Hi all, I'm doing my MSc on the possibility to use marine organisms as a bioindicator for metals. My model is - how tow factors : Time and Space (both explanatory categorical variable) affect metal concentration (Numeric response variable). most of my data seems normal once plotted, nonetheless normality test (Shapiro, KS etc.) point out the data is not normal.
To make a long story short - please, is there a non-parametric test I can use for this model? I know a similar question was raised here in the past and yet, I couldn't get a straight answer. Kruskal is good for one factor, so please try to avoid talking about it unless you disagree.
Thank you, appreciate all attempts to answer,
Ran
Dear Ran Tal ,
I don't think there is one. I think your only choices are to try to use a parametric test with a robust exception to its assumptions or to use two one way nonparametric tests, but this would not account for interaction effects.
I agree with Peter Samuels conclusion for references and code see :
https://www.google.com/search?q=robust+regression+r+tutorial&rlz=1C1CHBF_enUS917US917&oq=Robust+regression+R&aqs=chrome.3.0l7j69i61.25748j1j4&sourceid=chrome&ie=UTF-8
and the attached paper. Best, David Booth
Hi, thank you all so much for your answers and attachments.
Peter Samuels & David Eugene Booth Indeed I thought of this option, so it wiil be Kruskal wallis right? and I read that once I'm doing parametric test , interaction might be useless because non-parametric test uses ranks and not real data, but that's night be wrong. What do you think?
Ömer Sari I did checked the skewness, in the first experiment it ranges between 0.4-0.6, and in the second one it most of the data is around 0.5 . Anyhow, I did plot histogram and QQ on the residuals and it seemed fine...
thanks again, have a lovely day!
Ran
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology