STATISTICAL POWER IN NONRECURSIVE LINEAR-MODELS

In nonrecursive models, estimates of simultaneous relationships are often subject to high sampling variability. In this paper, we apply classical procedures for computing statistical power to the issue of sampling variability in estimates of reciprocal causal effects. Using a model of married women's attitudes regarding work and family size as an example, we show how the power to detect nonrecursive relationships depends on the model's parametric structure. Specifically, we show how the power of statistical tests depends on the strength of instrumental variables, the number of overidentifying restrictions, and the covariation among disturbances. We conclude by discussing the implications of our results for applications of nonrecursive models in the social sciences.