Stats question
Christie Wilcox
Wednesday, 01 April 2009 18:35 UTC
Hey there
I’m trying to work up the stats for a series of viability assays and I have a question.
The protocol is that I pool the cells together for a given exp, then separate them evenly into different tubes and give them different treatments, with one tube left alone for ‘control’ viability. I did a number of these, and the basal viability levels can be variable. So now that I’m looking at stats (using sigma stat), I want to know if I should use a regular ANOVA or a Repeated Measures ANOVA to compare the different treatments because each treatment’s viability is relative to the control level. The data is set up so that each exp is in line in the rows, with the different treatments in line in the columns.
I hope that made sense. Anyhow, if anyone could let me know which is better (I am thinking RM, but I don’t want to use a more powerful test than I’m able).
THANKS!
-
Replies
-
Hi Christie,
I would suggest posting this in the Nature Protocols forum as well. -
I am not sure repeated measures is appropriate here. RM is used to test for differences between treatments when the response variable for each case has been measured more than once (e.g. on different days).
-
I’m no expert but I believe Raf is correct. RM is for non-independent measurements of the same experimental sample (kind of analogous to using a paired vs. unpaired T-test, in a way). You have independent measurements of different samples. That they were derived from a different initial culture is, I think, not important.
Disclaimer – I’m not a statistician ;)
-
Phoo. I meant:
That they were derived from the same initial culture… [blah blah blah]
-
Hm… I see what you’re saying. I guess my question is this:
Don’t the data violate the implied “independence of samples” of a regular ANOVA since, let’s say during Exp. 1 the basal viability is 90% and a few weeks later, Exp 3 the basal viability is 75%? The different treatments’ viability is tied to how viable the cells are that day, and that is highly variable. Isn’t that what a RM ANOVA is supposed to account for? Do different treatments to the same batch of cells (with their inherent viability) count as repeated measures of the same “individual” of sorts? I know it’s not exactly the same as measuring a person’s height at different ages or something, but it seems to me there ought to be a way to statistically account for the variation in control viability between experiments.
-
I guess, to explain the exp. better, I’m looking at cells at 3 time points after treating them with one of three drugs and hydrogen peroxide to kill them, to see if the drugs prevent their death. The most “death” that occurs, though, is around 20% loss. Those that die do it about the same amount every exp no matter the base line, so they drop by 20% whether they start at 90% or at 75%. But the starting point makes the data highly variable at that last time point- 90 down to 70 v. 75 down to 55, etc. Is the repeated measures the right way to account for that initial viability?
I was thinking of it because I know you use a RM when you have drug studies in people where you’ve matched a group for whatever factors you’re looking at (aka get together 6 women with the same depression scores) and then give each one different drugs/placebos to see how it affects their mood. You don’t have to have the repeat in the exact individual giving them each drug in turn to see how it affects them, you can count the group of matched people as an individual…
-
let’s say during Exp. 1 the basal viability is 90% and a few weeks later, Exp 3 the basal viability is 75%?
I would include basal viability as a covariate.
-
