NCJ Number
52441
Journal
Psychological Bulletin Volume: 84 Issue: 3 Dated: (MAY 1977) Pages: 489-502
Date Published
1977
Length
14 pages
Annotation
NOTING THAT TRADITIONAL STATISTICAL ANALYSES FOR TIME-SERIES EXPERIMENTS HAVE BEEN LIMITED BY THE USE OF A SINGLE CASE MODEL, THIS ARTICLE PROPOSES ALTERNATIVE PROCEDURES FOR USE WITH A LARGE SAMPLE SIZE.
Abstract
THE STATISTICAL ANALYSES UNIQUELY SUITABLE FOR CROSS-SECTIONAL TIME-SERIES EXPERIMENTS ARE DEVELOPED. THESE DESIGNS INVOLVE A SEQUENCE OF OBSERVATIONS OVER TIME, AS HAS BEEN CUSTOMARY, BUT WITH THE NUMBER OF SUBJECTS OR STATISTICAL CASES BEING RELATIVELY LARGE (THE BOX-JENKINS TRADITIONAL APPROACH OPERATES WITH A SINGLE CASE). THIS INCREASE IN THE NUMBER OF CASES PROVIDES FOR GREATER GENERALIZABILITY OF RESULTS. INTERRUPTED TIME SERIES, EQUIVALENT TIME SAMPLES, AND MULTIPLE TIME SERIES ARE THE ALTERNATIVE ANALYTICAL PROCEDURES DISCUSSED, WITH ALL CONSIDERED WITHIN A MULTIPLE REGRESSION FRAMEWORK. THE INTERRUPTED TIME SERIES IS CONSIDERED THE SIMPLEST DESIGN, AS IT INVOLVES A SINGLE INTERVENTION ACROSS ALL CROSS-SECTIONAL CASES. IN THIS DESIGN, THE CAUSAL INFERENCE IS BASED ON DISCOVERING AN ABRUPT CHANGE IN THE VALUES TAKEN BY THE DEPENDENT VARIABLE. WHILE THE GREATEST THREAT TO INTERNAL VALIDITY FOR THE INTERRUPTED TIME SERIES IS THE LACK OF A CONTROL FOR HISTORY, THE EQUIVALENT TIME SAMPLES DESIGN APPLIES THE INTERVENTION AT SEVERAL DISTINCT POINTS IN TIME, THUS RULING OUT HISTORY AS A THREAT TO VALIDITY INSOFAR AS IT WOULD REQUIRE AN UNUSUAL SEQUENCE OF COINCIDENCES FOR THE HISTORICAL EXPLANATION TO CARRY ANY WEIGHT. THE MULTIPLE TIME SERIES BY EMPLOYING A CONTROL GROUP ALSO AVOIDS HISTORICAL THREATS TO VALIDITY. EACH OF THE DESIGNS IS DISCUSSED AND ILLUSTRATED USING DUMMY VARIABLE REGRESSION EQUATIONS. A GENERALIZED LEAST SQUARES ESTIMATION PROCEDURE IS OUTLINED AS A SUITABLE ALTERNATIVE TO THE BOX AND JENKINS APPROACH, AND SOME OF THE SPECIAL ADVANTAGES OF THE DESIGNS ARE BRIEFLY DISCUSSED. (RCB)