The Association Between Negative and Positive Affect and Alcohol Use: An Ambulatory Study

Abstract

Objective:

Negative and positive affect are proposed to play a crucial role in alcohol use and the development of alcohol use disorder. Results from ambulatory studies that measure momentary affect and subsequent alcohol use have been mixed, particularly regarding negative affect. We attempted to identify within-person moderators (i.e., time between assessments, prior blood alcohol content) that might explain mixed results.

Method:

We examined the association between self-reported affect and an objective measure of alcohol consumption (measured via a transdermal ankle bracelet) in a sample of heavy social drinkers across 7 days of ambulatory assessment.

Results:

Our results showed that negative affect was negatively related to later drinking, whereas positive affect was positively related to later drinking. The results showed that these effects were stronger for amount consumed when affect was assessed closer rather than farther in time.

Conclusions:

These findings are important for understanding affect as an antecedent to alcohol use, which may ultimately have implications for the development of alcohol use disorder.

Affect plays a crucial role in many theories of alcohol use and etiology of alcohol use disorder (e.g., Conger, 1956; Fairbairn & Sayette, 2013). Research shows that self-reports of drinking to reduce negative affect and increase or prolong positive affect are associated with frequency of drinking and drinking problems (Patrick et al., 2011; Studer et al., 2016). Neuroscientific models emphasize the importance of positive reinforcement (e.g., increasing positive affect) for heavy users and negative reinforcement (e.g., decreasing negative affect) for dependent users (Koob, 2009). Despite the prominent theoretical role for affect, empirical data linking momentary affective states to drinking behavior have been mixed (e.g., Dvorak et al., 2016; Treloar et al., 2015).

In terms of negative affect, a recent meta-analysis found a significant, but small, effect of negative affect inductions on increased alcohol use under laboratory conditions (Bresin et al., 2018). Ambulatory studies—in which researchers measure momentary affect and alcohol use in participants’ natural drinking environments—have generally not found strong evidence of a positive association between prior negative affect and later alcohol use. Dvorak et al. (2016) found no significant association between daytime negative affect and nighttime self-reports of drinking or number of drinks consumed in college undergraduates (see Pisetsky et al., 2016, for similar null results in a sample of women with eating disorders and Armeli et al., 2007, for results with college students). Simons et al. (2014) found that daytime negative affect was not significantly related to evening abstaining from drinking but was positively related to the number of drinks consumed for college undergraduates. In a community sample, Treloar et al. (2015) found a negative association, in that negative affect decreased in the hours leading up to self-reported alcohol consumption.

In contrast to negative affect, ambulatory studies have generally found that higher positive affect is related to later drinking (e.g., Armeli et al., 2007; Dvorak et al., 2016; Treloar et al., 2015). This is consistent with research and theory identifying the role of prolonging positive affect as an additional motive for alcohol consumption (e.g., Fairbairn et al., 2015). This is also in line with theories of addiction that emphasize the role of the reward of positive affect in substance use among nondependent individuals (Koob, 2009) and incentive models of substance use (e.g., Robinson & Berridge, 2001).

The lack of consistent findings for negative affect may indicate that there are important moderators of the effect. Although some studies have identified individual differences that change the strength of the relation between affect and alcohol use (e.g., emotional differentiation; Kashdan et al., 2010), few studies have examined within-person moderators.

Prior studies have varied in the time between the assessment of affect and alcohol use. Studies that have found no significant association for negative affect have examined how affect during the afternoon is related to nighttime drinking, often reported the next day (e.g., Dvorak et al., 2016). Treloar et al. (2015), who found a negative association, used a more comprehensive assessment strategy with up to five random prompts a day and hourly post-drink assessments, giving a greater coverage of moment-to-moment changes. Because affect motivates individuals for action (e.g., Lang et al., 1990), it is likely that affect is more motivating for, and therefore predictive of, behavior that occurs close in temporal proximity. Thus, it is possible that the relation between affect and alcohol use is stronger the closer in time affect is assessed in relation to the time alcohol use is measured. This study used an ambulatory protocol that allowed for us to examine how moment-to-moment changes in affect were related to alcohol use in the next 2 hours.

