Colorado data investigation

Higher gas prices in Colorado are linked to fewer crashes. Here is what that really means.

Name: Colorado Gas Prices vs. Crash Counts Analysis
Creator: CrashStory

CrashStory tested monthly Colorado crash counts against official Colorado gas prices. The strongest result is not “gas prices magically make roads safe.” It is narrower and more useful: after adjusting for seasonality and time trend, a 10-cent higher gas price one month earlier is associated with about 176 fewer crashes in the following month.

Updated 2026-03-15Formal window January 2023 to January 202637 full months, 36 lagged-model monthsEIA release 3/10/2026

Read the story Download CSV Download JSON report

Crash signal

176

fewer crashes associated with a 10-cent higher gas price one month earlier

Injury signal

fewer injuries associated with the same 10-cent move

Typical month

10,097

average monthly Colorado crashes in the formal January 2023 to January 2026 window

Live coverage

593,392

Colorado crashes in CrashStory's live summary through 2026-02-26

Short answer

In the formal model window from January 2023 through January 2026, Colorado’s lagged model points to fewer crashes and fewer injuries when gas gets more expensive.

A 10-cent higher gas price one month earlier is associated with roughly 176 fewer crashes and 63 fewer injuries in the next month. The fatality model also points negative, but because fatalities are much rarer and more volatile, that estimate is treated as directional rather than headline-safe.

This is still an association, not proof that higher gas prices directly cause fewer crashes. But it is a meaningful enough pattern to say that expensive fuel appears to reduce road exposure or trip volume enough to show up in crash counts.

What a 10-cent move means in a typical month

Crashes

1.74% of a typical month

Average month: 10,097. Modeled change from a 10-cent higher gas price one month earlier: 176 fewer.

Injuries

2.00% of a typical month

Average month: 3,153. Modeled change from a 10-cent higher gas price one month earlier: 63 fewer.

Fatalities

3.32% of a typical month

Average month: 70.8. Modeled change from a 10-cent higher gas price one month earlier: 2.4 fewer.

Why the raw pattern can fool you

If you split the lagged-price months into a low-price quartile and a high-price quartile, the high-price months actually show more raw crashes on average, not fewer.

Bottom-price months in this window average 9,773.3 crashes. Top-price months average 10,434.6.

That does not kill the hypothesis. It shows why seasonality matters. Higher gas months often line up with spring and summer travel months when road exposure is higher. That is why the adjusted models and year-over-year views matter more than a raw “high gas vs low gas” split.

This is an unadjusted split. It is useful for showing why raw comparisons can mislead, because high-gas months also tend to land in higher-travel spring and summer periods.

Why the formal window starts in 2023

CrashStory’s public monthly series currently exposes the most recent 60 months, from April 2021 through February 2026.

The monthly crash level steps up sharply in 2023. Before 2023, the public series averages about 1,486.8 crashes per month. From 2023 onward, it averages about 9,953.4.

Mixing those two regimes would blur the analysis, so the formal models use the cleaner post-break window from January 2023 through January 2026.

Charts that make the story easier to read

These charts answer different questions. One shows scale, one shows the cleaner year-over-year relationship, one shows the raw series, and one shows why Colorado gas should be compared with Colorado crashes rather than a generic U.S. price line.

Average monthly crashes, injuries, and fatalities compared with the modeled effect of a ten-cent gas-price increase. — How big the modeled effect is
This is the simplest magnitude chart on the page. It translates the model into the size of a typical month so the estimate is easier to understand.

Scatter plot of year-over-year gas-price changes versus crash-count changes in Colorado. — The cleaner relationship is year over year
Comparing each month with the same month a year earlier strips out some seasonality. That is why the year-over-year crash correlation is -0.54, even though the raw same-month views can look muddled.

Indexed Colorado gas prices, crash counts, and fatalities. — Crash counts soften more clearly than fatalities do
Indexed lines make the asymmetry visible. Crash counts move more smoothly and show a clearer negative pattern than fatalities, which are much noisier month to month.

Raw monthly lag correlations between gas prices and Colorado crash outcomes. — Raw lag correlations are not the final answer
The raw correlations stay positive because seasonality is strong. That does not contradict the adjusted model. It shows why controlling for calendar month matters.

Colorado and U.S. monthly gas prices over time. — Colorado broadly tracks the national gas story
Colorado usually moves with the national market, but the local line still matters because this page is testing Colorado crashes against Colorado prices.

What the statistics mean

Lagged 1 month

This month's crashes are compared with last month's gas price. That helps test whether driving changes after people feel the price shock.

Year-over-year change

A month is compared with the same month one year earlier. That reduces the distortion from seasonality, like summer driving and winter weather.

p-value

This is a measure of how surprising the estimate would be if there were really no relationship at all. Lower is stronger evidence, but it is not proof of causation.

