Within-Person and Between-Person Risk Perceptions Predict Driving After Drinking: An Ecological Momentary Assessment Study

Abstract

Background:

Alcohol-impaired driving (AID) remains a significant public health concern, accounting for approximately 30% of vehicle fatalities annually in the United States. Despite widely negative attitudes toward AID, many adults report driving after drinking. Day-to-day changes in how people think about AID may explain this inconsistency. The current study used ecological momentary assessment (EMA) to examine how both stable patterns and day-to-day changes in perceived danger of driving, willingness to drive, and subjective intoxication predict real-world driving after drinking.

Methods:

Participants (N = 113) from a Midwestern community completed six weeks of EMA data collection. Daily morning surveys assessed the previous night’s drinking and transportation decisions, while repeated evening surveys measured perceived danger of driving, willingness to drive, subjective intoxication, and breath alcohol concentration (BrAC) using portable breathalyzer devices. Mixed-effects models examined day- and person-level cognitions as predictors of driving behavior while accounting for alcohol consumption via BrAC.

Results:

Participants reported driving on nearly half of drinking days. On days when people perceived driving as more dangerous and felt more intoxicated than usual, they were less likely to drive. Moreover, on days when people were more willing to drive than usual, they were more likely to drive. These patterns held after accounting for BrAC. Additionally, people who typically experience stronger subjective effects of alcohol were less likely to drive on days when their breathalyzer readings were high (BrAC ≥ 0.05%).

Conclusions:

Day-to-day changes in how people perceive the dangers of driving and how intoxicated they feel predict whether they drive after drinking, even at high BrAC. People who typically experience strong subjective effects from alcohol appear especially responsive to objective indicators of high intoxication levels. Findings support developing real-time interventions that combine subjective feelings of intoxication with objective measures like breathalyzers to promote safer transportation choices.

Introduction

Alcohol-impaired driving (AID) continues to be a major public health concern in the United States, accounting for 30% of traffic fatalities, totaling 12,429 deaths in 2023 (National Center for Statistics and Analysis, 2025). In the United States, AID is legally defined using a per se blood alcohol concentration (BAC) limit of 0.08% in most states, although the National Transportation Safety Board recommends a change to a 0.05% BAC limit, in line with that of many other countries (National Transportation Safety Board, 2013). There are an estimated 110 million AID events in the US every year, resulting in $52 billion in economic costs annually (Blincoe et al., 2015, Centers for Disease Control and Prevention [CDC], 2022).

Despite widely negative views of AID, approximately 20% of driving age adults report driving within 2 hours of drinking in the past year (Drew et al., 2010). Adults are more likely than other age cohorts to engage in AID (Paschall, 2003), with nearly 3 million adults (ages 21 to 25) reporting AID every year (Center for Behavioral Health Statistics and Quality [CBHSQ]. 2018). One possible explanation for this inconsistency is within-person variability in AID cognitions – that despite generally negative views of AID, individuals may engage in AID when they either underestimate their current risk of driving or misperceive their current level of intoxication.

The establishment of a per se BAC limit for legally defining AID has had a significant impact on AID decision making (Fell and Voas, 2014). However, the effectiveness of per se laws on AID decision making depends, in part, on individuals’ ability to estimate their BAC when making driving decisions after drinking (Davis-Stober et al., 2019). Studies have shown that most individuals are poor estimators of their current BAC (Beirness, 1987, Eby et al., 2017, Liguori et al., 1999), especially at high levels (Beirness et al., 1993).

Many AID prevention efforts are predicated on increasing the perceived likelihood of negative consequences to decrease rates of this behavior (Piquero and Paternoster, 1998, Shults et al., 2001). However, this approach faces a fundamental challenge: alcohol intoxication itself decreases the perceived danger of risky behaviors (Davis et al., 2007, Fromme et al., 1999, Fromme et al., 1997, Maisto et al., 2002), including AID (Morris et al., 2014). This undermines prevention effectiveness, as individuals at higher BACs are more likely to underestimate the perceived risk of driving at their current level of intoxication (Amlung et al., 2014, Morris et al., 2014).

Individual cognitions play a central role in shaping AID behavior. Cross-sectional studies consistently show that general attitudes regarding AID (e.g., expected AID penalties; willingness to engage in AID) strongly predict AID engagement (Baum, 2000), and individuals are more likely to engage in AID when they perceive it as less dangerous (Bingham et al., 2007). Longitudinal research further demonstrates that permissive attitudes toward AID uniquely predict increased AID frequency over time (McCarthy and Pedersen, 2009). These studies typically assess attitudes toward AID measured at a single time point while individuals are sober, reflecting general beliefs and evaluations about the behavior held outside of drinking contexts. This approach cannot explain why the same individual sometimes drives after drinking and sometimes does not—that is, within-person variability in drinking and driving behavior. Aggregating repeated momentary assessments across multiple drinking episodes, as done in ecological momentary assessment (EMA) research, provides a conceptually different type of person-level measure that reflects an individual’s typical cognitive state during actual drinking episodes rather than stable attitudes held while sober. Moreover, EMA designs enable examination of day-to-day variation within individuals that may impact situational decisions about transportation.

Few studies have examined within-subjects influences on AID. Quinn and Fromme (2012) used a daily diary approach and found that drinking more alcohol than usual coupled with lower subjective intoxication predicted a greater likelihood of driving after drinking. In a two-week pilot study (N = 36) using EMA, Motschman et al. (2020) examined within-person, event-level variability in AID perceptions during real-world drinking episodes. Results indicated that willingness to drive after consuming alcohol was strongly predicted by variability in both subjective intoxication and perceived danger, over and above breath alcohol concentration (BrAC), an objective metric of alcohol intoxication.

The present study aimed to replicate and extend findings from Motschman et al. (2020), using a 6-week EMA protocol with an independent sample to examine the influence of cognitions on the decision to drive after consuming alcohol. Beyond the extended assessment period, this study makes two key extensions. First, whereas Motschman et al. (2020) examined momentary willingness to drive as the primary outcome, we examine self-reported driving behavior on the following morning, providing a more direct assessment of real-world transportation decisions. Second, we test cross-level interactions between person-level characteristics and daily contextual factors, addressing recent calls to understand individual heterogeneity in alcohol-related decision-making, including the decision to drive after consuming alcohol (Davis-Stober et al., 2019). Specifically, we tested whether daily variability in self-reported subjective intoxication, perceived danger of driving, and willingness to drive during drinking episodes predicted driving after drinking. BrAC was included as an objective indicator of intoxication in-the-moment, measured using mobile breathalyzer devices (KHN Solutions Inc., 2018). We tested whether subjective intoxication, perceived danger of driving, and willingness to drive predicted driving over and above BrAC level, and whether the association between these variables and driving likelihood differed as a function of BrAC.

These data were collected as part of a parent study that tested the effect of BrAC warnings on driving decisions (see Haney et al., 2024). However, because all participants in the current analyses received identical warning protocols during their EMA period, and because our research questions concern the fundamental relationships between cognitions, intoxication, and driving behavior rather than intervention effects per se, these data are well-suited to address our aims. We focus on understanding basic cognitive-behavioral relationships during naturally occurring drinking episodes rather than testing intervention efficacy or mechanisms of behavior change, which has been examined elsewhere (see Haney et al., 2024). Additionally, because all participants in the current study received BrAC warnings when their readings exceeded 0.05%, we conducted post-hoc analyses using the minimal assessment control condition (which did not receive warnings) to explore whether warnings may have influenced the relationship between subjective intoxication and driving after drinking behavior.

Materials and Methods

Participants

Participants were recruited from a large Midwestern university and the surrounding community through email listservs, social media advertisements, and flyers. Interested individuals completed a phone interview to determine eligibility. To be eligible, participants had to be at least 21 years old, have consumed at least five (for males) or four (for females) standard alcoholic beverages on a single occasion within the past six months, have regular access to a car and drive frequently, and reside at least two miles from their typical drinking location(s). Exclusion criteria included current pregnancy or lactation, a body mass index (BMI) outside the range of 18–30 (to reduce variability in alcohol metabolism), a history of treatment for substance use or psychiatric disorders, or medical or medication-related contraindications to alcohol consumption. Table 1 reports additional demographic information.

Table 1.

Descriptive Statistics of Participant Demographics and Study Variables (N = 113)

Variable	M	SD	Range	n	%
Age	25.04	4.00	21 to 35
Sex
Female				67	59.3
Male				46	40.7
Racea/Ethnicity
Asian				3	2.7
Black				9	8.0
Latinx/Hispanic				2	1.8
Middle Eastern/North African				1	0.9
Multiracial				7	6.2
Other				1	0.9
White				98	86.7
Post-COVID sample				83	73.5
Study Variablesb
Perceived Danger of Driving	2.69	2.00	1 to 10
Willingness to Drive After Drinking					53.2
Subjective Intoxication	2.49	1.54	1 to 8

Note. M = mean. SD = standard deviation.

^aParticipants could select all that apply.

^bStudy variables represent day-level averages across all observations.

Procedure

All procedures were approved by the University of Missouri Institutional Review Board. Participants were informed about measures taken to ensure confidentiality (e.g., secure data storage, data de-identification) prior to their phone screen and again during informed consent. These data were collected as part of a broader 12-week study that combined an alcohol administration laboratory session with six weeks of EMA. The alcohol laboratory session occurred at baseline and involved a controlled beverage alcohol administration procedure designed to assess acute tolerance to alcohol’s effects on alcohol-impaired driving decision making. This laboratory component was discontinued after March 2020 due to COVID-19 restrictions. Importantly, the laboratory session occurred before EMA data collection began (either immediately after or following a 6-week delay, depending on condition assignment) and is not expected to have influenced EMA responses given the temporal separation and the lack of any feedback or intervention provided during the laboratory session.

The current study focused on understanding fundamental associations between cognitions and driving behavior in naturalistic drinking contexts. However, a supplementary aim of the parent study was to test the effect of BrAC warnings given when a recorded breathalyzer reading was ≥ 0.05% (see Haney et al., 2024). All participants included in our analyses received these BrAC warnings, providing a consistent context for examining how subjective intoxication, perceived danger, and willingness to drive predict actual driving behavior. Importantly, the study protocol did not attempt to teach protective behavioral strategies or provide motivational feedback beyond the BrAC-triggered warnings. As such, these data capture naturally occurring variation in real-world drinking occasions and driving decisions, with the warning protocol providing a consistent safety-oriented context that does not fundamentally alter the nature of the decisions participants face.

Participants were randomized into three EMA conditions. Two were BrAC warning conditions, one starting immediately after baseline and the other after a six-week delay. These two warning conditions differed only in the timing of EMA initiation relative to baseline (immediate vs. 6-week delay), but both received identical warning protocols during their six-week EMA data collection period. Because both groups had equivalent EMA experiences - the same prompting schedule, warning thresholds, and assessment measures - and because the baseline laboratory session was temporally separated from EMA data collection, we had no theoretical expectation that condition would predict EMA responses. To empirically confirm this, we tested whether condition was associated with any study variable and found no significant associations (all ps > .05). We therefore report models without this covariate, following best practices of model parsimony. The third was a minimal assessment control condition that did not receive BrAC warnings. The minimal assessment control condition also did not answer questions about either perceived danger of driving or willingness to drive, so participants from this condition were not included in primary analyses for the current study.

Participants completed six weeks of EMA data collection, which included one daily morning survey (available from 5 a.m. to 1 p.m.), and up to nine repeated evening surveys (6 p.m. to 2 a.m.) each day. Prior to March 1, 2020, participants received evening EMA prompts between 6 p.m. and 12 a.m. Following March 1, 2020, the protocol was modified to extend prompts to 1 a.m. and 2 a.m. This modification was made in response to preliminary data indicating that certain participants experienced drinking episodes that continued beyond midnight. Of the 113 participants included in the analytic sample, 30 (26.5%) were enrolled prior to this protocol change and 83 (73.5%) were enrolled after it was implemented. To account for potential systematic differences in drinking and driving behavior due to the pandemic, we included a binary COVID-era covariate (pre- vs. post-March 1, 2020) in all models.

To promote compliance with the EMA protocol, staff members demonstrated how to respond to prompts on the EMA survey application (TigerAware; Morrison et al., 2018) and how to use the portable breathalyzer device (BACtrack Mobile Pro). Evening surveys were delivered via automated prompts every hour during this window to capture drinking episodes at multiple time points throughout typical drinking occasions. Regardless of drinking status, participants were instructed to provide a breathalyzer sample during each evening survey. Participants were instructed to wait 15 minutes after drinking alcohol and to rinse their mouth out with water before completing the reading to minimize residual mouth alcohol. BrAC readings were not visible to participants, but all participants included in the primary analyses received a warning cautioning against driving if their BrAC was ≥ 0.05%. When participants reported not drinking, questions about AID-related cognitions were replaced with additional questions about current mood to equalize time requirement across evening reports.

Participants were compensated up to $60 per week based on a survey compliance rate of 80% or more. Average compliance across both survey types was 84.1%. Morning surveys measured whether participants drank the previous evening, the number of drinks consumed, drinking context (e.g., drinking companions and location), and method of transportation to get home (e.g., drove self; used a designated driver). Evening surveys measured drinking status, the number of drinks consumed, drinking context (e.g., drinking companions and location), subjective intoxication, perceived danger of driving, willingness to drive, and BrAC.

Our study was designed to include brief, repeated evening surveys to increase temporal resolution while minimizing participant burden. Evening EMA retention was evaluated using eligible evening prompts, defined as days on which at least one eligible evening prompt was recorded for a participant. On average, participants completed 70.7% of eligible EMA days (SD = 0.35). The mean number of study days completed was 30.47 (SD = 12.54). Retention to later study periods was summarized based on documented eligibility (i.e., eligible prompts recorded during the relevant window). Among participants with documented week-6 eligibility (n = 90), 84.4% completed at least one evening survey during week 6 (days 35–41), and among participants with documented eligibility through the end of week 4 (n = 106), 90.6% provided data through at least 4 weeks.

Measures

Demographics

Participants completed a baseline questionnaire assessing demographic information, including age, sex, gender identity, race, ethnicity, education level, employment status, driving characteristics, and alcohol use.

Perceived Danger of Driving

At each evening survey where drinking was reported, participants were asked to report on their perceived dangerousness of driving. Using a visual analog scale, participants were asked to report how dangerous they felt it was for them to drive “RIGHT NOW” on a scale from 1 (not at all) to 10 (extremely).

Willingness to Drive After Drinking

At each evening survey where drinking was reported, participants were asked to report whether they would drive based on how they felt “RIGHT NOW” using a dichotomous response scale (1 = yes; 0 = no).

Subjective Intoxication

At each evening survey where drinking was reported, participants were asked to report how intoxicated they felt “RIGHT NOW” using a visual analog scale from 1 (not at all) to 10 (more intoxicated than you’ve ever been).

Breath Alcohol Concentration (BrAC)

BrAC was assessed during every evening report using the BACtrack Mobile Pro portable breathalyzer (KHN Solutions Inc., 2018). Participants were asked to “please stop drinking for 15 minutes and rinse your mouth with water.” After 15 minutes, participants were asked to provide a breathalyzer sample by ensuring their Bluetooth was activated and to blow when prompted. If participants’ recorded BrAC was ≥ 0.05%, they received the following warning: “Your BAC is at the point where you should NOT drive. Please seek another way home.” BrAC readings were not visible to participants.

The BACtrack Mobile Pro portable breathalyzer has good agreement with standard breathalyzers, with a slight bias to over-report; bias = .008 g/dl, 95% CI [.0062, .0096] (Riordan et al., 2017). Delgado et al. (2021) showed that the BACtrack Mobile Pro produces comparable results to peak phlebotomy blood alcohol concentration (BAC) and a police-grade sensor.

Driving After Drinking

At each daily morning survey where drinking was reported the previous night, participants were asked “how did you get home last night.” Participants were asked to select one option, including “I drove,” “someone else drove,” “taxi/uber,” “public transportation,” “walked,” “bicycle,” or “I only drank at home.” Participants who selected “I drove” were coded as 1 for driving after drinking; all other responses except “I only drank at home” were coded as 0.

Analytic Method

A total of 207 participants were recruited for the study. Among these, 69 participants were part of a minimal assessment condition that excluded questions about the perceived danger of driving and willingness to drive. These participants were not included in the current analyses. Additionally, 25 participants were dropped for lacking at least one drinking episode accompanied by a subsequent morning report to verify whether driving occurred, resulting in 113 participants and 634 (633 for subjective intoxication) drinking days.

To examine associations among focal study variables, Pearson correlations were computed separately at the day level (using raw daily observations) and the person level (using participant means across drinking episodes), consistent with the multilevel decomposition of within- and between-person variance used in primary analyses. Correlations involving BrAC ≥ 0.05% are point-biserial given its binary coding, though mathematically equivalent to Pearson’s r.

All analyses involved mixed-effects modeling to account for the nested and repeated structure of the data (moments [level-1] nested within days [level-2] nested within individual [level-3]). Separate models examined within- and between-person perceived danger of AID, willingness to drive after drinking, and subjective intoxication as the primary predictors examining day-level AID behavior. All models included an objective measurement of BrAC.

Post-hoc analyses evaluated whether BrAC warnings influenced the subjective intoxication model results. We re-ran an identical model using only the minimal assessment condition, where BrAC warnings were not provided. This restricted analysis included 55 participants and 279 daily observations, with the reduced sample size reflecting incomplete responses between evening and morning reports and exclusion of participants that never endorsed any driving after drinking.

Results

Participants reported consuming alcohol on an average of 11.00 days (SD = 8.58) during the study, with a mean of 2.64 drinks per drinking day (SD = 2.26). The average BrAC was 0.04% (SD = 0.04%), and BrAC was ≥ 0.05% on 24.4% of drinking days. Drinking episodes were more frequent on weekends, with 58.7% of drinking days occurring Friday through Sunday. Participants reported driving (vs. other transportation methods) on 46.7% of drinking days and were less likely to drive when their BrAC was ≥ 0.05% (all ORs < 1.0), though this association was not statistically significant in any model. Table 1 presents additional descriptive statistics of study variables.

At the day and person levels, all focal variables were significantly intercorrelated (all ps < .01). Across day and person levels, perceived danger and subjective intoxication were strongly positively correlated (rs = .86 and .88), while willingness to drive was strongly negatively correlated with both perceived danger (rs = −.71, −.73) and subjective intoxication (rs = −.67, −.67). BrAC was positively correlated with perceived danger and subjective intoxication and negatively correlated with willingness to drive at both levels (day-level rs = .37, .42, −.38; person-level rs = .54, .69, −.57). Given these intercorrelations, each cognition was modeled separately.

We estimated an empty means logistic mixed-effects model with a random intercept to quantify the degree of clustering within individuals. The intraclass correlation coefficient (ICC) was 0.233, indicating that 23.3% of the variance in driving following alcohol consumption was attributable to between-person differences, while 76.7% was attributable to within-person variation across study days. This pattern suggests that although individuals differ in their tendency to drive after drinking, a substantial portion of the variability is within-person, supporting our use of mixed-effects models.

Cognitions Predicting Driving After Drinking

Table 2 presents findings from the model examining the association between perceived danger of driving and driving after drinking. Daily within-person perceived danger of driving was associated with driving after drinking, such that when an individual reported higher daily perceived danger than usual, they were less likely to drive after drinking that day. Among the covariates, weekend drinking days were associated with a lower likelihood of driving after drinking. No other main effects or interactions were statistically significant.

Table 2.

Mixed-Effects Models of AID Cognitions and Subjective Intoxication Predicting Driving After Drinking (N = 113)

