Help with Logit model specification (pooled, panel FE, panel RE)

Giovanna Ortolani

Join Date: Jun 2023
Posts: 10

Help with Logit model specification (pooled, panel FE, panel RE)

29 Apr 2024, 09:27

Dear all,

I am using longitudinal data from the SHARE survey (Survey of Health, Ageing and Retirement in Europe) to run a model on the covariates of transition from work into retirement (for a sample aged 50yo and over). Covariates include: one's risk of poverty in wave t-1, education, sex, age groups, equivalised household size, self-reported health status, work type, marital status, and extra working household member.

My co-authors and I thought it best to estimate a binary choice model like Logit model instead of a linear probability model. We start with a simple reg, estat hettest (heteroskedasticity confirmed), followed by logit ..., vce(cl mergeid).

Next, we estimate conditional xtlogit ..., fe and xtlogit ..., re, followed by hausman fe re. As p-value is zero, the panel Logit with FE would be more appropriate. Note that all models include wave- and country-fixed effects. Results and dataex follow after the questions.

My questions are:

As hausman fe re does not allow for robust nor clustered S.E., I am afraid that the results might not be precise in my case. Being this an unbalanced panel, I cannot easily implement the "robust Hausman test" proposed by Cameron & Trivedi. What would you recommend? For instance, coefficients of pov_risk_t_1, a dependent var of interest, change significantly from one model to another.
In the xtlogit ..., fe results, educ and male variables are omitted, along with all country-FE. In principle, I understand why this happens, but I am puzzled about not being able to include country-FE with individual FE. Also, if I stick to xtlogit ..., fe, I would end up with 13,000 obs (versus 43,000 with xtlogit ..., re). How can I take this constraint into consideration?

If you need any further info, please let me know. It is the first time I am dealing with non-linear models and I am struggling to put theory into practice. Any comments and/or feedback are very much welcome!

Code:

global cov_pov_risk2 pov_risk_t_1 educ male i.age_grp hhsize_eqh_sr sphus_poor i.work_type i.marital_status hhmemb_work 

*** OLS
reg trans $cov_pov_risk2 i.wave i.country if trans==0 | trans==1
rvfplot
estat hettest //p-value=0, can reject null of homoskedasticity

reg trans $cov_pov_risk2 i.wave i.country if trans==0 | trans==1, vce(cl mergeid)
eststo ols_trans_all
capture drop insample
gen insample=1 if e(sample)==1

*** POOLED LOGIT
logit trans $cov_pov_risk2 i.wave i.country if insample==1, vce(cl mergeid)
margins, dydx($cov_pov_risk2) post
outreg2 using "$result/transition_25-04-24.doc", replace ctitle(Pooled OLS) keep($cov_pov_risk2) label dec(3) pdec(3) addtext(Wave FE, YES, Country FE, YES)

*** HAUSMAN TEST

** PANEL LOGIT, FE
xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, fe
est sto fe

** PANEL LOGIT, RE
xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, re
est sto re

hausman fe re //sigmamore option not allowed (?)

Code:

                 ---- Coefficients ----
             |      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
             |       fe           re         Difference       Std. err.
-------------+----------------------------------------------------------------
pov_risk_t_1 |   -.6016172    -.1795393       -.4220778        .2444998
     age_grp |
          2  |   -.7621384     1.532987       -2.295125        .5458784
          3  |    .7511475     3.644596       -2.893449        .7114418
          4  |    2.048137     5.536591       -3.488453         .850413
          5  |    .2238864     5.504902       -5.281015        .9802493
          6  |   -2.020909     5.611854       -7.632763        1.162086
hhsize_eqh~r |   -1.527002    -.6933263       -.8336755        .7235039
  sphus_poor |   -.0086558     .2115868       -.2202427        .2288058
   work_type |
          2  |   -.7272602    -.0266944       -.7005657        .5069954
          3  |   -.1223119    -.4053891        .2830772        .6502483
marital_st~s |
          4  |    .8187897     .5042493        .3145404        2.837579
          5  |    1.590961     .6787202        .9122412        .7105594
 hhmemb_work |    7.846142     3.672941        4.173201        .4753171
        wave |
          4  |    4.198946     .3118012        3.887145        .5569936
          5  |    7.219535    -.5716207        7.791155        .6539124
          6  |    9.880225    -.7593481        10.63957        .7250568
          7  |    12.43963    -.7975258        13.23716        .7884817
          8  |    15.73511    -.6655215        16.40063        .8880205
------------------------------------------------------------------------------
                        b = Consistent under H0 and Ha; obtained from xtlogit.
         B = Inconsistent under Ha, efficient under H0; obtained from xtlogit.

