The Language of Subjective Alcohol Effects: Do Young Adults Vary in their Feelings of Intoxication?

Abstract

Among young adults, subjective feelings of alcohol’s effects often guide risky decision-making. The majority of studies measuring subjective effects have used singular indices (“How drunk do you feel?”) which limits our understanding of young adults’ full range of subjective states and their individual differences in subjective effects language. Toward a more in-depth understanding of the heterogeneity among alcohol users based on their subjective experiences of alcohol’s effects, we identified latent classes of individuals based on their self-generated language describing feelings after drinking and compared these classes across demographic and drinking characteristics. Participants (n = 323, 54% women, 68% White, ages 18–25 years) were recruited using Amazon’s Mechanical Turk (MTurk). Participants listed words they would use to describe how they feel after drinking low, moderate, and heavy amounts of alcohol. Four latent classes of young adults emerged: “Happy Drinkers” (31%) primarily reported feeling “happy” when drinking; “Relaxed Drinkers” (24%) reported feeling happy, relaxed, and buzzed; “Buzzed Drinkers” (18%) reported feeling buzzed and dizzy; and “Multi-Experience Drinkers” (27%) reported feeling buzzed, tipsy, drunk, and wasted. Relaxed Drinkers indicated heavier alcohol use and Buzzed Drinkers reported lower drinking frequency. Classes did not differ by demographic characteristics. Young adult alcohol users can be distinguished based on the language they use to describe their feelings of intoxication. To continue to advance our understanding of subjective effects, it is necessary to take into account the full range of language used and how this language differs by young adult drinking behavior.

Young adulthood is a high-risk period for hazardous alcohol use with 18 to 25 year-olds reporting the highest prevalence of heavy alcohol use of any age group (Substance Abuse and Mental Health Services Administration, 2018). An estimated 37% of young adults report binge drinking at least once in the past month and 10% report binge drinking on five or more days in the past month. Excessive alcohol use among young adults is a major public health concern and is linked to physical and mental health problems, legal trouble, driving under the influence, injury, assault, and even death (White & Hingson, 2013).

Understanding the prevalence and patterning of problematic alcohol use is critical to the advancement of prevention and intervention techniques targeting young adult drinkers most at risk. Objective measures of alcohol intoxication can offer important insights into such behavioral patterns, but a growing body of research suggests it is equally – if not more – important to understand subjective experiences of intoxication. A daily diary investigation assessing actual versus perceived intoxication found that college students’ risk for driving while under the influence was highest when individuals were intoxicated but did not feel intoxicated (Quinn & Fromme, 2012). Thus, subjective feelings of intoxication may serve to guide risky decision-making more than one’s objective levels.

To best measure subjective intoxication, we must consider young adults’ language of subjective intoxication, or the ways in which they describe varying levels of impairment. There are a wide range of terms used to communicate subjective alcohol use experiences, with these terms continually evolving over time and often differing by culture and environment (Thickett et al., 2013). Early work by Levine (1981) reviewed the vocabulary of drunkenness and found a staggering number of synonyms for the term “drunk”, referencing over 350 terms in the Dictionary of American Slang (Wentworth & Flexner, 1975), over 200 terms compiled by Benjamin Franklin (Larson, 1937), and close to 900 terms in The American Thesaurus of Slang (Berrey & Van Den Bark, 1953).

Despite evidence of an exceedingly large amount of terms to describe an alcohol use experience, contemporary research has typically used a singular term, such as “drunk” to assess subjective intoxication on metrics ranging from 0 (not at all drunk) to 100 (extremely drunk). A large body of work suggests this approach may not necessarily capture the full range of young adults’ subjective intoxication. The Subjective Effects of Alcohol Scale (Morean, Corbin, & Treat, 2013), the Biphasic Alcohol Effects Scale (Martin, Earleywine, Musty, Perrine, & Swift, 1993), and the Subjective High Assessment Scale (Judd, Hubbard, Janowsky, Huey, & Atewall, 1977; Schuckit & Gold, 1988) highlight this complexity by measuring multiple dimensions of perceived effects from alcohol use. These assessments include stimulation and sedation as well as an individual’s affective state. Work by Levitt, Sher, and Bartholow (2009) showcased the wide variety of terms college students use to describe their subjective states and that each word may differentially represent the level of subjective effects experienced. Specifically, Levitt et al. provided college students with a variety of words describing subjective impairment. Researchers assessed participants’ familiarity and usage of each term at various levels of drinking. Results showed that intoxication-related terms, particularly the word “drunk”, are differentially understood and do not necessarily distinguish moderate and high levels of impairment. Recent work by Linden-Carmichael, Masters, and Lanza (2020) built upon this research by asking participants to self-generate and rank-order words they would use to describe varying states of alcohol intoxication. Participants expressed a wide range of terms that fit along a continuum of subjective states (ranging from “slightly buzzed” to “tipsy/happy” to “drunk” to “wasted”).

Young adults have a large vocabulary for describing subjective effects of alcohol experienced and this language can quantitatively map onto the amount of alcohol consumed. However, it is unknown whether individuals cluster together in terms of their own language of subjective states, and whether these terminology groupings are indicative of particular types of drinkers. Subgroups of young adults may exist based on the language they use to describe the subjective effects they experience, possibly due to the historic association between alcohol use behavior and certain demographic or environmental characteristics. Kerr, Yi, and Moreno (2018) examined gender differences in the language that college students use to describe alcohol use on social media and found that women were more likely than men to reference terms relating to their level of intoxication, such as using the term “drunk” in their social media posts. Levitt et al. (2009) found that men were more likely to use words with a forceful or violent connotation (e.g., “ripped” and “hammered”), while women were more likely to use more subtle terms to describe their alcohol use (e.g., “loopy” and “tipsy”). Gleaning from research on reasons for drinking (Kuntsche, Knibbe, Gmel, & Engels, 2005), intoxication terminology may also differ by age and college status. While the majority of alcohol use motives, including social and coping reasons, become less prevalent from age 18 on, using alcohol to relax and to sleep becomes more common throughout young adulthood (Patrick & Schulenberg, 2011; Patrick et al., 2011). Further, the relatively limited research available on differences in motivations between college- and non-college-attending young adults suggests that non-college individuals may have an increased tendency to drink to self-medicate or cope relative to college students (Barnett et al., 2003), potentially related to differing life circumstances (Lau-Barraco, Linden-Carmichael, Hequembourg, & Pribesh, 2017). Lastly, subjective responses to alcohol may differ by race and ethnicity (Pedersen & McCarthy, 2009; Richner, Corbin, & Menary, 2018). Specifically, black participants were found to experience increased stimulation after drinking alcohol compared to white participants, and Hispanic/Latino participants were found to experience stronger sedative effects compared to white participants; such differences may be illuminated in motivations and language used to describe intoxication.

Young adults have an extensive vocabulary to describe various states of intoxication, suggesting that metrics of subjective effects or intoxication that use singular terms (e.g., drunk) may be inadequate for obtaining a comprehensive understanding of subjective effects experienced. In addition to the variability in language used, participants may also differ in their own drinking experiences and thus use different sets of words to describe these experiences. In advancing our metrics to best assess subjective effects, it is critical that we also identify potential heterogeneity in subjective drinking experiences among young adults. Thus, the current study used recently collected (Linden-Carmichael et al., 2020) crowd-sourced data from young adults that obtained self-generated words used to describe their own subjective states at various levels of intoxication. Our first aim was to use these self-generated words to identify underlying latent classes of young adults based on the terms they used to describe subjective alcohol effects. Our second aim was to compare these latent classes of young adult drinkers in terms of typical drinking behavior and demographic characteristics (i.e., sex, age, race/ethnicity, college status).

Method

Participants and Procedure

Participants (n = 323) were recruited using Amazon’s Mechanical Turk (MTurk). MTurk is an online crowdsourcing marketplace where “requesters” (researchers) can pay “workers” (participants) to complete generally short, online tasks. MTurk allows requesters to evaluate workers’ performance in terms of data quality and degree of completion. Participants whose work is approved receive payment for participation. MTurk has been widely used in many clinical, experimental, and survey-based studies (Woods, Velasco, Levitan, Wan, & Spence, 2015) including studies focused on addictive behaviors (Amlung, Reed, Morris, Aston, Metrik, & MacKillop, 2019; Strickland & Stoops, 2019).

Eligibility criteria in the current study included (1) aged 18 – 25 years, (2) past-month heavy episodic drinking (4+/5+ drinks in one sitting for women/men), (3) past-month simultaneous use of alcohol and marijuana (use of both substances so that the effects overlap), (4) residence in the U.S., (5) native/fluent English speaker, (6) an MTurk task approval rating of >95%, and (7) completion of 50+ tasks on MTurk previously. After providing informed consent and completing a screening questionnaire, eligible participants were asked to complete an online survey that took approximately 10 minutes to complete. Three attention checks were placed throughout the survey (“Please select ‘monthly’ for this item”; “How many days are in an average week?”; “How many legs does a typical cat have?”); participants’ work was approved if they answered 2+ attention checks correctly and if they provided sensible responses. MTurk participants were compensated $1.00. The university’s institutional review board approved the study. Data were collected from July to September 2018. Sample characteristics are presented in Table 1. Additional details from the parent study can be found in Linden-Carmichael et al. (2020).

Table 1.

Descriptive Statistics across Demographic Characteristics and Drinking Behavior (n = 323)

Demographic Characteristics
Sex
% male (n = 149)	46.3%
% female (n = 173)	53.7%
% missing (n = 1)	<0.1%
M Age (SD)	23.0 (1.8)
Race/ethnicity
White (n = 221)	68.4%
Black (n = 48)	14.9%
Asian (n = 17)	5.3%
American Indian/Alaska Native (n = 8)	2.5%
Native Hawaiian or Other Pacific Islander (n = 1)	0.3%
Multiracial (n = 21)	6.5%
Hispanic (any race; n = 52)	16.1%
Missing (n = 7)	2.2%
College
% currently enrolled/graduated (n = 239)	74.0%
% never enrolled/previously enrolled but did not graduate (n = 84)	26.0%
DDQ – Drinking Behavior
M Drinking Quantity (SD)	14.66 (9.83)
M Drinking Frequency (SD)	4.08 (1.81)
% Binge Drinking Days (SD)	39.2% (34.5%)
High-Intensity Drinking
% reporting 8+/10+ (women/men) drinks (n = 40)	12.4%
% not reporting 8+/10+ drinks (n = 282)	87.6%
% missing (n = 1)	<0.1%
M Hours Spent Drinking (SD)	13.14 (9.27)

Note. DDQ = Daily Drinking Questionnaire. Drinking quantity, frequency, and hours spent drinking refer to reports during a typical week during the past three months. Binge drinking is defined as 4+/5+ (women/men) drinks on 1+ days on the DDQ. High-intensity drinking was coded as 1 = reported 8+/10+ (women/men) drinks on 1+ days on the DDQ, 0 = did not report 8+/10+ drinks on the DDQ.

Measures

Word generation.

Participants were asked to write one- or two-word phrases to describe their level of impairment when consuming only alcohol and not using any other substances. Specifically, participants were instructed to provide at least two (maximum of four) feelings for lighter (“When I’m drinking a smaller amount of alcohol [maybe one or two drinks], I would describe my level of impairment as feeling…”), moderate (“When I’m drinking a moderate amount of alcohol [maybe three or four drinks], I would describe my level of impairment as feeling…”), and heavier occasions of alcohol alone (“When I’m drinking a heavier amount of alcohol [maybe five or more drinks], I would describe my level of impairment as feeling…”).

Alcohol use indicators.

Several drinking indicators were derived from the Daily Drinking Questionnaire (DDQ; Collins, Parks, & Marlatt, 1985). The DDQ uses a calendar grid to assess the number of standard alcoholic drinks consumed and the number of hours spent drinking on a typical Monday through Sunday during the past three months. Drinking quantity was assessed by summing the total number of self-reported drinks consumed during a typical week. Hours spent drinking was assessed by totaling the number of hours spent drinking during a typical week. Drinking frequency was calculated by summing the number of drinking days during a typical week. Proportion of binge drinking days were calculated by dividing the number of days in a typical week involving 4+/5+ (women/men; National Institute on Alcohol Abuse and Alcoholism, 2004) drinks by their drinking frequency. High-intensity drinking was treated dichotomously; participants were categorized as a high-intensity drinker if they reported consuming 8+/10+ drinks (Patrick, 2016) on 1+ day during a typical week.

Demographic information.

Participants self-reported their sex, age, and college education status. Individuals were categorized as college-attenders if they were currently attending college or if they graduated from college. Individuals who reported never having enrolled in a four-year university or had attended in the past and did not graduate were categorized as non-college-attenders.

Analytic Strategy

Latent class analysis (LCA; Collins & Lanza, 2010) was used to identify underlying subgroups of participants who shared common language endorsement patterns. Each indicator was binary, reflecting whether participants generated a particular word (e.g., “buzzed”). Parameters estimated include latent class membership probabilities, reflecting the relative class sizes, and the probability of endorsing each word indicator conditional on latent class membership. Models with one through six latent classes were considered. Model identification of each was examined using 100 random sets of starting values. From identified models, an optimal number of classes was selected based on information criteria and class separation and interpretation. The bootstrap likelihood ratio test (BLRT) was also used to compare fit of k classes relative to model fit with k + 1 classes. All models were estimated using SAS PROC LCA (Lanza, Dziak, Huang, Wagner, & Collins, 2015) and the LCABootstrap macro (Dziak & Lanza, 2016).

To examine the relationship between class membership and individual characteristics, we implemented the corresponding SAS macro for the BCH 3-step approach (Dziak, Bray, & Wagner, 2017). This approach adjusts for the uncertainty introduced by possible misclassification (Bolck, Croon, & Hagenaars, 2004). The macro provides an overall test of association between class membership and each covariate of interest, as well as all pairwise comparisons between two classes on levels (or prevalence) of the covariate. To describe overall associations between class membership and each covariate, the BCH approach was used separately for each covariate of interest.

Prior to conducting analyses, language data generated by participants were cleaned to be consistent across participants (e.g., “buzzed”, “buzz”, and “buzzing” were converted to “buzzed”). After cleaning the data from the open-ended responses, we examined the frequency with which each word was generated across all participants. The most frequently generated twenty words in this sample were considered potential indicators in a latent class model. Because of sparseness in the contingency table, data reduction was necessary to achieve model identification. To select indicators, factor analysis was used for data reduction. Specifically, certain words appeared to have very similar meanings (e.g., relaxed and calm); factor analysis was used to examine the extent to which certain word pairs tended to occur together, suggesting that these words could be combined. We combined the following words that were similar in meaning: fun, funny, energetic (“fun/energized”); hammered and wasted (“wasted”); relaxed and calm (“relaxed”); light and warm (“warm”); good and happy (“happy”); blurry and dizzy (“dizzy”). No other words categories were combined. We removed two words from our analysis. “Nothing” was removed as we were interested in the effects of alcohol. “Blackout” was removed as blacking out is more of a consequence of drinking rather than a feeling. Finally, “loose” was examined due to questionable face validity, as it was unclear whether all participants were referring to being easy-going/relaxed or were suggestive of sexual promiscuity. We conducted sensitivity analyses to determine whether the four-class solution changed with and without adding the word “loose.” The addition of “loose” did not distinguish classes and did not change the class compositions in any way, and thus we also excluded this word from our final model.

Results

We included a set of 10 word indicators in our model: fun/energized, happy, warm, relaxed, tipsy, dizzy, buzzed, drunk, sleepy, and wasted. Models with 1 through 5 latent classes had adequate identification (fit statistics shown in Table 2). Based on the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC), the BLRT, as well as class interpretability, we selected a four-class model of intoxication language. Table 3 summarizes the class membership probabilities and item-response probabilities for each class. The prevalence of each word/word category for light, moderate, and heavy use occasions (“use intensity”) is presented next to each indicator. Class 1 comprised young adults who were likely to report feeling happy (most commonly a light use occasion word) after using alcohol (probability of endorsement of 0.60), but were not likely to endorse any other feeling; these individuals were labeled Happy Drinkers (31% of the sample). Class 2 was made up of individuals who were likely to report feeling happy, relaxed, and buzzed (primarily light use occasion words) than other feelings after drinking alcohol. The indicator for feeling relaxed had the highest probability of endorsement (0.75); thus, this class was labeled Relaxed Drinkers (24%). Class 3 comprised young adults who were likely to endorse only two items: feelings of dizzy (most commonly reflecting heavy occasions) and buzzed (most commonly light occasions), with a very high probability of endorsing feeling buzzed (0.98). This class was labeled the Buzzed Drinkers (18%). Finally, individuals in Class 4 had high probabilities of reporting feeling tipsy (mostly moderate), buzzed (mostly light), drunk (mostly heavy), and wasted (only heavy). As this class used words describing light, moderate, and heavy use occasions, this class was labeled Multi-Experience Drinkers (27%).

Table 2.

Fit Statistics for Latent Class Analysis Models

Class	Percent Agreement	AIC	BIC	ABIC	Entropy	BLRT p
1	100%	703.71	741.49	709.77	1.00	---
2	100%	492.74	572.08	505.47	0.71	.01
3	47%	475.54	596.42	494.92	0.70	.01
4	96%	451.10	613.54	477.15	0.75	.01
5	27%	450.42	654.41	483.13	0.79	.39
6	7%	450.28	695.83	489.65	0.84	.29

Note. AIC = Akaike Information Criteria, BIC = Bayesian Information Criteria, ABIC = adjusted Bayesian Information Criteria. BLRT = bootstrap likelihood ratio test (compares k classes to k + 1 classes). Percent agreement refers to the proportion of initial starting values that converged on the maximum likelihood solution.

Table 3.

Item-Response Probabilities for Four-Class Model of Young Adult Subjective Intoxication Language

Indicator	Prevalence by Use Intensity	Overall Prevalence	Class 1 “Happy Drinkers” (31%)	Class 2 “Relaxed Drinkers” (24%)	Class 3 “Buzzed Drinkers” (18%)	Class 4 “Multi-Experience Drinkers” (27%)
Fun/Energized	Light: 22%	.20	.36	.33	.06	.01
	Moderate: 56%
	Heavy: 22%
Happy	Light: 46%	.46	.60	.51	.30	.38
	Moderate: 42%
	Heavy: 12%
Warm	Light: 76%	.17	.21	.06	.30	.14
	Moderate: 18%
	Heavy: 6%
Relaxed	Light: 72%	.33	.36	.75	.16	.01
	Moderate: 25%
	Heavy: 3%
Tipsy	Light: 38%	.32	.03	.27	.16	.78
	Moderate: 59%
	Heavy: 3%
Dizzy	Light: 21%	.27	.41	.00	.69	.05
	Moderate: 29%
	Heavy: 51%
Buzzed	Light: 53%	.58	.02	.67	.98	.88
	Moderate: 41%
	Heavy: 7%
Drunk	Light: 3%	.35	.02	.41	.39	.66
	Moderate: 31%
	Heavy: 66%
Sleepy	Light: 21%	.17	.31	.11	.19	.06
	Moderate: 26%
	Heavy: 53%
Wasted	Light: 0%	.20	.00	.14	.08	.54
	Moderate: 0%
	Heavy: 100%

Note. In the second column, the prevalence of each word is presented by level of intensity; the most prevalent is highlighted for emphasis. Item-response probabilities ≥.50 are bolded for emphasis in the fourth through seventh column.

Characteristics Predicting Class Membership

Several aspects of drinking behaviors differed significantly across classes (see Table 4). The overall test of association between both typical drinking quantity (χ²=12.57, 3 df, p<.01) and drinking frequency (χ²=12.42, 3 df, p<.01) and class membership were statistically significant. Typical drinking quantity was significantly higher for individuals in the Relaxed Drinkers class (19.0 drinks) compared to Happy Drinkers (14.4 drinks), Multi-Experience Drinkers (12.0 drinks), and Buzzed Drinkers (13.9 drinks). In contrast, typical drinking frequency was significantly lower in the Multi-Experience Drinkers class (3.5 days) compared to Happy Drinkers (4.1 days), Relaxed Drinkers (4.5 days), and Buzzed Drinkers (4.3 days).

Table 4.

Demographic Characteristic and Drinking Behavior Comparisons across Latent Classes

	Overall Test χ2	Class 1 “Happy Drinkers”	Class 2 “Relaxed Drinkers”	Class 3 “Buzzed Drinkers”	Class 4 “Multi-Experience Drinkers”
Demographics
Sex (% female)	0.82	52.3%	52.9%	50.1%	58.5%
Age (years)	1.95	22.9	23.3	23.1	22.8
Race (% white)	0.94	65.2%	73.7%	70.0%	68.2%
College Status (% college)	5.341	73.2%	77.1%	85.9%	64.0%
DDQ - Drinking Behavior
Drinking Quantity (drinks/wk)	12.57*	14.4a	19.0b	13.9a	12.0a
Drinking Frequency (days/wk)	12.42*	4.1a	4.5a	4.3a	3.5b
% Binge Drinking Days	0.99	38.6%	41.7%	34.7%	40.8%
% High-Intensity Drinkers	3.01	12%	18%	4%	13%
Hours Spent Drinking	7.292	12.9	16.3	12.5	11.1

Note.

^*p < .01. Sex was coded as 1 = female, 0 = male. Race was coded as 1 = white (largest group), 0 = Non-white. College status was coded as 1 = currently enrolled/graduated and 0 = never enrolled/previously enrolled but did not graduate. DDQ = Daily Drinking Questionnaire. Drinking quantity, frequency, and hours spent drinking refer to reports during a typical week during the past three months. Binge drinking is defined as 4+/5+ (women/men) drinks on 1+ days on the DDQ. High-intensity drinking was coded as 1 = reported 8+/10+ (women/men) drinks on 1+ days on the DDQ, 0 = did not report 8+/10+ drinks on the DDQ. Classes that share the same alphabetic superscript are not significantly different from one another.

¹The overall test of association between college status and class membership was not significant, however the pairwise difference between Class 3 (85.9% college) and Class 4 (64.0% college) was significant (p<.05).

²The overall test of association was not significant, however the pairwise difference between Class 2 (16.3 hours) and Class 4 (11.1 hours) was significant (p<.05).

No significant association was found between class membership and demographic characteristics (proportion of females, mean age, proportion of whites, college status), the percent of typical binge drinking days, the percent of high-intensity drinkers, or mean number of hours per week spent drinking.¹

Discussion

A growing body of evidence suggests that to best understand behavioral alcohol use patterns and decision-making exhibited by young adults under the influence of alcohol, we must not only measure objective levels of impairment but also their subjective feelings of alcohol’s effects (Quinn & Fromme, 2012). Singular metrics of subjective effects or intoxication (e.g., “How drunk did/do you feel?”) may not fully capture the range of one’s subjective experiences (Levitt et al., 2009). To maximize this range, a recent metric was proposed for assessing subjective intoxication by placing anchors reflecting contemporary subjective intoxication language (e.g., buzzed, tipsy) along a continuum (Linden-Carmichael et al., 2020). Young adults may, however, vary in the words they use to describe different levels of impairment (Levitt et al., 2009). To better understand these potential differences across individuals, the current study aimed to uncover latent classes of young adult alcohol users based on the language they use to describe a range of their subjective states, and to compare these classes by typical drinking behavior and demographic characteristics.

Our findings revealed four subgroups of young adult drinkers based on their personal language of alcohol’s subjective effects. Consistent with prior work (Levine, 1981; Levitt et al., 2009), we found an extensive vocabulary, with some individuals reporting more words reflecting perceived changes in affect (e.g., fun, happy) and others providing more words related to physiological or behavioral outcomes of drinking (e.g., sleepy, dizzy). “Happy Drinkers” were defined by their use of the term, “happy” when describing their subjective states. “Relaxed Drinkers” predominately used the terms “buzzed”, “relaxed”, and “happy” in describing their subjective intoxication. “Happy Drinkers” and “Relaxed Drinkers” were both primarily classified by language used to describe lighter drinking episodes. “Buzzed Drinkers” mostly provided language related to feeling “buzzed” or “dizzy.” Generally, “buzzed” reflects lighter episodes and “dizzy” reflects heavier use episodes; this class may have bimodal drinking experiences. Finally, “Multi-Experience Drinkers” reported words that are more classically focused on the effects one feels from alcohol (“buzzed”, “tipsy”, “drunk”, and “wasted”). Interestingly, these four words reflected self-described light, moderate, and heavy use occasions and corresponded to the anchors used in the metric developed by Linden-Carmichael et al. (2020) from these data, suggesting these drinkers have a wide range of drinking episodes. The language used in these classes may shed light on each classes’ motivations for drinking as found in prior person-centered approaches examining profiles of drinking motives (Cadigan, Martens, & Herman, 2015; Coffman, Patrick, Palen, Rhoades, & Ventura, 2007). For instance, it is possible that individuals belonging to the Relaxed Drinkers class are motivated more to drink to cope with negative emotions or for the purposes of stress reduction through lighter drinking occasions, whereas Multi-Experience Drinkers may drink more for the purposes of getting drunk.

Our four classes were compared on typical drinking behavior and demographic factors. The Relaxed Drinkers class exhibited a significantly higher typical number of drinks in a week (19 drinks) compared to other classes. This group also reported approximately 16 hours of drinking per week with 42% of drinking days involving excessive alcohol use. These findings are in line with prior work finding that individuals who endorse higher motivations to cope as well as higher positive reinforcing motives exhibit the heaviest alcohol use and alcohol-related problems (Cadigan et al., 2015). Across these studies, young adults who report language suggesting positive reinforcement (“happy”) and coping language (“relax”) may reflect an at-risk group for heavier drinking and increased negative consequences. Comparisons across classes in the current study also indicated that Multi-Experience Drinkers had significantly fewer drinking days during a typical week relative to other latent classes. Although non-significantly different across classes, the Multi-Experience Drinkers reported the second-highest proportion of binge drinking days but the lowest frequency of drinking days. Additionally, this class reported words found at the higher-end of impairment continuum (i.e., “wasted”, “drunk”; Linden-Carmichael et al., 2020), suggesting that individuals in this class may have greater motivations to occasionally drink for the purposes of getting drunk than other classes.

Latent classes did not differ overall by sex, age, race/ethnicity, or college status. One exception emerged when examining pairwise comparisons specifically, with a greater proportion of college-attending young adults in the Single-Experience Class (85.9%) relative to the Multi-Experience Class (64.0%). As Multi-Experience Drinkers were distinguished from other classes in their use of high-intoxication level vocabulary, this finding adds to the growing body of literature that excessive alcohol use, such as high-intensity drinking, is not necessarily only a college phenomenon (Lanza & Collins, 2006; Linden-Carmichael & Lanza, 2018). In general, however, the lack of differences in demographic variables by class is surprising in light of findings that men and women differ on whether they use certain words to describe their impairment (e.g., women use words like “tipsy” more often than men; Levitt et al., 2009). Similar to prior work finding no differences in drinking motivation profiles for adolescent boys and girls (Coffman et al., 2007), it is possible that gender differences are more common for individual words but are less apparent when examining classes more broadly based on full range of vocabulary. Overall, our findings suggest that differences in subjective alcohol effects language are far more distinguishing with respect to the amount and frequency of drinking than other individual-level characteristics.

The current study added to the prior literature on subjective response in important ways. One major contribution is the identification of groups of individuals that differ in terms of the words they use to describe their subjective experiences of alcohol intoxication. Indeed, prior work has demonstrated heterogeneity in the subjective responses to alcohol and that subjective responses are associated with alcohol use outcomes (e.g., King, McNamara, Hasin, & Cao, 2014; Morean & Corbin, 2010; Quinn & Fromme, 2011). Converse to identifying types of subjective responses that exist, our person-centered approach of latent class analysis allows us to identify groups of people based on the language they naturally use to describe their subjective responses. Moreover, although prior scales for measuring subjective effects are multidimensional, our analysis allows for a comprehensive examination of the intersection of word use and the subtypes of people who use various descriptors. A second major contribution is that our participants generated words they would use to describe varying states of intoxication. By using self-generated language as latent class indicators, we are able to capture a wide range of subjective responses that may not necessarily be articulated in established questionnaires such as the Subjective Effects of Alcohol Scale or Biphasic Alcohol Effects Scale (Martin et al., 1993; Morean et al., 2013).

