-xtevent- Stata module to estimate and visualize linear panel event-study models updated on SSC

[Jorge Eduardo Perez Perez]

Thanks to Kit Baum, -xtevent- 2.0.0 is now available on SSC.

-xtevent- provides commands to estimate linear panel event-study models, create event-study plots, and conduct hypothesis tests after estimation. It implements the plotting suggestions and many of the estimators discussed in Freyaldenhoven et al. (2021).

-xtevent- 2.0.0 adds new features:

• Option nostaggered has been deprecated. There are new options to handle imputation of the policy variable in scenarios where it may be missing for some units or time periods. In -xtevent- 1.0.0 the default was to impute the policy variable at missing values and at the beginning and end of each timeseries in a manner consistent with staggered adoption. This default behavior has changed to not imputing the policy variable. To obtain similar results to those using -xtevent- 1.0.0 please use the option impute(stag).
• Options for trend extrapolation when the variable is not binary
• Bug fixes

We are now developing -xtevent- on Github. Check the Github page to see new releases, to follow development, to report bugs, and to request features:

https://github.com/JMSLab/xtevent

See general information about -xtevent- in the initial post:

https://www.statalist.org/forums/for...ailable-on-ssc

Here's the companion paper:

https://www.nber.org/papers/w29170

Here's an ungated version:

https://jorgeperezperez.com/files/EventStudy.pdf

[Hideto Koizumi]

Dear Jorge Eduardo Pérez Pérez,
Thank you very much for making this package available!

I must be misunderstanding what xtevent does, and I am having troubles with reghdfe option. Somehow, it drops many more observations than default, and I find a large discrepancy in standard errors between reghdfe result and default.

I installed it through

 net install xtevent, from("https://raw.githubusercontent.com/JMSLab/xtevent/master")

And I noticed a large discrepancy in standard errors between reghdfe result and default.
I used the example dataset

 *** setup
webuse nlswork, clear * year variable has many missing observations * Create a time variable that ignores the gaps
by idcode (year): gen time=_n
xtset idcode time

In particular, I compare

. xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , pol(union) w(3) cluster(idcode) plot

Using options panelvar and timevar from xtset

No proxy or instruments provided. Implementing OLS estimator
note: 12.time omitted because of collinearity.

Linear regression, absorbing indicators             Number of obs     =    421
Absorbed variable: idcode                           No. of categories =    373
                                                    F(19, 372)        =   5.87
                                                    Prob > F          = 0.0000
                                                    R-squared         = 0.9929
                                                    Adj R-squared     = 0.8979
                                                    Root MSE          = 0.1308

                                      (Std. err. adjusted for 373 clusters in idcode)
-------------------------------------------------------------------------------------
                    |               Robust
            ln_wage | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
--------------------+----------------------------------------------------------------
           _k_eq_m4 |    .042523   .8339359     0.05   0.959    -1.597296    1.682343
           _k_eq_m3 |   .0296822   .5998528     0.05   0.961    -1.149845     1.20921
           _k_eq_m2 |  -.0793981   .4049379    -0.20   0.845    -.8756523    .7168561
           _k_eq_p0 |   .0497396   .2800496     0.18   0.859     -.500939    .6004183
           _k_eq_p1 |   .0292423   .5159146     0.06   0.955    -.9852324    1.043717
           _k_eq_p2 |  -.1286404   .7467273    -0.17   0.863    -1.596976    1.339695
           _k_eq_p3 |  -.2677541   .7636168    -0.35   0.726    -1.769301    1.233793
           _k_eq_p4 |  -.1567835   .8617116    -0.18   0.856     -1.85122    1.537653
                age |   .0886796   .3813574     0.23   0.816    -.6612068    .8385661
                    |
        c.age#c.age |  -.0012995   .0041129    -0.32   0.752    -.0093869    .0067879
                    |
            ttl_exp |   -.162915   .5205288    -0.31   0.754    -1.186463    .8606329
                    |
c.ttl_exp#c.ttl_exp |   .0014087   .0070491     0.20   0.842    -.0124523    .0152697
                    |
             tenure |   .0045693   .0389143     0.12   0.907    -.0719503    .0810889
                    |
               time |
                 6  |   .2856202   .5228734     0.55   0.585    -.7425378    1.313778
                 7  |   .5344666   .8962087     0.60   0.551    -1.227804    2.296737
                 8  |   .6918545   1.258341     0.55   0.583    -1.782498    3.166207
                 9  |    .975965   1.772744     0.55   0.582    -2.509892    4.461822
                10  |   1.246936   2.153808     0.58   0.563    -2.988229    5.482101
                11  |    1.41785   2.445418     0.58   0.562    -3.390725    6.226425
                12  |          0  (omitted)
                    |
              _cons |   1.398587   6.385553     0.22   0.827    -11.15772    13.95489
-------------------------------------------------------------------------------------

against

.. xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , pol(union) w(3) cluster(idcode) plot reghdfe

Using options panelvar and timevar from xtset

No proxy or instruments provided. Implementing OLS estimator
(dropped 350 singleton observations)
(MWFE estimator converged in 7 iterations)

HDFE Linear regression                            Number of obs   =         71
Absorbing 2 HDFE groups                           F(  13,     22) =       4.70
Statistics robust to heteroskedasticity           Prob > F        =     0.0007
                                                  R-squared       =     0.9406
                                                  Adj R-squared   =     0.8514
                                                  Within R-sq.    =     0.3040
Number of clusters (idcode)  =         23         Root MSE        =     0.1331

                                       (Std. err. adjusted for 23 clusters in idcode)
-------------------------------------------------------------------------------------
                    |               Robust
            ln_wage | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
--------------------+----------------------------------------------------------------
           _k_eq_m4 |    .042523   .2621444     0.16   0.873    -.5011312    .5861773
           _k_eq_m3 |   .0296822   .1885613     0.16   0.876    -.3613701    .4207344
           _k_eq_m2 |  -.0793981   .1272906    -0.62   0.539    -.3433827    .1845864
           _k_eq_p0 |   .0497396   .0880325     0.57   0.578    -.1328285    .2323078
           _k_eq_p1 |   .0292423   .1621757     0.18   0.859    -.3070895    .3655741
           _k_eq_p2 |  -.1286404   .2347307    -0.55   0.589    -.6154422    .3581613
           _k_eq_p3 |  -.2677541   .2400399    -1.12   0.277    -.7655663    .2300581
           _k_eq_p4 |  -.1567835   .2708756    -0.58   0.569    -.7185451    .4049781
                age |   .0886796   .1198782     0.74   0.467    -.1599324    .3372917
                    |
        c.age#c.age |  -.0012995   .0012929    -1.01   0.326    -.0039808    .0013817
                    |
            ttl_exp |   -.162915   .1636261    -1.00   0.330    -.5022549    .1764249
                    |
c.ttl_exp#c.ttl_exp |   .0014087   .0022158     0.64   0.532    -.0031867    .0060041
                    |
             tenure |   .0045693   .0122326     0.37   0.712    -.0207995     .029938
              _cons |   1.964052   2.039996     0.96   0.346    -2.266641    6.194746
-------------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
      idcode |        23          23           0    *|
        time |         7           0           7     |
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

Why does xtevent drops 350 observations only under reghdfe option??

I would very much appreciate your response.
Kindest,
Hideto