Individual Differences in Affect Dynamics and Alcohol-Related Outcomes

Abstract

Objective:

To examine whether individual differences in intensive longitudinal data-derived affective dynamics (i.e., positive and negative affect variability and inertia and positive affect-negative affect bipolarity) – posited to be indicative of emotion dysregulation – are uniquely related to drinking level and affect-regulation drinking motives after controlling for mean levels of affective states.

Method:

We used a large sample of college student drinkers (N = 1640, 54% women) who reported on their affective states, drinking levels and drinking motives daily for 30 days using a web-based daily diary. We then calculated from the daily data positive and negative affect variability, inertia, affect bipolarity and mean levels of affect and used these as predictors of average drinking level and affect-regulation drinking motives (assessed using both retrospective and daily reporting methods).

Results:

Findings from dynamic structural equation models indicated that mean levels of affect were uniquely related to drinking motives, but not drinking level. Few dynamic affect predictors were uniquely related to outcomes in the predicted direction after controlling for mean affect levels.

Conclusion:

Our results add to the inconsistent literature regarding the associations between affective dynamics and alcohol-related outcomes, suggesting that any effects of these indicators, after controlling for mean affect levels, might be more complex than can be detected with simple linear models.

Motivational models of alcohol use posit that regulating positive and negative affective states are core reasons for drinking (Cooper et al., 2016; Cox & Klinger, 1988; Weiss et al., 2022) and that affect regulation-related drinking motives – especially drinking to cope with negative affect – are influenced, in part, by broader difficulties in regulating emotions (Cooper et al., 2016; Dvorak et al. 2015; Tragesser et al., 2007; Votaw et al. 2021). A parallel area of research posits that problematic emotion regulation is not only indicated by high levels of negative affect and low levels of positive affect, but by greater affect variability (Eid & Diener, 1999), affect inertia or resistance to change (e.g., Koval et al., 2013), and the degree to which positive and negative affect are correlated within person, i.e., bipolarity (e.g., Dejonckheere et al., 2018; Houben et al., 2015). Moreover, this line of research advances the notion that such processes are best assessed from intensive reports of affective states as they unfold in everyday life.

To date, relatively few studies have used intensive longitudinal designs to examine how individual differences in affect dynamics are related to alcohol-related outcomes and findings from studies that have are inconsistent (e.g., Gottfredson & Hussong, 2013; Peacock et al., 2015, Rankin & Maggs, 2006). In addition, the lack of focus on affect-regulation drinking motives in these studies is unfortunate given their theoretical connection to emotion dysregulation and their posited role as distal antecedents for alcohol use disorders (Cooper et al., 2016). Thus, the central aim of the present study was to evaluate the associations between intensive longitudinal data-derived affect dynamics individual difference indicators and affect-regulation drinking motives in a large sample of college student drinkers, a population at high risk for drinking-related problems (Hingson et al., 2017). Similar to previous studies, we also examined the association between individual differences in affective dynamics and drinking level.

Dynamic affect indictors of emotion regulation

Research examining individual differences in affect dynamics indicative of emotion dysregulation has been furthered by the increasing use of intensive longitudinal research designs in which affective states are reported daily or multiple times per day as they unfold in everyday life. These close to real-time ecologically valid reports of affective states allow researchers to directly calculate indicators of dynamic affect processes – such as day to day variability and inertia – rather than having individuals subjectively retrospect on these complex processes. In fact, research shows that intensive longitudinal data-derived dynamic affect indicators are weakly associated with one-time self-report assessments of affective dynamics such as variability (e.g., Sperry & Kwapil, 2020), suggesting that they measure distinct constructs. Evidence further indicates that intensive longitudinal data-derived dynamic affect indicators – such as affect variability and affect bipolarity (Dejonckheere et al., 2018) – have trait-like qualities and are stable across time (e.g., Eid & & Diener, 1999; Penner et al., 1994).

