Emerging Adult Identity Development, Alcohol Use, and Alcohol-related Problems During the Transition out of College

Abstract

Alcohol use generally peaks during the early twenties and declines with age. These declines, referred to as “maturing out,” are presumed to result from the acquisition of adult roles (e.g. marriage, employment) incompatible with alcohol use. Recent empirical evidence suggests that variables other than role transitions (e.g. personality) may also be important in understanding this process. Changes in identity that occur during emerging adulthood may also be linked to the process of maturing out of heavy drinking, though no studies have yet addressed this possibility. Utilizing data from a large sample of graduating college students (N = 907) during senior year (wave 1) and the two following years (waves 2-3), the current study examined relations between aspects of emerging adult identity and drinking outcomes (alcohol use and problems). Using time varying covariate growth models, results indicated that several facets of emerging adult identity conferred risk for the failure to mature out of heavy drinking and alcohol-related problems. Experimentation/possibilities emerged as a significant risk factor for both heavy drinking and alcohol problems, but these effects diminished considerably when accounting for personality risk. In contrast, although small in magnitude, effects of self-focus on heavy drinking and negativity/instability on alcohol-related problems were relatively independent of effects of other established predictors. The effect for negativity/instability was evident only at the final wave. The findings have important implications for theories of “maturing out” and may ultimately inform tailoring or refinement of prevention/intervention approaches for emerging adults.

Despite widespread prevention efforts, alcohol remains a large part of collegiate life and a major public health concern (Sher, Bartholow, & Nanda, 2001; Slutske, 2005; Wechsler, Lee, Kuo, Seibring, Nelson, & Lee, 2002). Evidence suggests that aspects of the college environment contribute to high rates of alcohol use. For example, the rate of binge drinking (five or more drinks during a single occasion) among college students in 2013 was 35% (Johnston, O’Malley, Bachman, Schulenberg, & Miech, 2014). Moreover, Wechsler and Nelson (2008) demonstrated that students who binge drink are at higher risk for alcohol-related problems including unplanned and unprotected sexual activity, injuries, and problems with campus or local police. Slutske (2005) also found that college students were more likely to be diagnosed with an alcohol use disorder (abuse and dependence) relative to non-students.

Although a high level of alcohol use occurs during college, use generally peaks during the early twenties and declines with age (Jochman & Fromme, 2010). Several factors appear to facilitate this “maturing out” process (e.g., marriage, parenthood), but relatively little is known about these normative reductions in use relative to the substantial literature on factors contributing to increases in drinking during the transition from adolescence and into emerging adulthood. Furthermore, despite normative reductions in alcohol use with age, a significant minority of young adults fail to mature out of this pattern of heavy drinking and go on to experience clinically significant problems (Jackson, Sher, Gotham, and Wood, 2001). Thus, it is important to identify factors that might facilitate the maturing out process. As such, the aim of this study was to examine dimensions of emerging adult identity during the transition out of college in relation to alcohol outcomes (use and related problems).

Evidence suggests that maturing out of heavy drinking is, at least in part, a result of the acquisition of adult roles and responsibilities such as marriage and/or parenthood, end of formal education and start of employment, change in living arrangements, and financial independence (Bachman et al., 2002; Eitle, Taylor, & Eitle, 2010; Fleming, White, & Catalano, 2010; Kretsch & Harden, 2014; Staff, Greene, Maggs, & Schoon, 2014). These relations may be a result of both role selection and role socialization. In support of the importance of role selection, Yamaguchi and Kandel (1985) found that marijuana users were more likely to postpone marriage due to incompatible roles of marriage and substance use. Conversely, and consistent with role socialization, Gotham, Sher, and Wood (2003) found that being married reduced the likelihood of a later Alcohol Use Disorder. Labouvie (1996) examined both role selection and role socialization and found evidence supporting both mechanisms. Those who were married or became married and those who were parents or became parents had more friends who were married or parents at a later time point. At the same time, those who married or had children had lower levels of alcohol use at later time points. Further, Lee, Chassin, and MacKinnon (2015) found that those with heavier drinking patterns pre-marriage had the greatest decreases after marriage reflecting role related maturing out processes.

Employment has also been identified as a role that may be associated with maturing out, but research on this topic has provided less consistent evidence relative to research on marriage and parenthood (Gotham et al., 2003). For example, Bachman, Wadsworth, O’Malley, Johnston, and Schulenberg (1997) actually found small increases in alcohol use in the last 30 days among full-time employees, though this group did report small (though not significant) decreases in levels of binge drinking over time. Vergés et al. (2012) found dynamic effects of employment status on risk for alcohol dependence over time. They found that becoming employed reduced the likelihood of a continued diagnosis of alcohol dependence among older but not younger adult males, whereas becoming employed was predictive of the emergence of alcohol dependence for those without a previous diagnosis of alcohol dependence. In summarizing the overall pattern of findings regarding role transitions in their study, Vergés et al. (2012) concluded that “role transitions are associated with the course of alcohol disorders throughout the life course but that the effects differ as a function of age, sex, and type of transition” (p. 10).

Although selection and socialization into adult roles is a key component to maturing out of heavy drinking, there is evidence that the natural process of personality maturation may also play a unique role (Ashenhurst, Harden, Corbin, & Fromme, 2015; Littlefield, Sher, & Wood, 2009, 2010a, 2010b; Steinberg et al., 2008; Quinn & Harden, 2013). For example, impulsivity and sensation seeking increase during childhood, and early to middle adolescence, respectively (Steinberg et al., 2008), and contribute to experimentation with alcohol and other drugs. While impulsivity steadily decreases starting around age 10, sensation seeking peaks in mid adolescence and then declines into adulthood (Steinberg et al., 2008). Recent studies suggest that changes in personality are linked with changes in drinking behavior. For example, Quinn and Harden (2013) found that slower decreases in impulsivity were related to larger increases in alcohol and substance use, while slower decreases in sensation seeking were uniquely related to increases in alcohol use. In another longitudinal study, Littlefield, Sher, and Wood (2009) examined relations between personality traits of impulsivity, neuroticism, and extraversion and problematic alcohol involvement between the ages of 18 and 35. They found that decreases in impulsivity and neuroticism were associated with decreases in problematic alcohol use, even after accounting for effects of marriage and parenthood. In a recent study that examined effects of both role socialization and personality, there was evidence that effects depended upon the developmental period, with stronger effects of role transitions in early young adulthood and stronger effects of personality in later young adulthood (Lee, Ellingson, & Sher, 2015).

Literature on identity development may also inform our understanding of the process of maturing out of heavy drinking. Erikson’s seminal work (Erikson, 1968) suggests that identity exploration (identity versus identity confusion) is a key aspect of adolescent development, and alcohol and other drug use is a common way in which adolescents seek to explore their identities and appear more “adult”. The tendency for exploration may also be related to personality maturation (e.g. increases in sensation seeking) that occurs during this period (Schwartz, Donnellan, Ravert, Luyckx, and Zamboanga, 2013). Ultimately, young people must decide which behaviors to incorporate and which to move beyond (e.g., substance use) in establishing their identity. Moving beyond adolescence, Erikson suggests that identity development shifts to a focus on intimacy vs. isolation, a period during which individuals seek to develop long-term relationships. Thus, Erikson’s theory also has clear implications for selection of roles that contribute to the maturing out process.

Marcia (1966) extended the work of Erikson in his ego identity status paradigm, which facilitated empirical evaluation of Erikson’s theory. Based on a combination of levels of exploration and commitment, Marcia created four identity statuses; Achievement (high exploration, high commitment), Moratorium (high exploration, low commitment), Foreclosure (low exploration, high commitment), and Diffusion (low exploration, low commitment). Thus, he proposed that one must both explore possible identities and ultimately commit to beliefs, values, and long-term goals to successfully achieve desired outcomes. Supporting the potential link between identity development and substance use, recent studies by Bishop, Weisgram, Holleque, Lund and Wheeler-Anderson (2005) and Schwartz et al. (2011) found that individuals characterized by Marcia’s achievement identity status engaged in less substance use.

Recent influential theoretical work by Arnett (2000) builds upon earlier identity theories and argues for a new and distinct developmental period that has emerged from societal level changes. Arnett termed this period emerging adulthood, identifying young people between the ages of 18-25, though for some, this period may extend through the twenties (Arnett, 2005). Arnett argues that delays in marriage, parenthood, and other adult roles in industrialized societies has extended the transition to adulthood, creating a period “characterized by change and exploration” (Arnett, 2005, p. 479). Consistent with this model, over the past 50 years, the median age of marriage has increased by five years to age 27 for men and 25 for women, while simultaneously extending the age of first parenthood from the early twenties to the late twenties. Given that the period of emerging adulthood corresponds with peak risk for heavy drinking and related problems, Arnett’s theory may be of particular relevance to understanding the phenomenon of maturing out. In describing the period of emerging adulthood, Arnett proposed five dimensions of emerging adult identity (Arnett, 2005, 2006). A brief description of each dimension and its relation to earlier theories of identity development is provided below.

Identity exploration constitutes the extent to which an individual explores various feelings or experiences before reaching adulthood. Arnett (2006) explains that these experiences are the basis for life decisions. This dimension of Arnett’s theory is closely aligned with the focus on identity exploration in both Erikson’s and Marcia’s models of identity development. Arnett (2005) also considers emerging adulthood a time of experimentation/possibilities, when individuals have the liberty to make extensive changes in their lives. Although similar to identity exploration in some ways, Arnett clearly identifies this dimension of identity as being of particular relevance to the period of emerging adulthood. Whereas adolescents and children have limited choice in the manner in which they live, emerging adults can generally live the way they see fit. Thus, they are able to explore a variety of behaviors without impediment, including high risk behaviors like substance use.

The dimension of negativity/instability involves the extent to which young people are able to maintain confidence amid the many changes that occur during emerging adulthood. Geographic instability is common during this period with rates of moving increasing rapidly beginning at age 18 and peaking in the mid-twenties. For many, emerging adulthood also marks the first time they are able to completely govern themselves (e.g. live on their own). Although Erikson identified negativity and instability that can occur when young people fail to successfully navigate critical stages of identity development, Arnett suggests that these experiences are a normative part of identity development during this critical period.

Arnett also characterizes emerging adulthood as a period of freedom and self-focus. Ultimately, emerging adults must move from self-focus to other focus as they take on more adult social roles like marriage and full-time employment. This shift is characteristic of Erikson’s stage of intimacy vs. isolation and aligns with Marcia’s ideas about the importance of commitment. Finally, Arnett suggests that emerging adults report feeling as if they are not quite an adult, but not quite an adolescent, a dimension of identify development referred to as feeling in-between. Arnett (2000) demonstrated that the qualities most critical to these feelings involve challenges related to “accepting responsibility for one’s self, making independent decisions, and becoming financially independent” (p. 472-3). The concept of being caught in between adolescence and adulthood is perhaps the most unique to Arnett’s theory although it clearly captures the struggle to commit to adult roles.

In summary, Arnett’s theories build from prior models of identity development while extending them in important ways. Perhaps the most novel aspects involve the ideas of extended adolescence and the feeling of being caught between two important periods of development. Further, while prior theories tended to focus on one or two critical tasks for successful movement through stages of identity development, Arnett characterized multiple dimensions along which individuals might vary during the period of emerging adulthood without placing special emphasis on particular accomplishments for successfully navigating this developmental transition. Given high levels of alcohol use and abuse during emerging adulthood and conceptual links between emerging adult identity and other key determinants of maturation out of heavy drinking (e.g., personality maturation, adult role adoption), the goal of the current study was to examine relations between dimensions of emerging adult identity and alcohol-related outcomes.

Recent studies support the idea that Arnett’s dimensions of emerging adult identity may be important in understanding risk for substance use. For example, Allem et al. (2013) found a marginally significant relation between experimentation and past month marijuana use. In another study, Smith, Bahar, Cleeland, and Davis (2014) found significant bivariate correlations between feeling in between and substance use frequency and between negativity/instability and substance related problems. Finally, Lisha et al. (2014) found significant relations between a measure comprising items from experimentation and self-focus dimensions and cigarette, alcohol, and marijuana use. In the one longitudinal study we were able to identify, Little et al. (2013) found that lower levels of experimentation were associated with greater likelihood of marijuana cessation one year later. However, when controlling for baseline levels of marijuana use this effect became marginally significant, suggesting that the effect was due in part to the relation between lower experimentation and less marijuana use at baseline. Although these studies provide preliminary evidence for the importance of emerging adult identity in relation to substance use, the average ages in these studies ranged from 17 to 21 and none of the studies explored potential age-related differences in effects as participants moved further into emerging adulthood. Given evidence for important age-related differences in effects of role transitions (Vergés et al., 2012) and the relative influences of role transitions and personality maturation (Lee et al., 2015), this is an important gap in the literature.

Thus, the current study sought to build upon prior work by examining relations between dimensions of emerging adult identity and both heavy drinking and related problems using three waves of longitudinal data collected in the fall of the fourth year of college (~ 22 years of age) and the two years following (~23-24 years of age). To isolate unique influences of emerging adult identity, we controlled for effects of personality (sensation seeking and impulsivity) and role transitions (e.g. entry into committed relationships and employment). Although there is likely a trait component to emerging adult identity (e.g., some individuals are likely to be higher than others in experimentation or negativity across stages of development), we were most interested in understanding the extent to which within person variation (e.g. state aspects of emerging adult identity across time) relate to drinking behavior and problems. Thus, we examined emerging adult identity dimensions as predictors of deviations from expected age-related declines in heavy drinking and problems. Based on results of prior studies, we hypothesized that higher levels of experimentation/possibilities, self-focus, and feeling in-between would be related to heavier alcohol use and that higher levels of negativity/instability would be related to more alcohol-related problems. Based on the finding by Lee et al. (2015) that intrapersonal influences become more important than contextual influences as individual move further into young adulthood, we also expected effects of emerging adult identity to become stronger as individuals moved toward the end of the period of emerging adulthood. For example, high levels of experimentation may be less normative and therefore more indicative of a pathological process for a 24 year old who is more than a year beyond graduation than for a 22 year old who is still enrolled in college. Although we had specific hypotheses about only 3 of the 5 dimensions of emerging adult identity for heavy drinking and 1 of 5 for alcohol-related problems, we examined all 5 dimensions as predictors of heavy drinking and problems given the lack of prior longitudinal studies controlling for other known predictors of maturing out.

Method

Participants and Procedures

The “UT Experience!” originally recruited all incoming first time freshman at The University of Texas (UT) at Austin in the summer before matriculation. Data for the current analyses come from participants who were assigned to a longitudinal condition with 10 waves of data collection from summer prior to college entry until 2 years after senior year (N = 2,245; 60% female). Eligible participants were unmarried, first-time college students between the ages of 17 and 19. Of the total sample, 53.9% were Caucasian, 18.4% Asian American or Hawaiian Pacific Islander, 15.2% Hispanic/Latino, 4.1% African American, .1% American Indian/Alaska Native, and 6.7% multi-ethnic, 1.6% declined to report race/ethnicity. For more detailed information about recruitment, see Hatzenbuehler, Corbin, and Fromme (2008).

Of the 2,245 participants who completed the initial survey, a total of 1,857 completed one of the three surveys on which the current analyses were based (senior year of high school and the fall of the first and second year after graduation). These represented waves 8-10 of the original study but we refer to them as waves 1-3 in the remainder of the paper given that waves 1-7 of the original study were not included in the analyses. The sample was further restricted to those who graduated in four years to allow for examination of the transition out of college. Of the 1,857 participants who completed at least one survey across these three time points, 907 (48.9%) met inclusion criteria of completing college in four years (see Table 1 for descriptive statistics and Table 2 for correlations among all study variables). In order to be counted as four year graduates, participants had to report that they were graduating in their fourth year (wave 1) and confirm that they graduated at wave 2 (one year later). Thus, 48.9% is likely an underestimate of the true graduation rate as those who indicated that they were graduating at wave 1, but did not complete the survey at wave 2 were not counted as four year graduates. The average age in year 4 of college for the final sample of 907 was 21.76 (SD = .35), and the average age 2 years later was 23.76 (SD = .34).

Table 1.

Descriptive Statistics

Descriptive Statistics
Variable	N	M or % (SD)	Range
Gender
Male	316	34.80
Female	591	65.20
Age at Wave 1	907	21.76 (.35)	20.53-23.24
Relationship Status-Not in relationship
Wave 1	400	48.7
Wave 2	298	47.2
Wave 3	260	41.8
Relationship Status- In a committed relationship
Wave 1	385	46.9
Wave 2	267	42.2
Wave 3	266	42.8
Relationship Status- Married or engaged
Wave 1	36	4.4
Wave 2	67	10.6
Wave 3	96	15.4
Employment Status- Unemployed
Wave 1	349	38.6
Wave 2	111	16.1
Wave 3	107	16.3
Employment Status- Part-time
Wave 1	527	58.3
Wave2	148	21.5
Wave 3	78	11.9
Employment Status- Full-time
Wave 1	28	3.1
Wave 2	429	62.4
Wave 3	472	71.8
IDEA- Experimentation
Wave 8	862	3.10 (.68)	1-4
Wave 9	664	3.00 (.71)	1-4
Wave 10	635	2.89 (.81)	1-4
IDEA- Self-Focus
Wave 8	863	3.37 (.57)	1-4
Wave 9	663	3.37 (.68)	1-4
Wave 10	635	3.39 (.64)	1-4
IDEA- Negativity/Instability
Wave 8	864	3.09 (.71)	1-4
Wave 9	665	2.83 (.84)	1-4
Wave 10	634	2.82 (.83)	1-4
IDEA- Identity Exploration
Wave 8	862	3.40 (.65)	1-4
Wave 9	662	3.24 (.76)	1-4
Wave 10	635	3.21 (.76)	1-4
IDEA- Feeling in-between
Wave 8	861	3.30 (.64)	1-4
Wave 9	660	3.14 (.77)	1-4
Wave 10	633	3.04 (.82)	1-4
Binge Drinking
Wave 8	905	4.25 (6.79)	0-54
Wave 9	689	3.58 (6.58)	0-50
Wave 10	654	3.09 (5.37)	0-45
Drunk
Wave 8	905	3.75 (6.28)	0-60
Wave 9	689	2.92 (5.47)	0-60
Wave 10	653	2.78 (5.01)	0-47
Alcohol-related problems
Wave 8	892	2.84 (5.55)	0-54
Wave 9	682	1.80 (3.85)	0-31
Wave 10	649	1.83 (4.53)	0-43

Table 2.

Correlations among All Variables in the Model

Measure	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Gender	1	−.01	.02	−.08	−.04	.01	−.02	−.12**	−.08	−.14***	.14***	.12**	.06	.05
2. Employment Status 3A		1	−.59***	−.04	−.05	.03	−.08*	.15***	−.05	.06	.01	.03	.06	.12**
3. Employment Status 3B			1	.07	.07	−.01	.10**	−.23***	−.02	−.10*	−.06	−.04	.00	−.05
4. Relationship Status 3A				1	−.37***	−.02	.06	.02	.04	.07	.03	−.06	−.04	−.04
5. Relationship Status 3B					1	−.24***	−.17***	.00	−.13**	−.17***	−.15**	−.11**	−.19***	−.16***
6. Experimentation						1	.46***	.16***	.47***	.32***	.23***	.17***	.17***	.12***
7. Self-Focus							1	.08*	.41***	.22***	.04	−.01	.18***	.10***
8. Negativity/Instability								1	.18***	.26***	.11*	−.03	.05	.14***
9. Identity Exploration									1	.52***	.09*	.03	.08	.05
10. Feeling in Between										1	.04	.08*	.05	.07
11. Sensation Seeking											1	.50***	.21***	.18***
12. Impulsivity												1	.15***	.18***
13. Heavy Drinking													1	.61***
14. Alcohol-related Problems														1

Demographics

Basic demographics variables included gender, ethnicity/race, and age.

Predictor Variables

Dimensions of Emerging Adult Identity

Emerging adulthood identity was assessed using the Inventory of the Dimensions of Emerging Adulthood (IDEA; Reifman et al., 2007). The inventory includes 31 items assessing 5 dimensions including; experimentation/possibilities, self-focus, negativity/instability, identity exploration, and feeling in-between. A sixth dimension (other-focused), serves as a counterpoint to self-focus but was not included in the current study (Reifman et al., 2007). All items are measured on a four point Likert type scale with response options ranging from “strongly disagree” to “strongly agree.” In the current study, fifteen of the 31 original items were included to capture the 5 original dimensions using the 3 highest loading items from each subscale. Sample items include: “A time of learning to think for yourself” and “a time of trying out new things.” In the original measurement development, internal consistency coefficients (alpha) ranged from .70 to .85 and test-retest reliability ranged from .64 to .76. Internal consistency reliability in the current study was good (.76 to .87 at wave 1).

Relationship Status

Relationship status was assessed with a question that provided 7 response options including: “not dating,” “dating but not exclusively,” “dating exclusively,” engaged,” “married,” “other,” and “divorced.” In an attempt to avoid exclusion of participants, we attempted to reclassify those who provided a response of “other” into one of the six categories based on the information provided by the participant. We were able to recode 5 of 13 at wave 1, 3 of 8 at wave 2, and 4 or 6 at wave 3. Participants who indicated that they were “divorced” or who provided a response of “other” that could not be recoded into one of the other categories were excluded from the analyses (n=15). For purposes of this study, responses were reduced to three categories: “not dating” and “dating, but not exclusively” were combined into a group that was not in a committed relationship; “dating exclusively” comprised the second class of participants who were in a committed relationship but not engaged or married; “engaged,” and “married” were combined into the final group. Two dummy codes were created to contrast participants who were not in a committed relationship against those who were a) in a committed relationship but not engaged/married and b) engaged/married.

Employment Status

Employment status was assessed with a question that included 3 response options: no, part-time, and full-time. As with relationship status, two dummy codes were created contrasting those who were not employed against those who were a) employed part-time, and b) employed full-time.

Sensation Seeking and Impulsivity

Sensation seeking and impulsivity were assessed at each wave using 19 true-false items from the Zuckerman-Kuhlman Personality Questionnaire (ZKPQ; Zuckerman, Kuhlman, Joireman, Teta, Kraft, 1993). Sensation seeking is characterized by the willingness to take risks for the sake of varied and intense experiences and impulsivity reflects acting without forethought. In the present study, internal consistency reliabilities were good for both sensation seeking (alphas of .79 to .81) and impulsivity (alphas of .74 to .79).

Outcome Variables

Heavy Drinking

Heavy drinking was assessed with two items: 1) frequency of getting drunk (not just a little high) and 2) frequency of binge drinking. Frequency of getting drunk was assessed with the single item: “During the past 3 months, how many times did you get drunk (not just a little high) on alcohol?” (Jackson, Sher, Gotham, & Wood, 2001). Binge drinking was also assessed with a single item: “During the past three months, how many times did you have four/five (for women/men) or more drinks at a sitting?” (Wechsler & Isaac, 1992). The mean of the two measures was computed as a composite measure of heavy drinking.

Alcohol problems

Alcohol-related problems were assessed using the Rutgers Alcohol Problem Index (RAPI; White and Labouvie, 1989). The measure includes 23 items assessing the frequency of negative consequences resulting from drinking in the prior 3 months. Example items include: “relatives avoided you,” “Missed a day (or part of a day) of school or work,” and “not able to do your homework or study for a test.” Items were scored on a 1 (0 times) to 5 (more than 10 times) scale. The internal consistency reliability for the scale in the current study was .91 at wave 1 (year 4).

Data Analytic Plan

Prior to conducting the primary analyses, distributions of all variables were examined for non-normality and transformations were conducted as necessary. For the primary analyses, time varying covariate growth models were conducted in Mplus 7.0 (Muthén & Muthén, 2013) using bootstrapping (5,000 bootstraps) and Full Information Maximum Likelihood (FIML) estimation to manage missing data. Initially, we tested an unconditional growth model in which loadings from the intercept latent variable to heavy drinking at each wave were set to one. The latent variable for the slope represented linear growth in heavy drinking over time, with loadings set to 0, 1, and 2 for waves 1-3 respectively. In addition to examining fit of the unconditional growth model, we examined residual variances for heavy drinking at each wave to ensure that there was sufficient residual variability to allow for meaningful prediction within our time-varying covariate (TVC) models. We also examined means for heavy drinking and alcohol problem slopes to verify expected age-related decreases in these outcomes.

Next we proceeded to testing the TVC models. We initially entered gender as a time invariant predictor of the intercept and slope, and the five IDEA scores as the only time varying predictors of residual variability in heavy drinking at each wave. This allowed us to examine the extent to which IDEA scores predicted time-specific deviations from the expected trajectory of heavy drinking based on the underlying growth model. Next, we tested models with the addition of time-varying covariates including relationship status, employment status, sensation seeking, and impulsivity, to determine unique effects of emerging adult identity, accounting for other well established correlates of maturing out. Correlations among the time-specific predictors and between gender and the time-specific predictors were freely estimated. Within both unadjusted and covariate adjusted models, we first compared models that constrained IDEA effects to equality to models that allowed effects to vary across time. A significant difference in model fit indicated that the magnitude of effects differed across time. In the absence of evidence for time specific effects, we presented only averaged effects across time. If model fit differed between constrained and unconstrained models, we presented the magnitude of effects at all time-points. The same procedures were followed for the alcohol-related problems outcome variable.

Results

Evaluation of Variable Distributions

The heavy drinking composite measure and alcohol-related problems were positively skewed (skewness values ranging from 2.53 to 4.69) and were therefore log transformed. Resulting skewness values suggested that the log-transformations were successful in normalizing the distributions of these variables (all skewness values < 1.5).

Growth Models for Heavy Drinking

Unconditional Model

The unconditional growth model for heavy drinking displayed excellent model fit, χ² (1 df) = .66, p = .41; comparative fit index (CFI) = 1.00; root mean square error approximation (RMSEA) = 0.00, Standardized Root Mean Square Residual (SRMR) = .01. The mean for the slope was −.05 (SE = .008), p < .001, indicating significant decreases in heavy drinking over time, and examination of residual variability in heavy drinking at each wave indicated that there was significant remaining variability that might be accounted for by the predictor variables in the TVC models (all p values < .001).

Time-Varying Model without Covariates

The growth model for heavy drinking with only the IDEA scores as time variant predictors displayed excellent fit, χ² (32 df) = 37.12; CFI = 1.0; RMSEA = .01, SRMR = .03. Gender was not a significant predictor of either intercept or slope (p values > .53). Tests of constrained (constraining effects of individual IDEA scores to equivalence across time) vs. unconstrained models found no significant effects (all p values > .35). Thus, averaged effects across time were examined. Note that ranges are provided for standardized effects as standardized coefficients vary somewhat even in constrained models due to differences in the variances of the predictor and outcome measures across time. Experimentation/possibilities emerged as a significant risk factor, such that those with higher levels of experimentation were heavier drinkers than would be expected based on their underlying trajectories (b = .067, SE = .016, p < .001). Standardized coefficients ranged from .08 to .11. Self-focus also emerged as a significant risk factor (b = .053, SE = .017, p = .002), with standardized coefficients ranging from .06 to .07. Those with higher levels of self-focus reported heavier drinking than would be expected based on their underlying trajectories. Effects of negativity/instability, identity exploration, and feeling in between were not statistically significant (p values > .34).

Time Varying Model with Covariates

The model with the addition of covariates provided good fit to the data, χ² (68 df) = 114.44; CFI = .96; RMSEA = .03, SRMR = .02. With respect to the time-varying covariates, relationship status emerged as a protective factor and significantly predicted residual variability in heavy drinking at all 3 waves. Relative to those not in a relationship, individuals who were in a relationship but not engaged or married reported less drinking at all three time points (standardized coefficients = −.06, −.07, and −.14, respectively) than would be expected based on their underlying trajectory (all p values < .05). Effect sizes were similar when comparing those who were married or engaged to those not in a relationship (standardized coefficients = −.09, −.09, and −.12; all p values <= .001). Employment status (part time and full time employment each contrasted against no employment) was not a significant predictor of heavy drinking at any wave (all p values > .18). Although impulsivity did not predict residual variability in heavy drinking at any wave (p values > .28), sensation seeking was associated with residual variability in heavy drinking at all waves (standardized coefficients = .13, .12, and .08; p values < .05).

As in the unadjusted models, tests of constrained vs. unconstrained models identified no time-varying effects of the IDEA scores (all p values > .22). Thus, averaged effects of IDEA scores are reported here. See Figure 1 for standardized coefficients for all time-specific effects from the unconstrained model. Accounting for significant effects of relationship status and sensation seeking, experimentation remained a significant predictor (b = .044, SE = .166, p = .007), though the magnitude of the effect was reduced (standardized coefficients of .06 to .07) relative to the unadjusted model (.08 to .11). Self-focus also remained a significant risk factor (b = .053, SE = .017, p = .002), and the standardized effects (.06 to .07) were not reduced with the addition of the covariates. Effects of negativity/instability, identity exploration, and feeling in between remained non-significant when accounting for effects of the covariates (all p values > .38).

Figure 1.

Time-Varying Covariate Model for Heavy Drinking. Imp = Impulsivity, SS = Sensation Seeking, Emp Xa = Unemployed vs. Part-time Employed, Emp Xb = Unemployed vs. Full-time Employed, Rel Xa = Not in a Relationship vs. Committed Relationship but not Engaged/Married, Rel Xb = Not in a Relationship vs. Engaged/Married, IE = Identity Exploration, SF = Self Focus, EXP = Experimentation/Possibilities, FIB = Feeling in-between, NEG = Negativity/Instability.

Growth Models for Alcohol-related Problems

Unconditional Model

The unconditional growth model for alcohol-related problems displayed good model fit based on most fit indices, CFI = .99, SRMR = .02. However, the chi-square test was significant, χ² (df = 1) = 9.54, =.002, and the RMSEA indicated marginal fit with a value of .10. The mean for the slope was −.053 (SE = .007), p < .001 indicating significant decreases in alcohol-related problems over time, and there was significant variance in alcohol-related problems at each wave that was not accounted for by the underlying trajectories.

Time Varying Model without Covariates

The TVC growth model for alcohol-related problems without covariates displayed excellent overall model fit, χ² (32 df) = 44.37, p = .07; CFI = .98; RMSEA = .02, SRMR = .03. The time invariant covariate of gender was not a significant predictor of the intercept or slope for alcohol-related problems (p values > .52). With respect to IDEA scores, comparison of a model that constrained effects of negativity to equality across the three time-points and one that freely estimated time-specific effects yielded a significant difference in model fit, Δχ² (2 df) = 7.02, p = .03. Thus, effects for negativity/instability were examined separately at each time point. Negativity/instability emerged as a significant predictor at wave 3 (standardized coefficient = .14, p < .001) but not at either wave 1 (standardized coefficient = .02, p = .52) or wave 2 (standardized coefficient = .05, p = .16).

Tests of constrained vs. unconstrained models did not identify significant differences for any of the other IDEA scores (all p values > .25), indicating a lack of time-specific effects. Thus averaged effects across time are reported for these variables. See Figure 2 for standardized coefficients for all time-specific effects from the unconstrained model. Experimentation was a significant predictor of alcohol-related problems (b = .042, SE = .012, p = 001), with standardized effects ranging from .07 to .09. Participants with higher levels of experimentation were at heightened risk for alcohol-related problems than would be expected based on the underlying trajectory. Effects of self-focus, identity exploration, and feeling in between were not statistically significant (all p values > .49).

Figure 2.

Time-Varying Covariate Model for Alcohol-related Problems. Imp = Impulsivity, SS = Sensation Seeking, Emp Xa = Unemployed vs. Part-time Employed, Emp Xb = Unemployed vs. Full-time Employed, Rel Xa = Not in a Relationship vs. Committed Relationship but not Engaged/Married, Rel Xb = Not in a Relationship vs. Engaged/Married, IE = Identity Exploration, SF = Self Focus, EXP = Experimentation/Possibilities, FIB = Feeling in-between, NEG = Negativity/Instability.

Time Varying Model with Covariates

Although the Chi-square test was significant in the model including covariates, χ² (68 df) = 108.43, p = .001, all other indicators suggested good model fit; CFI = .95; RMSEA = .03, SRMR = .02. Effects of relationship status were generally consistent with the heavy drinking model, though effect sizes were smaller, and only engagement/marriage was significantly protective at waves 2 and 3. Effects of employment status were inconsistent, with full-time employment associated with reduced risk for alcohol problems at wave 2 only (standardized coefficient = −.09), and part-time employment associated with increased risk for alcohol problems at wave 3 only (standardized coefficient = .09). Although both sensation seeking and impulsivity were associated with greater risk for alcohol-related problems at all waves (standardized coefficients ranging from .02 to .12), effects for sensation seeking were significant only at waves 1 and 2, and effects of impulsivity were significant only at waves 1 and 3.

As in the model without covariates, comparison of fit between models that constrained and freely estimated the negativity/instability paths identified a significant difference in model fit, Δχ² (2 df) = 9.39, p = .01. As in the unadjusted model, the effect of negativity/instability was significant at wave 3 (standardized coefficient = .15, p < .001) but not at wave 1 (standardized coefficient = .02, p = .48) or wave 2 (standardized coefficient = .03, p = .43). Comparisons of constrained and unconstrained models for the other IDEA scores were not significant (all p values > .42), indicating a lack of time specific effects. Averaged across time, the effect of experimentation was reduced to marginal significance (b = .024, SE = .013, p = .06) with the inclusion of the covariates, and the standardized coefficients were substantially reduced (.04 to .05) relative to the unadjusted model (.07 to .09). Effects of self-focus, identity exploration, and feeling in between remained non-significant when accounting for effects of the covariates (all p values > .48).

Discussion

The aim of this study was to examine dimensions of emerging adult identity in relation to drinking outcomes during the transition out of college, controlling for the influences of personality and role transitions. Specifically, we predicted that experimentation, self-focus, and feeling in-between would predict heavier drinking and that negativity/instability would predict more alcohol-related problems, with stronger effects emerging with time. The results of the TVC growth models provided some support for study hypotheses.

The experimentation/possibilities dimension of emerging identity was a significant predictor of both heavy drinking and alcohol-related problems and these effects were consistent across time. However, when controlling for effects of covariates, standardized effects were reduced by 25% to 50%, and became non-significant (p = .06) for alcohol-related problems. Significant effects of self-focus on alcohol use were also consistent across time. However, unlike experimentation, the magnitude of the effects was unchanged by the inclusion of the covariates. Negativity/instability was the only facet of emerging adult identity that demonstrated differential effects across time, emerging as a significant predictor of alcohol-related problems at only wave 3. As with self-focus, the magnitude of this relation was consistent across models with and without covariates. Whereas at least partial support was found for other study hypotheses, feeling in-between was not a significant predictor at any wave, calling into question its relevance as a risk factor for heavy drinking and related problems.

In many ways, the findings of the current study were consistent with those of prior studies demonstrating links between experimentation and use of alcohol, tobacco, and marijuana (Allem et al., 2013; Lisha et al., 2014; Little et al., 2013). However, it is important to note that ours is the first to examine these relations using longitudinal data and to examine differences in the magnitude of effects across an important developmental transition. Further, we controlled for other well-established predictors of maturing out including employment, relationship status, sensation seeking, and impulsivity, and this aspect of the study design shed new light on findings of prior studies. Effects of experimentation were reduced by one-fourth to one-half with the inclusion of the covariates, and effects on alcohol-related problems were no longer significant.

Examination of zero order correlations suggests that the overlap was largely between experimentation and the personality variables. Within time-point correlations between experimentation and sensation seeking ranged from .26 to .30 and correlations with impulsivity ranged from .14 to .18. Thus, prior studies demonstrating effects of experimentation likely overestimated the unique influence of emerging adult identity given the overlap with other known personality predictors of risk. This is not to say that personality risk is important and emerging adult identity is not as there may be important interplay among these risk/protective factors. For example, maturation in personality with brain development in emerging adulthood may set the stage for emerging adult identity development, adoption of adult roles, and subsequent maturing out of heavy drinking. Prior studies have shown that other psychosocial risk factors like positive alcohol expectancies serve to mediate effects of trait personality risk on drinking outcomes (Corbin, Iwamoto, & Fromme, 2011). Thus, it is possible that sensation seeking and impulsivity have indirect effects on drinking behavior through a delayed or slower process of emerging adult identity development. This is an interesting question for future research.

While small, effects of self-focus on heavy drinking were not impacted by other known predictors of maturing out. This is interesting as one might have expected a reduction in effects of self-focus after accounting for relationship status. Although the results suggest that effects of self-focus are not accounted for by role transitions, they are consistent with the idea that committed relationships might indirectly contribute to reductions in alcohol use/problems through reduced self-focus. Although testing this type of mediated effect was beyond the scope of the current study, the zero-order correlation between marriage and self-focus was robust (r = −.24), and effects of self-focus remained when accounting for direct effects of marriage. Thus, pre-requisites for testing an indirect effect were met and deserve attention in future studies.

The link between negativity/instability and alcohol-related problems is highly consistent with theoretical models of risk which highlight the role of drinking to cope with negative emotion as a risk factor for alcohol-related problems (Corbin, Farmer, Nolen-Hoekesma, 2013; Cooper, Russell, Skinner, Frone, & Mudar, 1992). Our results are also consistent with those of Smith et al. (2014) who found that negativity/instability was significantly correlated with substance-related problems. Further, Luyckx, De Witte, and Goossens (2011) found that negativity/instability significantly predicted depressive symptoms, which have been shown to increase risk for alcohol-related problems (Gonzalez, Reynolds, & Skewes, 2011). The fact that negativity/instability did not emerge as a significant predictor until wave 3 is also interesting. It may be that stresses associated with the transition out of college (e.g. new work demands, development of new social networks) contribute to this association. It will be particularly important to examine effects of negativity/instability on alcohol problems in later adulthood and among individuals reporting clinically significant alcohol problems.

Although time-specific effects were demonstrated for negativity/instability, we found no additional support for our hypothesis that effects of emerging adult identity would increase with time. The attenuation of the effects of experimentation when controlling for personality risk suggests that this aspect of emerging adult identity may be closely tied to personality maturation which is typically more pronounced in adolescence (at least for sensation seeking; See Steinberg et al., 2008). Thus, evidence for time-specific effects of experimentation may be more likely during the transition into rather than out of emerging adulthood. With respect to self-focus, effects did not increase with time despite major role changes (e.g., entry into committed relationships), and significant relations between these role changes and self-focus. This suggests that, although there may be normative changes in self-focus with the adoption of adult social roles, there may also be a more stable, trait-like aspect of self-focus that confers risk for alcohol use, regardless of developmental stage. Of course, further studies of self-focus across additional developmental periods are needed to directly test this hypothesis.

Whereas the findings have potentially important implications, they must be considered in light of several limitations. Participants were from a single university, a large percentage of participants were excluded from analyses because they did not graduate in four years, and participants were roughly age 24 at the last wave. Thus, it is not clear the extent to which findings will generalize to college students more broadly and to later stages of the transition out of emerging adulthood. The later emerging effects of negativity in the current study highlight the need for future studies that extend into later stages of emerging adulthood. The analyses also relied entirely on self-report data. Although self-reports of high risk behaviors like alcohol use may be biased, several studies have shown that self-reports of alcohol use are reliable (Babor, Steinberg, Anton, & Del Boca, 2000; LaForge, Borsari, & Baer, 2005) and do not generally represent underestimates of actual use.

It is also important to acknowledge that, although the analyses were longitudinal, relations between the time-varying predictors and residual variance in drinking and problems were cross-sectional. Thus, it is quite possible that alcohol use/problems at a given time-point would also predict deviations away from the underlying trajectory of emerging adult identity development. Such reciprocal relations would be quite consistent with the literature on relations between personality variables and drinking outcomes (Littlefield, Sher, & Wood, 2009). Our analytic approach also did not examine how changes in emerging adult identity relate to changes in drinking outcomes. If there is considerable variability in rates of emerging adult identity development across individuals, it will be important to understand the extent to which these differences are associated with changes in drinking outcomes. It is also possible to model relations between growth processes and time-specific residual effects simultaneously (e.g. Autoregressive Latent Trajectory models; Bollen & Curran, 2004), or to simultaneously model separate between and within-group processes (Latent Curve Models with Structured Residuals; Curran, Howard, Bainter, Lane, & McGinley, 2014), which represent important directions for future research. Unfortunately, with only 3 data points and a large number of TVCs, such models were beyond the scope of the current study.

It is also important to note the small magnitudes of the significant effects observed in this study, and the fact that we examined a large number of parameters without control for multiple comparisons. Although we tested effects of each of the five IDEA subscales on both alcohol use and related problems, we had a priori hypotheses for only a subset of these parameters, many of which were supported. Findings for more exploratory analyses require future replication before we can draw firm conclusions. Regarding the small magnitude of the effects of emerging adult identity dimensions, measurement limitations may have been a contributing factor. In the context of a large battery of survey measures, we were only able to include 3 items for each IDEA dimension. Although internal consistency reliability was good for these brief measures, effects may have been larger if the full IDEA measure was included. Our measure of employment status was also limited in that we did not capture the types of jobs in which participants were working. It seems likely that different types of jobs would be differentially related to risk for or protection against heavy drinking and related problems. Thus, we may have underestimated effects of both emerging adult identity and employment status in the current study.

Finally, given the attenuation of relations between emerging adult identity facets and drinking outcomes when controlling for effects of personality, it is important to recognize that we were unable to control for other personality variables of interest. Neuroticism and negative urgency have been linked to changes in drinking behavior during emerging adulthood (Littlefield, Sher, & Wood, 2009, 2010a, 2010b; Settles, Cyders, & Smith, 2010) and it seems quite possible that accounting for these personality variables might attenuate relations between negativity/instability and alcohol-related problems in the same way that controlling for sensation seeking and impulsivity attenuated relations between experimentation and alcohol use. Including these personality variables represent an important future direction as our results suggest that prior studies may have over-estimated the unique effects of experimentation on substance use.

Despite the aforementioned limitations, the current study adds to the literature by demonstrating complex relations between emerging adult identity and drinking outcomes. The longitudinal nature of the study and control for other known predictors of maturing out clarify our understanding of the unique effects of emerging adult identity across a critical developmental transition. Nonetheless, longer longitudinal studies are needed to determine if effects persist or perhaps strengthen as individuals move further into young adulthood. It will also be important to examine the extent to which changes in emerging adult identity relate to changes in drinking behavior. Finally, it will be important to control for additional personality variables (e.g., neuroticism and negative urgency) and to identify potential mechanisms (e.g. drinking motives) through which emerging adult identity facets contribute to drinking outcomes. We hope the results of the current study will encourage others to examine these important questions.