Fixed effect with ivregress or ivreg2 command.

Xiaoke Ye

Join Date: Jan 2019

Posts: 33
#1

Fixed effect with ivregress or ivreg2 command.

07 Feb 2019, 00:59

Quick question, is it true that in order to use fixed effect with ivregress or ivreg2, I can ONLY put dummy variables into my main equation?
Because I must use

Code:

weakivtest

which is a postestimation command ONLY with either ivregress or ivreg2.
But I have more than 3k groups. Does that mean there is no way to put cross-section fixed effect with the two commands?
I have tried to modify

Code:

set maxvar

and

Code:

set matsize

, none of them worked.
Tags: None

Carlo Lazzaro

Join Date: Apr 2014
Posts: 17613

07 Feb 2019, 01:33

Xiaoke:
in order to use -fvvarlist- notation with the user-written programme -ivreg2-, you should prefix your code with -xi:-:

Code:

. webuse hsng2
. ivregress 2sls rent pcturban (hsngval = faminc i.region)

Instrumental variables (2SLS) regression          Number of obs   =         50
                                                  Wald chi2(2)    =      90.76
                                                  Prob > chi2     =     0.0000
                                                  R-squared       =     0.5989
                                                  Root MSE        =     22.166

------------------------------------------------------------------------------
        rent |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     hsngval |   .0022398   .0003284     6.82   0.000     .0015961    .0028836
    pcturban |    .081516   .2987652     0.27   0.785     -.504053     .667085
       _cons |   120.7065   15.22839     7.93   0.000     90.85942    150.5536
------------------------------------------------------------------------------
Instrumented:  hsngval
Instruments:   pcturban faminc 2.region 3.region 4.region

. xi: ivreg2 rent pcturban (hsngval  = faminc i.region)
i.region          _Iregion_1-4        (naturally coded; _Iregion_1 omitted)

IV (2SLS) estimation
--------------------

Estimates efficient for homoskedasticity only
Statistics consistent for homoskedasticity only

                                                      Number of obs =       50
                                                      F(  2,    47) =    42.66
                                                      Prob > F      =   0.0000
Total (centered) SS     =     61243.12                Centered R2   =   0.5989
Total (uncentered) SS   =      2816856                Uncentered R2 =   0.9913
Residual SS             =  24565.71669                Root MSE      =    22.17

------------------------------------------------------------------------------
        rent |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     hsngval |   .0022398   .0003284     6.82   0.000     .0015961    .0028836
    pcturban |    .081516   .2987652     0.27   0.785     -.504053     .667085
       _cons |   120.7065   15.22839     7.93   0.000     90.85942    150.5536
------------------------------------------------------------------------------
Underidentification test (Anderson canon. corr. LM statistic):          27.364
                                                   Chi-sq(4) P-val =    0.0000
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic):               13.298
Stock-Yogo weak ID test critical values:  5% maximal IV relative bias    16.85
                                         10% maximal IV relative bias    10.27
                                         20% maximal IV relative bias     6.71
                                         30% maximal IV relative bias     5.34
                                         10% maximal IV size             24.58
                                         15% maximal IV size             13.96
                                         20% maximal IV size             10.26
                                         25% maximal IV size              8.31
Source: Stock-Yogo (2005).  Reproduced by permission.
------------------------------------------------------------------------------
Sargan statistic (overidentification test of all instruments):          11.288
                                                   Chi-sq(3) P-val =    0.0103
------------------------------------------------------------------------------
Instrumented:         hsngval
Included instruments: pcturban
Excluded instruments: faminc _Iregion_2 _Iregion_3 _Iregion_4
------------------------------------------------------------------------------

. weakivtest
(obs=50)

Montiel-Pflueger robust weak instrument test
--------------------------------------------
Effective F statistic:       13.298
Confidence level alpha:          5%
--------------------------------------------

--------------------------------------------
Critical Values             TSLS      LIML
--------------------------------------------
% of Worst Case Bias
tau=5%                    16.720    10.231
tau=10%                   10.231     6.701
tau=20%                    6.701     4.749
tau=30%                    5.421     4.035
--------------------------------------------

.

Never heard about 3000 dummies to be included in a regression model: can't you go for a more parsimonious model?
As an aside, as per FAQ, please note that you're requested (for good reasons) to report the source where non-official Stata commands come from. Thanks.

