Hi,
can someone please help me with their expertise. To analyse a dataset with 2 IV´s (2 levels each) and 1 DV (time - measured at 4 points, i. e. 4 DV?) I found the two-way repeated measures ANOVA might be the right choice. However, some data are not normally distributed and I can´t make sense of the Friedman test, since it doesn´t allow me to enter the two IV´s, what am I doing wrong?
After reading a few days, I was wondering whether I should run a MANOVA instead, but this doesn´t solve for the data not being normally distributed?
I am using SPSS.
Thank you very much for your support.
Hello Anne Fischer. First, let me remind you of George Box's famous comment on normal distributions and straight lines--see the attached image. ;-)
Second, you say the DV is "time". Are you talking about response time (RT) measured in milliseconds? Or some other measure of time? And is there just one measure for each of the 4 occasions? Or do you have several raw measures that allow you to compute a mean or median for each occasion? I ask, because if your measure is RT, and you have just one measure per occasion, then the distribution will almost certainly be positively skewed.
Finally, if the residuals from your intended RM ANOVA model are positively skewed, you could try using GENLINMIXED with error distribution = GAMMA.
- https://www.ibm.com/docs/en/spss-statistics/28.0.0?topic=genlinmixed-target-options-subcommand-command
- https://en.wikipedia.org/wiki/Gamma_distribution
I hope this helps.
It just occurred to me that GENLIN with generalized estimating equations (GEE) is another option you could explore. It also has an option to use a GAMMA error distribution. And it might be a bit less fiddly than GENLINMIXED.
- https://www.ibm.com/docs/en/spss-statistics/28.0.0?topic=genlin-model-subcommand-command
Hi Bruce Weaver, hi Jos Feys,
thank you very much for your answers. I can confirm the time measured at all 4 timepoints for the DV is another measure of time (time consumed for a given task), alternatively likert scale data, for a set of participants, so can build a mean per timepoint but values are - due to the nature of the test- likely to increase over the course of the 4 timepoints, hence your assumption about them being positively skewed makes sense to me. Since I am comparing 2 trials with these 4 timepoints for my DV and the 2 IV´s, the difference in DV results resulting from the IV might as well impact normality of data.
I understand now this is eventually not considered a non-parametric dataset in the normal sense and will explore the options you both have provided in your answers. You have helped me a lot, thank you both very much, your support is greatly appreciated!
Hello Anne,
I think you should run a MIX ANOVA test, because in repeated measures ANOVA, you only have one group, but when you measure two groups at four different times, you have to run a MIX ANOVA.
We only have one group in repeated measurement ANOVA.
Best wishes
Esmail Shirdel Havar
The only things I would add are: even if appropriate the repeated measures MANOVA requires balanced data, ie no missing data thus one more reason to use either a mixed model or a GEE based model. In choosing between the two one must take into account whether a population average or a person level model is appropriate or desired. Mixed gives person level estimates while GEE has population averaged estimates.
- Badges
- Science topic