Hi everyone. I am in the process of making RDAs between faunistic and environmental tables. What I don't understand is that depending on the order in which I place my environmental variables (explanatory variables) in the model, the significant variables that emerge are not necessarily the same. Could someone explain why? Is there a way to fix this problem?
Than you very much!
Dear Julien,
Depending on the way you organize your sequence of analysis, this is normal to change. A bit of what you are describing seems to be a mathematical organization that is related to a possible next step for sensitivity analysis. When you change the way you do things, different outcomes are normal to be observed. If you are looking for specific variables, you may have to create an auxiliary "control" mechanism or a "monitoring" tool/mechanism.
In my experience, when I focused on enhancing resilience to the maximum level possible, that maximum result became a constant, and only time, costs/investments were able to serve for the ultimate sensitivity analysis. The inverse such as making funding a constant variable, then resilience and time would change.
Therefore, my suggestion is for you to verify how you are organizing the problem you are trying to model.
The distribution of your species in the abundance matrix will have different interactions with the environmental data matrix. Thus the position ij of the species matrix is variable, while the values of the environmental data matrix are fixed. If you implement the RDA with the same matrixes (species and environmental), the way the environmental variables are ordered in the matrix should not interfere with the significance.
Another reason is the function used to have best model fit.
Before conducting RDA and CCA we first need to assess multicollinearity of variables. Different ways to calculate this exist, such as step fucntion or vif (variance inflation factor) in R software. If you use stepwise, for isntance, terms are subtracted (backwards or forward) and depending on the sequence/order you write terms, you will have different results.
Thanks all for your answers. I have indeed assesed colinearity before performing the RDA using VIF.
I don't use CCA because this method use de Khi² distance, which translate the variation of the compositional profile between objects. Conversly, RDA combine regression and PCA which allows the study of the variation in the level of abundance between objects which is basically my question.
When I performed the complete model :
rda(fauna matrix ~ A + B+ C + D+ E + F)
A and B were significant but not the others.
If i do:
rda(fauna matrix ~ E + F+ A+ B+ C+ D)
it is now E and F that are significant.
However, in order to identify the best explanatory variables, I did a variable selection by creating a null model (i.e. origin only) and then performing a forward selection from the null model and a backward selection from the full model (all variables), simultaneously.
The resulting simplified model is always the same regardless of the initial order of my explanatory variables!
- Badges
- Science topic
- Similar topics
- Biological Science
- Faunistics