An Expanded Daily Alcohol Expectancies Measure: Results on the Multilevel Factor Structure and Psychometric Properties

Abstract

Objective:

Alcohol expectancies are beliefs people have about the likelihood of experiencing various positive or negative consequences related to alcohol use. Expectancies have most commonly been treated as traitlike characteristics of individuals, but some researchers have assessed expectancies as state-level characteristics that vary within persons across days. Previous work developed a 13-item daily alcohol expectancies measure. This study evaluated an expanded version of that measure that includes 10 additional expectancy items.

Method:

Participants were 2- and 4-year college students (N = 201; 63.7% female; 55.2% White non-Hispanic; 75.1% 4-year students) randomized to the control group of a longitudinal study designed to test the efficacy of a just-in-time adaptive intervention delivered via mobile app to reduce high-risk alcohol use. Multilevel exploratory factor analysis was used to determine the factor structure at the daily and person levels. Multilevel models were used to evaluate the convergent validity of the resulting subscales.

Results:

Two factors, broadly representing positive and negative alcohol expectancies, were retained at the daily and person levels. Composite reliability (omega) estimates ranged from .85 to .96 and suggested that the reliability of the resulting subscales was good to strong. Associations between the daily expectancy subscales and baseline scores on an established expectancies measure provided preliminary evidence of convergent validity.

Conclusions:

Findings indicate that this expanded 23-item daily alcohol expectancies measure is psychometrically sound. This measure is appropriate for use in daily or just-in-time expectancy challenge interventions. It is suitable for use among 2- and 4-year college students who drink alcohol regularly and occasionally in heavy quantities and who experience alcohol-related negative consequences.

Alcohol expectancies are beliefs a person holds about how likely they are to experience certain alcohol-related consequences, and expectancies are typically conceptualized as either positive (e.g., will feel less stress) or negative (e.g., will experience a hangover) (Fromme et al., 1993). A large body of research has shown that alcohol expectancies, particularly positive expectancies, are associated with increased alcohol use and consequences (Lee et al., 2020; Patrick et al., 2010; Smit et al., 2018). Alcohol expectancies have most commonly been assessed as relatively stable, traitlike characteristics with people differing in the extent to which they endorse each expectancy belief (e.g., Leigh & Stacy, 2004; Patrick et al., 2010; Waddell et al., 2022). However, research using daily designs has shown that expectancies also fluctuate within people across days (e.g., Butler et al., 2010; Lee et al., 2020; Rhew et al., 2021), which suggests that expectancies also have statelike properties that may be influenced by temporal and/or contextual factors. As researchers continue using daily designs to examine both within- and between-person associations between expectancies and alcohol use and consequences, it is important to have reliable and valid measures of expectancies that can capture both within-person fluctuations in expectancies across days and between-person differences in average levels of expectancies.

We previously developed a psychometrically sound, brief 13-item measure for assessing daily alcohol expectancies among college students (Lee et al., 2015). We found that the majority of variance in expectancies was attributable to within-person, day-to-day fluctuations relative to between-person differences in average responses (i.e., intraclass correlation coefficients [ICCs]: .30–.42). Results of multilevel exploratory factor analysis (EFA) indicated that the expectancy items loaded on two factors, broadly representing positive and negative expectancies, at both the daily and person levels. The subscales resulting from the four factors (positive and negative expectancies at each level) had good to strong reliability and demonstrated convergent and discriminant validity with respect to baseline scores on an established measure of alcohol expectancies for college students.

However, quantitative and qualitative (e.g., focus groups) research since that measure was published suggests that there are expectancies that are meaningful to college students and associated with alcohol use outcomes that are not captured by the measure. For instance, intentions for blacking out, which are influenced by one's expectancies (among other factors), are associated with drinking more frequently and in greater quantities, experiencing more negative consequences, and exhibiting more depressive symptoms (Miller et al., 2020). Expectancies about blacking out are associated with how positively or negatively young adults rate the experience of blacking out (Merrill et al., 2019). Texting (including sexting, or sending sexually explicit messages/images) and posting to social media while intoxicated are phenomena facilitated by smartphones that often result in regrettable social consequences (Dunne & Katz, 2015). Such behaviors appear to be somewhat common in community samples of college students (Dunne & Katz, 2015; Florimbio et al., 2018; Perez et al., 2021), and sex-related alcohol expectancies (e.g., feeling sexually excited) appear to mediate the association between alcohol use and sexting (Florimbio et al., 2018). Therefore, our prior daily expectancies measure is not fully up-to-date or comprehensive because it does not contain expectancies related to severe physical harm (e.g., blacking out), sexuality (e.g., having a greater desire for sex), or using technology while intoxicated (e.g., sexting).

