The effects of test reliability on relationships between measures of life events and depression.

Abstract

The paper develops a measurement model for assessing the reliabilities of life event inventories. This method is applied to life event data collected in a longitudinal study of a sample of over 1100 New Zealand children and their families. The findings suggest that the life event inventory was of low-tomodest reliability, with reliability estimates ranging from 0.67 to 0.72. The reliability estimates were applied to the correlations between life event and depression measures collected at two time periods spaced 1 year apart to obtain estimates of the correlations between measures corrected for the effects of test reliability. On the basis of observed and corrected correlations, a non-recursive path model was fitted to the data to examine the effects of measurement error on the coefficients of the model. This analysis showed the presence of quite marked differences in the coefficients of the fitted model. In particular, the model solved assuming that the data were of perfect reliability suggested the presence of reciprocal causation between life event and depression measures, whereas when test reliability was taken into account, the model suggested a uni-directional pattern of causation, in which life event measures influenced depression measures. Nonetheless, both models showed that the predominant direction of association was from life events to depression rather than vice versa. Various theoretical implications of these findings are discussed.