Alcohol Demand as a Predictor of Drinking Behavior in the Natural Environment

Abstract

Background:

Alcohol demand, a measure of alcohol’s reinforcing value, is associated with greater alcohol consumption and alcohol-related problems. Although alcohol demand has primarily been evaluated as a “trait-like,” individual difference measure, recent evidence indicates that demand exhibits meaningful short-term fluctuations.

Aims:

To test whether moment-to-moment fluctuations in alcohol demand in individuals’ natural drinking environments predicted drinking occurrence, drinking continuation, and drinking quantity.

Design:

Baseline laboratory or Zoom session and 6 weeks of ecological momentary assessment (EMA).

Setting:

Individuals’ natural drinking environments in Columbia, MO.

Participants:

89 young adults (56% female; M age=24.8) participated from November 2018 through October 2020. Participants reported 14.5 drinking days (SD=8.1) and 4.1 drinks per occasion (SD=2.5) during EMA.

Measurements:

Participants completed the Alcohol Purchase Task at baseline. Following, participants reported on their alcohol demand (breakpoint, O_max, intensity) and drinking behavior during EMA at daily, timed prompts from 6:00pm to 2:00am. They provided breathalyzer samples using a BACtrack Mobile^® Pro. Models tested concurrent and prospective (lagged) associations between alcohol demand and drinking occurrence and drinking continuation after drinking initiation. Additional models tested concurrent associations between demand and breath alcohol concentrations (BrACs).

Findings:

Higher alcohol demand was associated with higher odds of drinking and continued drinking for all demand indices at the momentary (ORs=1.27–1.56, ps≤.03) and day-level (ORs=2.14–3.39, ps<.001). Additionally, lagged demand predicted higher odds of drinking occurrence and continuation at the following prompt (ORs=1.32–1.53, ps≤.004). Higher alcohol demand was associated with higher BrACs at the momentary (bs=0.0011–0.0026, ps≤.03) and day-level (bs=0.0053–0.0062, ps<.001). At the person-level, findings varied depending on the demand measure.

Conclusions:

Alcohol demand is associated with both when and how much individuals drink in their natural drinking environments. Elevations in alcohol demand are associated with increased likelihood of drinking and continuing to drink, and greater total alcohol consumption, both within and across drinking days.

INTRODUCTION

The reinforcing value of a drug is thought to contribute to its use and potential misuse (1). As addiction develops, a drug acquires greater reinforcement value relative to both earlier experiences with the drug and non-drug rewards (e.g., food, sex, pleasurable activities). This “hallmark” shift in drug reinforcement value is captured in several DSM-5 substance use disorder diagnostic criteria (e.g., giving up important social, occupational, or recreational activities because of substance use) (2). Drug reinforcement value can therefore provide insight into processes that motivate substance use and lead to problematic use and/or addiction (3).

Behavioral economic demand quantifies the relative reinforcing value of drugs by comparing a drug’s valuation to non-drug rewards such as money (4). For example, alcohol’s relative reinforcing value, termed alcohol demand (5,6), is often assessed using an “alcohol purchase task”, where participants indicate hypothetical quantities of alcohol they would consume across a range of prices. Demand tasks provide an analogue to operant reinforcement tasks (7) without necessitating consumption, allowing for administration in settings in which consumption may raise ethical concerns, such as treatment settings (8,9). Individual differences in alcohol demand are associated with drinking quantity, frequency of heavy drinking, and alcohol-related problems (6,10–13). Elevated demand may also serve as a prognostic indicator of treatment response, for example, predicting worse treatment outcomes for a brief alcohol intervention (13). Individual differences in demand may therefore capture an important, “trait-like” index of drug reinforcement value, indicative of substance use severity and predictive of treatment response.

Demand is also susceptible to state-dependent changes in response to environmental contingencies that alter drug reinforcement value (14). For instance, affective states that acutely increase (e.g., abstinence, craving) or decrease (e.g., drug satiation) a drug’s reinforcement value should reflect corresponding changes in demand (4,15), and in turn, drug consumption. Research on changes in demand has primarily involved laboratory experiments and interventions. A recent meta-analysis demonstrated that drug cues, increased reinforcer magnitude (i.e., larger drug doses), and stress/negative affect induction increased alcohol demand, whereas pharmacotherapies and behavioral interventions reduced alcohol demand (4). These findings indicate that demand is sensitive to fluctuations as a result of state-based, motivational influences.

