LOG transformation of Percentage Independent variable

[Arindam Pal]

Hi,

In my regression I have used a ratio term (in %) as the independent variable. That value ranges from 0.0001% to 99.95%. Can I take the log of this variable to normalize the results.

Thanks in advance

[Nick Cox]

You can take the logarithm; it is perfectly well defined for such a range, as zero and negative values are not apparent. Whether the result will be nearer a normal or Gaussian in shape depends on the data. In principle, a bounded variable cannot be Gaussian. But also in principle, and more importantly, there is no assumption that an independent variable [that's what you said; presumably it's what you meant] should be Gaussian in regression. That's not even an assumption about the dependent variable. So, don't feel obliged to worry about transforming that variable unless it is a good idea on other grounds.

Although it is far from dead, better alternatives to the traditional terminology of dependent and independent variables have been available for at least 50 years. You can choose from response, outcome, ... or predictor, covariate, explanatory variables, and yet others. One of several objections to dependent and independent as terms is that people quite often get them the wrong way round, which may have happened here.

(Side note: You are resisting the suggestion that you use a full real name in this forum. That's your prerogative, but it does diminish your chances of getting a reply. Some people ignore posters not using full real names, as a matter of personal principle; others partially discriminate against such. These are personal choices, but be informed that they exist.)