Announcement

Collapse
No announcement yet.
X
Collapse
  • Page of 1
  • Filter
  • Time
  • Show
Clear All
new posts
  • #1

    How to include regional-specific time trends in the FE-IV model using the code ivreghdfe?

    Dear Statalisters:

    I want to control for regional-specific time trends in my FE-IV model (individual fixed effects and standard errors are cluster at individual level), so I tried several ways of coding:

    i).
    Code:
    ivreghdfe y (x=z) c.year##i.region, absorb(pid) cluster(pid)
    ii).
    Code:
    ivreghdfe y (x=z), absorb (c.year##i.region) cluster(pid)
    For the male sample, these two codes report the same results, however, for the student sample they report different results.

    I further used xtivreg, the code is as below:

    Code:
    xtset pid year
xtivreg y (x=z) c.year##i.region, vce(cluster pid) fe
    xtivreg reposts result different from the results from both above two codes.

    Can anyone tell me how I can correctly include regional-specific time trends in the FE-IV model?
    Thank you very much!!
    Last edited by Yanni Shen; 08 Aug 2022, 00:40.
    Tags: FE-IV, fixed effects, time trends
  • #2
    ivreghdfe is from SSC (FAQ Advice #12).

    c.year##i.region
    This includes year as a continuous variable, regional dummies and region-specific time trends and is equivalent to


    Code:
    c.year i.region c.year#i.region
    [.] So the region-specific time trends are given by just

    Code:
    c.year#i.region

    Code:
    ivreghdfe y (x=z) c.year##i.region, absorb(pid) cluster(pid)
    ii).
    Code:
    ivreghdfe y (x=z), absorb(c.year##i.region) cluster(pid)
    For the male sample, these two codes report the same results, however, for the student sample they report different results.

    The models are completely different. In one, you have individual effects captured by absorbing the variable pid whereas in the second, you do not absorb this variable. Only when the omitted variable within -absorb()- is collinear with the included variables will you get the same results.
    • 1 like

    Comment

    • #3
      Andrew Musau Thank you soooo much for your help!!

      Now I get the idea that the code to include region-specific time trends is c.year#i.region rather than c.year##i.provcd!

      ->The models are completely different. In one, you have individual effects captured by absorbing the variable pid whereas in the second, you do not absorb this variable.
      I'm sorry for my mistyping. The second model is
      Code:
      ivreghdfe y (x=z), absorb(pid c.year##i.region) cluster(pid)
      It includes individual fixed effects as in model 1, the only difference is whether put c.year##i.region in the parentheses of absorb or out of the parentheses.
      Because I don't know the difference, so I tried both: for the male sample, these two report the same results, however, for the student sample they report different results.
      I find this weird and don't know how to explain it... It would be really appreciated if you can help me with this, thanks!!

      Comment

      • #4
        I think you need to show the entire Stata output for both estimations. Place these within CODE delimiters.

        Comment

        • #5
          Andrew Musau Thank you so much!

          Results when put c.wave##i.provcd within the parentheses of absorb
          Code:
          ivreghdfe bmi (inttimeg4=bdbd) $ctvar if student==1, absorb(pid c.wave##i.provcd) cluster(pid)
(dropped 836 singleton observations)
(MWFE estimator converged in 49 iterations)

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

Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on pid

Number of clusters (pid) =         299                Number of obs =      622
                                                      F( 14,   298) =     1.90
                                                      Prob > F      =   0.0260
Total (centered) SS     =  351.3946003                Centered R2   =   0.0168
Total (uncentered) SS   =  351.3946003                Uncentered R2 =   0.0168
Residual SS             =  345.4887106                Root MSE      =    .7862

