multicollinearity.

[hussein bataineh]

Hi

Please i want to know if the if i have a multicollinearity problem as i read that VIF need to be bellow 10to decide whether we have multicollinearity or not ( well below the suggested threshold of 10 for the risk of multicollinearity (Cohen, Cohen, West, & Aiken, 2003). when i run VIF i found two variable have high VIF .these two are control variable and i don't need to drop any of them as both are important . meanly these high collinearity is between dividends_w and KZ_w .

my question is 8.27 ad 6.81 are acceptable in research .

. vif 
           
    Variable |       VIF       1/VIF  
-------------+----------------------
 dividends_w |      8.27    0.120855
        KZ_w |      6.81    0.146884
ROA_four_s~w |      4.38    0.228517
FREE_CASH_~w |      4.32    0.231710
  Leverage_w |      3.44    0.290465
 Firm_Size_w |      3.17    0.315829
Platn_Size_w |      2.78    0.359668
Sustaina~e_w |      2.16    0.462509
Book_to_Ma~w |      2.03    0.493062
Fund_Statu~w |      1.70    0.586539
CSR_Commit~e |      1.61    0.621333
       UNION |      1.43    0.697416
   UNDERFUND |      1.36    0.736990
Contr~sion_w |      1.28    0.783570
changes_di~w |      1.25    0.801855
Governance~w |      1.24    0.809712
CASh_FLOW_~w |      1.22    0.821751
csopresence1 |      1.22    0.822189
    DCAPEX_w |      1.15    0.870170
ACTUAL_RET~w |      1.06    0.941865
  TAX_rate_w |      1.03    0.968479
-------------+----------------------
    Mean VIF |      2.52

[Clyde Schechter]

I have some better reading for you. Find a copy of Arthur Goldberger's textbook A Course in Econometrics. He has an entire chapter devoted to the highly overrated issue of multicolinearity which will explain to you in a more entertaining and detailed way what I am about to say.

The only impact that multicolinearity has on a regression analysis is to increase the standard errors of the coefficients of the variables that are involved in the multicolinearity. You have already identified that those variables are not the key explanatory variables in your regression: they are covariates added to adjust for possible confounding effects. So you may, as a result of their correlation) get imprecise estimates of their coefficients. But, by definition, since they are not the key variables of the model, there is no reason to care about this. It is not part of your research goals to estimate their effects on the outcome--if it were, you wouldn't be calling them "control" variables. So even if their VIFs were in the hundreds, this would not be a problem. Run your regression as it is, and then move on. Don't waste another second thinking about this.