I have a scatter plot where the data are distributed very unevenly.

If you can imagine the ares of the scatter plot divided into 4 quadrants, about half of the data (i.e., half of the data points) like in the bottom-left or southwest quadrant.

Apart from deleting the other half of the data as "outliers", is there a transformation or technique to represent this distribution?

Edit: I am looking for a general rule-of-thumb -- or heuristic if the $10 word makes you feel better -- similar in concept to the type of transformation heuristic that textbooks provide and practitioners use when faced with non-normally distributed data.

The point -- like distribution transformations -- is to get the data to act more linearly, because many parametric statistical techniques are linear "under the hood".

Thanks in advance!

  • $\begingroup$ Hi, Penguin Knight - thanks for your advice. With all due respect, most textbooks give transformations for right-skewed and left-skewed distributions, and distributions that are exponential or inverse exponential, etc. without reference to the type of data, nor to the specific type of analysis. That said, your point re: the type of analysis is well-taken; most of the transformations I mentioned in the previous sentence are intended to make the data act more linearly, since a great many parametric statistical techniques are linear "under the hood". I will update. $\endgroup$ – user2621147 Mar 12 '14 at 14:07
  • $\begingroup$ Penguin_Knight: Your point is taken. However, rules of thumb exist for a reason: as a starting off point and guide for where to start and what can safely be ignored. I am assuming you don't try transformations involving pi and sines/cosines as a first step? $\endgroup$ – user2621147 Mar 12 '14 at 14:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.