Extending the prototype willingness model to daily life: examining risk-conducive contexts and drinking among adolescents and young adults

Abstract

The Prototype Willingness Model (PWM) emphasizes the importance of risk-conducive contexts in shaping behavior, yet these contextual processes have not been examined at the daily level. This study, in line with the PWM, examined the extent to which risk-conducive contexts (e.g., personal cannabis use, others’ cannabis use, others’ drinking, others’ drinking game engagement, being in new environments or with new people) increase alcohol use on days when individuals are more willing to engage in alcohol-related behaviors. Participants (N = 996, 57% female, 44.4% White, M_age = 20.00, SD = 3.22) were recruited in Texas (United States) for a longitudinal ecological momentary assessment (EMA) study examining cognitions and alcohol use. Results of a hurdle negative binomial model indicated on days participants reported elevated (i.e., higher than their own average level) willingness to engage in alcohol use, they had greater odds of drinking and consumed more drinks. Drinking odds increased on days they were around others who drank and decreased when around new people. Two significant interactions showed willingness was positively associated with the predicted probability of consuming any alcohol and the number of drinks consumed, and the strength of the association became stronger when the participant was around others who were drinking. Findings extend the PWM by demonstrating how theoretically-based contextual processes operate at the daily level, informing intervention targets to reduce risky alcohol use.

1. Introduction

Early prevention of alcohol use and related consequences among young people is a public health priority (NIAAA, 2024), particularly in the United States (U.S.) where the legal minimum drinking age is 21. Past-year alcohol use in the U.S. is reported by 26.1% of 10th graders and 41.7% of 12th graders (Miech et al., 2025). Further, 62.3% of young adults (19–30) drank alcohol in the past month, with 25.4% having engaged in heavy episodic drinking (4+/5 + drinks for females/males) in the past two weeks (Patrick et al., 2025). Excessive alcohol use is associated with myriad negative consequences for young people (White & Hingson, 2013). Given that risky alcohol-related patterns form during adolescence and young adulthood (Arria et al., 2016), research identifying risk factors for alcohol-related decision making is imperative.

1.1. Prototype willingness model (PWM) of adolescent and young adult risk behavior

The PWM is a dual process theory that posits health-risk decision making occurs via two pathways: one based on heuristics and affect (and therefore reactive to social situations) and one based on planned, analytic reasoning (Davies & Todd, 2021; Gerrard et al., 2008). The reasoned pathway, grounded in the Theory of Reasoned Action (Fishbein & Ajzen, 1975), proposes that intentions mediate the effects of attitudes (positive or negative evaluations of a behavior; Armitage & Connor, 2001) and injunctive norms (perceived social approval of a behavior; Cialdini et al., 1991) on behavior. The social reaction pathway emphasizes behavioral willingness (openness to engaging in a risk behavior if the opportunity arises; Gerrard et al., 2008), which is posited to mediate associations between perceived vulnerability (perceived likelihood of experiencing negative consequences; Gerrard et al., 2008), descriptive norms (perceptions of how often peers engage in the behavior; Cialdini et al., 1991), prototype similarity (perceived similarity to the typical person who engages in the risk behavior; Gerrard et al., 2008), and behavior. Meta-analytic evidence demonstrates that willingness accounts for unique and additional variance in health behaviors beyond intentions, with the strongest willingness-behavior associations observed for alcohol use (Todd et al., 2016). The social reaction pathway is ideal for predicting adolescent and young adult health-risk behaviors as it focuses on behavior that is reactive to risk-conducive contexts (i.e., circumstances that facilitate but do not require or demand risky behaviors; Davies & Todd, 2021; Gerrard et al., 2008; Litt & Lewis, 2013).

