Correcting Exaggerated Drinking Norms with a Mobile Message Delivery System: Selective Prevention with Heavy Drinking First-Year College Students

Abstract

Many efficacious interventions designed to reduce college student drinking aim to correct misperceptions of peers’ drinking behavior. The present study tests the efficacy of a novel delivery strategy, namely text messages, for promoting pro-moderation descriptive and injunctive drinking norms. Participants included 121 college students who were randomly assigned to receive daily text messages containing accurate drinking norms (experimental group, n=61) or historical facts (control group, n =60) for 10 weeks following a baseline assessment. Participants completed 3-month post-baseline and 6-month post-baseline follow-up assessments. The 3-month assessment revealed that pro-moderation text messages were effective for reducing peak consumption and alcohol consequences. Changes in descriptive norms and injunctive norms aligned with these two behavioral outcomes. The intervention group reported perceiving others as drinking less on their heaviest drinking day and perceived others as being less approving of alcohol-related consequences than the control group. The intervention group also reported more peer approval of using protective behavioral strategies. Yet, intervention effects were not maintained. None of the outcome measures differed by condition at the 6-month post-baseline assessment. Thus, the intervention had short-term effects on self-reported drinking behavior as well as on perceptions of drinking norms. However, the behavioral changes were not maintained when participants were assessed in the second semester after the daily text messages intervention had stopped.

The first year in college is a time of transition as students learn how to socialize in a new environment. The theory of emerging adulthood (ages 18-25) indicates that this time of life includes the important developmental task of exploring and crafting an adult identity (Arnett, 2005). For many emerging adults who attend college, this exploration takes place during socialization with peers and may involve experimentation with alcohol and other drugs. Indeed, the transition to college is associated with both onset of and increases in high-risk drinking (Borsari, Murphy, & Barnett, 2007; Fromme, Corbin, & Kruse, 2008). Because high-intensity drinking (8+/10+ drinks for women/men) (Patrick, Terry- McElrath, Kloska, & Schulenberg, 2016), and related consequences such as blackouts (N. P. Barnett et al., 2014) and sexual assault (Krebs, Lindquist, Warner, Fisher, & Martin, 2009) occur at alarming levels in the first year, this is a priority time for college alcohol prevention.

Social norms provide informative guideposts during times of transition. One of the strongest correlates of high-risk drinking among young college-attending adults is perceived norms. Descriptive norms (DN, the perception of what others do) and injunctive norms (IN, the perception of what others approve of) are positively associated with drinking (Borsari & Carey, 2003; Lee, Geisner, Lewis, Neighbors, & Larimer, 2007), and often account for more variance in drinking behavior than other established cognitive predictors such as expectancies and motives (Neighbors, Lee, Lewis, Fossos, & Larimer, 2007). Perceived drinking norms also predict alcohol consumption over time. In particular, both IN (Larimer, Turner, Mallett, & Geisner, 2004; Mollen, Rimal, Ruiter, Jang, & Kok, 2013) and DN (Wardell & Read, 2013), prospectively predicted drinking, controlling for baseline drinking. Further, there is evidence that college student drinking conforms to perceived norms over time (Lewis, Litt, & Neighbors, 2015; Wardell & Read, 2013).

A critical element of the relationship between perceived norms and behavior is the reliable finding that perceived norms tend to be exaggerated relative to actual norms. Specifically, estimates of others’ drinking (DN) usually exceed reports of one’s own behavior, and estimates of others’ approval of drinking behaviors (IN) are usually more permissive than one’s own attitudes (Borsari & Carey, 2003). Self-other discrepancies in the perceived approval of campus drinking (Neighbors et al., 2008; Suls & Green, 2003) and drinking consequences (DeMartini, Carey, Lao, & Luciano, 2011; Lewis et al., 2010) have been documented as well. In contrast, students endorse more personal approval of protective behavioral strategies (i.e., behaviors designed to mitigate the adverse consequences of drinking) than they ascribe to others (DeMartini et al., 2011). The observation that privately held attitudes tend to be more conservative than perceived peer attitudes (i.e., “pluralistic ignorance”), is seen in the context of other risk behaviors as well, including smoking, drug use, and hooking up (Hines, Saris, & Throckmorton- Belzer, 2002; Lambert, Kahn, & Apple, 2003).

These discrepancies can have adverse effects on individuals and the community. Self-other differences in DN predict increased drinking over time, suggesting that students conform to their (mis)perception that peers are engaging in heavier drinking than they really are (Carey, Borsari, Carey, & Maisto, 2006). Perceptions of self-other differences in IN can also serve to perpetuate a permissive drinking environment, whereby individual students who do not share the perceived approval of excessive drinking feel in the minority (Miller & Prentice, 2016), and those holding pro-moderation attitudes do not express their opinions for fear of social isolation (Glynn, Hayes, & Shanahan, 1997). Over time opinions perceived to be in the majority are expressed while those perceived to be in the minority are not (Matthes, 2014). Exaggerated perceived norms are thereby perpetuated and continue to influence drinking decisions.

