Pairwise Rasch model item parameter recovery under sparse data conditions

In social science research the occurrence of missing values is a prevalent issue. In addition to that the use of multi-level response formats often results in low cell frequencies, especially under the condition of higher proportions of missing values and small datasets. This in turn may lead to problems for parameter identification in probabilistic item response models, applying the Conditional Maximum Likelihood (CML) or Marginal Maximum Likelihood (MML) method. While listwise deletion or pairwise deletion usually results in a significant reduction of the sample size and may thus impair efficiency, most methods of data imputation require preceding assumptions about the data loss mechanism, which often can not be checked. Consequently an alternative non-iterative approach to item parameter recovery is introduced. This will be outlined using a minimal example. This approach, named PAIR, is based on conditional pairwise item category comparisons. In the present study data from the NEO-PI-R inventory were analyzed using three different algorithms for item parameter recovery (PAIR, CML, MML) each under a continuing rise of missing data. The resulting item parameter estimates are compared with regard to their accuracy of the point estimates as well as the size of their respective standard errors. The results are discussed comparatively under the condition of the increasing proportion of missing values by simulation. The results suggest that even at higher levels of missing data the PAIR approach leads to stable item parameter estimates.     «

In social science research the occurrence of missing values is a prevalent issue. In addition to that the use of multi-level response formats often results in low cell frequencies, especially under the condition of higher proportions of missing values and small datasets. This in turn may lead to problems for parameter identification in probabilistic item response models, applying the Conditional Maximum Likelihood (CML) or Marginal Maximum Likelihood (MML) method. While listwise deletion or pair...     »