------------------------------------------------------------------------------
             |               Robust
         bmi |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   inttimeg4 |  -.7185779   9.444969    -0.08   0.939    -19.30587    17.86871
       age20 |          0  (omitted)
       age30 |   1.095523   3.851325     0.28   0.776    -6.483718    8.674763
       age40 |   2.379844   4.405836     0.54   0.589    -6.290649    11.05034
        educ |  -.0606307   .3501335    -0.17   0.863    -.7496782    .6284169
     married |  -1.245148   1.223803    -1.02   0.310    -3.653539    1.163242
  familysize |    .063717   .2207936     0.29   0.773    -.3707952    .4982292
       rural |   -.500087   .7608098    -0.66   0.511    -1.997328    .9971537
        work |   .0958174   .9021097     0.11   0.915    -1.679495     1.87113
        linc |  -.0016748   .1090295    -0.02   0.988    -.2162401    .2128904
         GDP |  -.0000801   .0004231    -0.19   0.850    -.0009127    .0007525
    unemploy |  -.7969562   .8894302    -0.90   0.371    -2.547316    .9534037
     finbudg |   .0000895   .0002728     0.33   0.743    -.0004474    .0006264
      over65 |   .1406953   .3412964     0.41   0.680    -.5309612    .8123518
      hltins |   .0000403   .0001281     0.31   0.753    -.0002118    .0002924
------------------------------------------------------------------------------
Underidentification test (Kleibergen-Paap rk LM statistic):              0.295
                                                   Chi-sq(1) P-val =    0.5871
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic):                0.440
                         (Kleibergen-Paap rk Wald F statistic):          0.265
Stock-Yogo weak ID test critical values: 10% maximal IV size             16.38
                                         15% maximal IV size              8.96
                                         20% maximal IV size              6.66
                                         25% maximal IV size              5.53
Source: Stock-Yogo (2005).  Reproduced by permission.
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
------------------------------------------------------------------------------
Warning: estimated covariance matrix of moment conditions not of full rank.
         overidentification statistic not reported, and standard errors and
         model tests should be interpreted with caution.
Possible causes:
         number of clusters insufficient to calculate robust covariance matrix
         singleton dummy variable (dummy with one 1 and N-1 0s or vice versa)
partial option may address problem.
------------------------------------------------------------------------------
Collinearities detected among instruments: 1 instrument(s) dropped
Instrumented:         inttimeg4
Included instruments: age20 age30 age40 educ married familysize rural work linc
                      GDP unemploy finbudg over65 hltins
Excluded instruments: bdbd
Partialled-out:       _cons
                      nb: total SS, model F and R2s are after partialling-out;
                          any small-sample adjustments include partialled-out
                          variables in regressor count K
------------------------------------------------------------------------------

Absorbed degrees of freedom:
-------------------------------------------------------+
   Absorbed FE | Categories  - Redundant  = Num. Coefs |
---------------+---------------------------------------|
           pid |       299         299           0    *|
        provcd |        25           1          24     |
 provcd#c.wave |        25           0          25    ?|
-------------------------------------------------------+
? = number of redundant parameters may be higher
* = FE nested within cluster; treated as redundant for DoF computation

          Results when put c.wave##i.provcd out of the parentheses of absorb
          Code:
          ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave##i.provcd if student==1, absorb(pid ) cluster(pid)
(dropped 834 singleton observations)
Warning - collinearities detected
Vars dropped:       65.provcd#c.wave
(MWFE estimator converged in 1 iterations)

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

Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on pid

Number of clusters (pid) =         300                Number of obs =      624
                                                      F( 60,   299) =  7.0e+06
                                                      Prob > F      =   0.0000
Total (centered) SS     =  409.6982207                Centered R2   =  -0.3873
Total (uncentered) SS   =  409.6982207                Uncentered R2 =  -0.3873
Residual SS             =  568.3777212                Root MSE      =    1.005

