Dear Statalist users,
I am trying to fit a linear regression model (Stata 15.0) with 5 different exposure variables (in the same model) and some covariates. The model looks like this:
Code:
-> regress PCT_5mC logAs logHg logMn logPb logCd i.Sex i.Parity i.Maternal_edu

      Source |       SS           df       MS      Number of obs   =       631
-------------+----------------------------------   F(8, 622)       =      4.84
       Model |  .357290237         8   .04466128   Prob > F        =    0.0000
    Residual |  5.74385813       622  .009234499   R-squared       =    0.0586
-------------+----------------------------------   Adj R-squared   =    0.0465
       Total |  6.10114836       630  .009684362   Root MSE        =     .0961

------------------------------------------------------------------------------
     PCT_5mC |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       logAs |   .0106746   .0049339     2.16   0.031     .0009856    .0203637
       logHg |  -.0142802   .0071945    -1.98   0.048    -.0284086   -.0001519
       logMn |  -.0002981    .009837    -0.03   0.976    -.0196158    .0190196
       logPb |  -.0097966   .0091545    -1.07   0.285     -.027774    .0081809
       logCd |   .0087327   .0055965     1.56   0.119    -.0022577    .0197231
             |
         Sex |
        Boy  |  -.0409371    .007703    -5.31   0.000     -.056064   -.0258101
             |
      Parity |
         1+  |   .0167926   .0124899     1.34   0.179    -.0077349    .0413201
             |
Maternal_edu |
 =<12 years  |   .0078305   .0091661     0.85   0.393    -.0101698    .0258307
       _cons |   3.703645   .0315254   117.48   0.000     3.641736    3.765554
------------------------------------------------------------------------------
As you can see, both exposure variables logAs and logHg are statistically significant. However, these two variables are correlated (Spearman's rank=0.59). So I wondered whether this correlation would cause any trouble. The VIF statistics does not look too bad, I think:
Code:
    Variable |       VIF       1/VIF 
-------------+----------------------
       logAs |      1.54    0.647819
       logHg |      1.59    0.628908
       logMn |      1.02    0.976892
       logPb |      1.14    0.880986
       logCd |      1.11    0.899054
       1.Sex |      1.01    0.988661
    1.Parity |      1.02    0.975679
1.Maternal~u |      1.05    0.948319
-------------+----------------------
    Mean VIF |      1.19
However, removing the logHg variable, I get the following results for logAs:
Code:
-> regress PCT_5mC logAs logMn logPb logCd i.Sex i.Parity i.Maternal_edu

      Source |       SS           df       MS      Number of obs   =       631
-------------+----------------------------------   F(7, 623)       =      4.94
       Model |  .320908086         7  .045844012   Prob > F        =    0.0000
    Residual |  5.78024028       623  .009278074   R-squared       =    0.0526
-------------+----------------------------------   Adj R-squared   =    0.0420
       Total |  6.10114836       630  .009684362   Root MSE        =    .09632

------------------------------------------------------------------------------
     PCT_5mC |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       logAs |   .0051144   .0040711     1.26   0.209    -.0028803     .013109
       logMn |   .0009909   .0098387     0.10   0.920    -.0183301    .0203118
       logPb |  -.0125702   .0090685    -1.39   0.166    -.0303787    .0052384
       logCd |   .0086075   .0056094     1.53   0.125     -.002408     .019623
             |
         Sex |
        Boy  |  -.0400819    .007709    -5.20   0.000    -.0552207   -.0249431
             |
      Parity |
         1+  |   .0158463   .0125102     1.27   0.206     -.008721    .0404136
             |
Maternal_edu |
 =<12 years  |   .0060554   .0091439     0.66   0.508    -.0119011     .024012
       _cons |   3.706495   .0315669   117.42   0.000     3.644505    3.768486
------------------------------------------------------------------------------
The coef for logAs has been reduced, and it is no longer statistically significant.

Should I be concerned, or can I trust my findings, regarding logAs and logHg?


Best,
Kjell