Abstract:
The purposes of this research were 1) to compare bootstrapped statistics of equating mixed-format test of nonequivalent groups with anchor test equating design among scoring methods 2) to compare root mean squared difference (RMSD) of equating mixed-format test of nonequivalent groups with anchor test equating design between Levine method and CLE (Chained linear equating) method. The sample were 1,607 students obtained by multi-stage random sampling from Matayomsuksa 4 student of Office of Bangkok Educational Service Area Zone I. They were used for data collecting with 1) two parallel mixed - format tests of the Physics, the work and energy for Matayomsuksa 4. There were new form and reference form. New form consisted 16 multiple-choice items and 8 construct-response items, their reliability values were .634 and .824. Reference form consisted 16 multiple-choice items and 8 construct-response items, with reliability value of .602 and .872. Those form had 8 multiple-choice items and 4 construct-response items in anchor test. 2) rubric scoring used for constructed-response items in Physics test, new and reference forms, had inter-rater reliability with Pearsons product moment correlation were .96 and .97 at the .05 level of significance. 3) two raters, who had consistency between two raters of two tests with Pearsons product moment correlation were .86 - .99 and .84 1.00 at the .05 level of significance. Three scoring methods were 1) constructed-response anchor items scoring methods with no trend scoring method 2) constructed-response anchor items scoring methods with full anchor test trend scoring method 3) constructed-response anchor items scoring methods with half anchor test trend scoring method. Equating score with linear equating method and analyzing bootstrapped statistics employed program R and analyzing Root Mean Squared Difference employed Microsoft excel. The results of the study revealed that 1) constructed-response anchor items scoring methods with half anchor test trend scoring method with CLE (Chained linear equating) method resulted in smallest Root Mean Squared Error (RMSE), Equating error and Bias were .802, .251 and .840. 2) For this study, CLE (Chained linear equating) method had Root Mean Squared Difference (RMSD) from scoring methods of 1, 2 and 3 method were 1.421, 1.481 and 1.481, which lower than those of equating with Levine method were 1.717, 1.804 and 1.793.