Here - Tilburg University
Here - Tilburg University
Here - Tilburg University
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Author and presenter<br />
Gorter, Rosalie; VUmc, Dept. of Epidemiology and Biostatistics, EMGO+<br />
institute of health and care research<br />
Authors<br />
Rosalie Gorter; Martijn W. Heymans; Jos W.R. Twisk; VUmc, Dept. of<br />
Epidemiology and Biostatistics, Emgo+ Institute of Health and Care Research.<br />
Michiel R. de Boer, VU <strong>University</strong>, Faculty of Earth and Life Sciences, Institute for<br />
Health Sciences, Dept.of Methodology and Applied Biostatistics.<br />
Rien van der Leeden, Leiden <strong>University</strong>, Faculty of Social Sciences, Institute<br />
Psychology, Methodology & Statistics.<br />
Title<br />
Comparing the performance of software packages in estimating the parameters<br />
of multilevel IRT models for longitudinal data<br />
Background<br />
Many questionnaires used in patient research consist of items with a likert<br />
answering scale. An example is the increasing utilization of quality of life<br />
questionnaires in epidemiological and medical research. When the answers on<br />
such questionnaire are used as outcome variable, usually a score is attached to<br />
the answering categories and these scores are than added in order to obtain a<br />
total score for the construct. A theoretically more appropriate way of analyzing<br />
these data is by using an IRT model that estimates item and person parameters.<br />
An adjacent category (ordinal) logit model can be used to estimate the<br />
probability of a person to choose a specific category given his or her level of the<br />
latent variable theta. In addition to using IRT specific software packages for such<br />
analysis, the models can also be formulated as hierarchical models and analyzed<br />
with general software packages. An important advantage of this reformulation is<br />
that levels can be added for analyzing longitudinal or otherwise clustered data.<br />
There are several different software packages for fitting ordinal logit models<br />
which are capable of estimating the parameters for this type of longitudinal data.<br />
However, these packages use different estimation methods which may lead to<br />
different estimates depending on the combination of parameter specific<br />
characteristics of the data such as sample size, item and person characteristics.