Hi, I'm working on a reverse diff-in-diff analysis on sentencing after a circuit split was resolved by the Supreme Court. There was a period of 6-18 months (depending on the circuit) after circuits had decided a very particular sentencing question that can be summarized as follows: "In light of the limited sentencing discretion created by the Supreme Court's Booker decision, do sentencing judges have the discretion to ignore congress's 100-to-1 sentence disparity for crack cocaine, "and before the Supreme Court resolved the split in favor of judicial discretion in Kimbrough. There are 12 years of data after Kimbrough.

I have two questions about best practices for the length of time analyzed on both sides of a diff-in-diff. Last time I used a (normal, not-reversed) diff-in-diff analysis on the same dataset, I used exactly one sentencing year on both sides of the analysis. I'm now wondering if I should change that up. First, can I use varying lengths of time per circuit for the pre-convergence portion (6-18 months)? Or should I limit my pre-convergence groups to the period after which the last circuit made their decision (6 months)? I'll have thousands of convictions in either case, so I'm not worried about power. Second, how much should I limit the post convergence data? Should I use just one year, or all 12 years available to me, or somewhere in between? Again, power is not the issue, as there are no shortage of cocaine convictions in federal court.

Thanks in advance for any help.