eventdd – Estimate panel event study models and generate event study plots

eventdd is a Stata program which estimates panel event study models and generates event-study plots. eventdd additionally includes a number of post-estimation procedures such as joint tests of significance of event lages and leads.

Panel event study models seek to estimtate the impact of some policy or event adopted in certain groups and time periods, by comparing evolution of outcomes in adoption groups to groups which do not adopt the event. Specifically, the panel event study is estimated as:

$$y_{gt}= \alpha + \sum_{j=2}^J\beta_j(Lead_j)_{gt} + \sum_{k=0}^K \gamma_k (Lag_k)_{gt} + \mu_g + \lambda_t + X_{gt}^\prime+\varepsilon_{gt}$$

where $y_{gt}$ refers to outcomes for group $g$ and time period $t$, $\mu_g$ and $\lambda_t$ refer to group and time-fixed effects respectively, and, optionally, $X_{gt}$ refers to time-varying controls. Lags and leads refer to indicators for periods pre- and post-event occurrence, and are defined as follows:

$$(Lead\ J)_{gt}= \mathbb{1}[t\leq Event_g-J]$$ $$(Lead\ j)_{gt}= \mathbb{1}[t = Event_g-j] \text{ for } j\in\{2,\ldots,J-1\}$$ $$(Lag\ k)_{gt}= \mathbb{1}[t = Event_g+k] \text{ for } k\in\{0,\ldots,K-1\}$$ $$(Lead\ K)_{gt}= \mathbb{1}[t\leq Event_g+K]$$

where $\mathbb{1}$ is the indicator function, indicating that observations take 1 if $\mathbb{1}(\cdot)$ is true, and zero otherwise. Leads and lags are thus binary variables indicating that the given group was a given number of periods away from the event of interest in the respective time period. $J$ and $K$ leads and lags are included respectively leads and lags "accumulate" leads or lags beyond $J$ and $K$ periods. A single lead or lag variable is omitted to capture the baseline difference between groups where the event does and does not occur. In the specification above, as standard, this baseline omitted case is the first lead (one period prior to the reform), where $j = 1$.

eventdd permits a range of estimation procedures such as regression (regress) with fixed effects manually indicated, xtreg, and reghdfe (Correia, 2017), as well as a range of inference procedures. Standard errors should be clustered by group, but if a small number of groups is present, wild cluster bootstrap should be preferred. This is available via the boottest command (Roodman et al., 2019). Event study models can also be estimated over some other variable, which presents two (or more) event studies on a single plot. Examples of usage are provided below.

Further details can be found in the accompanying Stata Journal paper (Clarke & Tapia-Schythe, 2021) here, or a pre-print version, available here. Full details of the command are documented in this paper and the help file available with this command. An overview of syntax is available below, but please refer to the Stata Journal paper or the help file for full details including returned elements and supported post-estimation tests.

To install directly into Stata:

ssc install eventdd, replace

or using net install command:

net install eventdd, from("https://raw.githubusercontent.com/damiancclarke/eventdd/master") replace

Syntax

eventdd depvar [indepvars] [if] [in] [weight], timevar(varname) [options]

where options are:

• timevar(varname): Specifies the standardized time variable relative to the event of interest. This is required.
• over(varname): Indicates that multiple event studies should be estimated and plotted, where a separate event study is plotted over each level of the variable indicated. The indicated variable must be discrete, but can take greater than two levels, and can be either string or numerical format. Resulting graphs will be combined in a single plot.
• ci(type, ...): Specifies the type of graph for confidence intervals: rarea (with area shading), rcap (with capped spikes) or rline (with lines), and also the graphing options: twoway rarea for rarea (eg area), twoway rcap for rcap (eg line) or twoway rline for rline (eg connect) which should be passed to the resulting event study graph. rcap is the default graph type. If multiple event studies are produced using the over() option, and separate options are desired for each set of confidence intervals on each event study this can be requested using g1(), g2(), and so forth to pass options, with values corresponding to orderd levels of the variable indicated in the over() option.
• jitter(#): Only for use if over() is specified. Allows for each event study to be slightly shifted on graphical output to avoid super-imposition. Scalar value between 0 and 1 indicated in jitter indicates the distance on the horizontal axis to shift each event study.
• baseline(#): Specifies the baseline period relative to the moment of the event of interest; the default is -1.
• level(#): Set confidence level; default is level(95).
• accum: Accumulates periods beyond indicated leads and lags into a final coefficient/confidence interval.
• leads(#): Specifies the number of leads which should be included. This is required when specifying the accum, keepbal or inrange options, otherwise all possible lags will be plotted.
• lags(#): Specifies the number of lags which should be included. This is required when specifying the accum, keepbal or inrange options, otherwise all possible lags will be plotted.
• noend: Requests that end points are suppressed from graphs. This is only available if specifying the accum option.
• keepbal(varname): Indicates that only units which are balanced in the panel should be kept, where varname indicates the panel variable (eg state).
• method(type, [absorb(absvars)] * ...): Specifies the estimation method: ols (with Stata's regress command), fe (with Stata's xtreg, fe command) or hdfe (with the user-written reghdfe command), and also any additional estimation options and vce options which should be passed to the event study model, (eg robust or clustered sandwich estimator of variance). The absorb(absvars) sub-option is only required wh fying the hdfe option. ols is the default estimation method.
• wboot: Requests that confidence intervals be estimated by wild bootstrap. This requires the user-written boottest command. This may not be combined with the hdfe option.
• wboot_op(string): Specifies any options for wild bootstrap estimation, (eg seed(), bootcluster()). These will be passed to the boottest command. In the case of using the level option, this should only be specified in command syntax. nograph option is already specified.
• balanced: Requests that only balanced periods in which all units have data be shown in the plot.
• inrange: Requests that only specified periods in leads and lags be shown in the plot.
• noline: Requests that line at -1 on the x-axis is suppressed from graphs.
• graph_op(string): Specifies any general options in twoway options which should be passed to the resulting event study graph, (eg title, axis, labels, legends).
• coef_op(string): Specifies any options for coefficients in scatter which should be passed to the resulting event study graph, (eg marker). If multiple event studies are produced using the over() option, and separate options are desired for each set of coefficients on each event study this can be requested using g1(), g2(), and so forth to pass options, with values corresponding to orderd levels of the variable indicated in the over() option.
• endpoints_op(string): Specifies any options for end points coefficients in scatter which should be passed to the resulting event study graph, (eg marker). This is only available if specifying the accum option.
• keepdummies: Generate dummies of leads and lags. Required save the data before using or data in memory would be lost. This option is necessary to perform joint significance tests using wild or score bootstrap with the postestimation commands.

Examples

Stevenson & Wolfers (2006), No Fault Divorce and Suicides

webuse set www.damianclarke.net/stata/
webuse bacon_example.dta, clear

gen timeToTreat = year - _nfd

eventdd asmrs pcinc asmrh cases i.year, timevar(timeToTreat) method(fe, cluster(stfips)) graph_op(ytitle("Suicides per 1m Women") xlabel(-15(5)15) legend(order(2 "Point Estimate" 1 "95% CI") pos(6) rows(1)))  lags(15) leads(15) accum


. eventdd asmrs pcinc asmrh cases i.year, timevar(timeToTreat) method(fe, cluster(stfips)) graph_op(ytitle("Suicides per 1m Women") xlabel(-15(5)15) legend(order(2 "Point Estimate" 1 "95% CI") pos(6) rows(1))) lags(15) leads(15) accum


Fixed-effects (within) regression               Number of obs     =      1,617
Group variable: stfips                          Number of groups  =         49

R-squared:                                      Obs per group:
     Within  = 0.3846                                         min =         33
     Between = 0.0002                                         avg =       33.0
     Overall = 0.1151                                         max =         33

                                                F(48, 48)         =          .
corr(u_i, Xb) = -0.2138                         Prob > F          =          .

