|
Size: 375
Comment:
|
Size: 1885
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 2: | Line 2: |
[TAKEN FROM A REPLY BY JEREMY MILES ON THE PSYCH-POSTGRADS E-MAIL LIST] Doing post hoc tests with repeated measures data is really tricky if you want to be parametric, because your results are extraordinarily sensitive to violations of sphericity. (Usually for sphericity assumptions, we use p < 0.05 as a cutoff to decide if we have significant sphericity, for post hoc tests, it's been suggested by some people that we use p <0.5 (yes, that's zero-point-five) instead. This is why SPSS doesn't give the option of doing post hoc tests with repeated measures data. [PW NOTE:In the post-hoc graduate statistics talk we suggest using CBU spreadsheets to get around this (See the Grad talks page).] Actually, that's not quite true, because using a multilevel modeling approach to repeated measures allows you to relax the sphericity/compound symmetry assumption and still do post hoc tests, by specifying a different correlation structure. It's only relatively recently that this has been added to SPSS though. People (and programs) at the 'cutting edge' of statistics don't do repeated measures anova, they do multilevel models instead. For example R, which is (probably) the newest and most advanced stats package around just didn't do repeated measures anova for a long time (I think it's been added now). Generalized estimating equations (recently added to SPSS) and Huber-White sandwich estimates (sort of possible in SPSS) are also other ways to avoid repeated measures anova. |
|
| Line 11: | Line 25: |
Simple Main Effects
[TAKEN FROM A REPLY BY JEREMY MILES ON THE PSYCH-POSTGRADS E-MAIL LIST]
Doing post hoc tests with repeated measures data is really tricky if you want to be parametric, because your results are extraordinarily sensitive to violations of sphericity. (Usually for sphericity assumptions, we use p < 0.05 as a cutoff to decide if we have significant sphericity, for post hoc tests, it's been suggested by some people that we use p <0.5 (yes, that's zero-point-five) instead. This is why SPSS doesn't give the option of doing post hoc tests with repeated measures data. [PW NOTE:In the post-hoc graduate statistics talk we suggest using CBU spreadsheets to get around this (See the Grad talks page).]
Actually, that's not quite true, because using a multilevel modeling approach to repeated measures allows you to relax the sphericity/compound symmetry assumption and still do post hoc tests, by specifying a different correlation structure. It's only relatively recently that this has been added to SPSS though.
People (and programs) at the 'cutting edge' of statistics don't do repeated measures anova, they do multilevel models instead. For example R, which is (probably) the newest and most advanced stats package around just didn't do repeated measures anova for a long time (I think it's been added now).
Generalized estimating equations (recently added to SPSS) and Huber-White sandwich estimates (sort of possible in SPSS) are also other ways to avoid repeated measures anova.
[Last updated on 27 November, 2003]
[wiki:FAQ Return to Statistics FAQ page]
[wiki:CbuStatistics Return to Statistics main page]
[http://www.mrc-cbu.cam.ac.uk/ Return to CBU main page]
These pages are maintained by [mailto:ian.nimmo-smith@mrc-cbu.cam.ac.uk Ian Nimmo-Smith] and [mailto:peter.watson@mrc-cbu.cam.ac.uk Peter Watson]