CONSTRUCTION OF A SURROGATE VARIABLE FOR IMPOTENCE IN THE MASSACHUSETTS MALE AGING STUDY

Uncollected data must be filled in after the fact in many epidemiologi cal studies. We compared several methods for constructing a surrogate variable for erectile potency, based on responses to related questions , in the Massachusetts Male Aging Study (MMAS). Quadratic discriminant analysis performed best among the procedures tested, achieving high r ates of correct classification in cross-validation of training data an d (according to posterior estimates) in classification of MMAS subject s. Two versions of the surrogate variable were constructed, one involv ing a discrete classification and the other involving a set of continu ous probabilities. The two versions were used to investigate medical c orrelates of impotence in MMAS and produced similar results, although the continuous version appeared to provide higher statistical power. W e conclude that construction of a surrogate variable is a viable and p otentially beneficial option in epidemiological investigations.