I am working on my thesis and I have a survey that has true/false and multiple choice questions. From what I understand, KR20 is used for dichotomous variables to estimate reliability coefficient. Also, if multiple choice questions are not equally weighted, I can use KR20. Am I correct in doing KR20 for both trues/false and multiple choice items? Any suggestions and input is greatly appreciated. If you can suggest references that would be great. Thank you!
Dear Tatyana,
You're not wrong about the use of the KR-20 for dichotomous questions. However, the content in question has been widely used only for this purpose. The Cronbach's alpha has been used to assess the reliability of the instrument particularly in the case of an instrument with questions of Likert's type. In your case we are facing a multiple instrument choice questions has no equivalent weights suggest the strategy Item Response Theory (IRT), to the detriment of psychometric classical analyzes of reliability.
Regards,
Rodrigo Maia
Dear Tatyana,
KR20 is good in the case of dichotomized variables to estimate internal consistency of test (it is one of the figures of test reliability), but definitely not to multiple choice items. I agree with Rodrigo Maia, IRT could be very helpful for you. But when I were you I would try to avoid different form of items in the same test. Maybe in one (your) survey exactly are a few instruments. Then each should have separate assessment.
Greetings,
Slawomir Pasikowski
PS. :) There are an exception in the case of KR20 and position of multiple choice, as I think. Each option such a position could be converted into a dummy variable (0, 1). Then you would have as many items as options, and it give you chance to the use of KR20. Of course, this require changing the all multiple choice questions for the dummy variable and consider which position of the survey could be joined without risk of sens lose. But if in this domain will not be obstacles then the assessment of the internal consistency with KR20 becomes possible. Depends on situation, to whole such a test or to its subscales (the clusters of positions).
The KR20 is appropriate as the most restrictive type of items in your instrument are dichotomous. However, a question you should consider is whether or not your instrument is measuring only one scale. If you check through factor analysis or other means and find that the instrument has more than one scale, you should compute a reliability on each scale.
Thank you very much for all your responses!!!!!
With regards to the multiple items, I re-coded them into 0 and 1 as correct or incorrect, and now they are dichotomous as well.
Dr. McLean.
Thank you for your response. You suggested factor analysis. Can I do factor analysis of dichotomous variables? If so, what procedures should I use?
Thank you so much for your time.
Tatyana Aultman
- Badges
- Science topic