I am conducting a study where 4 groups are measured over 4 different time points (t0, t30, t60 and t90). I am mainly interested in what happens within groups. I am carrying out RM_ANOVA, I got the mean differences between the groups but I want to represent the differences in percentage.
Any idea what is the best method to do this?
Log-transforming the DV (assuming it must always be positive) might get you close to what you want. See the discussion and examples on this UCLA page:
https://stats.oarc.ucla.edu/other/mult-pkg/faq/general/faqhow-do-i-interpret-a-regression-model-when-some-variables-are-log-transformed/
Me is confused. If your DV is a percentage how can the difference between the means of the groups not be a %? If the mean of t60=60% and that of t30=30% would it not be 60%-30%=30%? Wouldn't it be easier to fit a beta-regression and perform a pairiwise comparison over the estimate coefficients. Or are you intersted in the overall variance?
Wim Kaijser and Jos Feys, I wonder if you looked at the UCLA page I pointed to. If not, take a quick look at the section with this heading:
Outcome variable is log transformed
Then search for the word percent. This is what I understood Abhilash Ludhiadch to be asking about. But until he posts again, I guess we won't know for sure. ;-)
Bruce Weaver makes sense too. Did you focus on this because log transformation of y is easier to grasp then when working the logit function. Or, because we can now also have values >100% e.g. 120% and assuming normality? If the latter is the case beta-regression is indeed out. I was thinking in fractions not percentages🙈.
Hi Wim Kaijser. I think you understood the OP to say that the DV is a percentage, and if it is, then you could convert that to a proportion (p) and use logit(p). But I did not understand the DV to be a percentage. I understood that the OP wants to convert differences between groups on whatever the original scale is to percentage differences. And provided the original raw values are constrained to be positive, the method shown on that UCLA page ought to give the desired result, or something that can easily be converted to the desired result. But again, the OP needs to respond and indicate if any of us are understanding the question correctly. ;-)
Abhilash Ludhiadch in what quantity/units is the dependent variable measured? And as a percentage of what would you want to represent them?
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers