compared in Table II (self-rated health) and Table III (HUI). Gender interaction terms
were included in a separate regression model of health for all elderly persons combined to
determine statistically significant gender differences in the regression coefficients, as
indicated in the last column (labeled Gender Gap) of these tables. Since age is a wellknown
determinant of health in later life, it is also included/controlled for in this analysis.
Age is a categorical variable, divided into 5-year intervals and recoded here into number
of years by taking the mid-point of each category (e.g., 65-69 = 67). Hence, age, along
with income, education, smoking, social support, and all stress-related variables, are
treated as continuous variables in the regression analysis. BMI and living arrangement
are treated as categorical data and therefore entered into the analysis as “dummy”
variables - the reference categories are BMI: overweight and living arrangement: living
alone.
Two methods are used to deal with missing cases. First, for HUI, income,
education, smoking, BMI, and social support index variables - all containing relatively
few missing cases – missing data are replaced by the mean of each variable. Second, the
NPHS allows proxy reporting for some variables. Since the stress-related variables are
applicable to non-proxy respondents only, there were some missing cases on the stress
variables; these cases were excluded from the analysis.
Limitation Certain limitations with the analysis and data used here must be
acknowledged. First, the NPHS household datafile does not cover persons residing in
institutions, most of whom are older women. Relatedly, the gender-bias in mortality (i.e.,
men compared to women at middle ages are more likely to suffer from life-threatening
chronic health conditions such as diabetes and heart disease, and therefore have a higher
7
probability of being deceased by older age) may produce a healthier population of elderly
men. Second, NPHS data, like most other health studies and data, are based on subjects’
responses to health-related questions. It is possible, therefore, that respondents'
perceptions of their health differ somewhat from diagnosed health problems among
Canadians. It is also possible that observed gender differences in health are to some
extent attributable to differential health-reporting behaviours of men and women. Third,
it is difficult to establish causality between social forces and health because of the nature
(i.e., cross-sectional) of the data used here. While it is possible that health status shapes
social resources to some extent, we presume, based on previous research, that social
forces have a causal influence on health (we therefore feel confident in describing these
social forces as predictors or determinants of health in this paper). Despite these
limitations, the NPHS is the only representative national dataset to consider the full
complement of social structural, behavioural, and psychosocial determinants of health,
and is the best available Canadian dataset for this particular study.
Results
Table I presents bivariate relationships between gender and socio-economic,
lifestyle, psychosocial, and health variables. Men have significantly higher levels of
income, education, smoking, marriage, and financial stress. Women have significantly
higher levels of insufficient weight, social support, and personal stress. Elderly men and
women assess their health in a similar manner, however, average HUI scores are
significantly different (0.84 for men and 0.81 for women, p