I am working on a project that is looking at the discriminative properties of 6, 5-point Likert items in a scale. I am using the ltm package in R to examine this within a item response theory (irt) paradigm, using a graded response model (grm).
For each item, I am given a "discrimination value" that corresponds with the slope of the ICC at each level given the ability level. For my six items, the values are (approximately): 2.1, 1.8, 1.3, 0.8, 2.7, 2.3.
How would I interpret these? Is there a cut-off for "good" discriminative properties? Are there values related to "weak", "medium", or "strong" levels of discrimination?
Thanks for your help!
Best Answer
I spent many more hours on this problem and eventually came up with guidelines presented in Baker (2001):
Baker, F. (2001). The basics of item response theory. ERIC Clearinghouse on Assessment and Evaluation, University of Maryland College Park, MD.
It seems that 'a' representing an item's discrimination, has values that are synonymous for both dichotomous and polytomous data. Discrimination is classified as 'none','low','moderate', 'high', and 'perfect'. The ranges are as follows: 0 is 'none' 0.01-0.34 is 'very low'; 0.35-0.64 is 'low'; 0.65-1.34 is 'moderate'; 1.35-1.69 is 'high'; => 1.7 is 'very high' + infinity is 'perfect'