Although a myriad intensive longitudinal data-derived dynamic indicators have been proposed, Dejonckheere et al.’s (2019) meta-analysis of 14 commonly examined dynamic indicators suggests that many of them are highly intercorrelated – generally falling into the domains characterized by variability, inertia (i.e., time-dependent affect carry-over from one measures period to the next) and affect-bipolarity (i.e., i.e., the strength of within-person association between positive and negative affect) – and that they tend to be associated with mean levels of affect. Results from Dejonckheere et al.’s simultaneous regressions predicting individual differences in levels of depression, borderline personality disorder symptoms, and life satisfaction from these dynamic affect predictors showed that few of them explained variance in outcomes controlling for mean affect levels and basic measures of variability (i.e., standard deviation).

Affect dynamics and alcohol use

Despite the key role that emotion dysregulation plays in theoretical models of problematic alcohol use, relatively few studies to date have examined whether individual differences in intensive longitudinal data-derived indicators of affect dynamics are related to commonly examined drinking-related outcomes, especially affect-regulation drinking motives. Furthermore, the studies that have tend to have small sample sizes, generally focus only on affect variability, rather than other affect dynamic indicators, and do not always control for mean affect levels in predicting drinking outcomes. Finally, findings from these studies have produced somewhat inconsistent findings.

For example, Rankin and Maggs’s (2006) study of 202 first-year college students found that individuals who had higher levels of week-level negative affect variability drank more on average and this effect held after controlling for mean negative affect levels. Mohr et al.’s (2015) study of 47 community adults found that individuals who displayed higher levels of positive and negative affect variability drank more overall. However, positive and negative affect variability indicators were examined in separate models, thus making their unique contributions unclear. Using a large sample of college students (examined in the present study), Weiss et al. (2018) found that daily affect variability mediated the effect of early life trauma on heavy drinking day frequency; however, they did not control for mean affect levels. Peacock et al.’s (2015) ecological momentary assessment study of 53 community adults found no association between daily affect variability and drinking level, and when analyses were restricted to affect reports occurring prior to drinking each day, results indicated that individuals with higher arousal-related affect variability drank less often. Finally, Jahng et al. (2011) found that between-day negative affect variability was higher among women with borderline personality disorder who drank compared to non-drinkers, but the opposite pattern was found among women with major depressive disorder/dysthymic disorder, with non-drinkers compared to drinkers showing greater between-day negative affect variability.

More relevant to the primary aim of our study, only one study, to our knowledge, has examined whether individual differences in intensive longitudinal data-derived affect dynamics were related to affect regulation drinking motives. Specifically, Gottfredson and Hussong (2013) found among 86 college students, that individuals with higher levels of overall affect variability (derived from both positive and negative affect), but not higher mean levels of either positive affect or negative affect, had higher levels of drinking to cope (DTC) motivation and more frequent daily alcohol use.

The present study

We attempted to further existing research examining how individual differences in intensive longitudinal data-derived affect dynamics are related to alcohol-related outcomes in several ways. Previously published research examining the association between intensive longitudinal data-derived dynamic affect indicators and alcohol-related outcomes has generally been conducted with relatively small sample sizes (cf., Weiss et al. 2018) which can produce biased effect sizes (Schäfer & Schwarz, 2019). In the present study, we examined a large sample (N = 1640) of college student drinkers who reported on their affective states daily for 30 days, thus providing high power to detect small effects.

Inconsistent results from previous studies also might be due to the focus on drinking level, which especially among young adults is strongly influenced by factors other than affect-regulation such as peer influences and norms (Borsari & Carey, 2001; Neighbors et al., 2016), alcohol availability and cost (Shih et al., 2015), and living arrangements (Wechsler & Nelson, 2008). We posited that affect dynamics indicative of emotion dysregulation would be more strongly related to affect-regulation drinking motives, given their theoretical link to deficits in emotion-regulation processes (Cooper et al., 2016; Tragesser et al., 2007; Votaw et al., 2021). Thus, we attempted to replicate Gottfredson and Hussong’s (2013) findings showing that individual differences in affect variability were related to DTC motivation and extend this research by also examining positive emotion enhancement drinking motivation.

Finally, previous studies examining the association between intensive longitudinal data-derived affect dynamics and alcohol-related outcomes have mainly focused on individual differences in positive and negative affect variability. In the present study, we simultaneously examined individual differences in multiple types of dynamic indicators as predictors of drinking-related outcomes to evaluate their incremental validity beyond mean affect levels. We focused on the dynamic dimensions related to the factors identified in Dejonckheere et al.’s (2019) meta-analysis including positive and negative affect variability and inertia and positive affect-negative affect bipolarity. Given that higher levels of affect variability and inertia and stronger negative associations between positive and negative affect (i.e., greater bipolarity) are conceptualized to indicate problematic emotion regulation (Dejonckheere et al., 2019; Houben et al., 2015), we predicted that they would be associated with higher overall drinking levels and both DTC and enhancement motivation after controlling for mean affect levels.

Method

Participants and procedure

Procedures were approved by the university institutional review board. Undergraduates at a large university were recruited from the psychology department research pool and through campus-wide email advertisements to participate in a study of daily life and alcohol use. Prospective participants needed to be at least 18 years of age, to have drunk alcohol at least twice in the past 30 days, and to have no past treatment for alcohol problems.

Participants provided informed consent and then completed an online baseline survey assessing demographics. Approximately two weeks later, participants completed the daily diary portion of the study in which they logged on to a secure website each day for 30 days to complete a brief survey between the hours of 2:30 – 7:00 PM. Relevant to our study, participants were asked each day to report on their previous night’s and current day’s drinking and drinking motives, and their current affective states.

We recruited 1818 students, 178 of whom either had missing data on their baseline survey or who failed to adhere to minimum daily reporting of 15 days. The final sample included 1,640 students, 54% women, and the average age was 19.2 (SD=1.5), most 79% were Caucasian. Included and excluded participants did not differ on age or ethnicity, but the final sample was comprised of a greater percentage of women compared to the excluded students (35%, p<.001).

Measures

Daily drinking and motives.

On each of the 30 daily diary days, participants logged in to a secure Internet-based daily survey to report how many drinks (responses: 0 to >15) they had in social (interacting with others) and non-social (alone; not interacting with others) contexts separately for the previous evening (i.e., after they completed the prior day’s survey) and for today (up to the reporting time). One drink is listed as “one 12-oz. can or bottle of beer, one 4-oz. glass of wine, one 12-oz. wine cooler or 1-oz. of liquor straight or in a mixed drink.” Given the low base-rate for non-social drinking (i.e., it was reported only 2.9% of the evenings and 0.8% of the days), we summed the social and non-social drinks to create a total number of drinks variable. To avoid extreme values from exerting undue influences, we recoded days with greater than 15 drinks reported to 15 (this was only 0.9% of the daily observations). We then aggregated these values over the number of available reporting days to create a mean daily drinking variable.

On days when participants reported alcohol use for either the previous night or the current day, they were then queried about their reasons for drinking in that period using a slightly modified version of Cooper’s (1994) Reasons for Drinking scale. Participants were asked whether they drank for the following reasons (responding with a 3-point scale [0 = no, 1 = somewhat, 2 = definitely]). To balance out response demands, participants were queried on reasons for not drinking on days when no consumption was reported. DTC motivation was assessed with the items “to forget my ongoing problems/worries,” “to feel less depressed,” “to feel less nervous,” “to avoid dealing with my ongoing problems,” “to cheer up,” and “to feel more confident/sure of myself”.” The main alteration was that in the original scale, drinking to reduce anxiety and depression is assessed with a single item; in the present study, this was separated into two items for aims unrelated to the present study. Enhancement motivation was assessed with “because I like the pleasant feeling” and “to have fun.” Composite scores were created across all day and night drinking occasions by averaging together the relevant items. Internal consistency (alpha) was .85 for the DTC motivation scale and .68 for enhancement motivation. We then calculated the mean values of both scales across all daily reports; if participants did not have any reports of drinking (and thus drinking motives) during the daily portion of the study, we assigned them a value of zero.

Daily affect.

We assessed daily affect using items derived from the Positive and Negative Affect Schedule- Expanded (PANAS-X; Watson & Clark, 1988) and Larsen and Diener’s (1992) mood circumplex. Participants were asked to rate how much does each of the following words describe how you felt today from the time they woke up until now? They responded using a 5-point scale (1 = “not at all” to 5 = “extremely”]). Daily negative affect was assessed with the items nervous, dejected, irritable, hostile, sad, angry, anxious, and tense. Positive affect was assessed with the items cheerful, happy, excited, relaxed, enthusiastic, content, calm, and energetic. Overall negative affect (NA) and positive affect (PA) scale scores were created by averaging together the relevant items. Internal consistency (alpha) was .86 for the PA scale and .92 for NA scale.

Analytic strategy

To assess the associations between the dynamic affect predictors and the drinking outcomes (i.e., retrospective and mean daily reports of drinking level, DTC motivation and enhancement motivation – each examined separately) we used dynamic structural equation modeling (DSEM) in Mplus v8.6 (Asparouhov et al., 2018). This approach offers advantages over traditional linear mixed models for estimating the effects PA and NA variance and inertia in that it can easily model heterogeneity and unequal time intervals and produces less biased estimates (Asparouhov & Muthen, 2019). In our models, affect variability was the log transformed variance of PA and NA. Inertia was the autoregressive effect of PA (and NA) on itself using a one-day lag. We also estimated and saved each participant’s correlation between negative and positive affect, then merged this data to be used as an additional predictor in the DSEM model.

Each outcome was evaluated in three sequential models: one to assess the bivariate relationship between each of five dynamic affect indicators (PA variability, PA inertia, NA variability, NA inertia, and PA-NA correlation [affect bipolarity]) and mean levels of PA and NA with each of the drinking outcomes. A second model included all five dynamic affect measures simultaneously to assess their unique effects in predicting the drinking outcomes. Finally, the third model also included mean levels of PA and NA in addition to the dynamic measures to assess their unique contributions in predicted the drinking outcomes. We used the Bayes estimator using two MCMC chains with 1000 iterations; convergence and stationarity were checked with autocorrelation and trace plots. Statistical significance was assessed using one-tailed p-values with a .025 cutoff from the posterior parameter distributions. The reason for a one-tailed test was the use of Bayesian estimation, which uses a generated posterior distribution instead of a theoretical null distribution centered at zero. Because the posterior is not centered at zero there cannot be two tails of extreme values equal distance from the null value of zero. Instead, the proportion of values from the posterior distribution that differ in sign from the point (median) estimate functions as the p-value. Given our large sample size, we had high power to detect even small unique effects in our final model including all seven predictors. Specifically, assuming a modest variance accounted for by our overall model (e.g., 10%), we had power of .98 to detect a small effect (i.e., unique variance explained of 1%) and power of .81 to detect an even smaller effect (i.e., unique variance explained of .5%). The data and MPLUS code for the final models can be found at https://osf.io/dwaqs/.

Results

Participants reported drinking on 16.9% of the reporting days, and 91.1% of the sample reported drinking on at least one day. On days when drinking occurred, participants consumed a mean of 5.8 drinks (SD = 4.1). Table 1 shows the descriptive statistics for the five dynamic affect measures, the mean levels of affect, and the alcohol-related outcomes (mean motives and daily drinking level). The inertia (autoregressive) effects for both negative and positive affect were of moderate strength and differed from zero (ps<.001), indicating that, on average, there was significant affect carryover from day to day. Likewise, the variance for both negative and positive affect differed from zero (ps <.001). The average correlation between negative and positive affect (bipolarity) was r = −.33 and differed from zero (p<.001); however, for 16% of the participants the correlation was positive. All measures had a positively skewed distribution except average positive mood.

Table 1.

Descriptive statistics

Measure	M	SD	2.5%	25%	Median	75%	97.5%
NA Inertia	.269	.126	.056	.176	.258	.346	.543
PA Inertia	.322	.121	.101	.242	.317	.389	.598
NA Variance	.179	.185	.004	.055	.124	.241	.632
PA Variance	.345	.207	.078	.204	.297	.440	.862
Bipolarity	−.328	.366	−.821	−.607	−.403	−.126	.627
NA Level	1.46	0.35	1.01	1.20	1.38	1.65	2.36
PA Level	2.54	0.57	1.41	2.16	2.54	2.92	3.73
Mean Daily drinking level	1.16	1.27	0.00	0.27	0.74	1.66	4.76
Mean Daily DTC motivation	0.21	0.30	0.00	0.00	0.08	0.29	1.08
Mean Daily DTE motivation	1.05	0.60	0.00	0.63	1.07	1.50	2.00

Note. NA = negative affect, PA = positive affect. DTC = drinking to cope, DTE = drinking to enhance. Lower levels of bipolarity correspond to stronger negative correlations, i.e., greater bipolarity.

The correlation between the dynamic affect measures and the affect mean levels are shown in Table 2. The dynamic measures showed low- to moderate-sized correlations with the strongest association being between PA variance and NA variance. Mean levels of PA and NA were generally weakly correlated with the dynamic affect measures with the exception of mean levels of NA and NA variance, which is likely due to the fact that individuals with low mean levels of NA showed little variability. Overall, these associations were generally consistent with Dejonckheere et al.’s (2019) findings.

Table 2.

Correlation Matrix among Dynamic, Bipolarity, Mean Level, and Outcomes Measures

	NA Inertia	PA Inertia	NA Variance	PA Variance	Bipolarity	NA Level
NA Inertia	--
PA Inertia	.261**	--
NA Variance	.115**	−.068**	--
PA Variance	−.040	−.099**	.408**	--
Bipolarity	−.004	.037	−.263**	−.268**	--
NA Level	.218**	−.083**	.638**	.069**	−.052*	--
PA Level	−.005	−.050*	−.073*	.144**	−.197**	−.045

Note. NA = negative affect, PA = positive affect; Lower levels of bipolarity correspond to stronger negative correlations, i.e., greater bipolarity.

^*p<.05

^**p<.01

Table 3 shows the results from the DSEM models predicting the alcohol-outcomes. For average drinking level, only bipolarity was significantly related at the bivariate level (model 1) and after controlling the other dynamic measures (model 2) and mean affect levels (model 3). This finding, however, was opposite to prediction, with the positive coefficient indicating that less bipolarity (i.e., weaker negative/more positive PA-NA correlations) was associated with higher drinking levels.

Table 3.

Models predicting mean daily drinking-related outcomes

	Model 1					Model 2					Model 3
Average Drinks	β	SE*	LL	UL	P ^	β	SE	LL	UL	P ^	β	SE	LL	UL	P ^
NA inertia	−.038	.043	−.128	.033	.160	−.025	.041	−.111	.053	.268	−.034	.040	−.102	.044	.235
PA inertia	−.067	.042	−.153	.001	.030	−.050	.042	−.137	.031	.115	−.048	.043	−.128	.028	.145
NA variance	−.007	.027	−.060	.041	.405	.010	.029	−.045	.061	.363	−.006	.041	−.090	.068	.455
PA variance	.008	.028	−.050	.055	.405	.027	.029	−.018	.087	.147	.036	.034	−.035	.097	.170
Bipolarity	.075	.024	.038	.131	<.001	.090	.026	.038	.143	<.001	.094	.025	.047	.145	<.001
NA level	.023	.025	−.043	.070	.195						.027	.041	−.053	.107	.265
PA level	.012	.024	−.023	.059	.290						.030	.027	−.027	.082	.130
Coping motives
NA inertia	.111	.037	.038	.184	<.001	.199	.038	.130	.276	<.001	.037	.032	−.024	.094	.135
PA inertia	−.060	.040	−.156	.007	.045	−.107	.039	−.168	−.028	.005	−.035	.034	−.109	.031	.115
NA variance	.325	.024	.275	.366	<.001	.360	.023	.313	.398	<.001	−.033	.035	−.100	.038	.200
PA variance	.007	.028	−.048	.054	.425	−.124	.027	−.171	−.069	<.001	−.001	.028	−.055	.053	.480
Bipolarity	.002	.024	−.039	.059	.460	.067	.023	.021	.113	<.001	.011	.022	−.033	.058	.330
NA level	.498	.020	.459	.539	<.001						.540	.028	.487	.596	<.001
PA level	−.104	.025	−.145	−.055	<.001						−.074	.024	−.124	−.032	<.001
Enhance motives
NA inertia	−.017	.040	−.098	.058	.345	.010	.038	−.053	.099	.380	.007	.041	−.066	.100	.430
PA inertia	−.048	.039	−.128	.029	.110	−.006	.041	−.094	.067	.415	−.011	.038	−.091	.056	.400
NA variance	.062	.027	.005	.109	.010	.009	.027	−.047	.062	.360	.036	.040	−.048	.104	.215
PA variance	.072	.029	.013	.121	.005	.021	.029	−.040	.074	.240	.009	.034	−.059	.067	.375
Bipolarity	−.176	.024	−.225	−.135	<.001	−.172	.025	−.040	.074	<.001	−.148	.026	−.199	−.093	<.001
NA level	.030	.025	−.029	.078	.120						−.005	.039	−.071	.080	.460
PA level	.126	.024	.083	.170	<.001						.106	.027	.050	.163	<.001

