Introduction

This research extends previous investigations on sample size requirements for calibrating items using the Partial Credit Model. This research will provide guidance on sample sizes necessary for accurate item parameter recovery when there is an internal disordering of contiguous category thresholds.

Objectives

This research will extend previous studies into adequate sample sizes for model parameter recovery for the PCM. Previous studies evaluated items have an internal monotonic scaling; however, it is potentially unrealistic to assume all items in will neatly fit this assumption. This research extends prior results to evaluate the effects of internal disordering of thresholds between contiguous categories and the effect of sample size on recovering the disordering.

Design/Methodology

A Monte Carlo simulation will be conducted for this investigation. The variable of interest will be overall sample size (N = 200 to 800 by 100). Factors to be manipulated:

• Percent of items with a single disordered step ranging from 5% to 20%, by 5%;
• Number of thresholds per item ranging from two to four by one, i.e. items with three to five score categories.

Results

Simulations are on-going at this point and results will be completed by the end of 2015. Preliminary results indicate that sample sizes above 250 provide adequate parameter recovery when items are not in the extremes of the ability distribution.

Conclusions

The practical significance of this paper will be to provide practitioners with information on sample size requirements as they begin planning related to field testing polytomous items to supplement their existing programs. Knowing a target sample size to adequately calibrate items will provide practitioners with knowledge for planning and costing their studies.