1
# VIF
        PI     BI 
    1.68431 1.68431

# cor.test
Pearson's product-moment correlation

data:  df$PI and df$BI
t = 42.464, df = 2635, p-value < 2.2e-16
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 0.6141774 0.6595292
sample estimates:

     cor 
0.637405

It seems like there might not be multicollinearity according to VIF, even though the predictors are correlated.

Moreover, in the regression, they have negative coefficients at each model on the dependent variable, but if they are in one regression model, BI turns as positive.

Is there multicollinearity?

POC
  • 346
  • 1
  • 8
  • 23
Nayeon
  • 27
  • 4

1 Answers1

1

Multicollinearity problem is related to an excessive correlation of two predictors. In this cases, it does not seem excessive according to the VIF.

For the flipping sign of the effect, you probably have confounding or suppressor effect in your model. I recommend checking here, for further details.

POC
  • 346
  • 1
  • 8
  • 23
  • Although they have positive correlation, they could be suppressors each other? I'll check. Thank you – Nayeon Mar 05 '21 at 00:03
  • 1
    Yes, it can. It's not their correlation that is most important here for these effect, but their relations to the dependent variables also. – POC Mar 05 '21 at 00:26