| Summary: | 碩士 === 國立臺中教育大學 === 教育測驗統計研究所 === 99 === Abstract
This research is to estimate the root mean square error (RMSE) of ability parameters and item parameters by BILOG-MG and SCORIGHT when the ability distribution is normal or bimodal in balanced incomplete block design(BIB). The factors taken into consideration include the following: the sample sizes (5460 and 7566), the number of examinees (27 and 45), the proportion of the testlet items (1/3 and 2/3) and the variances of the testlet effects (0, 0.5 and 1).
The results of this research show that:
1.The RMSE decreases as the sample size or the number of examinees increase when the ability distribution is normal.
2.The RMSE decreases as the sample size decrease or the number of examinees increase when the ability distribution is bimodal.
3.When the ability distribution is normal, compare with BILOG-MG and SCORIGHT simulation, the RMSE of the ability, the item discrimination and difficulty parameters are smaller under SCORIGHT simulation, but the RMSE of the pseudo-chance parameter is random.
4.When the ability distribution is bimodal, compare with BILOG-MG and SCORIGHT simulation, the RMSE of the pseudo-chance and the item difficulty parameters are smaller under BILOG-MG simulation, but the RMSE of the ability parameters and the item discrimination are random.
Key word：BIB、testlet
|
|---|