Ample evidence supports the efficacy of interventions designed to correct exaggerated DN. Personalized normative feedback (delivered via computer or in person) is included in many efficacious interventions designed to reduce college student drinking (Carey, Scott-Sheldon, Carey, & DeMartini, 2007; Carey, Scott-Sheldon, Elliott, Garey, & Carey, 2012; Cronce & Larimer, 2011; Lewis & Neighbors, 2006). Nearly all these studies present accurate DN to correct misperceptions of peer drinking behavior, in one or two exposures. Importantly, mediation analyses support DN as a mechanism of change, such that interventions that successfully reduce elevated DN lead to lower alcohol consumption (Reid & Carey, 2015).

A growing body of literature suggests that correcting exaggerated IN may also be a viable prevention strategy, yet comparatively fewer prevention interventions have employed IN feedback with the goal of reducing risky drinking. Two early studies used facilitator-led group discussions with mixed results (L. A. Barnett, Far, Mauss, & Miller, 1996; Schroeder & Prentice, 1998). A recent review of college drinking interventions found weak evidence for mediation by IN (Reid & Carey, 2015), but most of the reviewed interventions did not attempt to change IN, and those that included IN manipulations failed to successfully change IN (i.e., the “a” path in mediation). Thus, strong tests of the potential for a successful IN manipulation to facilitate change in drinking are lacking. Other studies have demonstrated the malleability of IN. For example, one study demonstrated that IN changed immediately following presentation of corrective information about college students’ attitudes about drinking (Prince & Carey, 2010). Similarly, an RCT that delivered face-to-face IN feedback on peer approval of consequences reduced perceptions of IN and consumption and consequences at a 1-month follow-up, relative to an assessment-only control (Prince, Maisto, Rice, & Carey, 2015). Taken together, correcting inaccurate IN remains a promising direction for college alcohol abuse prevention.

The idea that addressing both DN and IN within one intervention has potential to change norms-driven behavior is supported by theory. The theory of normative social behavior (Rimal & Real, 2005) holds that the influence of DN on behavior is greatest in the context of supportive IN. Thus, norms feedback is most persuasive when DN and IN align in presenting a consistent message (Reid, Cialdini, & Aiken, 2010). The focus theory of normative conduct (Cialdini, Kallgren, & Reno, 1991) holds that norms are likely to influence behavior when they are made a salient focus of attention, which is possible when normative feedback is repeated over time. Thus, correcting exaggerated DN and IN has a sound theoretical basis as a prevention strategy, but this combination strategy has been underutilized in alcohol abuse prevention interventions.

The first year of college is an optimal time to deliver a corrective norms intervention, because impressions of campus-specific norms are just forming. The new college environment provides multiple sources of information about the prevalence and acceptability of drinking, due to selective exposure to and sharing of pro-drinking norms (Kitts, 2003). Prevention strategies that counteract the formation of exaggerated DN and IN during the formative first year may reduce the perceived peer expectations for high risk drinking, which may reduce future harms. Therefore, the purpose of this study is to evaluate a novel approach to delivering a corrective norms intervention prevention to first year students.

The typical method for correcting exaggerated drinking norms involves one or two doses of personalized normative feedback delivered by computer or in-person. We implemented an alternate method of delivering corrective information about DN and IN: text messages consisting of accurate campus-based data on peer drinking behaviors and attitudes, experiences of alcohol-related consequences, as well as peer use and endorsement of protective behavioral strategies. This approach takes advantage of the fact that text messaging affords nearly universal reach among young adults (Pew Research Center, 2019), and offers multiple exposures to pro-moderation content to compete with exposure to risky drinking and peer approval that maintains exaggerated unhealthy norms. A small-scale pilot test demonstrated the feasibility and acceptability of daily text messages containing generic drinking norms over 28 days with second year students (Merrill, Boyle, Barnett, & Carey, 2018), but no statistically significant changes in drinking behavior were observed. The present study provides a stronger test with longer exposure, campus-specific norms, and a larger sample of students getting adjusted to college.

Our primary hypothesis was that regular exposure to campus-specific pro-moderation norms delivered daily via text messages during the first semester of college would result in lower alcohol consumption and problems than a structurally equivalent attention control condition. Outcomes for this preliminary efficacy trial were assessed at three-months (post-intervention) and six-months (follow-up) after baseline. Secondary hypotheses predicted changes in hypothesized mediators of the intervention. Specifically, because they were represented in the content of the text messages, we predicted that perceived DN and IN related to each drinking domain presented (i.e., consumption, use of protective behavioral strategies, and consequences) would change in the direction of pro-moderation norms in the experimental group but not in the control group at the three-month post-intervention assessment.

