0

I am running a study looking at the effect of Vitamin C on muscle function. For my study design, I had 2 groups: A treatment (vit C) and a control (no treatment) group. Pre and post measurements were taken for Muscle function, and I was wondering what is the best statistical test to perform? Subjects were randomly assigned to either group.

dipetkov
  • 9,805
Connor3351
  • 11

1 Answers1

1

A 2x2 mixed analysis of variance would probably work well, assuming your muscle function outcome variable is continuous. Your experimental variable would be your between subjects factor and your pre-/post- measurements of muscle function would be your within subject factor.

jsakaluk
  • 5,514
  • 1
  • 23
  • 47
  • Ok thanks, yes, the muscle function outcome variable is continuous. Would a paired t-test work in this case as well, or is the fact that there are different subjects in the two groups makes the paired t-test an unappropriate statistical test in this instance? – Connor3351 Mar 27 '15 at 16:37
  • No, not to answer the question you want to answer. A paired t-test would only allow you to test whether muscle function changes from pre-to-post, irrespective of your participants' experimentally assigned Vitamin C condition.

    A 2x2 mixed ANOVA, alternatively, will allow you to examine whether your two experimental conditions are different at either pre- and/or post-muscle function measurement. Presumably you are anticipating differences between conditions at the post-test assessment.

     – jsakaluk Mar 27 '15 at 16:50
  • 1
    Oh, ok. Yes, I'm expecting the Vit C group muscle function post-test assessment results to be different to the placebo group. What would be done in the case of non normally distributed results? Can a 2x2 mixed ANOVA still be run? – Connor3351 Mar 27 '15 at 17:06
  • It depends how non-normal the distribution of your response variable is--ANOVA is robust to some violations of assumptions (e.g., normality). You should check your skewness/kurtosis statistics of your response variable. Also, are your treatment group sizes relatively equal? Unequal cell sizes exacerbate the severity of assumption violations. – jsakaluk Mar 27 '15 at 17:09
  • So the response variable is the muscle funtion in this case, right? Regarding the treatment group sizes, both groups contain almost the same amount of participants. – Connor3351 Mar 27 '15 at 17:21
  • Yes, sorry, I use response/outcome/dependent variable interchangeably. Equal groups is nice, though you should still check and interpret your skewness/kurtosis statistics. Many texts recommend that skewness should be less than 2, and kurtosis less than 7. – jsakaluk Mar 27 '15 at 17:25
  • Ok, thanks. Let's assume, for example, that skewness was less than 2 and that kurtosis was less than 7. What would be done in that case? – Connor3351 Mar 27 '15 at 17:36
  • In that case, and since your group sizes are equal, probably what I have recommended: a 2x2 mixed ANOVA. – jsakaluk Mar 27 '15 at 17:37
  • I actually meant if skewness was more than 2 and kurtosis was more than 7..Sorry for asking so many questions by the way hehe – Connor3351 Mar 27 '15 at 17:43
  • Then you would likely want to use some sort of non-parametric test (I'm sure there is one--not my area of expertise), or if your response variable was better characterized by some other random component (e.g., a negative-binomial distribution, instead of a normal distribution), then you would probably want to use some sort of multilevel generalized linear model. In either case, you should probably seek help in a separate question about that specifically, as the current title for your question doesn't scream "non-parametric" or "multilevel generalized linear model"--at least not to me. – jsakaluk Mar 27 '15 at 17:49