It is about an experimental study, where subjects were tested twice at baseline (without supplementation) and after a week with the supplemented and placebo groups.
Each test consisted of measuring some variables at rest, followed by a load test, and then re-measuring the variables at a fatigue state.
Do I need to use Anova repeated messures?
I ask for your help and suggestion, thanks.
Yes, you could run a repeated measures ANOVA, or an ANOVA on the change scores. Yet, a lot of statisticians would advise to run an ANCOVA, i.e. a regression: post = pre + conditions. But first do inspect the normality of distributions of your measures with histograms and/or QQ-plots (maybe also the Shapiro test, although this test is known to be biased by size [the larger the size, the more prone to --maybe falsely-- indicating the distribution is not normal]). For the ANCOVA, a.o. the assumption is that the relationship between the outcome and the covariate must be the same in all groups. I.o.w., there must be no interaction between the conditions (treatments) and the covariate (the pre-measure).
Jos Feys thanks for answer, your reasoning is very clear and helpful.
If your data follows a normal distribution, you can use the paired t-test if you have just 2 repeated measures (pre and post). If you have 2 or more repeated measures, you can use repeated measures test. In the event that your data does not present normality and the transformation attempts are not successful, you will have to use non-parametric tests although the latter are less robust.
Ouorou Ganni Mariel Guera my data follows a normal distribution. I was considering paired t-test, but i have 2 repeated measures in first and also 2 in second trail, so i think that maybe Anova repeated measures is more appropriate. thx for your sugestion
I quote from Stokes et al. (1995) Categorical Data Analysis Using the SAS System, p. 156:
"The validity of an ANCOVA depends on several assumptions, including normality of error terms, equality of error variances for different treatments, equality of slopes for different treatment regression lines (say correlations between pre and post), and linearity of regression. For situations in which these assumptions may not be satisfied, Quade (1967) proposed the use of rank analysis of covariance."
This rank analysis of covariance (RANCOVA) is an ANCOVA on the ranks of the pre and post-measures. This implies to first compute the ranks of the response (post) variable and the covariate (pre) in the combined groups (treatments).
Quade, D. (1967). Rank analysis of covariance. Journal of American Statistical Association, 62, 1187-1200.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA