Identification of factors associated with various types of impaired driving

Vehicle crashes are among the leading causes of death worldwide, resulting in over 1 million deaths and 20 million severe injuries annually. Human factors play a crucial role, with about 30% of U.S. motor vehicle fatalities due to alcohol-impaired drivers, and fatigued and distracted drivers contributing an even higher share of fatalities than DUI. Impairment is not only due to alcohol or drugs but also distraction, fatigue, and emotional conditions. The literature exhibits definitional ambiguity regarding impaired driving; here DUI is defined as driving under the influence of drugs or alcohol. Past studies have also considered impairment arising from medical or cognitive conditions. Fatigue and drowsiness degrade attention, reaction time, and judgment, leading to severe crashes. Study objectives: (1) Highlight that impaired driving behaviors such as distraction, fatigue, and emotional conditions are as risky as DUI and warrant attention, using literature to show similarities with DUI. These impaired driving types accounted for almost half of fatalities in Wyoming during the study period, despite comprising about 10% of crashes. (2) Investigate similarities and dissimilarities in factors associated with DUI versus other impaired driving actions by disaggregating impaired drivers and estimating a unified multinomial model. Prior research suggests predictors of one risky behavior may not predict others. (3) Include predictors that can help foresee or suggest driver impairment by indirectly incorporating unseen factors likely associated with drivers’ attributes (e.g., vehicle type, environmental characteristics). The study focuses on associations, not causation, across DUI, fatigue, distraction, and emotional conditions, using crash data from Wyoming, a mountainous western U.S. state. This work challenges inconclusive definitions of impaired driving by comprehensively examining associated factors across multiple impairment types.

The literature is organized by impairment categories considered in this study: DUI, fatigue, distraction, and emotional conditions. Driving under the influence (DUI): Alcohol-related crashes incur over $40 billion annually. Systematic reviews show drugs like cocaine and cannabis significantly increase risk of severe and fatal crashes. Demographically, ages 24–34, males, whites, and in-state drivers show higher alcohol/drug involvement. Psychiatric comorbidity and criminal history are risk factors for DUI arrests. Despite extensive work on substance involvement and crash severity, fewer studies examine factors associated with DUI itself. Fatigued drivers: Fatigued driving is common; about half of U.S. adult drivers report drowsiness and 20% admit falling asleep while driving. Fatigue produces impairments comparable to moderate alcohol consumption; sleep deprivation slows reaction time more than a moderate blood alcohol level in some studies. Drowsiness is associated with disengagement and unsafe operation in simulators. Risk factors for sleep-related crashes include multiple jobs and night shifts. Both fatigue and alcohol impair physical/psychomotor characteristics. More research is needed on factors associated with fatigued driving. Distraction: Distraction can pose risks comparable to DUI and involves diverting attention from essential driving tasks. Common distractions include lack of concentration, adjusting in-vehicle equipment, looking outside the vehicle, and talking to passengers. Inattentive drivers show higher injury severity risk (e.g., +9.7% in single-vehicle crashes; +14.6% injury probability in two-vehicle crashes). Structural equation modeling with naturalistic driving data found higher distraction likelihood among younger drivers, at night, and in unclear weather. Cellphone use shifts gaze behavior and prompts speed adaptations depending on phone condition. Emotional condition: Emotional states (anger, sadness, depression) recorded at crash time can affect driving. Anger correlates with aggressive and risky driving and violations. Mood influences risk perception and attitudes; negative emotions increase risky driving. Feeling depressed and taking medications have been linked to higher crash involvement. Emotions affect risk perception and driving choices; situational anger relates to diminished control over incidents. Although direct studies comparing DUI with emotional states are limited, both can produce impaired driving behaviors. Overall, while many studies connect impaired driving types to crash severity, fewer examine factors associated with impaired-driving conditions themselves. This study addresses that gap in a unified framework.

Study design and framework: The study analyzes associations (not causation) between multiple factors and impaired driving categories: DUI (drugs/alcohol), fatigued, distracted, emotional, versus normal (reference). The framework treats drivers’ demographic characteristics as proxies for traits and uses vehicle type and environmental/roadway context as proxies for attitudes. Drivers’ actions at the time of crash are included as independent variables to capture behavioral manifestations potentially linked to impairment and other unobserved confounders. Data: Wyoming Department of Transportation (WYDOT) crash data from 2015–2019 were used. The dataset includes a driver condition indicator (DUI, distraction, fatigue, emotional, normal) and numerous explanatory variables. Variables retained as important include vehicle type (passenger car, pickup, SUV), environmental/temporal factors (icy road, dark lighting, non-clear weather, AADT, weekend, work zone, highway type), driver actions (drive too fast for conditions, failure to keep proper lane, follow too close, over-speeding, negotiating a curve), and driver characteristics (seatbelt use, license validity, citation record, gender, age, residency/vicinity). Descriptive stats indicate, for example, passenger cars 22%, pickups 21%, SUVs 11%; dark lighting 46%; weekend 27%. Response counts: DUI 1,615 (4%), fatigued 1,382 (3%), distracted 642 (2%), emotional 2,133 (5%), normal 34,162 (86%). Statistical analysis: A multinomial logit (MNL) model was estimated using the vector generalized linear model (VGLM) framework and solved via iteratively reweighted least squares (IRLS) with Fisher scoring (implemented in R’s VGAM package). The approach minimizes deviance and iterates until convergence. The model uses a transformed working response and applies Cholesky decomposition of the weight matrix to convert generalized least squares to ordinary least squares on transformed data. Constraint matrices allow excluding parameters, equating coefficients across categories, or estimating all freely to improve robustness and interpretability compared with standard MNL solved solely by maximum likelihood. Log-likelihood was computed from the multinomial PMF. A modified MNL with selective coefficient adjustments (equality constraints/removals) yielded improved goodness-of-fit over the standard MNL. The normal driver condition served as the reference category. Ethics and availability: Data are available upon request; no human subjects protocols were applicable as this is secondary analysis of de-identified crash records.

Descriptive context: Combined impaired-driving categories (DUI, fatigue, distraction, emotional) accounted for <10% of crashes but nearly half of fatalities in Wyoming during the study period. Response distribution: DUI 1,615 (4%), fatigued 1,382 (3%), distracted 642 (2%), emotional 2,133 (5%), normal 34,162 (86%). Vehicle type:

• SUVs were negatively associated with all impaired driving types (e.g., DUI Est. −1.05, p<0.05; fatigued −0.72, p<0.05; distracted −0.25, p=0.06; emotional −0.16, p=0.1), suggesting relatively safer driver profiles.
• Pickups were less likely to be impaired overall (e.g., DUI +0.50, p<0.05; fatigued −0.29, p<0.05; distracted −0.59, p<0.05; emotional −0.02, p=0.7). Passenger cars were positively associated with most impairment types (DUI +0.78, p<0.05; fatigued +0.19, p=0.005; emotional +0.35, p<0.05) but not distraction (−0.14, p<0.05). Environmental/temporal factors:
• Non-clear weather was negatively associated with DUI (−0.56), fatigue (−1.24), and distraction (−0.80) (all p<0.05); not significant for emotional (+0.03, p=0.7).
• Dark lighting was positively associated with DUI (+0.22, p<0.05) and emotional (+0.39, p<0.05), non-significant for fatigue (−0.04, p=0.5), and negatively associated with distraction (−0.81, p<0.05).
• Icy road conditions were negatively associated with all impaired categories (DUI −1.02; fatigue −2.10; distraction −1.31; emotional −0.74; all p<0.05).
• Higher traffic (AADT) was negatively associated with nearly all impaired driving (DUI −6.7e−05; fatigue −9.2e−05; distracted −9.8e−05; emotional −4.2e−05; all p<0.05).
• Weekends increased DUI (+0.13, p=0.03) but decreased fatigued (−0.29), distracted (−0.37), and emotional (−0.06, n.s.) driving.
• DUI was more likely on interstates (Highway type +0.58, p<0.005). Work zones were negatively associated with DUI (−0.49, p<0.005) and fatigue. Driver actions:
• Failure to keep proper lane was strongly and positively associated with all impaired categories (DUI +1.39; fatigue +1.84; distracted +1.46; emotional +1.09; all p<0.05).
• Driving too fast for conditions was negatively associated with DUI (−0.31, p=0.008), fatigue (−1.13, p<0.05), and emotional (−0.17, p=0.06), and not significant for distraction (−0.29, p=0.1).
• Following too close was negatively associated with DUI (−0.57, p<0.05), fatigue (−0.69, p<0.05), and emotional (−0.33, p=0.02), but positively associated with distraction (+0.76, p<0.05).
• Over-speeding was negatively associated with DUI (−0.77, p<0.05), distraction (−0.66, p<0.05), and emotional (−0.10, p=0.05), and positively associated with fatigue (+0.14, p=0.02). Negotiating a curve was positively associated with DUI (+0.65) and fatigue (+0.33), not significant for distraction, and positive for emotional (+0.20; all p-values as reported). Driver characteristics:
• Not wearing a seatbelt was strongly negatively associated (belted drivers less likely to be impaired) across all categories (DUI −2.98; fatigue −1.85; distracted −2.05; emotional −2.80; all p<0.05).
• Invalid driver’s license increased likelihood across all impaired types (DUI +1.53; fatigue +1.12; distracted +0.98; emotional +1.02; all p<0.05).
• Citation record (0–1 tickets vs others) was positively associated with DUI (+0.37, p<0.05) and negatively with other impairments (fatigue −1.86; distracted −1.60; emotional −0.94; all p<0.05).
• Female drivers were less likely to be impaired across categories (DUI −1.04; fatigue −0.91; distracted −0.41; emotional −0.61; all p<0.05).
• Younger age (≤42) increased impairment likelihood relative to older drivers for DUI (−1.19 indicates older drivers less likely), fatigue (−0.73), distraction (−0.75), but emotional showed a positive coefficient for older drivers (+0.22), indicating younger drivers less likely emotional impairment relative to older.
• Wyoming residency/vicinity effects suggested residents were generally less likely to be impaired (e.g., Residency −0.22, p=0.03; Vicinity negative across categories with small variations). Modeling: A modified MNL with constrained/equalized coefficients across categories improved fit versus a standard MNL (goodness-of-fit improvements noted). Overall, results suggest impaired drivers often avoid suboptimal conditions, DUI correlates with nighttime/weekends and short-term control deficits (lane keeping, curve negotiation), and seatbelt use, license validity, and lane-keeping violations are strong, consistent indicators across impairment types.

