🔒 All inputs are processed locally in your browser. Nothing is transmitted, stored, or logged. Inputs disappear when you close the tab.
SELECTED STUDIES · METHODS · AL-187 · #173
Parallel-Trends & Placebo Checker
A single pre-period and a single post-period is enough to compute a difference-in-differences estimate, but it is not enough to test the assumption that estimate depends on: that the treatment and control groups would have moved in parallel absent treatment. Enter three or more pre-treatment periods for both groups to get a pre-trend slope gap with a t-test, a period-by-period event-study table, and an in-time placebo test that reruns the estimate at a fake treatment date inside the pre-period. A classic single pre/post 2×2 tab ships for completeness only.
Method: closed-form ordinary least squares on your entered period values, with a numerically implemented t-distribution CDF for the t-tests. No live data. The built-in worked example is an illustrative synthetic panel, not a historical dataset.
Group Labels
Period Panel
Pre-Treatment Rates by Period
Pre-Trend Slope Gap (Pooled Regression, Group × Period Interaction)
Uses the same period panel entered in the Pre-Trend Test tab. The baseline period is the last pre-treatment period before the treatment-start selection.
Period-by-Period Treatment-Minus-Control Difference, Relative to Baseline
▶ How to Read This
Uses the pre-treatment periods from the Pre-Trend Test tab only. Pick a period inside the pre-treatment window to treat as a fake treatment date; everything from that period onward inside the pre-window is treated as the fake post-period.
Fake Treatment Date
Placebo Difference-in-Differences Estimate
This tab computes the classic single pre-period, single post-period 2×2 difference-in-differences estimate from four group means. It exists only so all four modes live in one file. This design is already served by other difference-in-differences calculators in this suite; the pre-trend, event-study, and placebo tabs above are the actual contribution of this tool. Four means alone carry no variance information, so no standard error or significance test is reported here.