                                (Std. err. adjusted for 49 clusters in stfips)
------------------------------------------------------------------------------
             |               Robust
       asmrs | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
       pcinc |  -.0010834   .0004016    -2.70   0.010    -.0018908   -.0002759
       asmrh |   1.148921   .5916515     1.94   0.058    -.0406743    2.338516
       cases |  -194.0857   134.5432    -1.44   0.156    -464.6029    76.43154
             |
        year |
       1965  |   5.953814   1.997673     2.98   0.005     1.937224    9.970405
       1966  |   3.654006   2.323974     1.57   0.122    -1.018658     8.32667
       1967  |    5.17208   2.304469     2.24   0.029     .5386341    9.805525
       1968  |   6.133335   2.455151     2.50   0.016     1.196924    11.06975
       1969  |   9.226593   3.110661     2.97   0.005     2.972189      15.481
       1970  |   11.25014   3.736967     3.01   0.004     3.736465    18.76382
       1971  |   18.10162   3.500476     5.17   0.000     11.06344    25.13979
       1972  |   14.55401   4.063154     3.58   0.001     6.384495    22.72353
       1973  |   16.59578   4.637619     3.58   0.001     7.271226    25.92034
       1974  |   17.45827   4.471475     3.90   0.000      8.46777    26.44878
       1975  |   20.93109   4.572276     4.58   0.000     11.73791    30.12427
       1976  |   17.23318   5.158467     3.34   0.002     6.861382    27.60497
       1977  |   19.26182   4.751529     4.05   0.000     9.708227     28.8154
       1978  |   15.95673   5.233447     3.05   0.004     5.434177    26.47928
       1979  |   14.37324   5.869618     2.45   0.018     2.571583     26.1749
       1980  |   11.58301    5.18697     2.23   0.030     1.153912    22.01212
       1981  |   14.28482   5.707315     2.50   0.016     2.809492    25.76014
       1982  |   11.28251   4.998155     2.26   0.029     1.233043    21.33197
       1983  |   10.74783    5.47755     1.96   0.056    -.2655208    21.76118
       1984  |   13.66639   5.721671     2.39   0.021     2.162195    25.17058
       1985  |   9.867876   6.168195     1.60   0.116    -2.534111    22.26986
       1986  |   15.17414   6.102758     2.49   0.016     2.903725    27.44456
       1987  |   16.46542   5.954447     2.77   0.008     4.493206    28.43764
       1988  |   14.22775   6.457503     2.20   0.032     1.244072    27.21143
       1989  |   13.51503   7.118574     1.90   0.064    -.7978169    27.82789
       1990  |   14.33783   6.929579     2.07   0.044     .4049788    28.27068
       1991  |   13.33514   6.816121     1.96   0.056    -.3695851    27.03987
       1992  |   12.48164   7.135318     1.75   0.087    -1.864873    26.82816
       1993  |   15.18881   7.411584     2.05   0.046     .2868251     30.0908
       1994  |   14.51899   7.469569     1.94   0.058    -.4995817    29.53757
       1995  |   12.96002   7.679496     1.69   0.098    -2.480639    28.40068
       1996  |   13.29211   7.955388     1.67   0.101     -2.70327    29.28749
             |
      lead15 |  -2.635591    4.28278    -0.62   0.541     -11.2467    5.975515
      lead14 |   4.119924   5.294739     0.78   0.440    -6.525863    14.76571
      lead13 |  -1.219058   4.859393    -0.25   0.803    -10.98952    8.551405
      lead12 |  -.1276021   6.852116    -0.02   0.985    -13.90471     13.6495
      lead11 |  -9.485776   4.018745    -2.36   0.022      -17.566   -1.405548
      lead10 |  -1.133975   4.904681    -0.23   0.818     -10.9955    8.727548
       lead9 |  -4.792218   3.658702    -1.31   0.196    -12.14853    2.564096
       lead8 |   -2.81221   3.937852    -0.71   0.479    -10.72979    5.105373
       lead7 |  -1.248013   4.360872    -0.29   0.776    -10.01613    7.520109
       lead6 |  -.7797544   2.980252    -0.26   0.795    -6.771954    5.212445
       lead5 |  -2.814077   2.615867    -1.08   0.287     -8.07363    2.445477
       lead4 |   .1759406   2.393439     0.07   0.942    -4.636391    4.988273
       lead3 |  -2.355844   2.951173    -0.80   0.429    -8.289574    3.577887
       lead2 |  -.5340147   2.507715    -0.21   0.832    -5.576115    4.508085
        lag0 |    .276763    2.71115     0.10   0.919    -5.174369    5.727895
        lag1 |  -1.616236   2.923493    -0.55   0.583    -7.494312     4.26184
        lag2 |  -1.692049   3.877192    -0.44   0.664    -9.487667    6.103568
        lag3 |  -.7508698   2.841648    -0.26   0.793    -6.464387    4.962647
        lag4 |  -2.960066   2.824574    -1.05   0.300    -8.639253     2.71912
        lag5 |   -2.38612   2.754708    -0.87   0.391    -7.924831     3.15259
        lag6 |  -3.297617   3.565538    -0.92   0.360    -10.46661    3.871378
        lag7 |  -5.131507   3.416243    -1.50   0.140    -12.00032    1.737311
        lag8 |  -6.981646   3.095878    -2.26   0.029    -13.20633   -.7569651
        lag9 |  -4.833179    3.09219    -1.56   0.125    -11.05044    1.384086
       lag10 |  -8.824309   3.682567    -2.40   0.021    -16.22861   -1.420012
       lag11 |  -7.337769   3.631075    -2.02   0.049    -14.63853   -.0370029
       lag12 |  -6.305412   4.066619    -1.55   0.128     -14.4819    1.871073
       lag13 |  -8.520276   3.951522    -2.16   0.036    -16.46534   -.5752084
       lag14 |  -6.850119   3.860921    -1.77   0.082    -14.61302    .9127823
       lag15 |  -9.341169   4.086061    -2.29   0.027    -17.55675   -1.125592
       _cons |   72.02055   7.006356    10.28   0.000     57.93332    86.10777
-------------+----------------------------------------------------------------
     sigma_u |  15.758038
     sigma_e |  10.797527
         rho |   .6804992   (fraction of variance due to u_i)
------------------------------------------------------------------------------

