0

If you had two models - one with a better fit, and one with predictors of higher significance - which one would you choose?

Keep in mind that both models are considered a good fit. One has a 3% better predictive capabilities. And also all the same predictors are significant in both models, but in one model (with a slightly worse fit) they are slightly more significant.

IvLi
  • 631
  • 1
  • 8
  • 12
  • 4
    What do you want to do with the model? Do you want to use it to explain some variable? Or do you want to produce forecasts of the variable? – soakley Jun 19 '13 at 00:18
  • 4
    It depends on what the model is *for*. If the point is to predict, you'd care more about predictive accuracy. If the point is something else, you may not care about slightly better predictive performance. – Glen_b Jun 19 '13 at 00:23
  • 1
    Please note that a good fit to the training data and predictive ability are not at all the same thing. Your question seems to suggest it. – Glen_b Jun 19 '13 at 00:51
  • Thx for the that info! My main aim is to predict, so by your comment I guess I am going to choose the model with variables of greater significance. Right? I will use the model to test a few hypothesis. – IvLi Jun 19 '13 at 14:24
  • Have a look at this question for a better understanding of data properties that impact the two measures: https://stats.stackexchange.com/questions/369554/significance-vs-goodness-of-fit-in-regression – DotPi May 11 '20 at 12:54

0 Answers0