Abstract
A common and perverse problem facing those who evaluate human service programs
is the difficulty in getting random assignment of clients to control and experimental
groups, so that experimental or quasi-experimental research designs can be meaningfully applied. This article demonstrates the use of a technique (covariance adjustment) that statistically manipulates independent variables so as to give an approximation
of random assignment on those variables. In this article, we adapt the work of Alwin and
Sullivan (1976) to an actual data set from public alcohol treatment programs. We found
some very significant differences between the adjusted and nonadjusted treatment out comes, demonstrating the need for some type of pretreatment controls in the absence of
random assignment. The covariance adjustment technique and its assumptions are dis cussed leading to the conclusion that the technique is a very workable resolution of the
random assignment problem. We also demonstrate how the technique yields some valuable information generally not available when random assignment is used: namely, the
identification and weighting of certain selection biases as they relate to the dependent
variable.
