Data & Research / Holiday crash statistics
Do Arizona Holidays Have More Alcohol-Related Fatal Crashes Than Expected?
Six major holidays, six years of ADOT data, and one question: how do actual alcohol-related fatal-crash counts compare with what their reporting hours would ordinarily predict?
Do the holidays show an overall increase?
Across all six holidays and all six years combined, the observed number of alcohol-related fatal crashes was nearly identical to the modeled number expected for the exact days and hours included in their reporting periods.
Important: ADOT recorded 118 deaths in these alcohol-related fatal crashes, but this model estimates expected fatal crashes, not expected deaths. A difference this small is not evidence of a meaningful overall increase or decrease. These six selected holiday periods are not the full calendar year, and the proxy is not adjusted for vehicle miles driven or uncertainty in its estimated time pattern.
Which holidays were above or below the model's expectation?
Percent difference between observed and modeled expected alcohol-related fatal crashes for each holiday's six reporting periods (2019–2024). Sorted from greatest positive difference to greatest negative difference. Expectations adjust for the included weekdays, hours and period lengths.
Actual / expected shows the raw number of alcohol-related fatal crashes followed by the modeled expected number, summed across each holiday's six reporting periods. The center line means observed equals modeled expected. The bar extends right in hot pink when observed crashes exceed expectation and left in muted purple when they fall below it. Each bar shows one holiday's percentage difference—not a measure of driving risk, a causal holiday effect, or a proven statistically significant result.
The modeled differences range from about +19% for New Year's to −22% for Christmas. With only 10–20 observed alcohol-related fatal crashes per holiday over six years, differences of just a few crashes can substantially change these percentages. The ordering is descriptive, not a ranking of genuinely safer or more dangerous holidays.
How to interpret the holiday comparisons
Why a straight holiday comparison isn't fair
In our six-year analysis of all alcohol-related crashes, 43.1% happened on Saturday or Sunday; including Friday raises that share to 58.6%. Our hourly research also shows a strong evening and overnight concentration. A long weekend is therefore expected to have more crashes than a similarly long midweek period, even without a holiday.
Thanksgiving always includes Friday, Saturday and Sunday. In 2024, Christmas included none of them. Comparing total crashes or merely dividing by the number of calendar days ignores that difference.
Our modeled expectation accounts for the actual weekday and hour of every hour in each holiday reporting window. It uses the pattern of all alcohol-related crashes as a proxy, adjusted to each year's totals and scaled to its annual alcohol-related fatal-crash count. This is substantially more informative than raw counts, but it still does not account for traffic volume or different probabilities that crashes will be fatal.
Exactly which days does ADOT count?
A holiday is more than the calendar date itself. ADOT's fatal-crash reporting window starts at 6 p.m. before the designated holiday period and ends at 5:59 a.m. after it. ADOT labels a Friday-to-Tuesday window as four days, although the actual elapsed time is 3½ days (84 hours). The model uses the real hours, not the rounded day labels.
Here are the official 2024 reporting windows. New Year's, July 4th and Christmas shift between weekdays in other years; our calculations use the dates from each of the six annual reports.
| Holiday | Start, 6 p.m. | End, 5:59 a.m. | Elapsed time |
|---|---|---|---|
| New Year's | Fri Dec. 29, 2023 | Tue Jan. 2, 2024 | 84 hours |
| Memorial Day | Fri May 24 | Tue May 28 | 84 hours |
| July 4th | Wed July 3 | Mon July 8 | 108 hours |
| Labor Day | Fri Aug. 30 | Tue Sept. 3 | 84 hours |
| Thanksgiving | Wed Nov. 27 | Mon Dec. 2 | 108 hours |
| Christmas | Tue Dec. 24 | Thu Dec. 26 | 36 hours |
The examples above are for 2024 only. Download all 36 exact holiday windows (2019–2024).
Related research: day, hour and month
Source tables, exact calculations and downloads
What we counted
Observed counts: ADOT Arizona Motor Vehicle Crash Facts Table 1-9, “Holiday Crash Statistics,” for each year 2019–2024. These are crashes in which alcohol was reported as involved and at least one person died—not the total number of deaths or a count of legal DUI convictions.
Time-pattern inputs: ADOT Table 6-14, “Alcohol Related Crashes by Hour and Day of Week.” For each year we used the reported counts of all alcohol-related crashes by weekday and by hour. ADOT 2024 Table 6-14 provides a complete 7-day × 24-hour joint pattern used as a starting template. Each year's total number of alcohol-related fatal crashes is used to scale the model (234, 163, 215, 202, 308 and 312 respectively).
How the expected counts are estimated
We adjust the 2024 joint weekday/hour pattern to match each year's own seven weekday and 24 hourly all alcohol-related crash totals using iterative proportional fitting. We scale the fitted distribution to the corresponding year's total of alcohol-related fatal crashes, divide each weekday/hour count by the number of occurrences of that weekday in the year (52 or 53), and add the expected counts for every hour contained in that year's exact ADOT holiday window. For a reporting window crossing New Year's, the named report year's baseline is applied to the full window.
This model assumes that the usual all-crash weekday-hour pattern is a reasonable proxy for the fatal-crash timing pattern; that assumption can be imperfect. As a sensitivity check, a simpler model using independent weekday and hour distributions produced very similar six-year results. Neither is a traffic-volume-adjusted risk model or a statistical test of causation. The holiday itself contributes to the historical baseline, so the estimates cannot cleanly isolate a holiday effect. The model also assumes all alcohol-related crash timing patterns approximate fatal-crash timing. Small fatal-crash counts mean the percentages are uncertain. We have not estimated confidence intervals that include uncertainty in the model.
| Holiday | Observed fatal crashes | Modeled expected | Observed / expected |
|---|---|---|---|
| New Year's | 20 | 16.79 | 1.19× |
| July 4th | 18 | 16.22 | 1.11× |
| Labor Day | 19 | 17.41 | 1.09× |
| Thanksgiving | 20 | 21.51 | 0.93× |
| Memorial Day | 14 | 17.41 | 0.80× |
| Christmas | 10 | 12.82 | 0.78× |
Download observed-vs-expected holiday summary (CSV)
Download 36 year-by-holiday calculations (CSV)
Download original ADOT holiday fatal-crash observations (CSV)
Download joint weekday-hour template (CSV) · Annual weekday totals · Annual hourly totals · Annual fatal-crash counts
View the complete reproducible calculation code on GitHub.