Effect	Perceived Danger of Driving			Willingness to Drive			Subjective Intoxication
	OR	95% CI	p	OR	95% CI	p	OR	95% CI	p
Intercept	2.35	[1.09, 5.07]	.029	1.77	[0.85, 3.65]	.125	1.92	[0.83, 4.48]	.130
Level-2 (Between-Person) Main Effects
Between-person perceived danger	0.70	[0.44, 1.12]	.136	—	—	—	—	—	—
Between-person willingness	—	—	—	1.02	[1.00, 1.04]	.057	—	—	—
Between-person subjective intoxication	—	—	—	—	—	—	0.97	[0.48, 1.94]	.922
Level-1 (Within-Person) Main Effects
Within-person perceived danger	0.63	[0.47, 0.84]	.002	—	—	—	—	—	—
Within-person willingness	—	—	—	1.03	[1.02, 1.04]	<.001	—	—	—
Within-person subjective intoxication	—	—	—	—	—	—	0.62	[0.44, 0.87]	.006
BrAC ≥ 0.05% (ref = < 0.05%)	0.68	[0.44, 1.06]	.089	0.81	[0.52, 1.26]	.345	0.72	[0.45, 1.16]	.176
Interactions
Between-person predictor × BrAC	0.61	[0.36, 1.02]	.062	1.02	[1.00, 1.04]	.107	0.47	[0.25, 0.87]	.017
Within-person predictor × BrAC	1.09	[0.79, 1.52]	.602	1.00	[0.99, 1.02]	.788	1.05	[0.71, 1.55]	.809
Covariates
Age (mean centered)	1.04	[0.97, 1.11]	.308	1.04	[0.96, 1.12]	.323	1.04	[0.97, 1.12]	.251
Post-COVID-19 (ref = Pre-COVID-19)	0.67	[0.30, 1.50]	.332	0.90	[0.44, 1.88]	.788	0.91	[0.38, 2.17]	.825
Weekend (ref = weekday)	0.65	[0.44, 0.95]	.028	0.66	[0.44, 0.98]	.038	0.61	[0.41, 0.89]	.010
Male (ref = female)	1.33	[0.76, 2.33]	.326	1.29	[0.71, 2.33]	.404	1.35	[0.76, 2.39]	.303

Note. AID = alcohol-impaired driving. BrAC = breath alcohol concentration. SE = standard error. OR = odds ratio. CI = confidence interval. Significant effects at p < .05 are bolded. — = predictor was not included in the model.

Table 2 presents findings from the model examining the association between willingness to drive and driving after drinking. Daily within-person willingness to drive was associated with an increased likelihood of driving after drinking that day. This suggests that on days when individuals were more willing to drive than usual, their likelihood of driving after drinking increased. Weekend drinking days were associated with a lower likelihood of driving after drinking. No other main effects or interactions were statistically significant.

Table 2 presents findings from the model examining the association between subjective intoxication and driving after drinking. Daily within-person subjective intoxication was associated with driving after drinking, such that when an individual reported higher daily subjective intoxication than usual, they were less likely to drive after drinking that day. The interaction between between-person subjective intoxication and BrAC was significant, such that at higher BrAC the negative association between between-person subjective intoxication and driving was significant and negative (simple slope for between-person subjective intoxication when BrAC was ≥ 0.05: OR = 0.456, 95% CI = [0.239, 0.868], p = .017), while at lower BrAC, the association between between-person subjective intoxication and driving was not significant (simple slope for between-person subjective intoxication when BrAC was < 0.05: OR = 0.966, 95% CI = [0.482, 1.936], p = .922). Figure 1 shows a plot of this interaction as the predicted probability of driving by subjective intoxication and BrAC level. Weekend drinking days were associated with a lower likelihood of driving after drinking. No other main effects or interactions were statistically significant.

Figure 1. Predicted Probability of Driving by Subjective Intoxication and BrAC level.

Note. The solid line represents predicted probabilities when BrAC < 0.05%, and the dashed line represents predicted probabilities when BrAC ≥ 0.05%. Subjective intoxication is displayed at values corresponding to the sample mean and ±1 and ±2 standard deviations. Predictions were derived from the mixed-effects logistic regression model controlling for within-person subjective intoxication, age, sex, weekend, and post-COVID status.

Post-Hoc Analysis: Subjective Intoxication in the Control Condition

Prior research from this study found that BrAC warnings significantly reduced the likelihood of AID behavior (Haney et al., 2024). To assess whether warnings influenced the results of our primary subjective intoxication model, we re-ran the mixed-effects analysis described above, restricting the sample to participants in the minimal assessment (control) condition. Because both experimental and control groups completed evening surveys assessing subjective intoxication, this follow-up analysis allowed us to test whether the relationships between subjective intoxication, BrAC, and driving behavior observed in the primary model were influenced by the presence of BrAC warnings. In the control condition, neither between-person nor within-person subjective intoxication, nor their interactions with BrAC, significantly predicted driving after drinking. Among the covariates, only age emerged as a significant predictor.

