statistics

[bitterchocolate]

hi,
i was wondering if it was better to use log-transformed data or untransformed data when making comparisons between two groups.
also should we perform a 2-sample t test using log-transformed or untransformed data and why?
thanks..

 

[quark]

Hi,

Are the data skewed? You need a reason for log transformation. There are other types of transformations too.

If you decide to do a transformation, then you should do the hypothesis test using transformed data.

 

[JohnM]

IMHO, I've never been a "fan" of transforming data, because interpreting the results is always problematic (telling a client about the log of the data points will always get you a puzzled look --> not good).

If the data are skewed badly enough, then you may want to consider:

(1) finding probable causes for the "outliers" and removing them, or
(2) using a nonparametric method

However, if there's a really, really good reason, then there are virtually unlimited ways to transform data - just make sure you pick the one that meets your analytical needs....