Event-study figure of independent-poll polling error in ±4 weeks around each self-sponsored poll. Independent bins hover near zero error; the self-sponsored event-day point sits at +7.4 pp (SE 0.93). Visualizes the within-candidate trajectory test.
Results
Per-bin means of polling error (poll % − final vote %) for independent polls of the same candidate, by week relative to the self-sponsored poll. Each event = one self-sponsored poll; events without any independent poll in the ±4-week window are excluded so the t=0 self-sponsored marker is drawn on the same event set as the indep trajectory.
| Week | n_events | Mean error (pp) | SE |
|---|---|---|---|
| −4 | 19 | −1.74 | 2.88 |
| −3 | 28 | −1.28 | 2.56 |
| −2 | 32 | +1.21 | 2.26 |
| −1 | 38 | −1.44 | 1.30 |
| 0 (self) | 117 | +7.44 | 0.93 |
| +1 | 45 | −1.70 | 1.53 |
| +2 | 20 | +1.03 | 2.30 |
| +3 | 12 | +2.45 | 2.18 |
| +4 | 15 | +4.02 | 2.62 |
Interpretation
The visual punchline. Independent polls of the same candidate in the same race cluster within ±2 pp of zero error in the four weeks before AND the four weeks after the self-sponsored event. The self-sponsored point sits at +7.4 pp — six standard errors above the surrounding trajectory.
Closes the "real momentum" alternative mechanically: for the sponsored high to reflect a real spike in support, that spike would have to begin after the prior independent poll (week −1 already near zero), peak exactly on the sponsored-poll field-end day, and decay by the next independent poll (week +1 back near zero). A one-day spike is not a normal pattern of campaign momentum.
The mild post-event upward drift (+4 pp at week +4) is consistent with the candidate's actual share rising toward the final-vote reference as election day approaches; not noise around the sponsored event itself.
Follow-ups
- A variant restricted to within-race FE (subtract race-week mean of indep error in same window) might tighten the pre-event baseline further, but the unconditional version is cleaner for the headline figure.
- Could split by sponsor-route (A vs B vs C) once the methodology decomposition lands.