VIF error

[Ashvinder Singh]

Hello everyone,

I'm new to Stata so I'm still figuring out the ropes. I have the following problem.

I am on Stata 14, and I am trying to get the VIF for some X variables. First, I get an error r(199) - "command VIF not defined by VIF.ado". So, I try again and I get another error, this time r(301) - "not appropriate after regress, nocons; use option uncentered to get uncentered VIFs".

I am doing a fixed effects regression ("areg"). Can someone help?

Thanks.

Ashvinder

[Carlo Lazzaro]

Ashvinder:
welocme to the list.
-estat vif- is not allowed after -areg-.
However, you can go -estat vce, corr- as an alternative.
The following toy-example covers that issue comparing -are- with -regress-:

. use http://www.stata-press.com/data/r14/auto2
(1978 Automobile Data)

. areg mpg weight gear_ratio, absorb(rep78)

Linear regression, absorbing indicators         Number of obs     =         69
                                                F(   2,     62)   =      41.64
                                                Prob > F          =     0.0000
                                                R-squared         =     0.6734
                                                Adj R-squared     =     0.6418
                                                Root MSE          =     3.5109

------------------------------------------------------------------------------
         mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      weight |  -.0051031   .0009206    -5.54   0.000    -.0069433    -.003263
  gear_ratio |    .901478   1.565552     0.58   0.567    -2.228015    4.030971
       _cons |   34.05889   7.056383     4.83   0.000     19.95338     48.1644
-------------+----------------------------------------------------------------
       rep78 |          F(4, 62) =      1.117   0.356           (5 categories)

. estat vif
estat vif not valid
r(321);

. estat vce, corr

Correlation matrix of coefficients of areg model

        e(V) |   weight  gear_r~o     _cons
-------------+------------------------------
      weight |   1.0000                    
  gear_ratio |   0.7544    1.0000          
       _cons |  -0.8976   -0.9639    1.0000

. reg mpg weight gear_ratio

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(2, 71)        =     66.38
       Model |  1592.05392         2  796.026958   Prob > F        =    0.0000
    Residual |  851.405544        71  11.9916274   R-squared       =    0.6516
-------------+----------------------------------   Adj R-squared   =    0.6417
       Total |  2443.45946        73  33.4720474   Root MSE        =    3.4629

------------------------------------------------------------------------------
         mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      weight |  -.0059643   .0008013    -7.44   0.000    -.0075621   -.0043665
  gear_ratio |   .0994882   1.364894     0.07   0.942    -2.622033     2.82101
       _cons |   39.00643   6.169955     6.32   0.000     26.70389    51.30898
------------------------------------------------------------------------------

. estat vce, corr

Correlation matrix of coefficients of regress model

        e(V) |   weight  gear_r~o     _cons
-------------+------------------------------
      weight |   1.0000                    
  gear_ratio |   0.7593    1.0000          
       _cons |  -0.8985   -0.9647    1.0000

. estat vif

    Variable |       VIF       1/VIF 
-------------+----------------------
  gear_ratio |      2.36    0.423527
      weight |      2.36    0.423527
-------------+----------------------
    Mean VIF |      2.36

[Ashvinder Singh]

Dear Carlo,

Many thanks for your wishes and for the illustration. This solves it - however, just to check, when I run -estat vce, corr- or -estat vif- , somehow my y variable is also included whereas in your illustration above, it does not appear:

1. Is there another line in the code?
2. Does the inclusion of the y variable affects the correlation between the x variables?
3. If I change the y variable, will the correlation between the x variables change as well?

Thank you.

Ashvinder

[Bruce Weaver]

VIF and tolerance concern the explanatory variables only. So one solution to your problem would be to use Phil Ender's collin program. Type findit collin to find and install it. Then, using a slight modification of Carlo's example:

. clear

. use http://www.stata-press.com/data/r14/auto2
(1978 Automobile Data)

. areg mpg weight length gear_ratio, absorb(rep78)

Linear regression, absorbing indicators         Number of obs     =         69
                                                F(   3,     61)   =      30.71
                                                Prob > F          =     0.0000
                                                R-squared         =     0.6952
                                                Adj R-squared     =     0.6602
                                                Root MSE          =     3.4197

------------------------------------------------------------------------------
         mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      weight |  -.0014644   .0019614    -0.75   0.458    -.0053865    .0024576
      length |  -.1234458   .0591825    -2.09   0.041    -.2417885   -.0051032
  gear_ratio |   1.383746   1.542316     0.90   0.373    -1.700307    4.467798
       _cons |   44.82346   8.594935     5.22   0.000     27.63683    62.01009
-------------+----------------------------------------------------------------
       rep78 |          F(4, 61) =      1.400   0.245           (5 categories)

. collin weight length gear_ratio if e(sample)
(obs=69)

  Collinearity Diagnostics

                        SQRT                   R-
  Variable      VIF     VIF    Tolerance    Squared
----------------------------------------------------
    weight     12.75    3.57    0.0784      0.9216
    length     10.02    3.17    0.0998      0.9002
gear_ratio      2.72    1.65    0.3682      0.6318
----------------------------------------------------
  Mean VIF      8.50

                           Cond
        Eigenval          Index
---------------------------------
    1     3.9172          1.0000
    2     0.0793          7.0290
    3     0.0026         38.8199
    4     0.0010         64.1869
---------------------------------
 Condition Number        64.1869
 Eigenvalues & Cond Index computed from scaled raw sscp (w/ intercept)
 Det(correlation matrix)    0.0374

The if e(sample) you see there ensures that collin uses the same cases (or the same estimation sample) as the preceding areg command.

HTH.

[Ashvinder Singh]

Dear Bruce - thank you so much for the clarification. It's helped me a lot!

Ashvinder

[Sybil Wang]

Dear Carlo,

I am confused about the command "-estat vce, corr-" you mentioned here. I noticed that the correlation matrix are different when this command and "corr" are used. What is the defference between them? Thanks for your help!

Ashvinder:
welocme to the list.
-estat vif- is not allowed after -areg-.
However, you can go -estat vce, corr- as an alternative.

[/CODE]

Best regards,
Sybil

[Carlo Lazzaro]

Sybil:
welcome to the list.
Whereas -corr- gives you back the correlation between variables, -estat vce, corr- gives you back the correlation between regression coefficients. That's why they are different.

[Mallikarjuna Tupakula]

Hi Carlo Lazzaro I am new to stata
How to store results of vif and corr, vce in excel

[Carlo Lazzaro]

Mallikarjuna:
provided that I never challenged myself with such a task, you may want to take a look at: https://www.statalist.org/forums/for...-vifs-to-excel