Few studies have examined within-person, moment-to-moment fluctuations in demand. Two recent laboratory studies examined within-person fluctuations in demand under a moderate dose of alcohol (target breath alcohol concentration [BrAC]=.10 g%). Amlung and colleagues found that both alcohol demand and craving increased as BrAC ascended and decreased as BrAC descended (16). In a follow-up study, Motschman and colleagues tested whether alcohol-induced craving played a role in enhancing alcohol’s reinforcement value, and found that momentary increases in craving predicted elevated alcohol demand, as did increases in alcohol’s stimulating effects (15). Although these studies implicate demand as a promising indicator of drinking motivation, these studies did not assess drinking intentions or behavior directly. The laboratory setting is also limited by a lack of contextual factors that may alter drug reinforcement magnitude and its motivational significance.

The present study is the first investigation of dynamic changes in alcohol demand and their prediction of drinking outside the laboratory. Studying demand in natural drinking environments provides two important advantages over prior research. First, alcohol’s reinforcement value may depend, in part, on elements of the drinking environment (e.g., a stimulating bar setting) or drinking experience (e.g., social reinforcers that augment alcohol’s valuation). The natural environment setting allows for authentic variation in environmental contingencies that may influence drinking motivation. Second, the natural environment setting allows for unconstrained alcohol consumption, and thus for direct tests of demand’s ability to predict subsequent consumption. Previous studies support correspondence between demand for hypothetical and actual alcohol rewards (17,18), but we are not aware of any studies to date that have assessed momentary fluctuations in alcohol demand in the natural environment, nor its prediction of drinking behavior (demand has been examined at the day-level as an outcome variable; (19)).

The present study was a 6-week ecological momentary assessment (EMA) investigation of individuals who drank moderate-to-heavy amounts. The primary aim was to examine alcohol demand as a predictor of three aspects of drinking behavior: 1) engaging in drinking at all (occurrence), 2) continued drinking once alcohol was on board (continuation), and 3) the amount of alcohol consumed (consumption), providing insight into different motivational mechanisms. To fully disentangle alcohol demand’s motivational influence on drinking behavior, we examined fluctuations in demand within a given day and drinking episode, in addition to day-to-day fluctuations and individual differences in demand. We also tested alcohol demand as a prospective (lagged) predictor of drinking behavior when feasible. Based on the cumulative research suggesting that alcohol demand is a relevant index of alcohol use motivation, tracks with fluctuations in craving in the laboratory, and can be acutely increased under certain manipulations (e.g., drug cues, stress), we hypothesized that increases in alcohol demand would positively predict drinking behavior in the natural environment. A secondary aim was to examine associations between alcohol demand indices collected over repeated EMA assessments and their associations with traditional, “trait-like” indices of alcohol demand (APT) to assess construct overlap (6). We considered these analyses exploratory because no prior studies have examined momentary fluctuations in alcohol demand in EMA.

An important and novel aspect of this study was the use of mobile breathalyzers as a biological measure of alcohol consumption. Traditional EMA studies rely on self-reported drink quantity and alcoholic content to generate estimated BACs (20), which can be fallible (21). Alcohol absorption, distribution, and elimination, and factors that influence these processes (e.g., stomach content, beverages mixers) are also unaccounted for and may produce problematic estimates. Mobile breathalyzers circumvent these issues, providing a more precise quantitative measure of alcohol consumption.

METHODS

Participants

Participants were recruited from a large, Midwestern university and surrounding community through fliers and email advertisements. Eligibility criteria included: consuming alcohol at least 2 days per week and at least 1 binge drinking occasion within the past 6 months (≥4/5 standard drinks [12 oz. beer, 5 oz. wine, or 1.5 oz. liquor] within 2 hours for women/men), living ≥2 miles from one’s typical drinking locations, having a body mass index ≥18 and <35, not pregnant/nursing, not using medications contraindicated to alcohol consumption, no history of treatment for psychiatric or substance use disorders, and no alcohol-related medical conditions. Of the 109 participants who completed the initial session, 89 completed the 6-week EMA protocol (81.6% completion rate)¹ and were included in analyses (Table 1).

