I have a variable having Z score of skewness nearly 5, After transformation through Inverse Distribution Function is SPSS it reduced to .07 i.e acceptable range. Now, can i run regression and correlation same as if could apply before transformation. Is there different strategy to apply these tests?
Yes, the analysis proceeds as if you had no transformation. Interpreting the output may become more difficult. Be careful in the discussion. A=1/B will result in "1/B" decreasing as B increases given B1, and "1/B" increasing as B decreases for -1>B>1.
The following are ok:
1) Report the raw means (=not transformed) and standard deviations, and report the statistical results from the transformed data. One could report raw means and 95% confidence intervals.
2) Report the LSMeans and standard errors of the transformed data along with the result from the statistical model using the transformed data.
The following are bad:
1) Calculate the LSMeans and SE, back-transform them into the original units and report the statistics from the transformed data.
2) Calculate LSMeans and SE using raw data, and report the statistics from the transformed data.
Do not transform
Article TO DETERMINE SKEWNESS, MEAN AND DEVIATION WITH A NEW APPROAC...
See page 55 of the book linked below. You can use unstandardized or the z-score transformed for regression etc.
http://hbanaszak.mjr.uw.edu.pl/TempTxt/Timothy%20Keith-Multiple%20Regression%20and%20Beyond_%20An%20Introduction%20to%20Multiple%20Regression%20and%20Structural%20Equation%20Modeling-Routledge%20(2014).pdf
Dear Muhammad
Please, read in the following link of researchgate, I hope will be useful for you
https://www.researchgate.net/post/Why_do_we_do_transformation_before_data_analysis
Good Luck
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS