NCJ Number
69236
Date Published
1979
Length
72 pages
Annotation
A BROAD INTRODUCTION TO VARIOUS STATISTICAL TECHNIQUES AVAILABLE FOR ANALYSIS OF PANEL DATA, THIS PAPER DISCUSSES ANALYZING OBSERVATIONS TAKEN ON MANY CASES, INCLUDING BOTH DISCRETE AND CONTINUOUS VARIABLES.
Abstract
THE FOCUS OF THIS PAPER IS ANALYSIS, AS OPPOSED TO PROBLEMS OF PANEL SAMPLING OR DESIGN. THE ANALYSIS OF DISCRETE VARIABLES, EITHER DICHOTOMIES OR POLYTOMIES, IS DETAILED. THE FIRST CLASS OF PROBABILITY MODELS DISCUSSED IS MARKOV CHAINS FOR THE ANALYSIS OF PANEL DATA. STEADY STATE, SECOND-ORDER PROCESS, MARKOV MODELS FOR HETERGENEOUS POPULATIONS, AND THE MOVER-STAYER MODEL ARE DESCRIBED. SINCE THE MARKOV APPROACH IS LIMITED (MARKOV CHAINS DO NOT EXPLAIN WHY INDIVIDUALS CHANGE OVER TIME AND HAVE A GENERAL INABILITY TO DEAL WITH MEASUREMENT ERROR), TWO MODIFICATIONS OF THE MARKOV SCHEME HAVE BEEN DEVELOPED. WIGGINS MODELS HAVE THE ADVANTAGE OF BEING RELATIVELY FLEXIBLE, ENABLING THE RESEARCHER TO INCORPORTATE HIS OR HER A PRIORI ASSUMPTIONS ABOUT CHANGE AT BOTH MANIFEST AND LATENT LEVELS. THE COLEMAN MODEL IS SIMILAR TO THE WIGGINS MODEL; HOWEVER, IT DIFFERS BOTH IN TERMS OF THE THEORIES OF ATTITUDE CHANGE AND IN THE PARAMETERS TO BE ESTIMATED. THE COLEMAN APPROACH POSSESSES A NUMBER OF DISTINCT ADVANTAGES OVER BOTH SIMPLE MARKOV CHAIN MODELS AND THE WIGGINS VARIANT. A FINAL ALTERNATIVE APPROACH DESCRIBED IS THE GOODMAN LOG-LINEAR MODEL. AFTER THE DISCUSSION OF MODELS USED FOR THE ANALYSIS OF DISCRETE VARIABLES, THE PAPER EXTENDS THE DISCUSSION TO CONTINUOUS (OR INTERVAL LEVEL) VARIABLES. THIS SECTION PROVIDES AN EXAMINATION OF REGRESSION EFFECTS, AUTOCORRELATION, AND MEASUREMENT ERROR AND INTRODUCES TECHNIQUES BY THESE, WILEY AND WILEY, AND JORESKOG FOR ANALYSIS OF PANEL DATA. ALL THE TECHNIQUES ARE BASED ON CAUSAL MODELING AND REGRESSION ANALYSIS. A FAMILIARITY WITH MULTIPLE REGRESSION ANALYSIS IS ASSUMED. TABLES, GRAPHS, SCHEMATIC REPRESENTATION, FOOTNOTES AND APPROXIMATELY 70 REFERENCES ARE INCLUDED.