How to calculate att.avg for a subgroup of the treated units and obtain the standard error, using all observations?

[scarlettdeng22]

Basically, how to replicate Table 3 in your paper. Could you help?

[feliuserra]

Hi, I was wondering exactly the same thing.

My first guess were two options: One, only include the treated values for each of the columns of Table 3 on separate gsynths, this way we would obtain standard errors from all the rest of the observations that are non-treated; or two, include them all at once, and then average the subgroups of the treatment over the different groups since we can access the counterfactuals created separately. Despite I have only seen the counterfactuals estimated through the plots, we should be able to obtain different for the different subgroups, although I'm not sure about how the standard errors would be computed.

The point here is to esimate heterogeneous effects by stratifying or some similar technique. Am I wrong?

Any help would be appreciated, if someone can give any hint it would be great. I can potentially make a PR if needed.

Thanks

[feliuserra]

Hi,

I guess that this has been solved with the new version of the package (1.1.4), where you can estimate the cumulative effect, but also sub-groups effects with confidence intervals. Note this on the new version (https://yiqingxu.org/software/gsynth/gsynth_examples.html):

"One can also calculate average treatment effect for a sub-group by specifying unit names in option id. By specifying cumu = FALSE, average treatment effects (rather than cumulative effects) at each period will be returned. Note that in this case, parametric bootstrap procedure is needed for uncertainty estimates."

[scarlettdeng22]

Thank you so much for sharing this! I’ll check that.

Thanks, Xueting

On Apr 9, 2020, at 7:22 AM, Feliu Serra Burriel wrote:

 Hi,

I guess that this has been solved with the new version of the package (1.1.4), where you can estimate the cumulative effect, but also sub-groups effects with confidence intervals. Note this on the new version (https://yiqingxu.org/software/gsynth/gsynth_examples.html):

"One can also calculate average treatment effect for a sub-group by specifying unit names in option id. By specifying cumu = FALSE, average treatment effects (rather than cumulative effects) at each period will be returned. Note that in this case, parametric bootstrap procedure is needed for uncertainty estimates."

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[feliuserra]

Hi, don't get me wrong, by no means I am collaborating on the development of this software, I am just a regular user. But yeah, I believe your question has been somehow answered. Also I believe that the first comment I made was wrong. Anyway, cheers to the authors.

Feliu

[scarlettdeng22]

Hi,

Thank you for clarifying and providing updates! You’re right, cheers for the author and the package!

Xueting

On Apr 9, 2020, at 10:20 AM, Feliu Serra Burriel notifications@github.com wrote:

 Hi, don't get me wrong, by no means I am collaborating on the development of this software, I am just a regular user. But yeah, I believe your question has been somehow answered. Also I believe that the first comment I made was wrong. Anyway, cheers to the authors.

Feliu

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[Katebellis]

Hi all,

I was wondering whether anyone knows how to get the att.avg and standard error for a subgroup, rather than the att at each period, using cumuEff?

Many thanks in advance