I have two groups (unpaired) of plant biomass measurements and want to know whether there is a test that can be used to determine statistical significance where assumptions are not violated. The values are independent but not normally distributed (nor are variances) due to several outliers and a large number of zeroes (~1/3 of the dataset).
I have run a t-test and wilcoxon rank test which both indicate a significant difference at the 99% CI, and just wanted to check there isn't a better way.
I would recommend a zero inflated poission regression. This model allow an excess of zero counts. T-test does not sound appropiate, the way you describe the data.
Poisson is likely too strikt on the variance. I bet the data will be over-disperesed. I would consider a quasi-Poisson or a Gamma model (both are available with the possibility to handle zero-inflation).
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero