Summary

We often express the binomial count data as proportions, but it’s usually more informative to keep them in count form. Here,the numbers of successes and failures sum to give the binomial denominator, which we can use to give the larger binomial trials more weight in the analysis. With GLMs for binomial count data, the mean is modeled on the logit scale using the symmetric S-shaped logistic curve (although other choices of link function are available), while the binomial variance is largest at intermediate proportions and declines toward both zero and one.