Effect Size in fitglme

Hey, I am using a model which I have estimated with fitglme. I want to report effect sizes and confidence intervals. Would the best way be to report OR terms and their respective intervals?

I am calculating them the following way:

[beta,~,stats] = fixedEffects(glme_best);
CI = coefCI(glme_best);
names = glme_best.CoefficientNames(:);
OR = exp(beta);
CI_OR = exp(CI);

Is that correct or is there a better way to directly get OR terms in Matlab?

Best!

[Torsten]

We must know more about your model in order to understand why you take "exp" of beta and CI.

[Laurie König]

opts = {'Distribution','Binomial','Link','logit','FitMethod','Laplace'};

formula_fix = 'Correct ~ 1 + Trial_z';

mA = fitglme(T, [formula_fix, ' + (1|Subj)'], opts{:}); --> That would be an example for a model

[William Rose]

@Laurie König,

It would help me to see some sample data to understand this better. For example, I am interetsted to know if Trial_z is a continuous variable or categorical, and if categorical, is it ordinal, i.e. ordered?

It appears to me, from your original question and your response to @Torsten, that Correct is a binomially-distributed (i.e. 0 or 1) variable which you are predicting. You use a linear mixed-effects model with a logit link. The logit link function is appropriate, since Correct has a binomial distribution. The linear model has an intercept, a fixed-effect slope for the variable "Trial_z", and random intercepts for the random effect "Subj".

when you do

[beta,~,stats] = fixedEffects(glme_best);

you will extract the slope associated with the fixed effect "Trial_z", and related statistics.

Is my understanding above correct?

Since the link function is the logistic function (see below), it does not seem useful to compute exp(beta).

where .

If the variable being predicted ("Correct") had a Poisson distribution, then a log link function would be used. In this case, I suppose one might be interested in exp(beta). But that does not seem to be the case here.