Subjective intoxication predicts alcohol-related consequences at equivalent alcohol concentrations in young adults using ecological momentary assessment and alcohol sensors

Abstract

Objective:

Subjective intoxication (SI) when drinking may serve as an internal barometer of whether to continue drinking or engage in potentially unsafe behavior. Mobile assessments offer the potential to use SI as a prospective risk indicator during drinking episodes; little evidence exists for the validity of real-time SI measures. We test correspondence of SI with estimated BAC (eBAC) and transdermal alcohol concentration (TAC) in young adults’ natural settings. We provide a novel test of whether SI features (peak and mean SI) uniquely predict consequences adjusting for alcohol concentration.

Method:

222 heavy drinking young adults (Mage=22.3, 64% female, 79% non-Hispanic white, 84% undergraduates) participated in a 6-day study that used ecological momentary assessment of drinking and TAC sensors. SI was assessed every 30 minutes during drinking episodes. Multilevel modeling was used to test hypotheses.

Results:

Momentary SI and eBAC had moderate associations at the moment- and day-levels (standardized βs=0.5–0.6); SI was moderately associated with TAC at the day-level (βs=0.5). Associations between SI and alcohol concentration varied widely between persons and across days. Day-level SI features predicted consequences when adjusting for alcohol concentration (IRRs=1.29–1.70).

Conclusions:

Our two-item SI measure shows evidence of validity in real-world settings with heavy drinking young adults. SI was significantly correlated with alcohol concentration and was a unique predictor of consequences. The strength of these associations varied greatly across persons and days. Real-time SI measurement may be useful in preventive interventions but continued research is needed into when and for whom momentary SI is most predictive of risk.

Young-adult drinking is a major public health problem. Approximately 1 in 3 students report heavy episodic drinking in the past month (4+/5+ drinks in two hours for females/males) and over 10% are consuming two to three times these amounts (Hingson & White, 2013; Patrick et al., 2013, 2016; Schulenberg et al., 2018; White & Hingson, 2013). Young adults who engage in these frequent risky drinking behaviors are vulnerable to alcohol-related consequences. Prediction and prevention of consequences among young adults who drink heavily is a continued focus of research. Studies have primarily focused on number of drinks as a predictor of consequences, but the accuracy of self-reported drinking measures may lessen during heavy drinking episodes (Northcote & Livingston, 2011) and may lessen predictive power of self-reported consumption measures as a result. Researchers have suggested that direct measurements of intoxication may avoid these problems and aid in predicting consequences. Direct measurements of intoxication can be device-based (e.g., transdermal alcohol concentration [TAC] sensors) or concentration based (e.g., blood alcohol concentration; BAC). Evidence supports the predictive validity of intoxication measures. For example, TAC and estimated BAC (eBAC) measured in real time (via sensors and ecological momentary assessment (EMA), respectively) have been shown to predict alcohol-related consequences in young adults’ natural settings (Carpenter & Merrill, 2021; Russell et al., 2022). The real-time measurement offered by these may allow preventionists to better anticipate and prevent alcohol-related consequences in young adults’ natural settings and may lead to the development of new and powerful mobile interventions.

Subjective intoxication (SI) is another potential candidate for prediction of alcohol-related consequenes in natural settings (Andreasson, 2016; Pearson et al., 2016). SI has been measured in multiple ways, including rating scales across multiple dimensions of subjective response to alcohol (e.g., ratings of sluggishness, arousal; Morean & Corbin, 2010; Ray et al., 2009), self-estimation of BAC (e.g., what is your current BAC?; Aston & Liguori, 2013), and rating scales of perceived intoxication intensity (how drunk, how buzzed, how intoxicated did you feel last night?; e.g., Linden-Carmichael & Calhoun, 2022). Other work investigating SI has included sober persons answering questions about their expected intoxication in hypothetical scenarios (Turrisi et al., 1988; Turrisi & Jaccard, 1991; Turrisi & Wiersma, 1999). In the current paper, we focus on rating scales of perceived SI intensity because they represent a straightforward approach to obtaining how intoxicated a person is feeling at any given point or over any given time period. SI intensity ratings when drinking may serve as an internal barometer of whether to continue or cease drinking and influence their perceived capacity to engage in certain behaviors such as driving. They may therefore serve as real-time predictors of risk. How intoxicated a person feels at any given point is influenced by many factors, both biological and contextual. A large body of literature has demonstrated that persons with a family history of alcohol use disorder report lower SI intensities at equivalent BACs compared to those without a family history (e.g., Eng et al., 2005; Pollock, 1992; Schuckit, 1984). Heath et al. (1999) found that genetic factors explain approximately 60% of the variability in SI. SI may also differ by factors such as sex and tolerance (Amlung et al., 2014; Mallett et al., 2009; Marczinski & Fillmore, 2009). There is evidence that social context and location play a role in SI. Laboratory research indicates that SI is rated higher when drinking with friends versus alone (Corbin et al., 2021; Kirkpatrick & de Wit, 2013) and the rewarding effects of alcohol are rated higher in a bar-like setting compared to a traditional laboratory setting (e.g., with computers, filing cabinets; Corbin et al., 2015). SI measured during real-world drinking episodes found similar results (Fischer et al., 2023), indicating the importance of contextual factors.

Previous evidence supports the validity of SI measures for risk prediction. In a large daily diary prospective cohort study of college students, Quinn and Fromme (2011) found that daily retrospective reports of SI predicted alcohol-related consequences over and above eBAC. Quinn and Fromme (2012) also reported that according to retrospective daily diary reports, students were more likely to drive after drinking when they had high eBACs but perceived themselves as less intoxicated (low SI) via significant interaction effects. SI may also inform the distance a person is willing to drive, should they choose to drive after drinking, based on an alcohol administration study (Motschman et al., 2020). Using retrospective recollection of their 21^st birthday celebrations, Wetherill and Fromme (2009) found SI to be a significant predictor of blackouts and hangovers in college students. Marino and Fromme (2018) found frequency of high SI in the past 3 months (reported via survey) to be prospectively associated with blackouts after college years among a large sample of young adults followed through and 2 years after college. These studies suggest the potential of SI as an important predictor of consequences, but gaps remain regarding the utility of SI to serve as a real-time risk indicator.

Despite significant advances in mobile technology, there has been little work testing whether and to what extent real-time measurements of SI (a) correlate with alcohol concentration measurements (e.g., TAC and eBAC), and (b) uniquely predict alcohol-related consequences adjusting for these. Advancements in mobile technology allow in-depth, real-time measurement of SI intensity as drinking events unfold in young adults’ natural environments. SI intensity is a dynamic psychological state that changes frequently throughout a drinking event, and differs both between persons and across drinking occasions. Given the simplicity of in-the-moment SI intensity measurement (typically a simple Likert scale), such knowledge may help guide the development and delivery of real-time, adaptive interventions.

Present Study

The present study uses an intensive event-level study to address two research aims. First, we examine the relationship between momentary SI and alcohol concentration intoxication measures collected during drinking episodes (eBAC from EMA and TAC from a wearable sensor). While previous literature has examined the relationships between SI and eBAC at daily- and person-levels, our study will be the first to study momentary associations between the two in the natural environment including “random” slopes. Alcohol administration paradigms indicate that the strength of association between SI and BAC varies between people (for review, Morean & Corbin, 2010), thus using “random” slopes in multilevel modeling frameworks allow us to estimate this variability.

Second, we examine the ability of SI (mean SI and peak SI) to predict consequences over and above alcohol concentration measures. Greater SI may be the initial consequence of high BAC and may therefore be a more proximal indicator of physical impairment or compromised judgment than alcohol concentration measures (Andreasson, 2016). SI may foretell the experience of physical consequences like hangover or illness and engagement in risky or disagreeable activities that lead to conflict or injury. Our use of EMA to repeatedly measure SI during multiple drinking episodes allowed us to take a novel approach to predicting consequences in natural settings. This will help us understand whether SI offers unique information for the prediction of alcohol-related consequences that is not captured in alcohol concentration assessments (e.g., support its use as a risk indicator in natural settings).

Our investigation was guided by the following hypotheses.

H1: SI and alcohol concentration measures (eBAC and TAC) each measure alcohol intoxication, but different aspects. We therefore hypothesized that SI would be positively – but moderately – associated with alcohol concentration measures (eBAC and TAC) on average. We also hypothesized that the strength of the associations between SI and alcohol concentration measures would vary significantly across days and persons. Evidence strongly supports SI as a stronger correlate of alcohol concentration for some individuals than others (e.g., Schuckit, 1984). Less evidence shows that subjective-alcohol concentration associations will vary by situation, but we hypothesize that differences in social and physical drinking environments will affect links between SI and alcohol concentration measures. Significant variability in subjective-alcohol concentration associations across persons and days will support these hypotheses.

H2: Evidence shows that even at equivalent levels of alcohol concentration intoxication, higher SI may lead to greater consequence risk (e.g., Wetherill and Fromme, 2009). We therefore hypothesized that SI would have unique predictive validity for consequences even at equivalent alcohol concentration levels.

This work extends the previous literature by examining SI during drinking episodes, allowing for assessment and characterization while people are actively drinking. This helps translate the lab paradigm into real life, where young adults drink to levels that would be unethical for researchers to dose. Most previous work has used retrospective daily SI, which may be subject to forgetting and skewed towards the peak experience of intoxication of that drinking day. Understanding the moment-level associations between subjective and alcohol concentration is important because it could inform our ability to anticipate, and possibly prevent, alcohol-related consequences in natural settings.

Method

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. Materials and analysis code for this study are available by emailing the corresponding author. This study was not preregistered.

Participants

Participants included 222 young adults aged 21–29 (M age (SD) = 22.3 (1.3), 64% female, 79% non-Hispanic white, 84% undergraduate) who regularly engaged in heavy episodic drinking, defined as 4+/5+ drinks on a single occasion for females/males. Full eligibility criteria are described below (see Recruitment and Screening). The majority of participants (90.5%) were full-time students, with 93% of students being undergraduates and the remaining 7% of students being graduate, professional, or “other” students. This study was approved by the Pennsylvania State University Institutional Review Board (Study #5561). Participants provided written informed consent to participate.

Recruitment and Screening

Participants were recruited near the campus of a large northeastern U.S. university using flyers. Participants completed a screening survey prior to enrollment. To be eligible, participants needed to: (1) be between the ages of 21–29, (2) have engaged in heavy episodic drinking at least weekly on average during the past calendar year or typically engaged in heavy episodic drinking at least weekly on average during the academic year, and (3) be sufficiently proficient in written English to complete study procedures (Russell et al., 2022). Of the 531 individuals who completed the screening survey, 419 were eligible. Time and resource limitations prevented invitations to all eligible participants; invitations were sent in the order in which screening surveys were received. Of 343 invited, 222 completed the study. No evidence of bias was observed comparing those completing versus not completing participation by sex, race/ethnicity, student status, or past-two-week heavy episodic drinking (ps ≥ .10).

Procedure

The study consisted of five 24-hour periods spanning six consecutive days, beginning on a Wednesday and ending the following Monday. The study was designed to intensively track heavy drinking during a single social weekend (i.e., Thursday – Saturday; Finlay et al., 2012) among young adults who frequently engaged in heavy drinking. The study included baseline and endpoint assessments, three times daily EMA reports (to capture same-day pre- and post-drinking reports and previous-day drinking reports), self-initiated EMA drinking episode surveys, and transdermal alcohol sensors. We limited the study duration to six days to avoid overburdening participants and maximize data quality. Participants started with a baseline appointment in our research laboratory, at which time they provided consent, followed by completion of a 30-minute survey assessing demographics, drinking-related habits (e.g., drinking history, past-year drinking and consequences), personality, and other psychosocial factors (e.g., stress, social support). Following survey completion, an alcohol monitor was fitted to the participants’ ankle and they received training on the mobile phone EMA surveys. Immediately following the baseline session, participants began the field protocol, during which they wore a transdermal alcohol monitor on their ankle (the Secure Continuous Remote Alcohol Monitoring-Continuous Alcohol Monitor or SCRAM-CAM) and carried a mobile phone on which they completed the three daily EMA surveys and self-initiated EMA drinking episode surveys. Participants completed the endpoint assessment in our laboratory when they returned their devices. Participants were compensated up to $110 for their participation. Compensation was not dependent on the number of self-initiated drinking episode surveys completed. This was done so as not to directly incentivize alcohol consumption.

Ecological Momentary Assessment (EMA) Protocol

Participants were provided with an Android smartphone to use for EMA. The phone was equipped with a customized EMA survey app designed by specialist programmers at the Survey Research Center at the authors’ academic institution. The app contained two survey types. The first was the scheduled EMA surveys, which prompted participants to complete morning (scheduled for 10 AM), afternoon (4 PM), and evening (9 PM) surveys (afternoon and evening survey data not included in the present study). Participants could also opt to receive scheduled prompts one hour later (11 AM, 5 PM, 10 PM, 26% chose this option). Participants could respond immediately upon being prompted or could self-initiate surveys to allow for flexibility if they began drinking before the programmed time and to accommodate unexpected changes in their schedule. The morning report uniquely asked about the previous days’ drinking and alcohol-related consequences and same-day drinking intentions (see Measures). The second EMA type was the self-initiated drinking survey sequence, in which participants were instructed to initiate a “drinking survey” EMA as they were starting their first drink. Participants then responded to half-hourly prompts following the initial drink report. Drinking episode prompts continued every 30 minutes until (a) participants reported that they had not consumed alcohol in the past 30 minutes and that they were finished drinking for the next 2–3 hours, or (b) participants missed three consecutive prompts.

