Examining testlet effects on the PIRLS 2006 assessment

We fit 10 PIRLS testlets with standard item response and testlet response theory model. The variances of testlet effects are examined and compared. Item parameters and ability parameters estimated from the two different models are compared. The results indicate that variances of testlet effects range from .168 to .489. Item parameters as well as ability parameter estimated from two approaches correlated highly. When local dependence is ignored errors of estimation in the difficulty parameter and cutoff parameter are negligible. However, estimates of both discriminate parameter and guessing parameter are biased. The standard error of ability parameter is underestimated when the local independence is incorrectly assumed.