0
$\begingroup$

Data transformations like logarithmic as a part of preprocessing during model preparation inherently changes the data. Doesn't that mean we are making model with entirely new data?

$\endgroup$
  • 1
    $\begingroup$ Suppose you report acidity of a solution as pH and another chemist reports it as a concentration of hydrogen ions: are you working with the same or "entirely new" data? I believe most people would say the data are the same (ignoring any quibbles concerning the meaning of pH and its relationship to the H ion concentration). What will differ are the implicit assumptions in your models that are invoked to analyze and understand variation. Transformations--particularly nonlinear ones--can have profound effects on those implicit assumptions. $\endgroup$ – whuber Oct 16 '18 at 16:27
1
$\begingroup$

I don't think it makes new data, it just changes the way to interpret and perceive the result such as log transformation it is used when we want to change the scale of our model from the collection unit to percentage or it is used in the case that we want to fit some exponential function data with linear model by using log to transform it.

$\endgroup$
-1
$\begingroup$

Transformations When (and why) should you take the log of a distribution (of numbers)? are sometimes taken when a theory is to be tested empirically .. otherwise transformations are like drugs ..some are good for and some not. Knowledgeable statisticians know that tests of significance are all about the distribution pf the residuals from a model NOT the original data thus in the absence of a theory one should stay from any and all transformations (differencing, percentages et al ) unless warranted/suggested by model residuals.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.