Negative and positive affect may play different roles in the initiation of drinking and the continuation of drinking once it has begun. For instance, individuals may start drinking because they are experiencing negative affect; however, if negative affect occurs after drinking has begun, it may influence one to stop drinking. Conversely, an increase in positive affect during drinking may increase later alcohol consumption because of the association between positive affect and appetitive behavior (Lang et al., 1990). The relation between affect and alcohol use may also be influenced by whether the individual is in the ascending or descending limb of the blood alcohol concentration (BAC) curve (Rueger & King, 2013). Research shows that there are biphasic effects of the subjective experience of alcohol, with more activation/elation effects before the peak of BAC and more deactivation/sedation effects after the peak (Earleywine & Erblich, 1996). It is possible that positive affect, and perhaps negative affect, may be a stronger predictor of drinking behavior before the BAC peak (vs. after the peak) because elation is an activating emotion that will lead to action. Conversely, affect may be less related to drinking behavior after BAC has reached its peak because of the sedation effects, which may lead to inaction.

This study also builds on the literature, which has relied on self-reported drinking, by using a novel objective measure of alcohol consumption, the Secure Remote Alcohol Monitoring System (SCRAM; Alcohol Monitoring Systems, Inc., Littleton, CO). SCRAM is an ankle bracelet that continuously measures transdermal alcohol concentration (TAC; i.e., the excretion of alcohol through the skin). At the time of data collection, SCRAM was the most reliable and valid transdermal alcohol sensor available for research (Fairbairn et al., 2018b; Leffingwell et al., 2013). Moreover, we used the best practice for estimating BAC from TAC by collecting simultaneous readings of breath alcohol concentration (BrAC) and TAC in a laboratory alcohol administration session. This allows for individualized calibration for each device–participant pairing, and thus addresses many limitations of TAC, which primarily center on its indirect relation to blood alcohol level (Luczak & Rosen, 2014; Rosen et al., 2014).

The goal of this study was to extend the literature by examining how recent affect is related to later alcohol use in a sample of heavy social drinkers across 7 days of ambulatory assessment.¹ Moreover, we sought to identify within-person moderators that might influence the relation between affect and alcohol use. Based on theory and laboratory studies (Bresin et al., 2018), we predicted that negative affect at the prior prompt would be positively related to alcohol use at the next prompt. In line with prior ambulatory studies (e.g., Dvorak et al., 2016; Pisetsky et al., 2016), we also predicted that positive affect at the prior prompt would be positively related to alcohol use at the next prompt.

In terms of moderators, we predicted that the relation between affect and alcohol use would be stronger when it was assessed closer (vs. further) in time. We also predicted that negative affect would have a stronger relation with drinking when prior BAC was low, whereas positive affect would have a stronger relation with drinking when prior BAC was high. Given that there was limited research on limb of the BAC curve, we explored this as a moderator. In line with recent recommendations to reduce the influence of researcher degrees of freedom (Simmons et al., 2011), we report results for both supported and unsupported hypotheses, using supplemental material to detail additional analyses. (Supplemental material appears as an online-only addendum to the article on the journal’s website.)

Method

Participants

Participants were 48 heavy social drinkers (24 women; 37 full-time students), with an average age of 22.6 years of age (range: 21–28). (See supplemental material for a power analysis.) The racial/ethnic breakdown was 56% European American, 13% African American, 6% Hispanic, and 17% Asian. This sample and these data were used in Fairbairn et al. (2018a) to address a distinct, but related, question about how consumption of alcohol affects positive and negative affect and whether this was moderated by social context. Importantly, this previous article included no information about the relation between affect and subsequent alcohol consumption.

Participants were recruited via advertisements for a paid alcohol research study. Eligibility criteria were designed for the laboratory alcohol dosing session, which was used to calibrate the SCRAM bracelet. Eligibility criteria (e.g., age, body mass index) and ineligibility criteria (i.e., current alcohol use disorder, medical conditions that interact with alcohol, pregnancy in women) were assessed via phone and in-person interviews.

All participants met the criteria for at-risk drinking as defined by the National Institute on Alcohol Abuse and Alcoholism (2017) criteria of at least one occasion of 3 or more drinks for women and 4 or more drinks for men in a single day. Participants reported an average of 4.09 (SD = 3.01, maximum = 12) binge episodes in the last month. Participants reported drinking an average of two to three times per week and 3.8 (SD = 1.7) drinks/occasion.

Procedure

Participants who met eligibility criteria completed three laboratory visits and 1 week of ambulatory assessment. The first visit familiarized participants with the ambulatory assessment procedures, whereas one of the subsequent visits involved the administration of alcohol to calibrate the transdermal sensors. (See supplemental materials and Fairbairn et al., 2018b, for details.) In the first visit, participants were fitted with transdermal sensors and oriented to the ambulatory assessment procedures. Participants were informed that they would be completing surveys in response to random prompts, six per day between noon and midnight, for the next 7 days. Prompts were programmed to occur at random times roughly 2 hours apart, although the exact time was allowed to vary randomly. After receiving a prompt, participants had to respond within 15 minutes, reporting their current mood along with other measures not relevant to the current study (Fairbairn et al., 2018a, 2018b). Prompts were delivered through the Metricwire survey app (Trafford, 2016) directly on participants’ smartphones. For participants who did not own a smartphone, an iPod touch was provided.

Measures

Affect was assessed with a 10-item measure, which included five negative affect items (nervous, sad, irritated, lonely, bored) and five positive affect items (upbeat, content, happy, euphoric, energized). Participants rated items as to how they felt in that current moment (i.e., right now). Internal consistency was calculated using equations provided by Bonito et al. (2012) for nested data. Both scales showed adequate internal consistency between persons (negative affect = .93; positive affect = .96), although, as in prior research (Nezlek, 2017), the reliability was somewhat lower at the occasion level (negative affect = .44; positive affect = .71). Items were taken from previous studies (e.g., Fairbairn & Sayette, 2013). (See supplemental materials for more details on item identification.)

Alcohol consumption was assessed with the SCRAM bracelet. This device fits around the ankle and was used to measure TAC. Readings derived from transdermal sensors were translated into estimates of BrAC (eBrAC) using BrAC Estimator software. BrAC Estimator is a Matlab code based on a first principles forward model for the transport of alcohol from the blood through the skin and measurement by a transdermal sensor (Rosen et al., 2014). The parameters in the model are estimated, or tuned, to the particular device and participant using BrAC and TAC data collected during the alcohol laboratory session. A prior study with these data found a strong positive correlation between self-reported number of drinks on the prior day and SCRAM eBrAC (β = .75; Fairbairn et al., 2018b).

Data analytic plan

Because of the nested nature of the data, multilevel modeling was used (Singer & Willett, 2003). Because eBrAC, the primary outcome variable, was positively skewed with a high number of zeros, we used zero-inflated gamma extensions of multilevel modeling, which simultaneously fit two models to the data. The zero-inflation portion of the model is a logistic regression model predicting zeros, in our case, zero-eBrAC (i.e., not drinking) versus nonzero-eBrAC (i.e., drinking) occasions. The gamma portion of the model explains variation among the non-zeros, in our case, eBrAC among drinking occasions. Thus, the zero-inflation model explains whether participants drank or not, and the gamma model explains how much alcohol they consumed per drinking occasion.

We tested several models. In each, the Level 1, within-subject model contained positive or negative affect at the prior prompt within the same day and several covariates that have been found to be associated with alcohol use in prior work: gender (effect coded), number of drinking days in the last 30 days reported at baseline, time of day, and day of week (effect coded).² All Level 1 variables were centered within person so that slopes could be interpreted as within-subject effects, and all Level 2 variables were grand-mean centered. We also ran additional models that tested moderators of the relation between affect and alcohol use. First, we included the time between assessment of affect and alcohol (centered within person). Second, we included eBrAC at the prior assessment as a moderator. Because zero was a meaningful value for eBrAC, we did not center this variable.