Most PWM research focuses on between-person differences as it stems from cross-sectional and longitudinal research and indicates that individuals with more favorable alcohol-related PWM cognitions tend to report increased alcohol use (Glenn et al., 2023; Litt et al., 2014; Pomery et al., 2009; Rivis et al., 2006; Zimmermann & Sieverding, 2010). However, this prior research has predominantly used hypothetical scenarios to assess willingness (e.g., “Suppose you were in a situation where…how willing would you be…”; Davies & Todd, 2021), which are limited in their ability to capture reactive cognitions in the naturalistic contexts in which they occur, thereby reducing ecological validity and obscuring the situational specificity proposed by the PWM. Ecological momentary assessment (EMA) addresses these limitations as it can provide a more ecologically valid test of the willingness to behavior pathway. Although limited, existing daily findings indicate that days with elevated willingness are associated with greater likelihood and consumption of alcohol (e.g., Boyle et al., 2023; Lewis et al., 2020, 2022, 2023, 2026; Litt et al., 2024; LoParco et al., 2022; Patrick et al., 2016; Waldron et al., 2023).

1.2. Risk-conducive contexts and the PWM at the daily level

Unlike intentions, which involve effortful consideration of behavior, willingness is thought to be more contextual (Davies et al., 2021; Gerrard et al., 2008). As such, individuals may be more likely to engage in a behavior when they are both willing to engage in that behavior and when risk-conducive circumstances arise (Davies & Todd, 2021; Gerrard et al., 2008). Risk-conducive contexts may therefore moderate the pathway (See Fig. 1) from willingness to behavior. However, the PWM literature has not examined this notion and identifying contextual moderators is imperative, as it can advance understanding of when willingness translates into behavior at the daily level, thus allowing for a more ecologically valid and contextually grounded test of the PWM.

Fig. 1.

Proposed moderation effect (purple text) of risk-conductive contexts on the associations between willingness and drinking behavior within the Prototype Willingness Model’s Social Reaction Pathway (blue text).

Daily-level research shows that contextual factors can amplify alcohol-related risk behaviors (Cross et al., 2025; Lewis et al., 2023). Adolescents and young adults frequently consume alcohol in social settings where peer influences are highly salient (Cox et al., 2019; Fairbairn & Sayette, 2014; Goodell et al., 2022). Event-level research indicates that alcohol consumption increases when individuals are around peers who are drinking heavily or engaging in other risk behaviors (O’Grady et al., 2011; Patrick et al., 2016; Stevens et al., 2021). Further, the level of risk in a given situation is at least partially determined by one’s own behavior. For example, engaging in cannabis use is associated with heavier drinking and more negative consequences when substances are used simultaneously (Daros et al., 2022; Lee et al., 2017). Familiarity with one’s environment is another important contextual factor. Although findings are mixed, some evidence suggests young adults may drink more heavily with close peers due to increased comfort (Merrill et al., 2021), while others report heightened drinking in less familiar social situations, viewing alcohol as a means to ease social tension (Calhoun et al., 2024; Fairbairn & Bresin, 2017). In unfamiliar settings, young adults are also more likely to engage in high-risk drinking behaviors, such as participating in drinking games which pose elevated risk (Lewis et al., 2023; Zamboanga et al., 2021). The physical environment itself can further moderate risk as unfamiliar or new contexts introduce novel cues that intensify alcohol’s effects, whereas familiar environments may elicit compensatory physiological responses that mitigate those effects (Siegel, 2001, 2005; Siegel & Ramos, 2002).

1.3. The present study

Theoretically grounded in the PWM, the present study provides the first ecologically valid test of whether risk-conducive contexts moderate the daily-level associations between alcohol-related willingness and alcohol use. For main effects, we hypothesized that on days individuals’ willingness is higher than usual and on days involvement in riskier contexts (i.e., when one is using cannabis, when others are using cannabis, when others are drinking, when others are playing drinking games, and when in new environments and with new people) is higher than usual, participants would be more likely to drink (H1a) and would consume more drinks (H1b). Further, we hypothesized a within-person interaction between willingness and risk-conducive contexts such that the association between willingness and use will be stronger on days when in high-risk contexts (H1c).

2. Methods

2.1. Participants

Adolescents and young adults (N = 1,009) were recruited in Texas for a longitudinal measurement burst ecological momentary assessment (EMA) study that used a 3-week EMA burst design with bursts repeated quarterly across 12 months. Each EMA burst consisted of morning and afternoon surveys (up to 8 surveys per week). The larger study included the following key eligibility criteria: (1) being 15 to 25 years of age; (2) if age 18 or older: drink alcohol at least 1x/month over the last six months; if age 15–17: no drinking criterion was applied given the study’s aims that focus on willingness and intentions, which are particularly relevant for younger adolescents who have not initiated drinking; and (3) reside in Texas.

The analytic sample consisted of 996 participants who completed baseline, training, started Burst 1, completed at least 1 daily survey, and had valid values for willingness and key contextual variables used in the analytical model (see 2.4 Analytic Plan). The average age of participants at baseline was 20.00 (SD = 3.22), and 569 (57.1%) participants identified as female biological sex. The race and ethnicity for the analytic sample was 44.4% non-Hispanic White, 29.0% Hispanic, 12.7% non-Hispanic Asian, 10.1% non-Hispanic Black, and 3.8% Other. Data for the current analyses were collected at baseline and in the morning surveys of EMA Bursts 1–4. Only morning data were used in the present analyses as they captured both participants’ reports of the previous day’s drinking and contexts, as well as their willingness to drink later in the day.

2.2. Procedure

All procedures were approved by the University’s Institutional Review Board, and no adverse events occurred. Recruitment methods included advertisements, flyers, and social media ads that did not include specific dollar amounts to avoid undue influence or coercion. Individuals were first asked to consent/assent and then were directed to a brief online eligibility survey. Individuals aged 15–17 years old provided contact information for at least one parent/guardian to provide consent. Participants could earn up to $408 across the full year-long study. Participants completed an online baseline survey and a training session with a staff member via the Zoom platform to review study procedures. The EMA surveys began the first Thursday after participants completed the training session, and Bursts 2–4 occurred quarterly over 12 months. Each burst consisted of surveys on 3 consecutive weekends plus two random weekdays (up to 8 surveys per week). Participants received survey invitations via email and text messages on Thursday (afternoon survey only), Friday and Saturday (morning and afternoon surveys), and Sunday (morning survey only). Participants chose a 3-hour window between 6am and 12 pm to complete the morning survey (analyzed here), and the morning survey window could differ for weekdays and weekends. Participants completed the afternoon survey in a 1-hour window that occurred randomly between 1 pm and 6 pm. In each week of the 3-week burst, participants received a single afternoon survey on a random weekday between Monday and Wednesday, which was always followed by a morning survey the next day. Morning survey completion rates were 88.0%, 79.7%, 73.6%, and 68.9% for Bursts 1–4, respectively.

2.3. Measures

2.3.1. Daily predictors

2.3.1.1. Willingness.

Nine items assessed participants’ willingness to engage in alcohol-related behaviors for the upcoming day (e.g., drink with friends, get drunk), using the prompt: “If a situation arises tonight where you have the opportunity, how willing (i.e., open) are you to…” The response scale ranged from 0 (Not at All Willing) to 4 (Extremely Willing), and a mean score was calculated (within-person ω = 0.87; between-person ω = 0.92).

2.3.1.2. Daily risk-conducive contexts.

Participants reported whether they used cannabis the previous day. Participants also responded to several items assessing whether other people they were around the prior day engaged in a risk behavior (drank alcohol, played drinking games, used cannabis). Responses were: 1 = Yes, 0 = No, or 2 = Don’t Know, and “2″ was recoded as “0.” Finally, two items assessed participants’ familiarity with their environment (“Did you go to a new location yesterday?” and “Did you spend time with people you just met yesterday?”). Responses were: 1 = Yes and 0 = No.