Table 1.

Demographics and Drinking Characteristics

Characteristic	M(SD), %, or Mdn
Age	24.8 (3.8)
Female	55.7%
Race/ethnicitya
Native American/Alaska Native	1.1%
Asian	6.7%
Black	10.1%
Latino/Hispanic	2.2%
Middle Eastern/North African	1.1%
Pacific Islander/Native Hawaiian	1.1%
White	87.6%
≥ High school education	100.0%
Current student (part/full-time)	66.3%
Household income	$25,001 - $35,000
Alcohol use in past 30 days
Frequency of drinking days	2x/week
Drinks per day on drinking days	3–4 drinks
Binge drinking days	2–3 days

^aPercentages total greater than 100% because some participants endorsed multiple races/ethnicities.

Measures

Retrospective drinking behavior

Participants reported on the number of drinking days, number of binge drinking days, and typical drink quantity consumed per day using items from the National Institute on Alcohol Abuse and Alcoholism Task Force on Recommended Alcohol Questions (22).

Alcohol Purchase Task (APT)

The Alcohol Purchase Task (6) was used as a conventional measure of trait-level alcohol demand. Participants reported how many standard alcoholic drinks they would consume at drink prices ranging from free to $30 (USD) in a hypothetical, typical drinking scenario. Participants were instructed they would need to consume all drinks purchased, could not save/stockpile drinks nor consume alcohol elsewhere afterward. Three observed alcohol demand indices were examined: breakpoint, the first price at which consumption reduced to zero, O_max, the maximum alcohol expenditure, and intensity, the number of drinks consumed when free. Breakpoint was recorded as the maximum price ($30) when participants did not have a price at which consumption reduced to zero (12.4% of sample).

EMA alcohol demand

As in previous studies (15,16,23), state-dependent alcohol demand was assessed with three items that paralleled observed demand indices from the trait APT. However, participants answered based on how they felt “right now.” Alcohol demand indices included: breakpoint, the maximum amount participants would pay for a single drink, ranging $0 to $30 in $2 increments, O_max, the maximum total amount participants would spend on drinking, ranging $0 to $60 in $5 increments, and intensity, the number of drinks consumed when free, ranging 0 to 20 in 1-drink increments.

EMA drinking behavior

Participants reported whether they had consumed alcohol (yes/no) since the previous evening survey prompt. Drinking occurrence was defined as either endorsement of drinking (“yes”) or a positive BrAC reading. When drinking was endorsed, participants also reported on the total number of drinks consumed up to that point.

Breath alcohol concentration (BrAC)

BrAC was assessed using a BACtrack Mobile Pro portable breathalyzer (24), which has good agreement with standard breathalyzers (bias=+.008 g/dl, 95% CI=.0062, .0096) (25). BrAC readings were transmitted as data within the EMA survey app and not displayed to participants during the study.

Procedure

Study overview

This study was part of an ongoing clinical trial to reduce alcohol-impaired driving behavior in young adults (see ClinicalTrials.gov ID #NCT03846050 for full details). Participants were randomized to one of three intervention conditions. Participants in the intervention groups differed in the onset of EMA (immediate/delayed) and completed an EMA protocol that included questions about alcohol-impaired driving intentions/behaviors. These groups received the EMA breathalyzer intervention, which provided feedback that it was unsafe to drive when BrAC was ≥.050g%. This message was contingent on BrAC readings ≥.050g%, but participants were unaware of this cutoff or their actual BrACs and were only told that it was unsafe to drive. Participants in the control condition did not receive breathalyzer feedback nor answer alcohol-impaired driving questions during EMA but received an equivalent length EMA protocol. Intervention condition was included as a covariate in all analyses.

Baseline session

Participants were recruited from November 2018 through October 2020. Participants recruited prior to COVID-related restrictions (44.9%) attended a laboratory alcohol administration session. Participants recruited post-COVID (55.1%) attended their baseline session via Zoom. Participants provided informed consent and completed baseline computerized questionnaires. (Pre-COVID alcohol administration data are not included in the present study.)

EMA protocol

Experimenters oriented participants to the protocol prior to their scheduled EMA period. Participants were issued a portable breathalyzer and iPhone 5se with the TigerAware survey application or downloaded it onto their personal iPhone (26). Experimenters provided instructions on responding to EMA survey prompts and using the portable breathalyzers (e.g., wait 15 min after drinking and rinse mouth with water prior to providing samples).

