Synthetic Control Method and In Space Placebo

[Thomas Chauvin]

Hello,

I am using STATA 14, an the synth package for conducting the synthetic control by Jens Hainmueller: https://web.stanford.edu/~jhain/synthpage.html

I am trying to conduct an "in Space" placebo test, and have followed STATALIST user Michael Jankowski's guide, which can be found: https://www.statalist.org/forums/for...control-method

As I have understood, this method calculates the difference between the actual unit and the synthetic control unit. If the graph is around zero, it means that the synthetic and the actual unit basically is the same. As you can see from "Synthetic Control France", it has a quite good fit. However, the plaebo graph seems to be different than the France graph (Orange is representing the difference between actual France and the synthetic control). Am I doing something wrong here? I tried to to the exact same thing with Abadie and Hainmuellers "smoking" dataset, and looks good compared to their synthetic control.

Not sure if I'm clear enough, but I hope you'll understand my question.

Cheers

[Jesse Wursten]

The orange line gives you the *difference* between france and synthetic france. The initial graph gives you the evolution of france and synthetic france separately. So I'd say the graphs look correct?

[Thomas Chauvin]

The orange line gives you the *difference* between france and synthetic france. The initial graph gives you the evolution of france and synthetic france separately. So I'd say the graphs look correct?

Hi Jesse, thanks for the speedy reply!

If you see the graph called "Synthetic Control France", we can see that the synthetic and the actual France is nearly identical all the way up to 2010, where there are some deviations. So, doesn't that mean that the Orange line should be very close to zero up to 2010, and then fluctuate a little bit? We observe from that graph that the orange line goes under zero in the middle of 2005 and 2010, whereas there are close to no deviations between these two according to "Synthetic Control France". We can also see that the actual France is below it's synthetic from 2014, while the orange graph indicates it is higher since it is above zero. Have I misunderstood the graph here? I though that if the synthetic and the actual was identical, the orange graph should just lay on zero and not fluctuate around it.

[Jesse Wursten]

It's hard to say because of the different scales. I recommend looking at the actual values produced by synth (iirc the actual and synthetic values are printed in the command window) and then comparing those to the value of your effect variable in the final dataset (the one used to create the graph). If the values of effect equal the difference between actual and synthetic, then it's correct. If not, something was wrongly specified somewhere.