No abstract

In this paper we study a procedure for modelling the strength of a linear relationship between bivariate random variables in terms of a grouping variable. Regression analysis with indicator variables is the usual method of modelling the association between a dependent variable Y and an independent variable X over a grouping variable Z. Pearson's correlations give a measure of the strength of the associationat each level of the grouping variable. In this paper we show that weighted least squares can be used, with ***Fisher's z‐transformation, to fit a model in which the dependent variable is Pearson's correlation and the independent variable is the grouping variable. The fitted model provides a smooth function of the strength of association across levels of the grouping variable.

Using the population of St John's, Newfoundland, we did a constructive replication of previous studies testing the association between health practices and health status. A telephone questionnaire was applied to all adults in a probabilistic sample of households (3300 subjects, 85% response rate). Several health practices and preventive behaviours indexes were developed and tested. An additive index of six practices (breakfast, sleeping, drinking, smoking, weight and exercise) using the Alameda County definitions, and an additive score of five practices (excluding breakfast and with revised definitions for smoking, drinking and exercise) were analysed using log-linear models; there was association between the indexes and self-assessed health status when controlling for sex, age and education, with the exception of the six-practice index in males. The association between these two indexes and several variables and constructs of health status showed that the best relationships were with self-assessed health status and with a construct including variables measuring subjective health.

No abstract

No abstract