The use of proxies in health surveys. Substantive and policy implications.

It has traditionally been assumed that obtaining health and illness behavior data by proxy on household members who are not present during the actual interview presents no significant threats to the internal validity of subsequent analyses. Using data on the 78,439 adults interviewed in person or by proxy (about 37% of the total adult sample) as part of the 1978 Health Interview Survey, the authors examine whether the use of proxy respondents alters the results of empirical assessments of the behavioral model of health services utilization or the policy implications that can be derived from it. Two important findings emerge from these analyses. On the one hand, failing to consider explicitly the possibility of a proxy effect (i.e., including a proxy variable in the analyses) does not alter the effect parameters estimated for the behavioral model. On the other hand, failing to consider explicitly the effect of using proxies appears to underestimate slightly physician and hospital contact rates (by 4% and 2%, respectively), as well as the volume of physician utilization. The substantive and policy implications of these findings are discussed, as are two alternative explanations that suggest that proxy-respondents simply use fewer health services either because they are "too busy" or because they are in slightly better health.