Event Study analysis in Stata - by Samsun - YouTube

Hello everyone, welcome back to our  channel. um this is Samsun. Today I will  

um talk about how you can  run an even study in Stata.

So there are lots of paper which is talk about  even studies you can find um i include some of  

them over here you can see. um and um so  before i proceed um so you can say what is  

even study? So i would say even study is another  version of uh difference-in-difference estimation.  

um and when uh do we need to use even study?  um so when um i would say when there is a  

variation in treatment time timing or  heterogeneous treatment effect. um this is um  

like the treatment occurs in different time for  different reason, place or a group of people.  

So that time um you can use an even study  instead of difference-in-difference,  

the regular difference-in-difference  estimation. Then um you can find um  

I have I mean it will be a good analysis  than difference-in-difference uh estimation.  

So today I will show you an example from uh  Stata. um um so if um and i'll uh use this command  

uh event dd. So if you don't have this command  installed in your Stata then you can just write  

that much um "ssc install eventdd", then it  will install in your computer. In my computer,  

it's already installed, so i can just um use comma  replace. So then it will install on my computer.  

Okay, you can see, it's installed  now. Now um I am going to load data  

from Stata. And this data is from  Stevenson and Wolfers (2006) um analysis on  

no-fault divorce reforms and female suicide and  you can find these from Web. So yeah i will run  

that much to load my data. So you can see  the load the data is loaded in my Stata.

Okay, so um I am going to browse uh to look at  how my data looks like. So you can see the first  

uh variable is um uh state um fips code,  that means they are states um and state code  

and then year um that means the year of  survey and this um _nfd uh this is a no-fault  

divorce onset that means uh um the time for each  state take a no-fault divorce reform. And um so  

if you look at this variable _nfd then you can see  um it's different for it is different for each um  

state. So maybe you can just tab stfips _nfd,  then you can see better. Yeah, here you can see  

a better um you can understand better. uh you  can see like this um state um 20 state code 20,  

the reform happened in 1960 um 1969 and  you can see this state code for 45 also  

happened in 1969. So yeah so you can see  um also you can see um and you can see the  

the reform happened from 1969 to 1988  and it's different for each state.

Okay, and um so our uh and you can see  this post means the post period of um  

of this reform uh or the I mean the period  begins from the reform to post um period.  

So for example, you can see  for state 1 um here is some  

some survey time but the reform happened in  1971. So until you can see until the is 1971  

for those year, the post is equal to zero and  from 1971 you can see the post equal to one um  

yeah for all of the period  which is um a greater than 1971.

so it is um true for all of the states.

And now um yeah so that means  post is our treatment dummy.  

And um our dependent variable is the women's  suicide rate, um women suicide um mortality.  

So this is our asmrs is our independent variable  sorry dependent variable uh our outcome variable.  

And uh this uh pcinc, asmrh, cases , weight,  and copop all of them are independent variables.  

So this one is per capita income, this is the  cases is the AFDC cases, copop is population.  

Okay, so now I am going to um

i am going to create a standard event time  variable uh which will give us um the time  

and the passage of the event for each state. So  what I'm doing um I am subtracting this reform  

time from our actual year. So if I run  this one, then um and if you browse again  

then you can see this variable now creating like  this year minus _nfd. So this is if 1964 minus  

1971, which is giving us negative 7 that means  7 years before um before the reform, okay.  

And like this one is the same year so this  is zero. And those are the post periods,  

so those are the positive years you can look at.

um now um so uh it is good for even studies,  we sometimes graph and we see how um  

this reform looks like for um every year. So um  so this is the command I will be using to find the  

even study plot uh which is eventdd. And this is  our the first one is our dependent variable or our  

outcome variable which is female suicide and those  three variables are independent or covariates and  

i will use a year and state fixed  effects. And this timevar(timetoTreat)  

this is the time uh will give us the  for every every period from the reform  

um period yeah every time from the reform and  I will be clustered at the state level. And  

those are the graph those are the options  for graphs. Now if I run this command,  

so you can see a nice graph for each period  of time the suicide per 1 million women  

before the reform and after the reform. So  these um you can see this solid uh vertical line  

um this is the baseline uh which is  omitted um in the in our analysis and um  

and also it indicates that the one year  

before to the adoption of the reform in each  reforming state. Okay, and these those are the the  

point um estimates are displayed and these  are the 95 percent confidence interval,  

okay. And um yeah and then you can see from the  graph how much is different for each period for  

each state it's very different um okay. So um so  you can um you can show a regression result for  

um for this even studies. So if you don't  specify the estimation method, eventdd  

uses Stata's regress command to estimate the  model by the ordinary least squares regression.  

um so if you specify the ols option, yeah you  will um obtain the same result over here method  

um over here method fe so you can use  um ols instead of fe. So um and if you  

specify this one like here I  specify fe that means it will  

um give us a fixed effect estimator uh from  these eventdd commands. um So if I run this one,  

so then this first one uh it gives us this  one is giving us the ols regression for this  

this one. You can see this is here  it's saying linear regression. But  

for this one uh I have specified that this method  will be a fixed effect and uh the estimator will  

be the fixed effect estimator. So then um  this is for the graph, so if I run this one,  

this will give us the graph for  from the fixed effect estimator  

and um for the model you can use uh that much um  until this the cluster uh cluster um state code.

so

then

yeah so you can see this is our result and you can  see it's a fixed effect uh regression analysis,  

so this coefficient now is um for  uh is a fixed effect estimator.

yeah and you can see those are for  the main um independent variable,  

these are the year fixed effects, and those are  the lead and lags. um that means that before um  

this the lead will be the after uh reform period  and this lag will be the pre pre-reformed period  

okay. uh also you can do a lots of things with  this eventdd command. um so I mean you you can  

specify as you need as you need for your research  um so if you run this command "help eventdd"

then you can see a lot of  information you can get for "eventdd"

and um yeah lots of information um you can use uh  from it. Yep, so I think uh this video might be  

helpful for you guys um. Please subscribe to our  channel, like our videos, share with your friends.  

Stay tuned!! Thanks for watching.  Bye. Please, Please, Please Subscribe.