Just a comment. There are two issues here. One is, do test scores predict performance for the subset of students admitted? The other is, do test scores predict which students should be admitted? You don't have further data on the students screened out, which affects your ability to test the validity of the screening test. It's likely that any correlation you compute on the students who passed the screening will underestimate the correlation that would be computed across the full range. That assumes of course that the relationship between test score and course performance is linear.
Cook & Campbell discuss this issue in their book on quasi-experimental design; I don't remember what they suggested since I have not had occasion to use this. As to the specific question of computing a correlation with a categorical variable (grade) and a quantitative one (test score), I would suggest you treat grade as ordinal (which it certainly is), and compute Spearman's r, or treat it as quantitative (since after all we do compute GPA) and compute Pearson's r. If you wanted to correlation grades in English IA with grades in English A, there is a statistic comparable to phi for larger 2-way chi square designs, and it is computed using chi. Howell's textbook (the more advanced one) provides it; I can email the reference and/or the formula to you tomorrow if you remind me. Charlotte >I have been asked by the VP to look at the validity of the screening tests used by >the English and Math Departments. The general idea is to look at the scores on the >two tests and see how well they predict subsequent grades. Another task would be to >compare the grades students get in English 1A with the grades they get in English A >e.g. what percentage of student who get an C in IA pass English A. I know I can do >sorting for Grades in A by Grades in 1A and get Chi squares with any of my stats >programs. My question is, however, What test of correlation do I use to compare a >numerical score on a test to a letter grade? Would I be better off converting >numerical scores into quintiles? I am somewhat vague about the details but I know >that ties affect correlations. By using either method I will end up with a lot of >ties in the since the range of grades is only 1-5. I would appreciate a reference to >the authority on whatever method you suggest because I know the math department will >be livid with rage that I am doing this. The psych department got into a knockdown >dragout ugly battle with the math folk over our offering a stats class in the psych >curriculum. The head honcho stats guy stated flatly that psychologists don't know how >to use statistics, teach statistics or interpret stats. A couple of them still won't >talk to us! > > > >Harry Avis PhD >Sierra College >Rocklin, CA 95677 >email: [EMAIL PROTECTED] > >Life is opinion - Marcus Aurelius >There is nothing that is good or bad, but that thinking makes it so - Shakespeare > > >_________________________________________________________________ >Send and receive Hotmail on your mobile device: http://mobile.msn.com > > >--- >You are currently subscribed to tips as: [EMAIL PROTECTED] >To unsubscribe send a blank email to [EMAIL PROTECTED] ==================================================================== Charlotte F. Manly, Ph.D. Psychological & Brain Sciences Assistant Professor 317 Life Sciences Bldg ph: (502) 852-8162 University of Louisville fax: (502) 852-8904 Louisville, KY 40292 [EMAIL PROTECTED] http://www.louisville.edu/a-s/psychology/ http://www.louisville.edu/~cfmanl01 --- You are currently subscribed to tips as: [EMAIL PROTECTED] To unsubscribe send a blank email to [EMAIL PROTECTED]
