[EMAIL PROTECTED] (Erin) wrote in message news:<[EMAIL PROTECTED]>... > > I have a study with two groups (patient, therapist) who were scored by > 3 raters using a nonverbal coding system. For simplicity let's say > the coding system has 3 ratings (positive, neutral, negative). These > categories were assessed at 2 time points (high, low). What would be > the best way to analyze this data? Specifically I'd like to know if > there is a significant difference between the high and low conditions > (taking each group seperately). I'd also like to know if there is an > interaction between time point and rating. I tried using a simple two > factor anova but my data is not independent, as the same people are > scored in both conditions. I have done several analyses using simple > paried t-test, but am at a loss as to what my next step should be. > Would a repeated measures anova be appropriate?
Your description is ambiguous. Do you have 1 dependent variable, whose level of measurement is discrete ordinal with 3 categories, with each subject getting 1 rating from each rater in each condition; or do you have 3 dependent variables, whose levels of measurement are unspecified, with each subject getting 3 ratings from each rater in each condition? Also, were all the ratings done by the same raters? If not, then how many raters were there, and which ones rated which subjects under which conditions? Assuming that all the ratings were done by the same raters, then you have a mixed between-and-within design, with 1 between-subjects factor (Group: patient, therapist) and 2 within-subject factors (Condition: high, low; Rater: rater1, rater2, rater3). Group and Condition are fixed effects, Rater and Subject_within_Group are random effects. Assuming that you have only 1 d.v., the easiest path is to treat it as an interval-level measure, with values +1,0,-1, and do an anova. See Myers & Well, Research Design & Statistical Analysis; chap 9 (1st ed) or chap 14 (2nd ed). However, if there are strong floor or ceiling effects then you should consider switching to an ordered-polytomy logistic model. I don't have a reference that will tell you how to do logistic analyses that parallel those in Myers & Well. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
