I'm researching microplastic abundance and distribution in sandy beaches. I compared microplastic abundance between the intertidal and supralittoral zones and also between 4 beaches. My data is not normal, so I used non-parametric K-W test (+ Dunn's multiple comparison poshoc) to compare beaches and then conducted 4 Mann-Whitney U tests to compare zones in each beach. From my point of view, these were appropriate because I can easily see differences between zones in the four beaches individually. However, my revisor said it would be better to conduct a 2-factor GLM (factor 1 being zone, factor 2 being beach).
What does he mean by GLM? General lineal model? or GeneralizED lineal model? It would be my first time running this kind of test due to my lack of expertise. In fact, I was thinking about keeping my K-W and Mann-Whitney tests and argue that conducting these will give a better overview of the microplastic distribution of the beaches individually. In my results, as the microplastic abundance can vary a lot from beach to beach (for example, beach 1 having a mean abundance of 10 and beach 2 having a mean concnetration of 200), I think the robustness of the test is not my major concern. Can anyone give me some recommendations?
GLM stands for Generalized Linear Model which is a flexible generalization of the general linear model (which assumes a normal distribution of the residuals). What ever model you fitted, make sure to test the corresponding model assumption are met. Here is a link which explains the GLM concept in an easy way:
http://www.statsoft.com/textbook/generalized-linear-models
When you run the GLM model with the two factors as your independent variables, it allows you to explore the relationship between the outcome variable and one of the independent variables while also controlling for the effect of the other independent variable on the outcome variable. May be your adviser is thinking of this.
Hello Gabriel,
A model that includes the main effect of beach (4 levels), the main effect of zone (2 levels), and beach x zone interaction would potentially do a better job of explaining differences in your DV measure than just looking at what differences you see among beaches alone then zone differences for individual beaches. I suspect that was the reason for the reviewer's suggestion.
Of course, you can always forgo a reviewer's suggestion, but you'll have to convince the editor (or whomever is making the final, action decision) that your approach is sound.
You could always look at whether a data transformation might make the data better conform to the normality assumption (specifically, that model residuals or "errors" are normally distributed). As well, you could use exact/resampling/bootstrap tests instead of parametric tests to avoid the issue with normality.
Good luck with your work.
In SAS parlance, "glm" means "general linear model", whereas in, say, R parlance, it means "generalized linear model", so it's difficult to know which hey mean unless they tell you. It probably doesn't matter to their point, though.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics