Compare adj-R2
in [SAS]
|
Prev: Zip to Congressional District?
Next: Urgent Position: (JMP/SAS Expert), San Jose CA 6 months Contract
From: Steve Denham on 2 Feb 2010 15:24 Since you are fitting the same model to different data sets, there doesn't seem to me to be a reason to compare adjusted R2. You could compare the R2. A simple approach is to just look at the significance of the F test for the regression, as the Fvalue is equal to (R2 divided by the number of regressors) in turn divided by ((1-R2) divided by the total number of observations minus the number of regressors minus one). or F= (Rsq/k)/((1-Rsq)/(n-k-1)). You could look at the significance of each regression. A much more complicated way would be to consider the cases as replications of the study, since the model being fit is identical. The existing model terms become fixed effects, and replication could be treated as a random effect. Using the COVTEST option in PROC GLIMMIX would give a likelihood ratio test of whether there was additional variation over the residual error due to replication--in other words, does accounting for replication explain any additional variation. If yes, then the R2 values are different. I await the comments of others on this approach... Steve Denham Associate Director, Biostatistics MPI Research, Inc. ----- Original Message ---- From: Nash Will <nashwill(a)GMAIL.COM> To: SAS-L(a)LISTSERV.UGA.EDU Sent: Tue, February 2, 2010 9:16:12 AM Subject: Compare adj-R2 Hello everyone I've a problem while comparing the results of two regressions. These two Regs are performed with different datasets and I want to compare the adj-R2 (coefficients are not my concerns) I know Vuong test can help to compare R2 if using the same dataset. But since I use different dataset but the same regression model, I wonder if any proc in SAS can test whether an adj-R2 is significantly different from another one. Many thanks. Will |