1
$\begingroup$

I did a stepwise regrssion analysis to predict energy expenditure using the variables, height, weight, age, gender and energy intake. The final model contains the variables gender and weight. Now does this final model take into account gender by weight interaction? Or do I have to construct a new equation from this final model, that will accomodate interaction?

$\endgroup$
5
$\begingroup$

Stepwise regression has a number of bad qualities (just look e.g. here or here and this is one of them. With only 5 variables (plus one interaction), unless your sample size is small you can include all the IVs. Why would you do this?

1) It is clear, even to a non-expert, that all your potential IVs are included for strong theoretical reasons. Finding that one of them was not related to energy expenditure would be remarkable. In fact, it would be more remarkable than finding that all of them were related.

2) Even if a variable is not significant and even if its coefficient is small, it might affect the relationship between the DV and other IVs.

$\endgroup$
  • $\begingroup$ Thank you for your reply. I used step wise regression to prove the argument that weight predicts most of the variance in total energy expenditure. So are you saying that a simple multiple regression would do? If so, what about the interaction factor? $\endgroup$ – user46697 Jun 23 '14 at 12:13
  • 1
    $\begingroup$ What I would do, in your situation, supposing that there was no issue with overfitting, is run one multiple regression with all 5 IVs and one interaction, which seems to be what you hypothesized. $\endgroup$ – Peter Flom Jun 23 '14 at 12:15

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.