Subjective and Behavioral Impulsivity Differentially Moderate Within- and Between-Person Associations Between Physical Activity and Alcohol Consumption

Abstract

Objective:

Evidence indicates a counterintuitive positive relationship between physical activity and alcohol consumption, suggesting that people who engage in more physical activity consume more alcohol. Impulsivity, which has a well-documented role in alcohol use disorders, has been shown to moderate the between-person physical activity–drinking association among emerging adults. However, only a handful of studies have explored within-person associations of physical activity and drinking and potential moderators of this relationship. The current study evaluated the effects of both subjective and behavioral impulsivity on the within- and between-person association between physical activity and alcohol consumption among college students.

Method:

Undergraduate students (N = 250) between ages 18 and 25 years were asked to report their daily physical activity and drinking over 21 days. Physical activity was also recorded objectively through Pacer, a smart-phone app. Subjective impulsivity was assessed using the UPPS-P Impulsive Behavior Scale, and behavioral impulsivity was evaluated using the Balloon Analogue Risk Task.

Results:

Within- and between-subject physical activity–drinking associations were differentially moderated by behavioral impulsivity and self-reported impulsivity. For instance, behavioral impulsivity moderated the within-person association between drinking and self-reported vigorous physical activity, whereas negative urgency moderated the between-person association between drinking and objective physical activity.

Conclusions:

Impulsivity, whether measured subjectively or behaviorally, significantly moderates the physical activity–alcohol consumption association. Importantly, this effect operates differently when predicting variation in behavior within individuals as compared with predicting differences in behavior between individuals.

College drinking exhibits an initial marked increase in the first few years (Sher & Rutledge, 2007), leading to a range of deleterious consequences. The pre-frontal cortex—a brain region largely responsible for self-control, reasoning, and planning—has not completed its development in emerging adults, which puts them at a higher risk of alcohol-related neurobiological impairments (Spear, 2018). Therefore, it is important to address problem drinking early in this population.

In contrast to the negative consequences of alcohol consumption (AC), the health benefits of physical activity (PA) are manifold. Engaging in exercise is associated with widespread neural benefits (West et al., 2019). Young adults who undertake more PA demonstrate better mental health (Chekroud et al., 2018), report better quality sleep (Collings et al., 2015), and have a lower risk of developing obesity (Wareham et al., 2005). However, there appears to be a significant decline in PA during the first few years of university (Corder et al., 2019). Therefore, it is important to understand how PA and AC covary among college students in order to inform health intervention strategies usually aiming to increase PA and decrease alcohol intake.

There is a well-established yet paradoxical relationship between PA and AC, indicating that people who engage in more PA consume more alcohol (Conroy et al., 2015; French et al., 2009; Leichliter et al., 1998; Lisha et al., 2011; Piazza-Gardner & Barry, 2012). This between-person association has been consistently observed among university students (Kokotailo et al., 1996; Leichliter et al., 1998). Of note, Henderson et al. (2020) demonstrated that social motives for drinking and both negative and positive affect significantly moderated this PA-AC relation, suggesting the importance of taking into account key explanatory factors when evaluating the PA-AC association. Indeed, PA and AC appear to be associated psychosocially, in that a number of motives and moderators tend to influence their covariance. For instance, Leasure and Neighbors (2014) demonstrated that PA and AC are more strongly linked in higher sensation seekers.

Impulsivity (IMP) is a heterogeneous construct constituted of multiple dimensions that are important to consider in the framework of health behaviors (Cyders et al., 2007; Dick et al., 2010; Stamates & Lau-Barraco, 2017). Five dimensions of subjective IMP assessed through the UPPS-P Impulsive Behavior Scale (negative urgency [NU], lack of perseverance [PERS], lack of premeditation [PREM], sensation seeking [SS], and positive urgency [PU]; Cyders & Smith, 2007; Cyders et al., 2007) have been developed, validated, and evaluated in the context of health behaviors. NU is an urge to act impulsively in response to negative affect (Whiteside & Lynam, 2001). PREM is characterized as the inclination to act rashly while ignoring the consequences of these actions. PERS is characterized by a low sense of responsibility and conscientiousness and the inability to concentrate on or finish a task (Whiteside & Lynam, 2001). SS is the search for new and thrilling experiences (Whiteside & Lynam, 2001). PU is the tendency to act rashly in response to positive mood (Cyders et al., 2007).

A number of studies have revealed a robust association between subjective IMP and the development of alcohol use disorder (AUD) in young adults (Bø et al., 2016; de Wit, 2009; Di Nicola et al., 2015). Spear (2018) reviewed the effects of adolescent drinking on the brain and behavior and reported that adolescents higher on IMP were more at risk of developing an AUD in later adulthood. Thus, it is crucial to understand how IMP is implicated in the PA-AC relation to better inform intervention strategies for college students.

In addition, the literature often separates subjective IMP (typically measured through self-report questionnaires) from behavioral IMP (typically measured using behavioral tasks; Dick et al., 2010). Findings corroborate weak correlations between subjective and behavioral IMP (Dick et al., 2010; King et al., 2014). To date, only a handful of studies have explored the effects of IMP on the PA-AC relation. Leasure and Neighbors (2014) showed that IMP moderated the between-person association of PA and AC and, specifically, that more highly impulsive people consumed more alcohol and engaged in more PA than their less impulsive counterparts. Conversely, Cho et al. (2020) demonstrated no significant effects of IMP on the within-subject PA-AC association. Of note, the above-mentioned studies have only examined subjective IMP. However, since behavioral IMP also predicts alcohol use among emerging adults (Stamates & Lau-Barraco, 2017), it also merits further examination as a potential moderator of the PA-AC relationship. Furthermore, research on the within-subject association of PA and AC is still in its infancy, with the existing data being inconclusive. Hence, it is important to explore factors that may influence both the within- and between-person analyses to better comprehend the PA-AC association.

The current study evaluated the effects of subjective and behavioral IMP on both the within- and between-person associations of AC and PA among college students. Given that subjective IMP has previously been demonstrated to moderate the between-subject PA-AC association (Leasure & Neighbors, 2014), we hypothesized that (a) subjective IMP will significantly moderate both the between- and within-person PA-AC relation. Furthermore, as the potential moderating effect of behavioral IMP on the PA-AC relation has not been previously examined, we hypothesized that (b) behavioral IMP will significantly moderate both the between-and within-person PA-AC relation.

Method

Procedures

The study was reviewed and approved by the universities’ institutional review boards. Participants provided informed consent and undertook an orientation session before the start of the project. Participants were informed that they would be taking part in a 3-week online diary study, in which they had to report their daily PA and AC and during which time their PA was continuously monitored by a fitness app. A 21-day period is sufficient to model rapid daily variations in PA and AC and not too long a time to lead to higher attrition (Conroy et al., 2015). Participants were also asked to complete baseline and follow-up surveys assessing subjective and behavioral IMP. Data collection took place over three semesters, starting in October 2018 until December 2019 (preceding the COVID-19 outbreak).

Participants

Participants included 250 undergraduate students (74% female) between 18 and 25 years of age (M = 20.3, SD = 1.9) recruited from two Southwestern universities. The study sample was ethnically diverse, with 40% Hispanic/Latino and 60% non-Hispanic, 53% White/Caucasian, 21% Black/African American, 12% other, 7% Asian, 3% Native American/ American Indian, 3% multi-ethnic, and less than 1% Native Hawaiian/Pacific Islander. Exclusion criteria included being less than 18 years or more than 25 years of age, consuming less than one alcoholic beverage per week, and not owning a smartphone. Compensation for eligible participants consisted of 10 extra course credits that served as an incentive for the entire study participation, including downloading the Pacer app and having it recording PA in the background.

Measures

Alcohol consumption. Participants completed a 21-day online diary; at the end of each day, a link containing the diary was sent to them. Participants were invited to report the number and type of standard drinks consumed on the same day the link was sent and on the day before (1 day defined as the hours between 12 a.m. and 11:59 p.m.). Standard measurements of distinct types of alcoholic beverages were described by definitions and images at the beginning of each diary record (adapted from Boynton & Richman, 2014). AC was measured through quantity (total daily servings) and frequency (i.e., drinking/no drinking [yes/no variable]).

Physical activity. PA was assessed through self-reports in a daily diary, as well as objectively. Having both methods of self-report and objective recordings of PA may help circumvent self-report bias (Abrantes et al., 2017) and missing data from forgetting to wear the device (Schuna et al., 2013). Participants were invited to report the time (in minutes) engaging in moderate or vigorous PA, as well as the type of PA. Examples of moderate activities included brisk walking and yoga; examples of vigorous activities included running and swimming. We then calculated the metabolic equivalent (METS) based on the intensity of PA reported. PA intensity was quantified as 3 METS < moderate < 6 METS, and vigorous > 6 METS (Giffuni et al., 2012).

It is typical to examine duration and intensity of exercise separately, as the nature of each variable's relationship with drinking appears to differ (Abrantes et al., 2017; Graupensperger et al., 2020). Objective PA was recorded continuously through Pacer, a fitness smartphone app (Pacer Health, 2022). We assisted participants in downloading the app into their smartphones during orientation. Reported parameters by Pacer are steps taken, distance traveled, calories expended, and total time spent doing PA (Pacer Health, 2022). Smartphone-based assessments represent a valid and accurate measure of objective PA (Höchsmann et al., 2018; Presset et al., 2018). Forgetting or the inability to wear the phone during waking hours (6 A.M.–11 P.M.) was coded as “non-wear” time and characterized by 0 activity counts for 60 or more consecutive minutes (Henderson et al., 2020). Average PA was calculated as the person-specific means across all days centered at the grand mean. Daily PA was centered at the person-specific means.

Impulsivity. Subjective IMP was evaluated using the UPPS-P Impulsive Behavior Scale (Cyders & Smith, 2007; Lynam et al., 2007), consisting of five facets of IMP (NU, PU, PREM, PERS, and SS). Behavioral IMP was evaluated using the original version of the Balloon Analogue Risk Task (BART; Lejuez et al., 2002). Although the BART was initially developed as a risk-taking task, it has also been developed to assess impulsive choice in the laboratory (Weafer et al., 2013). Impulsive choice has been consistently associated with the risk of AUD (King et al., 2014). Consisting of 30 trials, the BART requires participants to press a key that will pump a balloon, and each pump is rewarded with $0.05 (fake money); the more times the balloon is pumped, the more money is earned. The caveat is that the balloon is programmed to pop after a distinct number of pumps, and, if the balloon pops, the participants “lose” the money.

Statistical analyses

Preliminary analysis indicated that the distribution of drinking quantity deviated significantly from normal. Moreover, our drinking quantity data were zero-inflated; therefore, the best model to examine PA-AC associations (for alcohol quantity as the outcome) was a hierarchical linear model with a zero-inflated negative binomial distribution. Drinking frequency was a binary variable, coded “1” if participants drank on a particular day and “0” if participants did not drink.

We used hierarchical linear modeling to fit the data structure with days (i.e., Level 1) nested within people (i.e., Level 2; Snijders & Bosker, 1999) and to examine interaction effects of IMP on between- and within-person PA-AC associations. We used separate models for AC quantity and AC frequency as outcome variables, and for PA variables as predictors. Both subjective and behavioral IMPs were treated as continuous variables (lower to higher levels) and were treated as separate variables in the analyses. Behavioral IMP represented the score obtained on the BART for each participant, and the subscales of subjective IMP coded as SS, PU, NU, PERS, and PREM were scored based on a number of items representing each in the UPPS-P measure. Moreover, given that the PA-AC relation may differ depending on PA intensity (Graupensperger et al., 2020; Leasure & Neighbors, 2014), our models distinguished moderate and vigorous PA. We centered the terms comprising the interaction and multiplied them together (Aiken & West, 1991) in separate data analytical steps for subjective and behavioral IMP.

To examine our first hypothesis (subjective IMP is expected to moderate the between- and within-person PA-AC relation), we interacted the five facets of subjective IMP (SS, PU, NU, PERS, and PREM) with PA. To examine our second hypothesis (behavioral IMP is expected to moderate the between- and within-person PA-AC relation), we interacted BART with PA. Under the assumption that data were missing at random, hierarchical linear modeling handles missing data points for Level 1 variables. Regarding the Level 2 variable (IMP) missing data represented less than 4% of the total data; thus, no action was taken to deal with this missingness.

Results

Descriptive analyses

Correlations, means, standard deviations, and ranges among total/average variables across all days, and IMP variables are displayed in Table 1. Because these correlations do not take into account the multilevel structure of the data, it is advisable that they be interpreted descriptively rather than inferentially (Conroy et al., 2015). Drinking variables were all positively correlated with each other. All daily PA variables were significantly correlated with each other. Gender (female coded 0) was significantly correlated with daily drinking variables and vigorous PA. The daily drinks variable was significantly correlated with moderate PA time and moderate PA METS.

Table 1.

Correlations, means, standard deviations, and ranges among average/total variables across all days and baseline impulsivity variables

Variable	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.	11.	12.	13.	14.	15.
1. Gender	-	.10	.12	.007	.007	.25***	.25***	.13	.29***	.05	-.07	-.08	-.04	.41***	.45***
2. ACquant		-	.67***	.14*	.14*	.07	.07	-.1	.07***	.04*	.05**	.00	.04**	-.07	-.04
3. ACfreq			-	.06	.06	-.03	-.03	-.1	.08***	.04**	.06**	.05	.04**	.07	.11
4. ModPA (time)				-	1.00***	.41***	.41***	-.03	.05	.15*	.10	.05	.02	.15	.14
5. ModPA (METS)					-	.41***	.41***	-.03	.05	.15*	.10	.05	.02	.15	.14
6. VigPA (time)						-	1.00***	.004	.02	.11	-.02	-.02	.05	.33**	.29**
7. VigPA (METS)							-	.004	.02	.11	-.02	-.02	.05	.33**	.29**
8. BART								-	.14*	-.02	-.06	-.11	-.14	-.12	-.09
9. SS									-	.28***	.17**	-.10	.19*	.16	.26**
10. PU										-	.67***	.34***	.42***	.04	.06
11. NU											-	.39***	.42***	-.01	.00
12. PERS												-	.48***	.04	.02
13. PREM													-	-.03	-.00
14. ObjPA (time)														-	. 86***
15. ObjPA (METS)															-
M	0.26	12.91	6.00	54.22	162.65	16.67	100.03	24.10	2.67	1.82	2.21	1.91	1.83	53.68	112.22
SD	0.44	12.48	4.37	87.68	192.77	41.97	164.50	12.81	0.67	0.61	0.61	0.47	0.48	24.82	47.08
Minimum	0.00	0.00	0.00	0.00	0.00	0.00	0.00	1.73	1.17	1.00	1.00	1.00	1.00	3.2	30.23
Maximum	1.00	76.50	20.00	602.86	1,809.00	163.85	983.08	62.00	4.00	3.86	3.83	3.80	3.91	134.19	300.42

Notes: ACquant = sum of total daily drinks; ACfreq = drinking frequency across all days; ModPA (time) = person-specific mean of moderate physical activity time; ModPA (METS) = person-specific mean of moderate PA metabolic equivalent minutes; VigPA (time) = person-specific mean of vigorous PA time; VigPA (METS) = person-specific mean of moderate PA MET minutes; BART = behavioral impulsivity reported at baseline; SS = sensation seeking; PU = positive urgency; NU = negative urgency; PERS = lack of perseverance; PREM = lack of premeditation; ObjPA (time) = person-specific mean of objective PA minutes; ObjPA (METS) = person-specific mean of objective PA METS.

^*p < .05;

^**p < .01;

^***p < .0001.

Within- and between-person analyses

Negative binomial multilevel model coefficients examining within- and between-person associations between PA and AC are depicted in Tables 2–4. The between-subject variance in average drinking quantity was smaller in the final adjusted interaction effects models (${\sigma }_{\mathrm{u0}}^2$ = .37) compared with the between-subject variance in average drinking quantity in the unconditional model (${\sigma }_{\mathrm{u0}}^2$ = .56), indicating that the predictor variables and interactions helped account for some of the variability in average drinking quantity. The between-person variance in drinking frequency was smaller in the final adjusted models (${\sigma }_{\mathrm{u0}}^2$ = .88 compared with ${\sigma }_{\mathrm{u0}}^2$ = 1.07 in the unconditional model), indicating that the predictor variables helped explain some of the variability in drinking frequency.

Table 2.

Negative binomial multilevel model coefficients examining between- and within-person associations between physical activity (PA; self-reported in diary) and drinking quantity

Parameters	Unadjusted model				Adjusted model
	γ	SE	t	P	γ	SE	t	P
Diary PA time model
Intercept	-0.54***	0.06	-8.85	<.0001	-1.07***	0.08	-13.42	<.0001
Average moderate PA	0.002	0.001	1.63	.10	0.002	0.0009	1.75	.08
Average vigorous PA	-0.0005	0.002	-0.21	.84	-0.001	0.002	-0.63	.53
Daily moderate PA	-0.0003	0.0006	-0.48	.63	-0.001	0.0006	-0.89	.37
Daily vigorous PA	-0.002	0.001	-1.31	.19	-0.001	0.001	-0.72	.47
Gender					0.17	0.14	1.24	.22
PU × Daily Moderate PA					0.004*	0.001	2.48	.014
Diary PA METS model
Intercept	-0.54***	0.06	-8.85	<.0001	-1.07***	0.08	-13.42	<.0001
Average moderate PA	0.0005	0.0003	1.63	.10	0.0005	0.0003	1.75	.08
Average vigorous PA	-0.00008	0.0004	-0.21	.84	-0.0002	0.0004	-0.63	.53
Daily moderate PA	-0.0001	0.0002	-0.48	.63	-0.0002	0.0002	-0.89	.37
Daily vigorous PA	-0.0003	0.0002	-1.31	.19	-0.0001	0.0002	-0.72	.47
Gender					0.17	0.14	1.24	.22
PU × Daily Moderate PA					0.001*	0.0005	2.49	.014

Notes: PU × daily moderate PA = the interaction effect between positive urgency and daily moderate physical activity on drinking quantity; METS = metabolic equivalent; PU = positive urgency.

^*p < .05;

^***p < .0001.

Table 4.

Negative binomial multilevel model coefficients examining between- and within-person associations between physical activity (PA; self-reported in diary) and drinking frequency

