A regression solution to the problem of criterion score comparability

Abstract

When the criterion measure in a study is the accumulation of responses or behaviors for an individual over time, there may be a problem in obtaining comparable scores if all individuals cannot be observed for the same amount of time. Past solutions to this problem have included prorating the criterion score and deleting individuals lacking a portion of the criterion information. In the present paper the problem of criterion score comparability is viewed as a missing data problem, and a solution based upon linear regression is proposed. The regression solution was evaluated by comparing it to the prorating and deletion methods in a monte carlo analysis. The population data were an actual data set collected during a study of illness aboard U.S. Navy ships, and a series of samples were created by randomly deleting individuals’ records and portions of records. Results of the analyses suggested that the regression method is more efficient than the alternative methods. Application of the regression method and the results obtained to other research settings is discussed.