Kind regards,
Carlo
(StataNow 18.5)

Comment

Xiaoke Ye

Join Date: Jan 2019
Posts: 33

07 Feb 2019, 02:31

Originally posted by Carlo Lazzaro View Post

Xiaoke:
in order to use -fvvarlist- notation with the user-written programme -ivreg2-, you should prefix your code with -xi:-:

Code:

. webuse hsng2
. ivregress 2sls rent pcturban (hsngval = faminc i.region)

Instrumental variables (2SLS) regression Number of obs = 50
Wald chi2(2) = 90.76
Prob > chi2 = 0.0000
R-squared = 0.5989
Root MSE = 22.166

------------------------------------------------------------------------------
rent | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
hsngval | .0022398 .0003284 6.82 0.000 .0015961 .0028836
pcturban | .081516 .2987652 0.27 0.785 -.504053 .667085
_cons | 120.7065 15.22839 7.93 0.000 90.85942 150.5536
------------------------------------------------------------------------------
Instrumented: hsngval
Instruments: pcturban faminc 2.region 3.region 4.region

. xi: ivreg2 rent pcturban (hsngval = faminc i.region)
i.region _Iregion_1-4 (naturally coded; _Iregion_1 omitted)

IV (2SLS) estimation
--------------------

Estimates efficient for homoskedasticity only
Statistics consistent for homoskedasticity only

Number of obs = 50
F( 2, 47) = 42.66
Prob > F = 0.0000
Total (centered) SS = 61243.12 Centered R2 = 0.5989
Total (uncentered) SS = 2816856 Uncentered R2 = 0.9913
Residual SS = 24565.71669 Root MSE = 22.17

------------------------------------------------------------------------------
rent | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
hsngval | .0022398 .0003284 6.82 0.000 .0015961 .0028836
pcturban | .081516 .2987652 0.27 0.785 -.504053 .667085
_cons | 120.7065 15.22839 7.93 0.000 90.85942 150.5536
------------------------------------------------------------------------------
Underidentification test (Anderson canon. corr. LM statistic): 27.364
Chi-sq(4) P-val = 0.0000
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic): 13.298
Stock-Yogo weak ID test critical values: 5% maximal IV relative bias 16.85
10% maximal IV relative bias 10.27
20% maximal IV relative bias 6.71
30% maximal IV relative bias 5.34
10% maximal IV size 24.58
15% maximal IV size 13.96
20% maximal IV size 10.26
25% maximal IV size 8.31
Source: Stock-Yogo (2005). Reproduced by permission.
------------------------------------------------------------------------------
Sargan statistic (overidentification test of all instruments): 11.288
Chi-sq(3) P-val = 0.0103
------------------------------------------------------------------------------
Instrumented: hsngval
Included instruments: pcturban
Excluded instruments: faminc _Iregion_2 _Iregion_3 _Iregion_4
------------------------------------------------------------------------------

. weakivtest
(obs=50)

Montiel-Pflueger robust weak instrument test
--------------------------------------------
Effective F statistic: 13.298
Confidence level alpha: 5%
--------------------------------------------

--------------------------------------------
Critical Values TSLS LIML
--------------------------------------------
% of Worst Case Bias
tau=5% 16.720 10.231
tau=10% 10.231 6.701
tau=20% 6.701 4.749
tau=30% 5.421 4.035
--------------------------------------------

.

Dear Mr Lazzaro

Thanks for your help and I appreciate it. However, I am still confused, with

Code:

 xi:ivreg2

, can I simply use an option such as "fe" to include fixed effect at cross-section dimension?
My command is the following:

Code:

 xi: ivreg2 cta lntotal_assets lntotal_assetsquar cfta LnNoE  salegrowth leverage PriGdp markcapGdp Growth (chinapeindDE = chinapeindGR) i.year  i.bvid1, first vce(cluster nace)

As you can see, I tried to use two-way fixed effect at both time and cross-section level. However, I have more than 3000 groups, if I simply use i.bvid1 to create a dummy for each groups, it is not very feasible.
In other words, can I still include fixed effect with cross-section group without using dummy variable approach with xi:ivreg2

Last edited by Xiaoke Ye; 07 Feb 2019, 02:37.

Comment

Anil Rup

Join Date: Sep 2016

Posts: 13
#4

29 Nov 2020, 09:52

Hi Carlos,

This may be a dumb question but I was wondering when this command: ivregress 2sls rent pcturban (hsngval = faminc i.region)

is implemented, does that take into account that the fixed effects (i.region) are included in both stages? If not what would be the command to include FE in both stages.