Parameters	Unadjusted model				Adjusted model
	γ	SE	t	p	γ	SE	t	p
Diary PA time model
Intercept	-0.69*	0.33	-2.09	.04	-1.18**	0.35	-3.33	.001
Average moderate PA	0.001	0.001	1.12	.26	0.0001	0.001	1.01	.31
Average vigorous PA	-0.003	0.003	-0.93	.35	-0.005	0.003	-1.63	.10
Daily moderate PA	-0.0005	0.0006	-0.80	.43	-0.0006	0.0006	-1.00	.32
Daily vigorous PA	-0.0006	0.001	-0.53	.60	-0.0003	0.001	-0.26	.79
Gender					0.35*	0.17	2.05	.04
PU × Daily Moderate PA					0.003*	0.001	2.02	.04
BART × Daily Vigorous PA					-0.0003	0.0001	-2.32	.02
Diary PA METS model
Intercept	-0.69*	0.33	-2.09	.04	-1.18**	0.35	-3.33	.001
Average moderate PA	0.0005	0.0004	1.12	.26	0.0004	0.0004	1.01	.31
Average vigorous PA	-0.0005	0.0005	-0.93	.35	-0.0008	0.0005	-1.63	.10
Daily moderate PA	-0.0002	0.0002	-0.80	.43	-0.0002	0.0002	-1.0	.31
Daily vigorous PA	-0.0001	0.0002	-0.53	.60	-0.00005	0.0002	-0.26	.79
Gender					0.35*	0.17	2.05	.04
PU × Daily Moderate PA					0.0009*	0.0004	2.02	.04
BART × Daily Vigorous PA					-0.00005*	0.00002	-2.32	.02

Notes: PU × daily moderate PA = the interaction effect of positive urgency and daily moderate PA on drinking frequency; BART × daily vigorous PA = the interaction effect of the behavioral impulsivity reported at baseline and daily vigorous PA on drinking frequency; METS = metabolic equivalent; PU = positive urgency; BART = Balloon Analogue Risk Task.

^*p < .05;

^**p < .01.

Table 3.

Negative binomial multilevel model coefficients examining between- and within-person associations between physical activity (PA; objectively recorded) and drinking quantity

Parameters	Unadjusted model				Adjusted model
	γ	SE	t	p	γ	SE	t	p
Objective PA time model
Intercept	-0.59***	0.09	-6.93	<.0001	-1.15***	0.11	-10.23	<.0001
Average objective PA	-0.004	0.003	-1.11	.27	-0.006	0.004	-1.61	.11
Daily objective PA	-0.00001	0.002	-0.01	.99	0.0001	0.002	0.65	.51
Gender					0.61**	0.19	3.91	.002
PERS × Daily Objective PA					-0.004*	0.002	-2.10	.038
Objective PA METS model
Intercept	-0.57***	0.09	-6.55	<.0001	-1.13***	0.11	-9.66	<.0001
Average objective PA	-0.002	0.002	-1.06	.29	-0.003	0.002	-1.68	.09
Daily objective PA	-0.001	0.001	-1.33	.18	-0.0004	0.001	-0.40	.69
Gender					0.56**	0.20	2.78	.006
NU × Average Objective PA					-0.005*	0.003	-1.98	.05

Notes: PERS × daily objective PA = the interaction effect of lack of perseverance and daily time spent in physical activity (recorded objectively) on drinking quantity; METS = metabolic equivalent; NU × average objective PA = the interaction effect of negative urgency and average intensity of PA on drinking quantity; PERS = lack of perseverance; NU = negative urgency.

^*p < .05;

^**p < .01;

^***p < .0001.

Subjective impulsivity as a moderator

Interaction effects between the five dimensions of subjective IMP and PA revealed a significant interaction between NU and average objective PA METS on drinking quantity (γ = -0.009, SE = 0.004, t = -2.17, p = .03), indicating a moderating influence of NU on the between-person relation between drinking quantity and PA intensity (recorded objectively). The interaction plot depicting interaction effects of levels of NU on the relation between PA METS and quantity of drinking shows that on average, for individuals higher on NU, alcohol intake decreases as PA intensity increases (Figure 1).

Figure 1.