2.3.2. Daily alcohol use

Participants were asked, “Since the time you woke up to the time you went to sleep yesterday, did you drink alcohol?” with responses 1 = Yes and 0 = No. On days participants reported drinking the prior day, they were asked, “How many drinks did you consume yesterday?” with responses from 1 (1 Drink) to 15 (15 or More Drinks).

2.3.3. Daily covariates

Daily-level covariates were: Weekend (1 = Friday or Saturday, 0 = Otherwise), Week of Burst (0–2), and Burst number (0–3).

2.3.4. Baseline covariates

Age (in years) and biological sex (0 = Female and 1 = Male) were included.

2.4. Analytic plan

Willingness was assessed in reference to the upcoming evening. All risk-conducive context items were in reference to the previous day. Therefore, these variables were lagged to match the prior day’s willingness. Because of the multilevel data structure of the EMA design with days (Level 1) nested within participants (Level 2), mixed effects models with a participant-level random intercept were conducted. To accommodate zero-inflation for the count outcome of the number of drinks consumed, mixed effects hurdle models were estimated. Model comparisons indicated that the hurdle negative binomial model provided better fit than a hurdle Poisson alternative based on AIC and likelihood ratio tests, and was therefore used as the primary analytic model. Each model has a logistic regression portion estimating adjusted odds ratios (OR) for any drinks vs. no drinks consumed on a given day as well as a negative binomial regression portion estimating rate ratios (RR) for the number of drinks (Feng, 2021). Because model comparison supported heterogeneity in the within-person willingness effect, a random slope for daily willingness was included at the participant level.

To evaluate the moderating effects of risk conducive contexts on the daily-level associations between willingness and alcohol use, interaction terms between each of the six risk conducive factors and willingness were estimated in a single model. Twelve interaction terms (six for the logistic regression portion; six for the negative binomial regression portion) were tested. Analyses were performed in R (version 4.4.1). The mixed effects models were estimated using the “glmmTMB” R package (Brooks et al., 2017). In each statistical model, Level 2 covariates were grand-mean centered to estimate the effects of differences from person-level averages across participants. At Level 1, covariates were centered within-person to test daily fluctuations relative to participants’ own aggregated averages. All models included age, biological sex, weekend, week of burst, and burst number as covariates.

Given the top-coded response option for number of drinks, where 15 or more drinks was coded as 15, we conducted sensitivity analyses to evaluate whether this coding decision affected the results. All top-coded observations (15+) were recoded to higher values (16, 18, and 20) and the analytic models were re-estimated.

3. Results

3.1. Descriptive information

Table 1 provides the descriptive statistics for key variables. The final analytic sample included 996 participants across 17,392 days. Between persons, participants reported an average willingness of 0.61 (SD = 0.56) on a 0–4 response scale. Regarding alcohol use, participants reported an average 0.84 drinks per day (SD = 1.98) across all days, at the between-person level.

Table 1.

Descriptive statistics for key variables.

Predictor	N (people or days)	M (SD) or n (%)	Range
Person-Level
Male biological sex	996	0.57(0.57)	0–1
Age	996	20(3.22)	15–25
Willingness	996	0.61(0.56)	0–3.31
Cannabis use	996	0.12(0.24)	0–1
New location	996	0.26(0.19)	0–1
New people	996	0.18(0.17)	0–1
Others’ drinking	996	0.29(0.24)	0–1
Others’ playing drinking games	996	0.04(0.09)	0–1
Others’ cannabis use	996	0.13(0.22)	0–1
Number of drinks	996	0.85(1.25)	0–12.86
Daily-Level
Willingness	17,392	0.57(0.69)	0–4
Cannabis use	17,392	0.1(0.3)	0–1
New location	17,392	0.26(0.44)	0–1
New people	17,392	0.18(0.38)	0–1
Others’ drinking	17,392	0.3(0.46)	0–1
Others’ playing drinking games	17,392	0.04(0.2)	0–1
Others’ cannabis use	17,392	0.12(0.33)	0–1
Number of drinks	17,392	0.84(1.98)	0–15

Note. Male biological sex was coded as 0 for female and 1 for male.