---------------------------------------------------------------------------------------------------
                                  |               Robust
                              bmi |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------------------+----------------------------------------------------------------
                        inttimeg4 |  -2.776917   54.59785    -0.05   0.959    -110.2216    104.6678
                            age20 |          0   3.158351     0.00   1.000    -6.215412    6.215412
                            age30 |   1.684808   2.765546     0.61   0.543    -3.757592    7.127208
                            age40 |   3.485824          .        .       .            .           .
                             educ |   .0344442   1.562028     0.02   0.982    -3.039518    3.108406
                          married |   -1.37917   3.685965    -0.37   0.709     -8.63289    5.874551
                       familysize |   .0730965   .2328119     0.31   0.754     -.385061    .5312541
                            rural |  -.6074469   .9691196    -0.63   0.531    -2.514606    1.299712
                             work |  -.1189428   .8542204    -0.14   0.889    -1.799988    1.562103
                             linc |   .0207765   .3733503     0.06   0.956    -.7139508    .7555037
                              GDP |  -.0000893   .0008054    -0.11   0.912    -.0016742    .0014956
                         unemploy |  -.6515151   5.990465    -0.11   0.913    -12.44033     11.1373
                          finbudg |    .000234   .0039475     0.06   0.953    -.0075343    .0080024
                           over65 |   .0582966    .819276     0.07   0.943    -1.553981    1.670574
                           hltins |   .0000541   .0005797     0.09   0.926    -.0010868    .0011949
                             wave |   .1387621    .792673     0.18   0.861    -1.421163    1.698687
                                  |
                           provcd |
                         Beijing  |          0  (empty)
                         Tianjin  |          0  (omitted)
                           Hebei  |          0  (omitted)
                          Shanxi  |          0  (omitted)
                        Liaoning  |          0  (omitted)
                           Jilin  |          0  (omitted)
                    Heilongjiang  |          0  (omitted)
                        Shanghai  |          0  (omitted)
                         Jiangsu  |          0  (omitted)
                        Zhejiang  |          0  (omitted)
                           Anhui  |          0  (omitted)
                          Fujian  |          0  (omitted)
                         Jiangxi  |          0  (omitted)
                        Shandong  |          0  (omitted)
                           Henan  |          0  (omitted)
                           Hubei  |          0  (omitted)
                           Hunan  |          0  (omitted)
                       Guangdong  |          0  (omitted)
Guangxi Zhuang Autonomous Region  |          0  (omitted)
                       Chongqing  |          0  (omitted)
                         Sichuan  |          0  (omitted)
                         Guizhou  |          0  (omitted)
                          Yunnan  |          0  (omitted)
                         Shaanxi  |          0  (omitted)
                           Gansu  |          0  (omitted)
Xinjiang Uygur Autonomous Region  |  -5.053061   9.097552    -0.56   0.579     -22.9564    12.85028
                                  |
                    provcd#c.wave |
                         Beijing  |          0  (empty)
                         Tianjin  |  -.0009207   .0016891    -0.55   0.586    -.0042448    .0024034
                           Hebei  |  -.0022754    .015141    -0.15   0.881    -.0320718     .027521
                          Shanxi  |  -.0017281   .0146236    -0.12   0.906    -.0305064    .0270502
                        Liaoning  |  -.0020004   .0110623    -0.18   0.857    -.0237702    .0197693
                           Jilin  |   .1751903   4.111847     0.04   0.966    -7.916635    8.267016
                    Heilongjiang  |    -.00203   .0030207    -0.67   0.502    -.0079746    .0039145
                        Shanghai  |  -.0009557   .0052536    -0.18   0.856    -.0112944    .0093831
                         Jiangsu  |   4.27e-06   .0132042     0.00   1.000    -.0259807    .0259893
                        Zhejiang  |  -.0019899   .0010138    -1.96   0.051    -.0039849    5.12e-06
