how to use xtreg after ebalance

[Alice Yang]

Dear Statalist,

I'm using ebalance (from SSC) for entropy balancing, I wonder how to apply xtreg after I obtain _webal?

What I did

ebalance treat controls, targets(3) keep(baltable) replace
xtreg y treat controls i.year [aw=_webal] ,fe vce(cluster firm)

and I get

weight must be constant within firm
r(199);

I also tried pweight and fweight, but still get the same message that weight must be constant within firm. The examples I saw all use reg rather than xtreg. If I want want to run two-way fixed effects model after entropy balancing, how can I achieve this?

Thanks a lot for any suggestions!

[Carlo Lazzaro]

Alice:
you can go -regress-:

. use "https://www.stata-press.com/data/r17/nlswork.dta"
(National Longitudinal Survey of Young Women, 14-24 years old in 1968)

. regress ln_wage i.idcode i.year if idcode<=3

      Source |       SS           df       MS      Number of obs   =        39
-------------+----------------------------------   F(16, 22)       =      2.84
       Model |  3.48635949        16  .217897468   Prob > F        =    0.0122
    Residual |  1.68937946        22  .076789976   R-squared       =    0.6736
-------------+----------------------------------   Adj R-squared   =    0.4362
       Total |  5.17573896        38  .136203657   Root MSE        =    .27711

------------------------------------------------------------------------------
     ln_wage | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
      idcode |
          2  |  -.3898423   .1155629    -3.37   0.003     -.629505   -.1501795
          3  |  -.4648596   .1120424    -4.15   0.000    -.6972212   -.2324979
             |
        year |
         69  |    .208967   .3918928     0.53   0.599    -.6037689    1.021703
         70  |  -.2747772   .3439816    -0.80   0.433    -.9881514    .4385969
         71  |  -.3613911    .326316    -1.11   0.280    -1.038129    .3153467
         72  |  -.2056973    .326316    -0.63   0.535    -.8824352    .4710406
         73  |  -.0310461    .326316    -0.10   0.925     -.707784    .6456917
         75  |   .0416271    .326316     0.13   0.900    -.6351107     .718365
         77  |   .0358937    .326316     0.11   0.913    -.6408441    .7126316
         78  |   .2433199    .326316     0.75   0.464    -.4334179    .9200578
         80  |   .2726139    .326316     0.84   0.412    -.4041239    .9493518
         82  |   .1747839   .3439816     0.51   0.616    -.5385903    .8881581
         83  |   .2924489    .326316     0.90   0.380    -.3842889    .9691868
         85  |   .3712589    .326316     1.14   0.267     -.305479    1.047997
         87  |   .2960361    .326316     0.91   0.374    -.3807017     .972774
         88  |   .3038639    .326316     0.93   0.362    -.3728739    .9806018
             |
       _cons |   1.958421   .2989038     6.55   0.000     1.338532    2.578309
------------------------------------------------------------------------------

. xtreg ln_wage i.year if idcode<=3, fe

Fixed-effects (within) regression               Number of obs     =         39
Group variable: idcode                          Number of groups  =          3

R-squared:                                      Obs per group:
     Within  = 0.5446                                         min =         12
     Between = 0.2670                                         avg =       13.0
     Overall = 0.3678                                         max =         15

                                                F(14,22)          =       1.88
corr(u_i, Xb) = -0.0356                         Prob > F          =     0.0897

------------------------------------------------------------------------------
     ln_wage | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        year |
         69  |    .208967   .3918928     0.53   0.599    -.6037689    1.021703
         70  |  -.2747772   .3439816    -0.80   0.433    -.9881514    .4385969
         71  |  -.3613911    .326316    -1.11   0.280    -1.038129    .3153467
         72  |  -.2056973    .326316    -0.63   0.535    -.8824352    .4710406
         73  |  -.0310461    .326316    -0.10   0.925     -.707784    .6456917
         75  |   .0416271    .326316     0.13   0.900    -.6351107     .718365
         77  |   .0358937    .326316     0.11   0.913    -.6408441    .7126316
         78  |   .2433199    .326316     0.75   0.464    -.4334179    .9200578
         80  |   .2726139    .326316     0.84   0.412    -.4041239    .9493518
         82  |   .1747839   .3439816     0.51   0.616    -.5385903    .8881581
         83  |   .2924489    .326316     0.90   0.380    -.3842889    .9691868
         85  |   .3712589    .326316     1.14   0.267     -.305479    1.047997
         87  |   .2960361    .326316     0.91   0.374    -.3807017     .972774
         88  |   .3038639    .326316     0.93   0.362    -.3728739    .9806018
             |
       _cons |   1.659677   .2833366     5.86   0.000     1.072073    2.247281
-------------+----------------------------------------------------------------
     sigma_u |  .24956596
     sigma_e |  .27711004
         rho |  .44784468   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(2, 22) = 9.64                       Prob > F = 0.0010

.

Assuming that you are interested in -idcode- and -i.year- fixed effects, you can see that the two approaches give the same results as far as the shared predictors are concerned.
It's up to you to add weights- in -regress-.
A word of caution: despite being feasible if you go -regress-, [aw] should be consistent with the methodology in force in your reserach field..

[Alice Yang]

Hi Carlo,

Thanks a lot for your reply!

I tried reg with i.firm and i.year, it works. But I have more than 28,000 firms, so the table goes too long. I think -reghdfe- would also do the same, and I can absorb firm and year fixed effects. My code is

reghdfe y treat controls [aw=_webal], absorb(firm year) vce(cluster firm)

I suppose this is the same as

reg y treat controls i.firm i.year [aw=_webal] ,vce(cluster firm)

As for the weight to use, I actually don't know which one is appropriate. I followed other examples to use aweight. This is a robustness check, my main test used PSM (what weight does PSM use? is it aweight?), and I want to keep the two methods consistent in the choice of weight.

Thanks,
Alice

[Carlo Lazzaro]

Alice:
thanks for clarifying.
Yes, the community-contributed module -reghdfe- is the way to go there (and your -reghdfe- code equals your -regress- specification).
As far as I know, PSM use frequency or sampling weights, though.

[Alice Yang]

Hi Carlo,

Thanks for your reply, I will try fweight or pweight instead.

[Mohamed Mahmoud]

Carlo Lazzaro

Dear Carlo I am wondering if I can apply two step System GMM after entropy balancing using xtabond2 , as I tried to use the following command but always the prob > F = . and the F of the model is missing

The code used:

 xtabond2 depvar  l.depvar   treat  controls  i.year i.ind [aw=_webal] , gmm(treat controls ,collapse) iv(i.year  i.ind , equation(level)) nodiffsargan twostep robust orthogonal small

Do you think is there any errors with my code ?

[Carlo Lazzaro]

Mohamed:
sorry, but I do not use dynamic panel data regression.

[Mohamed Mahmoud]

Many Thanks Carlo Lazzaro

Dear Sebastian Kripfganz Do you think system GMM could works after entropy balancing ? As I tried many times and it does not works


Many thanks

[Sebastian Kripfganz]

I do not know enough about entropy rebalancing to give a helpful answer. I am not sure it preserves the dynamic structure of the model, which would be a problem.