Driving/Data

Baseline rates (verbatim; used directly in models)

The following U.S. national rates are taken verbatim from authoritative sources and are **not** converted here. Conversions to per-miles exposure are performed only inside the RiskModel calculations below.

• Fatalities: **1.26 deaths per 100 million vehicle miles traveled (VMT)** in 2023 (national estimate).

 * NHTSA 2023 Traffic Fatalities Estimates  
 * IIHS: Fatality statistics

• Injuries: **75 injuries per 100 million VMT** in 2022 (police-reported injuries, national estimate).

 * NHTSA: Traffic Safety Facts 2022

Distance Options

Distance choice	Miles
100 miles	100
1,000 miles	1000
100,000 miles	100000

These rows provide user-friendly exposure choices. The RiskModels convert the per-100M-VMT base rates into expected counts for the chosen miles.

Time of Day (modifier)

Time of day	Fatality risk multiplier	Injury risk multiplier
Day	1.0	1.0
Night	3.56	1.47

Per-mile risk is higher at night: in 2022, ~53.9% of fatalities and ~32.9% of injury crashes occurred during roughly ~25% of VMT (nighttime). Using exposure-adjusted ratios yields ≈3.56× (fatalities) and ≈1.47× (injuries).

• Night/day crash distribution and injuries: NHTSA 2022
• Background on time-of-day involvement: NHTSA
• Exposure assumption (≈25% of VMT at night): FHWA Lighting Guidance
• Supplemental analysis: University of Michigan per-mile risk

Seat-belt Usage (modifier)

Seat-belt usage	Fatality risk multiplier	Injury risk multiplier
Worn (car)	1.0	1.0
Not worn (car)	1.82	2.0
Worn (SUV/van/truck)	1.0	1.0
Not worn (SUV/van/truck)	2.5	2.857

Seat belts reduce fatal injury risk by ~45% in cars and ~60% in light trucks; moderate-to-critical injury by ~50% in cars and ~65% in light trucks. The multipliers above are the inverse of those reductions (i.e., increased risk when not wearing a belt).

• Effectiveness summary: IIHS: Seat belts
• Technical background: NHTSA: Seat belt effectiveness

Alcohol-impaired driving frequency (share of miles while impaired)

How often does the driver drive after drinking too much?	Approximate share of miles while impaired
Never	0
A few times per year	0.001
About once a month	0.005
About once a week	0.02
Several times a week	0.05

These shares approximate the fraction of total miles driven while impaired (not the fraction of people). They translate intuitive frequencies (“a few times per year”, “once a month”, etc.) into small exposure shares so the multiplier applies only to that portion of miles.

Calibration notes (assumptions; adjust as needed):

• “A few times per year” ≈ 0.1% of miles (e.g., ~10–20 impaired miles in ~10,000 annual miles).
• “About once a month” ≈ 0.5% of miles.
• “About once a week” ≈ 2% of miles.
• “Several times a week” ≈ 5% of miles.

Background/prevalence:

• CDC summary: self-reported alcohol-impaired driving episodes occur at a non-zero rate in the population.

Phone-based visual–manual distraction (share of miles while actively interacting)

How often does the driver interact with a phone (dialing/texting/etc.) while driving?	Approximate share of miles while actively interacting
None	0
One interaction per trip	0.005
Several interactions per trip	0.02
Lots of interactions per trip	0.05

These options reflect self-reported patterns of phone use while driving, translated into approximate shares of total miles driven while actively interacting with the phone (visual–manual tasks). The RiskModels apply a higher crash risk **only** to that fraction of miles, using a multiplier of ~6.0× for active distraction (consistent with naturalistic driving studies).

Evidence background:

• Visual–manual phone use (dialing, texting) is strongly associated with crash risk, with multipliers between ~6× and 12× depending on task.
• Talking (handheld or hands-free) is not consistently linked to higher crash risk, so it is not modeled here.

Speeding (Modifier)

Speeding status	Relative crash severity multiplier
At or below limit	1.0
Above limit	2.0

Speeding increases crash severity and overall risk—doubling speed quadruples kinetic energy, and speeding was involved in ~29% of U.S. traffic fatalities. Using a simplified 2× severity multiplier as a proxy.

• NHTSA: Speeding-related fatalities ~29%

---

  RiskModel: Driving/Data:fatality_model
Calculation: 1 - (2.718281828 ^ ( - ( (1.26 / 100000000) * distance_miles * time_fatality_multiplier * belt_fatality_multiplier * ( (1 - alcohol_frequency_share) + (alcohol_frequency_share * 4.0) ) * ( (1 - distraction_frequency_share) + (distraction_frequency_share * 6.0) ) * speeding_multiplier ) ))
    Content: 
Your estimated chance of being in a fatal crash is {{One_In_X|{result}}}.

  RiskModel: Driving/Data:injury_model
Calculation: 1 - (2.718281828 ^ ( - ( (75 / 100000000) * distance_miles * time_injury_multiplier * belt_injury_multiplier * ( (1 - alcohol_frequency_share) + (alcohol_frequency_share * 4.0) ) * ( (1 - distraction_frequency_share) + (distraction_frequency_share * 6.0) ) * speeding_multiplier ) ))
    Content: 
Your estimated chance of being injured in a crash is {{One_In_X|{result}}}.

  RiskModel: Driving/Data:any_crash_model
Calculation: 1 - (2.718281828 ^ ( - ( (185.54886112876233 / 100000000) * distance_miles * time_injury_multiplier * ( (1 - alcohol_frequency_share) + (alcohol_frequency_share * 4.0) ) * ( (1 - distraction_frequency_share) + (distraction_frequency_share * 6.0) ) * speeding_multiplier ) ))
    Content: 
Your estimated chance of being in a police-reported crash is {{One_In_X|{result}}}.

Calculation note: Each model computes an expected count λ from the verbatim base rates (per 100 million vehicle miles traveled), scaled by distance and modifiers, then converts to a probability \(p = 1 - e^{-λ}\) using \(2.718281828^{-\lambda}\). Alcohol and phone-based visual–manual distraction are applied via exposure shares: the model blends \((1 - \text{share})\times 1 + \text{share}\times 4.0\) for alcohol and \((1 - \text{share})\times 1 + \text{share}\times 6.0\) for distraction so elevated risk applies only to those fractions of miles.

---

Note on uncertainty: These are national averages and approximate modifiers. Actual risks vary by state, roadway type (urban/rural), weather, vehicle, driver demographics, and year-to-year changes. Modifiers are population-level and not individual predictions.

Initially generated by GPT-5 Thinking