Method

Participants and Recruitment

Participants (N= 121) were first-semester students at a residential 4-year university in the northeastern United States. Eligibility criteria consisted of (a) first-year students, (b) being between 18 and 20 years of age, (c) meeting NIAAA criteria for risky drinking (for men, >4 drinks in a day or >14 in a week; for women >3 drinks in a day or >7 in a week) in the past 30 days, (c) possession of a mobile phone with text messaging capacity, and (d) use of text messaging at least weekly. Participants were excluded if they reported (a) being in treatment for any substance use disorder or it had been recommended to them that they should seek treatment for drug and alcohol use, or (b) plans to be out of the country or otherwise without cellular service for more than 3 days during the fall semester. The University’s Institutional Review Board approved study procedures.

Eligibility was determined via an online screening survey; the hyperlink to the screening survey was distributed during the first few weeks of the fall semester primarily by email to all first-year students enrolled at the university, and secondarily on flyers around campus and on university social media sites. Based on their responses to the screening questions, eligible participants were directed to an online consent form and asked to provide contact information. If consented, they were then emailed a link to an online baseline survey which assessed a range of demographic characteristics, alcohol use and consequences, and social and health behaviors. See Figure 1 for the CONSORT flow diagram.

Figure 1.

Consort Diagram

The 121 students who were eligible and consented to participate averaged 18 years of age (SD = 0.33), and 50% identified as female. With regard to race and ethnicity, 58% identified as White, 8% as Black, 18% as Asian, 6% as multi-racial and 9% as other; 16% identified themselves as Hispanic. The demographic characteristics did not differ by condition.

Intervention Development

Text message development.

Our pilot work (Merrill et al., 2018) indicated that students were more interested in alcohol-related information specific to students at their own university than about the general student population. Based on these local data, and literature that indicates that normative feedback is most effective when relevant to the recipient (Borsari & Carey, 2003), we used campus-specific normative data to generate the text messages. The normative feedback provided in the text messages was based on a 2017 survey of 447 students attending the university where the present study was conducted. The normative survey deliberately gathered data on a wide range of behaviors and attitudes that might serve as sources of pro-moderation feedback on campus drinking; these included student (a) drinking behaviors, (b) campus alcohol culture, (c) protective behavioral strategies, (d) behaviors that occur when intoxicated, and (e) alcohol-related consequences (sources of campus DN), as well as student levels of approval for the domains listed (for campus IN). The 2017 survey collected data on both student behavior and attitudes as well as perceptions of peer behavior and attitudes, so that items with the greatest self-other discrepancies could be identified as candidates for corrective feedback.

A student advisory group of nine students who represented the diverse student body of the university helped to create normative feedback messages from the raw survey data. Messages included in the intervention were drinking data that had high pro-moderation endorsement, self-other discrepancies, and could be formulated into text messages that the student advisors felt were (1) important for their peers to know and (2) of interest to college students. The final set of 70 text messages developed for this study is available from the authors.

Procedure

Orientation.

Upon completion of the online baseline survey, participants attended an in-person group orientation. At orientation, study procedures were reviewed, and a sample text message was sent to ensure participants’ phones were compatible with the text message software. Participants were randomly assigned to condition (experimental, n=61 versus control, n=60) stratified by gender using online randomization software.

Intervention delivery and follow-ups.

Procedures within each of the two conditions were identical except for message content, in order to control for contact and daily delivery of brief text messages containing new information. All text messages were sent daily at about 7 pm each day for ten weeks. Participants in the control condition received messages containing a unique fact about “This Day in History” (e.g., Today [10/1] in 1890 Yosemite National Park was established). Participants in the experimental condition received a unique message describing an accurate descriptive or injunctive norm (e.g., Can you have a good time at a party without getting wasted? 9 out of 10 of your peers think so.); see Table 2 for additional examples. See Supplemental Materials for the full text library. To confirm message delivery and enhance attention to messages, participants were expected to respond to each text message with an interest rating from 1 (Not at all) to 5 (Extremely interesting).

Table 2.

Mean Interest Rating for Text Messages by Domain

	Rating
	M	SD
Drinking Levels	3.37a,b,c	0.80
Sample DN text: It‘s not every weekend - half of university students drink less than once a week.
Sample IN text: Do you remember what you did last night?
Most (>80%) of university students are opposed to getting black out drunk.
Intoxicated Behaviors	3.13a,d,e	0.88
Sample DN text: Don’t feel guilty about grabbing a condom off your RPL’s door, 85% of university students have never taken the risk of unprotected sex even while drunk.
Sample IN text: Keep calm and carry on: 99% of university students think it is unacceptable to become physically aggressive when intoxicated.
Protective Behavioral Strategies	2.95b,d,f,g	0.90
Sample DN text: More than 8 out of 10 of university students eat before or while they drink at least half the time to avoid getting too intoxicated.
Sample IN text: Plan ahead: 9 out of 10 university students think it‘s a good idea to set a limit on how many drinks they have before going out.
Campus Alcohol Culture	3.33e,f,h	0.84
Sample DN text: Over 90% of university students report they can AND do have a good time with friends without consuming alcohol.
Sample IN text: Be one of the 98% of university students who respect other folks’ decision to drink or not.
Consequences	3.13c,g,h	0.92
Sample DN text: Blackouts aren’t the norm. Only 15% of university students report having blacked out in the last month.
Sample IN text: Time management is key in college. Nearly 95% of university students would not recommend spending too much time drinking.

