Anyone out there know how to program SAS to compute a simple correlation coefficient (Pearson’s r will do) that corrects for unreliability on the criterion? And while we’re on it, do you have information on when it’s appropriate to do such a correction? I can’t seem to find the topic in any of the stats textbooks I have handy.
If you can help, leave a comment or e-mail me. Thanks!
I don’t think SAS does unreliability corrections, but I could be wrong. I remember having to use a program developed by Hunter when I was in Frank Schmidt’s M&O methods class, but that was four years ago so things may have changed since then. There is a section on correcting for unreliability in Nunnally & Bernstein (1994), “Psychometric Theory” Chapter 7. I’m not sure when it’s appropriate – I just know in Schmidt’s class we corrected everything!
The correction for attenuation given by NB is r’12 = r12/sqrt(r11 r22), yielding the expected correlation between two perfectly reliable variables. Also you can compute what the correlation would be if reliability were increased to a particular amount by multiplying the correlation by sqrt(r’xx r’yy)/sqrt(rxx ryy) where rxx is the reliability for x, ryy is the reliability for y, and r’xx and r’yy are the expected changed reliabilities for each.
I hope this helps. Sorry I don’t know more about it – we don’t correct for anything in my area!
Really? Cool. For what were you taking a class on research methods from Frank Schmidt?
Why does that not surprise me? 🙂
Thanks for the tip! I think I may have a copy of Nunnally & Bernstein lying around here somewhere, else I may be able to get it from a library. It’s good to have a mother who works in the Interlibrary Loan Department of a university library.
I’m glad it was of some help! I was in Frank’s class the first year of my PhD program because I’m doing behavioral research in managerial accounting, which is definitely *not* the focus of the accounting department here! My advisor felt the M&O methods class would teach me more of the methods I needed for my research. He was right – I never would have learned about factor analysis, path analysis, SEM, hierarchical modeling, or meta-analysis if it hadn’t been for that class, and those techniques are very useful in studying managerial behavior. It seems like in accounting, everyone just runs regression after regression and ANOVA after ANOVA! Plus, it was fun having Frank as a professor, hearing all his ranting about significance testing and such.