This shows how much of the month-to-month variation the model explains. It is about fit, not about whether the relationship is causal.

Outcome	Model	Change per 10c	p-value	R2	Months
Crashes	Same month	-132.48	0.002	0.725	37
Crashes	Lagged 1 month	-175.88	<0.0001	0.7758	36
Injuries	Same month	-54.22	<0.0001	0.8663	37
Injuries	Lagged 1 month	-63.03	<0.0001	0.8838	36
Fatalities	Same month	-0.69	0.3341	0.758	37
Fatalities	Lagged 1 month	-2.35	0.0001	0.8142	36

What we can say, and what we cannot

After month-of-year controls and a linear trend, a 10-cent rise in Colorado gas prices one month earlier is associated with about 176 fewer crashes (p=<0.0001).
The same-month crash model also points negative, at roughly 132 fewer crashes per 10-cent increase (p=0.002).
The lagged fatality model also points negative, at about 2.4 fewer fatalities per 10-cent increase one month earlier (p=0.0001), but fatalities are rarer and this build still does not adjust for miles driven, so that estimate is more fragile than the crash-count result.
Year-over-year gas-price changes and crash-count changes are negatively correlated (-0.54), which supports the discretionary-driving hypothesis more than the raw same-month correlations do.
The raw correlations are not enough on their own because gas prices and crash counts both move seasonally. The adjusted models are the more important read.

The main caution is that this page uses crash counts, not exposure-normalized crash rates per mile driven. So the cleanest interpretation is still about driving exposure and crash volume, not a direct claim that higher prices themselves make individual trips safer.

Why EIA is the backbone, not TomTom or Mapbox

TomTom and Mapbox are useful for station discovery and real-time location workflows, but they are not the historical statewide price backbone for this analysis.

In the current stack, TomTom returned station discovery data but a sample station-ID lookup still came back with HTTP 404. Mapbox returned station metadata but no price fields.

That makes the Energy Information Administration the correct historical source for the price side of this page.

Limits and sources

This build uses CrashStory crash counts, injuries, and fatalities, not exposure-normalized crash rates per mile driven.
The public CrashStory monthly series only exposes the most recent 60 months, and the count level steps up in 2023, so the formal models default to 2023-01 onward.
TomTom provides current station discovery and documents a fuel-prices product, but we do not have historical station-level fuel price snapshots in the current stack.
Mapbox station discovery is useful for POI lookup and routing, not fuel prices.
The current outputs show association, not causation. Weather, commuting patterns, the economy, and enforcement can all affect crash counts alongside gas prices.

EIA Colorado weekly gas prices

https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=PET&s=EMM_EPMRU_PTE_SCO_DPG&f=W

EIA U.S. weekly gas prices

https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=PET&s=EMM_EPMRU_PTE_NUS_DPG&f=W

EIA gasoline dashboard

https://www.eia.gov/petroleum/gasdiesel/index.php

TomTom Fuel Prices API docs

https://developer.tomtom.com/fuel-prices-api/documentation/fuel-prices-api/fuel-prices-api

TomTom Fuel Prices coverage

https://developer.tomtom.com/fuel-prices-api/coverage/fuel-prices-coverage

Mapbox Search Box category search docs

https://docs.mapbox.com/api/search/search-box/#category-search

The next question is harder: do fatalities fall too?

Colorado gives the cleanest crash-count story. The national follow-up tests whether the same logic survives when you move to fatality rates and exposure-adjusted data.

Read the national analysis Explore Colorado crash data

CrashStory

Analyzing Data...

Higher gas prices in Colorado are linked to fewer crashes. Here is what that really means.

Updated 2026-03-15Formal window January 2023 to January 202637 full months, 36 lagged-model monthsEIA release 3/10/2026

Outcome

Model

Change per 10c

p-value

Months

Crashes

Same month

-132.48

0.002

0.725

Crashes

Lagged 1 month

-175.88

<0.0001

0.7758

Injuries

Same month

-54.22

<0.0001

0.8663

Injuries

Lagged 1 month

-63.03

<0.0001

0.8838

Fatalities

Same month

-0.69

0.3341

0.758

Fatalities

Lagged 1 month

-2.35

0.0001

0.8142

Why EIA is the backbone, not TomTom or Mapbox

TomTom and Mapbox are useful for station discovery and real-time location workflows, but they are not the historical statewide price backbone for this analysis.

In the current stack, TomTom returned station discovery data but a sample station-ID lookup still came back with HTTP 404. Mapbox returned station metadata but no price fields.

That makes the Energy Information Administration the correct historical source for the price side of this page.