Note. Model 1: Bi-variate associations between dynamic and mean level affect measures and outcomes. Model 2: Dynamic measures entered simultaneously in the same model. Model 3: Dynamic measures and mean affect levels entered simultaneously in the same model. β = standardized regression coefficient.

^*SE is the Bayesian posterior SD.

^{^}One-sided p-value

In predicting DTC motivation, we found several significant bivariate associations (model 1), including positive associations with NA inertia, NA variance, and mean levels of NA, and negative associations with PA inertia and mean levels of PA. All of the dynamic indicators remained significant controlling for each other (model 2), and bipolarity became significant in this step, but not in the direction predicted. In model 3, only mean levels of PA and NA were significant when entered into the model with the dynamic affect predictors.

Finally, we found that enhancement motivation was positively related to NA and PA variance and mean positive affect levels and negatively associated with affect bipolarity at the bivariate level (model 1). Only affect bipolarity remained significant in model 2 and in model 3. In model 3, mean positive affect and was a unique positive predictor of enhancement motivation. As a check, we re-estimated all for the models reported in Table 3 after removing individuals (N = 146) who did not drink during the daily reporting. The pattern of significant findings was identical.

Discussion

We found relatively little evidence that individual differences in affect dynamics are associated with drinking-related outcomes independent of mean affect levels. Although several of our findings were consistent with our predictions at the bivariate level and after controlling for other dynamic predictors – e.g., negative affect variability and inertia being positively related to drinking to cope motivation, and variability in both types of affect were positively associated with enhancement drinking motivation – these effects were no longer significant when mean affect levels were controlled.

The null effects were unsurprising in the case of negative affect variability, given its strong correlation with mean levels of negative affect. However, negative affect inertia showed only a small to moderate overlap with mean negative affect levels, and positive affect variability and inertia were weakly related to mean levels of positive affect. All dynamic indicators were uniquely predictive of drinking to cope motivation when mean affect levels were not controlled but became non-significant in the final models including all of the predictors. Thus, it appears that the associations between these dynamic indicators and the affect-regulation drinking motives were, for the most part, due to overlap with mean affect levels. These null findings are broadly consistent with Dejonckheere et al.’s (2019) meta-analytic results regarding the associations between individual differences in affect dynamics and non-alcohol-related measures of well-being and add to a pattern of inconsistent results from similar studies examining alcohol-related outcomes.

We did find that affect bipolarity was uniquely predictive of drinking level and enhancement motivation after controlling for mean affect levels. However, the former effect was not in the predicted direction. Specifically, we found that individuals who showed weaker negative correlations between positive and negative affect (less bipolarity, i.e., less dysregulation) drank more. Although this might simply represent a spurious effect and needs replication, it could be indicative of processes distinct to this population. For example, among college students drinking tends to be more closely tied to social and normative factors such as drinking at parties, bars, and Greek organizations. Individuals who show problematic emotion regulation, at least as indicated by greater bipolarity, might be more likely to self-select out of such settings or might be less likely to be invited to such occasions given interpersonal difficulties related to their dysregulated affect, thus resulting in lower drinking levels.