Test of H0: Difference in coefficients not systematic

   chi2(18) = (b-B)'[(V_b-V_B)^(-1)](b-B)
            = 633.49
Prob > chi2 = 0.0000

Code:

. xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, fe
note: multiple positive outcomes within groups encountered.
note: 17,818 groups (30,342 obs) omitted because of all positive or
      all negative outcomes.
note: educ omitted because of no within-group variance.
note: male omitted because of no within-group variance.
note: 12.country omitted because of no within-group variance.
note: 13.country omitted because of no within-group variance.
note: 14.country omitted because of no within-group variance.
note: 15.country omitted because of no within-group variance.
note: 16.country omitted because of no within-group variance.
note: 17.country omitted because of no within-group variance.
note: 18.country omitted because of no within-group variance.
note: 19.country omitted because of no within-group variance.
note: 23.country omitted because of no within-group variance.
note: 28.country omitted because of no within-group variance.
note: 29.country omitted because of no within-group variance.
note: 31.country omitted because of no within-group variance.
note: 32.country omitted because of no within-group variance.
note: 33.country omitted because of no within-group variance.
note: 34.country omitted because of no within-group variance.
note: 35.country omitted because of no within-group variance.
note: 47.country omitted because of no within-group variance.
note: 48.country omitted because of no within-group variance.
note: 51.country omitted because of no within-group variance.
note: 53.country omitted because of no within-group variance.
note: 55.country omitted because of no within-group variance.
note: 57.country omitted because of no within-group variance.
note: 59.country omitted because of no within-group variance.
note: 61.country omitted because of no within-group variance.
note: 63.country omitted because of no within-group variance.

Iteration 0:  Log likelihood = -884.53166  
Iteration 1:  Log likelihood = -457.83119  
Iteration 2:  Log likelihood = -368.83669  
Iteration 3:  Log likelihood = -365.37058  
Iteration 4:  Log likelihood = -365.34699  
Iteration 5:  Log likelihood = -365.34698  

Conditional fixed-effects logistic regression       Number of obs    =  12,985
Group variable: panel                               Number of groups =   4,637

                                                    Obs per group:
                                                                 min =       2
                                                                 avg =     2.8
                                                                 max =       6

                                                    LR chi2(18)      = 8344.49
Log likelihood = -365.34698                         Prob > chi2      =  0.0000

