I am doing a study to analyze the effects of knocking out a particular gene on a certain behaviour in flies. After a log transformation, my observations have a normal distribution so I'm going to do a three-way ANOVA, with age as the first, sex as the second and genotype (mutant vs wild-type) as the third variable, also looking at interactions. The problem is, my mutants come from two slightly different strains that have never been shown to differ behaviourally but I should still check first to determine whether I need to distinguish between the two in the subsequent analysis or not. I'm wondering whether this could be done with a nested ANOVA (all the examples of nested ANOVAs I've seen look quite different from what I have, which is why I haven't tried it, but I'm probably just not getting it).

So far, what I've done is run a three-way ANOVA (type III ss) on just the mutants, with age as the first, sex as the second and genotype (mutant strain 1 vs mutant strain 2) as the third variable. Age, sex and the interaction of age and sex are both significant while mutant strain isn't, nor are any of the interaction terms containing it. My question is, can I just report the fact that the mutant strain has no significant effect and then only worry about the age and the sex when I'm running the ANOVA on all my flies (this time without distinguishing between the two strains of mutants) or do I have to report it and take it into account now too? And what if mutant strain had come out significant? Would that change anything as to whether or not I should report the significance of age and sex at this stage?

I would have just done a t-test to compare the two mutant strains but I was afraid there might be an effect that was masked by either age or sex as both seem to have an effect on the behaviour I'm looking at.

Thank you so much in advance, I know this is a newbie question (it's my very first real statistical analysis but I'm enjoying every second of it) but if any of you had a second to answer I'd be very grateful.

  • $\begingroup$ What is the subsequent analysis? $\endgroup$
    – Flask
    Dec 28 '13 at 21:23

An important (but sometimes overlooked) criteria for what makes a "good" model is parsimony - i.e. all else being equal, always choose the less complicated model.

Based on your description, it sounds like you should report the more simple model (age, sex) and then note somewhere else that you found no evidence of heterogeneity in these estimates by strain type.

There are certainly other ways to approach this kind of problem, but given your stated level of experience, this is what I would suggest.

  • $\begingroup$ Thank you for your reply! The problem is, I am only doing this as a preliminary test to see whether I have to take into account which strains the mutants are later on when I'll be running the ANOVA comparing the wild-types and the mutants. I will be including age and sex as variables anyway at that point so I'm just wondering whether I can for the moment just ignore the fact that I'm getting an effect with age and sex in my preliminary ANOVA too and just report the fact that there are no significant differences between the strains. $\endgroup$
    – Rosina
    Apr 1 '13 at 12:32
  • $\begingroup$ I'd say that this shows that the mutants don't differ when age and sex are controlled for. $\endgroup$ Dec 28 '13 at 16:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.