Open Hanh7 opened 12 months ago
Hi,
You can set target power to be anything you want. In the AER:I paper, I run simulations using both 0.5 (detect the pre-trend half the time) and 0.8 (detect it 80% of the time; a common threshold in power analyses).
J
On Thu, Sep 14, 2023 at 11:18 AM Hanh Pham Nguyen @.***> wrote:
Dear Dr. Roth,
I am reading your paper to understand why the targetpower in pretrends function is set to be 0.5. I wonder why it is not 0.7 or 0.79 or another number.
There is a sentence you said: "I find that linear violations of parallel trends that would be detected only 50 percent of the time can produce large biases in the treatment effects estimates and lead confidence intervals (CIs) to substantially under-cover the true effect. In the most extreme case, the bias from a trend detected only half the time is larger than the estimated treatment effect and a nominal 95 percent CI contains the true parameter only 24 percent of the time".
My question is that: The result of 50% power here comes from your analysis of characteristics of datasets in 12 papers (final sample) you red and surveyed. Am I correct?
— Reply to this email directly, view it on GitHub https://github.com/jonathandroth/pretrends/issues/5, or unsubscribe https://github.com/notifications/unsubscribe-auth/AE6EXFEZYRXGM3IUTFVWQ6LX2MN5ZANCNFSM6AAAAAA4YJGOWM . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Dear Dr. Roth,
I am reading your paper to understand why the targetpower in pretrends function is set to be 0.5. I wonder why it is not 0.7 or 0.79 or another number.
There is a sentence you said: "I find that linear violations of parallel trends that would be detected only 50 percent of the time can produce large biases in the treatment effects estimates and lead confidence intervals (CIs) to substantially under-cover the true effect. In the most extreme case, the bias from a trend detected only half the time is larger than the estimated treatment effect and a nominal 95 percent CI contains the true parameter only 24 percent of the time".
My question is that: The result of 50% power here comes from your analysis of characteristics of datasets in 12 papers (final sample) you red and surveyed. Am I correct?