Bhalotra et al. (2023) Reserved Seat Quota Adoption and Maternal Mortality

webuse set www.damianclarke.net/stata/
webuse quota_example.dta, clear

// Generate indicator for GDP in bottom quartile
sum lngdp, d
gen GDPp25 = lngdp<r(p25)

// Generate time to adoption
gen timeToTreat = year-quotaYear

// Event study examining impact of reserved seats on maternal mortality
eventdd lnmmrt i.year, timevar(timeToTreat) method(hdfe, absorb(country) cluster(country)) lags(10) leads(10) accum graph_op(ytitle("ln(Maternal Mortality)") legend(order(2 "Point Estimate" 1 "95% CI") pos(6) rows(1)) ylabel(, format("%04.2f")))

eventdd lnmmrt i.year, timevar(timeToTreat) method(hdfe, absorb(country) cluster(country)) lags(10) leads(10) accum over(GDPp25) jitter(0.2) graph_op(ytitle("ln(Maternal Mortality)") legend(pos(6) order(2 "Point Estimate (GDP {&ge} 25 p)" 5 "Point Estimate (GDP < 25 p)" 1 "95% CI") rows(1)) ylabel(, format("%04.2f"))) coef_op(g1(ms(Sh)) g2(ms(Oh))) ci(rarea, g1(color(gs12%30)) g2(color(gs12%50))) 

Authors

eventdd is written by Kathya Tapia Schythe (University of California, Davis) and Damian Clarke (University of Chile & University of Exeter).

If you use eventdd and would like to cite it, please cite either the paper and/or the command's RePEc citation. BiBTeX citations for the paper are

@article{ClarkeTapiaSchythe2021,
  author = {Damian Clarke and Kathya Tapia-Schythe},
  title ={Implementing the panel event study},
  journal = {The Stata Journal},
  volume = {21},
  number = {4},
  pages = {853-884},
  year = {2021},
  doi = {10.1177/1536867X211063144},
}

Clarke, D., & Tapia-Schythe, K. (2021). Implementing the panel event study. The Stata Journal, 21(4), 853-884. https://doi.org/10.1177/1536867X211063144

References

Sonia Bhalotra, Damian Clarke, Joseph Gomes and Atheendar Venkataramani, "Maternal Mortality and Women's Political Power", Journal of the European Economic Association 21(5):2172–2208 (2023).
Damian Clarke and Kathya Tapia Schythe, "Implementing the Panel Event Study", Stata Journal 21(4):853–884 (2021).
Sergio Correia "Linear Models with High-Dimensional Fixed Effects: An Efficient and Feasible Estimator" Working Paper. http://scorreia.com/research/hdfe.pdf (2017).
Betsey Stevenson and Justin Wolfers, "Bargaining in the Shadow of the Law: Divorce Laws and Family Distress", The Quarterly Journal of Economics, 121(1):267-288 (2006).
David Roodman, Morten Nielsen, James MacKinnon and Matthew Webb "Fast and wild: Bootstrap inference in Stata using boottest", The Stata Journal, 19(1),4-60 (2019).