Parallel Trends: Plotting Without Performing

Difference-in-differences leans on parallel trends in the absence of treatment. Visualizing pre-treatment dynamics helps readers assess plausibility, but plots can mislead if outcomes are noisy or samples shift. Bin event time thoughtfully: too many leads inflate noise; too few hide meaningful dynamics. Show raw means alongside model-based estimates when possible, and document sample restrictions that change composition. If pre-trends drift, discuss whether controls, alternative time windows, or alternative comparison groups are warranted — and document deviations from the preregistered plan transparently.