1

I have different results from a correlation table and a multiple regression model. I know that it is an effect of multicollinearity because correlations up to $.474$ exist between predictors, but this is normal in the context of my research area and I cannot remove or change any predictor.

Now I want to provide information on which predictors affect dependent variable and how (positively / negatively). So what is more accurate here, correlation or multiple regression?

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
Saleh Ahmadov
  • 13
  • 4

1 Answers1

1

There is no such thing as which is "more accurate" in the abstract. The answer depends on what question you are asking of these data. The correlations tell you about the relationship between each predictor and the dependent variable when you ignore the other variables, and the multiple regression tells you about the relationships when you are controlling for the other variables. They are both correct answers, but to different questions. It may help you to read my answer here: Is there a difference between 'controlling for' and 'ignoring' other variables in multiple regression?

Community
  • 1
gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
  • I want to know how each of these 6 predictor variables influence 1 dependent variable? All of them are Likert scales. – Saleh Ahmadov Aug 01 '15 at 16:56
  • It just depends on whether you want to know how they influence the DV together or separately. – gung - Reinstate Monica Aug 01 '15 at 17:13
  • so when you control for other variables, like in multiple regression model, it basically tells you how each of them influence on DV separately, while pairwise correlation cannot (because of confounding variables I guess)? Please tell me that I got it right. – Saleh Ahmadov Aug 01 '15 at 17:34
  • Multiple regression tells you how each variable influences the DV when controlling for the other variables. It may help to read my linked answer. – gung - Reinstate Monica Aug 01 '15 at 18:46