# Testing for normality on un-even sample size generating different levels of significance - SPSS

I'm quite new to stats, and still puzzled on uneven samples testing. I have a database containing quantities of plastic debris found on a beach. I have 15 types of debris (bottles, plastic tubes, caps, lids etc.), but with an uneven number of counts for each, and therefore different quantities. For example bottles have 14 data entries with a total number of 56 bottles found, lids have 5 data entries with a total of 37 lids found, and so on. This is a result of a clean-up effort where each participant reported total quantities of waste picked-up per plastic type, for example one person finding 10 bottles during his clean-up corresponding to one data entry. I was reading this post which answered some of my questions: Statistical tests for uneven groups

I've inserted the data into SPSS, and ran a normality test to understand more about the data. However; I have different levels of significance for each type, which in some cases reject the null hypothesis or not. I was wondering what are the next steps when it comes to data with different levels of significance in the output of the normality test? If there are more non-normal data should I carry on with non-parametric tests? Sorry if this question is not relevant, I've been trying to read more on this but haven't manage to find anything useful.

• The first thing to clarify is why you believe you need to test for normality especially since your values seem to be counts where we usually start from the assumption that they follow a Poisson distribution. Jul 13, 2020 at 14:19
• Because testing for normality would put me in the direction of what test to further use to observe any significant differences between the debris. Jul 13, 2020 at 14:21
• But they are counts. We know the null hypothesis that they are normally distributed is false so it is not logical to test it Jul 13, 2020 at 14:57
• Ok I see, it makes sense actually. I read that non-parametric tests are best suited for count data. I could however log transform the data and then test for normality? Jul 13, 2020 at 15:15