To examine whether the limb of the BAC curve affected the relation between affect and alcohol use, we included an effect-coded variable to denote whether it was the ascending or descending limb of the BAC curve, determined on whether eBrAC had reached its peak. We also conducted robustness analyses, which included current affect as a covariate. In the main text of the article, we focus on our theoretically derived, a priori hypotheses. The supplemental materials contain additional exploratory and follow-up analyses (e.g., limb of BAC curve, day of the week, and dynamic changes in affect). All data analyses were conducted in SAS 9.4 using Proc NLMIXED (SAS Institute Inc., Cary, NC).

Results

During the 7-day ambulatory assessment period, the majority (94%) of participants reported drinking at least once. On average, participants drank on 3.3 days (SD = 1.56), not including the laboratory visit. The average non-zero eBrAC at the time of ambulatory prompts was .04 (SD = .08, 95% CI [.03, .05]). The average peak eBrAC for drinking occasions was .11 (SD = .17, 95% CI [.10, .13]).³ Compliance with the ambulatory procedures was excellent. Participants responded to an average of 93.1% of prompts (SD = 10.6, minimum = 51, maximum = 100). Aside from one participant who experienced an equipment malfunction, all participants responded to at least 70% of prompts. There were no incidents of noncompliance with SCRAM monitoring. The average time between prompts within a day was 2.31 hours (SD = 1.27, Mdn = 2; interquartile range = 1). Preliminary analyses showed no evidence of reactivity to wearing the bracelet (see the supplemental material).

Negative affect at the prior time point was positively related to zero-eBrAC at the subsequent time point (b = 0.46, 95% CI [0.01, 0.83], p = .015) in the zero-inflation portion of the model. Stated differently, when negative affect was high, people were less likely to drink at the next assessment. This finding is counter to our hypotheses but is similar to the results of Treloar et al. (2016). For the gamma portion of the model, there was not a significant effect of negative affect, suggesting that for drinking occasions negative affect was not significantly related to the amount of alcohol consumed (see Table 1 for parameter estimate).

Table 1.

Regression coefficients and 95% confidence intervals for affect predicting alcohol consumption within day

	Zero inflation		Gamma
Variable	Coefficient	p	Coefficient	p
Negative affect
Intercept	0.74 [0.41, 1.07]	<.001	-3.60 [-3.81, -3.39]	<.001
Gender	0.47 [0.14, 0.80]	.007	0.02 [-0.15, 0.19]	.812
Drink days	-0.07 [-0.13, -0.01]	.019	0.02 [-0.01, 0.05]	.156
Time	-0.15 [-0.20, -0.10]	<.001	0.14 [0.07, 0.21]	<.001
Weekend	-0.55 [-0.71, -0.40]	<.001	0.45 [0.27, 0.63]	<.001
Lag negative affect	0.46 [0.10, 0.83]	.014	-0.03 [-0.53, 0.46]	.895
Positive affect
Intercept	0.74 [0.41, 1.87]	<.001	-3.61 [-3.82, -3.40]	<.001
Gender	0.47 [0.14, 0.80]	.006	0.03 [-0.15, 0.20]	.772
Drink days	-0.07 [-0.13, -0.01]	.018	0.02 [-0.01, 0.05]	.206
Time	-0.15 [-0.19, -0.10]	<.001	0.14 [0.07, 0.21]	<.001
Weekend	-0.53 [-0.68, -0.37]	<.001	0.44 [0.26, 0.62]	<.001
Lag positive affect	-0.35 [-0.55, -0.15]	.001	0.10 [-0.16, 0.37]	.437

Notes: p values are based on t(47). Gender was coded as 1 = men, -1 = women; drink days = number of drinking days during the 30 days before the study; time = time of day in hours; weekend was coded as 1 = weekend, -1 = weekday (centered within person).