Note. Means with the same subscript differ significantly at p <.05.

At the end of the 10-week intervention (3 months post-baseline), participants received an email containing a link to an online post-intervention survey, which included primary outcome measures plus questions about the (a) percentage of text messages received and read, (b) acceptability of text message intervention, and (c) relevance of text messages received. A link to the final online follow-up survey was sent 3 months later (6 months post-baseline). The 6-month follow-up was completed before the students left for spring break.

Participants received compensation of up to $90 for their participation: $25 for completing the baseline survey and $30 and $35 (respectively) for completing the post-intervention and 6-month follow-ups. In addition, all participants who rated at least 90% of the daily text messages were entered to win one of four $50 bonus payments that were raffled off at the end of the 10-week intervention period.

Measures

Demographic data gathered as part of the baseline survey included biological sex, gender identity, age, ethnicity, race, and Greek and athletic involvement. At each assessment, participants were provided with information that a standard drink included 12oz of beer, 5oz of 12% table wine; 12oz. of a wine cooler; or 1.25oz of 80-proof liquor. The timeframe for all drinking outcomes was the 30 days prior to each assessment.

Alcohol consumption.

Participants completed the Daily Drinking Questionnaire (Collins, Parks, & Marlatt, 1985) estimating the number of drinks they typically consumed each day of the week in the past 30 days. Number of drinks was summed across days and divided by number of drinking days to determine average drinks per drinking day. Participants reported the number of drinks they consumed on their heaviest drinking day in the past 30 days. They also reported how frequently they engaged in heavy episodic drinking (HED; 4+ drinks [females] and 5+ drinks [males] in a single drinking occasion) in the past 30 days.

Alcohol-related consequences.

Participants were asked to indicate whether they had experienced 24 alcohol-related consequences on the Brief Young Adult Alcohol Consequence Questionnaire (B-YAACQ) (Kahler, Strong, & Read, 2005) in the past 30 days. The alpha for the B-YAACQ across the three assessments in this study ranged from .66 to .82.

Descriptive and Injunctive norms (for consumption).

Participants completed an adapted version of the Daily Drinking Questionnaire (Collins et al., 1985) for both DN and IN. For DN they estimated the number of drinks a typical student of their same gender and university consumed each day of the week in the past 30 days. For IN they estimated the number of drinks a typical student of their same gender and university would consider to be acceptable to consume on each day of the week in the past 30 days. For both measures, number of drinks was summed across days and divided by number of drinking days to determine DN and IN for drinks per drinking day. In addition, participants were asked to estimate the maximum number of drinks that a typical student of their same gender and university drank on the day they drank the most in the past 30 days (DN), and the maximum number of drinks that a typical student of their same gender and university would find acceptable (IN), from 0 to 25+ drinks (peak drinks).

Descriptive and injunctive norms (for protective strategies).

Participants estimated how many of their university attending peers who drink used 8 protective behavioral strategies (PBS, e.g., alternating alcoholic and nonalcoholic beverages) before/during a drinking episode in the past 30 days (DN), and how approving a typical student of their same gender and university would be of using these 8 PBS before/during a drinking episode (IN).The specific 8 items selected to measure PBS norms were items that were included in the norms text messages in the intervention, derived from the several PBS questionnaires (DeMartini et al., 2013; Martens, 2007; Sugarman & Carey, 2007). For DN, participants responded using a 7-point scale: (0) None, (1) a few, (2) under half, (3) about half, (4) over half, (5) most, (6) all. Alphas ranged from .78-.85 across assessments. For IN, the following 6-point scale was used: (1) Strongly disapprove, (2) slightly disapprove, (3) slightly approve, (4) moderately approve, (6) strongly approve. Alphas ranged from .90-.93 across assessments.

Descriptive and injunctive norms (for consequences).

Participants estimated how many students from their university had experienced 11 consequences from the Brief Young Adult Alcohol Consequence Questionnaire (B-YAACQ) (Kahler et al., 2005) (DN) and how approving students from their university would be of experiencing these 11 consequences. The specific consequence items used to assess DN and IN represented the consequences that were included in the intervention text messages. For DN, participants responded using a 7-point scale: (0) None, (1) a few, (2) under half, (3) about half, (4) over half, (5) most, (6) all. Alphas ranged from .88-.90 across assessments. For IN, participants responded using a 6-point scale: (1) Strongly disapprove, (2) slightly disapprove, (3) slightly approve, (4) moderately approve, (6) strongly approve. Alphas ranged from .81-.90 across assessments.

