Missing Data in Psychiatric Research: A Solution

Presents a method of dealing with statistical analyses that are biased because of missing data. For instance, treatment outcome at the end of some Ss' episodes are often not observed, while basic descriptive variables are observed for all Ss, making the sample censored. The recommended technique yields consistent estimates when a sample is censored and is also applicable to problems of S attrition, nonresponse to questionnaires, and natural experiments, especially nonrandom assignment of patients to treatment modalities. (July 1983)