Daily- and Person-Level Associations Between Physical Activity and Alcohol Use Among College Students

Abstract

Emerging adults, particularly university students, who are physically active, drink more than their less physically active peers. We extended this between-person relationship to the within-person level of analysis, by examining whether students are more likely to drink on days when they exercise, and whether this within-person association remains after controlling for potential confounding factors. We also explored the temporal sequence of the physical activity (PA)-alcohol use association. University students (N = 426) completed a 30-day online diary. The small positive within-person association between PA and alcohol use was not retained after controlling for day of the week. However, previous day’s drinking was inversely associated with next day’s PA on weekdays. These findings suggest that the previously reported positive PA-alcohol association does not necessarily align with the within-person daily association. Future studies with more nuanced measurement strategies, such as ecological momentary assessment, are needed to better understand the association between PA and alcohol use.

Physically active people are more likely to engage in other health promoting behaviors (e.g., healthy eating) and less likely to engage in risky health behaviors (e.g., smoking; Dodd, Al-Nakeeb, Nevill, & Forshaw, 2010; Lippke, Nigg, & Maddock, 2012). Despite the overall inverse relationship between physical activity (PA) and risky health behaviors, the association between PA and alcohol use is an exception. Individuals who are physically active tend to drink more across the life span, including college students (Piazza-Gardner & Barry, 2012), a subpopulation of emerging adults (Arnett, 2016).

Emerging adults are an especially vulnerable population for alcohol problems, as alcohol intake peaks among individuals aged 18–25 years (Center for Behavioral Health Statistics and Quality, 2016), and a relatively large proportion of emerging adults’ deaths are attributed to alcohol (World Health Organization, 2015). Although intuitively appealing, efforts to increase PA as a way of reducing alcohol-related problems among emerging adults could be counterproductive, given consistent evidence showing that PA is positively related to alcohol use. For example, Dodd and colleagues (2010) differentiated university students’ multiple health behaviors into various clusters including a “low-risk” cluster characterized by low psychological stress, high fruit and vegetable intake, moderate alcohol intake, and high PA. A separate study of college students identified a “moderately healthy” cluster of college students who are unlikely to smoke and use illicit drugs, and more likely to eat fruits and vegetables, be physically active but also are highly likely to binge drink (i.e., ≥5 drinks in one setting; Kwan, Arbour-Nicitopoulos, Duku, & Faulkner, 2016).

This positive association between alcohol use and PA might be explained by shared biological bases—both alcohol and PA can activate reward mechanisms in the brain and regulate stress responses (Leasure, Neighbors, Henderson, & Young, 2015)—or social/interpersonal motivations—PA and drinking often occur as social activities (Cooper, 1994; Ryan, Frederick, Lepes, Rubio, & Sheldon, 1997). Furthermore, personality traits such as extroversion and impulsivity have been proposed to explain the association (Leasure et al., 2015); extroverted people, who are socially motivated, are more likely to engage in PA (Rhodes & Smith, 2006) and drink alcohol (Raynor & Levine, 2009). Also, exercising drinkers might be inclined to seek sensation-evoking activities, as drinking and PA can be rewarding stimuli (Leasure et al., 2015).

However, most previous studies examining PA and alcohol use have relied on cross-sectional designs to examine the relationship between recalled drinking (e.g., “How many times did you drink in the past month?”) and recalled PA (e.g., “How many minutes did you engage in PA in the past week?”). Such retrospective reports are vulnerable to recall error and bias (Tennen, Affleck, Armeli, & Carney, 2000), a particular concern for the study of PA and alcohol intake, which occur frequently among college students and with day-to-day variation in intensity (Conroy, Elavsky, Doerksen, & Maher, 2013; O’Grady, Cullum, Tennen, & Armeli, 2011). Also, the previously reported positive between-person association between PA and alcohol use does not necessarily inform us about within-person daily association, as it may reflect an artifact of aggregating temporally discontinuous behaviors (e.g., students may engage in PA on weekdays but drink mostly on weekends; Conroy et al., 2015).

