I am attempting to finish my stats on my Master's thesis and am having difficulties with which statistics test I should be using, and how to go about doing so. I have collected chl-a samples at 15 time intervals, for two separate conditions (sometimes I also have three). My sample sizes are unequal, and my data also fails normality tests. Does anyone have any idea which test if a two-way ANOVA would be the appropriate test for my case? I know that I would have to take into considering that my data is unbalanced, does anyone have some R script examples where they have done this before? I prefer to use SPSS but find it a bit limiting at times so I also use R.
$\begingroup$
$\endgroup$
4
-
1$\begingroup$ Anova is very robust to unbalanced designs, non-normality is a concern. Does your data meet the assumption of homogeneity of variances? Can you provide limited data lines so we can see what your data structure looks like? $\endgroup$– OliverFishCodeCommented Mar 8, 2019 at 23:05
-
3$\begingroup$ The assumption is the residuals are normally distributed not the "data". Plus it sounds like you have multiple observations on the same subject so you're going to need some type of repeated measures design if this is the case. Sample sizes not being equal is usually not an issue. $\endgroup$– GlenCommented Mar 8, 2019 at 23:26
-
1$\begingroup$ True on the residuals, but it doesn't mean the guassian distribution is a good fit for the data. chlorophyll A, which is what they are measuring can often be heavily skewed when measured through time. The gamma distribution may be a better fit. $\endgroup$– OliverFishCodeCommented Mar 9, 2019 at 2:58
-
1$\begingroup$ Looks like a good place for semiparametric ordinal regression models that do not assume a particular data distribution. See here for resources. $\endgroup$– Frank HarrellCommented Nov 25, 2023 at 13:09
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
The GENLINMIXED procedure for generalized linear mixed models in SPSS should be able to handle these data when you decide what distribution is appropriate, including unbalanced data and repeated measures.
