Hi,
I have binomial response data, and wish to do variance or deviance partitioning. First, is it OK to do this? second, if yes, how can be done.
I tried with GLM and GAM from R library Ecospat, but got negative values as well.
I have three groups of data (temperature, precipitation and others). I wish for variance or deviance explained by temperature and precipitation. How can I perform it.
Any clue will be highly appreciated.
Many thanks
Take a look at the last example here:
http://rcompanion.org/rcompanion/e_06.html
It uses a binary response and a single continuous independent variable.
It uses glm in R to build the model.
And then Anova in the car package to produce the analysis of deviance table.
And shows how to get a p-value for the whole model, and a pseudo R-squared for the model.
There is no agreed measure for this. Various pseudo R^2 measures exist but no single measure captures all properties of R^2. In addition R^2 itself is a problematic measure.
The fundamental problem here is that in a binomial distribution variance is a function of the mean so the property of variance explained is also a function of the mean. Within a given range one can approximate an answer and those situations I like the simplicity of correlating the model predictions with the outcome. You could then compare between models with and without predictors. However this is a personal preference - see attached for other options.
Chapter Pseudo-R2 and related measures. Online Supplement 4 to Serio...
Thank, but I am not only looking for R^2, I need to partition them by variables.
I could not find any such examples.
thanks.
If you want an analysis of deviance table like a standard analysis of variance table, then you want to do a logistic regression, and use software that will produce the table. The example I suggested will do this.
Otherwise, I don't understand the question.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS