Hi all,
I'm trying to determine the right stats test for the project I'm working on. The setup is:
- 3 different treatment groups e.g. A, B, and C
- The same treatment is applied across different days, on the same group e.g. On day 1, all the people in group A get the group A treatment. On day 2, group A still gets the group A treatment
- I have at least 2 dependent variables I'd like to test for significance. The DVs are continuous. I'm trying to see if either of them change significantly after the treatments, and also if the days themselves matter.
- In other words, I want to see if some combination of treatment + days affects the dependent variables
What's the appropriate test to use here?
I'm unsure if I should be using a repeated measures ANOVA, a MANOVA, or something else.
Edit: Update with a min reproducible example (see image)
Indeed, your design is not very clear, as Olga Krasko pointed out.
You seem to have a Between x Within x Within design.
Between: 3 treatment: A,B,C. That is clear.
1st Within: several days repeated measures. But how much?? and for every treatment the same number of repeated days?
2nd Within: at least 2 dependent variables. but how many exactly ,and again, the same number for each treatment? If these dependent variables are on the same scale (e.g. a 5-point Likert scale), then you can test the differences (+ occasional interaction with the other factors) with a repeated measures ANOVA, otherwise with a MANOVA
Olga Krasko Jos Feys thanks for the help, apologies for the lack of clarity.
I've created a min reproducible example in the attached image. Here we see 4 people assigned to different groups. They each get the same treatment daily, over 3 days. Each day, the week over week change in some dependent variable is measured. There might be null values sometimes.
The wash-out period is a day.
1st within: There is one treatment per day. It is the same treatment each day.
2nd within: The DVs are measuring week over week changes in a result, a continuous value. E.g. week over week change in the hours worked. It feels like I shouldn't use a likert scale here.
My background is not in research, so I'll likely be neglecting some other important factors. Let me know if I did!
You can run a Between (treatments) x Within (Days) Manova. If there are sign. effects (of treatments, days or the interaction) then you can use post hoc tests.
If the 'Null value's are missing values, you might want to replace these, e.g. with multiple imputations. If the 'change in hours worked' can be considered as the same continuous scale for all 3 DV's (X,Y,Z) then you can use a repeated measures ANOVA instead of a MANOVA. The repeated measures ANOVA corrects for sphericity.
I think your TVtwo (Days) should be considered to be a Within subjects (repeated measures/split-plot) factor. Maybe this can help:
https://stats.stackexchange.com/questions/351442/multivariate-analysis-for-repeated-measures-with-dependent-treatment-groups-and
See also the R script in attachment.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA