This study of person
fit in attitude surveys was undertaken in order to investigate
the influence of the inclusion of misfitting persons on item
parameter estimates in analyses using the Partial Credit
extension of the Rasch measurement model. It was hypothesised
that the inclusion of misfitting persons in data sets used for
the calibration of attitude survey instruments might
compromise the measurement properties of those instruments.
Using both actual and simulated data sets, the inclusion of
misfitting cases was found to reduce item variance. Several
characteristics of both item and person samples were found to
influence the proportion of cases identified as misfitting.
These characteristics must be considered before removing cases
that, according to customary practice, appear to misfit. The
residual based misfit indicators that are commonly reported in
Rasch analyses, the weighted and unweighted mean squares,
appear not to have the generality over all instruments nor the
precision required to make clear decisions on the retention or
elimination of cases from samples, and there is a need to seek
better misfit indicators.
Misfit, Attitude Surveys, Rasch