Participants completed 6 continuous weeks of EMA. They were instructed to complete morning reports daily upon waking and evening reports 4 or 5 times daily when prompted at scheduled times: 6:00pm, 8:00pm, 10:00pm, 12:00am, and 2:00am (2:00am assessment added post-COVID). Participants were able to initiate drinking reports at other times as needed. Alcohol demand, self-reported endorsement of drinking, and drinking behaviors (e.g., quantity, location, context) were collected at each prompt prior to participants providing BrAC samples. Participants provided BrAC samples at all assessments. After completing the study, participants were debriefed. Participants were compensated $15/hour for the baseline session, $60/week for EMA (full payment for ≥80% compliance), and $15 per interview for completing follow-up assessments. All procedures were approved by the university IRB.

RESULTS

EMA compliance and descriptives

The total possible number of surveys for each participant was n=41 for morning surveys and n=168 (pre-COVID) or n=210 (post-COVID) for evening surveys. Average compliance was 90.1% (SD=13.3%) for morning surveys and 57.6% (SD=16.9%) for evening surveys (9015 total evening surveys out of a possible 15,341 completed across all participants). The evening survey compliance rate is based on a relatively conservative calculation (i.e., potentially counting surveys after participants went to sleep). The fuller set of compliance calculations is available in the Supplemental Materials. On average, participants reported 14.5 drinking days (SD=8.1; range=1 to 33), consumed 4.1 standard drinks per occasion (SD=2.5; range=1.2 to 12.2), and had average BrACs of .036 g% on drinking days (SD=.026; range=.001 to .120) during the EMA period. Momentary BrACs were positively correlated with self-reported total drink counts, r=0.52, p<.001. Demand indices for EMA and the APT are shown in Table 2.

Table 2.

Alcohol Demand Indices in Ecological Momentary Assessment (EMA) and Alcohol Purchase Task (APT)

Alcohol Demand Index	EMA				APT
	All Moments		Drinking Moments
	M(SD)	Range	M(SD)	Range	M(SD)	Range
Breakpoint ($)	2.74 (4.16)	0 – 30	3.70 (4.76)	0 – 30	16.60 (7.14)	5 – 30
Omax ($)	6.64 (10.06)	0 – 60	9.13 (11.82)	0 – 60	22.58 (10.51)	8 – 64
Intensity (# drinks)	2.37 (3.58)	0 – 20	3.35 (4.23)	0 – 20	7.47 (3.94)	2 – 20

Note. EMA demand measures reflect Ms, SDs, and ranges across all moments and participants. APT measures reflect individual differences on the baseline APT.

Data analytic overview

Multilevel models and generalized linear mixed models tested alcohol demand indices as predictors of drinking behavior. Momentary observations were nested within days, and days nested within participants. PROC GLIMMIX Laplace estimation with between-within df in SAS 9.4 was used to predict drinking occurrence and continued drinking (yes/no). PROC MIXED robust maximum likelihood estimation with Satterthwaite df was used to predict BrAC during drinking episodes. Models included random intercepts at both levels and random slopes for day-level demand. Within-day random slopes either could not be estimated due to too little variability or did not account for significant variance. To disaggregate effects across levels of analysis, momentary and day-level demand predictors were centered within each day and person, respectively, and person means were centered on their respective sample averages (27). We included hours since the first evening report to account for an increased likelihood of drinking and achieving higher BrACs later in the evening. Additional covariates included biological sex (female=reference), intervention condition (control=reference), and COVID group (pre-COVID=reference).

Drinking occurrence

We used all evening reports (N=8352 observations included in fixed effects in initial non-lagged model) to examine associations between alcohol demand and drinking occurrence, of which 5070 observations had lagged demand data (final model). As hypothesized, higher momentary demand was associated with higher odds of drinking for all demand indices (Table 3). Additionally, higher lagged momentary demand predicted higher odds of drinking at the following report. At the day-level, higher demand on a given day was associated with increased odds of drinking. Drinking was more likely to occur later into the evening. Participants recruited post-COVID reported more drinking versus non-drinking moments, as did participants in the late-onset intervention condition, but in breakpoint and O_max models only. Findings did not conclusively indicate associations between person-level demand or sex and odds of drinking occurrence.

