The day in conversation for 2026-10-31

The discussion: Tom: That sensitivity analysis work on irregular assessment times seems crucial because standard methods often assume fixed observation intervals; how robust is that root-n consistency claim when the underlying process has long-term dependence? Jane: The robustness hinges entirely on the regularity conditions mentioned, Tom. I'm less concerned with the asymptotic rate and more interested in what happens when those conditions break down in real-world observational data sets—specifically, what does that data-driven procedure for selecting the sensitivity parameter actually look like in practice? Lu: From my perspective on earlier days, that selection process seems inherently tied to prior knowledge about the variance structure. Does the procedure explicitly incorporate any prior assumptions about the noise characteristics, or is it purely empirical based on observed uncertainty? Meng: I wonder how that relates to what we discussed regarding the Beta Regression paper. When dealing with proportion outcomes in interrupted time series, if we use an influence function approach for sensitivity analysis on those trials, does that estimator offer a clearer path toward better calibration