Transdermal Alcohol Concentration (TAC) Sensor Protocol

Participants wore the SCRAM-CAM anklet during wake and sleep hours. SCRAM-CAM uses self-generated air flow to capture transdermal ethanol evaporation. TAC is determined using fuel-cell technology (Fairbairn & Kang, 2019; van Egmond et al., 2020). After data collection, data are uploaded to the company’s server (SCRAMNet), which houses TAC data, records TAC “positives”, and tracks device wear compliance through skin temperature and sensor quality (infrared voltage) readings. SCRAMNet’s algorithms are designed to be conservative, missing a large percentage of light-to-moderate drinking days (Barnett et al., 2014; Roache et al., 2019). We applied validated research algorithms described by Roache et al., (2019), which increase sensitivity to light-to-moderate drinking with little impact on specificity. Compliance rates were high: only 2.0% of TAC data showed evidence of device removal or interference; these data points were clustered within a few individuals (n = 24) and removed. No evidence suggested compliance was associated with demographics (gender, age, race/ethnicity, student status) or AUDIT scores (ps > .05).

Social Days

Because drinking behavior does not conform to the midnight-to-midnight boundaries of a calendar day, day boundaries for TAC and EMA drinking episode data were redefined such that 10 AM marked the start of a new social day. 10 AM was chosen because it was the modal prompt time for the morning report which asked participants to reflect on their drinking the day/night before. If the morning report was provided before 10 AM, the social day boundary was adjusted to correspond to the time of the morning report (104 TAC and 7 episodic EMA observations were shifted).

Measures

Subjective Intoxication (SI)

Subjective intoxication (SI) was measured during the EMA drinking survey using two items on 5-point scales ranging from not at all (0) to extremely (4). The two items – “how buzzed are you right now?” and “how drunk are you right now?” were assessed every 30 minutes during ongoing drinking episodes. The SI scale was created by taking the mean of items at each assessment. Reliability analyses conducted using omega ($\omega$) from a multilevel confirmatory factor analysis (McDonald, 1999) showed that the reliability of the SI scale was high at each level (moment $\omega$ = 0.85, day $\omega$ = 0.93, person $\omega$ = 0.91). Combining the two SI items allowed us to a) more efficiently estimate subjective intoxication than a single item would have allowed, and b) examine the reliability of subjective intoxication measures in natural environments. We use moment-level and day-level SI in analyses that include outcomes measured in self-initiated surveys repeated every 30 minutes (i.e., eBAC). We use peak and mean SI in analyses that include outcomes at the daily level (i.e., TAC and alcohol-related consequences).

eBAC

eBAC was calculated during drinking episodes using $eBA{C}_{it}=\left[\left(c/2\right)*\left(GC/w\right)\right]-\left({\beta }_{60}*t\right)$ where $c$ = number of drinks consumed up to the current moment, $GC$ is a gender constant (7.5 for males and 9 for females), $w$ is bodyweight (pounds), ${\beta }_{60}$ is the population average hourly alcohol metabolic rate (−0.017 g/dl/hour), and $t$ is time in hours since drinking initiation (Hustad & Carey, 2005; Matthews & Miller, 1979). The first drink of each social day was the starting point from which drinks consumed and hours since initiation were accumulated. Drink reports were provided during EMA drinking surveys. A figure showing standard drink sizes was displayed every 30 minutes (12 oz of 5% beer; 5 oz of 12% wine; 1.5 oz of 40% liquor). Participants reported how many standard drinks they consumed since the last survey. First drink surveys initiating the EMA drinking survey sequence counted as one drink, unless the participant indicated that their first drink survey was not reporting the first drink of the event (n episodes=89; 15.3% of episodes), in which case participants were instead asked to report the time of the first drink and the number of drinks consumed since then using the standard drink figure. To account for consumption time of the first drink and thereby make eBAC more realistic, twenty minutes were added to all time latencies (t) following previous research (Piasecki et al., 2012). As in previous research, eBAC sometimes produced implausible values (Piasecki et al., 2012). eBAC values below 0 (n=213 observations, M(SD)= −0.04(0.04)) were set to 0 and high eBAC values were Winsorized at 0.35 (n=303 observations, M(SD)=0.41(0.03)). eBAC was summarized in daily analyses using (a) AUC, which represented the cumulative alcohol concentration according to eBAC that day and was calculated using $\Sigma \left[\frac{eBA{C}_{it}+eBA{C}_{i,t+1}}{2}\right]*\left[\mathit{\text{hour}}{s}_{i,t+1}-\mathit{\text{hour}}{s}_t\right]$, where eBAC and hours represent the eBAC reading and the time of the reading for person i at time t; and (b) peak eBAC, which was the maximum eBAC value recorded that day. eBAC AUC and peak were only calculated on social days with a self-initiated drinking survey. eBAC features were set to missing on days with no self-initiated drinking surveys (n days=824), leaving 532 eBAC-positive days.

TAC features: AUC and peak TAC

Two TAC features from each social day with TAC-positive episode data were used in the current analyses (Russell et al., 2022). If a TAC episode spanned multiple days, features were calculated separately for each day. If a day contained multiple TAC events, features were calculated using all data for the day. AUC represented the cumulative alcohol concentration that day and was calculated during TAC-positive windows using $\Sigma \left[\frac{TA{C}_{it}+TA{C}_{i,t+1}}{2}\right]*\left[\mathit{\text{hour}}{s}_{i,t+1}-\mathit{\text{hour}}{s}_t\right]$, where TAC and hours represent the TAC reading and the time of the reading for person i at time t (n drinking days with sufficient data = 692). Peak TAC was the maximum TAC value (n drinking days=694). For days with no TAC-positive events (n days=611), all features were coded as 0 if no evidence of non-compliance (removal or interference) was observed that day (n days=587) because these 0 readings could be considered valid indicators of non-drinking. Days with no TAC-positive events were set to missing if evidence of non-compliance was present, as these non-positive readings could have been due to sensor non-wear (n days=24). Combining the 587 TAC-negative days with the 694 TAC-positive days left 1281 days. For all TAC features, outliers were defined as values above the 99^th percentile. These were removed. The remaining 1274 days of TAC data across 218 persons were used in analysis.

Negative Alcohol-Related Consequences

Thirteen negative alcohol-related consequences were assessed via EMA on morning-report drinking days only. If participants endorsed drinking the day before in the morning survey (n days = 554), they then reported whether a set of alcohol-related consequences had occurred as a result of drinking yesterday/last night. Consequences were selected from the Importance of Consequences of Drinking-Short Form (Patrick & Maggs, 2011) and the Brief Young Adult Alcohol Consequences Questionnaire (adapted for daily use; Kahler et al., 2005). The specific consequences chosen were intended to provide a comprehensive listing across multiple domains, including physical symptoms (e.g., hangover, sick to your stomach, throw up), interpersonal conflict (e.g., get into an argument, get into a physical fight), safety risk (e.g., blackout, wake up in an unexpected place), sexual risk (e.g., have a sexual experience you wish you hadn’t), and miscellaneous (get into trouble with the police or campus authorities, find yourself in a situation where no one was sober enough to drive). Single consequences are presented in Supplemental Table 1 along with their frequencies across days and persons. Participants received a score each day representing the number of alcohol-related consequences they reported. Alcohol-related consequences were “led” one day (shifted up one row in the daily data file) prior to analysis to align day-level retrospective reports with prospectively collected TAC features.

Social Context and Location

Social context and location were assessed during the EMA drinking survey. For social context, participants were asked if they were alone or with someone. If particpants answered “with someone”, they were asked who they were with (not considered in the present study). For location, participants were asked where they were “right now”. Participants could indicate that they were home (apartment, dorm, house), a friend’s place, boyfriend’s/girlfriend’s place, party (house party, Greek party), bar/club, restaurant, walking somewhere, major entertainment event (sports, tailgate, concert), or other. Location was dichotomized as “outing event” (party, bar/club, restaurant, major entertainment event) or “other” to compare residential and other infrequently reported locations (e.g., walking somewhere, 2.7%) to locations where higher levels of alcohol consumption are expected (Fischer et al., 2023).

Statistical Analysis

All data analyses were conducted in R. Means, person-, day-, and moment-level SDs were generated from empty multilevel linear models using lme4::lmer. In empty multilevel models, the intercept corresponds to the random mean.

Relationship Between Subjective and Alcohol Concentration Measures

To assess H1, multilevel linear models were conducted to test the association between SI intensity and eBAC or TAC. SI and eBAC were measured during drinking episodes and had three levels of variation (moment, day, person). To examine their association, we used a three-level centering strategy to partition the variance of SI into moment, day, and person-levels. Momentary SI refers to SI collected during self-initiated EMA drinking episodes (every 30 minutes). Moment-level SI means SI varies from moment to moment, within the same drinking episode. Day-level SI means SI varies from day to day, within the same person. Person-level SI means SI varies from person to person. SI can also be summarized by its peak and mean. Peak SI can be interpreted as the highest SI reported each day, and mean SI as the average of all SIs reported each day. We include eBAC in addition to TAC because the precise timing of TAC is delayed relative to BAC and SI (Karns-Wright et al., 2017). This occurs because TAC can only be measured after alcohol is eliminated from the skin whereas other intoxication measures are measured prior to this. The delay length varies across persons and situations, which prevents clear interpretation of moment-level associations between TAC and other intoxication measures. We include eBAC as a momentary marker of alcohol concentration intoxication and include day-level features from TAC in our analyses. We calculate the peak and area under the curve (AUC) from each day with valid TAC data and test their associations with peak and mean SI at the day level. We estimate associations between SI and eBAC at both moment and day-levels. Raw SI values were centered on person-day-means (creating a moment-level SI variable), person-day-means were centered on person-means (creating a day-level SI variable), and person-means were centered on the grand mean (creating a person-level SI variable). TAC peak and AUC were day-level summaries of TAC curves that only varied at day and person levels.

To examine the association between SI and TAC features, two-level multilevel models were used with day-level SI peak or mean centered on person-means and SI person-means centered on the grand mean (creating a person-level SI peak or mean variable). Peak and mean SI were calculated from our repeated (momentary) SI measures. Peak SI represented the peak intoxication level the person felt while drinking that day. Mean SI represented the average SI level while drinking that day. Peak and mean SI correspond with alcohol concentration features, peak, and area under the curve (AUC), respectively, hence their calculation at the daily level. SI was treated as the outcome to help answer the question of “If one sees a certain SI score, what should one expect to see for eBAC and TAC?” We acknowledge that during a drinking episode, SI would come after eBAC and TAC, and do not imply temporality in these models.

To determine effect size, coefficients were standardized using the following formula $b*S{D}_x/S{D}_y$, with the appropriate SD used for the level of the association (moment, day, person). Random slopes at the moment- (for momentary SI) and/or day-level (for peak and mean SI) were included to allow for these relationships to differ across persons. Best linear unbiased predictors (BLUP) of these were plotted in a histogram and their distribution was described. Beta effect size cutoffs used for these tests were 0.3, 0.5, and 0.7 representing “weak,” “moderate,” and “strong” convergent validity (Abma et al., 2016).

For models containing momentary eBAC, moderation by social context and location were tested by including interaction terms between momentary SI and momentary and daily social context and location. For models containing daily TAC, moderation by social context and location were tested by including interaction terms between daily SI and daily social context and location.

Relationship Between SI and Alcohol-Related Consequences

To assess H2, sex (male=1, female=0), bodyweight, and social weekend (Sunday-Wednesday=0, Thursday-Saturday=1) were z-scored and included as covariates because their exclusion may bias estimates of associations between SI, eBAC, TAC, and consequences. Intoxication measures (SI peak and mean; eBAC peak and AUC, and TAC peak and AUC) were centered on person-means (Level 1) and their person-means were grand-mean centered (Level 2). All intoxication variables were then z-scored to allow comparison of effect sizes. Multilevel negative binomial models were used to predict total number of alcohol-related consequences from SI, controlling for alcohol concentration measures. Multilevel logistic models were used to predict individual alcohol-related consequences from peak or mean SI, controlling for alcohol concentration measures. IRR effect sizes estimated the proportional difference in the number of alcohol-related consequences with each unit difference in predictors. IRRs can be interpreted similarly to ORs, in that 95% confidence intervals that do not contain 1.0 are considered significant. Random slopes were generated for each model. Models were estimated in a Bayesian framework (with non-informative priors) using the brms package in R (Bürkner, 2017). Bayesian multilevel models facilitate convergence of models with correlated random effects compared to maximum likelihood. All model results are presented using the estimate and its 95% credible interval. Given their implication for the prediction and prevention of consequences, we focus on the momentary- and daily-level associations below. Person-level means were controlled for and associations are available in Tables 3–4.