Discussion

The present study used EMA to examine how intra- and inter-individual variability in cognitions predicted driving after drinking in real-world drinking environments. Our findings reinforce the importance of driving after drinking cognitions, supporting prior research (e.g., Motschman et al., 2020, Quinn and Fromme, 2012) and providing additional evidence that these cognitions are influential in the decision to drive after drinking. At the day-level, we found that higher within-person perceived danger of driving and subjective intoxication were associated with a lower likelihood of driving after drinking, whereas higher willingness to drive was associated with increased likelihood of driving, even after controlling for BrAC.

Our findings demonstrate that higher perceived danger of driving and willingness to drive during drinking episodes influence real-world driving behavior. Building on prior work from Morris et al. (2014) and Motschman et al. (2020), which highlighted within-person variability in these cognitions in laboratory and real-world settings, we demonstrate that higher daily variations in risk perceptions are robust predictors of driving behavior after consuming alcohol. This pattern is consistent with laboratory research using controlled alcohol administration, which demonstrates that intoxication reduces perceived danger of driving (Morris et al., 2014) and influences driving decisions based on subjective intoxication, perceived danger, and context (Motschman et al., 2020). Our EMA findings extend this laboratory work by showing these cognitive processes predict actual driving behavior in naturalistic drinking contexts where individuals face real transportation decisions and consequences. These findings align with emerging research that demonstrates the utility of examining cognitive factors, such as intentions and willingness, at both the event-level and between-person level to understand decision making for alcohol-related behaviors (e.g., Scaglione et al., 2022). This research highlights the importance of situationally varying risk appraisals, which may exert stronger influence on behavior than stable individual differences in risk perception. Moreover, our findings indicate that the decision to drive after drinking was more strongly impacted by higher within-person daily variability in these risk perceptions than by an objective indicator of intoxication (BrAC). Our findings offer preliminary ecological validity that dynamic risk appraisals are relevant predictors of drinking and driving behavior in real-world contexts, over and above a biomarker of intoxication level.

Across all three models, weekend drinking days (Friday through Sunday) were consistently associated with a significantly lower likelihood of driving after drinking compared to weekday drinking days. This pattern persisted after accounting for our AID-related cognitions, suggesting that day-of-week exerts an independent effect on transportation decisions beyond individual-level cognitive factors. For our sample of adults aged 21–35 recruited from a large Midwestern university and surrounding community, weekend drinking occasions may differ systematically from weekday drinking in ways that facilitate safer transportation planning—such as drinking in larger social groups or having greater flexibility to leave vehicles overnight. Future research should examine whether these day-of-week patterns generalize to other populations and geographic contexts, and whether targeted interventions accounting for these temporal patterns might enhance effectiveness.

Our findings also underscore the role of between-person differences in subjective intoxication in shaping drinking and driving behavior. Specifically, individuals who generally experience stronger subjective effects of alcohol were less likely to drive on days when their intoxication was objectively higher (BrAC ≥ 0.05%). This pattern suggests that a heightened general awareness of intoxication may serve as a protective factor, reducing the likelihood of engaging in risky driving behavior following alcohol use. These findings contribute unique insights into real-world drinking and driving and provide evidence for person-level heterogeneity in how individuals make alcohol-related decisions (Davis-Stober et al., 2019). Prior research found that social drinkers frequently underestimate their BAC, and these estimation errors have direct implications for driving decisions (Beirness, 1987). Individuals who underestimate their BAC during drinking episodes consume more alcohol, perceive themselves as less intoxicated than they actually are, and critically, are more likely to judge themselves capable of driving when their BAC exceeds legal limits (Beirness, 1987). Laboratory studies have demonstrated that this underestimation pattern translates directly to riskier driving behavior. Laude and Fillmore (2016) found that, independent of measured BAC, drivers who estimate lower BACs following alcohol administration display more dangerous simulated driving compared to those who estimate higher BACs. This heightened risk appears linked to reduced sensitivity to subjective intoxication cues, as experimental evidence shows that risky drinkers exhibit blunted subjective responses to alcohol (Fillmore and Van Dyke, 2020).

