I am analysing for differences in seagrass cover (cover) over time (period), specifically if the seagrass cover differed significantly from the baseline (periodbaseline1 and periodbaseline2).
The data was found to be non-normal (shapiro-wilkins) and variance was not homogenous (Levene's Test). The data was log-transformed (cover_log) and although the data remained non-normal, the variance was homogenous. A glm was carried out and the output showed that qr6 was significantly different from periodbaseline1 (P=0.04322). I carried out a pair-wise comparison with the the multcomp package as I also needed to compare qr6 against periodbaseline2. Unexpectedly, the result for qr6-periodbaseline1 changed to being not significantly different (p=0.4665). In a case such as this, which result should I be taking, the one from the glm output or the multcomp output?
I advise to use statistical tests of null models where you do not have to analyze the normal distribution or the homogeneity of variance, in addition in this type of statistics the null hypothesis is accepted. For comparison of two means the Monte Carlo permutation test and for comparison of multiple means PERMANOVA tests are non-parametric and you can adjust them for 10 000 permutations. You can make them in the Past. 2.35, PRIMER 6.0 or R 3.0.3.
I attach the website of Pats 3.25 so you can download it if you use it.
https://www.softpedia.com/get/Science-CAD/PAST.shtml
https://www.filecluster.es/programas/PAST-123690.html
It appears to also happen with qr 5 and 3. The problem is that you are performing t tests separately on each qr. You need one big data table that produces only one p value from your glm. And if this is < 0.05, only then should you perform pairwise comparisons
Thanks for the answers. Andrew Paul McKenzie Pegman I carried out an analysis of deviance: anova (glm_mod, test="Chisq"). There is a significance difference amongst the periods.
Enrique Reynaldo de la Cruz Thanks for the suggestion and references. I'll check out the Monte Carlo permutation test.
- Badges
- Science topic
- Similar topics
- Computer Science
- Computer Communications (Networks)
- Internet