I have a feeling that I'm probably asking the impossible, but I'd like to look at the effects of treatment, time, and treatment x time on 22 daily body mass measurements without having to consolidate them into temporal averages.
1. Repeated measures ANOVA does not leave me enough degrees of freedom to analyze the effect of time or its interaction with treatment with regard to body mass.
2. One-way ANOVA is out of the question because my design violates independence among the data points thanks to my 22 repeated mass measurements, which were taken from the same subjects.
3. A generalized linear mixed model appears to be calculating my degrees of freedom for comparison between the two treatment variables from n = the total number of mass measurements, which suggests the same one-way-ANOVA-like assumption of independence that doesn't fly in this case (even though I thought that mixed models were supposed to take repeated measures into account. Is it possible that I set it up incorrectly?).
4. Multiple independent t-tests at each time point open the multiple comparisons can of worms and don't offer me the metascopic picture that I am seeking.
Am I up the creek with a sample size too small for the analyses I want to run and no recourse beyond repeated-measures ANOVA and/or consolidation of some data points, or does there exist another way to investigate the effects of these two fixed factors?
Thanks so much for the advice and book chapter. The mixed model approach has given me the most plausible results so far, but the degrees of freedom that SPSS is giving me for each analysis are still too high even with the random effect added to my model unless I direct it to "use robust estimation to handle violations of model assumptions" (though I thought that I wasn't violating any assumptions since I built my model to take into account repeated measures). I'll see how R handles the task.
Hi Kelley,
I am not sure if your still interested in a response to this question, but I would suggest you consider one of two options: a GEE model or a time series model. Both will handle the autocorrelation of multiple observations, but time series models are better (or at least more often used) when sample sizes are small.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA