Testing Affect Regulation Models of Drinking Prior to and After Drinking Initiation Using Ecological Momentary Assessment

Abstract

Objective:

Affect regulation models of drinking state that affect motivates and reinforces drinking. Few studies have been able to elucidate the timing of these associations in natural settings. We tested positive affect (PA) and negative affect (NA) as predictors of drinking behavior, both prior to and during drinking episodes, and whether drinking predicted changes in affect during episodes.

Method:

222 regularly drinking young adults (21-29 years, 84% undergraduates), completed an ecological momentary assessment (EMA) protocol for five consecutive 24-hour periods stretching across six days (Wednesday–Monday). Participants provided PA and NA reports three times daily and every half hour during drinking episodes. Alcohol consumption reports were provided each morning and every half hour during drinking episodes.

Results:

Multilevel models showed that greater pre-drinking PA predicted higher odds of drinking, but not number of drinks consumed. Pre-drinking NA did not predict same day odds of drinking or drinks consumed. Episode-level results revealed different associations for PA and NA with drinking. Current PA did not predict drinks consumed over the next half hour; however, increased drinking was associated with greater increases in PA over the next half hour.. Higher NA predicted fewer drinks consumed in the next half hour and higher odds of the end of a drinking episode; however, increased drinking was not associated with changes in NA.

Conclusions:

PA increased following drinking during episodes. Our results suggest that a focus on PA prior to episodes and a focus on NA during episodes may interrupt processes leading to heavy drinking, and may therefore aid prevention efforts.

Introduction

Affect regulation models of alcohol use examine subjective affective experiences and their associations with alcohol use (Dvorak et al., 2018) and include two underlying tenets: (1) affect motivates drinking behavior, and (2) drinking enhances PA and/or diminishes NA, thereby reinforcing drinking. In these models, affect as a motive for drinking refers to affect promoting or leading to subsequent drinking behavior. Potential mechanisms underlying affect as a motive for drinking include anticipation of drinking and related expectations of alcohol effects, specific motivations for alcohol use (e.g., coping, social motivations), etc. (Cooper et al., 2008; Sher & Grekin, 2007; Wray et al., 2012). The reinforcing effects of alcohol may occur through positive reinforcement, in which individuals drink as a reward or to enhance positive experiences and PA (Cooper et al., 1992; Dvorak et al., 2018), or negative reinforcement, in which individuals drink to cope with NA or to reduce the effects of negative experiences (Cooper et al., 1992). Affect regulation models hold inherent assumptions regarding the temporality underlying the affect-alcohol use association, suggesting that an individual’s positive or negative affect motivates their alcohol consumption over short time frames, typically on the same day (Cooper et al., 1992; Treloar et al., 2015). Additionally, the effects of drinking on affect are typically assumed to occur immediately post-drinking (Cooper et al., 1992; Treloar et al., 2015). Therefore, purpose of this manuscript was to test the temporal assumptions underlying affect-alcohol use associations at the day-level and during ongoing drinking episodes.

Laboratory-based studies show that experimentally inducing PA can increase alcohol consumption under certain conditions or in some populations (e.g., males) (Dinc & Cooper, 2015; Stein et al., 2000; Wardell et al., 2012), and that alcohol consumption increases PA (Conrod et al., 2001; Wilkie & Stewart, 2005). Bresin et al.’s meta-analysis of lab-based manipulations found small-to-medium effects of NA on increasing alcohol consumption, and subsequent increases in alcohol craving, providing support for an association between NA and drinking behavior (Bresin et al., 2018). Similarly, a meta-analysis found evidence that alcohol use decreases NA and dampens the effect of stress on NA (Bresin, 2019), though one lab-based study found that alcohol use increased NA (Wilkie & Stewart, 2005). Lab-based research also suggests that affective responses to drinking may also be moderated by alcohol use expectancies (Grant & Stewart, 2007; Sitharthan et al., 2009; Wardell et al., 2012). These lab-based studies provide support for the internal validity of both tenets of affect regulation models of alcohol use. However, ecologically valid approaches are needed to account for the temporal nuances of affect regulation models and to examine the directionally of affect-alcohol use associations during theoretically relevant time periods.

The majority of previous studies have examined associations between positive and negative affect and drinking behavior across days, with fewer studies testing associations within days. With regard to PA, these studies found that individuals consumed more alcohol on days when they experienced greater PA (Armeli et al., 2000; Crooke et al., 2013; Howard et al., 2015; Mohr et al., 2005). Associations with NA were mixed, with some studies finding that individuals were more likely to consume alcohol on days with greater NA (Armeli et al., 2000, 2008), and other studies finding no associations between NA and drinking behavior across days (Crooke et al., 2013; Howard et al., 2015; Hussong et al., 2001). The results of these studies limit researchers to inferring that there likely exists a positive correlation between PA and alcohol use; however, the directionality of the association remains unclear, as the majority of these studies were not able to discern whether affect predicts alcohol use or vice-versa. This prohibits researchers from specifically testing or differentiating both of the underlying tenets of affect regulation models of alcohol use.

Other daily process studies have examined the associations between daytime positive and negative affect and drinking behaviors on the same day. These studies most commonly find a positive association between daytime PA and increased likelihood of drinking and greater volume of alcohol consumed on the same day (Dvorak & Simons, 2014; Mohr et al., 2008; Simons et al., 2005, 2010, 2014; Swendsen et al., 2000). These studies also suggest a positive association between greater daytime NA and likelihood of drinking, volume of alcohol consumed, and shorter time to drinking on the same day (Dvorak et al., 2014; Grant et al., 2009; Mohr et al., 2005; Simons et al., 2005, 2014). Overall, daytime PA positively correlates with nighttime drinking behaviors, which appears to support the affect regulation model’w’s daytime affect motivates their drinking behaviors on the same day (Cooper et al., 1992; Treloar et al., 2015). However, the inferences that can be drawn from these studies have two limitations. Firstly, they examine affect-alcohol use associations on days when an individual is drinking. Secondly, they do not specifically identify true pre-drinking affect. Without specifically identifying pre-drinking affect and examining affect during similar timeframes on non-drinking days, researchers cannot infer whether pre-drinking daytime affect is sufficiently different between drinking and non-drinking days to suggest that pre-drinking affect serves as a signal for same day alcohol use; thereby limiting the theoretical and intervention implications of study findings.

