Parallel Trends DID

[aish shukla]

Dear StataList Users,

I am using a Difference in Difference Model but I am facing a challenge in establishing the parallel trends assumption. What I have usually seen is that parallel trends assumption is established by plotting a trendline that visually shows the Treatment Control Groups to be parallel in the pre-treatment time period. I am wondering how to plot these trend lines accurately after considering the controls.
For example, if there is a dataset on house selling prices over many counties and one is looking at the effect of the launch of an online portal (exogenous launch in several counties) on the selling price as the treatment. Then in such a case, I am looking to establish parallel trends assumption by showing that in the pre-treatment period the selling prices of houses in control and treatment groups are parallel (after controlling for observables such as square footage, county crime rate, etc., Busy street (0/1), etc. ).

What would be the best way to formally establish the parallel trends assumption in such a case? In other words, how to account for the controls?

[Clyde Schechter]

Take your data set and drop all the post-intervention observations. Then graph housing prices against time in the two groups and observe whether the resulting curves appear roughly parallel. If you need to adjust for other variables, you can build a regression model of prices as a function of treatment group and those other variables, and time. Then use -margins- and -marginsplot- to graph the adjusted housing prices in both groups over time.

If you want more specific advice, please use the -dataex- command to post an example of your data, and explain the variable in your data.