Quantity of drinking as a function of the interaction between average objective physical activity metabolic equivalent (PA METS) and negative urgency (NU)

Cross-level interactions between subjective IMP and self-reported PA revealed that, of the five dimensions of IMP, only PU exhibited significant interactions with self-reported daily moderate PA time (γ = 0.002, SE = 0.0008, t = 2.47, p = .01) and METS (γ = 0.0007, SE = 0.0003, t = 2.47, p = .01) on drinking quantity as well as on drinking frequency (self-reported daily moderate PA time: γ = 0.003, SE = 0.001, t = 2.02, p = .043; and METS: γ = 0.0008, SE = 0.0004, t = 2.02, p = .043), suggesting a moderating influence of PU on the within-person PA-AC relation. Interaction plots revealed that for individuals higher in PU, as daily moderate PA time and METS increased, quantity of drinking also increased; and for lower levels of PU, as daily moderate PA time and METS increased, drinking quantity decreased (Figures 2 and 3).

Figure 2.

Quantity of drinking as a function of the interaction between daily moderate physical activity (PA) time and positive urgency (PU)

Figure 3.

Quantity of drinking as a function of the interaction between daily moderate physical activity metabolic equivalent (PA METS) and positive urgency (PU)

The same results applied to drinking frequency as the outcome variable. Interaction plots revealed that for individuals higher in PU, as daily moderate PA time and METS increased, frequency of drinking also increased and vice versa. A significant interaction effect was also shown between PERS and time spent in objective PA (γ = -0.004, SE = 0.002, t = -2.10, p = .038), and simple slopes revealed that on days when time spent in PA increased, AC decreased for participants higher on PERS. Conversely, individuals lower on PERS consumed more alcohol on days when they engaged in more PA.

Behavioral impulsivity as a moderator

Results from cross-level interactions showed that behavioral IMP (BART) had significant moderating effects on the within-person relation between drinking frequency and self-reported vigorous PA time (γ = -0.0003, SE = 0.0001, t = -2.32, p = .02) and METS (γ = -0.00005, SE = 0.00002, t = -2.32, p = .02). Interaction plots revealed that for individuals higher in behavioral IMP, as daily vigorous PA time and METS increased, AC decreased; and for lower levels of behavioral IMP, as daily vigorous PA time and METS increased, drinking frequency also increased.

Discussion

The present study assessed whether subjective and behavioral IMP had a moderating influence on the within- and between-person PA-AC associations across 21 days. Our findings indicated differential moderating effects of behavioral IMP and facets of subjective IMP on the PA-AC association, contingent upon drinking quantity or frequency, on the measurement of PA, and whether these relations were examined at the between- or within-person level. We showed consistent patterns of moderating influences of PU on the relationship between drinking quantity and moderate PA as well as drinking frequency and moderate PA. Since the hallmark of PU is that it combines IMP and positive affect (Cyders et al., 2007), it may be that those individuals who scored higher on PU engaged in more PA and consumed more alcohol because both activities inherently feel good and provide immediate rewards. These results are consistent with Leasure and Neighbors (2014), who reported that the PU dimension of subjective IMP significantly moderated the association between moderate intensity PA and AC at between-person levels. Furthermore, although prior longitudinal research has demonstrated that PU was the strongest predictor of drinking quantity among students in college (Settles et al., 2010), the current results are the first to indicate a significant moderating influence of PU on the within-subject relation between moderate AC and PA.

Our results also revealed a moderating effect of NU on the between-person association between drinking quantity and vigorous PA measured objectively in Pacer. Interaction plots showed that participants higher on NU consumed less alcohol when they engaged in higher PA intensity levels. In line with these findings, but contrary to our second hypothesis, we found that participants who scored higher on behavioral IMP decreased their drinking on days they increased their vigorous PA. One possible explanation for these individuals higher on NU and behavioral IMP is that more intense PA may have acted as a substitute for drinking. Kotbagi et al. (2017) showed that NU was robustly associated with extreme maladaptive exercise, often resulting in health impairments. Because NU is characterized by the urge to behave rashly in response to negative affect (Cyders & Smith, 2007; Cyders et al., 2007), Kotbagi et al. postulated that individuals higher on NU engaged in excessive levels of high-intensity exercise to alleviate negative emotion. Another potential explanation for these results is the hedonic substitution of drinking with PA. Both exercise and alcohol have rewarding properties since they share common neurobiological mechanisms of reward by activating the production of and increasing dopamine and endogenous opioids (Bardo & Compton, 2015; Lynch et al., 2013).

PERS was also demonstrated to moderate the within-person relation between drinking quantity and PA time measured objectively in Pacer. Scoring higher on PERS indicated a decline in daily AC as time spent engaging in PA increased. Smith and colleagues (2021) reported that those greater on PERS engage in less vigorous PA; therefore, it would be counterintuitive to speculate that those higher on PERS replaced AC, a health risk behavior, with PA, a health-promoting behavior. Of note, unlike with NU, the moderation effect by PERS was on the negative relation between AC and time spent in PA recorded objectively in Pacer, but not intensity of PA. Thus, it is possible that these participants engaged in only light activities (e.g., walking), which generally require minimal effort and commitment.

There is ample evidence suggesting that more impulsive people are at a higher risk of developing AUD (Gowin et al., 2017), and the current study demonstrated that IMP moderates the association between PA and AC. Therefore, interventions for AUD that focus on using exercise regimens (Giesen et al., 2015; Weinstock et al., 2016) may consider that if PA and AC are positively associated in individuals exhibiting higher levels of IMP, it may be counterproductive to use exercise as a treatment method for AUD. In addition, given the heterogeneous nature of IMP, being aware of whether subjective and/or behavioral IMP moderate the PA-AC relationship may provide important information on suitable tasks to administer for patients assigned to undergo an exercise program to help reduce drinking. For instance, Stamates and Lau-Barraco (2017) corroborated that among young adults, PU and NU were more tightly associated with alcohol-related problems. It is, then, significant to tease apart the distinct effects of IMP's multiple dimensions when examining PA-AC associations. Our nuanced findings on facets of IMP, PU, NU, and PERS are important in ensuring optimal health and identifying the most vulnerable students on campus.

Some noteworthy limitations of the current project are regarding the validity and generalizability of our findings. We recognize that one limitation involved our low drinking and predominantly female (74%) sample, which may limit generalizability to young men in other colleges. Therefore, future work may wish to focus on more varied student populations. Unlike prior studies using self-report measures requiring retrospective recall to collect PA and AC data, here we used an objective measure of PA, which may have circumvented self-report bias and inaccurate data due to retrospective recall. Furthermore, evidence suggests that smartphone-based measures of PA are sensitive enough to distinguish actual PA from noise (Höchsmann et al., 2018; Presset et al., 2018).

To add to the current literature, future studies may want to use a mixed methods design, whereby participants reporting high rates of PA and AC are then interviewed and asked about why they believe they engage in both health behaviors. Linking diverse socioeconomic backgrounds to the relationship between PA and AC may also be of future interest, as health behaviors are strongly associated with socioeconomic status (Petrovic et al., 2018). Furthermore, subsequent studies should explore whether the current findings are consistent across distinct socioeconomic scenarios.

Taken together, our findings highlight differences in the moderating effects of subjective versus behavioral IMP on the PA-AC association, and these effects operate differently when predicting variation in behavior at within-person levels compared with the between-person levels. Because the direct effects of PA on AC are not always observed, it is important to probe other influences (i.e., IMP) on the PA-AC relationship, which was the objective of the current research. Our results may provide additional groundwork for developing health guidelines and suitable multibehavior therapies when a health-promoting behavior (i.e., exercise) and a health-jeopardizing behavior (i.e., alcohol misuse) systematically covary among emerging adults in college. It is advisable that regimens intended to integrate PA in AUD treatment identify methods in which the inclusion of PA does not concurrently raise drinking.