                           Anhui  |   -.001627   .0024569    -0.66   0.508     -.006462     .003208
                          Fujian  |  -.0003294    .029397    -0.01   0.991    -.0581806    .0575219
                         Jiangxi  |  -.0016595   .0043595    -0.38   0.704    -.0102387    .0069197
                        Shandong  |   -.001501   .0300378    -0.05   0.960    -.0606132    .0576113
                           Henan  |  -.0016981   .0278318    -0.06   0.951    -.0564692     .053073
                           Hubei  |  -.0018102   .0041874    -0.43   0.666    -.0100507    .0064303
                           Hunan  |  -.0023242   .0054073    -0.43   0.668    -.0129654    .0083169
                       Guangdong  |    .139863   .5158146     0.27   0.786    -.8752238     1.15495
Guangxi Zhuang Autonomous Region  |  -.0017409    .003177    -0.55   0.584     -.007993    .0045112
                       Chongqing  |  -.0005156   .0220691    -0.02   0.981     -.043946    .0429148
                         Sichuan  |  -.0017686   .0272625    -0.06   0.948    -.0554193     .051882
                         Guizhou  |  -.5974209   1.442167    -0.41   0.679    -3.435504    2.240662
                          Yunnan  |   .0539856   1.766206     0.03   0.976    -3.421784    3.529755
                         Shaanxi  |   -.001436   .0082411    -0.17   0.862    -.0176539    .0147819
                           Gansu  |  -.0025182   .0025988    -0.97   0.333    -.0076325    .0025961
Xinjiang Uygur Autonomous Region  |          0  (empty)
---------------------------------------------------------------------------------------------------
Underidentification test (Kleibergen-Paap rk LM statistic):              0.395
                                                   Chi-sq(1) P-val =    0.5299
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic):                0.844
                         (Kleibergen-Paap rk Wald F statistic):          0.352
Stock-Yogo weak ID test critical values: 10% maximal IV size             16.38
                                         15% maximal IV size              8.96
                                         20% maximal IV size              6.66
                                         25% maximal IV size              5.53
Source: Stock-Yogo (2005).  Reproduced by permission.
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
------------------------------------------------------------------------------
Warning: estimated covariance matrix of moment conditions not of full rank.
         overidentification statistic not reported, and standard errors and
         model tests should be interpreted with caution.
Possible causes:
         number of clusters insufficient to calculate robust covariance matrix
         singleton dummy variable (dummy with one 1 and N-1 0s or vice versa)
partial option may address problem.
------------------------------------------------------------------------------
Collinearities detected among instruments: 5 instrument(s) dropped
Instrumented:         inttimeg4
Included instruments: age20 age30 age40 educ married familysize rural work linc
                      GDP unemploy finbudg over65 hltins wave 12.provcd
                      13.provcd 14.provcd 21.provcd 22.provcd 23.provcd
                      31.provcd 32.provcd 33.provcd 34.provcd 35.provcd
                      36.provcd 37.provcd 41.provcd 42.provcd 43.provcd
                      44.provcd 45.provcd 50.provcd 51.provcd 52.provcd
                      53.provcd 61.provcd 62.provcd 65.provcd 12.provcd#c.wave
                      13.provcd#c.wave 14.provcd#c.wave 21.provcd#c.wave
                      22.provcd#c.wave 23.provcd#c.wave 31.provcd#c.wave
                      32.provcd#c.wave 33.provcd#c.wave 34.provcd#c.wave
                      35.provcd#c.wave 36.provcd#c.wave 37.provcd#c.wave
                      41.provcd#c.wave 42.provcd#c.wave 43.provcd#c.wave
                      44.provcd#c.wave 45.provcd#c.wave 50.provcd#c.wave
                      51.provcd#c.wave 52.provcd#c.wave 53.provcd#c.wave
                      61.provcd#c.wave 62.provcd#c.wave
Excluded instruments: bdbd
Partialled-out:       _cons
                      nb: total SS, model F and R2s are after partialling-out;
                          any small-sample adjustments include partialled-out
                          variables in regressor count K
Dropped collinear:    65.provcd#c.wave
------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
         pid |       300         300           0    *|
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