A few daily process studies help address the question of whether daytime affect motivates same day drinking behavior by examining pre-drinking affect on drinking versus non-drinking days, and by examining how affect changes throughout the pre-drinking time period (Dvorak et al., 2018; Russell et al., 2020; Treloar et al., 2015). These studies found that pre-drinking PA was higher and/or increased prior to drinking on drinking versus non-drinking days (Dvorak et al., 2018; Russell et al., 2020; Treloar et al., 2015); whereas, PA decreased or showed less change throughout the day on non-drinking versus drinking days (Russell et al., 2020; Treloar et al., 2015). Findings for NA are less clear, with studies finding no differences in pre-drinking NA or lower pre-drinking NA on drinking versus non-drinking days (Dvorak et al., 2018; Russell et al., 2020). NA also decreased significantly prior to drinking on drinking days; whereas, there were no changes in NA on non-drinking days (Russell et al., 2020; Treloar et al., 2015). The aforementioned studies support the affect regulation model’s first tenet that affect motivates drinking behavior by demonstrating clear differences between pre-drinking PA on drinking versus non-drinking days, suggesting that pre-drinking PA in particular might be useful for predicting same day drinking behavior. One advantage of these studies is their use of intensive, repeated measures for examining trajectories of change in pre- and post-drinking affect, providing information that can inform theory and characterize affect-alcohol use associations. However, rather than relying on measuring an individual’s trajectory of change in pre-drinking affect, researchers could benefit from identifying whether a single or average measure of pre-drinking affect predicts the odds of drinking or the volume of alcohol consumed on drinking versus non-drinking days. This could be useful for indicating whether pre-drinking affect can serve as a signal for same day drinking behavior and for indicating whether just-in-time interventions that occur prior to drinking onset can use a single or average measure of pre-drinking affect to target drinking initiation, drinking volume, or both.

The current study reports on a sample of 222 regularly drinking young adults (ages 21-29) who provided ecological momentary assessments (EMA) on affect and drinking before, during, and after drinking episodes for five consecutive days. The design of this study allows for nuanced temporal examinations of affect-alcohol use associations in a natural setting. Using this design, we set out to answer the following research questions:

1. Does pre-drinking positive or negative affect predict the odds of drinking on any given day?

2. Does pre-drinking positive or negative affect predict the number of drinks consumed on a drinking day?

These questions address the first tenet of affect regulation models, which state that pre-drinking affect may motivate alcohol consumption. Here we test whether elevated pre-drinking affect serves as a signal for the initiation and intensity of drinking later in the day, which could inform interventions that occur prior to the onset of drinking. We also set out to answer the following question:

• 3.Does positive or negative affect predict subsequent number of drinks consumed during an ongoing drinking episode?

This research question will specifically test the first tenet of affect regulation models during every 30-minutes within an ongoing drinking episode. This extends beyond the previously mentioned studies by using a more discrete timeframe during an ongoing drinking episode and will help address whether the drinking episode is a temporally relevant timeframe over which affect impacts drinking behaviors. These three research questions address the first tenet of affect regulation models, affect as motive for drinking behavior; however, they do not address the second tenet of affect regulation models, whether drinking enhances PA and/or diminishes NA.

Previous research examining the affect enhancing effects of alcohol is sparse and occurred over a variety of time frames. These studies generally found that PA increased post-drinking across days (Crooke et al., 2013), on average (Dvorak et al., 2018), after the first drink (Treloar et al., 2015), or within the first two hours after drinking initiation (Russell et al., 2020). Post-drinking PA is also higher on drinking versus non-drinking days (Dvorak et al., 2018). Overall, these studies suggest that drinking immediately enhances PA, supporting the second tenet of affect regulation models of alcohol use.

Studies examining the effects of drinking on NA across multiple days or a single day suggest that NA either decreases (Gorka et al., 2017; Hussong et al., 2001) or does not change post-drinking (Crooke et al., 2013). Examining the immediate effects of drinking suggests that NA decreases after the first drink or within the first hour post-drinking (Russell et al., 2020; Treloar et al., 2015). Post-drinking NA also appears to be lower on drinking versus non-drinking days (Russell et al., 2020), and only changes on drinking days (Russell et al., 2020; Treloar et al., 2015). These findings suggest that drinking may diminish NA, supporting the second tenet of affect regulation models of alcohol use. While the previous studies examined post-drinking changes in PA and NA, they did not specifically examine these associations during each 30-minute time period during an ongoing drinking episode. Understanding affect-alcohol use associations during this discrete time period could provide unique information about the temporality underlying affect-alcohol use associations. Therefore, we asked the question:

• 4.Does the number of drinks an individual consumes predict their positive or negative affect during an ongoing drinking episode?

This research question addresses the second tenet of the affect regulation model within a temporally relevant and useful timeframe. Examining affect-alcohol use associations during drinking episodes, as explored through research questions 3 and 4, will provide novel information that has the potential to inform just-in-time interventions that could occur after the onset of drinking, specifically within the context of an ongoing drinking episode. Overall, this study will provide unique information about both tenets of the affect regulation model of alcohol use within the temporally relevant timeframes of a single day and every 30-minutes during an ongoing drinking episode, which will inform theory and provide practically relevant information for intervention development.

Methods

Participants

Participants included a non-clinical sample of 222 young adults aged 21-29 (M age (SD) = 22.3 (1.3), 64% female, 78% non-Hispanic white) who engaged in regular heavy episodic drinking, defined as consuming 4+/5+ drinks for females/males in a row at least once a week in the past calendar or academic year. Most participants (90.5%) were full-time students. Of these students, 93% were undergraduates and the remaining 7% held graduate, professional, or “other” student status. This study was approved by the Pennsylvania State University Institutional Review Board. Participants provided informed consent to participate in the study.

Recruitment and screening

Participants were recruited on and around the campus of a large northeastern US university using flyers containing a study description and our study email address. When participants emailed, research staff responded with a brief study description and a screening survey link. Participants were eligible if they were between ages 21-29 and if they reported weekly heavy episodic drinking over the past year on the screening survey. 531 individuals completed the screening survey and 419 (78.9%) were eligible. Eligible versus ineligible participants were more likely to be male (35% versus 26%, p=0.066), white (83% versus 72%, p=0.008), and undergraduates (93% versus 73%, p <0.001). 419 participants were eligible and 343 were invited to participate. Time and resource limitations prevented invitations to all eligible participants; invitations were sent in the order in which screening surveys were received. Of the 343 invited, 222 completed the study. Of the 121 who were invited but did not participate, 97 did not respond to invitations or did not attend the appointment, 4 graduated or moved away, 2 were cancelled by the study team due to space limitations, 17 were cancelled due to the COVID-19 university shutdown, and one withdrew their participation at baseline. No evidence of bias was observed when comparing those who completed participation versus those who did not on dimensions of sex, race/ethnicity, student status, or past-two-week binge drinking frequency (all ps ≥ .10).

Procedure

The study lasted five consecutive days, beginning on a Wednesday and ending on the following Monday. The study was designed to track alcohol misuse intensively during the social weekend, and included baseline and endpoint assessments, three-times daily EMA (to capture pre- and post-drinking reports where possible and obtain retrospective previous-day drinking reports), drinking event-based EMA sequences, and transdermal sensor wear. Due to the protocol’s intensive nature, we limited its duration to five days in order to avoid overburdening participants and maximize data quality. We chose the days of Wednesday through Monday to capture all days of the “social weekend” (e.g., Thursday - Saturday; (Finlay et al., 2012)). Participants began by attending the baseline appointment in our research laboratory. Participants were consented before privately completing a 30-minute questionnaire containing measures of background/demographics, alcohol-related habits (e.g., drinking history, past-year drinking and consequences), personality, and other psychosocial factors (e.g., stress, social support). Following questionnaire completion, participants were fitted for the alcohol monitoring anklet and received training on EMA mobile phone surveys. Following the baseline session, participants immediately began the five-day field protocol. During the field protocol, participants wore a transdermal alcohol monitor on their ankle (the Secure Continuous Remote Alcohol Monitoring-Continuous Alcohol Monitor or SCRAM-CAM; data are not presented in the current report) and carried a mobile phone on which they were prompted to complete three EMA surveys daily, along with an event-initiated drinking survey sequence. Participants were compensated up to $110 for their participation.

Ecological momentary assessment protocol

Participants were provided with an Android-based smartphone that they carried with them for the five days of the study. The device was equipped with our EMA survey app, which contained two survey types. The first was the scheduled EMA survey, which prompted participants to complete surveys in the morning (scheduled for 10 AM), afternoon (4 PM), and evening (9 PM). Participants could also choose to have these prompts occur one hour later (11 AM, 5 PM, 10 PM, 26% chose this option). Participants could respond directly at prompt times or self-initiate these diaries if the prompt was missed. Compliance with reports was high: 94% of scheduled EMA reports were provided and 93% of scheduled EMA reports were provided within two hours of the scheduled prompt time (median absolute time difference = 17.1 minutes; IQR = 0.3, 44.6 minutes). Morning, afternoon, and evening reports tended to be completed within a few minutes. Median completion times were 3.9 minutes (IQR = 2.5, 4.3) for the morning report, 1.7 minutes for the afternoon report (IQR = 0.9, 2.0), and 2.4 minutes for the evening report (IQR = 1.3, 2.9). The morning report uniquely queried about the previous days’ drinking and alcohol-related consequences. The second EMA survey type was the event-initiated drinking EMA, in which participants were instructed to initiate a “drinking survey” EMA as they were starting their first drink. The initiated drinking survey was counted as 1 drink if the participant reported that it was their first drink of the episode; participants then reported the number of drinks they consumed in 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. EMA drinking episode prompts were very brief and tended to be completed quickly (median completion time = 1 minute, IQR = 0.67, 2). Ninety-two percent of participants provided EMA drinking episode data; these participants 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).

To incentivize compliance, participants were compensated $2 for each scheduled EMA report they provided, with a $5 bonus provided at the end of the study if they completed 90% or more of their scheduled EMA reports (morning, afternoon, and evening). Participants were not compensated for drinking episode EMA reports so as not to inadvertently incentivize drinking. Because drinking often begins prior to and extends past midnight, day boundaries for drinking episode EMA were redefined such that 10 AM marked the start of a new day. 10 AM was chosen as it corresponded to modal prompt time for the morning report, which asked participants to reflect on their drinking the day/night before. Compliance with EMA reports was high: EMA drinking episode data was provided on 89% of morning-report identified drinking days, and 84.3% of days with an EMA drinking episode were corroborated by the TAC sensor, a detection rate consistent with prior research (van Egmond et al., 2020).

Measures

Affect

Positive and negative affect were assessed during each scheduled morning, afternoon, and evening report, and during drinking episode reports using the positive and negative affect scale (PANAS) (Watson et al., 1988), adapted for EMA. Questions in the morning, afternoon, and evening referred to the person’s specific affective experiences across that time period (e.g., “this morning” for morning EMA); drinking episode affect reports referred to affective experiences “right now”. Participants were asked to rate each affect item on a 5-point scale from not at all (0) to extremely (4). Morning, afternoon, and evening survey items for PA included happy, lively, energetic, cheerful, calm, and relaxed. Morning, afternoon, and evening survey items for NA included sad, angry, depressed, nervous, tense, hostile, and lonely. During the drinking episode, PA and NA scales were reduced in size to offset increased reporting burden associated with half-hourly reporting. PA during drinking episodes was measured using a single item “happy”. NA during drinking episodes was measured using sad, angry, and anxious. To assess scale reliability, Cronbach’s alphas (α) were calculated at the within-episode, between-episode, and between-person levels for PA and NA measured during drinking episodes, and at the within-person and between-person levels for PA and NA measured during morning, afternoon, and evening surveys. Cronbach’s alphas for morning, afternoon, and evening PA were good at the within-person (α = 0.81–0.84) and between-person (α = 0.88–0.90) levels. For drinking episode PA, Cronbach’s alphas were not calculated due to using a single item to measure PA.¹ For morning NA, Cronbach’s alphas for morning, afternoon, and evening NA were acceptable at the within-person (α = 0.77–0.82) and between-person (α = 0.86–0.87) levels. For drinking episode NA, Cronbach’s alphas were fair at the within-episode (α = 0.61) and between-episode levels (0.63) and acceptable at the between-person level (α = 0.77).¹ Correlations were examined between- and within-person for happy with sad (r = −0.40 and r = −0.19, respectively), angry (r = −0.26 and r = −0.14, respectively), and anxious (r = −0.25 and r = −0.22, respectively).¹ The generally acceptable reliability of the PA and NA items suggests that items on the scales were acting as central items, and the low between- and within-person correlations suggests that PA and NA are not simply the inverse of one another (Diener & Emmons, 1984).

Alcohol consumption.

During the morning survey, participants reported their intention to drink (“Do you intend to drink today/tonight?”; Response options: Yes, No), and total number of drinks they had consumed during the prior day, with response options ranging from 0 to 15 (or more, coded at 15). When participants reporting having consumed 1 or more drinks during the prior day via the morning survey, the prior EMA diary day was classified as a drinking day (n = 554 days; 42.0% of diary days). When participants reporting having consumed 0 drinks during the prior day via the morning survey, the prior EMA diary day was classified as a nondrinking day (n = 766 days; 58.0% of diary days). The grand mean number of drinks consumed across all days was 2.75 (SD = 3.65, median = 1.00, mode = 0.00); the grand mean number of drinks consumed across all drinking days was 5.30 (SD = 3.50, median = 5.00, mode = 2.00). As mentioned above, participants also reported on the number of drinks consumed in the past half-hour during the drinking episode survey sequence (see Table 2).

Table 2.

Descriptive Statistics for Momentary Drink and Affect Variables During an Ongoing Drinking Episode^a

	Range (Min – Max)	Grand Mean	Between-Person	Between-Episode	Within-Episode
Momentary Drinks	0 – 15	1.2
Standard Deviation			1.1	0.7	0.7
ICCb			0.12	0.14	0.75
Momentary Positive Affect	0 – 6	4.4
Standard Deviation			1.3	1.0	1.0
ICCb			0.40	0.16	0.44
Momentary Negative Affect	0 – 6	0.5
Standard Deviation			0.8	0.6	0.6
ICCb			0.49	0.13	0.38

^aN=204.

^bICC=Intraclass correlation coefficient. Calculated from the intercept-only models.

Data Preparation

For the purpose of examining research questions 1 and 2, study days were separated into drinking and nondrinking days. Within both drinking and nondrinking days, we divided the day into pre- and post-drinking segments. On drinking days, we used the time a person reported that they started drinking on the given day (i.e., the time of EMA drinking episode initiation) or the first time the SCRAM anklet detected alcohol, using whichever value occurred first temporally (i.e., earlier in time), to divide the day into pre- and post-drinking segments. On drinking days, if both the EMA drinking episode initiation value and SCRAM anklet value were missing, we used the mean time that each person reported they started drinking on drinking days to divide the day into pre- and post-drinking segments. For nondrinking days, we used the mean time that each person reported they started drinking on drinking days, as has been done in previous research (Dvorak et al., 2018; Russell et al., 2020). After dividing days into pre- and post-drinking segments, mean positive and negative affect during pre-drinking segments was calculated using morning, afternoon, and/or evening EMA reports of affect, as appropriate. In cases where participants only had a single pre-drinking affect report, due either to drinking initiation occurring prior to the afternoon report or due to missing data, affect for that single report was used to represent their pre-drinking affect. Using this approach, pre-drinking affect was calculated for all drinking and nondrinking days. If participants were missing pre-drinking affect reports for a given day, that day was not included in the day-level analyses. For example, if a participant’s drinking start time was 1PM and they were missing their morning affect report (the only affect report prior to 1PM), then that day of data was not included in the day-level analyses for research questions 1 & 2. Missing data for morning, afternoon, or evening reports did not impact episode-level analyses, as episode-level analyses relied exclusively on EMA reports that occurred every 30-minutes during drinking episodes. Eleven individuals reported no drinking episodes, prohibiting any within-person examination of pre-drinking affect predicting the odds of drinking on drinking days, as well as any within-episode examinations of associations between affect and drinking. By default, these individuals were excluded from the analyses for research questions 2, 3, and 4.

Statistical Analyses

Two-level multilevel models with days (Level 1) nested within people (Level 2) tested research questions 1 and 2, with separate models estimated for positive and negative affect. For research question 1 (Does pre-drinking positive or negative affect predict the odds of drinking on any given day?), we used multilevel logistic regression models, with average pre-drinking affect as the predictor of the log odds of it being a drinking day (based on the morning EMA report). Pre-drinking affect on a given day was centered on the person-mean affect to test the within-person, day-level association. The effect of day of the week was entered as a covariate into the model by creating a dichotomous variable comparing the effect of social weekends (i.e., Thursday, Friday, Saturday) versus social weekdays (i.e. Sunday, Monday, Tuesday, Wednesday). The effect of intending to drink that day was entered as a covariate into the model as a dichotomous variable by comparing the effect of not intending to drink versus intending to drink that day. Odds ratios (OR) and 95% confidence intervals (CI) were calculated by exponentiating the model estimates and associated CIs extracted from the logistic regression models.

For research question 2 (Does pre-drinking positive or negative affect predict the number of drinks consumed on a drinking day?), we used multilevel Poisson regression models, with pre-drinking affect (person-mean centered) as the predictor of number of drinks consumed on drinking days (based on the morning EMA report). The effects of social weekends versus social weekdays and intending to drink versus not intending to drink were entered as covariates into the model. Incident Rate Ratios (IRR) and 95% CIs were calculated by exponentiating the model estimates and associated CIs extracted from the Poisson regression models.

Three-level multilevel models with moments (Level 1) nested in episodes (Level 2) and episodes nested within people (Level 3) tested research questions 3 and 4, with separate models estimated for positive and negative affect. For research question 3 (Does positive or negative affect predict subsequent number of drinks consumed during an ongoing drinking episode?), we lagged affect one measurement occasion to test whether affect was a predictor of the number of drinks reported over the next half hour. Lagged affect was centered on the mean of the episode, representing the within-episode, within-person effect of lagged affect on drinks. The models also included episode mean affect centered on the participant mean (between-episode, within-person effect) and participant mean affect centered on the sample grand-mean (between-person effect). Momentary lagged drinks, centered on the episode mean, was also entered into the model. Controlling for lagged drinks over the last half hour allowed us to examine whether affect in the last half hour was associated with change in drinks consumed over the same time period. The models examined the effect of time, in hours since the start of the drinking episode, on number of drinks consumed, thereby controlling for the within-episode trajectory of affect and drinking. Random slopes for lagged affect (LaggedAffect_tij) at the episode and person levels were tested using likelihood ratio tests. Standardized betas (β) were calculated using β = b_x*[s_x/s_y], where b_x is the fixed effect of x, s_x is the standard deviation of x, and s_y is the standard deviation of y. Standard deviations of the appropriate level (within-episode, between-episode, between-person) were used to calculate betas.

For research question 4 (Does the number of drinks an individual consumes predict their positive or negative affect during an ongoing drinking episode?), momentary reports of drinks consumed over the past half hour was the predictor of affect (positive or negative). Momentary drinks consumed was centered on the mean of the episode, representing the within-episode, within-person effect of momentary drinks on affect. The models also included episode mean drinks centered on the participant mean (between-episode, within-person effect) and participant mean drinks centered on the sample grand-mean (between-person effect). Momentary lagged affect (positive or negative), centered to the episode-mean, was also entered into the model. Controlling for lagged affect over the last half hour allowed us to examine whether the number of drinks consumed in the last half hour was associated with a change in affect over the same time period. The models examined the effect of time, in hours since the start of the drinking episode, on affect (positive or negative), thereby controlling for the within-episode trajectory of affect and drinking. Random slopes for momentary drinks (Drinks_tij) at the episode and person levels were tested using likelihood ratio tests. Standardized betas were calculated at the within-episode, between-episode, and between-person levels using standard deviations at the appropriate level. We used R version 4.0.1 (R: The R Project for Statistical Computing, 2019) to estimate the models using the nlme package for the linear regressions and the glmer functions of the lme4 package for the logistic and Poisson regressions. Statistical significance was set at p < .05.

Results

Participant Characteristics

The sample consisted of 222 participants. Table 1 summarizes the demographic characteristics for the sample. Table 2 provides the descriptive statistics for the momentary, episode-level variables included in the multi-level models testing research questions 3 and 4, including the grand mean, standard deviations, and intraclass correlations. During the past half hour during drinking episodes, the sample consumed a mean of 1.2 ± 1.1 drinks (median = 1, interquartile range: 0.5, 1.5), had a mean momentary PA of 4.4 ± 1.3, and a mean momentary NA of 0.5 ± 0.8.

Table 1.

Participant Demographics

Demographics	N=222
Age in years (Mean ± SD)	22.3 ± 1.3
Sex (n (%))
Male	81 (36.5)
Female	141 (63.5)
Race, non-Hispanic ethnicity (n (%))
White	175 (78.8)
Asian	15 (6.8)
Black	8 (3.6)
Native American	0 (0.0)
Mixed	8 (3.6)
Race, Hispanic ethnicity (n (%))
White	11 (5.0)
Asian	0 (0.0)
Black	0 (0.0)
Native American	1 (0.5)
Mixed	3 (1.4)
Missing	1 (0.5)
Student Status (n (%))
Undergraduate student	187 (84.2)
Graduate student	14 (6.3)
Non-student	21 (9.5)

Research Question 1: Does pre-drinking positive or negative affect predict the odds of drinking on any given day?

Participants’ average pre-drinking PA was 3.0 ± 1.0 and their average pre-drinking NA was 0.7 ± 0.7. Table 3 includes the model outcomes for the logistic regression models and Poisson regression models examining pre-drinking PA and NA predicting the odds of drinking and the total number of drinks consumed on drinking days, respectively. Each model included two levels, with days nested in participants. As shown in Table 3, pre-drinking PA significantly predicted the odds of drinking, with each one unit increase in PA above the participant mean being associated with a 45% increase in the odds of drinking (OR = 1.45, 95% CI: [1.13, 1.86]). In contrast, pre-drinking NA did not predict increased odds of drinking (OR = 1.05, 95% CI: [0.73, 1.52]). The odds of drinking were higher on weekend days in both models (PA model: OR = 5.21, 95% CI: [3.49, 7.79]; NA model: OR = 5.10, 95% CI: [3.42, 7.59]). The odds were higher on days when participants intended to drink in both models (PA model: OR = 9.45, 95% CI: [6.57, 13.61]; NA model: OR = 9.45, 95% CI: [6.58, 13.57]).

Table 3.

Multi-level Models with Day-level Affect Predicting the Odds of Drinking and Number of Drinks Consumed

	Odds of Drinkinga	Drinks Consumedb
Positive Affect Models	OR [95% CI]	IRR [95% CI]
Model Predictors
Intercept	0.09 [0.06, 0.13]**	2.52 [2.07, 3.06]**
Positive Affect (participant-centered)	1.45 [1.13, 1.86]**	1.06 [0.99, 1.14]
Weekend Dayc	5.21 [3.49, 7.79]**	1.29 [1.07, 1.56]**
Intending to Drinkd	9.45 [6.57, 13.61]**	1.59 [1.36, 1.87]**
	Odds of Drinkinga	Drinks Consumedb
Negative Affect Models	OR [95% CI]	IRR [95% CI]
Model Predictors
Intercept	0.09 [0.06, 0.13]**	2.53 [2.08, 3.07]**
Negative Affect (participant-centered)	1.05 [0.73, 1.52]	0.97 [0.86, 1.08]
Weekend Dayc	5.10 [3.42, 7.59]**	1.28 [1.07, 1.55]**
Intending to Drinkd	9.45 [6.58, 13.57]**	1.60 [1.37, 1.88]**

Notes:

^**p < .01;

^*p < .05. OR=Odds ratio. CI=Confidence Interval. IRR=Incident Rate Ratio.

^aThe logistic regression model included 902 observations nested in 211 participants.

^bThe Poisson model included 376 observations nested in 184 participants.

^cThe effect of social weekend days (Thursday, Friday, Saturday) compared to social weekdays (Sunday, Monday, Tuesday, Wednesday).

^dThe effect of the participant intending to drink that day.

Research Question 2: Does pre-drinking positive or negative affect predict the number of drinks consumed on a drinking day?

As shown in Table 3, pre-drinking PA and NA did not predict the number of drinks consumed on drinking days (PA: IRR = 1.06, 95% CI: [0.99, 1.14]; NA: IRR = 0.97, 95% CI: [0.86, 1.08]). The number of drinks consumed on drinking days was higher on weekends in both models (PA model: IRR = 1.29, 95% CI: [1.07, 1.56]; NA model: IRR = 1.28, 95% CI: [1.07, 1.88]). The number of drinks consumed on drinking days was higher when participants intended to drink in both models (PA model: OR = 1.59, 95% CI: [1.36, 1.87]; NA model: OR = 1.60, 95% CI: [1.37, 1.88]).

Research Question 3: Does positive or negative affect predict subsequent number of drinks consumed during an ongoing drinking episode?

Positive affect

Table 4 includes the model outcomes for lagged PA or NA predicting drinks consumed over next half hour. Likelihood ratio tests indicated that including random slopes for momentary lagged positive affect (LaggedAffect_tij) at the episode or person levels did not significantly improve model fit; therefore, the final model for PA did not include random slopes for momentary lagged affect at the episode or person levels. Table 4 shows no significant within-episode/within-person effect of lagged PA, such that momentary lagged PA did not predict drinks consumed over the next half hour (b = −0.02, β = −0.02, p = .36). Time (hours elapsed since the start of the drinking episode) did not predict drinks consumed over the next half hour (b = −0.01, β = −0.04, p = .07). ^2,3

Table 4.

Multi-level Models with Momentary Affect Predicting Subsequent Number of Drinks Consumed during a Drinking Episode

	Drinks Modelsa
Positive Affect - Model Parameters	b	β
Model Predictors
Intercept	1.15**
Lagged Momentary Positive Affect (episode-centered)b	−0.007	−0.01
Episode Mean Positive Affect (participant-centered)c	0.08	0.04
Participant Mean Positive Affect (sample-centered)d	0.02	0.02
Lagged Drinks (episode-centered)e	0.22**	0.20
Timef	−0.01	−0.04
Negative Affect - Model Parameters	b	β
Model Predictors
Intercept	1.15**
Lagged Momentary Negative Affect (episode-centered)b	−0.12*	−0.06
Episode Mean Negative Affect (participant-centered)c	−0.04	−0.001
Participant Mean Negative Affect (sample-centered)d	−0.10	−0.06
Lagged Drinks (episode-centered)e	0.21**	0.20
Timef	−0.01	−0.04

Notes:

^**p < .01;

^*p < .05. b=Unstandardized beta. β=Standardized Beta.

^aThe models included 2507 observations nested in 497 episodes, which were nested in 194 participants. The positive affect model had no random effects of lagged positive affect. The negative affect model included random effects of lagged negative affect at the episode and participant levels.

^bThis represents the within-episode/within-person effect of lagged affect on drinks consumed.

^cThis represents the between-episode/within-person effect of affect on drinks consumed.

^dThis represents the between-person effect of affect on drinks consumed.

^eThis represents the within-episode/within-person effect of momentary lagged drinks on current drinks.

^fTime represents hours elapsed since the start of the drinking episode.

Negative affect

Likelihood ratio tests indicated that including random slopes for momentary lagged negative affect at the episode and person levels significantly improved model fit; therefore, the final model for NA included a random slope for momentary lagged affect (LaggedAffect_tij) centered to the episode mean at the episode and person levels. As shown in Table 4, there was a significant within-episode/within-person effect of lagged NA, such that each unit increase in momentary NA above the episode mean predicted significantly fewer drinks consumed over the next half hour (b = −0.14, β = −0.06, p = .007). Time (hours elapsed since the start of the drinking episode) did not predict drinks consumed over the next half hour (b = −0.01, β = −0.04, p = .08). ^2,3,4

Research Question 4: Does the number of drinks an individual consumes predict their positive or negative affect during an ongoing drinking episode?

Positive affect

Table 5 includes the model outcomes for drinks predicting PA and NA over the next half hour, including unstandardized and standardized betas. The final models for PA and NA included a random effect of momentary drinks (Drinks_tij) centered to the episode mean at the episode and participant levels. As shown in Table 5, there was a significant within-episode/within-person effect of momentary drinks on PA, such that each one unit increase in momentary drinks consumed above the episode mean predicted significantly greater PA over the next half hour (b = 0.16, β = 0.18, p < .001). Time (hours elapsed since the start of the drinking episode) did not predict PA over the next half hour (b = 0.002, β = 0.008, p = .81). ²

Table 5.

Multi-level Models with Momentary Drinks Predicting Subsequent Positive and Negative Affect During a Drinking Episode

	Positive Affecta		Negative Affectb
Model Parameters	b	β	b	β
Model Predictors
Intercept	4.26**		0.47**
Momentary Drinks (episode-centered)c	0.16**	0.18	−0.007	−0.01
Episode Mean Drinks (participant-centered)d	0.18*	0.11	−0.03	−0.03
Participant Mean Drinks (sample-centered)e	0.13	0.08	−0.10	−0.13
Lagged Positive Affect (episode-centered)f	0.14**	0.13	-	-
Lagged Negative Affect (episode-centered)f	-	-	0.16**	0.15
Timeg	0.002	0.008	0.01*	0.08

Notes:

^**p < .01;

^*p < .05. b=Unstandardized beta. β=Standardized beta.

^aThe model included 2488 observations nested in 496 episodes, which were nested in 194 participants. A random effect for momentary drinks was included at the episode and participant levels.

^bThe model included 2487 observations nested in 496 episodes, which were nested in 194 participants. A random effect for momentary drinks was included at the episode and participant levels.

^cThis represents the within-episode/within-person effect of momentary drinks on affect.

^dThis represents the between-episode/within-person effect of drinks on affect.

^eThis represents the between-person effect of drinks on affect.

^fThis represents the within-episode/within-person effect of momentary lagged affect on current affect.

^gTime represents the number of hours elapsed since the start of the drinking episode.

Negative affect

As shown in Table 5, there was no within-episode/within-person effect of momentary drinks on NA, such that momentary drinks consumed did not predict NA over the next half hour (b = −0.007, β = −0.01, p = .64). Time (hours elapsed since the start of the drinking episode) predicted greater NA over the next half hour (b = 0.01, β = 0.08, p = .01). ^2,5

Post-Hoc Analyses: Does positive or negative affect predict the odds of it being the end of a drinking episode.

Based on the suggestions of anonymous reviewers, we conducted post-hoc analyses using logistic regression models to examine PA and NA predicting the end of a drinking episode. An observation was coded as the end of a drinking episode if there was no temporally subsequent drinking episode prompt within the next three hours. Each model includes episode mean-centered momentary affect, participant mean-centered episode affect, sample mean-centered participant affect, and time as covariates, with episodes nested in participants. As shown in Supplemental Table 1, each one unit increase in PA above the episode mean was associated with a 32% decrease in the odds of it being the end of a drinking episode (OR = 0.68, 95% CI: [0.58, 0.78]). In contrast, each one unit increase in NA above the episode mean was associated with a 39% increase in the odds of it being the end of a drinking episode (OR = 1.39, 95% CI: [1.08, 1.80)). Time (hours elapsed since the start of the drinking episode) predicted increased odds of it being the end of a drinking episode in both models (PA model: OR = 1.14, 95% CI: [1.08, 1.19]; NA model: OR = 1.12, 95% CI: [1.07, 1.18]). ^6,7

Discussion

This study tested affect regulation models of alcohol use within days and drinking episodes using EMA in the everyday lives of young adults (predominantly college students). Models testing pre-drinking affect as a predictor of drinking showed that greater pre-drinking PA predicted greater odds of drinking on a given day, but did not predict the number of drinks consumed on drinking days; whereas, pre-drinking NA predicted neither the odds of drinking on a given day nor the number of drinks consumed on drinking days. Models testing affect-alcohol use associations within ongoing drinking episodes showed that current PA did not predict continued drinking over the next half hour, but PA did predict lower odds of a drinking episode ending, and half-hour periods of higher drinking predicted increases in PA. In contrast, having higher than average NA at a given moment predicted consuming fewer drinks over the next half hour and predicted higher odds of a drinking episode ending, but the number of drinks consumed in the last half hour did not predict NA. Importantly, time, expressed as the number of hours elapsed since the start of the drinking episode, did not affect outcomes, suggesting that affect-alcohol use associations within the context of an ongoing drinking episode are not impacted by the length of the drinking episode. This study is unique in examining both tenets of the affect regulation model of alcohol use within ongoing drinking episodes at the high temporal resolution of every half hour and examining affect predicting the end of a drinking episode, while simultaneously examining whether pre-drinking affect predicts same day drinking behavior, thus providing temporally relevant information about the affect regulation model of alcohol use.

Our findings contribute to the literature testing whether affect predicts drinking behavior by examining whether pre-drinking affect predicts the odds of drinking or number of drinks consumed on the same day, allowing us to infer whether pre-drinking affect can serve as a signal of same day drinking behaviors. In this sample of young adults, we found that greater pre-drinking PA predicted greater odds of drinking on the same day, even after controlling for an individual’s intention to drink. This supports the first tenet of affect regulation models – that PA motivates same day drinking behavior. These results correspond with previous research in which pre-drinking PA was higher on drinking versus non-drinking days (Dvorak et al., 2018; Russell et al., 2020), as well as with studies finding positive associations between daytime PA and same day drinking behaviors (Dvorak & Simons, 2014; Mohr et al., 2008; Simons et al., 2005, 2010, 2014; Swendsen et al., 2000). Overall, ours and other studies support that pre-drinking PA appears to be a reliable signal for the initiation of same day drinking. This has important implications for interventions, suggesting that interventions could use a single or average measure of pre-drinking PA as a signal for same day drinking initiation. Such interventions could aim to prompt additional questions during momentary assessments when an individual’s pre-drinking PA is higher than usual. These assessments could help researchers better understand the social, emotional, or other contextual variables that may help account for higher than usual PA within an individual. Such information could inform subsequent just-in-time adaptive interventions prior to drinking that are aimed at preventing drinking onset or providing contingency management strategies to reduce negative consequences of drinking, particularly in high risk populations like heavy drinking college students, who constituted the majority of our sample.

In contrast to our PA findings, pre-drinking NA was not associated with the odds of drinking or number of drinks consumed, corresponding with previous studies finding no differences in pre-drinking anxiety, anger, or sadness on drinking versus non-drinking days (Dvorak et al. 2018) and no associations between daytime depressed or anxious mood and the odds of drinking (Grant 2009, Dvorak & Simons 2014). However, some previous studies found positive associations between daytime NA and drinking behavior (Grant et al., 2009; Mohr et al., 2005; Simons et al., 2005, 2014, Dvorak 2014). These contrasting findings could be due to the fact that, with the exception of Simons et al. 2014, the aforementioned studies could not temporally distinguish pre- and post-drinking NA, and did not compare drinking to non-drinking days, introducing the chance that study findings were due to reverse causality, such that greater daytime NA was a result rather than a cause of drinking. These mixed findings suggest that drinking in young adult students and non-students appears to be less strongly predicted by pre-drinking NA versus PA. However, it is possible that NA predicts drinking initiation among some subsets of individuals. For example our participants (who were not selected based on affective vulnerability), reported relatively low NA, but NA may predict drinking initiation among individuals with higher levels of NA or who experience high levels of affective variability (a form of emotional dysregulation), as previous studies suggest that greater variability in NA is associated with greater odds of drinking and higher levels of alcohol consumption (Gottfredson & Hussong, 2013; Mohr et al., 2015; Shadur et al., 2015). Studies also indicate that drinking motives, such as drinking for coping, enhancing, or social purposes, moderate the association between NA and drinking (Armeli et al., 2008, 2010; Dvorak et al., 2014; Grant et al., 2009; Hussong et al., 2005; Mohr et al., 2005; Todd et al., 2005, 2009). Additionally, social support (Hussong et al., 2001), negative and positive urgency (Simons et al., 2010; Wray et al., 2012), and executive functioning (Dvorak & Simons, 2014) may moderate the associations between NA and drinking. Therefore, while it is possible that NA may be an important predictor of drinking among some subsets of individuals, interventionists interested in examining pre-drinking affect as a signal for same day drinking behaviors may benefit from focusing on pre-drinking PA as a more consistent indicator of subsequent drinking behavior among the majority of non-clinical, heavy drinking young adults.

