Hello,
I have a data set of morphological measurements of the same character from 4 different species (all different sample sizes). Just looking at the data in boxplot format, I can see that all of the data have huge overlaps of their interquartile ranges and their means are all relatively close to one another. Also, if I conduct multiple t-test comparisons between the different groups, I get high p-values, agreeing with my visual observation that no, the measurements do not significantly vary among species. HOWEVER, when I do what I believe I'm supposed to do and run an ANOVA I get a P-value of zero, or I get a super super tiny p-value (1.3e-26), suggesting that at least one of the groups is different. Whats going on here? What is the appropriate statistical procedure that I should do? I just want to do the correct statistics to show what is pretty obvious, this morphological variable does not vary significantly among species.
Thanks
Kate
Because they are testing different hypotheses. The ANOVA F test is an omnibus test. Best wishes, D. Booth
Hello Kathryn. Please provide the sample size (N), mean and SD for each group. That will allow interested readers to see if they can duplicate your findings, and may provide some insight into what is going on.
It is a little hard to imagine why you are getting those results. You may try to run a post hoc to see where the differences are. If the overall omnibus is significant, a post hoc would show which two (or more) means are different. You are right to question your results as it does not make sense for there to be large differences in p-values. Of course they will be different because they are testing different hypothesis and power is different, but going from very large to essentially 0 indicates that something may be off.
You are correct to run an ANOVA for testing difference among multiple groups as opposed to many individual t-test, which inflates type I error. I think looking at the post hoc test will be the quickest way to see what is going on between groups. Best of luck!
Dear Kathryn , what do you need to compare regarding parameters : x` or variances. ANOVA is to compare between 3 groups or more and it checks statistical difference between variances with no respect to location of mean (x`). While T test is to compare between 2 groups and it checks statistical difference between x`. So as doctor David said : they are testing different hypotheses. And doctor Bruce said : what is the sample size. Please put in consideration if you will choose parametric or non-parametric statistical test.
Hi Greg, I think that giving the formulae to compute the F are not the same as those to compute the student t, that is not surprising that for some samples they don't behave exactly the same way relative to their 0.05 threshold values (< or >).
Anyway go for permutation, it's non parametric and super simple.
Best to Zelics
Julien Lagarde why are you answering to me lol? But agreed, permutation - the way to go :)
Kathryn, as mentioned by others, the difference in sample size is most likely at the root of the difference you get here: the variance of all 'groups' will be taken into account in the ANOVA, not in the paired t-tests
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size