I'm Building a logistic regression model and one of my independent variables is very skewed at zero. How do you suggest that i deal with this situation?
$\begingroup$
$\endgroup$
6
-
$\begingroup$ Why do you think that this is a problem? Is the independent variable in question continuous or categorical? $\endgroup$– EdMJun 8, 2016 at 15:24
-
$\begingroup$ some sample data/ reproducible example would help $\endgroup$– AntoineJun 8, 2016 at 15:27
-
$\begingroup$ It is continuous. 25% of the values are zero. I want to know the best practice od deal with it. Should create a dummy variable indicating zero and non-zero and add it to the model along with the original variable $\endgroup$– IamNotLegendJun 8, 2016 at 15:55
-
$\begingroup$ If i take a log transformation of the data the data looks normally dist wxcept for the spike at 0. Btw I'm taking log(1+x) $\endgroup$– IamNotLegendJun 8, 2016 at 16:33
-
2$\begingroup$ Why would you care whether the log of (1+x) looks normal? $\endgroup$– Glen_bJun 8, 2016 at 17:32
|
Show 1 more comment
1 Answer
$\begingroup$
$\endgroup$
The best practice is to not transform independent variables unless there is a substantive reason for doing so, such as aiding interpretation, dealing with known non-linearity or heteroscedasticity.
There is no distributional requirement or condition for independent variables, so, often, there is no need to do so.