Hello,
I am currently working on the data analysis for my MSc. project comparing probability of occurrence of a species between two different habitats using presence - absence data. Due to the design of the field study I decided to use GLMM with binomial distribution as I have various random effects that need to be accounted for. The model seems to be doing the job, however, the use of GLMM was not really a part of my stats module during my MSc. so I am not really sure how to report the results. I am using lme4 package in R console to analyze my data. Thank you
check http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html for a review of GLMM in general.
In particular, a GLMM is going to give you two parts:
the fixed effects, which are the same as the coefficients returned by GLM
the random effects, which -- assuming you didn't get into random slopes -- will act as additive terms to the linear predictor in the GLM. In this case, the random effect is to be added to the log odds ratio. Personally, I change the random effect (and it's 95% CI) into odds ratios via the exponential. Getting them is a bit annoying. Use the 'arm' package to get the se.ranef function. The 'sjPlot' is also useful, and you can extract the ggplot elements from the output.
The random effects are important in that you get an idea of how much spread there is among the individual components. You might, depending on what the confidence intervals look like, be able to say something about whether any terms are statistically distinct.
The ICC (random effect variance vs overall variance) isn't as easily interpretable as that from a linear mixed model.
I always recommend looking at other papers in your field to find examples. There is no accepted method for reporting the results. You could check my own pubs for examples; for example, my paper titled "Outcome Probability versus Magnitude" shows one method I've used, but my method varies depending on the journal. Good luck!
check http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html for a review of GLMM in general.
In particular, a GLMM is going to give you two parts:
the fixed effects, which are the same as the coefficients returned by GLM
the random effects, which -- assuming you didn't get into random slopes -- will act as additive terms to the linear predictor in the GLM. In this case, the random effect is to be added to the log odds ratio. Personally, I change the random effect (and it's 95% CI) into odds ratios via the exponential. Getting them is a bit annoying. Use the 'arm' package to get the se.ranef function. The 'sjPlot' is also useful, and you can extract the ggplot elements from the output.
The random effects are important in that you get an idea of how much spread there is among the individual components. You might, depending on what the confidence intervals look like, be able to say something about whether any terms are statistically distinct.
The ICC (random effect variance vs overall variance) isn't as easily interpretable as that from a linear mixed model.
I am not sure whether you are looking at an observational ecology study. i guess you have looked at the assumptions and how they apply.
The assessment of the random effects and the use of lme4 in r will give you some fixed effects output and some random. The random outputs are variances, which can be reported with their confidence intervals. I found a nice site that assist in looking at various models. if you have more than two independent variables of interest in the logistic model- you may have to look at choosing the appropriate model. the parsimonious model can be chosen. Model comparison is examine used Anova(mod1,mod1) . I think Anova is from the car package.. Where the mod1 and mod2 are the objects from fitting nested models in the lme4 framework.
This site is nice for assisting with model comparison and checking:
https://onlinecourses.science.psu.edu/stat504/node/157
Check chapter 7.4 in my book draft:
https://www.researchgate.net/project/Book-New-statistics-for-the-design-researcher
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics