Analysis of Ordinal Variables Part 3: Longitudinal Data
Karl G. Joreskog
December 12, 2001
Because of the mathematical content and length of this contribution to Karl's Corner, we are only making
it available as a PDF file. It can then be printed using Adobe's Acrobat Reader (version 4 and higher).
The zip file ordinal3.zip contains both the PDF
file and examples (input files and datasets).
The contents of this column are briefly described below.
Many papers have considered the specification of models incorporating causation and measurement errors in the
analysis of data from panel studies, and statistical models and methods for the analysis of longitudinal data.
The characteristic feature of a longitudinal design is that the same measurement instruments are used on the
same people at two or more occasions. The purpose of a longitudinal design or panel study is to assess the changes
that occur between the occasions, and to attribute these changes to certain background characteristics and events
existing or occurring before the first occasion and/or to various treatments and developments that occur after
the first occasion. Often, when the same variables are used repeatedly, there is a tendendency for the measurement
errors in these variables to correlate over time because of specific factors, memory or other retest effects. Hence
there is a need to consider models with correlated measurement errors.
The analysis of ordinal variables in longitudinal studies requires special techniques and procedures (available
in LISREL 8.51) which are different from those used with continuous longitudinal variables. This contribution by
Karl illustrates these techniques and procedures using the Political Action Panel Study for the USA which is a
two-wave panel study. The original USA sample consisted of 1719 cases interviewed in 1974. Five years later 933
of these cases were re-interviewed using the same six political efficacy items that were analyzed in part two of