--------------------------------------------------------------------------------------------
                     trans | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
---------------------------+----------------------------------------------------------------
              pov_risk_t_1 |  -.6016172   .2503352    -2.40   0.016    -1.092265   -.1109691
                      educ |          0  (omitted)
                      male |          0  (omitted)
                           |
                   age_grp |
                  55-59yo  |  -.7621384   .5653121    -1.35   0.178     -1.87013    .3458531
                  60-64yo  |   .7511475   .7270053     1.03   0.302    -.6737567    2.176052
                  65-69yo  |   2.048137   .8667529     2.36   0.018     .3493329    3.746942
                  70-74yo  |   .2238864   .9984175     0.22   0.823    -1.732976    2.180749
                    75+yo  |  -2.020909   1.181374    -1.71   0.087     -4.33636    .2945416
                           |
             hhsize_eqh_sr |  -1.527002   .7280251    -2.10   0.036    -2.953905   -.1000989
                sphus_poor |  -.0086558    .233234    -0.04   0.970    -.4657862    .4484745
                           |
                 work_type |
2. Public sector employee  |  -.7272602   .5089185    -1.43   0.153    -1.724722    .2702018
         3. Self-employed  |  -.1223119    .652457    -0.19   0.851    -1.401104     1.15648
                           |
            marital_status |
         2. Never married  |   .8187897   2.838588     0.29   0.773    -4.744741     6.38232
      3. Divorced/widowed  |   1.590961    .712656     2.23   0.026     .1941812    2.987742
                           |
               hhmemb_work |   7.846142   .4802585    16.34   0.000     6.904853    8.787431
                           |
                      wave |
         Wave 4 (2011/12)  |   4.198946   .5628121     7.46   0.000     3.095854    5.302037
            Wave 5 (2013)  |   7.219535   .6587072    10.96   0.000     5.928492    8.510577
            Wave 6 (2015)  |   9.880225   .7291923    13.55   0.000     8.451035    11.30942
         Wave 7 (2017/18)  |   12.43963   .7921748    15.70   0.000       10.887    13.99226
         Wave 8 (2019/20)  |   15.73511    .891713    17.65   0.000     13.98739    17.48284
                           |
                   country |
                  Germany  |          0  (omitted)
                   Sweden  |          0  (omitted)
              Netherlands  |          0  (omitted)
                    Spain  |          0  (omitted)
                    Italy  |          0  (omitted)
                   France  |          0  (omitted)
                  Denmark  |          0  (omitted)
                   Greece  |          0  (omitted)
                  Belgium  |          0  (omitted)
           Czech Republic  |          0  (omitted)
                   Poland  |          0  (omitted)
               Luxembourg  |          0  (omitted)
                  Hungary  |          0  (empty)
                 Portugal  |          0  (empty)
                 Slovenia  |          0  (omitted)
                  Estonia  |          0  (omitted)
                  Croatia  |          0  (omitted)
                Lithuania  |          0  (empty)
                 Bulgaria  |          0  (empty)
                   Cyprus  |          0  (empty)
                  Finland  |          0  (empty)
                   Latvia  |          0  (empty)
                    Malta  |          0  (empty)
                  Romania  |          0  (empty)
                 Slovakia  |          0  (empty)
--------------------------------------------------------------------------------------------

Code:

. xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, re //LR test of rho=0: chibar2(01
> ) = 2.27
note: 59.country != 0 predicts failure perfectly;
      59.country omitted and 18 obs not used.


Fitting comparison model:

Iteration 0:  Log likelihood = -21005.955  
Iteration 1:  Log likelihood = -12053.908  
Iteration 2:  Log likelihood = -10779.721  
Iteration 3:  Log likelihood = -10584.603  
Iteration 4:  Log likelihood = -10579.394  
Iteration 5:  Log likelihood = -10579.358  
Iteration 6:  Log likelihood = -10579.358  

Fitting full model:

tau =  0.0    Log likelihood = -10579.358
tau =  0.1    Log likelihood = -10580.232

Iteration 0:  Log likelihood = -10580.232  
Iteration 1:  Log likelihood = -10578.224  
Iteration 2:  Log likelihood = -10578.222  
Iteration 3:  Log likelihood = -10578.222  

Random-effects logistic regression                  Number of obs    =  43,309
Group variable: panel                               Number of groups =  22,437

Random effects u_i ~ Gaussian                       Obs per group:
                                                                 min =       1
                                                                 avg =     1.9
                                                                 max =       6

Integration method: mvaghermite                     Integration pts. =      12

                                                    Wald chi2(44)    = 3530.49
Log likelihood = -10578.222                         Prob > chi2      =  0.0000

