0

I understand you can graph the variable distributions but is there an actual best fit test?

Edit:Sorry about the wording. Basically what I did is run two linear regression models. One containing choice predictors and an outcome and the other with the log transformed outcome. I want to see which of these models gives a better fit to the data. Does that make more sense?

Sara A
  • 31
  • 2
    Variables don't fit data sets. Variables may fit a distribution. What are you trying to do? – Peter Flom Aug 04 '15 at 20:59
  • 1
    The decision to transform the original data depends on the objective. If the original data is all positive and you take the logs, then both sets of values are equivalent in information content. In other words, both "fit the dataset" exactly the same as they are the dataset. However, if you want to know if a transformation fits the assumptions of a particular model better than not transforming, you should consider looking how Box-Cox power transformations (among other possibilities) are assessed. – JimB Aug 04 '15 at 21:10
  • I think R^2 or coefficient of determination is what you're looking for and yes there is a fit test called "lack of fit test" – YouTah Aug 04 '15 at 21:12
  • And I'm not pushing Box-Cox transformations. It's just that if you study those, you'll see how different transformations are assessed. – JimB Aug 04 '15 at 21:13

1 Answers1

-1

I guess you mean does it fit another variable better. You can use a simple correlation test for this.

Dirk N
  • 283