Microlongitudinal (daily diary) study designs limit recall error and bias while capturing behavior close to its real-time occurrence. To date, very few studies have examined the within-person association between PA and alcohol use, and results have been mixed. The adult participants in Conroy and colleagues’ (2015) study completed three 21-day measurement bursts at the end of day. Individuals in this study drank more than usual on days during which they engaged in more PA than usual. However, in a 14-day diary study of first-year college students, there was no main effect of daily PA on daily alcohol use, although there was an interaction effect: First-year students drank less than usual on weekends during which they engaged in more PA than usual (Finlay, Ram, Maggs, & Caldwell, 2012).

The current study is a secondary analysis of a larger investigation designed to understand emerging adults’ alcohol use. A microlongitudinal design was used to examine the day-level within-person association between PA and alcohol use among college students. We predicted a positive association between PA and drinking and examined whether this association was retained after controlling for a possible confounder, day of the week. In addition, we investigated whether the PA-drinking association at the daily within-person level is moderated by gender and personality traits. As men consume more alcohol (O’Malley & Johnston, 2002) and engage in more PA than women (Troiano et al., 2008), and extroverted and impulsive people are more likely to engage in PA (Leasure et al., 2015; Raynor & Levine, 2009; Rhodes & Smith, 2006) and drink alcohol (Raynor & Levine, 2009), we predicted that the PA-drinking association would be stronger among men and those who are more extroverted and impulsive.

Finally, we wanted to extend previous findings in an exploratory manner by examining the temporal sequence of the within-person association between PA and alcohol use. For example, students may drink after participating in team sports, or they may exercise after drinking to compensate for calorie intake from drinking. Thus, understanding the temporal relationship between PA and drinking may be helpful in developing interventions to decrease alcohol consumption or increase PA in college students.

Method

Participants and Procedure

Undergraduate students (N = 575) from a large northeastern university were asked to complete an online baseline survey including demographics and personality traits followed approximately 2 weeks later by a 30-day daily diary on a secure website. They completed the daily survey (about 5 min) between 2:30 p.m. and 7:00 p.m. each day to reduce variation in reporting times and to coincide with the end of the school day. This research was approved by the university institutional review board. Informed consent was obtained from all participants.

We excluded data from five participants who did not complete a baseline questionnaire and 65 students who did not complete at least 15 daily surveys (i.e., half of the surveys). Seventy-nine participants who reported either no alcohol use or no PA during the 30 days were more likely to be women (p = .019) and reported lower impulsivity (p = .034). They were excluded from the analysis to avoid the biases that may be introduced by examining the within-person association for individuals who had no variability in PA and alcohol use during the daily surveys. Thus, the final sample included 426 students. In the final sample, the number of daily surveys completed was unrelated to neuroticism (r = −.00, p = .939) and impulsivity (r = −.01, p = .856). However, women completed more daily surveys than men, t(424) = 2.84, p = .005. Mean age was 18.80 years (SD = 1.15) and about half were women (50.2%) and freshman (56.6%). Most were non-Hispanic Whites (88.3%) followed by Asians/Pacific Islanders (5.9%), Blacks (2.8%), Hispanics (2.3%), and Other (0.7%).

Measures

Daily-level measures included alcohol use and PA. First, participants were asked to report last night’s activities defined as the period after taking the previous day’s survey (or after 6 p.m. yesterday) until going to sleep. Then, they were asked to report today’s activities defined as from the period that they woke up until the time they were completing that day’s diary entry. Previous night’s PA was measured with 1 item, “How many hours did you exercise or play sports last night?” from 0 to 12+. Today’s PA was measured with 2 items: exercised (yes/no) and played sports (yes/no). If students positively endorsed at least 1 of the 2 items, they were coded yes. Participants reported how many drinks they consumed using a 17-point scale from 0 to 15+. A standard drink was defined as one 12-oz. can or bottle of beer or wine cooler, one 5-oz. glass of wine, or a 1-oz. measure of liquor straight or in a mixed drink. Drinking and PA responses were dichotomized to investigate whether students are more likely to drink on days when they exercise. Finally, day of the week was dichotomized based on a previous study (Finlay et al., 2012): social weekend (Thursdays to Saturdays) and weekday (Sundays to Wednesdays).