Table 3.

Results of Multilevel Linear Models Showing Prediction of Objective Measures from SI

	b (95% CI)	Standardized b
Prediction of momentary eBAC from SI
Fixed Effects
Momentary SI	0.04 (0.04, 0.05)	0.51 (SD=0.19)
Daily-mean SI	0.05 (0.04, 0.06)	0.62 (SD=0.24)
Person-mean SI	0.04 (0.04, 0.05)	0.58
Intercept	0.10 (0.10, 0.11)
Day-Level Random Effects
Intercept SD	0.04 (0.03, 0.04)
Momentary SI Slope SD	0.03 (0.02, 0.03)
Correlation (intercept and moment SI slope)	0.59 (0.43, 0.73)
Person-Level Random Effects
Intercept SD	0.05 (0.04, 0.05)
Moment SI Slope SD	0.02 (0.01, 0.03)
Day SI Slope SD	0.03 (0.02, 0.04)
Correlation (intercept and moment SI slope)	0.72 (0.48, 0.93)
Correlation (intercept and day SI slope)	0.93 (0.78, 0.99)
Correlation (moment and day SI slope)	0.63 (0.24, 0.92)
Prediction of daily peak TAC from daily peak SI
Fixed Effects
Peak SI	0.05 (0.04, 0.06)	0.56 (SD=0.15)
Person-mean peak SI	0.05 (0.03, 0.06)	0.75
Intercept	0.13 (0.12, 0.14)
Random Effects
Intercept SD	0.07 (0.06, 0.08)
Slope SD	0.02 (0.01, 0.03)
Correlation (intercept and peak SI slope)	0.80 (0.38, 0.99)
Prediction of daily TAC-AUC from daily mean SI
Fixed Effects
Daily-mean mean SI	0.62 (0.48, 0.75)	0.57 (SD=0.28)
Person-mean mean SI	0.46 (0.31, 0.60)	0.71
Intercept	0.88 (0.78, 0.99)
Random Effects
Intercept SD	0.61 (0.52, 0.70)
Slope SD	0.43 (0.28, 0.59)
Correlation (intercept and day-mean SI slope)	0.85 (0.58, 0.99)

Note: AUC=area under the curve, CI=confidence interval or credible interval, eBAC=estimated blood alcohol concentration, SD=standard deviation, SI=subjective intoxication, TAC=transdermal alcohol concentration. Estimates bolded represent p<0.05. The standardized beta for the Level 1 factors (i.e., momentary SI or peak/mean SI) represents the mean random slope with SD.

Table 4.

Results of Multilevel Negative Binomial Models Showing Prediction of Alcohol-Related Consequences from SI, Controlling for Objective Measures

	Model 1 IRR (95% CI)	Model 2 IRR (95% CI)	Model 3 IRR (95% CI)	Model 4 IRR (95% CI)
	Adjusting for eBAC peak	Adjusting for TAC peak	Adjusting for eBAC AUC	Adjusting for TAC AUC
Fixed Effects
SI	1.70 (1.29, 2.23)	1.53 (1.22, 2.00)	1.54 (1.25, 1.96)	1.29 (1.07, 1.56)
Person-mean SI	1.72 (1.35, 2.21)	1.62 (1.26, 2.11)	1.42 (1.16, 1.76)	1.38 (1.11, 1.73)
Objective measure (eBAC or TAC)	1.30 (1.08, 1.58)	1.55 (1.31, 1.85)	1.24 (1.06, 1.46)	1.56 (1.35, 1.83)
Person-mean objective measure (eBAC or TAC)	1.35 (1.05, 1.74)	1.20 (0.96, 1.51)	1.64 (1.36, 2.03)	1.24 (1.00, 1.55)
Male	1.06 (0.81, 1.37)	0.89 (0.68, 1.16)	1.09 (0.85, 1.39)	0.90 (0.68, 1.18)
Weight (pounds)	1.02 (0.78, 1.32)	0.97 (0.74, 1.28)	1.01 (0.78, 1.28)	0.93 (0.69, 1.23)
Social weekend	1.03 (0.84, 1.28)	1.02 (0.83, 1.26)	1.07 (0.86, 1.34)	1.00 (0.81, 1.24)
Intercept	0.24 (0.16, 0.34)	0.16 (0.11, 0.23)	0.29 (0.20, 0.40)	0.19 (0.13, 0.27)
Random Effects
Intercept SD	0.83 (0.48, 1.17)	0.88 (0.58, 1.20)	0.60 (0.09, 0.97)	0.85 (0.57, 1.14)
SI Slope SD	0.35 (0.05, 0.64)	0.25 (0.02, 0.53)	0.26 (0.02, 0.57)	0.15 (0.01, 0.41)
Correlation (intercept and SI slope)	−0.63 (−0.98, 0.18)	−0.43 (−0.97, 0.68)	−0.24 (−0.95, 0.82)	0.09 (−0.88, 0.93)
Dispersion	7.61 (1.22, 46.96)	17.03 (1.70, 112.85)	3.45 (0.89, 15.66)	12.38 (1.45, 76.15)

Note: ARC=alcohol-related consequences, AUC=area under the curve, CI=credible interval, eBAC=estimated blood alcohol concentration, SI=subjective intoxication, SD=standard deviation, TAC=transdermal alcohol concentration. Estimates bolded represent p<0.05

Results

Descriptive Statistics

Compliance to surveys was high; 94% of scheduled morning, afternoon, and evening reports were completed, with 93% completed within two hours of the prompt time (median absolute time difference = 17.1 minutes; IQR = 0.3, 44.6 minutes). Morning, afternoon, and evening surveys were completed within 3.9 minutes (IQR = 2.5, 4.3), 1.7 minutes (IQR = 0.9, 2.0), and 2.4 minutes (IQR = 1.3, 2.9) of starting, respectively. EMA drinking survey prompts were very brief and tended to be completed quickly (median completion time = 1 minute, IQR = 0.67, 2). Self-initiated EMA drinking episodes were recorded by 92% of participants, who provided a median of 3 days with EMA drinking episode data (IQR = 2, 3). Days with EMA drinking episode data contained a median of 6 drink reports (IQR = 4, 9).

Table 1 describes participants’ demographics, baseline substance use, and drinking over the 6-day study period. Participants had an average AUDIT score of 11.9 (SD = 4.4) and the majority did not report past 30-day tobacco or cannabis use. During the study, the average number of drinks per day was 2.8 (SD = 2.2) and there were an average of 1.5 HED days per participant (SD = 1.3).

Table 1.

Participant Characteristics (N = 222)

	Frequency (%)
Sex
Male	81 (36.5%)
Female	141 (63.5%)
Age
Mean (SD)	22.3 (1.3)
Race/Ethnicity
White	185 (78.8%)
Mixed	20 (9.0%)
Asian	16 (6.8%)
Black	8 (3.6%)
Hispanic	4 (1.8%)
Student Status
Not a full time student	21 (9.5%)
Undergraduate	187 (84.2%)
Graduate	11 (5.0%)
Professional	2 (0.9%)
Other	1 (0.5%)
Greek Affiliation
Yes	40 (20.1%)
No	159 (79.9%)
AUDIT Scores
Mean (SD)	11.9 (4.4)
Past 30-Day Tobacco Use a
Yes	82 (36.9%)
No	140 (63.1%)
Past 30-Day Cannabis Use
Yes	72 (32.4%)
No	150 (67.6%)
Standard Drinks Per Day b
Mean (SD)	2.8 (2.2)
Number of HED Days b
Mean (SD)	1.5 (1.3)

Note: AUDIT = Alcohol Use Disorders Identification Test, HED = heavy episodic drinking (4+/5+ drinks in two hours for females/males), SD=standard deviation

^aTobacco use refers to cigarette or e-cigarettes

^bAlcohol use over the 6-day study period

Table 2 shows descriptive statistics for SI, eBAC, peak TAC, TAC-AUC, and alcohol-related consequences. Across all EMA drinking surveys, participants reported a mean SI of 1.11 (Median=1.0, IQR: 0.0–2.0) and a mean eBAC of 0.09 (Median=0.08, IQR: 0.03–0.17). Across all days, participants had a mean peak TAC of 0.06. (Median=0.01, IQR: 0.0–0.09) and a mean TAC-AUC of 0.43 (Median=0.01, IQR: 0.0–0.39). Across all drinking days, participants reported a mean of 0.53 (Median=0.0, IQR: 0.0–1.0) alcohol-related consequences.

Table 2.

Descriptive Statistics for Study Variables

	N persons	N days	N moments	Mean	Person-level SD	Day-level SD	Moment-level SD
SI	203	529	3234	1.11	0.56	0.53	0.81
Standard Drinks	204	536	3655	1.19	0.37	0.43	0.87
eBAC	202	532	4697	0.09	0.04	0.06	0.07
Peak TAC	218	1274	--	0.06	0.04	0.09	--
TAC-AUC	218	1272	--	0.43	0.35	0.75	--
ARCs	209	546	--	0.53	0.44	0.93	--

Note: ARC=alcohol related consequences, AUC=area under the curve, eBAC=estimated blood alcohol concentration, SD=standard deviation, SI=subjective intoxication, TAC=transdermal alcohol concentration

Associations Between Subjective Intoxication and Alcohol Concentration

Results of the full multilevel linear models and their standardized betas are presented in Table 3. For H1, two main findings are apparent concering associations between SI and eBAC. First, the average moment-level association between SI and eBAC was significant and of moderate magnitude (standardized b=0.51). Significant random slopes suggested that the moment-level association between SI and eBAC differed significantly across days (Figure 1a). For momentary SI and eBAC, the day-level SD suggests most slopes are between 0.32–0.70, the minimum and maximum slopes suggest the moment-level (within-episode) SI-eBAC association was as weak as −0.03 on some days and as high as 1.33 for others. Second, a similar magnitude of association was found at the day level – days with higher SI were moderately correlated with days higher in eBAC (standardized b=.62). A significant random slope suggested that the magnitude of this association was stronger for some people than it was for others (Figure 1b). For daily SI and eBAC, the person-level SD suggests most slopes are between 0.38–0.86, the minimum and maximum slopes suggests the association between day-mean SI and eBAC goes as low as 0.12 and as high as 1.45. These results show that SI ratings among young adults are moderately linked with eBAC ratings, and for some days and persons, these links are stronger than for others. A significant interaction effect was observed between momentary SI and momentary location (b = −0.02, 95% CI = −0.02, −0.00). This indicates a weaker association between SI and eBAC when students were “out” versus not. We did not detect any moderation by social context.

Figure 1.

.Distributions of Random Slopes Between Subjective Intoxication (SI) and Alcohol Concentration Measures.

A) Random slope between momentary SI and estimated blood alcohol concentration (eBAC). B) Random slope between daily SI and eBAC. C) Random slope between peak SI and peak transdermal alcohol concentration (TAC). D) Random slope between mean SI and TAC-AUC (area under the curve).

Table 3 also shows associations between SI and TAC features. For H1, two main findings are apparent regarding associations between SI and TAC. First, SI peak and mean were moderately associated with corresponding TAC features (peak and AUC) at the day level (standardized b = .56, .57 respectively). Second, significant random slopes suggest that this association was larger for some young adults versus others (Figures 1c, 1d). For daily peak SI and peak TAC, the SD suggests most slopes are between 0.41–0.71, the minimum and maximum slopes suggest the association goes as low as 0.19 for some and as high as 0.94 for others. For daily mean SI and TAC-AUC, the SD suggests most slopes are between 0.29–0.85, the minimum and maximum slopes suggest it goes as low as 0.02 and as high as 1.5. No other significant interaction effects were observed.

Daily Subjective Intoxication as a Predictor of Next Morning Alcohol-Related Consequences

Table 4 includes the results of four multilevel negative binomial models testing H2, including alcohol concentration (eBAC or TAC), sex, bodyweight, and social weekend as covariates.

Peak SI predicted consequences the next morning above peak eBAC (Table 4, model 1: IRR=1.70, 95% CI: 1.29, 2.23) and peak TAC (Table 4, model 2: IRR=1.53, 95% CI: 1.22, 2.00) at the day-level. Mean SI predicted total consequences above eBAC-AUC (Table 4, model 3: IRR=1.54, 95% CI: 1.25, 1.96) and TAC-AUC (Table 4, model 4: IRR=1.29, 95% CI: 1.07, 1.56) at the day-level. This suggests that SI features derived from repeated SI measurement during drinking events offer useful and unique information in the prediction of consequences even after adjusting for alcohol concentration measures. Significant random slopes were observed, suggesting the day-level relationships between SI features and consequences vary by person. The slope SD between day-level peak SI and consequences (controlling for peak eBAC) was 0.35, between day-level peak SI and consequences (controlling for peak TAC) was 0.25, between day-level mean SI and consequences (controlling for eBAC AUC) was 0.26, and between day-level mean SI and consequences (controlling for TAC AUC) was 0.15.

Single Consequence Models

Supplemental Tables 2a through 2d show multilevel logistic regression models of SI in predicting the six most prevalent consequences (hangover, being sick to your stomach, experiencing an argument, throwing up, being in bad physical “shape”, and blacking out). SI predicted next-morning hangovers above alcohol concentration at the day-level in most cases, with the exception of mean SI which did not predict hangovers when adjusting for TAC AUC. SI (peak or mean) did not consistently predict other individual consequences. The large 95% CIs for some single consequences (e.g., blackout) may have been driven by low base rates. Significant random slopes were observed for all models, suggesting that the predictive power of peak and mean SI for single consequences varied substantially across people.

Discussion

The present study sought to test two main hypotheses. First, we examined the relationship between momentary SI and alcohol concentration measures. We observed full support for H1. We found moderate associations between momentary (moment- and day-level) SI and eBAC (standardized bs in the 0.5–0.6s) and moderate associations between day-level SI and TAC (standardized bs in the 0.5s). This suggests that on average, SI serves as a straightforward indicator of intoxication. We also observed significant random slopes for nearly all SI associatons, indicating that the degree to which SI tracks alcohol concentration metrics varies both by day and by person. The variability in the strength of momentary SI-eBAC associations across days may be explained by a variety of factors at the day level, including the setting in which drinking occurs, the people with whom one is drinking, the day of the week, etc. The day-level SI-eBAC and the day-level SI-TAC associations also varied across persons, which indicates that day-level peak and mean SI may track eBAC and TAC features more closely on average for some people than for others. The differences in associations between SI and alcohol concentration measures may have been anticipated by past research and theory, suggesting that SI and alcohol concentration may be more tightly connected on average for some people versus others, due in part to both biological and experiential factors (e.g., family history of AUD, sex, tolerance). One laboratory-based study suggests that personality traits, such as impulsivity, may moderate the association between SI and alcohol concentration (Westman et al., 2016). Future studies should examine whether impulsivity and other personality traits act as moderators in naturalist study designs. Laboratory research has also revealed that social context may play a role in moderating subjective intoxication (Corbin et al., 2021; Kirkpatrick & de Wit, 2013). Similar findings have been reported in a naturalistic EMA-based study (Fischer et al., 2023). Our finding of daily variation in momentary SI-eBAC associations suggests that influences in the strength of these associations may go beyond biological or psychological individual differences, and may be partly driven by the social and physical context in which a person’s alcohol consumption is occurring. We did not, however, detect any moderation by social context. Future research should test a set of theoretically specified factors as predictors of stronger versus weaker SI-alcohol concentration links at both day and person levels to address the complexity of influences. This could help push the field forward by informing not only among whom SI most strongly tracks alcohol concentration measures, but also in what settings.

We also examined the relationship between day-level SI and alcohol-related consequences (H2), adjusting for alcohol concentration measures. This offers a test of whether SI ratings have unique predictive value relative to alcohol concentration measures, which we hypothesized based on the proximal nature of SI to consequence risk (e.g., Andreasson, 2016). Our tests are novel because we calculated features from momentary measures of SI to isolate the dynamics of SI experience that may be important for consequence risk in natural environments. Our results indicate that both peak and mean SI were significant predictors of number of consequences even when adjusting for features from alcohol concentration measures (eBAC peak and AUC, or TAC peak and AUC) at the day-level. This suggests that on days with equivalent alcohol concentration, days in which a person feels more intoxicated tend to carry higher risk for alcohol-related consequences. It is possible that SI may serve as a mediator in the relationship between eBAC and consequences. We also observed that these associations were stronger for some people than for others, suggesting that some persons’ perceptions of their intoxication are more strongly linked to consequence risk than others. This variability is important as it suggests that SI may have different predictive value for some individuals than for others, and suggests the need for future research aimed at identifying the individual characteristics that explain this variability. Supplemental analyses of single consequence outcomes showed that peak and mean SI consistently predicted next morning hangovers at equivalent alcohol concentrations levels at the day-level. Other consequences were not consistently predicted. This difficulty in prediction of single consequences beyond hangovers may have been due to their relatively low base rates.

Identifying scenarios when the relationship between subjective intoxication and alcohol concentration is weaker may aid in the prevention of severe alcohol-related consequences like drunk driving (Quinn & Fromme, 2012; Turrisi & Jaccard, 1991). Situational predictors of weak assocations might include bar drinking, pregaming, playing drinking games, or drinking with peers/friends and should be included in future studies. Understanding the person-level factors that are associated with weaker assocations could also be an important factor in prevention and intervention efforts. This notion is supported by the low level of response model that suggests individuals who respond less to the effects of alcohol are more likely to drink heavily, experience alcohol-related consequences, and develop alcohol use disorders (Schuckit, 1994; Schuckit et al., 2011; Schuckit & Smith, 2000, 2001; Trim et al., 2009).

Peak and mean SI summarized from momentary SI collected in real-time served as significant predictors of alcohol-related consequences. This indicates that SI may be a valuable measure to incorporate in a mobile prevention intervention (e.g., one which uses ecological momentary intervention; EMI). EMI is an extension of EMA that provides personalized intervention based on EMA responses (Balaskas et al., 2021). On its own, SI could be used to detect when an individual may be at highest risk for experiencing consequences and intervened on. We measured SI using a simple, two-question Likert scale which can be completed quickly. Our evidence shows that on average, this is useful information, but its degree of usefulness varies by day and by person, and future research is needed to identify the potential causes of this variability. Previous work has compared subjective intoxication ratings between real-world and laboratory settings in the same people and reported moderate-to-good correspondence (Fridberg et al., 2021). This further suggests that EMA-collected SI could be a valuable tool for risk assessment. In conjunction with a wearable sensor, incongruencies could be detected between SI and alcohol concentration. Using EMI, researchers could observe whether SI is “tracking” alcohol concentration measures (e.g., if BAC increases, does SI also increase in strong correspondence?). With the use of a real-time model, researchers could be notified if SI is not tracking well and could employ intervention. This would be especially helpful for individuals who rate their SI as low compared to their alcohol concentration. Individuals could be alerted that they may be more intoxicated than they realize, drawing awareness to how much alcohol they have consumed and potentially prevent serious consequences. Such an alert may include information on how much they have consumed and their current alcohol concentration. Other important information to provide might include if they have rapid rise rates to alcohol concentration and how to reduce their speed. A pilot EMI study that combined EMA with wearable sensors (electrodermal activity) found the intervention raised consciousness about the quantity of alcohol they consume (Leonard et al., 2017), and thus may work to decrease consumption and related harms. A similar approach could be applied for subjective intoxication.

Our results suggest that even when biological intoxication levels (i.e., eBAC or TAC) are equivalent, feeling more buzzed or drunk on one occasion versus another is important for consequence risk prediction. A person’s self-perceptions of their own intoxication level appear to be responding not only to physical cues associated with their own intoxication levels but others as well. What these other cues might be is not currenly clear and would be an intriguing avenue for future research. Possible explanations could include other substance use, food consumption, sleep, hydration, or a wide array of other physical, psychological, and contextual factors. Future research should test the mechanisms outside of biological intoxication through which SI may foretell consequence risk in natural settings.

Peak and mean SI also predicted next morning hangovers at equivalent alcohol concentration levels but did not consistently predict other individual consequences. Peak and mean SI did, however, predict more consequences overall. More than one in five drinking days resulted in a hangover and more than two in five persons experienced at least one during the six-day study period (Supplemental Table 1). One possible explanation for the ability of peak and mean SI to predict hangovers is that feeling high SI intensity suggests higher alcohol toxicity at the biological level, which leads to its unique association. It is likely that we were underpowered to detect significant associations between peak and mean SI and other single alcohol-related consequences given their much lower base rates (see Supplemental Table 1). Research focusing on longer periods of time including more high risk days may be needed to fully explore the relationship between SI and other individual consequences.

The findings also have implications regarding the utility of using TAC sensors to measure alcohol use and related consequences in field settings. Much of the recent TAC-related literature focuses on the ability to detect drinking (e.g., Barnett et al., 2014; Roache et al., 2019), correlations between drink counts and TAC (e.g., Courtney et al., 2022; Richards et al., 2021, 2022), or the translation of TAC into BAC (e.g., Dougherty et al., 2012; Fairbairn et al., 2018, 2020). Fewer studies to date have used TAC sensors to predict alcohol-related consequences (e.g., Russell et al., 2022), and these prediction models controlled for self-reported drink count. This is the first study to demonstrate: a) the relationship between TAC and SI, and b) compare the potential for TAC with subjective intoxication and eBAC in predicting consequences. Both peak TAC and TAC-AUC predicted consequences at the daily level (while controlling for SI). TAC sensors offer objective, continuous measurement of alcohol use, but they cannot be interpreted like blood alcohol concentration (BAC) values, which are grams of alcohol per deciliter of blood volume. TAC values from the SCRAM-CAM device offer a measure of grams per alcohol per 1470 l of air. Although it is not equivalent, TAC functions in the same way as BAC. TAC measures from SCRAM-CAM function well as a relative metric for both between- and within-person comparisons, in that higher values mean greater biological intoxication. TAC features show strong correlations with drinking self-reports, in that more drinking is correlated significantly and strongly with higher TAC (Russell et al., 2022). This area is still growing. Future directions could include TAC lab studies examining: 1) SI at different TAC levels, and/or 2) results on objective performance tests for coordination, speed of response, and accurarcy of recall.

Limitations

The following limitations should be considered. First, while we used a reliable, two-question measure of SI, this measure did not differentiate between the different types of alcohol responses (rewarding, stimulant, and sedative). This was partly due to our desire to test the effectiveness of a straightforward measure of SI that could be deployed easily during real-world drinking episodes. Our scale only included five response options for SI ratings because we wanted to minimize reporting burden, as we were asking this every 30 minutes while participants were drinking. However, the use of only five response options may have limited our resolution of measurement. Future research might investigate the impact of varying the number of response options in naturalistic SI assessments. Second, our sample is limited to 21–29 year olds who were primarily Non-Hispanic White, and in college. It is unclear how results would generalize to young adults of different races and ethnicities, to older individuals, and to individuals with different patterns of alcohol use and/or those with alcohol use disorder. We did not include younger college students (aged 18–20) in our study to avoid collecting objective drinking data (i.e., TAC) among participants for whom alcohol consumption was illegal. As students drink throughout college, they may gain tolerance which could affect their alcohol experiences. Future studies should examine momentary SI among more diverse populations. Third, due to low frequency of reported individual consequences other than hangovers, our models could not reliably predict the occurrence of most consequences individually. Although peak and mean SI did predict more total consequences experienced on a given day, studies with longer duration may be needed to test the predictive validity of SI for specific consequences outside hangovers. Fourth, the current study only examined negative consequences. Future research should consider how SI is related to positive consequences as well. This is important because young adults often drink for enhancement and this motive has been associated with more alcohol consumption and alcohol-related problems (Kuntsche et al., 2005; Lyvers et al., 2010). Fifth, it may be challenging for participants to calculate the number of standard drinks they consumed at each time point, especially for unusual drink sizes. This could lead to inaccurate reports. Future studies may consider assessing number, size, and type of alcohol consumed rather than asking the participant to calculate standard drinks themselves (Fridberg et al., 2021). Sixth, the current study spanned six days, and therefore may not have represented a typical week. Our relatively short study design was enacted to keep engagement in our intensive measurement scheme high. Future research may follow participants for longer periods of time to achieve a more accurate estimate of participants’ average drinking habits. Seventh, we did not include substance use in our models. Substance use may impact SI, however, we did not observe any significant associations between substance use and SI over and above alcohol use in our analyses. Eighth, for many of our analyses we included eBAC. eBAC is not a completely objective measure, as it relies on participant self-report and thus may be subject to misreporting. Including eBAC, however, allowed us to examine momentary associations, a novel aspect of this study. Ninth, it is possible we did not have the power to detect moderation effects by context. Future studies should examine moderation by context in larger samples over a longer period of time.

Conclusions

Our results suggest that momentary SI is a valid indicator of intoxication and a unique predictor of consequence risk. SI showed moderate evidence of convergent validity with eBAC and TAC on average across moments and days, but the magnitude of these associations varied widely across days and persons. Our results support real-time measurement of SI in etiological studies and suggest that more work is needed to aid our understanding of when and for whom SI measures may most closely track alcohol concentration and predict consequences. Future advances in this area may assist in the development of targeted mobile preventive interventions to potentially anticipate and reduce consequences in young adults’ natural settings.

Public Health Significance.

This study indicates that momentary subjective intoxication is a valid indicator of alcohol concentration and a unique predictor of consequence risk. Our results suggest that real-time measurement of subjective intoxication may be a straightforward and effective way to anticipate and reduce consequences in natural settings, but suggest that the correspondence of subjective intoxication with alcohol concentration measures and consequence risk varies across persons and days.