My dataset has repeated observations on patients with patient-level costs captured at regular intervals over a period of time. I model how these costs vary over time using time-invariant patient characteristics (xi) and time indicators (quarters). This works with a GLM model with a gamma log link (see 1 below). However, when I use a generalized linear mixed model to account for non-independence of observations by the same patient, again with a gamma log link (see 2 below), I get the following error message:
option family(gamma) invalid;
cost <= 0 (le 0) in some observations
Is meglm not able to model zero costs? I am using Stata 14.1. I would appreciate any comments or references.
(1)
glm cost quarter1-quarter8 x1-x5 , family(gamma) link(log) vce(cluster patient_id) eform
(2)
meglm cost quarter1-quarter8 x1-x5 || patient_id:, family(gamma) link(log) eform
option family(gamma) invalid;
cost <= 0 (le 0) in some observations
Is meglm not able to model zero costs? I am using Stata 14.1. I would appreciate any comments or references.
(1)
glm cost quarter1-quarter8 x1-x5 , family(gamma) link(log) vce(cluster patient_id) eform
(2)
meglm cost quarter1-quarter8 x1-x5 || patient_id:, family(gamma) link(log) eform