In contrast to the risks associated with blunted subjective response, our EMA findings show that heightened subjective intoxication functions protectively. Individuals who typically experienced stronger subjective effects of alcohol demonstrate lower odds of driving on high-intoxication days, suggesting that subjective awareness serves as an internal restraint on risky decision-making. This protective pattern has been observed in other natural drinking environments. Field research among bar patrons shows that stronger subjective feelings of intoxication were associated with better awareness of exceeding the legal driving threshold (Rossheim et al., 2017). Recent EMA research using continuous transdermal alcohol monitoring corroborates this protective pattern, demonstrating that subjective intoxication predicts reduced alcohol-related consequences (e.g., driving after drinking) at equivalent objective alcohol concentrations (Richards et al., 2024). These converging findings suggest that individuals who maintain sensitivity to subjective intoxication cues may be better positioned to make safer decisions about driving after drinking. People who consistently experience stronger subjective effects of alcohol may be more attuned to these cues and, as a result, may adopt more cautious decision-making strategies. Such individuals may rely on personal “consumption rules”—prioritizing subjective cues over external factors like transportation costs (McCarthy et al., 2021)—when deciding whether to drive.

We conducted an exploratory post-hoc analysis examining whether the between-person subjective intoxication by BrAC interaction observed in our primary analyses emerged in the minimal assessment control condition without BrAC warnings, given prior research from this study showing that BrAC warnings can lower the likelihood of driving (Haney et al., 2024). The interaction was not significant in the control condition. This pattern of results may indicate a subtle effect of warnings, such that individuals who typically experience stronger subjective effects are especially responsive to high BrAC readings. However, this inference is limited by the relatively modest power to detect an interaction in the control condition (n = 55 compared to n = 113 in primary analyses). Future research is needed to determine the potential value of interventions that combine internal cues (e.g., subjective intoxication) with external indicators of intoxication (e.g., BrAC warnings) to promote safer transportation choices.

Our study offers new insights into the within- and between-person processes underlying drinking and driving but is not without limitations. First, the study did not include an assessment of BrAC immediately prior to driving and so did not fully capture an individual’s level of intoxication at the moment they decided to drive. Future studies could include a BrAC assessment immediately prior to the driving event or use personal monitoring devices to improve the precision of the temporal link between BrAC and driving. Additionally, although BrAC provides an objective estimate of alcohol concentration, it does not directly index behavioral impairment. Future research could pair BrAC with performance-based measures of driving-relevant impairment (e.g., psychomotor, divided-attention tasks) or passive sensing approaches to more directly assess functional impairment in the natural environment. Second, although we conducted a post hoc analysis with participants who did not receive BrAC warnings, restricting the sample reduced our statistical power (n = 55) and may have impacted the interaction effect. Given that BrAC warnings themselves can deter driving (Haney et al., 2024), future studies should further investigate how such warnings alter subjective intoxication and driving decisions in larger samples. Finally, our focus on individual-level cognitive factors did not account for contextual influences (e.g., drinking location and companions) that could clarify when and why perceived danger, willingness to drive, and subjective intoxication most strongly influence driving behavior.

The study period encompassed both pre- and post-COVID-19 onset, which likely influenced drinking patterns, social contexts, and transportation decisions. Stay-at-home orders contributed to reduced social gatherings, temporary closure of bars and restaurants, and increased remote work, which may have decreased both drinking occasions requiring travel and the perceived necessity of driving to drinking venues. Whether participation occurred pre or post-COVID onset was included as a covariate in all models (see Table 2). Nevertheless, the generalizability of our findings to typical pre-pandemic drinking contexts remains uncertain, and the base rate of driving after drinking may have been lower during pandemic restrictions. Future research should examine whether these cognitive-behavioral relationships persist in fully reopened post-pandemic environments with unrestricted social drinking venues.

Our results underscore the importance of investigating within-person changes in drinking and driving cognitions, suggesting that targeted, real-time strategies may be effective in mitigating drinking and driving behavior. Moreover, our findings show that stable appraisals of subjective intoxication influence drinking and driving behavior and interact with high levels of daily drinking. Recognizing both dynamic and stable processes can inform more comprehensive interventions that address long-term subjective appraisals while adapting to daily variations in risk perception, ultimately promoting safer driving outcomes.