Announcement

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

    Problem with -margins- after -xtlogit-

    Dear all,

    I estimate a panel Logit model on the covariates of transition from work into retirement (for a sample aged 50yo and over) using data from the SHARE survey (Survey of Health, Ageing and Retirement in Europe). 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.

    I am using -margins- command after -xtlogit-, but Stata reports that it cannot estimate it for categorical variables such as sex, age groups, marital status and work type. Why this happens? Does someone know how, and if, this could be fixed?

    Thanks!

    Code:
    xtlogit trans $cov_pov_risk2 i.wave i.country if insample==1, vce(cl mergeid)
note: 59.country != 0 predicts failure perfectly;
      59.country omitted and 18 obs not used.


Fitting comparison model:

Iteration 0:  Log pseudolikelihood =  -19819.09  
Iteration 1:  Log pseudolikelihood = -11436.338  
Iteration 2:  Log pseudolikelihood = -10233.414  
Iteration 3:  Log pseudolikelihood = -10039.908  
Iteration 4:  Log pseudolikelihood = -10034.707  
Iteration 5:  Log pseudolikelihood = -10034.671  
Iteration 6:  Log pseudolikelihood = -10034.671  

Fitting full model:

tau =  0.0    Log pseudolikelihood = -10034.671
tau =  0.1    Log pseudolikelihood = -10035.455

Iteration 0:  Log pseudolikelihood = -10035.455  
Iteration 1:  Log pseudolikelihood =  -10033.58  
Iteration 2:  Log pseudolikelihood = -10033.576  
Iteration 3:  Log pseudolikelihood = -10033.576  

Calculating robust standard errors ...

Random-effects logistic regression                  Number of obs    =  41,108
Group variable: panel                               Number of groups =  21,691

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

Integration method: mvaghermite                     Integration pts. =      12

                                                    Wald chi2(44)    = 3775.88
Log pseudolikelihood = -10033.576                   Prob > chi2      =  0.0000

                                         (Std. err. adjusted for 21,691 clusters in mergeid)
--------------------------------------------------------------------------------------------
                           |               Robust
                     trans | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
---------------------------+----------------------------------------------------------------
              pov_risk_t_1 |   -.166522   .0554566    -3.00   0.003    -.2752149   -.0578291
                      educ |  -.0480365   .0051522    -9.32   0.000    -.0581347   -.0379383
                      male |  -.0036535   .0403659    -0.09   0.928    -.0827692    .0754621
                           |
                   age_grp |
                  55-59yo  |   1.583443   .1441991    10.98   0.000     1.300818    1.866068
                  60-64yo  |   3.665995   .1458866    25.13   0.000     3.380062    3.951927
                  65-69yo  |    5.52666    .163362    33.83   0.000     5.206477    5.846844
                  70-74yo  |   5.521011   .1849095    29.86   0.000     5.158595    5.883426
                    75+yo  |   5.636382   .2081366    27.08   0.000     5.228442    6.044323
                           |
             hhsize_eqh_sr |   -.688005   .0841462    -8.18   0.000    -.8529286   -.5230814
                sphus_poor |   .2017044    .046549     4.33   0.000     .1104699    .2929388
                           |
                 work_type |
2. Public sector employee  |  -.0362042   .0449182    -0.81   0.420    -.1242423    .0518339
         3. Self-employed  |  -.3962109   .0567451    -6.98   0.000    -.5074292   -.2849926
                           |
            marital_status |
         2. Never married  |   .5132168   .0761686     6.74   0.000     .3639292    .6625045
      3. Divorced/widowed  |   .6816849   .0563723    12.09   0.000     .5711973    .7921726
                           |
               hhmemb_work |   3.681871   .0693236    53.11   0.000     3.545999    3.817743
                           |
                      wave |
         Wave 4 (2011/12)  |   .3374674   .0840367     4.02   0.000     .1727584    .5021764
            Wave 5 (2013)  |  -.5325465   .0808804    -6.58   0.000    -.6910691   -.3740239
            Wave 6 (2015)  |  -.7306933   .0770652    -9.48   0.000    -.8817383   -.5796483
         Wave 7 (2017/18)  |  -.7648985   .0768926    -9.95   0.000    -.9156052   -.6141918
         Wave 8 (2019/20)  |  -.6286996   .0854089    -7.36   0.000    -.7960979   -.4613014
                           |
                   country |
                  Germany  |  -1.132057   .1076993   -10.51   0.000    -1.343143   -.9209697
                   Sweden  |   -1.45545    .101693   -14.31   0.000    -1.654765   -1.256136
              Netherlands  |   -1.05716   .2297129    -4.60   0.000    -1.507389   -.6069306
                    Spain  |  -1.180997   .1105569   -10.68   0.000    -1.397684   -.9643094
                    Italy  |  -.9605274   .1118705    -8.59   0.000    -1.179789   -.7412653
                   France  |  -.4165643   .1011144    -4.12   0.000    -.6147449   -.2183838
                  Denmark  |  -1.660126   .1037472   -16.00   0.000    -1.863466   -1.456785
                   Greece  |  -1.926768   .1433675   -13.44   0.000    -2.207763   -1.645773
                  Belgium  |  -.5056569   .0967097    -5.23   0.000    -.6952045   -.3161094
           Czech Republic  |   .0129918   .1045713     0.12   0.901    -.1919641    .2179477
                   Poland  |  -.4235347   .1624915    -2.61   0.009    -.7420123   -.1050572
               Luxembourg  |    .796625    .164644     4.84   0.000     .4739288    1.119321
                  Hungary  |  -.2479898   .3837071    -0.65   0.518    -1.000042    .5040623
                 Portugal  |  -1.773635   .4457738    -3.98   0.000    -2.647336   -.8999345
                 Slovenia  |   .2080824   .1367434     1.52   0.128    -.0599297    .4760945
                  Estonia  |  -2.076609    .108053   -19.22   0.000    -2.288389   -1.864829
                  Croatia  |  -.9230151    .202877    -4.55   0.000    -1.320647   -.5253835
                Lithuania  |  -1.530839   .3172675    -4.83   0.000    -2.152672   -.9090057
                 Bulgaria  |  -1.756475   .4795923    -3.66   0.000    -2.696459   -.8164917
                   Cyprus  |  -1.829712   .6442928    -2.84   0.005    -3.092503   -.5669214
                  Finland  |   -1.37054   .3897093    -3.52   0.000    -2.134356   -.6067239
                   Latvia  |  -2.485857   .6674459    -3.72   0.000    -3.794027   -1.177687
                    Malta  |          0  (empty)
                  Romania  |  -1.502531   .6942346    -2.16   0.030    -2.863206   -.1418559
                 Slovakia  |  -1.277278    .785677    -1.63   0.104    -2.817177    .2626202
                           |
                     _cons |  -2.592475   .2180242   -11.89   0.000    -3.019795   -2.165155
---------------------------+----------------------------------------------------------------
                  /lnsig2u |  -2.120559   .6699549                     -3.433647   -.8074715
---------------------------+----------------------------------------------------------------
                   sigma_u |    .346359   .1160225                      .1796359    .6678206
                       rho |   .0351819   .0227411                      .0097133    .1193795
--------------------------------------------------------------------------------------------
    Code:
    margins, dydx(pov_risk_t_1 educ i.male i.age_grp hhsize_eqh_sr sphus_poor i.work_type i.marital_status hhmemb_work) post 

Average marginal effects                                Number of obs = 11,875
Model VCE: Bootstrap

Expression: Pr(trans|fixed effect is 0), predict(pu0)
dy/dx wrt:  pov_risk_t_1 educ 1.male 2.age_grp 3.age_grp 4.age_grp 5.age_grp 6.age_grp hhsize_eqh_sr sphus_poor 2.work_type 3.work_type 4.marital_status 5.marital_status hhmemb_work