I appreciate an answer. Thank you.

Anil
Comment

Carlo Lazzaro

Join Date: Apr 2014
Posts: 17613

29 Nov 2020, 10:53

Anil:
you cannot iobtain coefficient for -i.region- in both stages of 2SLS if this regressor is an instrument:

Code:

use https://www.stata-press.com/data/r16/hsng
. ivregress 2sls rent pcturban (hsngval = faminc i.region), first

First-stage regressions
-----------------------

                                                Number of obs     =         50
                                                F(   5,     44)   =      19.66
                                                Prob > F          =     0.0000
                                                R-squared         =     0.6908
                                                Adj R-squared     =     0.6557
                                                Root MSE          =  9253.4821

------------------------------------------------------------------------------
     hsngval |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    pcturban |   182.2201   115.0167     1.58   0.120    -49.58092    414.0211
      faminc |   2.731324   .6818931     4.01   0.000     1.357058    4.105589
             |
      region |
    N Cntrl  |  -5095.038   4122.112    -1.24   0.223    -13402.61    3212.533
      South  |   -1778.05   4072.691    -0.44   0.665    -9986.019    6429.919
       West  |   13413.79   4048.141     3.31   0.002     5255.296    21572.28
             |
       _cons |  -18671.87   11995.48    -1.56   0.127    -42847.17    5503.439
------------------------------------------------------------------------------


Instrumental variables (2SLS) regression          Number of obs   =         50
                                                  Wald chi2(2)    =      90.76
                                                  Prob > chi2     =     0.0000
                                                  R-squared       =     0.5989
                                                  Root MSE        =     22.166

------------------------------------------------------------------------------
        rent |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     hsngval |   .0022398   .0003284     6.82   0.000     .0015961    .0028836
    pcturban |    .081516   .2987652     0.27   0.785     -.504053     .667085
       _cons |   120.7065   15.22839     7.93   0.000     90.85942    150.5536
------------------------------------------------------------------------------
Instrumented:  hsngval
Instruments:   pcturban faminc 2.region 3.region 4.region
.

Kind regards,
Carlo
(StataNow 18.5)

Comment

Anil Rup

Join Date: Sep 2016

Posts: 13
#6

30 Nov 2020, 09:00

Thank you, Carlos. If I do the following instead : vregress 2sls rent pcturban (hsngval = faminc) i.region, first

I can recover fixed regions effects but the coefficient of the instrumented changes (bigger). In my specific case, the size of the coefficient becomes outrageously large. I was wondering your insight on this. Thanks.

ivregress 2sls rent pcturban (hsngval = faminc) i.region, first

First-stage regressions
-----------------------

Number of obs = 50
F( 5, 44) = 19.66
Prob > F = 0.0000
R-squared = 0.6908
Adj R-squared = 0.6557
Root MSE = 9253.4821

------------------------------------------------------------------------------
hsngval | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
pcturban | 182.2201 115.0167 1.58 0.120 -49.58092 414.0211
|
region |
N Cntrl | -5095.038 4122.112 -1.24 0.223 -13402.61 3212.533
South | -1778.05 4072.691 -0.44 0.665 -9986.019 6429.919
West | 13413.79 4048.141 3.31 0.002 5255.296 21572.28
|
faminc | 2.731324 .6818931 4.01 0.000 1.357058 4.105589
_cons | -18671.87 11995.48 -1.56 0.127 -42847.17 5503.439
------------------------------------------------------------------------------

Instrumental variables (2SLS) regression Number of obs = 50
Wald chi2(5) = 51.68
Prob > chi2 = 0.0000
R-squared = 0.2079
Root MSE = 31.147

------------------------------------------------------------------------------
rent | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
hsngval | .0038683 .0008403 4.60 0.000 .0022212 .0055153
pcturban | -.4980121 .4859064 -1.02 0.305 -1.450371 .4543469
|
region |
N Cntrl | 1.528672 14.20338 0.11 0.914 -26.30944 29.36679
South | 7.74279 14.17355 0.55 0.585 -20.03686 35.52244
West | -40.61235 18.3958 -2.21 0.027 -76.66746 -4.557233
|
_cons | 88.26681 29.7293 2.97 0.003 29.99845 146.5352
------------------------------------------------------------------------------
Instrumented: hsngval
Instruments: pcturban 2.region 3.region 4.region faminc
Comment

Announcement