Dear Pascale,

before performing the analysis, it is necessary to check the distribution of variables. You can calculate this by Shapiro-Wilk test or the Kolmogorow-Smirnow test. If the parameters are normally distributed, you can choose the t-test for analysis the both elephant groups. If the parameters are not normally distributed, you have to take the Mann-Whitney-U test.

Good luck for your study!

Best

Tanja

Pascale,

The way you question is stated it's hard to tell if you have enough data to perform a statistical test. How large is your sample size? How many n per Elephant population per season?

If the question(s) you'd like to get at is: Does the home range (m2 / ha) vary between Elephant herd 1 and Elephant herd 2, do home ranges of Elephant populations vary over seasons, and do home ranges between Elephant population 1 and Elephant population 2 vary over seasons? Then you are really in need of a two-wayANOVA (or related test) not just a one-way ANOVA. With the two-way you can test the hypotheses, (1) main effect of Elephant population, (2) main effect of season, and (3) the interaction of Elephant population and season.

I'm going to disagree with Tanja, in that the assumptions of normality should be examined on the residuals of a general linear test–not just the raw distributions. There's suggestion that other, possibly better, ways to check normality assumptions exist rather than distribution shape tests (i.e., residual plots, simulation). Nevertheless, using hypothesis tests to check your assumptions is better than ignoring assumptions altogether. I'd be sure to also check for homogeneity of variance, i.e., Bartlett's test. Independence is a little trickier... hopefully you have a good experimental design, but you might think about temporal correlation. Take a look at a related discussion here: https://stats.stackexchange.com/questions/6350/anova-assumption-normality-normal-distribution-of-residuals

If you feel the assumptions for classical ANOVA aren't reasonably met, you can use a randomization/permutation/bootstrap approach or a nonparametric approach.

With a little more detail of your data, I and others may be able to offer better or more precise advice.

Best,

Caleb

if your data noramality distributed and the samples are homogentiy, and you have two samples only then you can use two independent T-Test. But if no, then you have to use nonparametric tests

Dear all,

in generall, data should be tested concerning the distribution type and after this, simple tests like the t-test or Mann-Whitney-U test is appropriate to analyze differences between to groups. Of course, after analyzing with simple tests more sophisticated test - for example multivariate analylizes - can be conducted. But I disagree with Caleb - his recommendations do not meet the daily practice in performing statistical analyses.

Best

Tanja