--------------------------------------------------------------------------------------------
                     trans | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
---------------------------+----------------------------------------------------------------
              pov_risk_t_1 |  -.1795393   .0537363    -3.34   0.001    -.2848605   -.0742182
                      educ |   -.045703   .0050548    -9.04   0.000    -.0556102   -.0357958
                      male |   .0069455   .0383919     0.18   0.856    -.0683012    .0821921
                           |
                   age_grp |
                  55-59yo  |   1.532987   .1469509    10.43   0.000     1.244968    1.821005
                  60-64yo  |   3.644596   .1496238    24.36   0.000     3.351339    3.937853
                  65-69yo  |   5.536591   .1675063    33.05   0.000     5.208284    5.864897
                  70-74yo  |   5.504902   .1896016    29.03   0.000     5.133289    5.876514
                    75+yo  |   5.611854   .2126046    26.40   0.000     5.195157    6.028551
                           |
             hhsize_eqh_sr |  -.6933263   .0810102    -8.56   0.000    -.8521035   -.5345491
                sphus_poor |   .2115868   .0452328     4.68   0.000     .1229321    .3002416
                           |
                 work_type |
2. Public sector employee  |  -.0266944   .0442009    -0.60   0.546    -.1133267    .0599378
         3. Self-employed  |  -.4053891   .0536401    -7.56   0.000    -.5105218   -.3002564
                           |
            marital_status |
         2. Never married  |   .5042493   .0756994     6.66   0.000     .3558811    .6526175
      3. Divorced/widowed  |   .6787202   .0546258    12.42   0.000     .5716556    .7857848
                           |
               hhmemb_work |   3.672941   .0687164    53.45   0.000     3.538259    3.807623
                           |
                      wave |
         Wave 4 (2011/12)  |   .3118012   .0807195     3.86   0.000     .1535939    .4700084
            Wave 5 (2013)  |  -.5716207   .0793329    -7.21   0.000    -.7271103   -.4161311
            Wave 6 (2015)  |  -.7593481   .0775501    -9.79   0.000    -.9113435   -.6073526
         Wave 7 (2017/18)  |  -.7975258   .0764036   -10.44   0.000    -.9472741   -.6477775
         Wave 8 (2019/20)  |  -.6655215   .0810649    -8.21   0.000    -.8244058   -.5066372
                           |
                   country |
                  Germany  |  -1.168965     .10466   -11.17   0.000    -1.374095    -.963835
                   Sweden  |  -1.480859   .1034137   -14.32   0.000    -1.683546   -1.278172
              Netherlands  |  -1.273757   .1367407    -9.32   0.000    -1.541764    -1.00575
                    Spain  |  -1.186293   .1140347   -10.40   0.000    -1.409797   -.9627893
                    Italy  |  -.9683722   .1125943    -8.60   0.000    -1.189053   -.7476915
                   France  |  -.4131417   .1014051    -4.07   0.000     -.611892   -.2143913
                  Denmark  |  -1.683413   .1049886   -16.03   0.000    -1.889186   -1.477639
                   Greece  |  -1.922236   .1445765   -13.30   0.000    -2.205601   -1.638871
                  Belgium  |  -.5099378   .0995423    -5.12   0.000    -.7050371   -.3148385
           Czech Republic  |   .0235277   .1025167     0.23   0.818    -.1774014    .2244568
                   Poland  |  -.4236044   .1582189    -2.68   0.007    -.7337077   -.1135011
               Luxembourg  |   .8005673   .1579398     5.07   0.000     .4910109    1.110124
                  Hungary  |  -.2384578   .3374907    -0.71   0.480    -.8999274    .4230119
                 Portugal  |  -1.744507   .3401498    -5.13   0.000    -2.411188   -1.077825
                 Slovenia  |   .2154525   .1305904     1.65   0.099    -.0404999    .4714049
                  Estonia  |  -2.091223   .1080082   -19.36   0.000    -2.302915   -1.879531
                  Croatia  |  -.9198809   .2152789    -4.27   0.000     -1.34182   -.4979419
                Lithuania  |  -1.531874   .3288001    -4.66   0.000     -2.17631   -.8874374
                 Bulgaria  |  -1.750336   .5283161    -3.31   0.001    -2.785816    -.714855
                   Cyprus  |     -1.822   .8435301    -2.16   0.031    -3.475289   -.1687116
                  Finland  |  -1.361717   .4022767    -3.39   0.001    -2.150165   -.5732691
                   Latvia  |  -2.491216   .6616125    -3.77   0.000    -3.787953    -1.19448
                    Malta  |          0  (empty)
                  Romania  |  -1.484375   .6202134    -2.39   0.017    -2.699971   -.2687794
                 Slovakia  |   -1.40925   .3980663    -3.54   0.000    -2.189446   -.6290545
                           |
                     _cons |  -2.556813   .2192327   -11.66   0.000    -2.986501   -2.127124
---------------------------+----------------------------------------------------------------
                  /lnsig2u |   -2.16503   .7041196                     -3.545079   -.7849811
---------------------------+----------------------------------------------------------------
                   sigma_u |   .3387425   .1192576                       .169901    .6753727
                       rho |   .0337032   .0229313                       .008698    .1217642
--------------------------------------------------------------------------------------------
LR test of rho=0: chibar2(01) = 2.27                   Prob >= chibar2 = 0.066

