0

Question: Can we use unstandardized coefficients derived from linear regression to remove the effect of an independent variable on the dependent variable?

I have a large dataset and suspect that an (unwanted) independent variable (IV) influences the hundreds of dependent values (DV) under research. All variables are ratio/interval. I need to remove this effect before continuing further analysis and would like to create a new, corrected estimated dataset. My idea has been to calculate the regression coefficient through linear regression, between the IV and each of the DV. If the effect of the IV (X) on DV (Y) turns out to be significant, consequently, I will calculate a new estimated Y that subtracts the regression coefficient multiplied by the IV value in to a corrected estimated Y value for the new dataset. 

Y^new = Y^old - bX

Y = dependent variable
X = independent variable
b = unstandardized regression coefficient

Is this method appropriate? What should I do for IV-DV that are not significantly correlated?

Chame Leon
  • 1
  • 1
  • You can get better guidance on cross validated on this question – Gaurav Nov 09 '15 at 06:39
  • Thank you, I didn't know about that resource – Chame Leon Nov 09 '15 at 07:04
  • 1
    No, do not follow this path. Instead, you should include the covariate in all subsequent analyses. Also be warned that ordinary linear least squares regression is not appropriate if your dependent is a ratio. – Roland Nov 09 '15 at 11:55

0 Answers0