3.2. Daily-level association between willingness and alcohol use

Table 2 shows the results for testing the daily-level associations between willingness and alcohol use from the mixed effects hurdle negative binomial model. The top panel shows the logistic regression portion testing associations between each predictor and the probability of any alcohol use, whereas the bottom panel shows the zero-truncated negative binomial regression portion testing associations between each predictor and the number of drinks consumed.

Table 2.

Mixed effects zero-inflated negative binomial models testing daily associations between willingness and number of drinks.

	Predictor: willingness
	OR/RR	95% CI LL, UL
Main Effects Model
Logistic Regression Portion (Consuming Any Drinks)
Intercept	0.05***	0.04, 0.07
Person-level (Between-person)
Male biological sex	1.05	0.85, 1.31
Age	3.32***	2.93, 3.77
Willingness (GMC)	1.68***	1.49, 1.90
Cannabis use (GMC)	1.09	0.95, 1.25
New location (GMC)	1.07	0.95, 1.21
New people (GMC)	0.84**	0.74, 0.95
Others’ drinking (GMC)	3.14***	2.77, 3.56
Others’ drinking games (GMC)	1.08	0.97, 1.20
Others’ cannabis use (GMC)	0.94	0.81, 1.09
Daily-level (Within-person)
Weekend	1.57***	1.29, 1.91
Week of burst	1.02	0.97, 1.08
Burst	1.1**	1.04, 1.16
Willingness (CWP)	2.92***	2.47, 3.46
Cannabis use (CWP)	1.13	0.88, 1.47
New location (CWP)	1.52***	1.33, 1.74
New people (CWP)	1.18*	1.01, 1.39
Others’ drinking (CWP)	22.88***	19.89, 26.32
Others’ drinking games (CWP)	1.76***	1.35, 2.29
Others’ cannabis use (CWP)	0.98	0.80, 1.20
Within-Person (CWP) Interaction Terms
Willingness × Cannabis use	0.85	0.53, 1.37
Willingness × New location	1.12	0.82, 1.51
Willingness × New people	1.29	0.91, 1.82
Willingness × Others’ drinking	1.46*	1.07, 1.99
Willingness × Others’ drinking games	1.19	0.71, 2.00
Willingness × Others’ cannabis use	0.9	0.59, 1.36
Truncated Negative Binomial Regression Portion (Number of Drinks)
Intercept	1.2**	1.06, 1.36
Person-level (Between-person)
Male Biological Sex	1.28***	1.17, 1.40
Age	1.05	0.99, 1.12
Willingness (GMC)	1.15***	1.09, 1.21
Cannabis use (GMC)	0.99	0.93, 1.04
New location (GMC)	0.95*	0.91, 1.00
New people (GMC)	1.04	0.99, 1.09
Others’ drinking (GMC)	1.23***	1.17, 1.30
Others’ drinking games (GMC)	1.06**	1.02, 1.10
Others’ cannabis use (GMC)	1.02	0.96, 1.08
Daily-level (Within-person)
Weekend	1.22***	1.10, 1.35
Week of Burst	1.02*	1.00, 1.04
Burst	1.03*	1.01, 1.05
Willingness (CWP)	1.44***	1.34, 1.56
Cannabis use (CWP)	0.95	0.86, 1.05
New location (CWP)	1.08**	1.03, 1.14
New people (CWP)	1.22***	1.15, 1.30
Others’ drinking (CWP)	1.37***	1.28, 1.48
Others’ drinking games (CWP)	1.4***	1.30, 1.51
Others’ cannabis use (CWP)	1.07	1.00, 1.15
Within-Person (CWP) Interaction Terms
Willingness × Cannabis use	0.94	0.83, 1.07
Willingness × New location	1.01	0.92, 1.10
Willingness × New people	0.97	0.88, 1.07
Willingness × Others’ drinking	1.19**	1.05, 1.35
Willingness × Others’ drinking games	0.91	0.82, 1.00
Willingness × Others’ cannabis use	1.02	0.91, 1.14

Note. The dispersion parameter from the final hurdle negative binomial model was 0.152. OR = odds ratio; RR = rate ratio; LL = lower limit; UL = upper limit.