Study Feedback.

At the post-intervention assessment, participants were asked about how often the text messages were interesting, relevant, and helpful measured on a 5-point scale from 1 (never) to 5 (always) and were asked about their overall rating of participating in the study. Furthermore, a question about how often participants received and read the text messages was asked to describe intervention dose.

Analysis Plan

First, for descriptive purposes, t-tests were conducted to compare baseline measures by condition, and we conducted a repeated measures ANOVA model to examine if there were differences across interest rating for the five drinking domain text messages: (1) drinking level, (2) intoxicated behaviors, (3) PBS, (4) campus alcohol culture, and (5) consequences. Following significant omnibus tests, paired t-tests were conducted to examine where differences existed.

Hierarchical linear models (HLM) were run to test our substantive hypotheses regarding primary outcomes at 3- and 6-months, using the HLM 7.02 program (Raudenbush, Bryk, & Congdon, 2013). These analyses relied on intent-to-treat principles, with N=121 participants with three potential observations each. As such, the person-period data set for HLM included 363 potential observations. Though all participants could be included in model estimation, due to a small amount of attrition at each time point (Figure 1), a total of 344 data points (95%) were available for analysis. All outcomes were normally distributed. Fully unconditional HLM models (i.e., no predictors) were run first, in order to determine intraclass correlations (ICCs) for outcomes, which provides information on the percentage of variation in each outcome at the between-person level. Next, in primary models, two time components (Time3M coded 0, 1, 0; Time6M coded 0, 0, 1) at Level 1 (within-person level) were used to represent change from baseline to 3-month and 6-month follow-ups, respectively. For each outcome, we added condition at Level 2 (between-person level) as a predictor of the intercept (i.e., effect of group on the outcome at baseline) and both time effects (i.e., effect of group on change in the outcome between baseline and 3-months and between baseline and 6-months). Additionally, we controlled for sex in all models. This effect was centered so that intercepts could be interpreted collapsing across gender. In exploratory models, we then examined potential moderating effects of sex on group differences in outcomes over time. We estimated random intercepts for observations clustered within individuals. Each random slope was tested and retained when significant, which was the case for the Time3M slope of peak drinks and consequences. In reporting effect significance, we relied on robust standard errors.

Finally, linear regression models were used to examine the potential impact of the intervention on intermediate outcomes (DN and IN for drinks per drinking day, peak drinks, PBS, and consequences) at the 3 month post-intervention assessment. Models controlled for both sex and baseline levels of the outcome.

Results

Descriptive Data

Sample description and equivalence of groups at baseline.

Participant demographics and baseline levels of outcome variables are shown in Table 1. The two conditions were equivalent on demographics and drinking frequency (mode and median = 2x/week). The groups did not differ on any of the primary outcome variables.

Table 1.

Baseline sample demographics and drinking behaviors by condition

	Whole Sample	Experimental	Control	t/chi square	P
Age	18 (0.33)	18 (0.36)	18 (0.30)	0.79	.114
Female sex/gendera	61 (50.4%)	31 (50.8%)	30 (50.0%)	0.01	.928
Hispanic ethnicity	19 (15.7%)	10 (16.0%)	9 (15.0%)	0.04	.833
Race
White	70 (58.3%)	32 (52.5%)	38 (64.4%)	3.63	.459
Black	10 (8.3%)	5 (8.2%)	5 (8.5%)
Asian	22 (18.3%)	13 (21.3%)	9 (15.3%)
Multiracial	7 (5.8%)	3 (4.9%)	4 (6.8%)
Other	11 (9.2%)	8 (13.1%)	3 (5.1%)
Greek intentions	37 (30.6%)	16 (26.2%)	21 (35.0%)	1.10	.295
Varsity athlete	18 (14.9%)	6 (9.8%)	12 (20.0%)	2.50	.116
Club athlete	36 (29.8%)	19 (31.1%)	17 (28.3%)	0.12	.735
Intramural athlete	48 (39.7%)	25 (41.0%)	23 (38.3%)	0.09	.766
Drinking frequency
2= <1x per month	1 (0.8%)	0 (0%)	1 (1.7%)	5.84	.322
3=1x per month	4 (3.3%)	3 (4.9%)	1 (1.7%)
4=2-3x per month	6 (5.0%)	3 (4.9%)	3 (5.0%)
5=1x per week	21 (17.4%)	14 (23.0%)	7 (11.7%)
6=2x per week	65 (53.7%)	32 (52.5%)	33 (55.0%)
7=3+ per week	24 (19.8%)	9 (14.8%)	15 (25.0%)
Peak drinks	8.20 (3.60)	8.67 (3,85)	7.73 (3.29)	1.46b	.148
Heavy day frequency	4.75 (2.83)	4.77 (2.81)	4.73 (2.87)	0.07b	.943
Drinks per drinking day	4.71 (2.08)	4.66 (1.76)	4.77 (2.37)	−0.30b	.768
BYAACQ	4.65 (2.78)	4.95 (2.87)	4.35 (2.67)	1.19b	.236

Note. BYAACQ = Brief Young Adult Alcohol Consequences Questionnaire.

^aBoth biological sex and gender identity were assessed and aligned for all participants in this study.

^bequal variances not assumed

Exposure to text messages.

Ninety-five percent of the daily ratings of the text messages were completed. At the individual level, 99% completed more than half of all daily ratings, and 88% completed nearly all of the daily ratings (65 or more texts out of 70). On the 3-month survey, 89% of participants recalled receiving text messages daily. Most (96%) reported reading all text messages they received.

Feedback on the intervention experience.

Overall, participants rated their experience in the study as 3.6 (SD=0.9) on a 5-point scale. Interest ratings averaged 3.14 (SD=1.37) for the norms text messages, and 2.46 (SD=1.43) for the control messages. We examined the interest rating across the normative domains (drinking level, PBS, intoxicated behaviors, campus culture, and consequences), and found a significant between-domain effect (F[4]=11.02, p=.000). Pairwise comparisons were used to determine differences among the domains (see Table 2); interest levels were highest for DN and IN related to peer drinking levels and campus culture, followed by peer DN and IN about intoxicated behaviors and consequences, and lowest for DN and IN protective strategies. Participants in the intervention group rated the personal relevance of alcohol norms texts as 2.2 (SD=0.8) on a 5-point scale.

Substantive HLM Models Predicting Primary Outcomes

The ICC for both drinks per drinking day and peak drinks was .51, suggesting that 51% of the variance in each of these outcomes was between- and 49% was within-person. The ICC was .42 for binge frequency, and .30 for consequences. Results of HLM models predicting each primary outcome are shown in Table 3 and Figure 2, and briefly described below. In exploratory models, sex did not moderate intervention effects at either the 3- or 6- month assessment. As such, for parsimony, models without the inclusion of sex as a moderator are reported.

Table 3.

Hierarchical linear models predicting change between baseline and 3M (post-intervention) and 6M (follow-up) outcomes by condition

	Drinks per Drinking Day				Peak Drinks				Binge Frequency				Consequences
	B	SE	t	P	B	SE	t	P	B	SE	t	P	B	SE	t	P
BL Level (Control)	4.77	0.31	15.61	<0.001	7.73	0.40	19.11	<0.001	4.73	0.37	12.81	<0.001	4.35	0.34	12.72	<0.001
Effect of Intervention	−0.09	0.38	−0.24	0.808	0.87	0.58	1.49	0.140	0.04	0.52	0.08	0.937	0.60	0.50	1.20	0.233
Sex (centered)	1.08	0.31	3.53	<0.001	2.47	0.50	4.97	<0.001	0.46	0.41	1.13	0.261	−0.05	0.43	−0.12	0.903
Change BL to 3M (Control)	0.02	0.27	0.06	0.952	0.80	0.43	1.88	0.062	−0.51	0.43	−1.19	0.237	1.04	0.56	1.84	0.068
Effect of Intervention	−0.62	0.41	−1.52	0.129	−2.13	0.65	−3.25	0.002	−0.35	0.58	−0.60	0.552	−1.60	0.72	−2.22	0.029
Change BL to 6M (Control)	−0.74	0.28	−2.63	0.009	−0.91	0.39	−2.33	0.022	−1.52	0.40	−3.82	<0.001	−0.69	0.50	−1.39	0.169
Effect of Intervention	0.28	0.38	0.75	0.452	−0.03	0.61	−0.05	0.960	0.69	0.54	1.28	0.201	0.04	0.65	0.07	0.948

Figure 2.

HLM Model Implied Values of Primary Outcomes at Baseline, 3M, and 6M by Condition

3-month effects.

Between baseline and 3 months, there were no effects of time or intervention condition on drinks per drinking day or binge frequency. However, there were significant intervention effects on peak drinks and consequences, with both favoring the intervention over the control group (Cohen’s d were −.59 and −.43, respectively).

6-month effects.

Between baseline and 6 months, there was a significant decline in drinks per drinking day, peak drinks, and binge frequency; however, there were no effects of condition, suggesting that these outcomes changed equally across the two groups. There were no effects of time or intervention condition on consequences between baseline and 6 months.

Regression models predicting 3-month intermediate outcomes

Descriptive norms.

Findings of regression models predicting descriptive norm variables at 3 months are shown in Table 4. Controlling for baseline levels, there were no effects of condition at 3 months on descriptive norms for drinks per drinking day, PBS or consequences. However, there was a significant effect on descriptive norms for peak drinks, such that the intervention group had lower values at 3 months than control.

Table 4.

Regression models predicting 3M post-intervention outcomes: Descriptive norms

	Drinks per drinking day DN		Peak drinks DN		PBS DN		Consequences DN
	β	p	β	p	β	p	β	p
Intervention	.11	.209	−.16	.018	.12	.124	−.09	.307
Sex	.12	.218	.20	.006	−.06	.402	−.08	.361
BL level of outcome	.34	.001	.61	<.001	.60	.124	.37	<.001
Overall model	F= 7.316 Adj R2 = .145	<.001	F=35.947 Adj R2=.481	<.001	F=22.085 Adj R2 =.359	<.001	F=6.772 Adj R2 =.133	<.001

Note: BL=baseline, DN=descriptive norms, PBS=protective behavioral strategies

Injunctive norms.

Findings of regression models predicting injunctive norm variables at 3 months are shown in Table 5. There were no effects of condition at 3 months on injunctive norms for drinks per drinking day or peak drinks. However, there was a significant effect of condition on injunctive norms for PBS, such that the intervention condition reported higher perceived approval of PBS at 3 months than control. Additionally, there was a significant effect of condition on injunctive norms for consequences, such that the intervention condition reported lower perceived approval of consequences at 3 months than control.

Table 5.

Regression models predicting 3M post-intervention outcomes: Injunctive norms

	Drinks per drinking day IN		Peak drinks IN		PBS IN		Consequences IN
	β	P	β	P	β	P	β	P
Intervention	−.04	.624	.02	.767	.18	.039	−.18	.041
Sex	.17	.051	.31	.001	−.08	.360	.17	.052
BL level of outcome	.38	<.001	.31	.001	.44	<.001	.28	<.001
Overall model	F= 8.228 Adj R2 = .161	<.001	F =12.972 Adj R2 =.241	<.001	F=12.161 Adj R2 =.229	<.001	F=6.487 Adj R2=.127

Note: BL=baseline, DN=descriptive norms, PBS=protective behavioral strategies

Discussion

We conducted a randomized controlled trial to examine delivery of normative feedback to heavy drinking college students via text message. The text messages were campus specific, developed using student voices, and designed to provide accurate information on both DN and IN for a range of dimensions relevant to drinking, including drinking practices, consequences experienced, and use of protective behavioral strategies. This study confirmed the feasibility and acceptability of text messages as a strategy for delivering accurate campus drinking norms. Immediately after receiving 10 weeks of daily text messages, intervention participants reported reductions in peak consumption and alcohol-related consequences, relative to the attention control participants. Although we observed medium between-groups effect sizes on these outcomes at the 3-month post-intervention assessment, none of the outcome measures differed by condition at the 6-month follow-up assessment. Thus, the intervention had short-term effects on self-reported drinking behavior as well as on perceptions of campus drinking norms. However, the behavioral changes were not maintained when participants were assessed in the spring semester after the daily text messages had stopped.

The method of delivering corrective norms information used in this study differed from prior tests of norms-based interventions that rely on a single face-to-face or computer-facilitated delivery of personalized normative feedback, (e.g., Dotson, Dunn, & Bowers, 2015) or social norms marketing approaches that utilize mass media (e.g., DeJong, 2010). Text messaging capitalizes on the reliable delivery of information to targeted individuals and the high likelihood that young adults will view and read their text messages. Consistent with our earlier pilot study (Merrill et al., 2018) with second-year college students, students in their first semester in college reliably viewed the norms messages and reported acceptable levels of interest and relevance. This test of feasibility is important, given that the first year in college is high-risk for heavy drinking and consequences (Borsari et al., 2007), and that drinking norms formed pre-college are predictive of heavy drinking in the transition to college (Sher & Rutledge, 2007).

We observed intervention effects at 3 months only on peak drinks and consequences. This is notable given the relatively low internal consistency achieved on the B-YAACQ at the baseline assessment. On these two outcomes, the control group tended to increase from the beginning to the end of the first semester, whereas the intervention group did not. This suggests that first-year students who are already heavy drinkers when they arrive on campus in September are likely to end the semester drinking more heavily with associated increases in consequences, but that the exposure to moderate alcohol use norms minimizes this temporary elevated risk.

Following the participants into the second semester, we found that the differences between the intervention and control groups disappeared. Thus, the text-delivered intervention was effective at suppressing risky drinking and consequences only during the semester of active exposure. Notably, all participants reduced all three of the consumption variables from baseline to 6-months, so intervention effects may have been obscured by an overall effect of history or maturation on consumption or even assessment reactivity ((Walters, Vader, Harris, & Jouriles, 2009). Despite an initial apparent increase in alcohol involvement, the control group experienced a downward correction over the course of the academic year that eventually brought them closer to the intervention group. Although not a threat to internal validity because of the use of random assignment in this study (c.f. Shadish, Cook, & Campbell, 2002), these findings highlight that variations in drinking quantities and frequency are observed across the academic year (Del Boca, Darkes, Greenbaum, & Goldman, 2004; Hoeppner et al., 2012; Tremblay et al., 2010). It appears that the text messages communicating the acceptability and practice of moderate drinking by student peers prevented intervention participants from participating in extreme drinking during their first semester in college; however, spending more time in college has a similar (if delayed) effect.

We found partial support for putative mediators – changes in DN and IN – as we observed selected intervention effects in the predicted directions at three months. Specifically, the students who received the daily texts about campus alcohol norms reported lower perceptions of DN for peak drinks at the end of the intervention period. Similarly, students in the intervention group reported increased perceptions of peer approval of using protective behavioral strategies, and lower perceptions of peer approval for alcohol-related consequences. Consistent with the literature for other normative correction interventions focusing on DN (e.g., Lewis & Neighbors, 2006; Scott-Sheldon, Demartini, Carey, & Carey, 2009) or IN (Prince et al., 2015) the text-message delivery was successful in moving normative perceptions in the direction of moderation. The changes in DN for peak drinks and IN for consequences align with the two behavioral outcome measures on which intervention effects were seen. However, in this study several types of DN and IN norms were targeted, and not all changed as a function of exposure. This may be in part due to imprecise measurement of normative constructs that were developed for this study. Alternatively, it is possible that the text message content was more memorable or salient for certain types of norms and less so for others. However, the fact that the significant intervention effects were all in the predicted directions (reduction for some, increase for another) is promising support for the ability of the text message delivery format to change perceived norms, a reliable mechanism of behavior change for brief alcohol interventions in college contexts (Reid & Carey, 2015).

It is worth noting interventions delivered by text have achieved significant drinking reductions despite varying content and theoretical underpinnings. As one example, Haug and colleagues (2017; Haug et al., 2013) developed a 12-week program of individually tailored text messages that provided feedback on drinking patterns with normative comparisons, with different messages delivered according to tailoring variables such as sex, age and motivation to reduce drinking. Another example is the work of Suffoletto and colleagues (2014) who used interactive text messages to reduce weekend binge drinking; participants were prompted to set drinking goals and received responses tailored to goal attainment. Our campus norms-based text intervention offers promising results without personalized messages or tailored content. Thus, text messaging has proven to be a versatile delivery format for alcohol risk reduction interventions for young adults.

Limitations

The findings should be interpreted in light of study limitations. First, our sample was small, and not powered to detect small effects or to test mediation. However, we were able to demonstrate the feasibility of the text-based norms intervention, and promising short-term effects on peak drinks and consequences. These outcomes were supported by predicted changes in perceived norms (reductions in DN for peak drinks and IN for consequences, increases in IN for PBS), suggesting that a larger sample may provide support for the proposed mediational pathway. Second, we measured perceived norms at 3- and 6-month assessments using norms assessments that were developed for this study. We created item sets aligned with the normative messages delivered in the intervention; although these item sets were internally consistent, they were not psychometrically established measures and may have lacked sensitivity. Third, natural variations in drinking occur throughout the academic year (e.g., Tremblay et al., 2010), which may have affected estimates of change from when baseline assessments take place during the beginning of the semester, when consumption levels are highest (Hoeppner et al., 2012). Even though reductions would be expected from September to December even in the attention control condition, the observed intervention effects suggest that the intervention had an impact above and beyond the well-documented reductions associated with the end of the semester. We acknowledge that more frequent assessments would provide more sensitive estimates of follow-up drinking. Finally, participants evaluated the text messages as only modestly interesting and personally relevant. It is unclear whether these dimensions can be improved and whether the intervention would be more impactful with more personalized or engaging messages.

Future Directions

This preliminary study lays the groundwork for a larger RCT with more participants (and power) and longer follow-ups to demonstrate the durability of effects. Replication is needed in a larger, more diverse sample. The current study found promising evidence of intervention-related change on hypothesized mediators, including reducing descriptive norms for excessive drinking and perceived approval of alcohol consequences, and increasing perceived peer approval for use of protective behavioral strategies. Future research should be designed to test the full mediational pathways in order to evaluate the specific mechanisms by which normative feedback can influence drinking behavior. In addition, future research on this text-delivery framework for normative correction could evaluate the unique effects of DN vs IN feedback or different categories of feedback content (e.g., norms about consequences vs norms about protective strategies). Once a text-delivered intervention is successful in changing perceived norms and drinking behavior within efficacy studies, further implementation research would be needed to establish generalizable incentives for students to participate before it could be taken to scale on other campuses. In summary, this study provides promising evidence to support the continued study of daily text messages as a vehicle for delivering campus-based norms to first-semester college students.