Person-level PA was calculated as the number of PA days during the 30-day diary period. Also, gender and two personality traits, extroversion and impulsivity, were measured with the Revised NEO Personality InventoryV? (Costa & McCrae, 1992) at baseline. Participants responded to extroversion (12 items) and impulsivity (8 items), on a scale ranging from 1 (strongly disagree) to 7 (strongly agree). Cronbach’s α was .80 for extroversion and .63 for impulsivity.

Analytic Plan

We conducted multilevel logistic regressions using the full maximum likelihood method. Specifically, we examined the association between daily PA and daily alcohol use either that day or night (i.e., combined daytime and nighttime drinking to cover the entire day).

First, we tested an unadjusted (random intercept) model including the day-level PA and the person-level PA (grand-mean centered) to predict daily alcohol use. Then, we tested an adjusted (random slope) model, including day of the week at the day level. In this model, we also included the three between-person factors—gender, extroversion (grand-mean centered), and impulsivity (grand-mean centered)—to explain between-person variability in alcohol use (i.e., predicting the intercepts), and to model the cross-level interactions in which the association between daily PA and drinking is moderated by gender and personality traits. Changes in deviances (log-likelihood [−2LL] differences via χ² tests), Akaike information criteria, and Bayesian information criterion were calculated to examine model fit.

We also examined the following associations based on the same analytic plan: (1) temporal co-occurrence—(a) the association between daytime PA and daytime alcohol use and (b) between nighttime PA and nighttime alcohol use; (2) within-day temporal sequence—(a) the association between daytime PA and nighttime alcohol use and (b) between daytime alcohol use and nighttime PA; (3) the next-day lagged association—(a) between previous day’s PA and next day’s alcohol use and (b) between previous day’s alcohol use and next day’s PA.

On average, participants completed a mean of 28 days (SD = 2.22, range = 18–30; approximately 93% of the total days). In addition, nine individuals were omitted from analyses because of missing person-level data. Missing data were deleted listwise. Students who were excluded in the final analysis were more likely to be women (p = .014) and reported lower extroversion (p = .006) and lower impulsivity (p = .034) compared with those who were included.

Results

Descriptive statistics regarding personality traits, alcohol use, and PA are presented in Table 1. The PA-drinking association over the full day,¹ that is, any daytime or nighttime reports, is presented in Table 2. We found a positive association between daily PA and drinking in the unadjusted model, but it was no longer significant in the adjusted model that controlled for day of the week. Person-level PA was associated with daily drinking in neither the unadjusted nor adjusted model. There were no interaction effects between daily PA and day of the week. Also, no cross-level interaction was observed (see Table 2).

Table 1.

Participants’ Characteristics.

Variable	M (SD)	Range
Extroversion	59.43 (9.16)	31–84
Impulsivity	33.62 (6.54)	17–55
Alcohol usea	6.22 (4.16)	3–28
Daytime alcohol use	0.72 (2.01)	0–27
Nighttime alcohol use	5.69 (3.83)	0–28
PAb	11.64 (7.46)	1–29
Daytime PA	7.25 (6.52)	0–26
Nighttime PA	9.20 (6.81)	0–29
Number of drinks during daytime	0.14 (1.01)	0–16
Number of drinks during nighttime	1.42 (3.21)	0–16
Number of total drinks per day	1.47 (3.54)	0–32

^aAlcohol use = mean number of drinking days (combining daytime and nighttime reports) during the 30 diary days.

^bPhysical activity (PA) = mean number of PA days (combining daytime and nighttime reports) during the 30 diary days.

Table 2.

Multilevel Logistic Regression Predicting Alcohol Use From Physical Activity (Full Day).