There was a significant interaction between negative affect and time since the last assessment in the gamma portion of the model (b = 0.54, 95% CI [0.17, 0.91], p = .005) but not the zero-inflated portion of the model (b = -0.001, 95% CI [-0.30, 0.30], p = .994). The top panel of Figure 1 shows the relation between negative affect and eBrAC at different times between assessments. Follow-up tests found that there was a significant negative relation between negative affect and eBrAC when affect was reported roughly 1.25 hours prior (-1 SD; b = -0.78, 95% CI [-1.35, -0.22], p = .008). At +1 SD, however, the effect was not significant (∼3.8 hours: b = 0.41, 95% CI [-0.24, 1.07], p = .212). Consistent with our prediction, then, affect assessed closer in time was a significant predictor of later alcohol use, whereas affect assessed further back in time is not.

Figure 1.

Estimated blood alcohol content at the current time (y-axis) as a function of time since the last assessment (separate lines) and affect at the last assessment (x-axis; top panel) and estimated breath alcohol content (separate lines) and affect at the last assessment (x-axis; bottom panel).

For eBrAC at the prior assessment, the interaction was not significant for the zero-inflation model (b = -6.38, 95% CI [-41.34, 28.59], p = .715) but was for the gamma portion of the model (b = -18.92, 95% CI [-37.31, -0.54], p = .044). As shown in the bottom panel of Figure 1, when eBrAC at the prior prompt was zero, there was a marginally significant positive relation between negative affect and later eBrAC (b = 0.49, 95% CI [-0.07, 1.04], p = .082). The effects were not significant between .02 and .08 eBrAC (ps range from .648 to .117). At .1 eBrAC, there was a marginally significant negative relation between negative affect and eBrAC (b = -1.40, 95% CI [-3.03, 0.24], p = .093). It is also worth noting that in the zero-inflation part of the model, the relation between prior negative affect and drinking occasions was similar in size to the model without lag eBrAC, but not significant (b = 0.36, 95% CI [-0.05, 0.76], p = .086). This suggests that this effect was partially dissociable from effects of prior alcohol consumption on affect. There was not a significant interaction between whether participants were on the ascending or descending limb of the BAC curve for the zero-inflation or gamma models for negative affect or positive affect (also see Table 2).

Table 2.

Regression coefficients and 95% confidence intervals for affect and limb of blood alcohol concentration (BAC) curve predicting alcohol consumption within day

	Zero inflation		Gamma
Variable	Coefficient	p	Coefficient	p
Negative affect
Intercept	0.79 [0.46, 1.12]	<.001	-3.62 [-3.83, -3.40]	<.001
Gender	0.46 [0.13, 0.79]	.007	0.02 [-0.15, 0.20]	.804
Drink days	-0.07 [-0.13, -0.01]	.025	0.02 [-0.01, 0.05]	.155
Time	-0.16 [-0.21, -0.11]	<.001	0.14 [0.08, 0.21]	<.001
Weekend	-0.52 [-0.68, -0.36]	<.001	0.45 [0.26, 0.63]	<.001
Lag negative affect	0.46 [0.09, 0.82]	.016	-0.05 [-0.55, 0.45]	.841
BAC limb	-0.48 [-0.70, -0.26]	<.001	0.05 [-0.13, 0.23]	.569
Lag Negative Affect × BAC Limb	0.13 [-0.44, 0.70]	.649	0.10 [-0.39, 0.59]	.675
Positive affect
Intercept	0.79 [0.46, 1.12]	<.001	-3.62 [-3.84, -3.41]	<.001
Gender	0.47 [0.14, 0.80]	.006	0.02 [-0.15, 0.20]	.792
Drink days	-0.07 [-0.13, -0.01]	.024	0.02 [-0.01, 0.05]	.210
Time	-0.16 [-0.21, -0.11]	<.001	0.14 [0.07, 0.21]	<.001
Weekend	-0.49 [-0.65, -0.33]	<.001	0.44 [0.26, 0.62]	<.001
Lag positive affect	-0.37 [-0.58, -0.17]	.001	0.11 [-0.15, 0.37]	.415
BAC limb	-0.53 [-0.75, -0.31]	<.001	0.05 [-0.13, 0.23]	.585
Lag Positive Affect × BAC Limb	0.21 [-0.10, 0.52]	.178	0.10 [-0.25, 0.26]	.954

Notes: p values are based on t(47). Gender was coded as 1 = men, -1 = women; drink days = number of drinking days during the 30 days before the study; time = time of day in hours; weekend: 1 = weekend, -1 = weekday (centered within person); BAC limb = blood alcohol concentration limb, coded as 1 = ascending, -1 = descending.

