Thanks to Dr. J. Martin Bland and Dr. Peter L. Flom for their kind and valuable hints. My problem, however, remains whole and I would be very appreciative of any help. Am I allowed to include statement, gender, SES, and rater as independent variables? Here are the suggestions:
Dr. Bland wrote:
You do not say why you did all this. 30 observers each rated a lot of subjects. This seems like a very extravagant design. I would take a summary for each suject, say the median category on the Likert scale and do ordered logistic regression, or bite the bullet and average the thirty ratings, then you have a conitnuous measure and can do least squares regression..
Martin
Dr. Flom wrote:
<<<<
I collected two datasets. One from pupils and the other from teachers.
From pupils: 50 pupils were asked to reproduce a sentence without particular direction. Two weeks later the same pupils were asked to reproduced the same sentence following specific directions as how to form capital letters. The two samples were shown to 20 (not 30) raters
to rate any improvement on a Likert type scale from 1 to 5. I wanted to run t-test to confirm or infirm any improvement. A friend suggested to make it more interesting by running ordinal logistic regression taking into account the age of pupils, the gender of raters, the gender of
pupils, and SES.
>>>
So each rater made 50 ratings on each of two occassions?
To do what your friend suggested I think you need a nonlinear mixed model. In SAS, you'd want PROC NLMIXED, in R, the nlme library. This isn't simple. The problem is that the data are clustered, in that each rater rated multiple things....
<<<<
From teachers: twenty teachers were asked to rate on Likert type scale,1 to 5, six statements on pupil-teacher relation. Once again, I wanted to run t-test to confirm or infirm any difference between male and female teachers. This requires six different t-tests. However, based on
what was suggested for the pupils' data I thought it would interesting to run ordered logistic regression taking into account raters' seniority, gender, and possible relation among the statements. I really don't know if it makes sense to include statement as independent
variable. I am refering to a model like
>>>>
Here an ordinal logistic model seems appropriate.
HTH
Peter
|