I have the problem defining a "generalized linear model" with the lognormal distribution. Simply the glm function does not accept this distribution and I saw the suggestion to use the following command:
glm(y ~ a+b+c, family=gaussian(link="log"))
or
glm(log(y)~ a+b+c, family=gaussian)
Where y has lognormal distribution.
But I wondered if this is the best binding function, as the following message appears from erro glm.fit: algoritmo não convergiu in the first case.
Another option gamlss package, but this one has not explored.