Person-level predictors were grand-mean-centered (GMC). Daily-level predictors were centered-within-person (CWP). Number of persons = 996, Number of days = 17392.

^*p < 0.05

^**p < 0.01

^***p < 0.001.

In the logistic regression portion of the model (H1a), within-person results indicated that on days participants had elevated willingness to engage in alcohol and cannabis related behaviors, they had greater odds of drinking alcohol that day (OR with respect to 1-unit increase in the willingness scale: 2.92; 95% CI for OR: [2.47, 3.46]). Results also indicated participants had greater odds of drinking on days they went to new locations or met new people, and on days other people around them drank or played drinking games. Between-person results indicated participants who, on average across all study days, reported greater willingness, met fewer new people, and spent more time with people who drank generally had more alcohol use days.

In the negative binomial regression portion of the model (H1b), within-person results indicated that on days participants reported elevated (i.e., higher than their own average level) willingness to engage in alcohol and cannabis related behaviors, they drank a higher number of drinks (rate ratio (RR) with respect to 1-unit increase in the willingness scale: 1.44; 95% CI for RR: [1.34, 1.56]). Within-person results also indicated participants drank more on days they were in a new location or met new people, and on days other people around them drank, or played drinking games. Between-person results indicated participants who identified as male and who, on average across all study days, reported greater willingness, spent less time in new location, and spent more time with people who drank or played drinking games generally reported consuming more drinks.

Results of the interaction terms used to test hypothesis H1c showed that two of the twelve interaction terms were significant. The interaction between willingness and others’ drinking was statistically significant for the logistic regression portion of the model, indicating the effect of willingness was more strongly associated with the number of drinks consumed when other people around drank alcohol. Specifically, the interaction plot (part (A) of Fig. 2) indicates that willingness was positively associated with the predicted probability of consuming any alcohol, and the strength of the association became stronger when other people around drank alcohol. In addition, the same interaction was also statistically significant for the negative binomial portion, indicating that willingness was more strongly associated with the number of drinks consumed when other people around drank alcohol (part (B) of Fig. 2).

Fig. 2.

Moderation effects of whether other people that participant was around drank alcohol on the associations between willingness and drinking outcomes. Part A shows the predicted probability of any alcohol use, and Part B shows the predicted number of drinks. Because the moderator was a binary variable centered within person, the two lines represent the two meaningful levels for a typical participant (−0.5 = others not drinking; +0.5 = others drinking).

3.3. Sensitivity analyses

The top-coded 15+ drinks category accounted for 37 out of 17,392 observations (0.2%) in the analytic sample. Across all recoding schemes (15+ recoded to 16, 18, and 20), the significant interaction between willingness and others’ drinking remained statistically significant and showed similar effect sizes in both portions. In the logistic portion of the model, the interaction between willingness and others’ drinking was associated with greater odds of drinking on days when others were drinking (ORs ranged from 1.45 to 1.46 across sensitivity models; 95% CIs overlapped substantially with those from the primary analysis). In the count portion, the same interaction remained significant, with similar rate ratios across sensitivity models (RRs rounded to 1.19 across sensitivity models; 95% CIs again largely overlapping with those from the primary analysis). No new interactions were statistically significant in the sensitivity models.

