Exploratory data analysis and transformations?

Question

Hi,

I'm working on a data set where the outcome is a proportion. The majority of my observations are 0%, and hence I need to use a transformation.

My question is, should I carry out my exploratory analysis (i.e. scatterplots or whatever) on the RAW data or on the TRANSFORMED data? or do you only need to use a transformation when you are using more complex methods such as regression analysis which assumes the normal distribution?

Thanks,

Nicola

Answer 1

No, I'm not a statistician, but I obviously need your help.

Answer 2

Well, I'm not truly a statistician either. However, transformations are usually done for convenience -- e.g. transform a non-linear relationship to a linear one so that one can do linear regression -- not to somehow transform non-normal data to normally distributed.

Linear regression, in particular, doesn't depend on normally distributed data, although things like evaluating significance levels do.

Anyway, the short answer is you do everything on the raw data until such point as a transformation is suggested, and then the transformed data is used. For example, you might typically  do a scatter plot of the raw data, notice an obvious non-linear relationship, transform, do a scatter plot of the transformed data, and proceed with the transformed data.

Answer 3

Bóg psa

Bóg psa