I’m Rozaini Abu Bakar from University of Science Malaysia. I have a few questions,
1. Is there any minimum sample requirement to analyse using rasch model? Is it true that at least there must be more than 30 sample?
2. I currently conducting a study on the development of an assessment rubric for preservice teacher. In this case, I’m using multifacet to validate the rubric. I’m planning to use 15 raters and 4 examinees (preservice teacher) to assess their teaching practise using my rubric which is consists 50 items. In order to exceed minimum 30 sample, I’m calculated (15 raters x 4 preservice teacher x 50 items = 3000 ratings), which is more than 30. So, am I fulfilling the requirement of minimum sample size of rasch model?
3. What is the most effective way for these 15 raters to assess their 4 preservice teacher? How to link it in the control file?
4. Is it ok if I have more raters than the examinees. This is because I have refer to the book entitled ‘Rasch Analysis In Human Science’ written by William J.Boone, John R.Staver and Melissa S.Yale in chapter 20.
1. The minimum requirement for a Rasch analysis to construct meaning is usually 3 persons and 3 items. The number "30" is explained at www.rasch.org/rmt/rmt74m.htm - it applies when we want statistically-stable estimates of ability, difficulty, etc.
2. Your focus is the assessment rubric and its 50 items. So you need 30 observations of each item. You will have 15x4 = 60 observations of each item. Plenty!
3. We only need 30 observations, so we can have judging plan like: 15 raters xx--xx--xx--xx- examinee 1 -xx--xx--xx--xx examinee 2 --xx--xx--xx--x examinee 3 x--xx--xx--xx-- examinee 4 It doesn't matter if a few mistakes are made in implementing this design. There is plenty of connectivity.
The data could look like: rater number, examinee number, 1-50, (50 ratings)
or a judging plan where all 15 raters rate all 4 examinees, but with a judging plan for the 50 items, similar to the one above.
4. You know who is a "rater" and who is an "examinee". The Rasch model does not know. It only knows about numbers. So more raters than examinees is fine! You can think of the "examinee performances" as "tasks" for the raters to perform.