Code:

. hausman fe re //sigmamore option not allowed

                 ---- Coefficients ----
             |      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
             |       fe           re         Difference       Std. err.
-------------+----------------------------------------------------------------
pov_risk_t_1 |   -.6016172    -.1795393       -.4220778        .2444998
     age_grp |
          2  |   -.7621384     1.532987       -2.295125        .5458784
          3  |    .7511475     3.644596       -2.893449        .7114418
          4  |    2.048137     5.536591       -3.488453         .850413
          5  |    .2238864     5.504902       -5.281015        .9802493
          6  |   -2.020909     5.611854       -7.632763        1.162086
hhsize_eqh~r |   -1.527002    -.6933263       -.8336755        .7235039
  sphus_poor |   -.0086558     .2115868       -.2202427        .2288058
   work_type |
          2  |   -.7272602    -.0266944       -.7005657        .5069954
          3  |   -.1223119    -.4053891        .2830772        .6502483
marital_st~s |
          4  |    .8187897     .5042493        .3145404        2.837579
          5  |    1.590961     .6787202        .9122412        .7105594
 hhmemb_work |    7.846142     3.672941        4.173201        .4753171
        wave |
          4  |    4.198946     .3118012        3.887145        .5569936
          5  |    7.219535    -.5716207        7.791155        .6539124
          6  |    9.880225    -.7593481        10.63957        .7250568
          7  |    12.43963    -.7975258        13.23716        .7884817
          8  |    15.73511    -.6655215        16.40063        .8880205
------------------------------------------------------------------------------
                        b = Consistent under H0 and Ha; obtained from xtlogit.
         B = Inconsistent under Ha, efficient under H0; obtained from xtlogit.

Test of H0: Difference in coefficients not systematic

   chi2(18) = (b-B)'[(V_b-V_B)^(-1)](b-B)
            = 633.49
Prob > chi2 = 0.0000

Code:

