I am very new to regression analysis. I have a dependent variable and 9 independent variables. I want to build the best possible model from them, and I don't know when to log transfrom a variable.


marked as duplicate by whuber Jan 17 '14 at 22:33

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.


"Best possible model" is not well defined.

Whether you should log transform a variable depends on both statistical and substantive considerations. If the substantive reasons aren't met then it may be best to use a different regression method (but maybe not).

First, you can only take logs of variables that are always positive.

Second, you take logs of a variable when the relationships to that variable are multiplicative rather than additive. For example, money amounts are often logged. If your dependent variable is, say, income, then, in some sense, doubling of income rather than adding a fixed amount is what is important. E.g. if you make \$10,000 and get a \$3,000 raise, it's huge. If you make \$250,000 a year and get a \$3,000 raise, it isn't huge.

Statistically, logs are taken to remove right skewness. But "regular" OLS regression makes assumptions about the distribution of the residuals, not the variables themselves.

  • $\begingroup$ Can you clarify which of your reasons are substantive and which are statistical? $\endgroup$ – skeller88 Nov 27 '17 at 21:12
  • $\begingroup$ The first is a mathematical requirement. The second is substantive. The last is statistical. $\endgroup$ – Peter Flom Nov 27 '17 at 21:34

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