NCJ Number
56271
Journal
EVALUATION QUARTERLY Volume: 1 Issue: 3 Dated: (AUGUST 1977) Pages: 399-420
Date Published
1977
Length
22 pages
Annotation
A GENERAL AND POTENTIALLY USEFUL ALTERNATIVE TO COVARIANCE ANALYSIS (ANCOVA) FOR CONSIDERING PREEXISTING DIFFERENCES IN A NONEQUIVALENT CONTROL GROUP SITUATION IS PRESENTED.
Abstract
TRADITIONAL APPROACHES TO DISTRIBUTE THE EFFECTS OF CONFOUNDING FACTORS IN A NONEQUIVALENT CONTROL GROUP SITUATION, ONE OF THE MAIN PROBLEMS FACED BY SOCIAL PROGRAM EVALUATORS, CALCULATE A PARTIAL CORRELATION OR REGRESSION COEFFICIENT CONTROLLING FOR ONE OR MORE 'COVARIATES'. BUT THESE METHODS YIELD BIASED ESTIMATES OF EFFECT. IN THE ALTERNATIVE APPROACH PRESENTED HERE, HYPOTHESIZED FACTORS DO NOT HAVE TO BE ORTHOGONAL, ESTIMATES FOR THE UNRELIABILITY AND IRRELEVANCE OF COVARIATES (IN TERMS OF MEASUREMENT ERROR VARIANCE) NEED NOT BE AVAILABLE, AND IT IS NOT NECESSARY TO ESTIMATE FACTOR SCORES. THE MODEL FALLS UNDER THE REALM OF LINEAR STRUCTURAL RELATIONSHIPS INVOLVING MULTIPLE INDICATORS OF UNMEASURED VARIABLES. AS SUCH, A COMPUTER PROGRAM IS AVAILABLE TO CALCULATE MAXIMUM LIKELIHOOD OF ESTIMATES AND CHI-SQUARE GOODNESS OF FIT STATISTICS. THERE ARE INFINITE WAYS TO FORMULATE A CAUSAL MODEL, AND IT IS IMPORTANT FOR RESEARCHERS TO FORMULATE MANY MODELS SO THEY CAN DETERMINE WHETHER CONCLUSIONS WILL DIFFER IF A DIFFERENT SET OF ASSUMPTIONS IS ACCEPTED. IT IS ALSO IMPORTANT TO FOLLOW CERTAIN GENERAL GUIDELINES IN BUILDING MODELS WHEN RESEARCHERS HAVE LIMITED INFORMATION ABOUT THE CAUSAL PROCESS. THE GENERAL APPROACH IS ILLUSTRATED USING DATA FROM A HEAD START EVALUATION IN WHICH IT WAS CLEAR THAT THE CONTROL GROUP WAS FROM A HIGHER SOCIAL CLASS THAN THE EXPERIMENTAL GROUP. RESULTS FROM THE HEAD START EVALUATION ARE SUMMARIZED, AND SOME ALTERNATIVES TO ANCOVA SUGGESTED IN THE LITERATURE ARE BRIEFLY DESCRIBED. EQUATIONS RELEVANT TO THE ANALYSIS PROCEDURES AND REFERENCES ARE INCLUDED.