I wonder if you can use the typical approach, but with robust standard error estimates?
Hello everyone,
I have data of a variable from 4 sites, with 3 repetitions from each site. Tests revealed that I do have homoscedasticity but no normality. So I chose to use a test of permutations to ANOVA. A few days ago someone told me that I cannot do this test if I have 4 sites and only 3 repetitions. My questions are:
1. Is this true? and if so:
2. What other test should I choose?
Thank you very much,
Roy.
I wonder if you can use the typical approach, but with robust standard error estimates?
Stop cowardice, ban guns!
How did you test for normality? Was it on the raw response?
I don't have emotions and sometimes that makes me very sad.
Right. I was asking if you just did the test on the raw responses or were you looking at the residuals from the anova?
I don't have emotions and sometimes that makes me very sad.
Maybe this link can help. Here the number of replicates is also three, but with 24 observation and an extra factor. So I guess that your study is OK.
The test says:
"The data do not have to be normal (and these data certainly are not) or have any other specific distribution. But notice that requiring the errors of observations to be iid brings in the traditional assumption of homogeneity of variance."
Tweet |