This study proposed a Scale Value to represent the interval in Likert scales and then investigate its used by conducted am empirical research. Three subscale of Big Five Scale were revised as 4, 5, 7-point rating scales and administrated to college students. When the data were collected, descriptive analysis and CTT based reliability coefficient (Cronbach’s alpha) of the subscale scores, as well as the IRT approach for category and scale value, were conducted to find out the influence by number of options. For the result from descriptive statistics, there were not different among three level of number of options, though the adjusted scale score and SD of 7-point scale were slightly smaller than 4 and 5-point scale. For the result from reliability estimates, as we discussed last week, the Cronbach’s alpha did not harms by the discretization and thus it was expect that the Cronbach’s alpha would not yield significant difference between 4, 5, 7-point scale. While the scale values based on item response model, there are interesting results among the three different numbers of options. The scale value of the 7-point scale deviated more and drifted a lot in a reversed subscale which having important implications to practices.
Here are some questions or comments about the manipulation in the article. First, did the adjust procedures has impact on the comparison among the three subscales with different number of options? Second, based on our discussion last week, the discretization did influence the IRT-based reliability coefficient (pi) statistically. The author employed the CTT-based coefficient and avoided the discussion on the change of reliability. Third, the explanation of the reversed subscale seems still open to further investigation.