          Comment

          • #6
            You can only absorb indicators (aka dummy variables). In this case, you are absorbing some continuous variables which is not correct.

            c.wave##i.provcd
            is equivalent to

            c.wave i.provcd c.wave#i.provcd
            and therefore you can only absorb "i.provcd" in this instance. Therefore, you should obtain the same results from

            ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave##i.provcd if student==1, absorb(pid) cluster(pid)
            and

            Code:
            ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave c.wave#i.provcd ///
if student==1, absorb(pid i.provcd) cluster(pid)

            Comment

            • #7
              Andrew Musau Thank you so much! It really makes sense to me and I totally agree!

              However, the code you told me
              Code:
               
 ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave c.wave#i.provcd /// if student==1, absorb(pid i.provcd) cluster(pid)
              does not have the same result (see below) as the code
              ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave##i.provcd if student==1, absorb(pid) cluster(pid)
              Instead, it has the same results as the code
              ivreghdfe bmi (inttimeg4=bdbd) $ctvar if student==1, absorb(pid c.wave##i.provcd) cluster(pid)

              Code:
              ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave c.wave#i.provcd if student==1, absorb(pid provcd) cluste
> r(pid)
(dropped 836 singleton observations)
(MWFE estimator converged in 26 iterations)

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

Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on pid

Number of clusters (pid) =         299                Number of obs =      622
                                                      F( 39,   298) =  3.2e+06
                                                      Prob > F      =   0.0000
Total (centered) SS     =  388.4740314                Centered R2   =   0.1107
Total (uncentered) SS   =  388.4740314                Uncentered R2 =   0.1107
Residual SS             =  345.4887088                Root MSE      =    .7862