4. Discussion

The present study sought to explore moderating effects of risk-conducive contexts on the willingness to behavior pathway of the PWM (Gerrard et al., 2008) at the daily level. Our findings partially confirmed hypotheses such that on days where willingness was higher than normal, individuals were both more likely to consume any alcohol and to drink more heavily. Although mean willingness was low overall, its associations with drinking likelihood and quantity suggest that increases in even low levels of willingness are meaningfully associated with alcohol use. Results further showed on days where an individual was with new people, in a novel location, with others who were drinking, and others who played drinking games, individuals were both more likely to consume any alcohol and to consume greater amounts.

We found partial support for interactions between willingness and risk-conducive contexts, such that the associations between willingness and both drinking likelihood and quantity of drinks were strengthened when individuals were with others who were drinking. This pattern is consistent with event-level research showing that alcohol consumption increases in the presence of peers who are drinking heavily (O’Grady et al., 2011; Patrick et al., 2016; Stevens et al., 2021) as well as the broader social norms literature demonstrating that perceived and observed peer behavior is among the strongest predictors of alcohol use (Borsari & Carey, 2003; Watts et al., 2024).

The current findings extend the PWM by providing evidence that specific risk-conducive contexts, namely being around others who drink alcohol, moderate the willingness–behavior pathway at the daily level. This supports the PWM’s assertion that willingness operates in reactive ways sensitive to context (Davies & Todd, 2021; Gerrard et al., 2008; Litt & Lewis, 2013). To date, most PWM research has relied on hypothetical or retrospective assessments of willingness. As such, an ecologically valid test using daily-level data was necessary to evaluate whether specific contexts influence the willingness–behavior association in daily life. Findings demonstrate the value of incorporating daily contextual factors into dual-process models to more accurately reflect how health risk behaviors unfold in everyday life.

4.1. Clinical implications

Understanding the contexts in which individuals’ willingness is most likely to be associated with alcohol use offers valuable insight for intervention development. The present findings suggest that interventions may be particularly effective when delivered on days when individuals are both willing to drink and are around others who are drinking. These findings could inform just-in-time adaptive intervention strategies that teach coping skills, refusal strategies, or alternative activities specifically for these high-risk social situations. Future research should test these approaches in intervention trials that systematically target these high-risk social contexts. By aligning intervention delivery with socially high-risk days, these trials could establish whether context-focused, targeted interventions are more effective than approaches delivered irrespective of social circumstances.

4.2. Limitations and future directions

The study was conducted with a sample of adolescents and young adults living in Texas, thus the generalizability of our findings to other populations is uncertain. In addition, given that within the study design participants were not assessed every day of the week, it is likely that some drinking days were not assessed. Although EMA methods are better suited to capture reactive, in-the-moment cognitions such as willingness than retrospective or hypothetical measures, future research is needed to identify optimal approaches for assessing these constructs with greater precision and minimal reactivity. Another limitation was the use of single items to assess familiarity of locations and people and general reference to the previous day (versus specific drinking occasions). Future work could examine more nuanced information about the types of people (e.g., large groups vs. small groups, mixed gender groups) or locations (i.e., bar vs. party vs. own home) which may provide additional contextual information, particularly if there were several drinking contexts throughout the day. Requiring parental consent for 15–17-year-olds may have slightly affected recruitment or self-reporting; however, adolescents were assured their responses would remain confidential and not be shared with parents. Finally, given that drinking criteria were relatively low, future research should examine whether findings generalize to heavier-drinking adolescents and young adults to determine whether the willingness–behavior pathway and context-specific risk factors operate similarly across different drinking levels.

5. Conclusion

This study offered a rigorous examination of the PWM willingness–behavior pathway and how contextual factors may strengthen that association at the daily level among a diverse sample of adolescents and young adults in the U.S. Results highlight the important role of situational contexts in shaping drinking behavior and suggest potential targets for preventive intervention efforts aimed at reducing alcohol use. By identifying when and where willingness is most likely to translate into behavior, this research advances theoretical understanding and provides a foundation for the development of more context-sensitive, real-time prevention strategies.

Data availability

Data will be made available on request.