Hi,
I am trying to get the P-value associated with a glmer model from the binomial family within package lme4 in R. My model is the following:
glmer(Infection.status~origin+ (1|donationID), family=binomial)->q7H
where Infection status is a dummy variable with two levels, infected and uninfected I tried to get the P-value associated to the the explanatory variable origin but I get only the F-value and the degrees of freedom
(aov
A few comments:
Regarding your data- have you got it in a layout that suits what you are requesting. As above check your model do you need glmer or glm?
In the script, the assign goes from the right and assigns to the object on the left.
q7H
Thanks Caitlin,
I know what you mean, but this is not my question. Of course when I run summary I get the Z-value and P associated to the different levels of the explanatory variable.
However, when I want to check the P associated to the explanatory as a whole I get the value I showed in my question but WITHOUT the Pvalue associated, so I want to know which distribution is following the value. That I am afraid is not a F value distribution but a Chi-squared one. Once I know, I can simply go to a table with the df and check it.
Thanks
Dear Miranda,
thanks, I knew what you mean, but that´s not my question as I replied above to Kaitlin.
Indeed, there are things I am not agree with you. The assign can go any direction, is not relevant. And I think there is no need to have 2 columns to defin a dummy variable (with 2 levels)
lme4 doesn't give p values by default because the test statistics don't necessarily follow a t, F etc. distribution that closely.
The FAQ for GLMM should help with suggestions:
https://bbolker.github.io/mixedmodels-misc/glmmFAQ.html
Personally I nearly always use MCMC methods via rstan or bmrs (which both use lme4 syntax) these days.
For a model fitted with glmer, the summary output gives you already p-values: summary(q7H )
If you follow Thom's link, you'll see that a Wald-z-statistic is computed for the model parameters, which are approximations (and in this case, reflect the worst choice for calculating p-values, but still OK as approximation as long as you don't need to be too exact). The p-values you see are then:
2 * pnorm(abs(), lower.tail = F)
Hi,
I was in contact with Ben Bolker, the author of the lme4 package.
F tests don't really work for GLM(M)s. (F tests are based on the uncertainty of the estimate of the residual variance; typical (binomial & Poisson) GLMs don't estimate a residual variance at all. There is a theory of "Bartlett corrections", which are finite-size corrections for GLMs, but they're not widely used.)
To calculate the P-value associated with a variable we need to create a model without this variable and compare both models as follows using LRT test. The Chi-sqaure and P-value of such comparison is the P-value associated to the variable
Hope it helps
Hi, Mario Garrido
Thank you for your contact with the author of the package. I needed this answer a lot because I'm trying to calculate the value of the random effects and did not know how.
Hi Mario Garrido
I'm using the same function but with more variables:
m1
- Badges
- Science topic
- Similar topics
- Statistics
- Survey