Log tranformed dependent: PROC Mixed / NLMIXED?

[mragaa]

Hi everyone,

  I'm sorry in advance if this is a dumb question. Say I'm using panel data to model log(Speed) as the response and Turbo as a binary predictor (0= No, 1= Yes). My question:

1- Is Proc Mixed usable here, or do I have to use PROC NLMIXED?

2-  Would the results be valid and interpretable as they are in cross-sectional data (i.e. we get x% in speed for Turbo)?

SAS Version: 9.4

Thank you.

[Rick_SAS · Accepted solution]

1. You can use PROC MIXED and other linear regression procedures if you first use the DATA step to define logSpeed = log(speed).  Then use logSpeed as the response variable.

2. I am not sure what you are asking, but perhaps it has to do with the interpretation of the parameter estimates for Turbo. If you model

log(Speed) = Intercept + A*Turbo then 

Speed = exp(Intercept)*exp(A*Turbo)

Thus when Turbo=0,  the predicted speed is exp(Intercept).

When Turbo=1,  the predicted speed is exp(Intercept)*exp(A).

So the model predicts that using Turbo results in MULTIPLYING the speed by exp(A).

[mragaa]

Thanks a lot.