I have a sample that violates normality. I took measurements at baseline, immediately after an intervention, and 45 days after the intervention. The measurement at 45 days post intervention is affected by a covariate. I dont think I can run Quade's because the covariate does not affect all measurements. What test would be appropriate?
Thanks in advance!
I didn't really understand the question but an ANCOVA say with a 0,1 coding scheme somewhere may be lurking in the vicinity.. Good luck and stay safe, David Booth
David Eugene Booth thank you for your answer.
We ran a mental health intervention to reduce vicarious traumatization of healthcare workers treating COVID-19 patients. We measured baseline stress levels. Then we ran the intervention and measured stress levels immediately after the intervention. We again measured stress levels 45 days after the intervention, and this time we asked if the participants had continued their practices taught during the intervention or not. We want to test if there is a significant difference between the three measurements. We wanted to use the friedman test, but then whether or not the participants kept up with the practice affects the 45 days post intervention measurements. So, we dont know how to proceed now.
See if you can get these course notes on longitudinal model by Marie Davidian and read them code is included. D. Booth PS PROPER PRIOR PLANNING PREVENTS POOR PERFORMANCE
If I understand correctly this might be better as an ANCOVA with baseline stress (centred) as a covariate and including the compliance/continuation measure (centred) as a covariate and including the interaction between compliance and time point of the intervention. You can also test baselines stress interaction with time point or compliance.
The interaction between compliance and time point of the intervention should then pick up the impact of compliance on the final timepoint. It may also impact the first time point because compliance maybe motivational (i.e., people who engaged with the intervention may continue to engage for longer).
ANCOVA is an outdated statistical method and null hypothesis testing is an inadequate approach in this case. If your response variable violates the Normality assumption, go for a Generalized Linear Model, by selecting a matching response distribution. In my book, I present a decision chart for GLMs and describe how to deal with various sorts of measures:
https://schmettow.github.io/New_Stats/glm.html#elements-glm
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude