Sex as a source of heterogeneity in a mental health survey

We have presented a series of constructs which were developed to measure the relative mental health of the members of three small populations. These constructs were highly concrete, and each of the 331 respondents investigated could be assigned a numerical rating without the use of clinical judgements of trained personnel. These constructs proved to be intercorrelated to a large degree. It was demonstrated that the size of the correlation between constructs was not always an accurate reflection of relationships within the population, so that certain relationships which were highly significant for men proved to be insignificant for women, and vice versa. The heterogeneous nature of the population could be shown to lead to two kinds of errors when data was looked at only for the population as a whole. (1) Loss or diminution of relationships which were true for one group but not the other; (2) errors in the interpretation of the significance of a relationship. It was pointed out that these errors probably represent general difficulties in the analysis of data from psychiatric studies.