Synopsis: this checklist provides a decision tree to get to the right type of analysis
Vocabulary: You are typically interested in how something varies when you change other things.
- We call this “something” the “response”
- We call the other things “treatments” if you have manipulated the conditions in an experiment. We call all other variables that may vary across observations “predictors”.
Questions 1:
Is the response you are interested in univariate, multivariate, or something else?
-
Examples for a univariate response is single measurement such as plant biomass, count data like the number of eggs, binary data such as dead / not dead, etc.
-
Examples for a multivariate response may be the number of individuals of different species on each site.
-
Examples for something else may be a spatial pattern, a time series, or other things
Depeding on your answer, go to sections about univariate, multivariat or other analysis
Univariate analysis
OK, we have a situation where a 1-dimensional variable of interest (response) varies across our observations. To know what we have to do, we have to look now at the conditions in which the observations differ, i.e. the treatments or predictors.
- Observations differ only in discrete (non-continous) variables
- The observation differs in only one discret variable / treatments
- There are two different levels / groups / treatments for this variable, and I want to know whether the response differs between these levels –> t-test
- There are more than two discrete groups (treatments), and I want to know whether the response differs between groups –> a number of options, see XXX 3) The observations differ in se
- Observation differ in continous variables
Additional considerations:
- Oservations are not all statistically independentn, but there is reason to think (e.g. through spatial placement)