1

I have data that are non-normal and (strongly) negative skewed. The data also have high kurtosis and outliers. There appears to be a variety of options for transformation, but I cannot find a source that helps me determine the best option (or why to choose one over the other). In this case, should I

  1. reflect and transform

  2. perform an exponential transformation

  3. look into Box-Cox?

And how do I know which one is the right choice for my data?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Dawn
  • 53
  • 1
  • 5
  • 5
    Why do you want the data to be normal? What are you going to do with the data? – Peter Flom Jul 10 '13 at 18:07
  • 1
    Dawn, @Peter's comment is spot on. If you would like a sense of the scope of this subject, the thread at http://stats.stackexchange.com/questions/298/ covers the same ground (albeit with a narrower focus on the possibility of a log transform). Reading over it first may help you refocus your question. – whuber Jul 10 '13 at 18:27
  • I have survey responses that are all non-normal with negative skew. Essentially, for almost every question in the survey, respondents tended to agree with the statement provided. – Dawn Jul 11 '13 at 19:13
  • Whoops, it wouldnt let me edit. More than 10% of my data are outliers (about 25 out of 225 respondents). I've tried doing some basic EFA's and CFAs (using SPSS and AMOS) on established scales, items are not loading on the expected factors and loading are all very low. In AMOS, I'm getting an inadmissible solution (due to not positive definite covariance matrix). – Dawn Jul 11 '13 at 19:30

0 Answers0