1

I have a set of independent variables and one dependent variable. I am performing regression analysis and k-means with those variables and I am wondering the following:

1) After reading this explanation:

Can I mix different data transformations in the same model?

I am still wondering if it is all right to apply different transformations to different independent variables (for example,. square one var, perform 1/x on the other, etc.) in order to achieve normality. And could you please give a brief explanation or example of WHY is that legitimate or non-legitimate thing to do?

2) If independent variables are transformed, do I have to transform dependent variable as well? My dependent variable is a composite variable which is sum of z-scores of few other variables.

3) Also, do I have to calculate z-scores from that dependent variable, even though it's composed of sum of individual z-scores? In other words, if individual standardised features are summed, is the resulting sum still standardised?

4) Is normality of independent and dependent variable a necessity for: a) linear regression b) k-means? I came across different opinions, so I am unsure what to think.

Cheers

Community
  • 1
Luigi_zg
  • 11
  • 1

1 Answers1

0

In terms of regression and transformations you might want to review the following question and read the responses . When (and why) should you take the log of a distribution (of numbers)? . Hope this helps ....

Community
  • 1
IrishStat
  • 27,906
  • 5
  • 29
  • 55