The limited scope of the original measure may be of less concern for researchers whose primary interest is not on expectancies because they may favor the reduced participant burden of a brief measure. However, for researchers whose primary interest is on alcohol expectancies, such as those testing expectancy-related interventions using daily designs, the absence of items for certain domains may be more concerning. In addition to a more comprehensive measurement of latent expectancy variables (e.g., positive and negative subscales), having a more comprehensive set of expectancy items would broaden the range of potential item-level analyses. For instance, some researchers may be interested in examining associations between specific expectancy items (e.g., alcohol will make me feel sexier) and specific alcohol-related outcomes (e.g., engaging in risky sexual behaviors under the influence of alcohol).

The present study examined the psychometric properties of an expanded version of the daily alcohol expectancies measure previously developed and validated by Lee et al. (2015). The expanded measure included 14 new expectancy items (27 total, including the original 13) and items related to serious physical harm, sexuality, and the use of smartphones and social media. This expanded version of the daily alcohol expectancies measure was designed to be used in a daily just-in-time adaptive intervention (i.e., an intervention tailoring the type, timing, and/or amount of support to the moment and context in which it is most needed and when the participant is most likely to be receptive) focused on alcohol expectancies. The present study had three aims. In Aim 1, we examined the multilevel factor structure of the expanded daily expectancies measure. We hypothesized that the expanded measure would have a two-factor structure, broadly corresponding to positive and negative expectancies, at both the daily and person levels. In Aim 2, we estimated the multilevel reliability of the resulting subscales. We hypothesized that the reliability of the subscales would be good, as indicated by composite reliability estimates greater than .70 (Taber, 2018). In Aim 3, we conducted initial tests of convergent validity with an established alcohol expectancies measure assessed at baseline. We hypothesized that the expanded daily expectancy subscales would demonstrate convergent validity as evidenced by positive associations between similarly valenced (i.e., positive or negative) subscales of the established expectancies measure and the expanded daily expectancies measure.

Method

Participants and procedures

Participants were students from 2- and 4-year colleges in the Greater Seattle, WA, metropolitan area who were enrolled in a longitudinal study designed to test the efficacy of a daily ecological momentary intervention delivered via mobile app to reduce high-risk drinking. Eligible participants were 18–25 years old, enrolled at a 2- or 4-year college, reported drinking 2 or more days per week, reported at least one heavy episodic drinking occasion (4+/5+ drinks for females/males) in the last 2 weeks, and reported 4 or more negative alcohol-related consequences in the past month. The University of Washington Institutional Review Board approved this study.

Students from two 2-year colleges and one 4-year university were randomly selected from registrar's lists to be invited to participate in the larger study. Selected students were sent emails inviting them to participate in an online screening survey. Advertisements promoting the study were also placed on social media, primarily Instagram and Facebook, to recruit students in the local area. All participants who accessed the screening survey were presented with an information statement describing the screening process and asked to complete a 20-minute survey to determine eligibility. Screening survey completion occurred between January 2020 and February 2021. Of the 5,152 individuals who completed the screening survey, 581 (11.3%) met the eligibility criteria and were invited to complete the baseline survey.

The 30-to 45-minute baseline survey assessed alcohol expectancies and other psychosocial measures related to alcohol use. Of those invited, 500 (86.1%) completed the baseline survey and were invited to schedule a training session (in-person or via Zoom during the COVID-19 pandemic) with research staff to install the app and review study procedures. Before the training session, participants were randomized to one of two conditions: intervention or assessment-only control. A total of 408 students enrolled in the study: 205 in the intervention condition and 203 in the control condition.

After the training session, participants used the mobile app on their own smartphones (Android or iOS) to complete twice-daily assessments for the subsequent 21 days, once in the morning (from a self-selected 3-hour period between 8:00 A.M. and 1:00 P.M.) and once in the afternoon (from a self-selected 3-hour period between 3:00 P.M. and 7:00 P.M.). Alcohol expectancies were assessed in the afternoon surveys; therefore, only those surveys were used in these analyses. Participants completed three additional 14-day bursts of daily surveys 1, 6, and 12 months after the initial 21-day burst. Participants were compensated $10, $15, or $20 for completing the screening survey (the incentive was increased twice to assist with recruitment), $25 for the baseline survey, $2 per twice-daily survey, and a $7 weekly bonus if 12 or more of the 14 twice-daily surveys were completed each week.

