search

iDiv-sCom / jSDM-Help

A starting point for empirical scientists to ask questions about jSDM analyses
https://idiv-scom.github.io/jSDM-Help/
0 stars 0 forks source link

How to calculate p-values for environmental predictors in s-jSDM? #2

Open CaiWang0503 opened 4 years ago

CaiWang0503 commented 4 years ago

I'm confused about how to know the environmental factors are significant when I use sjsdm. Anybody know how to do that？

When using HMSC i can estimate the environment variables by variance partitioning.

florianhartig commented 4 years ago

Hi Hi CaiWang0503,

it depends what you mean by "significant". HMSC is a Bayesian software and thus calculates posterior distributions, not p-values. The typical way to interpret posteriors is in terms of their mean and their central quantiles (=credible interval), e.g. display the central 90% of the posterior. I would recommend that.

The closes thing to a p-value is possibly the certainty about the sign. We do this, for example, in Fornoff, F., Klein, A. M., Hartig, F., Benadi, G., Venjakob, C., Schaefer, H. M., & Ebeling, A. (2017). Functional flower traits and their diversity drive pollinator visitation. Oikos, 126(7), 1020-1030.

Comments regarding variation partitioning see below.

MaximilianPi commented 4 years ago

Hi CaiWang0503, in sjSDM() you can set se=TRUE to get standard errors (and p-values) or use getSe(model) to calculate the standard errors and p-values post-hoc. But please not that if you use regularization (lambda >0.0 parameter in env = linear()), the standard errors and p-values are not exact.

CaiWang0503 commented 4 years ago

hi MaximilianPi, I got the values by getSe(model)-----model with regularization, so the only values can use is that "Estimate" ?

CaiWang0503 commented 4 years ago

I just saw a new page explaining how to use the new features, very useful, thanks. I will try importance and anova. Thank you all.

florianhartig commented 4 years ago

Hi guys, sorry, I just realised that I misunderstood the question, I somehow thought you ask about HMSC.

As said by Max, for s-jSDM, single predictors uncertainties / p-values can be extracted via se = T, and for entire modules (= Environment), there is the experimental VP / anova feature, but CaiWang0503, note that there could be changes in the functions in the next days, because we are still working on this.

CaiWang0503 commented 4 years ago

Thanks florianhartig, I will wait for the update.