* Example generated by -dataex-. For more info, type help dataex
clear
input float(trans pov_risk_t_1 educ) byte male float(age_grp hhsize_eqh_sr sphus_poor work_type) byte marital_status float hhmemb_work str12 mergeid
. .  . 1 1 1.4142135 0 . 1 0 "AT-000327-01"
1 0  . 1 1 1.4142135 1 2 1 1 "AT-000327-01"
0 .  3 0 1 1.4142135 0 1 1 1 "AT-000327-02"
. .  3 0 2         1 0 3 5 0 "AT-000674-01"
1 1  3 0 3         1 0 3 5 0 "AT-000674-01"
. 0  3 0 3         1 0 . 5 0 "AT-000674-01"
. . 15 0 5         1 0 . 4 0 "AT-001215-01"
. 0 15 0 5         1 1 . 4 0 "AT-001215-01"
. 0 15 0 6         1 1 . 4 0 "AT-001215-01"
. 0 15 0 6         1 1 . 4 0 "AT-001215-01"
. 0 15 0 6         1 1 . 4 0 "AT-001215-01"
. . 11 0 2 1.4142135 0 . 1 0 "AT-001492-01"
. 0 11 0 3 1.4142135 0 . 1 0 "AT-001492-01"
. 0 11 0 3 1.4142135 0 . 1 1 "AT-001492-01"
. 0 11 0 4 1.4142135 0 . 1 0 "AT-001492-01"
. 0 11 0 4 1.4142135 0 . 1 0 "AT-001492-01"
. . 13 1 2 1.4142135 0 1 1 0 "AT-001492-02"
0 0 13 1 3 1.4142135 0 1 1 0 "AT-001492-02"
1 0 13 1 3 1.4142135 0 1 1 1 "AT-001492-02"
. 0 13 1 4 1.4142135 0 . 1 0 "AT-001492-02"
. 0 13 1 4 1.4142135 0 . 1 0 "AT-001492-02"
. .  . 0 3 1.4142135 0 . 1 0 "AT-001816-01"
. .  8 1 2 1.4142135 0 . 1 0 "AT-001816-02"
0 0  8 1 2 1.7320508 0 1 1 0 "AT-001816-02"
. .  . 0 6 1.4142135 0 1 1 0 "AT-001881-01"
. 0  . 0 6 1.4142135 1 . 1 0 "AT-001881-01"
. 0  . 0 6 1.4142135 1 . 5 0 "AT-001881-01"
. 1  . 0 6 1.4142135 0 . 5 0 "AT-001881-01"
. 0  . 0 6 1.4142135 1 . 5 0 "AT-001881-01"
. .  . 1 6 1.4142135 1 . 1 0 "AT-001881-02"
. 0  . 1 6 1.4142135 1 . 1 0 "AT-001881-02"
. 0  . 1 6 1.7320508 1 . 1 0 "AT-001881-02"
. 1  . 1 6 1.7320508 0 . 1 0 "AT-001881-02"
. 0  . 1 6 1.7320508 0 . 1 0 "AT-001881-02"
. . 10 0 5   2.44949 1 . 5 0 "AT-002132-01"
. 1 10 0 5         1 1 . 5 0 "AT-002132-01"
. . 20 1 2 1.7320508 1 . 1 0 "AT-002136-01"
. 0 20 1 3 1.7320508 0 . 1 0 "AT-002136-01"
. 0 20 1 3 1.7320508 1 . 1 0 "AT-002136-01"
. 0 20 1 4 1.4142135 1 . 1 0 "AT-002136-01"
. 0 20 1 4 1.4142135 1 . 1 0 "AT-002136-01"
. . 14 0 2 1.7320508 0 . 1 0 "AT-002136-03"
. 0 14 0 3 1.7320508 0 . 1 0 "AT-002136-03"
. 0 14 0 3 1.7320508 0 . 1 0 "AT-002136-03"
. 0 14 0 3 1.4142135 0 . 1 0 "AT-002136-03"
. 0 14 0 4 1.4142135 0 . 1 0 "AT-002136-03"
. .  . 1 6   2.44949 1 . 1 0 "AT-002180-02"
. 1  . 1 6   2.44949 1 . 1 0 "AT-002180-02"
. .  2 0 5   2.44949 1 . 1 0 "AT-002180-03"
. 1  2 0 5   2.44949 1 . 1 0 "AT-002180-03"
. 1  2 0 6  2.236068 1 . 5 0 "AT-002180-03"
. .  . 1 5 1.4142135 1 . 1 0 "AT-002355-01"
. .  . 0 5 1.4142135 1 . 1 0 "AT-002355-02"
. .  . 0 4 1.4142135 1 . 1 0 "AT-002525-01"
. 0  . 0 5 1.4142135 1 . 1 0 "AT-002525-01"
. 0  . 0 5 1.4142135 1 . 1 0 "AT-002525-01"
. 0  . 0 5 1.4142135 1 . 1 0 "AT-002525-01"
. 0  . 0 6 1.4142135 1 . 1 0 "AT-002525-01"
. .  . 1 4 1.4142135 1 . 1 0 "AT-002525-02"
. 0  . 1 5 1.4142135 1 . 1 0 "AT-002525-02"