The unpredicted findings for bipolarity and drinking level are further complicated by our results showing that, consistent with prediction, individuals with stronger bipolarity showed higher levels of enhancement motivation, which was positively related to drinking level in our sample (r = .30, p < .05) – a finding consistent with the past studies (Cooper et al., 2016). This raises the possibility that dysregulated affect – as indicated by greater bipolarity – might manifest in a greater tendency to drinking to enhance positive emotions when drinking occurs, but that it also results in lower drinking levels via possible disruption of social and normative factors that tend to promote drinking among college students. Again, replication is needed before any conclusions can be drawn from these complex findings. More generally, the effects of bipolarity were quite small, thus raising questions about the importance of this factor in these various drinking-related processes.

Although our findings might indicate that previous significant results from studies with small samples sizes might represent spurious effects, differences in sample demographics, measures, and research designs across these studies prevent us from drawing firm conclusions. For example, Mohr et al. (2015) examined a community sample of adults, as compared to our focus on college students, and found differential effects depending on the social and contextual nature of the drinking, i.e., negative affect variability was positively associated with mean levels of drinking with others away from home and negatively associated with mean levels of solitary drinking at home. We focused only on overall drinking because past studies show that most college student drinking tends to be social in nature (Gonzalez, 2012; Mohr et al., 2001) and our daily drinking reports were consistent with this. Indeed, non-social drinking was very infrequent in our sample.

Another design difference that might explain the inconsistent results across these studies is the frequency with which affective states were sampled. Specifically, we sampled affective states once per day. Several previous studies showing individual differences in affect variability related to drinking level sampled affect multiple times within-day (e.g., Gottfredson & Hussong, 2013; Mohr et al., 2015, cf. Rankin & Maggs, 2006). It might be that assessing affective states intensively within-day provides a more precise assessment of affect variation – or affect dynamics such as bipolarity (e.g., Dejonckheere et al., 2018) and inertia (e.g., Koval et al., 2013) – that is indicative of emotion dysregulation. Future studies using more fine-grained assessment of affective states are needed to test this notion.

Our findings regarding how trait-like individual differences in affect dynamics are related to alcohol use and drinking motives are not informative about the effects of state-like changes in such dynamics on these outcomes. For example, several studies that sampled both affective states and alcohol-related outcomes multiple times within day provide some support for the notion that within-person (within-day) increases in affect variability are related to higher levels of proximal drinking (e.g., Mohr et al., 2015; Gottfredson & Hussong, 2013; cf., Tovmasyan, Monk, Qureshi et al., 2022). However, we urge caution in over-interpreting results from a small number of studies employing relatively small person-level sample sizes and few within-day observations. Indeed, results from recent meta-analyses found null (Dora et al., 2022; Tovmasyan, Monk, Sawicka et al., 2022) or weak positive (Bresin et al., 2018; Tovmasyan, Monk, & Heim., 2022) within-person associations between negative affect and proximal levels of alcohol use. These findings highlight the possibly complex nature – i.e., complicated mediation and moderation effects – of any associations that might exist between affect-related processes and proximal drinking behavior. Future research employing both large sample sizes and frequent within-day sampling is needed to better evaluate associations between state-like changes in affect dynamics and drinking behavior and motivation.

Finally, several other limitations merit mentioning. Our sample was overwhelming Caucasian and from one university, thus limiting the generalizability of our findings. We did not exclude individuals with mood disorders; inclusion of such individuals could have increased noise in our assessment of the various affective dynamics we measured. Our enhancement drinking motivation measure demonstrated relatively low reliability at the daily level, which could have contributed to some of our null findings. However, our primary analyses focused on person-level associations, i.e., after aggregating daily reports across all reporting days. At this level, our measure showed strong correlations with other variables such as drinking to cope motivation and mean drinking level. Thus, it is our belief that the sub-optimal reliability at the day-level did not excessively attenuate the observed associations between this measure and relevant correlates.

To summarize, we found weak evidence that individual differences in intensive longitudinal data-derived dynamic measures were uniquely associated with affect-regulation drinking motives and drinking level after controlling for mean affect levels. Thus, at the very least, our results suggest that any role individual differences in affective dynamics play in explaining such outcomes might be more complex than what we detected in our additive linear models. For example, the relationship between these factors and drinking outcomes might be multiplicative in nature with the effect of the dynamic factors being contingent on the relative level of other dynamic factors, mean affect levels, and other individual difference factors not assessed here. Future research should test this possibility.