0

I am building a logistic regression model and one of my independent variable sis heavily skewed. Is it better to use the ln or the log10? Why?

And how to correct the skewness of a variable that contains a lot of zero values?

DroppingOff
  • 537
  • 1
  • 4
  • 12
  • 3
    Since changing the base of the logarithm only multiplies by a constant, it changes the scale, not the shape -- so it makes no difference at all which you choose. If you have a high proportion of values at some single value, you can't separate them by a transformation, so generally you can't do much. – Glen_b Jul 28 '14 at 12:42
  • Why would you want to change the skewness of an IV? – Glen_b Jul 28 '14 at 12:47
  • Isn't it better that all the variables are as normal as possible? – DroppingOff Jul 28 '14 at 12:57
  • The logarithm of zero is indeterminate in any case. – Nick Cox Jul 28 '14 at 13:26
  • This is two questions. I closed the first one as a duplicate. The second one is answered in many threads which can be found by searching our site for keywords zero+skew+regression. – whuber Jul 28 '14 at 13:50
  • Better for what reason? Logistic regression doesn't need the IVs to have a normal distribution. Indeed frequently they're binary. – Glen_b Jul 28 '14 at 14:01

0 Answers0