2
$\begingroup$

This question already has an answer here:

I have model where my dependent variable is Total money spend and then I have independent variable Income and some other variables. Is it okay to use log transformation on the dependent variable and one of the independent ones and keep the rest unchanged?

$\endgroup$

marked as duplicate by whuber r Dec 12 '18 at 18:41

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ Yes. You can use a combination of different transformations in one model. There is no reason not to if you find that such log transformations make sense and help with interpretation/modeling assumptions. $\endgroup$ – Heteroskedastic Jim Dec 12 '18 at 13:08
  • 1
    $\begingroup$ @HeteroskedasticJim, post your comment as an answer? $\endgroup$ – Ben Bolker Dec 12 '18 at 13:15
  • $\begingroup$ @BenBolker thought it too simple an answer, but done. $\endgroup$ – Heteroskedastic Jim Dec 12 '18 at 13:34
4
$\begingroup$

Yes. You can use a combination of different transformations in one model. There is no reason not to if you find that such log transformations make sense and help with interpretation or modeling assumptions like linearity.

$\endgroup$
  • 1
    $\begingroup$ A simple example is fitting a power function in a measured predictor $y = ax^b$ with also an indicator variable with values $0$ and $1$. Hence you might work with $\log y = b_1 + b_2 \log x + b_3\ \text{indicator}$, where $b_1= \log a$. Here the indicator does not need transformation and indeed any other transformation could only map $0$ and 1 to two constants, so there is no scope whatever for helpful transformation. $\endgroup$ – Nick Cox Dec 12 '18 at 15:40

Not the answer you're looking for? Browse other questions tagged or ask your own question.