I have tested three different emotions and their impact on time (three within group bisection points). The entire data is within group with gender as a between group design.
However, various assumptions have been violated for me to use a repeated measures ANOVA.
What would be a good alternative?
I think using a Generalized Estimations Equation (GEE) might be a reasonably good alternative and a method of analysis that is fairly easy to present to the end user - the reader of your paper. GEE accounts for interdependence between observations butcan be used to look at level differences (in this case, between gender and time points). To use the GEE you will however have to restructure your datafile so each participant has three rows, one for each time point. SPSS provides a user friendly interface for the GEE and its almost self-explanatory. I have used it in several of the papers on my profile so you can perhaps take a look.
I think using a Generalized Estimations Equation (GEE) might be a reasonably good alternative and a method of analysis that is fairly easy to present to the end user - the reader of your paper. GEE accounts for interdependence between observations butcan be used to look at level differences (in this case, between gender and time points). To use the GEE you will however have to restructure your datafile so each participant has three rows, one for each time point. SPSS provides a user friendly interface for the GEE and its almost self-explanatory. I have used it in several of the papers on my profile so you can perhaps take a look.
Friedman's Test is a fairly standard non-parametric statistic that's analogous to a repeated measures ANOVA. Most statistical packages support this test, though the post-hoc capabilities may be more limited.
If using SPSS, one possible solution may be to run Friedman's Test first and then run a standard repeated measures ANOVA for the post hoc if the main concerns are unequal variances. In SPSS you can select from many post-hoc tests that work with data violating the equal variance assumption. The Games-Howell test is liberal while the Tamhane T2 is conservative.
That said, if your concern with the repeated measures ANOVA is that Box's test is violated, SPSS provides some corrected statistics that get around this. In the tables, look for the Greenhouse-Geisser outputs, which may be used instead of the Sphericity Assumed row should Box's test be violated. If the Levene's statistic for the between-subjects factor is violated, one can then try running a one-way or non-repeated measures ANOVA to test between subjects effects using Welch's test or the Brown-Forsythe test (in SPSS these are options of the "One-way ANOVA").
I hope this helps!
I just wanted to clarify that my response concerns primarily the dependence between the within subject variable. The state at the three time-points will likely be correlated for your participants and this is something you will need to adress in your analysis.
I study behavior in Insects. A tool we use is proc Glimmix in SAS. You might also want to test the correlation between repeated samples. If they are not correlated then a repeated measures is not the right answer anyway. If they are correlated, then there is a problem.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA