1

I'm given a model of the form $E(\ln Y|X)=X\beta$.

Should I call it a linear model or loglinear model?

I'm assuming $Y$ is log-normally distributed.

Tim
  • 108,699
  • 20
  • 212
  • 390
GRS
  • 187
  • 2
  • 11

1 Answers1

4

$ E(\ln Y|X)=X\beta $ is a linear model of transformed variable $\ln Y$, more precisely, it is linear in parameters. As I noted in the comment, you can introduce additional variable $Z = \ln Y$ and then $ E(Z|X)=X\beta $ is just an ordinary regression model. You can make all kinds of transformations to your variables, but when talking about linearity of the model, we have in mind the relationship between variables.

Community
  • 1
Tim
  • 108,699
  • 20
  • 212
  • 390
  • I've got 2 models, one of the form $E(Y|X)$ and the other $E(\ln Y|X)$ where in both models I investigate $Y$. I was refering to $E(\ln Y|X)$ as the loglinear model, is this incorrect? – GRS Nov 24 '16 at 12:01
  • I can see that, given the transformation I can call it a linear model, but without the transformation, it is a loglinear one? – GRS Nov 24 '16 at 12:02
  • Both are linear for the reasons given above. "$\ln$" is not a part of your model, it's a transformation of the variable. – Tim Nov 24 '16 at 12:13
  • @GRS Check http://stats.stackexchange.com/questions/43930/choosing-between-lm-and-glm-for-a-log-transformed-response-variable to see the difference. – Tim Nov 24 '16 at 12:15
  • thanks, could you comment on this http://www.kenbenoit.net/courses/ME104/logmodels2.pdf ? It seems to classify them as loglinear – GRS Nov 24 '16 at 12:22
  • @GRS check http://stats.stackexchange.com/questions/86720/log-linear-regression-vs-logistic-regression – Tim Nov 24 '16 at 12:28