Research note · 13 November 2026 · Ilya Nikitin, PhD · The Null Institute
Independent reanalysis of publicly available historical data. Version 1.0; not peer-reviewed.
Is Friday the 13th dangerous? A famous British study from 1993 found 65 hospital admissions after transport accidents on Friday the 13th, versus 45 on Friday the 6th. That’s 44% more admissions. The finding has been repeated for decades.
We checked the underlying data rather than just the headline. The public OpenIntro dataset contains 61 observations from the original work, covering traffic, shopping and hospital admissions. They tell a more complicated story.
What the 61 observations show
| Measure | Friday the 6th | Friday the 13th | Difference |
|---|---|---|---|
| Hospital admissions after transport accidents (6 date pairs) | 45 | 65 | +44.4% |
| Vehicles counted on sampled M25 stretches (10 observations) | 1,283,853 | 1,265,495 | −1.43% |
| Shoppers counted at nine supermarkets (45 observations) | 223,673 | 225,765 | +0.94% |
These are different kinds of counts, not 61 independent Friday-the-13ths. The motorway figures come from two stretches of road in five months; the shop figures from nine stores across five months. None of these samples represents the whole country. The traffic data cannot be used as a direct denominator for admissions from hospitals serving a different population.
The important statistical catch
The admission numbers come from just six pairs of dates. Five of the six had more admissions on the 13th. Whether that amounts to persuasive statistical evidence changes with the test used:
- A conventional two-sided paired t-test gives p = 0.042, just under the commonly used 0.05 threshold.
- A two-sided exact paired Wilcoxon test gives p = 0.094, above that threshold.
- A one-sided Wilcoxon test gives p = 0.047.
In plain English: the apparent statistical significance is not robust across reasonable analytical choices. That does not establish an error in the original paper, or prove that accidents cannot increase. It means six paired observations cannot carry the sweeping claim that the date is inherently dangerous. These are exploratory sensitivity tests, not tests selected before the 1993 data were observed.
What more recent evidence says
A 2026 Ontario study examined 254,058 coroner-investigated deaths from 2009–2023. Across 27 Fridays the 13th, researchers counted 45.7 cases per day, against 46.4 across the whole period. They found no evidence of an increase. These were cases investigated by coroners, not all deaths in Ontario, and this was a different outcome from the British hospital study.
Our conclusion: The evidence does not show a consistent increase in bad outcomes on Friday the 13th. Superstition might change behaviour, including travel decisions, but the historical observations cannot establish why traffic was lower. “Fewer journeys” and “greater risk per journey” are different questions.
Data and reproduction
We checked all 61 rows of the OpenIntro CSV, confirmed that its difference column matches the two count columns, and independently calculated the totals and paired tests. To reproduce the admission check with Python, download the CSV as friday.csv and run:
import pandas as pd
from scipy.stats import ttest_rel, wilcoxon
d = pd.read_csv("friday.csv")
a = d[d["type"] == "accident"]
print(a["sixth"].sum(), a["thirteenth"].sum())
print(ttest_rel(a["thirteenth"], a["sixth"]).pvalue)
print(wilcoxon(a["thirteenth"] - a["sixth"], method="exact").pvalue)Sources: Scanlon et al., BMJ (1993) · OpenIntro data and variable definitions · Rosenthal & Dmetrichuk, CHANCE (2026).