. 0  . 1 5 1.4142135 1 . 1 0 "AT-002525-02"
. 0  . 1 6 1.4142135 1 . 1 0 "AT-002525-02"
. .  8 0 6 1.4142135 0 . 1 0 "AT-002573-01"
. 0  8 0 6 1.4142135 0 . 1 0 "AT-002573-01"
. .  8 1 4 1.4142135 0 . 1 0 "AT-002573-02"
. 0  8 1 5 1.4142135 0 . 1 0 "AT-002573-02"
. . 23 0 3         1 0 1 1 0 "AT-002800-01"
1 0 23 0 3         1 0 1 1 0 "AT-002800-01"
. 0 23 0 3         1 0 1 1 0 "AT-002800-01"
. 0 23 0 4         1 1 . 1 0 "AT-002800-01"
. .  8 0 6 1.4142135 0 . 5 0 "AT-002965-02"
. 1  8 0 6 1.4142135 1 . 5 0 "AT-002965-02"
. 1  8 0 6 1.4142135 1 . 5 0 "AT-002965-02"
. 1  8 0 6 1.4142135 1 . 5 0 "AT-002965-02"
. .  2 1 5 1.4142135 0 . 1 0 "AT-003194-01"
. 0  2 1 6 1.4142135 0 . 1 0 "AT-003194-01"
. 0  2 1 6 1.4142135 0 . 1 0 "AT-003194-01"
. 0  2 1 6 1.4142135 0 . 1 0 "AT-003194-01"
. 0  2 1 6 1.4142135 0 . 1 0 "AT-003194-01"
. .  2 0 5 1.4142135 1 . 1 0 "AT-003194-02"
. .  3 0 1 1.4142135 0 1 1 0 "AT-003683-01"
. .  3 1 1 1.4142135 0 1 1 0 "AT-003683-02"
. .  8 1 1 1.4142135 1 1 1 0 "AT-004234-01"
. 0  8 1 2 1.4142135 1 . 1 0 "AT-004234-01"
. 0  8 1 3 1.4142135 1 . 1 0 "AT-004234-01"
. .  8 0 1 1.4142135 0 1 1 0 "AT-004234-02"
0 0  8 0 2 1.4142135 1 1 1 0 "AT-004234-02"
. .  8 0 4         1 0 1 4 0 "AT-004379-01"
. 0  8 0 5         1 0 . 4 0 "AT-004379-01"
. 1  8 0 6         1 1 . 4 0 "AT-004379-01"
. .  8 1 6 1.4142135 1 . 1 0 "AT-004855-01"
. 1  8 1 6 1.4142135 1 . 1 0 "AT-004855-01"
. .  7 0 5 1.4142135 0 . 1 0 "AT-004855-02"
. 1  7 0 5 1.4142135 1 . 1 0 "AT-004855-02"
. 0  7 0 6 1.7320508 0 . 5 0 "AT-004855-02"
. 1  7 0 6 1.7320508 1 . 5 0 "AT-004855-02"
. 1  7 0 6 1.7320508 1 . 5 0 "AT-004855-02"
. 1  7 0 6 1.7320508 1 . 5 0 "AT-004855-02"
. 1  7 0 6 1.4142135 1 . 5 0 "AT-004855-02"
. .  3 1 4         1 0 . 5 0 "AT-004974-01"
end
label values trans trans
label def trans 0 "0. No transition", modify
label def trans 1 "1. Retirement", modify
label values pov_risk_t_1 pov_risk
label def pov_risk 0 "not in risk of poverty", modify
label def pov_risk 1 "in risk of poverty", modify
label values educ eduyears_mod
label values male gender
label def gender 0 "Female", modify
label def gender 1 "Male", modify
label values age_grp age_grp
label def age_grp 1 "50-54yo", modify
label def age_grp 2 "55-59yo", modify
label def age_grp 3 "60-64yo", modify
label def age_grp 4 "65-69yo", modify
label def age_grp 5 "70-74yo", modify
label def age_grp 6 "75+yo", modify
label values work_type work_type_val
label def work_type_val 1 "1. Private sector employee", modify
label def work_type_val 2 "2. Public sector employee", modify
label def work_type_val 3 "3. Self-employed", modify
label values marital_status marital_status
label def marital_status 1 "1. Married/in partnership", modify
label def marital_status 4 "2. Never married", modify
label def marital_status 5 "3. Divorced/widowed", modify

