Line chart parallel trends

[Joan Marti]

Hi,

I have data of the US election at county level since 2000 (variable "per_dem" - percentage vote for the democratic party), and am running a difference in difference analysis to check the impact of a treatment in the year 2016 ("treat=1"). In order to check for parallel trends, I want to create a line chart by year with two lines:

- Average % vote since 2000 for democratic party for counties treated in 2020
- Average % vote since 2000 for democratic party for counties not treated in 2020

Is there a simple way of telling stata to plot the yearly average of "per_dem" of counties with treat=1 and "per_dem" of countries with treat=2?

Thank you very much,

Joan

[Andrew Musau]

You need to collapse the data excluding county and graph. Below, I assume that the variable "treated2020" takes the value 1 in all years if a county is treated in the year 2020 and 0 otherwise.

preserve
collapse per_dem, by(year treated2020)
tw (line per_dem year if treated2020) (line per_dem year if  !treated2020)
restore

[Joan Marti]

Thank you very much Andrew, this works great but the problem is that my treatment variable only takes the value 1 for the year 2020 so I only have one line in the chart. How can I generate a treatment variable that takes the value of 1 for the treated counties also before the treatment took place?

Thank you again,

Joan

[Andrew Musau]

bys county: egen treated2020= max(treat==1 & year==2020)

[Joan Marti]

Thank you so much this works perfectly. One last question, my treatment has 4 different intensities so treated2020 can actually take the values 1, 2, 3 or 4. I tried to adapt your command to

#bys county: egen treated2020= max(treat==4 & year==2020)#

but it takes the value of 1 for all observations, how can I adapt the command so that the variable treated2020 can take the values 1, 2, 3 and 4?

Thanks again,

Joan

[Andrew Musau]

You can create indicators for each level separately:

forval i=1/4{
    bys county: egen t2020_`i'= max(treat==`i' & year==2020)
}

or if each county is associated with only one value in 2020,

bys county: egen treated2020= max(cond(year==2020, treat, .))

[Joan Marti]

Great, I managed to make it work with the above suggestion, thank you so much. Is there a way to also add the confidence intervals for each period?

[Andrew Musau]

twoway just graphs, it does not generate confidence intervals. If you have variables with the confidence intervals, sure.