Table 3.

Associations between Alcohol Demand and Drinking Occurrence

Effect	Breakpoint			Omax			Intensity
	b	OR (95% CI)	p	b	OR (95% CI)	p	b	OR (95% CI)	p
Intercept	−3.38			−3.40			−3.25
Lagged demand	0.36	1.43 (1.23, 1.65)	<.001	0.43	1.53 (1.30, 1.81)	<.001	0.30	1.35 (1.20, 1.51)	<.001
Momentary demand	0.33	1.39 (1.19, 1.64)	<.001	0.37	1.45 (1.23, 1.70)	<.001	0.27	1.31 (1.15, 1.48)	<.001
Day-level demand	1.20	3.33 (2.44, 4.55)	<.001	1.22	3.39 (2.52, 4.55)	<.001	1.19	3.28 (2.52, 4.25)	<.001
Person-level demand	−0.01	0.99 (0.79, 1.24)	.96	−0.01	1.00 (0.79, 1.26)	.96	0.02	1.02 (0.91, 1.14)	.77
Covariates
Hours after first report	0.15	1.16 (1.04, 1.29)	.008	0.16	1.18 (1.06, 1.31)	.003	0.16	1.18 (1.06, 1.31)	.002
Sexa	0.23	1.26 (0.65, 2.44)	.50	0.19	1.20 (0.61, 2.36)	.59	0.15	1.16 (0.58, 2.33)	.67
Early-onset interventionb	0.73	2.07 (0.88, 4.90)	.10	0.70	2.02 (0.87, 4.66)	.10	0.69	2.00 (0.85, 4.69)	.11
Late-onset interventionb	0.93	2.55 (1.12, 5.81)	.03	0.90	2.47 (1.05, 5.79)	.04	0.83	2.30 (0.97, 5.45)	.06
OVIDc	1.30	3.67 (1.69, 7.97)	.001	1.31	3.69 (1.68, 8.08)	<.001	1.25	3.50 (1.65, 7.42)	.001

Note. ORs are modeling the probability of drinking. ICCs = breakpoint, 0.434; O_max, 0.412; intensity, 0.470.

^aFemale = reference

^bControl group = reference

^cPre-COVID = reference

Drinking continuation

We used evening reports on drinking days only (N=2911) to examine associations between demand and likelihood of continuing to drink. Starting at the first drinking report of the evening, we calculated a lagged momentary demand variable, then removed the first drinking observation, resulting in 1421 remaining observations. For all demand indices, higher momentary alcohol demand was associated with higher odds of continuing to drink once initiated (Table 4). Higher lagged demand predicted higher odds of continued drinking at the following report. At the day-level, higher demand on a given day was associated with increased odds of continued drinking. At the person-level, higher demand O_max and intensity, on average across drinking days, were associated with higher odds of continued drinking. Continued drinking was less likely to occur later into the evening in the intensity model, though time was not conclusively linked to continued drinking in breakpoint and O_max models. Other covariates did not conclusively indicate associations with odds of continued drinking.

Table 4.

Associations between Alcohol Demand and Drinking Continuation

Effect	Breakpoint			Omax			Intensity
	b	OR (95% CI)	p	b	OR (95% CI)	p	b	OR (95% CI)	p
Intercept	0.27			0.25			0.66
Lagged demand	0.38	1.47 (1.13, 1.90)	.004	0.40	1.50 (1.18, 1.90)	.001	0.28	1.32 (1.12, 1.56)	.001
Momentary demand	0.44	1.55 (1.17, 2.05)	.002	0.44	1.56 (1.18, 2.06)	.002	0.24	1.27 (1.02, 1.56)	.03
Day-level demand	0.76	2.14 (1.52, 3.01)	<.001	0.81	2.25 (1.64, 3.08)	<.001	0.91	2.49 (1.71, 3.64)	<.001
Person-level demand	0.21	1.23 (0.92, 1.65)	.16	0.33	1.40 (1.05, 1.86)	.02	0.25	1.28 (1.09, 1.51)	.003
Covariates
Hours after first report	−0.18	0.83 (0.68, 1.02)	.07	−0.19	0.82 (0.67, 1.01)	.06	−0.24	0.79 (0.63, 0.98)	.03
Sexa	0.41	1.51 (0.71, 3.22)	.28	0.34	1.41 (0.66, 3.00)	.37	0.02	1.02 (0.47, 2.21)	.97
Early-onset interventionb	0.48	1.62 (0.63, 4.18)	.31	0.53	1.70 (0.68, 4.29)	.25	0.52	1.68 (0.68, 4.15)	.26
Late-onset interventionb	0.30	1.35 (0.56, 3.21)	.50	0.41	1.50 (0.64, 3.50)	.34	0.22	1.24 (0.53, 2.93)	.61
COVIDc	0.68	1.98 (0.88, 4.44)	.10	0.72	2.06 (0.93, 4.54)	.07	0.57	1.76 (0.79, 3.91)	.16

Note. ORs are modeling the probability of continued drinking. ICCs = breakpoint, .380; O_max, 0.395; intensity, 0.438.

^aFemale = reference

^bControl group = reference

^cPre-COVID = reference

Alcohol consumption (BrAC)

We used evening drinking reports with breathalyzer data (N=2065) to examine associations between demand and achieved intoxication levels. Lagged demand was not included in these models due to a substantial decrease in observations with lagged data available (N=883; 42.8% of BrAC observations). Higher momentary alcohol demand was associated with higher BrACs (Table 5). At the day-level, higher demand during a given episode was associated with higher average BrAC. At the person-level, higher demand intensity was associated with higher BrAC. Higher BrACs were achieved later into the evening and among participants recruited pre-COVID. Male participants had higher BrACs on average in the breakpoint and O_max models. Other covariates did not conclusively indicate associations with BrAC.

Table 5.

Associations between Alcohol Demand and Achieved Intoxication (Breath Alcohol Concentration [BrAC])

Effect	Breakpoint			Omax			Intensity
	b (95% CI)	t	p	b (95% CI)	t	p	b (95% CI)	t	p
Intercept	0.0348 (0.0230, 0.0466)	5.88	<.001	0.0344 (0.0228, 0.0459)	5.92	<.001	0.0347 (0.0232, 0.0462)	6.03	<.001
Momentary demand	0.0021 (0.0003, 0.0039)	2.34	.02	0.0026 (0.0009, 0.0043)	3.06	.002	0.0011 (0.0001, 0.0022)	2.15	.03
Day-level demand	0.0058 (0.0038, 0.0079)	5.74	<.001	0.0062 (0.0041, 0.0082)	6.11	<.001	0.0053 (0.0031, 0.0075)	4.89	<.001
Person-level demand	0.0016 (−0.0015, 0.0047)	1.04	.30	0.0022 (−0.0008, 0.0052)	1.46	.15	0.0018 (0.0002, 0.0034)	2.23	.03
Covariates
Hours after first report	0.0044 (0.0033, 0.0055)	7.99	<.001	0.0045 (0.0034, 0.0056)	8.07	<.001	0.0044 (0.0033, 0.0055)	8.05	<.001
Sexa	0.0106 (0.0008, 0.0205)	2.15	.04	0.0107 (0.0010, 0.0203)	2.19	.03	0.0087 (−0.0012, 0.0186)	1.76	.08
Early-onset interventionb	0.0096 (−0.0025, 0.0217)	1.58	.12	0.0097 (−0.0019, 0.0214)	1.66	.10	0.0109 (−0.0006, 0.0224)	1.89	.06
Late-onset interventionb	0.0025 (−0.0094, 0.0144)	0.41	.68	0.0032 (−0.0085, 0.0149)	0.54	.59	0.0046 (−0.0070, 0.0162)	0.80	.43
OVIDc	−0.0126 (−0.0228, −0.0025)	2.48	.02	−0.0122 (−0.0223, −0.0021)	2.40	.02	−0.0121 (−0.0219, −0.0024)	2.47	.02

Note. ICCs = breakpoint, 0.436; O_max, 0.452; intensity, 0.485.

^aFemale = reference

^bControl group = reference

^cPre-COVID = reference

Exploratory correlations between demand indices and drinking behavior

We examined construct overlap between alcohol demand indices assessed during EMA compared to those assessed with the conventional APT, and their associations with past month and EMA drinking behavior (Table 6). We created EMA averages within person across all moments and drinking moments only, respectively, to generate person-means for EMA alcohol demand, total number of drinks consumed, and BrAC (on drinking days). For drinking moments only, all EMA alcohol demand measures were positively correlated with past month drinking quantity and binge drinking frequency, EMA total drinks consumed and BrAC, and APT intensity. For all moments, all EMA demand indices were positively correlated with EMA total drinks consumed and BrAC. EMA intensity was positively correlated with past month drinking quantity and binge drinking frequency. Additionally, EMA O_max and intensity were positively correlated with APT intensity.

Table 6.

Correlations between Person-Level Means for Alcohol Demand in EMA, APT Demand Indices, and Drinking Behavior

	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.	11.
1. M Breakpoint (EMA)		.92**	.63**	.14	.16	.19	.01	.14	.13	.28*	.36**
2. M Omax (EMA)	.90**		.79**	.14	.19	.35**	.01	.20	.18	.31*	.38**
3. M Intensity (EMA)	.59**	.72**		.10	.20	.57**	.09	.27*	.22†	.36**	.35**
4. Breakpoint (APT)	.05	.00	−.08		.63**	−.08	.13	−.12	.01	−.16	−.07
5. Omax (APT)	.17	.18	.17	.63**		.30*	.19	.26*	.17	.17	.15
6. Intensity (APT)	.25†	.38**	.66**	−.08	.30*		.08	.67**	.56**	.53**	.42**
7. Past month frequency	−.04	−.04	.04	.13	.19	.08		.12	.39**	.16	.10
8. Past month quantity	.30*	.38**	.47**	−.12	.26*	.67**	.12		.67**	.72**	.42**
9. Past month binge frequency	.22†	.25†	.31*	.01	.17	.56**	.39**	.67**		.54**	.41**
10. M Number of drinks (EMA)	.40**	.47**	.58**	−.16	.17	.53**	.16	.72**	.54**		.75**
11. M BrAC (EMA)	.42**	.45**	.46**	−.06	.15	.43**	.10	.42**	.40**	.78**

Note. Correlations are between person-means averaged across all EMA moments (top-right triangle of table) or across drinking moments only (bottom-left triangle of table). Number of drinks and BrAC included values on drinking days only. Some correlation values are identical across the table because they are measures assessed once at baseline (APT, drinking in past month). APT = Alcohol Purchase Task; BrAC = breath alcohol concentration

^**p < .001,

^*p ≤ .01,

^†p < .05

Of the APT demand indices, only intensity was consistently positively correlated with drinking behavior assessed retrospectively or during EMA. Within measure type, EMA breakpoint, O_max, and intensity were all highly correlated, whereas for the APT, only the correlations between O_max and breakpoint and O_max and intensity reached significance.

DISCUSSION

We found that fluctuations in alcohol demand were associated with drinking in individuals’ natural drinking environments. As expected, higher momentary demand predicted greater likelihood of drinking and continuing to drink concurrently and prospectively (lagged associations). Additionally, higher demand on a given day was associated with drinking and continued drinking. These findings are consistent with studies evincing fluctuations in demand under acute alcohol intoxication in the laboratory (15,16), and complement studies finding increased demand in response to alcohol cues and stressors (4,16,23,28). Importantly, our findings extend prior research to demonstrate that fluctuations in demand are motivationally relevant, in that they predict concurrent and subsequent drinking behavior. To our knowledge, this is the first study to examine state-dependent alcohol demand in the natural environment and its ability to predict when individuals are more likely to drink.

We also found that higher momentary demand and demand within an episode were associated with greater BrACs. These associations implicate increased alcohol demand as a potential motivational influence on the total amount of alcohol consumed when individuals are drinking. Our findings provide preliminary support for the hypothesis that alcohol consumption can increase alcohol’s reinforcing value (15,16), in turn influencing additional consumption. However, our EMA protocol assessed behaviors at two-hour intervals, resulting in insufficient data to establish temporal precedence of demand predicting BrAC at later timepoints. Additional EMA studies with more densely assessed drinking episodes are needed to support this potential temporal association.