Limitations and Future Directions

There are several limitations to the current study. First, participants self-reported their typical alcohol use, which may be inaccurate due to social desirability and recall biases. Relatedly, participants’ self-reports of their subjective responses to alcohol’s effects may also be impacted by recall biases. Established subjective response measures have demonstrated a high agreement between retrospective subjective response to alcohol and laboratory-based alcohol administration subjective response (e.g., Fleming et al., 2016; Morean et al., 2013; Schuckit, Smith, & Tipp, 1997). While the current study took an initial step at examining how participants’ own words correspond to their levels of alcohol use, far more work is needed to psychometrically evaluate these self-generated word data in the prediction of alcohol use outcomes.

Second, the current study is derived from a larger study on young adult substance use, and thus our sample consisted of individuals who reported recent, simultaneous use of alcohol and marijuana. To ensure participants were reporting specifically on their subjective feelings for alcohol use, we asked participants to consider words they would use to describe their subjective states at low, moderate, and heavy occasions of (1) only alcohol and no other substances and (2) only marijuana and no other substances (see Linden-Carmichael et al., 2020). Thus, participants were asked to tease apart subjective feelings for each individual substance. Interestingly, when examining latent classes of individuals based on marijuana-specific feelings, little heterogeneity was found (data not shown here). This suggests that our sample of recent simultaneous alcohol and marijuana (SAM) users were more heterogeneous based on the words they use for describing subjective alcohol states than describing marijuana states. It should be noted, however, that relative to young adults who use only alcohol, young adults who report SAM use generally represent a higher-risk sample (Subbarman & Kerr, 2015; White, Kilmer, Fossos-Wong, Hayes, Sokolovsky, & Jackson, 2019); thus, our findings may not necessarily generalize to drinkers who do not use marijuana.

Third, data were collected from Amazon MTurk workers. Although prior work has found MTurk to be a sound platform from which to collect substance use data (Strickland & Stoops, 2019), MTurk participants are more likely to exhibit certain personality characteristics (e.g., lower social engagement) than the average population (McCredie & Morey, 2019), which could impact generalizability.

Fourth, our determination of binge drinking and high-intensity drinking behavior was based on behavior that occurred during a typical week. Such measurement may underestimate the prevalence of excessive alcohol use exhibited among our sample.

Fifth, in an effort to keep the MTurk survey brief, we did not assess several important constructs that may have illuminated additional differences among latent classes, including characteristics of their typical drinking environment (e.g., physical location, social setting) during low, moderate, and heavy episodes, individual traits such as mental health symptomatology (e.g., typical level of negative affect, depression, or anxiety), or indicators of alcohol use severity (e.g., acute negative consequences, alcohol use disorder status). Young adults’ language regarding intoxication may be heavily influenced by attributes about themselves and their surroundings; exploring this link is a key avenue for future research.

Finally, in collapsing words into categories, we had to infer the meaning of each word provided by participants. It is possible that certain words reflected different experiences that were not captured by the research team. Future work should expand upon this by conducting more in-depth qualitative interviews with young adults to understand their experiences under various degrees of intoxication.

Conclusions

Despite limitations, this study adds to a growing body of literature focused on the wide array of terms that young adults use to describe subjective alcohol effects. Our results showcase distinct classes of individuals based on the language they use, linking class membership to alcohol use behavior. These findings have important implications for research efforts and clinical work. Both self-report assessments of young adult alcohol use and alcohol interventions should include a wide range of terminology to describe alcohol intoxication and should be sensitive to individual differences in the interpretation of particular terms. Using certain language to describe the experience of alcohol use could act as a potential marker of high-risk drinking, assisting health professionals in identifying young adults engaging in risky alcohol use behavior. Future research should extend on the current findings by assessing whether subgroups of young adults who use particular intoxication-related terms are at increased risk for experiencing alcohol-related problems of the development of alcohol use disorder, providing deeper insight into the link between language, behavior, and consequences. In advancing our understanding of young adults’ experiences of alcohol’s effects, it is imperative to account for not only the full spectrum of language used, but how this language may differ based on individual characteristics and drinking behavior.

Public Health Significance:

We identified four subgroups of young adult drinkers based on their self-generated language describing subjective alcohol effects. As perceived effects may guide risky decision-making, it is important to consider the heterogeneity of alcohol users’ experiences and to integrate their own language into our measurement.