I have read a lot about interpretation of coefficients from Gamma-GLMs (using a log-link function), e. g. from this thread How to interpret parameters in GLM with family=Gamma , and found this to be very helpful.

However, I wonder what is the correct scientific term for the exp(coefficient) from a Gamma-GLM? From my understanding this would describe the ratio between two means, and could therefore be interpreted as a mean-ratio. This should be somewhat similar like when you would compare two odds in a logistic regression and call it an odds-ratio. Is this correct or is there any other more suitable term?


1 Answer 1


There is no scientific term in common usage for the exponential of a GLM coefficient. In some cases exp(coefficient) might represent the log-ratio of two group means, but not always. It depends on the experimental design and on how the GLM model has been parametrized.

In the simplest case that Group is a factor will two levels (A and B say) and a gamma GLM is fit by

fit <- glm(y ~ Group, family=Gamma(link="log"))

then exponentiating the second coefficient, exp(coef(fit)["GroupB"]), would indeed yield the ratio $\mu_B/\mu_A$ where $\mu_A$ and $\mu_B$ are the expected values for groups A and B respectively. In the bioinformatics contexts that I work in, exp(coef(fit))[2] might be called the fold-change and coef(fit)[2] the log-ratio or log-fold-change.

My advice would be to choose terminology appropriate for your specific context.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.