---------------------------------------------------------------------------------------------------
                                  |               Robust
                              bmi |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------------------+----------------------------------------------------------------
                        inttimeg4 |  -.7185779   9.444973    -0.08   0.939    -19.30587    17.86872
                            age20 |  -2.379844          .        .       .            .           .
                            age30 |  -1.284322   3.851326    -0.33   0.739    -8.863563     6.29492
                            age40 |          0   4.405837     0.00   1.000    -8.670495    8.670495
                             educ |  -.0606307   .3501337    -0.17   0.863    -.7496785    .6284171
                          married |  -1.245148   1.223803    -1.02   0.310    -3.653539    1.163242
                       familysize |    .063717   .2207937     0.29   0.773    -.3707954    .4982295
                            rural |   -.500087   .7608104    -0.66   0.511    -1.997329    .9971549
                             work |   .0958174   .9021101     0.11   0.915    -1.679496    1.871131
                             linc |  -.0016748   .1090295    -0.02   0.988    -.2162402    .2128905
                              GDP |  -.0000801   .0004231    -0.19   0.850    -.0009127    .0007525
                         unemploy |  -.7969562   .8894304    -0.90   0.371    -2.547316    .9534041
                          finbudg |   .0000895   .0002728     0.33   0.743    -.0004474    .0006264
                           over65 |   .1406953   .3412965     0.41   0.680    -.5309613     .812352
                           hltins |   .0000403   .0001281     0.31   0.753    -.0002118    .0002924
                             wave |   .1298071    1.12018     0.12   0.908    -2.074658    2.334273
                                  |
                    provcd#c.wave |
                         Beijing  |          0  (empty)
                         Tianjin  |   .2417993    2.45991     0.10   0.922    -4.599196    5.082795
                           Hebei  |  -.0359528   .9995317    -0.04   0.971    -2.002988    1.931082
                          Shanxi  |   .2280301   1.306942     0.17   0.862    -2.343976    2.800036
                        Liaoning  |  -.2542909   1.986464    -0.13   0.898    -4.163565    3.654983
                           Jilin  |   .4326121   2.332894     0.19   0.853    -4.158422    5.023647
                    Heilongjiang  |  -.1786883   1.038057    -0.17   0.863    -2.221539    1.864162
                        Shanghai  |  -.0718417   .1859726    -0.39   0.700    -.4378278    .2941443
                         Jiangsu  |      .3367   1.399953     0.24   0.810    -2.418347    3.091747
                        Zhejiang  |  -.4479951   1.621073    -0.28   0.782    -3.638197    2.742207
                           Anhui  |  -.0695772   .4891784    -0.14   0.887    -1.032259    .8931047
                          Fujian  |   .6924215   1.068973     0.65   0.518    -1.411272    2.796115
                         Jiangxi  |   .4399599   .8643629     0.51   0.611    -1.261069    2.140989
                        Shandong  |   .2236301   .2604126     0.86   0.391    -.2888506    .7361108
                           Henan  |   .1750232   .1749484     1.00   0.318    -.1692677    .5193141
                           Hubei  |  -.1453725   .4978909    -0.29   0.771      -1.1252    .8344551
                           Hunan  |  -.4561365   .5596887    -0.81   0.416     -1.55758    .6453065
                       Guangdong  |    .402145   1.762763     0.23   0.820    -3.066897    3.871187
Guangxi Zhuang Autonomous Region  |  -.0330182   .3892202    -0.08   0.932    -.7989867    .7329502
                       Chongqing  |   .1413529   .9334656     0.15   0.880    -1.695667    1.978373
                         Sichuan  |   .1521578   .2509067     0.61   0.545    -.3416157    .6459313
                         Guizhou  |  -.5232365   1.029698    -0.51   0.612    -2.549638    1.503165
                          Yunnan  |   .1415563   .9093735     0.16   0.876    -1.648051    1.931164
                         Shaanxi  |  -.1347621   1.644105    -0.08   0.935    -3.370289    3.100765
                           Gansu  |   .1255934   .9566727     0.13   0.896    -1.757097    2.008284
---------------------------------------------------------------------------------------------------
Underidentification test (Kleibergen-Paap rk LM statistic):              0.295
                                                   Chi-sq(1) P-val =    0.5871
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic):                0.440
                         (Kleibergen-Paap rk Wald F statistic):          0.265
Stock-Yogo weak ID test critical values: 10% maximal IV size             16.38
                                         15% maximal IV size              8.96
                                         20% maximal IV size              6.66
                                         25% maximal IV size              5.53
Source: Stock-Yogo (2005).  Reproduced by permission.
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
------------------------------------------------------------------------------
Warning: estimated covariance matrix of moment conditions not of full rank.
         overidentification statistic not reported, and standard errors and
         model tests should be interpreted with caution.
Possible causes:
         number of clusters insufficient to calculate robust covariance matrix
         singleton dummy variable (dummy with one 1 and N-1 0s or vice versa)
partial option may address problem.
------------------------------------------------------------------------------
Collinearities detected among instruments: 1 instrument(s) dropped
Instrumented:         inttimeg4
Included instruments: age20 age30 age40 educ married familysize rural work linc
                      GDP unemploy finbudg over65 hltins wave 12.provcd#c.wave
                      13.provcd#c.wave 14.provcd#c.wave 21.provcd#c.wave
                      22.provcd#c.wave 23.provcd#c.wave 31.provcd#c.wave
                      32.provcd#c.wave 33.provcd#c.wave 34.provcd#c.wave
                      35.provcd#c.wave 36.provcd#c.wave 37.provcd#c.wave
                      41.provcd#c.wave 42.provcd#c.wave 43.provcd#c.wave
                      44.provcd#c.wave 45.provcd#c.wave 50.provcd#c.wave
                      51.provcd#c.wave 52.provcd#c.wave 53.provcd#c.wave
                      61.provcd#c.wave 62.provcd#c.wave
Excluded instruments: bdbd
Partialled-out:       _cons
                      nb: total SS, model F and R2s are after partialling-out;
                          any small-sample adjustments include partialled-out
                          variables in regressor count K
------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
         pid |       299         299           0    *|
      provcd |        25           1          24     |
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

              Comment

              • #8
                Andrew Musau And weirdly, I find the c.wave#i.provcd have same result with c.wave##i.provcd when they are not in the parentheses of absorb

