You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When defining a GLM, the only situation in which you can write the response variable as a mean plus error is if the response is normally distributed. In the case of other distributions (e.g., binomial, Poisson), the representation $Y_i = \mu_i + \epsilon_i$ is incorrect. For example, if $Y_i$ is binary, then there is no named probability distribution you could put on the error term in order for $Y_i$ to only take on values 0 and 1.
Instead the correct specification of a GLM involves specifying the probability distribution for $Y_i$, e.g., Binomial, Poisson, etc. Then specifying the model equation involving the mean response $\mu_i = E(Y_i)$, as you did in your post: $g(\mu_i) = X_i' \beta$.
The text was updated successfully, but these errors were encountered:
I came across your blog post on GEEs: https://rlbarter.github.io/Practical-Statistics/2017/05/10/generalized-estimating-equations-gee/.
When defining a GLM, the only situation in which you can write the response variable as a mean plus error is if the response is normally distributed. In the case of other distributions (e.g., binomial, Poisson), the representation$Y_i = \mu_i + \epsilon_i$ is incorrect. For example, if $Y_i$ is binary, then there is no named probability distribution you could put on the error term in order for $Y_i$ to only take on values 0 and 1.
Instead the correct specification of a GLM involves specifying the probability distribution for$Y_i$ , e.g., Binomial, Poisson, etc. Then specifying the model equation involving the mean response $\mu_i = E(Y_i)$ , as you did in your post: $g(\mu_i) = X_i' \beta$ .
The text was updated successfully, but these errors were encountered: