vif for dummy variables like country and years dummie

[Badar Khalid]

Dear Statalist Users,

I have regression models in which have country and years dummy, when I check for vif, it comes more than 10 especially for country dummies (please see table below):

My question, do I need to account for country and years dummies' vif , and if no, how I can write the code to calculate the vif for the main variables in the model without calculating the vif for years and country dummies. Thank you

. vif

    Variable |       VIF       1/VIF 
-------------+----------------------
          UK |    411.49    0.002430
      Canada |    334.33    0.002991
   RuleofLaw |    237.65    0.004208
   Australia |    207.76    0.004813
      Sweden |    192.97    0.005182
     Finland |    191.48    0.005222
      Norway |    185.83    0.005381
      France |    179.49    0.005571
  Netherland |    151.43    0.006604
     Denmark |    118.09    0.008468
 Switzerland |    110.84    0.009022
       Spain |    100.39    0.009961
     Ireland |    100.05    0.009995
     Belgium |     70.60    0.014164
      Crisis |     11.53    0.086711
  PostCrisis |      9.11    0.109751
WidelyHel~20 |      8.73    0.114535
   Year_2012 |      8.15    0.122679
   Year_2011 |      8.12    0.123100
   Year_2010 |      7.88    0.126908
   Year_2009 |      7.66    0.130602
       Italy |      7.48    0.133657
FamilyCon~20 |      7.01    0.142677
  Industry_5 |      5.39    0.185422
  Industry_1 |      4.54    0.220200
  Industry_6 |      4.32    0.231591
  Industry_3 |      4.32    0.231726
StateCont~20 |      4.25    0.235516
  Industry_2 |      3.66    0.273000
  Industry_9 |      3.38    0.295897
PropertyPl~t |      2.91    0.343936

-------------+----------------------
    Mean VIF |     60.62

.

[Badar Khalid]

Any suggestion please

[Nick Cox]

On this evidence, you shouldn't be trying to code a report that doesn't indicate the problems with the model. You should be trying to fix the model.

I say on this evidence because, naturally, a table of VIF is all that you show us, not even the original model command.

(I suspect that others had similar thoughts, but pulled back from saying something limited, negative and (perhaps) obvious.)

[Badar Khalid]

Thank you for your reply and advice.

Below I show the command and the results as well.Any comments please. Thank you.

. regress BDI_16 L.(IO_Domestic $X $Y $I $C), robust cluster (CountryID)
note: L.Year_2006 omitted because of collinearity
note: L.Year_2012 omitted because of collinearity
note: L.Industry_7 omitted because of collinearity
note: L.Denmark omitted because of collinearity

Linear regression                               Number of obs     =      2,028
F(13, 14)         =          .
Prob > F          =          .
R-squared         =     0.2802
Root MSE          =      1.888

(Std. Err. adjusted for 15 clusters in CountryID)

Robust
BDI_16       Coef.   Std. Err.      t    P>t     [95% Conf. Interval]

IO_Domestic
L1.   -.0094184   .0044187    -2.13   0.051    -.0188956    .0000588
                         
LogTotalAsset
L1.    .3307708   .2527381     1.31   0.212    -.2112985    .8728402
                         
SalesGrowthRate
L1.    .0410934   .2105628     0.20   0.848    -.4105189    .4927057
                         
Leverage
L1.    .7224189   .6769358     1.07   0.304     -.729464    2.174302
                         
Cash
L1.   -.8727844   .9030625    -0.97   0.350    -2.809661    1.064092
                         
CapitalExpenditure
L1.    4.431041   1.413341     3.14   0.007     1.399725    7.462357
                         
MarkettoBookValue
L1.    .0475877   .0333987     1.42   0.176    -.0240454    .1192208
                         
ReturnonAsset
L1.    1.707071   .8869409     1.92   0.075    -.1952282     3.60937
                         
PropertyPlantandEquipment
L1.   -.8316488   .4172822    -1.99   0.066     -1.72663    .0633325
                         
AnalystCoverage
L1.    .0079802   .0120067     0.66   0.517    -.0177715    .0337319
                         
CrossListingDummy
L1.   -.1793173   .2398602    -0.75   0.467    -.6937662    .3351317
                         
RuleofLaw
L1.   -.0482036   .0495183    -0.97   0.347    -.1544096    .0580025
                         
Crisis
L1.    .4732932   .1834144     2.58   0.022     .0799085    .8666778
                         
PostCrisis
L1.    .0295949   .3080075     0.10   0.925    -.6310156    .6902054
                         
FamilyControlling_20
L1.    .9660318   .5111488     1.89   0.080    -.1302733    2.062337
                         
StateControlling_20
L1.    .3578423   .5773058     0.62   0.545    -.8803554     1.59604
                         
WidelyHeldat_20
L1.    .0758965   .3999505     0.19   0.852    -.7819121    .9337051
                         
Year_2006
L1.           0  (omitted)
                         
Year_2007
L1.    .0348819   .0940123     0.37   0.716    -.1667545    .2365183
                         
Year_2008
L1.    .0511911   .1863545     0.27   0.788    -.3484996    .4508818
                         
Year_2009
L1.   -.1675949   .2841458    -0.59   0.565     -.777027    .4418373
                         
Year_2010
L1.    .0202414    .313365     0.06   0.949    -.6518595    .6923424
                         
Year_2011
L1.    .3438802   .3338418     1.03   0.320    -.3721392      1.0599
                         
Year_2012
L1.           0  (omitted)
                         
Industry_1
L1.    -.303481   .5155339    -0.59   0.565    -1.409191    .8022294
                         
Industry_2
L1.    .4187083   .5662959     0.74   0.472    -.7958755    1.633292
                         
Industry_3
L1.    .5670687   .4605605     1.23   0.239    -.4207353    1.554873
                         
Industry_4
L1.   -.2064466   .5066541    -0.41   0.690    -1.293112    .8802182
                         
Industry_5
L1.   -.2500781   .4938907    -0.51   0.620    -1.309368     .809212
                         
Industry_6
L1.   -.6200133   .5523763    -1.12   0.281    -1.804743     .564716
                         
Industry_7
L1.           0  (omitted)
                         
Industry_8
L1.    .2358787   .6828852     0.35   0.735    -1.228764    1.700522
                         
Industry_9
L1.   -.4565187    .571185    -0.80   0.437    -1.681589    .7685514
                         
Australia
L1.   -.2780603   .3148107    -0.88   0.392     -.953262    .3971414
                         
Belgium
L1.   -.3280998    .509116    -0.64   0.530    -1.420045    .7638455
                         
Canada
L1.    .5330154   .2903987     1.84   0.088    -.0898279    1.155859
                         
Denmark
L1.           0  (omitted)
                         
Finland
L1.    .2531874   .2143378     1.18   0.257    -.2065215    .7128962
                         
France
L1.    .7798969    .477594     1.63   0.125    -.2444403    1.804234
                         
India
L1.    -4.27671   2.139008    -2.00   0.065    -8.864426    .3110067
                         
Ireland
L1.     .850713   .2382609     3.57   0.003     .3396943    1.361732
                         
Italy
L1.   -2.714751   1.854488    -1.46   0.165    -6.692231     1.26273
                         
Netherland
L1.    .7508282   .2529876     2.97   0.010     .2082238    1.293433
                         
Norway
L1.    2.066916   .1873225    11.03   0.000     1.665149    2.468683
                         
Spain
L1.   -.5712559   .7228659    -0.79   0.443    -2.121649    .9791373
                         
Sweden
L1.    2.212823   .1969369    11.24   0.000     1.790435     2.63521
                         
Switzerland
L1.    .1526121   .2107167     0.72   0.481    -.2993302    .6045544
                         
UK
L1.    1.084445   .3973656     2.73   0.016     .2321806    1.936709
                         
_cons    11.14169   5.192184     2.15   0.050     .0055651    22.27782


. vif

Variable        VIF       1/VIF 

RuleofLaw
L1.     236.34    0.004231
India
L1.     154.89    0.006456
Italy
L1.     130.11    0.007686
Spain
L1.      25.94    0.038549
UK
L1.      16.31    0.061305
France
L1.      15.85    0.063079
Crisis
L1.      10.38    0.096349
WidelyHel~20
L1.       8.82    0.113348
Belgium
L1.       8.70    0.114979
Year_2011
L1.       8.60    0.116228
Year_2010
L1.       8.19    0.122084
Year_2009
L1.       7.96    0.125601
Canada
L1.       7.42    0.134857
PostCrisis
L1.       7.26    0.137673
FamilyCon~20
L1.       7.05    0.141772
Australia
L1.       6.25    0.160112
Industry_5
L1.       5.55    0.180255
Industry_1
L1.       4.51    0.221547
Industry_6
L1.       4.39    0.227708
Industry_3
L1.       4.35    0.230066
StateCont~20
L1.       4.27    0.234294
Industry_2
L1.       3.70    0.269947
Netherland
L1.       3.50    0.285363
Industry_9
L1.       3.46    0.288724
Ireland
L1.       3.46    0.289197
Sweden
L1.       3.13    0.319937
PropertyPl~t
L1.       2.97    0.336200
Finland
L1.       2.97    0.337139
Norway
L1.       2.93    0.340816
Industry_8
L1.       2.86    0.349720
Industry_4
L1.       2.76    0.362388
Switzerland
L1.       2.75    0.363191
IO_Domestic
L1.       2.73    0.365655
CapitalExp~e
L1.       2.36    0.423463
Year_2008
L1.       2.33    0.428270
LogTotalAs~t
L1.       2.31    0.432587
AnalystCov~e
L1.       2.05    0.487339
Year_2007
L1.       1.80    0.554109
ReturnonAs~t
L1.       1.66    0.601411
MarkettoBo~e
L1.       1.66    0.602514
Leverage
L1.       1.65    0.607435
CrossListi~y
L1.       1.64    0.608804
Cash
L1.       1.38    0.724556
SalesGrowt~e
L1.       1.28    0.781414

Mean VIF      16.83

.

[Bruce Weaver]

I think generalized VIF (GVIF) would be more appropriate for categorical variables. I know there is an R function that can compute it, but I was not able to find anything comparable for Stata with a very cursory search. The closest I came was a conference paper by John Hendrickx and some co-authors and a Stata package called perturb. The conference paper suggests that perturb can compute GVIF--see Table 12. Type findit perturb to install.

Meanwhile, maybe someone else can point us to a newer or better Stata implementation of GVIF.

HTH.

[Badar Khalid]

Thank you Dear Bruce for your valuable guidance. I hope that some body can help me to find out GVIF command in STATA.

[Badar Khalid]

I have installed perturb using the following command : ssc install perturb

Can somebody help to find out the command for executing GVIF please. Thank you

[Nick Cox]

In my view you are focusing on trying a different way of measuring the size of the problem. That's like asking for a different thermometer when the patient clearly is running a high fever.

It would be better to find out what the problem is.

For example, you appear to be using a plain regression model on BDI

-- I don't know what this is, possibly

Beck Depression Inventory
Bee Diseases Insurance
Baltic Dry Index
...

-- you see that Googling can't triumph over my ignorance. But with economic data for several countries (??) some massive values would not be surprising and those could be acting as outliers and be one reason for apparent collinearity. Working on logarithmic scale (e.g.) might help.

A set of VIFs like that point to overfitting, as do many mediocre P-values. You could look at scatter plot matrices, correlation matrices, principal component analysis as ways of investigating the structure of collinearity.

Most important of all would be some economic thinking identifying a plausible simpler model but I can't help you there.