Interestingly, our findings regarding whether affect predicts drinking were reversed when examining those associations within ongoing drinking episodes. Specifically, PA did not predict the number of drinks consumed over the next half hour, though post-hoc analyses indicated that PA predicted lower odds of it being the end of a drinking episode.⁶ In contrast, experiencing greater NA predicted consuming fewer drinks over the next half hour and predicted higher odds of it being the end of a drinking episode.⁶ In combination with our day-level results, these findings suggest that NA may be associated with decreased drink consumption, as well as the end of a drinking episode, within the context of an ongoing drinking episode. Given that previous studies have not specifically examined momentary affect-drinking associations within the context of an ongoing drinking episode (after which an individual has already started drinking) we cannot directly compare our findings to previous research. However, it is possible that the contextual effect of NA on drinking within a drinking episode could help account for mixed findings in previous literature, with researchers finding positive (Dvorak et al., 2014; Grant et al., 2009; Mohr et al., 2005; Simons et al., 2005, 2014; Swendsen et al., 2000), negative (Grant et al., 2009; Simons et al., 2010; Swendsen et al., 2000), or null (Dvorak & Simons, 2014; Grant et al., 2009) associations between daytime NA and same day drinking behaviors. As previously mentioned, this NA-drinking association may also be moderated by variables like drinking for coping or social reasons (Armeli et al., 2008, 2010; Dvorak et al., 2014; Grant et al., 2009; Hussong et al., 2005; Mohr et al., 2005; Todd et al., 2005, 2009), the presence of friends while drinking (Mohr et al., 2005), or negative and positive urgency (Simons et al., 2010; Wray et al., 2012). This novel, unexpected finding suggests that NA may govern the amount of continued drinking during an ongoing drinking episode and the end of a drinking episode. It remains unclear why feeling good (PA) may govern drinking initiation, whereas feeling bad (NA) may govern decreased alcohol consumption or drinking termination, such that the worse individuals feel, the less they drink and the more likely they are to stop drinking. This novel finding suggests the need for additional research investigating the potential effects of individual and contextual characteristics on NA governing the amount of alcohol consumed and drinking termination during ongoing drinking episodes. Overall, our results examining affect predicting drinking imply that the temporality and directionality underlying the first tenet of affect regulation models, in which affect motivates drinking, may vary by type of affect and the time frame examined, with PA predicting drinking initiation over the longer timeframe of a day, and NA predicting less alcohol consumption over the shorter timeframe of an ongoing drinking episode.

We also found that consuming a higher than average number of drinks predicted increased PA during an ongoing drinking episode. These results correspond with previous research finding positive associations between drinking and subsequent PA across various time frames (Crooke et al., 2013; Dvorak et al., 2018; Russell et al., 2020; Treloar et al., 2015), and expand said findings within the unique context of an ongoing drinking episode. Our finding that drinking enhanced PA during an ongoing drinking episode supports the affect enhancing effects of drinking (i.e., the second tenet of affect regulation models) and suggests that drinking can enhance PA over a short timeframe, consequently providing immediate positive reinforcement for drinking as a means to enhance affect. When considering these results in combination with the post-hoc analyses indicating that PA predicted lower odds of it being the end of a drinking episode, this suggests a reinforcement cycle between drinking and PA. Specifically, drinking predicts greater PA, which then predicts higher odds of the drinking episode continuing, which permits continued drinking to enhance PA. However, given the finding that PA did not predict consuming more drinks during an episode, it is possible that other factors, such as social interactions, may play a role in PA prolonging a drinking episode. Future research should explore how drinking context interacts with affect in predicting the continuation of drinking episodes.

In contrast to our PA affect findings, drinks consumed did not change NA within an ongoing drinking episode, though there was a small effect of time on increasing NA during a drinking episode (β = 0.08). These findings contrast with previous studies in which NA decreased after the first drink or during the first hour post-drinking (Russell et al., 2020; Treloar et al., 2015), during the same day (Gorka et al., 2017), or over several days post-drinking (Hussong et al., 2001); however, they do correspond with one study finding that NA did not change over three days post-drinking (Crooke et al., 2013). The differences between ours and previous studies could be the timeframe over which we examined drinking effects on NA – every 30 minutes during an ongoing drinking episode – which is a shorter time frame than Treloar et al.’s (2015) and Russell et al.’s (2020) analyses of a study employing a comparable design. This suggests that the effects of drinking on NA may occur outside of the boundary of an ongoing drinking episode, effects that we would not have captured given our study design (Piasecki, 2019). In general, our results regarding drinking effects on NA and PA suggest that drinking immediately enhances PA but does not change NA within an ongoing drinking episode, thus providing partial support for the second tenet of the affect regulation model of alcohol use. Given the mixed literature regarding drinking effects on NA, additional research is warranted.

There were several strengths of our study. For example, our use of EMA enhanced the ecological validity of reports of positive and negative affect during drinking episodes, and our examination of episode- and day-level associations between affect and drinking allowed discovery of the nuanced temporality and directionality underlying affect-alcohol use associations, as well as an examination of both tenets of the affect regulation model. In general, our results support affect as a predictor of drinking, with PA predicting drinking initiation on the same day and NA predicting fewer drinks consumed during an ongoing drinking episode and drinking termination. The results also support the affect enhancing role of drinking on PA during an ongoing drinking episode and suggest a reinforcement cycle between drinking and PA. The knowledge gained from this study may be useful in guiding the development of drinking interventions for heavy drinking young adults or college students. For example, day-level pre-drinking PA could be used a signal for when to trigger just-in-time interventions prior to drinking onset to reduce alcohol consumption or provide contingency management strategies to reduce negative consequences of drinking.

As with all studies, our findings should be interpreted with appropriate consideration of limitations related to the study sample and methods. This sample consisted of non-clinical, heavy drinking young adults, the majority of whom were college students, and it is possible that associations between affect and drinking are different in other age groups, or at lower levels of alcohol consumption. Participants also reported low levels of NA, and results may differ among individuals who experience higher levels of NA. Another consideration regards using a PANAS-based measure of affect, which assumes that PA and NA dimensions are orthogonal. We also operationalized PA using a single item (happy); whereas, we operationalized NA via three items (sad, angry, anxious), items which differed from one another regarding arousal state (low arousal for sad, high arousal for angry and anxious). While results of our sensitivity analyses (see footnotes) suggest that using more items to operationalize PA and NA did not change our findings, we recognize that this measurement approach reflects one perspective on affect. It is possible that other models, such as the affective circumplex model (Posner et al., 2005), may yield different insights.

Conclusion

Affect regulation models of drinking suggest that affect motivates drinking behavior and that drinking is reinforced through the affect enhancing or diminishing effects of alcohol use. Our results are consistent with the affect regulation model of alcohol use, but add temporal and directional nuance, suggesting that (a) pre-drinking PA predicts same day drinking behaviors, (b) NA predicts drinking behaviors during drinking episodes, and (c) alcohol use immediately enhances PA during drinking episodes. These findings suggest that pre-drinking PA appears to signal the initiation of drinking later in the day; whereas, NA may govern decreased drinking during an ongoing drinking episode and the termination of drinking episodes, information which might aid in developing interventions to reduce drinking or drinking-related consequences in young adults or college students.

Public Health Significance:

This study indicates that low negative affect may foreshadow continued drink consumption during a drinking episode, and that drinking enhances positive affect during the next half hour within a single drinking episode. Additionally, this study highlights that pre-drinking positive affect may be a signal for drinking onset later the same day, which could inform the development of interventions targeting drinking in young adults.