For item response theory likert scale's optimization an application example

Item Response Theory (IRT) has been traditionally used to analyze tests with dichotomous items. Nevertheless, in the past years, some research has shown the utility of adopting IRT for polythomous scales. This study aimed to verify the advantages of using IRT when compared to Classic Test Theory (CTT), regarding the necessary amount of answer categories to measure the construct, by using partial credits method of analysis. For that reason, the scale was recoded into a smaller one, and the number of categories required depended on its specific characteristics. The sample was composed by 1,317 people, who answered a Brazilian scale for Agreeableness assessment in the Five Factor Model of Personality, which used originally a 7-point Likert scale. Its three subscales and three items were used to exemplify the procedure of recoding. The analysis of item/scale fit, structure and reliability were compared in both situations. The results indicate that using optimized categories, within a smaller number, has allowed measuring the construct and has maintained the psychometric parameters of the scale. Also, in some subscales, internal consistency and misfit measures were better than the original ones.