                Code:
                ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave#i.provcd if student==1, absorb(pid) cluster(pid)
(dropped 834 singleton observations)
(MWFE estimator converged in 1 iterations)

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

Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on pid

Number of clusters (pid) =         300                Number of obs =      624
                                                      F( 40,   299) =  9113.71
                                                      Prob > F      =   0.0000
Total (centered) SS     =  409.6982207                Centered R2   =  -0.3870
Total (uncentered) SS   =  409.6982207                Uncentered R2 =  -0.3870
Residual SS             =   568.236899                Root MSE      =    .9873

---------------------------------------------------------------------------------------------------
                                  |               Robust
                              bmi |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------------------+----------------------------------------------------------------
                        inttimeg4 |  -2.775984   7.302772    -0.38   0.704    -17.14733    11.59536
                            age20 |          0   2.413202     0.00   1.000    -4.749013    4.749013
                            age30 |   1.684723   1.615983     1.04   0.298    -1.495418    4.864864
                            age40 |   3.485781   1.685878     2.07   0.040     .1680915    6.803471
                             educ |   .0344186   .2408387     0.14   0.886     -.439535    .5083721
                          married |  -1.379109   1.241377    -1.11   0.267    -3.822053    1.063834
                       familysize |   .0730943   .1727982     0.42   0.673    -.2669604    .4131489
                            rural |  -.6074298   .3659925    -1.66   0.098    -1.327677    .1128177
                             work |  -.1189358   .5836753    -0.20   0.839    -1.267568    1.029696
                             linc |   .0207704   .0663176     0.31   0.754    -.1097381    .1512788
                              GDP |  -.0000893   .0001186    -0.75   0.452    -.0003227    .0001442
                         unemploy |  -.6514161   .9175017    -0.71   0.478    -2.456995    1.154163
                          finbudg |    .000234   .0006006     0.39   0.697     -.000948     .001416
                           over65 |   .0582852   .1863277     0.31   0.755    -.3083947     .424965
                           hltins |   .0000541   .0001035     0.52   0.602    -.0001497    .0002578
                                  |
                    provcd#c.wave |
                         Beijing  |   .1387726   .1896726     0.73   0.465    -.2344897    .5120348
                         Tianjin  |   .1378519   .1893205     0.73   0.467    -.2347176    .5104213
                           Hebei  |    .136497   .1880217     0.73   0.468    -.2335164    .5065104
                          Shanxi  |   .1370443   .1880873     0.73   0.467    -.2330982    .5071868
                        Liaoning  |    .136772   .1885574     0.73   0.469    -.2342957    .5078397
                           Jilin  |   .3140331   .7840028     0.40   0.689    -1.228829    1.856895
                    Heilongjiang  |   .1367426   .1891392     0.72   0.470    -.2354701    .5089552
                        Shanghai  |    .137817   .1900973     0.72   0.469    -.2362811    .5119152
                         Jiangsu  |   .1387767   .1886825     0.74   0.463    -.2325373    .5100906
                        Zhejiang  |   .1367827   .1894798     0.72   0.471    -.2361002    .5096656
                           Anhui  |   .1371456   .1893546     0.72   0.469    -.2354909     .509782
                          Fujian  |   .1384428   .1870715     0.74   0.460    -.2297007    .5065862
                         Jiangxi  |   .1371131   .1889681     0.73   0.469    -.2347629     .508989
                        Shandong  |   .1372712   .1871395     0.73   0.464    -.2310062    .5055485
                           Henan  |    .137074   .1872195     0.73   0.465    -.2313608    .5055088
                           Hubei  |   .1369624   .1891799     0.72   0.470    -.2353303     .509255
                           Hunan  |   .1364485     .18968     0.72   0.472    -.2368284    .5097254
                       Guangdong  |   .2786404   .4043806     0.69   0.491    -.5171521    1.074433
Guangxi Zhuang Autonomous Region  |   .1370317   .1891182     0.72   0.469    -.2351397    .5092031
