| |  Verfügbarkeit |
| Autoren | Rohwer, Götz |
|---|
| Titel | Making sense of missing answers in competence tests. |
|---|
| URL | https://www.neps-data.de/Portals/0/Working Papers/WP_XXX.pdf |
|---|
| Erscheinungsjahr | 2013 |
|---|
| Seitenzahl | 21 S. |
|---|
| Verlag | Bamberg: Otto-Friedrich-Univ. |
|---|
| Reihe | NEPS Working Papers. Band 30 |
|---|
| Dokumenttyp | Monographie; Discussion Paper / Working Paper / Konferenzbeitrag; online |
|---|
| Beigaben | Literaturangaben, Abbildungen, Tabellen |
|---|
| Sprache | englisch |
|---|
| Forschungsschwerpunkt | Bildungspanel (NEPS) |
|---|
| Schlagwörter | Bildungsforschung; Kompetenzmessung; Rasch-Modell; Antwortverhalten; Informationslücke; Schuljahr 05; Datenanalyse; Mathematische Kompetenz; Empirische Forschung; Quantitative Methode; Deutschland |
|---|
| Abstract | The paper discusses how to understand, and cope with, missing answers in competence tests. Starting point is the insight that missing answers in competence tests are not missing values in a normal sense (which hide sensible ‘true values’) but need an interpretation and evaluation. The paper then distinguishes two types of missing answers: A person is not able or motivated to produce a correct answer (type 1), or there was not enough time to produce a correct answer (type 2). It is argued that, if items are not of a multiple-choice type, missing answers of type 1 should be evaluated as wrong answers. It is shown that this does not contradict using a Rasch model for the construction of ability values and, in particular, does not lead to ‘biased estimates’. How to cope with missing answers of type 2 depends on which quantities one intends to estimate. Presupposing a standard Rasch model, missing answers of type 2 can be ignored when estimating item parameters. It is shown how this can be done with a conditional likelihood. With respect to ability values, it depends on whether one is interested in the ability to produce correct answers in the given time limit, or one intends to estimate ability values for situations without time restrictions. In the former case, missing answers of type 2 should be evaluated in the same way as missing answers of type 1. Given the latter intention, one needs to estimate a counterfactual score distribution for situations without time restrictions. The paper considers one possible estimation method that assumes observed numbers of correct and wrong responses being possibly censored observations, allowing one to use a two-dimensional Kaplan-Meier procedure. The estimated score distribution can then be used for multiple imputations. Finally, the paper illustrates the discussed methods with NEPS data on math competencies of 5th grade pupils. (DIPF/Orig.) |
|---|
| Förderkennzeichen | 01GJ0888 |
|---|