Parameter	Unadjusted Model			Adjusted Model
	OR	95% CI	SE	OR	95% CI	SE
Intercept	0.26***	[0.23, 0.29]	.02	0.06***	[0.04, 0.07]	.01
Day-level PA	1.19*	[1.04, 1.35]	.08	1.04	[0.79, 1.39]	.15
Person-level PA	1.01	[0.99, 1.02]	.01	0.99	[0.98, 1.01]	.01
Day of the week				9.60***	[7.90, 11.66]	.95
Gender				2.31***	[1.74, 3.07]	.34
Extroversion				1.01	[0.99, 1.02]	.01
Impulsivity				1.02	[0.99, 1.04]	.01
Daily PA × Day of the Week				1.25	[0.96, 1.62]	.17
Daily PA × Gender				0.87	[0.64, 1.19]	.14
Daily PA × Extroversion				0.99	[0.98, 1.01]	.01
Daily PA × Impulsivity				1.00	[0.98, 1.03]	.01
Random effect (estimate)
Intercept	0.71	[0.58, 0.88]	.08	0.90	[0.70, 1.15]	.11
Physical activity				0.25	[0.09, 0.70]	.13
−2LL	9,270.08			7,486.16
AIC	9,274.04			7,512.16
BIC	9,309.43			7,603.92

Note. Gender: 0 = female; 1 = male. PA = physical activity. Day of the week: 0 = weekday (Sunday to Wednesday); 1 = weekend (Thursday to Saturday); LL = log-likelihood; AIC = Akaike information criteria; BIC = Bayesian information criterion; OR = odds ratio; CI = confidence interval.

We conducted an additional analysis with binge drinking (≥4 drinks per day for women and ≥5 drinks per day for men) as an outcome; neither daily PA (odds ratio [OR] = 0.99, 95% confidence interval [CI] = [0.68, 1.43]) nor person-level PA (OR = 1.00, 95% CI [0.99, 1.02]) was associated with daily binge drinking in the adjusted model. There was a cross-level interaction between PA and gender (OR = 1.37, 95% CI [1.00, 1.88]). However, subgroup analysis showed that daily PA was associated with binge drinking in neither men (OR = 1.19, 95% CI [0.78, 1.81]) nor women (OR = 1.23, 95% CI [0.76, 2.00]).

Although we observed an inverse association between the day-level nighttime PA and nighttime drinking in the unadjusted model (OR = 0.86, 95% CI [0.77, 0.98]), it was not significant in the adjusted model (OR = 0.98, 95% CI [0.74,1.30]). The positive association between person-level nighttime PA and nighttime drinking in the unadjusted model (OR = 1.02, 95% CI [1.00, 1.03]) was not significant in the adjusted model (OR = 1.00, 95% CI [0.98, 1.01]). No significant cross-level interaction effects were observed (all ps > .05). Day-level daytime PA and daytime drinking were not associated in either the unadjusted (OR = 0.88, 95% CI [0.62, 1.23]) or adjusted model (OR = 0.60, 95% CI [0.22, 1.59]). The association between person-level daytime PA and daytime drinking was significant in neither the unadjusted (OR = 1.02, 95% CI [0.98, 1.06]) nor adjusted model (OR = 1.00, 95% CI [0.96, 1.04]). No significant cross-level interaction effects were observed (all ps > .05).

Regarding the within-day temporal sequence of PA and alcohol use, neither the unadjusted (OR = 1.10, 95% CI [0.95, 1.26]) nor adjusted (OR = 1.09, 95 % CI [0.85, 1.41]) model was consistent with an association between the day-level daytime PA and nighttime drinking. Likewise, no association between person-level daytime PA and nighttime drinking emerged in the unadjusted (OR = 1.00, 95% CI [0.98, 1.01]) or adjusted model (OR = 0.98, 95% CI [0.96, 1.00]). No significant cross-level interaction effects were found (all ps > .05). Day-level daytime drinking was not associated with nighttime PA in either the unadjusted (OR = 0.97, 95% CI [0.68, 1.39]) or adjusted model (OR = 0.75, 95% CI [0.30, 1.88]). However, person-level daytime drinking was positively associated with day-level nighttime PA in both the unadjusted (OR = 1.32, 95% CI [1.18, 1.47]) and adjusted models (OR = 1.23, 95% CI [1.10, 1.39]). No significant cross-level interaction effects were found (all ps > .05).

With respect to the next-day lagged association, day-level previous day’s PA was positively associated with next day’s drinking in the unadjusted model (OR = 1.20, 95% CI [1.06, 1.38]), but it was no longer significant in the adjusted model (OR = 0.79, 95% CI [0.59, 1.06]). Person-level previous day’s PA was associated with next day’s drinking in neither the unadjusted (OR = 1.00, 95% CI [0.98, 1.01]) nor adjusted model (OR = 0.99, 95% CI [0.97, 1.01]). No significant cross-level interaction effects were found (all ps >.05). Finally, day-level previous day’s drinking was negatively associated with next day’s PA (OR = 0.65, 95% CI [0.57, 0.74]) in the unadjusted model and it remained significant in the adjusted model (OR = 0.53, 95% CI [0.41, 0.69]). Person-level previous day’s drinking was positively associated with next day’s PA in the unadjusted model (OR = 1.06, 95% CI [1.02, 1.11]), but not in the adjusted model (OR = 1.01, 95% CI [0.96, 1.06]). There was a significant interaction between day-level drinking and day of the week (OR = 1.54, 95% CI [1.19, 1.98]). Subgroup analysis showed that previous day’s daily drinking was not associated with next day’s PA on social weekends (OR = 0.83, 95% CI [0.62, 1.11]), but it was negatively associated with next day’s PA on weekdays (OR = 0.54, 95% CI [0.40, 0.74]).

Discussion

The aim of the present study was to extend previous reports of a positive PA-drinking association in studies that examined the between-person level of analysis, by examining whether college students are more likely to drink on days when they exercise, and whether this within-person association is retained after controlling for potential confounding factors. We also explored the temporal sequence of the PA-alcohol use association. We observed that the positive association between daily PA and daily drinking was not retained when day of the week was taken into account, indicating that the effect of daily PA on daily drinking may be biased in the unadjusted model, partly because the effect of day of the week was attributed to PA. The absence of a main effect is consistent with the results in a previous study (Finlay et al., 2012), whereas we also found no interaction between daily PA and day of the week. Further, with regard to personality traits, there was no main effect of extroversion or impulsivity, and the PA-drinking association was not stronger for those who were more extroverted or impulsive, suggesting that explanations other than shared personality traits may be required to address any observed association between PA and alcohol use (Leasure et al., 2015). Taken together, our findings suggest that the previously reported positive between-person association between PA and alcohol use does not necessarily align with the within-person daily association.

However, our null findings may be due, in part, to our measurement approach. Our PA measures did not assess PA intensity and PA was coded as a dichotomous variable. Barry and Piazza-Gardner (2012) found that among college students, vigorous-intensity PA and strength training were positively associated with binge drinking, whereas moderate-intensity PA was negatively associated with binge drinking. Our study investigated whether college students were more likely to drink on days during which they exercised. We did not examine whether students drank more than usual on days during which they engaged in more PA than usual, which was investigated in a previous study (Conroy et al., 2015). Thus, this measurement limitation may have contributed to our failure to replicate some previous findings linking PA and alcohol use.

In our exploratory investigation of the temporal sequencing of PA and alcohol use, we found that students were less likely to engage in PA if they drank on the previous day. We speculate that they needed to sacrifice their exercise time to recover from drinking and to finish the previous day’s incomplete work due to drinking. The finding that previous day’s daily drinking was negatively associated with next day’s PA on weekdays, but not weekends, further supports this reasoning. Although a wide range of negative consequences of alcohol misuse among college students is well-documented (Perkins, 2002), this finding adds to the literature by documenting that alcohol use (when not necessarily misuse) may also predict a proximal negative consequence (no exercise) in students’ daily lives. However, note that weekday drinking is less common among college students (Woodyard & Hallam, 2010): In our sample, approximately 61% of drinking days were weekend days. Thus, the inverse association significant only during weekdays might also be due to different motivations between weekday drinking and weekend drinking, such that weekday drinking is associated with tension reduction (“feel calm”; Lau-Barraco, Braitman, Linden-Carmichael, & Stamates, 2016).

We also found a positive association between person-level alcohol use and PA: Students who reported more daytime drinking on average than others were more likely to engage in nighttime PA than others. This person-level finding is consistent with previous reports of a positive PA-drinking association in studies that have examined the between-person level of analysis. The finding that the person-level PA and drinking association was positive, whereas the day-level PA and drinking association was negative necessitates investigation across multiple scales nested in time simultaneously for a better understanding the PA-drinking relationship. Likewise, as drinking usually occurred at nighttime (only 2.9% of drinking days involved daytime drinking), we speculate that students who are more likely to drink during daytime than others might have different motivations for drinking, which might have influenced the positive association between-person level daytime drinking and nighttime PA. Future studies should employ more nuanced measurement strategies (e.g., ecological momentary assessment) to assess PA and alcohol use and their motivations in real time.

The present study has several limitations. First, our PA measures did not assess PA intensity and we analyzed PA and drinking as dichotomous variables. Also, the participants’ different conceptualizations of exercise might have influenced how they responded to the PA question. Future daily diary studies of PA and alcohol use should go beyond occurrence to assess intensity and perhaps use of objective PA measure (Plasqui, Bonomi, & Westerterp, 2013). Second, our nonexperimental design does not allow for causal inferences. Third, as our study did not assess real-time PA and alcohol use, it is also somewhat susceptible to recall error, though far less than traditional retrospective reports. Finally, our study was conducted with predominantly non-Hispanic White students from a large university. Thus, our findings require replication with racially/ethnically diverse samples and in other emerging adult populations such as 2-year college students and non-college-educated emerging adults (Arnett, 2016).

However, our study is novel in several ways. First, despite the hypothesized role of several personality traits in the PA-alcohol use association (Leasure et al., 2015), this is the first study that examined the between-person association between PA and alcohol use controlling for those personality traits (cf. Leasure & Neighbors, 2014) along with gender. Second, this is one of the few studies to investigate the within-person association between PA and alcohol use among college students, an important population in which to investigate the PA-drinking association (Raynor & Levine, 2009). In view of our fairly large study sample that yielded over 8,000 daily observations, we are confident that our null findings related to PA and alcohol use are not due to compromised statistical power (Bolger & Laurenceau, 2013). Third, we expanded previous studies by investigating the within-day temporal sequencing of PA and alcohol use and we evaluated next-day lagged models.

In conclusion, although we did not observe that students’ drinking and PA occur on the same day, there was a temporal sequence and a next-day lagged association in which earlier alcohol use influenced later PA. Our findings need to be replicated in future studies using fine-grained measures of PA and alcohol use and motives for both behaviors.

Author Biographies

Dalnim Cho, PhD, is a postdoctoral fellow at the Department of Health Disparities Research at the UT Texas MD Anderson Cancer Center. Her research investigates multiple levels of influence for health and health behavior changes to promote sustainable effects and reduce health disparities.

Stephen Armeli, PhD, is a social psychologist at Fairleigh Dickinson University. His research focuses on the daily stress and coping process and substance use.

Jeremiah Weinstock, PhD, is a licensed clinical psychologist and associate professor of psychology at Saint Louis University. His research focuses on the treatment of addictive behaviors, including the use of exercise as an intervention to reduce drinking and prevent relapse.

Howard Tennen, PhD, is board of trustees distinguished professor at the University of Connecticut. His research focuses on daily stress and coping processes in health and illness.