possibility that career instability causes long-term conditions cannot be dismissed due to the following limitations of this study. First, the only available health and well-being measure for the four waves of the SLID is a binary variable, long-term condition, which may not be sensitive enough to detect change in health status over time, therefore yielding results of no significance. Second, as a long-term condition tends to be chronic, long-term conditions caused by career instability is probably not an instant effect: the health effect of career instability may take time to develop and become manifest. In this study career instability was used only in the preceding one year to predict a long-term condition in a given year because of the fact that only four years of data was available. Third, the SLID sample is a subset of the sample used for the Labor Force Survey (LFS). The LFS sample is a probability sample drawn from an area frame and is based on a stratified multi-stage design. In short, SLID is a complex survey and design effects are present. Design effects represent the extent to which the sampling design has deviated from simple random sampling. While the sample weights provided by Statistics Canada were employed in the analysis and popular statistical software packages such as SAS can handle weighted data and can produce accurate estimates, the standard errors associated with these estimates are typically underestimated due to design effects.

In future, further waves of SLID data will become available for longitudinal analysis. Moreover, starting in 1997, the SLID has collected more variables on health and well-being, providing a richer data source for future study of the relationship between career instability and health.

Nevertheless, this research documents an important and very relevant trend of career instability and its relationship with long-term disability in the industrial world.

Bibliography

Allison, P.D. 1995. Survival Analysis using the SAS System: A Practical Guide. Cary, NC. SAS Institute Incorporation.

Betancourt, R.L. 1991. Retirement and Men's Physical and Mental Health. New York. Garland.

Bourbonnais, R., F. Meyer and G. Theriault. 1988. "Validity of self-reported work history." British Journal of Industrial Medicine. 45: 29 to 32.

Brisson, C. et al. 1991. "Validity of occupational histories obtained by interview with female workers." American Journal of Industrial Medicine. 19: 523 to 530.

Brower, P.S. and M.D. Attfield. 1998. "Reliability of reported occupational history information for US coal miners." American Journal of Epidemiology. 148, 9: 920 to 926.

Crimmins, E.M., S.L. Reynolds and Y. Saito. 1998. "Trends in health and ability to work, working-age population." Journals of Gerontology Series B, Psychological Sciences and Social Sciences. S31 to S39.

Diggle, P.J., K.Y. Liang and S.L. Zegar. 1994. Analysis of Longitudinal Data. Oxford. Cambridge University Press.

He, Y., A. Colantonio and V.W. Marshall. 2003. "Later life career disruption and health: An analysis of the General Social Survey." Canadian Journal on Aging. 22, 1: 45 to 58.

Iverson, L. et al. 1987. "Unemployment and mortality in Denmark, 1970-80." British Medical Journal (Clinical Research Ed.). 295, 6603: 879-884.

Iwatsubo, Y., F. Derriennic and B. Cassou. 1991. "Relationship between job mobility during working life and health status after retirement: a cross sectional study of 627 subjects living in the Paris area." British Journal of Industrial Medicine. 48: 721 to 728.

Jylha, M. et al. 1998. "Is self-rated health comparable across cultures and genders." Journal of Gerontology: Social Sciences. 53B, 3: S144 to S152.

Kalbfleisch J.D. and R.L. Prentice. 1980. The Statistical Analysis of Failure Time Data. New York. John Wiley and Sons.

Kingson, E.R. 1982. "The health of very early retirees." Social Security Bulletin. 45, 9: 3 to 9.

Kivinen, P. et al. 1998. "Self-rated health, physician-rated health and associated factors among elderly men: The finish cohorts of the seven countries study." Age and Aging. 27: 41 to 47.

Marshall, V.W. and P.J. Clarke. 1998. "Facilitating the transition from employment to retirement." In Canada Health Action: Building on the Legacy. 2: 171 211. Papers commissioned by the National Forum on Health, Determinants of Health: Adults and Seniors. Sainte-Foy, Quebec. Editions MultiMondes.

Marshall, V.W., P.J. Clarke and P.J. Ballantyne. 2001. "Instability in the retirement transition: Effects on health and well-being in a Canadian study." Research on Aging. 23, 4: 379 to 409.

Martikainen, P.T. and T. Valkonen. 1996. "Excess mortality of unemployed men and women during a period of rapidly increasing unemployment." Lancet. 348, 9032: 909 to 912.

Mathers, C.D. and D.J. Scholfield. 1998. "The health consequences of unemployment: The evidence." Medical Journal of Australia. 168: 178 to 182.

McDonald, L. 1994. "Retirement revisited: A secondary data analysis." Issues of an Aging Workforce: A Study to Inform Human Resources Policy Development. Working Paper. Toronto. University of Toronto. Centre for Studies of Aging.

Morris, J.K., D.G. Cook and A.G. Shaper. 1994. "Loss of employment and mortality." British Medical Journal. 308, 6937: 1135 to 1139.

Moser, K.A. et al. 1987. "Unemployment and mortality: comparison of the 1971 and 1981 longitudinal study census samples." British Medical Journal (Clinical Research Ed.). 294, 6564: 86 to 90.

Mutchler, J.E. et al. 1999. "Work transitions and health in later life." Journal of Gerontology: Social Sciences. 54B, 5: S525 to S261.

Mutchler, J.E. et al. 1997. "Pathways to labor force exit: Work transitions and work instability." Journal of Gerontology: Social Sciences. 52B, 1: S13 to S26.

Pavalko, E.K., G.H. Elder and E.C. Clipp. 1993. "Worklives and longevity: Insights from a life course perspective." Journal of Health and Social Behavior. 34: 363 to 380.

Reimers, C. and M. Honig. 1989. "The retirement process in the United States: Mobility among full-time work, partial retirement, and full retirement." Pp. 115 to 127 in Redefining the Process of Retirement: An International Perspective. W. Schmahl (ed.). Berlin. Springer-Verlag.

Rosenberg, C.R. 1993. "An analysis of the reliability of self-reported work histories from a cohort of workers exposed to polychlorinated biphenyls." British Journal of Industrial Medicine. 50: 822 to 826.

Ruhm, C.J. 1990. "Career jobs, bridge employment, and retirement." Pp. 92 to 107 in Bridges to Retirement: Older Workers in a Changing Labor Market. P. Doeringer (ed.). Ithaca, NY. ILR Press.

Schellenberg, G. 1994a. The Road to Retirement: Demographic and Economic Changes in the 90's. Ottawa. Centre for International Statistics. Canadian Council on Social Development.

Schellenberg, G. 1994b. Older Workers in the Canadian Labor Market. Ottawa. Centre for International Statistics. One Voice: The Canadian Seniors Network.

Smith, A.M., J.M. Shelley and L. Dennerstein. 1994. "Self-rated health: Biological continuum or social discontinuity?" Social Science and Medicine. 39, 1: 77 to 83.

---

Note

1. This research was supported by a grant from the National Health Research Development Program, Health Canada, and by a Natural Scieces and Engineering Research Council Scholarship granted to Yaohua H. He