How can I transform a variable (non linear transformation) such that its values are more evenly spread, that is reduce the peak in the middle of the histogram and move more into tails?
Asked
Active
Viewed 1,408 times
3
-
2The methods described [here](http://stats.stackexchange.com/q/10975) give one approach out of the many possible. – whuber Jul 01 '11 at 16:00
-
Thanks... I'll use arctan(x) to stay within 0 an 1 (Probabilistic framework)... Skewed probabilities I should say – user333 Jul 01 '11 at 20:29
1 Answers
3
The transform that most evens things out is the rank transform (just replace the data by the ranks). If there are no ties then the result is uniform.
If data is fairly normal (bell shaped) then the inverse of the normal distribution will spread towards uniform. Actually any s-shaped curve will tend to do this including the arctangent and inverse logit (center and choose an appropriate scale first).
Greg Snow
- 46,563
- 2
- 90
- 159