The term “impaired driving” is inconsistently defined, impeding effective policy and behavioral countermeasures. Risk perception and how situations are labeled shape decisions and behaviors; treating fatigue or inattention as less serious than DUI can reduce enforcement intensity and driver vigilance. Current regulations rigorously address DUI and commercial driver fatigue but provide limited mechanisms for non-commercial fatigue and distraction, often resulting in lesser violations (e.g., improper lane change) compared with DUI penalties. Aligning policies to recognize non-DUI impairments could discourage risky behaviors. Policy and enforcement implications include: sobriety-style checkpoints extended to fatigue and major distraction enforcement; high visibility patrols especially in favorable conditions when impaired drivers may choose to drive; using failure to keep proper lane as a key indicator triggering further driver condition checks; and prioritizing checks when encountering invalid licenses or unbelted drivers given their strong association with impairment. Enhanced public education and hazard warnings can reshape risk perceptions. Mass media campaigns that have reduced alcohol-impaired crashes could be extended to all impairment types. The results also suggest that autonomous vehicle technologies may mitigate risks for multiple impairment forms, but deeper understanding of behavioral processes is needed to quantify benefits. Overall, recognizing and addressing all impairment categories—not just DUI—can inform targeted enforcement, education, and engineering strategies to enhance roadway safety.

Key contributions: (i) Impaired drivers tend to compensate by avoiding less-than-optimal conditions (e.g., adverse weather, icy roads, high traffic), (ii) DUI differs meaningfully from other impairment types, indicating different underlying processes, (iii) license validity, seatbelt use, and failure to keep proper lane consistently signal impairment across categories and can serve as practical indicators, (iv) DUI drivers show deficits in shorter-term control tasks (lane keeping, curve negotiation) and are less associated with longer-term behaviors (driving too fast for conditions, speeding, following too close). The study recommends future research and policy to include distracted, emotional, and fatigued driving alongside DUI to reduce injury crashes and resolve definitional ambiguities around impaired driving.

Non-DUI impairments were not investigated using survey or driving simulation datasets, which could provide richer insights into drivers’ beliefs and behaviors when fatigued, distracted, or under emotional conditions. Thus, while associations are robust within crash data, complementary methods are needed to deepen understanding and inform interventions.