search

ebenmichael / augsynth

Augmented Synthetic Control Method
MIT License
147 stars 52 forks source link

Confidence interval that includes the value of zero #75

Open Germancampos opened 2 years ago

Germancampos commented 2 years ago

Hello everyone, again I want to ask for your help.

I am estimating with multysynth but my confidence interval on the graph includes the value of 0, is this enough proof that the intervention has no significant effect on treated units despite the synthetic control fitting very well and despite that there is a noticeable trend of impact after treatment?

In the package examples I saw that for augsynth there were ways to improve the confidence intervals, changing parameters in the main function or in the summary. Are there also options to improve confidence intervals in multysynth?

Thank you very much

tahokovi commented 1 year ago

I can't answer specifically about the package. However, I would consider the width of your confidence intervals before suggesting that your intervention has had no effect. If your confidence intervals include large positive and negative effects, I would be more inclined to say that is indicative of noisy data and statistical uncertainty rather than the absence of any effect of the intervention.