NCJ Number
58962
Journal
JOURNAL OF MULTIVARIATE BEHAVIORAL RESEARCH Volume: 12 Issue: 1 Dated: (JANUARY 1977) Pages: 111-131
Date Published
1977
Length
21 pages
Annotation
MULTIVARIATE PROCEDURES, BASED ON THE GENERAL LINEAR HYPOTHESIS, ARE OUTLINED FOR INFERRING TREATMENT EFFECTS IN QUASI-EXPERIMENTAL TIME-SERIES DESIGNS.
Abstract
ONE OF THE MOST FREQUENTLY ENCOUNTERED PROBLEMS IN SOCIAL SCIENCE RESEARCH IS THAT OF INFERRING TREATMENT OR INTERVENTION EFFECTS IN SETTINGS OVER WHICH THE RESEARCHER HAS LITTLE OR NO CONTROL. IN SUCH SITUATIONS, IT IS RECOMMENDED THAT QUASI-EXPERIMENTAL DESIGNS IN GENERAL, AND TIME-SERIES DESIGNS IN PARTICULAR, BE EMPLOYED IF REPEATED OBSERVATIONS ARE AVAILABLE. THE BASIC TIME-SERIES EXPERIMENT CONSISTS OF MAKING PERIODIC MEASUREMENTS ON A GROUP OF SUBJECTS, INTRODUCING AN EXPERIMENTAL CHANGE OR TREATMENT, AND MAKING PERIODIC OBSERVATIONS AND MEASUREMENTS ON THE GROUP AFTER TREATMENT. THE PROCEDURE FOR ESTABLISHING A TREATMENT EFFECT IS TO DETERMINE WHETHER OR NOT THE TREATMENT ACCOUNTS FOR ANY PRETREATMENT TO POSTTREATMENT ELEVATION IN THE TIME SERIES. A BASIC PROBLEM REMAINS, HOWEVER, WITH THE TIME-SERIES DESIGN CONCERNING THE STATISTICAL PROCEDURES THAT ARE CURRENTLY AVAILABLE FOR ANALYZING THE DATA. THE MULTIVARIATE PROCEDURES OUTLINED HEREIN TAKE INTO ACCOUNT THE DEPENDENT NATURE OF THE DATA OBTAINED IN TIME-SERIES EXPERIMENTS, AND REQUIRE COMPARING THE FITTED REGRESSION CURVES FOR THE PRETREATMENT AND POSTTREATMENT OBSERVATIONS, PROVIDING EXACT TESTS OF SIGNIFICANCE FOR THE VARIOUS HYPOTHESES OF INTEREST. STATISTICAL FORMULAS COMPRISE THE BULK OF THE DOCUMENT; REFERENCES ARE INCLUDED. (LWM)