Effects of Various Simulation Conditions on Latent-Trait Estimates: A Simulation Study

Abstract

The aim of this simulation study, determine the relationship between true latent scores and estimated latent scores by including various control variables and different statistical models. The study also aimed to compare the statistical models and determine the effects of different distribution types, response formats and sample sizes on latent score estimations. 108 different data bases, comprised of three different distribution types (positively skewed, normal, negatively skewed), three response formats (three-, five-and seven-level likert) and four different sample sizes (100, 250, 500, 1000) were used in the present study. Results show that, distribution types and response formats, in almost all simulations, have significant effect on determination coefficients. When the general performance of the models are evaluated, it can be said that MR and GRM display a better performance than the other models. Particularly in situations when the distribution is either negatively or positively skewed and when the sample size is small, these models display a rather good performance.

Psychometric properties of the Cultural Intelligence Scale based on item response theory

“…Fourth, the number of response categories (5) used in this study did not seem to affect the accuracy of ability or item information estimates due to the good fit of the model, the acceptable item discrimination values, and the large sample size. This is supported by the findings of several studies (Hauck Filho et al, 2014;Koğar, 2018;Maydeu-Olivares et al, 2009) that investigated the effect of number of response categories (3, 5, and 7) in rating scales, response distribution and difficulty level (moderate, too easy or too difficult for the sample assessed), and sample size on accuracy of latent score estimations within IRT-GRM framework. They reported that ability estimates were good and stable under different conditions, and that the ability estimates became with higher accuracy when the sample size increased (500 and more).…”