There are two treatment groups, as an independent variable, one group received drug A and another group received control article. The white blood cell numbers were observed at different time-points, such as 1 weeks, 2 weeks, 1 month, 6 months, 1 year and 2 years. So there are 2 curves, one is drug A, another is control, x-axis is time, y-axis is WBC. We want to see is there a difference between these 2 curves and the interaction between time and treatment.
Which statistic method I should use? My stat software is JMP.
Q1: The 2-way Anova has an assumption that the two independent factors must be categorical, but the time is numerical data, should I transfer the numerical data into categorical? Such as input day 1, day 10, day 30 instead of 1, 10, 30?
Q2: Another assumption of 2-way Anova is that each group should have the same size. But my data is a pool-up retrospective analysis where different time-point have different number of subjects. Will it still be suitable for 2-way Anova?
Q3: Will repeated measure Anova be better? My concern is that not all subjects repeated the WBC test in each time-point, such as some subjects only have the 1 month test, others only have the 1 year test.
Q4: Will Manova is an alternative choice? But there is only one dependent variable--WBC level, seems not qualify for Manova.
A mixed effects model is superior to what you have mentioned above. Another alternative might be growth curve modelling.
@ Mehmet Sinan Iyisoy
In the JMP guide video, it needs to designate one factor as Random to get the REML Method. Which factor do you think is the random one, the time or the treatment?
And do I have to transfer the time factor from numerical data into categorical? Such as input day 1, day 10, day 30 instead of 1, 10, 30? I saw on the guide both factors are categorical.
https://www.youtube.com/watch?v=Ww7n-JG7_fs
Thank you very much!
Subjects is the first candidate for a random factor. I generally use random intercepts models. Then it could be time. One can hardly say that treatement is random. Because it is generally not.
For considering time as categorical or not, I can say that if you have many levels of time, then it would be better to think it as numerical, since it would ease interpretation.
Hi, Suggest that you take a look at these notes on repeated measures type models. I think it will help you a lot if you do quite a bit of this. See this link. They are by Prof Marie Davidian.
https://www.stat.ncsu.edu/people/davidian/courses/st790/notes.html
Best wishes, David Booth
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs