Comparison of changes in a pretest-posttest design with
Likert scales
Christian Hennig, Daniel Müllensiefen and Jens Bargmann
June 2003
Abstract
Two methods for comparison of
the influence of a treatment on different discrete
variables are suggested, compared and applied to a dataset, which
concerns the influence of music on emotions and stems from a
questionnaire, where five emotions have been measured by ten questions
for each emotion on a five-point Likert scale. The question has been if
a certain piece of music induces anxiety to a significantly higher extent
than other emotions. The pretest values for
the emotions cannot be expected to be equally distributed,
and two methods to take this
into account are proposed.
The first one is a linear regression on the Likert mean
scores with pretest values as independent variables. The second one is a
t-test on new change scores, which are derived conditional on the pretest
values. It is shown that the second approach is more appropriate in the
present setup.
Download:
Compressed Postscript (109 Kb)
PDF (248 Kb).
Go back to the
Research Reports
from
Seminar für Statistik.