Our findings also provide preliminary external validation of state-dependent alcohol demand items. EMA demand indices were robustly associated with in-the-moment drinking, and correlations across moments indicated EMA measures may capture unique variance that the conventional APT does not. The only APT index consistently associated with consumption was intensity, which also correlated with EMA demand indices. When disaggregated at each level of analysis, however, individual differences in EMA demand were less robustly associated with drinking behavior, excepting demand intensity. Collectively, these findings suggest that, EMA demand intensity may better capture motivational processes relevant to individual differences in alcohol consumption, similar to APT intensity (12), whereas other demand indices may only predict within-person fluctuations in drinking behavior.

These results raise questions about whether state-dependent demand items capture the same underlying aspects of alcohol value as the trait-based APT. Correlations between state and trait indices were moderate for intensity, but negligible for O_max and breakpoint. A recent daily diary study (19) which examined associations between trait APT indices and daily alcohol demand also found a stronger association between daily and trait-level intensity than for breakpoint or O_max. This pattern may reflect differences in how these indices are generated; intensity is assessed via a single item in both assessment formats, whereas breakpoint and O_max are generated from observed values on the APT and assessed via single items in EMA. A related issue is whether EMA demand measures themselves capture unique aspects of demand, given high correlations between EMA demand indices. Merrill & Aston (19) reported a similarly high correlation between daily breakpoint and O_max, and smaller correlations between other indices (see also (23)). Determining whether EMA indices truly measure distinct aspects of alcohol demand should be addressed in future studies. EMA demand was also noticeably lower than demand observed on the APT in this study. Given that different time frames/scenarios were presented to participants across these assessments (demand “right now” versus anticipated demand in a typical drinking scenario), this is perhaps not surprising. Regardless, modest associations between EMA and APT indices suggest that state-dependent demand aggregated across moments may not map onto expected demand on the APT. This conclusion is based on limited data and requires further investigation.

There are several ways in which our findings may inform theory and clinical investigations. First, our findings suggest that changes in alcohol’s reinforcing value may serve as a motivational index of when individuals are most likely to drink and consume greater alcohol quantities, as well as when they may be less likely to drink and consume less. Prediction of alcohol use is often a goal of clinical research, both to understand development of addiction and AUD treatment effectiveness (29). Though conducted in a non-clinical sample, our results indicate that in-the-moment assessments of alcohol demand can provide insight into motivational processes that support AUD. Extending these findings to treatment-seeking samples is a necessary next step, as these items may operate differently in treatment-seeking populations.

Second, our study focused on one aspect of the valuation-consumption tradeoff, perceived alcohol value predicting consumption. As the cost to consume alcohol varies across drinking environments, we may expect individuals to drink more when alcohol is less expensive (e.g., during happy hour pricing) and less when alcohol is more expensive (e.g., at expensive drinking venues) (30). Individuals may be more likely to drink to excess when cost conditions are favorable, while policies that increase consumption costs can deter excessive consumption (31,32). Inexpensive drinking settings may also pose a greater threat to violating drinking intentions when attempting to abstain from or regulate consumption. The interaction between alcohol demand, actual monetary conditions, and consumption in different settings may reveal critical information about who and when individuals are most at-risk for excessive drinking.

Third, although we did not investigate alcohol craving, prior research indicates that craving predicts changes in alcohol demand under intoxication (15). Discerning the extent to which alcohol craving and demand operate as distinct or related motivational indices is important, given that craving also predicts alcohol consumption in naturalistic environments (33). Elucidating whether demand offers incremental validity in predicting drinking independent of (or via) craving is a valuable direction for future research.

Some limitations are important to consider. First, we did not examine contextual influences on alcohol demand (e.g., drinking companions, locations), which may partially account for alcohol demand’s associations with drinking behavior. Second, our compliance rate was lower than is typically seen in EMA studies of individuals who use substances (~75.1%, 34), though this includes assessments that may not have been completed during participants’ non-waking hours (e.g., 12:00am and 2:00am surveys, see Supplemental Materials). Finally, our sample included mostly highly educated individuals recruited from a university area. Generalization to more rural areas and broader populations of individuals is needed and appears feasible using a mobile technology-based design.

Overall, the present study indicates that alcohol demand assessed in-the-moment is associated with increased likelihood of drinking occurrence and continuation, and greater total alcohol consumption. These findings implicate state-dependent alcohol demand as a relevant motivational index of drinking behavior in the natural environment.