Tags: None

Mustafa Ugur Karakaplan

Join Date: Jun 2015

Posts: 26
#2

29 Apr 2024, 14:11

1. Regarding the discrepancy between fixed effects and random effects models and the inability to use the Hausman test with clustered standard errors:

You can estimate both models with clustered standard errors using the -vce(cluster mergeid)- option. After estimation, you can use the -estimates store- command to store the results, and then use -hausman- with the -alleqs- option to compare coefficients across the two models, even though it won't be a formal test:

Code:

xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, fe vce(cluster mergeid) estimates store fe_cluster xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, re vce(cluster mergeid) estimates store re_cluster hausman fe_cluster re_cluster, alleqs

For the bootstrap Hausman test, you can use Stata's -bootstrap- command. Here's a sketch of the code:

Code:

bootstrap diff=(_b[fe:]-_b[re:]), reps(100) cluster(mergeid) : /// (xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, fe) /// (xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, re)

This will give you the distribution of the differences in coefficients, which you can use to assess the significance of the differences.

2. Regarding the omitted variables and reduced sample size in the fixed effects model:

After running the fixed effects model, you can use -predict, residuals- to obtain the fixed effects. Then you can regress these on the time-invariant variables to see if there's a systematic relationship:

Code:

xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, fe predict fe, u reg fe educ male i.country

To compare the random effects results on the fixed effects sample, you can use an -if- condition based on -e(sample)- from the fixed effects model:

Code:

xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, fe estimates store fe xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1 & e(sample), re estimates store re_fe_sample estimates table fe re_fe_sample, b se

This will show you how much of the difference is due to the change in sample.

Remember, Stata's -margins- command can be very useful for interpreting the coefficients in non-linear models like logit. After your -xtlogit- command, you can use something like:

Code:

margins, dydx(*)

This will give you the average marginal effects, which are often more interpretable than the raw coefficients.
Comment
Giovanna Ortolani

Join Date: Jun 2023

Posts: 10
#3

30 Apr 2024, 05:51

Hello Mustafa,
thanks for your kind reply, but there is a problem with the suggested approach for point 1:

Originally posted by Mustafa Ugur Karakaplan View Post

1. Regarding the discrepancy between fixed effects and random effects models and the inability to use the Hausman test with clustered standard errors:

You can estimate both models with clustered standard errors using the -vce(cluster mergeid)- option.

As a matter of fact, -xtlogit ..., fe- does not support -vce(cluster clusterid)- option like; it needs to be either -vce(bootstrap)- or -vce(jackknife)- (also discussed in this post). The same limitation does not apply to -xtlogit ..., re-, though. I also read the meaning of -hauman fe re, alleqs- option but I don't understand how it fits this context.

On point 2, after running...

Code:

xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, fe predict fe, u reg fe educ male i.country

...I get:

Code:

. predict fe, u option u not allowed r(198);

I examined -help xtreg postestimation- but I cannot seem to find the mistake. Help anyone?

Another question on top of post #1:
How can I test the joint significance of i.country, and the joint significance of i.wave in a -xtlogit ..., fe- and -xtlogit ..., re-? Would -testparm i.wave- and -testparm i.country- be the correct approach?

Thank you all!
Comment
Giovanna Ortolani

Join Date: Jun 2023

Posts: 10
#4

23 May 2024, 10:45

Anyone could help with the previous doubt, please? Thank you in advance
Comment

Announcement

Help with Logit model specification (pooled, panel FE, panel RE)

Comment

Comment

Comment