Hi all,
I've had the same conflict with SPSS not providing a p-values for Mauchly's Test of Sphericity; this occurred for two different 3-WAY-RM-ANOVAs I recently ran.
In one of the 3-WAY-RM-ANOVAs I have a total of 89 subjects; I have 1 within-subject repeated measure (with 75 levels); I have 2 between subject factors (one with 2 levels and one with 6 levels). I attached screenshots of the data- and variable views for this ANOVA; no p-value was provided for Mauchly's Test of Sphericity.
In the second 3-WAY-RM-ANOVA, I have a total of 14 subjects; I have 1 within-subject repeated measure (with 75 levels); I have 2 between subject factors (each between subject factor has 2 levels).
Can anyone provide any suggestions?
I know there are questions very similar to this one that were posted previously; however, I have not found the comments to be applicable to my data, and both previous posts were last accessed over a year ago.
Any help that anyone could provide would be greatly appreciated!
Thanks,
Brenna
206-819-9647
You have a lot of levels for 89 subjects. I wonder if you tried subsetting your dataset to a smaller but similar experiment if you would have the same problem. This is just a guess. Best of Luck, David
Hi Brenna,
David Booth is pointing you in the right direction. The sample size is very small relative to the number of levels of the repeated measures factor (in both data sets), and certainly so for all of the related interaction terms that involve the repeated measures factor. That is, you have fewer cases (89) than the degrees of freedom for within-subjects effects such as B*R (factor B by the repeated measures factor R) (370) or the A*B*R (again, 370). In the second analysis, you have fewer cases (14) than repeated measures observations (75).
If it made sense to do so, consider collapsing the repeated measures scores into fewer values. Alternatively collect more data (though I suspect that won't be a desirable prospect). The last option would be to ignore the sphericity test, and use one of the epsilon adjusted values (perhaps Huynh-Feldt, which appears to work reasonably well) with the explanation for why you couldn't evaluate the assumption.
Good luck with your work.
Brenna, what is the RM variable that has 75 levels? The GLM command you are using treats it as a nominal variable. I doubt very much that it is purely nominal. Often, the repeated measures factor is Time, for example. Consider restructuring your data from WIDE to LONG (see VARSTOCASES command) and using MIXED to estimate your model. This will give you the option of treating Time (if that's what the variable is) as continuous. If you do that, it will use only one df rather than 74 df. If the effect of time is not linear, you can introduce further terms to achieve a better fit (e.g., Time2, Time3). Here is a UCLA page that may be helpful:
- https://stats.idre.ucla.edu/spss/seminars/spss-mixed-command/
HTH.
Thanks all!
David Booth -- surprisingly, I ran the same data set with 2 additional groups added in (and 2 additional levels to one of my between subject factors) and DID receive a Mauchly's p-value, so I don't think the issue has to do with the number of subjects relative to the number of RM levels per say (but thank you for the suggestion).
Bruce Weaver -- you are correct that my RM variable is time. I actually have 300 levels within this variable (I obtained samples every 30 seconds for 150min); however, SPSS will only accept 99 levels within the within-subject factor, so I had to bin my data into 2-minute bins to create just 75 levels. I'll take a crack at restructuring the data as you suggest, and thanks very much for the link!!
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA