time of post-testing and (3) the association of treatment assignment with evaluable versus non-evaluable status. To test for intervention effects, for each of the 17 outcome variables signed change scores (post-test minus baseline) were generated, and analysis of covariance was performed to determine whether change scores of the intervention group were more favourable than those of the control group. In these analyses,thestudycohortandthebaselinescoreontheoutcome variable were used as covariates, along with demographic or cognitive variables significantly related to the total sample’s baseline to post-test change scores. One-tailed statistical tests, conducted at the 0.05 a level, were used to assess the hypotheses of positive treatment effects. Onesided tests were used because (1) the intervention could be justifiably linked to an expectation of positive health outcomes based on existing theory29, (2) a positive experimental effect had previously been demonstrated in the first Well Elderly trial6e8 and (3) the directional hypotheses were prespecified prior to the