Positive affect at the prior time point was negatively related to a zero-eBrAC at the next time point (b = -0.35, 95% CI [-0.54, -0.15], p = .001), indicating that positive affect predicted a greater likelihood of later drinking. For drinking occasions, positive affect was not significantly related to eBrAC (b = 0.10, 95% CI [-0.16, 0.36], p = .437). Together these results suggest that positive affect may increase the chances of initiating drinking but does not predict drinking a greater quantity.

Positive affect significantly interacted with time since the last assessment (b = -0.20, 95% CI [-0.39, -0.02], p = .033) in the gamma model. The follow-up analyses showed that positive affect had a significant positive relation to eBrAC for assessments closer in time (-1 SD: b = 0.44, 95% CI [0.06, 0.82], p = .023) but not further back in time (+1 SD: b = -0.06, 95% CI [-0.37, 0.27], p = .728; Figure 2). Thus, similar to negative affect, positive affect was more predictive of drinking when it was assessed closer (vs. further) in time. Lag eBrAC did not significantly interact with positive affect to predict eBrAC (see Supplemental Table 2 for results). Positive affect was still significantly related to zero-BAC when lag eBrAC was in the model (b = -0.36, 95% CI [-0.56, -0.16], p = .001), suggesting that this effect is not fully explained by prior effects of alcohol.

Figure 2.

Estimated blood alcohol content (y-axis) at the current time as a function of time since the last assessment (separate lines) and positive affect at the last assessment (x-axis).

In general, the results were very similar when concurrent affect was added to the model. For instance, in the zero-inflation portion of the base models, prior negative affect was positively related to zero-eBrAC (b = 0.36, 95% CI [-0.02, 0.74], p = .064), and prior positive affect was negatively related to zero-eBrAC (b = -0.19, 95% CI [-0.41, 0.02], p = .075), although in both cases zero was in the confidence interval. For negative affect, the two significant moderator effects were still significant (time since the last prompt: b = 0.52, 95% CI [0.16, 0.90], p = .006; lag eBrAC: b = 17.98, 95% CI [-35.84, -0.12], p = .049). For positive affect, however, the interaction with time since the last prompt was no longer significant (b = -0.12, 95% CI [-0.32, 0.06], p = .170). Together these results show some robustness for our effects in that the effect size and confidence intervals were generally similar in size when including additional covariates. There were, however, some instances where p values were moved above the significance threshold. These changes in significance may be a function of the low informational value of p values (Cumming, 2014). They could also indicate bi-directional relations between affect and alcohol use.

Concurrent affect had much the same relation with eBrAC as prior affect. For negative affect, there was a positive relation with zero-eBrAC (b = 0.41, 95% CI [0.03, 0.78], p = .032) and a negative nonsignificant relation with eBrAC for drinking occasions (b = -0.31, 95% CI [-0.76, 0.14], p = .162). For positive affect, there was a negative relation with zero-eBrAC (b = -0.42, 95% CI [-0.62, -0.21], p < .001) and a positive relation with eBrAC for drinking occasions (b = 0.46, 95% CI [0.20, 0.73], p = .001).

Discussion

This study makes two novel contributions to the literature. First, we examined within-person moderators that may change the relation between affect and alcohol use. Second, we used a novel, objective measure of ambulatory alcohol use. Our results showed that negative affect was negatively, and positive affect was positively, related to later drinking. Moreover, these relations were stronger for amount consumed when affect was assessed closer rather than farther in time.

Consistent with prior studies (e.g., Dvorak et al., 2016; Simons et al., 2010), we found that higher momentary positive affect was positively related to later drinking and drinking a greater quantity at least when affect was assessed closer (vs. further) in time. These results are in line with a growing body of theories that suggest that people drink alcohol to prolong positive affect (Patrick et al., 2011). Given that the positive affect findings are replicable across samples, it appears that more work is needed to elucidate the role of positive affect in alcohol use and development of alcohol use problems.