--------------------------------------------------------------------------------------------
                           |            Delta-method
                           |      dy/dx   std. err.      z    P>|z|     [95% conf. interval]
---------------------------+----------------------------------------------------------------
              pov_risk_t_1 |   .0097081   .0068568     1.42   0.157    -.0037311    .0231472
                      educ |          0  (omitted)
                           |
                      male |
                     Male  |          .  (not estimable)
                           |
                   age_grp |
                  55-59yo  |          .  (not estimable)
                  60-64yo  |          .  (not estimable)
                  65-69yo  |          .  (not estimable)
                  70-74yo  |          .  (not estimable)
                    75+yo  |          .  (not estimable)
                           |
             hhsize_eqh_sr |  -.1381772   .0353437    -3.91   0.000    -.2074496   -.0689047
                sphus_poor |   .0029245    .006222     0.47   0.638    -.0092703    .0151193
                           |
                 work_type |
2. Public sector employee  |          .  (not estimable)
         3. Self-employed  |          .  (not estimable)
                           |
            marital_status |
         2. Never married  |          .  (not estimable)
      3. Divorced/widowed  |          .  (not estimable)
                           |
               hhmemb_work |   .2724671   .0771439     3.53   0.000     .1212678    .4236664
--------------------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
    Tags: None
  • #2
    It would probably be helpful to show us the command that produced $cov_pov_risk2. However, I am channeling Joao Santos Silva and pointing you towards aextlogit given the known problems of uninterpretable marginal effects estimates from fixed effects logit models. See the helpful presentation as well.
    • 2 likes

    Comment

    • #3
      Originally posted by Erik Ruzek View Post
      It would probably be helpful to show us the command that produced $cov_pov_risk2. However, I am channeling Joao Santos Silva and pointing you towards aextlogit given the known problems of uninterpretable marginal effects estimates from fixed effects logit models. See the helpful presentation as well.
      Thank you for the suggestion to use -aextlogit-, Erik! I will try it out.

      My apologies, here is the command producing $cov_pov_risk2. I was careful to use i.variable always, apart for male. Could it be something wrong with how these categorical variables were created?

      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

      Comment

      • #4
        It is unique to what you've done. I can get margins on dummy/cat variables using xtlogit without a problem.

        If all you're interested in is the margins, just use the LPM.

        Comment

        • #5
          Originally posted by George Ford View Post
          It is unique to what you've done. I can get margins on dummy/cat variables using xtlogit without a problem.

          If all you're interested in is the margins, just use the LPM.
          I have revised my code to recreate all dummies and categorical variables used in the regression and even changed the calculations in some instances. Still, -margins- after -xtlogit- do not calculate margins for the same variables mentioned in my post #3. I am not simply interested in the margins, but in comparing the models used (see my other post here) and how margins vary across them.

          Therefore, I would like to address this curious error in -margins- calculation. Please see below how each "faulty" variable is calculated. If anyone has any suggestions on how to make it work, please let me know.

          Code:
          generate agegrp=.
replace agegrp=1 if 50<=age & age<55
replace agegrp=2 if 55<=age & age<60
replace agegrp=3 if 60<=age & age<65
replace agegrp=4 if 65<=age & age<70
replace agegrp=5 if 70<=age & age<75
replace agegrp=6 if 75<=age

gen work_type=ep009__w
la var work_type "Type of work (private/public sector, self-employed)"
la de work_type_val 1"1. Private sector employee" 2"2. Public sector employee" 3"3. Self-employed"

gen marital_status0=marital_status 
la var marital_status0 "Marital status"
replace marital_status0=. if marital_status<0 //replace "no info", "don't know/refusal", and "implausible" as missing
replace marital_status0=1 if marital_status==2 | marital_status==3 //Married or in partnership
replace marital_status0=2 if marital_status==4 //Never married
replace marital_status0=3 if marital_status==5 //Divorced or widowed
la de marital_status0 1"1. Married/in partnership" 2"2. Never married" 3"3. Divorced/widowed"
la values marital_status0 marital_status0
la var married "Married dummy"
drop marital_status 
rename marital_status0 marital_status

          Comment

          • #6
            No obvious issues there.

            Without a dataex, I'll just toss some ideas out there.

            What's up with the error message related to 59.country?

            Run on a subset of countries (or one country), see if keeps doing it.

            Run:
            Code:
            margins, dydx(*) predict(pr)
margins, dydx(*) predict(xb)

            Replicate the margins error with a small subset of the data and then dataex that portion of the data.

            Also see
            HTML Code:
            https://www.statalist.org/forums/forum/general-stata-discussion/general/1304704-cannot-estimate-marginal-effect-after-xtlogit
https://www.statalist.org/forums/forum/general-stata-discussion/general/1410365-margins-after-xtlogit-re

            Comment

            Previous Next
            Working...
            X