When the home team loses, the police get busier

Every Super Bowl weekend the same statistic circulates. Domestic violence spikes on game day. The statistic as usually told is wrong. The real finding underneath it is stranger and better supported, and it is not about the Super Bowl at all.

The claim

David Card (UC Berkeley, later a Nobel laureate) and Gordon Dahl (UC San Diego) published “Family Violence and Football” in the Quarterly Journal of Economics in 2011. They found that when an NFL team that was expected to win loses, police reports of at-home violence by men against their wives and girlfriends in that team’s area rise by about 10 percent in the hours after the game. A loss the team was expected to suffer does nothing. An unexpected win does nothing either. Only the upset matters.

How they found it

The clever part is the point spread. Las Vegas bookmakers publish a pre-game line that summarises what everyone expects. Card and Dahl defined an upset loss as a home team that was favoured by four or more points and lost anyway. That gives you a clean split between “bad outcome” and “surprisingly bad outcome,” which is exactly the distinction their theory needs.

The violence data comes from the National Incident-Based Reporting System, a federal database of individual police reports. They pulled records from 763 police jurisdictions in the home states of six NFL teams, the Panthers, Lions, Patriots, Broncos, Chiefs and Titans, across regular seasons from 1995 to 2006. The outcome is incidents of violence by a man against an intimate partner at home, reported to police on the Sunday of a game, compared with Sundays when the local team was not playing. Controls include the point spread itself and the size of the local TV audience.

The timing is the strongest piece of evidence. The extra calls cluster in a narrow window from roughly the final hour of the game through the two hours after. Not the morning before, not the following day.

The mechanism

The authors frame it as a reference-point story. Fans set their expectations from the spread. A loss that was priced in hurts a little. A loss that was not priced in is a sharp negative surprise, and a small share of men act out on that surprise. Consistent with that, the effect is larger for games that matter more. Rivalry games roughly double it, 20 percent versus 8 percent for non-rivalry games. Games with playoff implications and games with a fourth-quarter collapse also push it up.

How much to believe it

Replicated, with a twist. A 2026 paper in the International Journal of Environmental Research and Public Health revisited the design 15 years later with newer NIBRS data and again found roughly a 10 percent rise after upset losses, while noting the relationship has shifted over time. Outside the NFL, several studies of soccer in the UK and elsewhere find spikes in domestic abuse calls around matches, though not always tied to surprise.

The strongest criticism is about the mechanism, not the spike. Ria Ivandic, Tom Kirchmaier, Neus Torres-Blas and Arash Saeidi matched individual-level domestic abuse call data from Greater Manchester to almost 800 Manchester United and City games from 2012 to 2019 (Journal of Public Economics, 2024). They found abuse calls fall about 5 percent during the two hours of a match, then climb and peak around 8.5 percent above normal roughly ten hours after kickoff. The whole rise comes from male perpetrators who had been drinking. Among perpetrators who had not, there was no significant increase. And on days when the bookmakers’ favourite lost, the rise was no bigger than on days the result went as predicted. In Manchester, alcohol explains the spike and emotional shock explains nothing.

That does not overturn Card and Dahl. NIBRS does not record whether the offender was drunk, so the US paper cannot separate the two. It does mean the “losing makes men violent” reading is weaker than it sounds. The honest summary is that game days raise domestic abuse in more than one country, that the timing in the US points at the final whistle, and that the best-measured study says booze is doing the work.

Sample and generalisation. Six teams and 763 jurisdictions is plenty for statistical power, but it is six teams, chosen because their states report to NIBRS. The effect is about reported incidents, which are a fraction of actual ones, and reporting itself could plausibly change on a game night.

Does it matter in real life? A 10 percent rise on a handful of Sundays a year is a small share of annual domestic violence. It is large enough for police scheduling, and it makes a strong case that violence responds to short-run triggers rather than only to long-run character, which is the finding the authors actually cared about.

What it means for you

If you are analysing event effects, the point-spread trick generalises. Any time you have a market or forecast that encodes expectations, you can split “bad” from “surprisingly bad,” and the surprise is often where the signal is.

Suggested chart

Hourly domestic abuse calls relative to a non-match baseline, from kickoff to 14 hours after, split by whether the perpetrator was recorded as drinking. The Manchester paper reports this in its event-study figures.

Sources

1. Card, D. and Dahl, G. B., 2011, “Family Violence and Football: The Effect of Unexpected Emotional Cues on Violent Behavior,” Quarterly Journal of Economics, 126(1), 103-143. academic.oup.com. Open PDF at davidcard.berkeley.edu.
2. Ivandic, R., Kirchmaier, T., Saeidi, A. and Torres-Blas, N., 2024, “Football, alcohol, and domestic abuse,” Journal of Public Economics. sciencedirect.com. Discussion paper version at cep.lse.ac.uk.
3. “‘Family Violence and Football’ at 15: Revisiting the Effect of Emotional Cues on Intimate Partner Violence,” 2026, International Journal of Environmental Research and Public Health, 23(6), 768. doi.org.