Our negative affect results were similar to those of Treloar et al. (2016), who found that negative affect decreased before alcohol consumption. Our study was more similar to Treloar et al. than other prior studies in two ways. First, we used a more intensive sampling procedure, which allowed us to model moment-to-moment affect and alcohol relations. Second, our sample was a mix of community members and students similar to Treloar et al. The finding of a negative relation between negative affect and alcohol use is contrary to many scientific and lay theories that posit a positive relation (e.g., Baker et al., 2004) and a meta-analysis that found a positive association in the lab (Bresin et al., 2018). One possible explanation for this unexpected finding is our use of a heavy social drinking sample. It is possible that the relation between negative affect and alcohol use may be positive in an alcohol-dependent sample (Koob, 2009). Another possibility is that participants expect or have had experiences where negative affect is made worse by consuming alcohol and therefore, avoid drinking when negative affect is high. Future research needs to determine whether the link between affect and alcohol is explained by magnifying emotions (e.g., negative becomes more negative) or whether among nondependent drinkers, alcohol is just more directly linked with positive affect than negative affect.

We also attempted to identify within-person moderators. We found that affect was a stronger predictor of the quantity of drinking when it was assessed closer in time. This fits with the idea that affect is more motivating for immediate action and may explain some of the mixed results across studies using different timeframes. We also found some evidence to suggest that, when eBrAC is zero, negative affect is associated with drinking a greater quantity but that, as eBrAC increases, it is related to drinking less. These results are in need of replication because the interaction was significant, but our simple slopes tests were only marginally significant. Regardless, the moderator analyses suggest that future ambulatory studies of negative affect and alcohol use should take care to consider between-person and within-person moderators. We also did not find any evidence that limb of the BAC curve affected how affect predicted future drinking. It is possible that the nature of drinking outside the lab may have increased the noise in the data. Unlike in the lab where a specific dose is provided over a relatively brief time, in their natural environment, people drink variable amounts over inconsistent time periods. This could suggest that other factors may need to be considered in understanding how the BAC curve affects alcohol consumption.

Limitations and strengths

There are several limitations that should be considered when interpreting our results. First, this sample of participants could be considered at elevated risk for an alcohol use disorder (National Institute on Alcohol Abuse and Alcoholism, 2017) but did not currently have an alcohol use disorder. Therefore, any extension of our results to individuals with more substantial drinking problems should be considered tenuous and in need of empirical testing. Second, although we had adequate power to detect small effects, the number of participants and number of days assessed may have limited the precision of our estimates, as suggested by our relatively wide confidence intervals. Replication in larger samples is needed. Finally, although TAC is the best objective measure of alcohol use, TAC is still an imperfect measure. Unlike BrAC, which has a direct relation with blood alcohol level, TAC has a much more complex relation that can change based on different environmental criteria. For instance, the relation between TAC and BrAC may change at higher doses of alcohol (Karns-Wright et al., 2017); however, the nature of these changes and what constitutes a high a dose has yet to be defined.

Our study also has some strengths that are worth noting. We used a field study, which allowed us to monitor participants’ alcohol use in natural drinking environments. We also combined self-reported ambulatory assessment with an objective measure of alcohol use (cf. Bertz et al., 2018). Our statistical modeling allowed us to look at alcohol use as two distinct processes: drinking occasions and amount consumed. The differing results across each part of the model suggest that this is an important distinction. In addition to these methodological advancements, our study highlights the importance of within-person moderators of the associations between affect and alcohol use. Far from providing a clear set of findings, our mixed results highlight the complexity of these relations and the limitations of the methods used to understand them. Still, these results suggest that momentary negative and positive affect play a role in drinking behaviors and present important directions for future research and theory development.

Acknowledgments

Catharine E. Fairbairn developed the overall study. Konrad Bresin contributed to the study design. Data were collected under the supervision of Catharine E. Fairbairn and Konrad Bresin. Konrad Bresin performed the data analyses and drafted the manuscript. Catharine E. Fairbairn provided critical revisions. Both authors approved of the final version of the article.