                       Chongqing  |   .1382566   .1877651     0.74   0.462    -.2312518    .5077651
                         Sichuan  |   .1370035   .1872924     0.73   0.465    -.2315747    .5055817
                         Guizhou  |  -.4586245   .3661949    -1.25   0.211     -1.17927    .2620214
                          Yunnan  |   .1927884   .4396925     0.44   0.661    -.6724955    1.058072
                         Shaanxi  |   .1373368     .18997     0.72   0.470    -.2365108    .5111843
                           Gansu  |   .1362544   .1894814     0.72   0.473    -.2366315    .5091404
Xinjiang Uygur Autonomous Region  |   .1362687    .189569     0.72   0.473    -.2367897    .5093272
---------------------------------------------------------------------------------------------------
Underidentification test (Kleibergen-Paap rk LM statistic):              0.395
                                                   Chi-sq(1) P-val =    0.5299
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic):                0.881
                         (Kleibergen-Paap rk Wald F statistic):          0.367
Stock-Yogo weak ID test critical values: 10% maximal IV size             16.38
                                         15% maximal IV size              8.96
                                         20% maximal IV size              6.66
                                         25% maximal IV size              5.53
Source: Stock-Yogo (2005).  Reproduced by permission.
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
------------------------------------------------------------------------------
Warning: estimated covariance matrix of moment conditions not of full rank.
         overidentification statistic not reported, and standard errors and
         model tests should be interpreted with caution.
Possible causes:
         number of clusters insufficient to calculate robust covariance matrix
         singleton dummy variable (dummy with one 1 and N-1 0s or vice versa)
partial option may address problem.
------------------------------------------------------------------------------
Collinearities detected among instruments: 1 instrument(s) dropped
Instrumented:         inttimeg4
Included instruments: age20 age30 age40 educ married familysize rural work linc
                      GDP unemploy finbudg over65 hltins 11b.provcd#c.wave
                      12.provcd#c.wave 13.provcd#c.wave 14.provcd#c.wave
                      21.provcd#c.wave 22.provcd#c.wave 23.provcd#c.wave
                      31.provcd#c.wave 32.provcd#c.wave 33.provcd#c.wave
                      34.provcd#c.wave 35.provcd#c.wave 36.provcd#c.wave
                      37.provcd#c.wave 41.provcd#c.wave 42.provcd#c.wave
                      43.provcd#c.wave 44.provcd#c.wave 45.provcd#c.wave
                      50.provcd#c.wave 51.provcd#c.wave 52.provcd#c.wave
                      53.provcd#c.wave 61.provcd#c.wave 62.provcd#c.wave
                      65.provcd#c.wave
Excluded instruments: bdbd
Partialled-out:       _cons
                      nb: total SS, model F and R2s are after partialling-out;
                          any small-sample adjustments include partialled-out
                          variables in regressor count K
------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
         pid |       300         300           0    *|
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

                Comment

                • #9
                  However, the code you told me
                  Code:
                  ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave c.wave#i.provcd /// if student==1, absorb(pid i.provcd) cluster(pid)
                  does not have the same result (see below) as the code
                  ivreghdfe bmi (inttimeg4=bdbd) $ctvar c.wave##i.provcd if student==1, absorb(pid) cluster(pid)
                  You have a lot of collinear variables. Make sure it is the same set of indicators being dropped across comparison models. For example, in one, age20 is dropped while in another, it's age40.

                  And weirdly, I find the c.wave#i.provcd have same result with c.wave##i.provcd when they are not in the parentheses of absorb
                  You can see that some indicators from the double hash specification are being dropped due to collinearity. Therefore, the net result is the same.


                  Comment

                  • #10
                    Andrew Musau Thank you very much for your kind reply.

                    I see! Thanks a lot!! Have a nice weekend!!

                    Comment

                    Previous Next
                    Working...
                    X