SDID time weights across 27 papers
All available cohort–outcome time-weight vectors from the first two tranches. Defaults are shown alone when they closely match the optimized reference. Otherwise, the plot and numeric table show both. Each paper opens with its previously preselected example; expand the paper to inspect every other outcome and cohort.
Display rule
Two vectors count as very close when no individual period changes by more than 0.1 percentage point and no more than 0.5% of total weight is reallocated (half the L1 distance). Both conditions must hold. Close examples use the default vector throughout the charts, numeric tables and CSV download. All its zeros and small weights are retained.
This is a presentation rule based on weight magnitudes, not a convergence or significance test. “Optimized” means more accurate numerical optimization at the same default penalty, including the completed high-precision refinements. Tiny weights and exact-zero support remain solver-sensitive.
Blue bars: author default · Orange line/points: optimized reference, when different
The CSV always includes all 754 returned examples under this display rule, irrespective of the search filter. This HTML contains everything needed for offline use.
60 unavailable attempts: zero default noise
These attempts have no substitute weights. They are separate from the 754 returned examples above.
Scope and interpretation
The figures concern the no-covariate SDID time-weight component. The author implementation is synthdid 0.0.9, commit 70c1ce3eac58e28c30b67435ca377bb48baa9b8a. The optimized reference retains the same tiny default penalty. Thirty-one blocks received 80-digit refinement across the two tranches; other references retain double precision. Objective-gap checks alone do not universally certify weight stability.
The second tranche uses complete-case subsets or shorter windows and explicitly shows retention. Some retained samples are very small. Original sample construction, repeated-cell aggregation and research-design caveats still apply; randomized audits and mixed-design/pending cases are labeled. These 754 blocks are dependent outcomes and cohorts from 27 papers, not independent or representative studies. No new fitting or change to original estimates was performed for this collection.
Horizontal axes count observed pre-periods, ending at −1. Hover or tap a plot for the original source-period label and weight. The values panel retains full exported precision. “Recent mass” means weight in the final ceil(20% × pre-period count) observations. Support counts use weight > 10⁻⁶ for description only; plotted and downloaded vectors are never thresholded.