Current analyses were limited to 201 participants randomized to the control condition and to the 9,572 afternoon surveys provided by those individuals. Two participants in the control condition did not participate in any daily surveys and are not included in the analyses here. In the analytic sample, the mean age of participants at screening was 20.79 years (SD = 1.60), 63.7% reported their biological sex as female, and 75.1% attended a 4-year university. Regarding race, 63.1% of the sample identified as White/Caucasian, 18.2% Asian/Asian American, 8.1% multiracial, 4.5% Black/African American, 4.5% other, 1.0% American Indian/Alaskan Native, and 0.5% Native Hawaiian/Pacific Islander. Regarding ethnicity, 16.4% of the sample identified as Hispanic/Latino(a). Regarding gender identity, 61.7% of participants identified as female, 36.8% as male, and 1.5% as other. Regarding sexual orientation, 73.6% of participants identified as heterosexual, 12.9% as bisexual, 5.5% as questioning, 3.0% as gay, 3.0% as queer, 1.0% as lesbian, and 1.0% as other. Sample participants completed 74.0% of afternoon surveys, and 77.3% responded to at least one afternoon survey in each of the four bursts.

Measures

Daily alcohol expectancies. Each afternoon, participants were asked, “If you were to drink tonight, how likely would you be to feel or do the following things as a result of your drinking?” Participants were presented with a list of 27 expectancy items (Table 1). Response options were (1) very unlikely, (2) unlikely, (3) somewhat unlikely, (4) somewhat likely, (5) likely, and (6) very likely. There were 13 expectancy items (7 positive and 6 negative) in the measure developed by Lee et al. (2015). Here, we added 14 expectancy items (e.g., have a blackout, drunk text/dial, feel sexier than usual, spend too much money). New items were selected based on focus groups and interviews with college students as well as a review of the literature for effects or consequences of alcohol use likely experienced by college students.

Table 1.

Descriptive statistics

Expectancy item	n	M	SD	Mdn	ICC
Items in daily alcohol expectancies measure developed in Lee et al. (2015)
Feel more relaxed	9,207	3.73	1.48	4	.43
Be in a better mood	9,172	3.47	1.48	4	.42
Feel a buzz	9,170	3.46	1.64	4	.39
Be more sociable	9,181	3.39	1.57	4	.43
Feel more energetic	9,138	2.92	1.50	3	.48
Express your feelings more easily	9,135	2.89	1.49	3	.49
Have a hangover	9,178	2.04	1.33	2	.48
Do something that embarrassed you	9,133	1.94	1.17	2	.52
Feel nauseated or vomit	9,155	1.78	1.12	1	.49
Be rude or obnoxious	9,134	1.72	0.99	1	.56
Not remembering what you did while drinking	9,157	1.62	0.99	1	.55
Become aggressive	9,170	1.54	0.90	1	.62
Hurt or injure yourself by accident	9,162	1.48	0.83	1	.57
Items added in the expanded daily alcohol expectancies measure
Be unable to study	9,145	2.98	1.73	3	.45
Forget about your worries	9,121	2.58	1.48	2	.50
Have more desire for sex	9,160	2.54	1.47	2	.48
Feel more attractive	9,131	2.54	1.42	2	.54
Not get up and do what you were supposed to do	9,135	2.46	1.56	2	.58
Feel sexier than usual	9,129	2.39	1.38	2	.53
Be overly emotional	9,131	2.35	1.35	2	.53
Spend too much money	9,119	1.96	1.21	1	.55
Drunk text/dial	9,139	1.71	1.08	1	.58
Post something on social media that you wouldn't normally post (e.g., Snapchat)	9,132	1.59	0.97	1	.65
Have a blackout	9,132	1.42	0.82	1	.58
Get into a serious fight	9,136	1.40	0.75	1	.61
Pass out or faint suddenly	9,132	1.37	0.75	1	.62
Damage property on purpose	9,127	1.37	0.75	1	.65

Notes: N = 201 participants. ICC = intraclass correlation coefficient. Response options for expectancy items: 1 = very unlikely; 2 = unlikely; 3 = somewhat unlikely; 4 = somewhat likely; 5 = likely; 6 = very likely.

Baseline alcohol expectancies. General alcohol expectancies were assessed at baseline using the Brief Comprehensive Effects of Alcohol questionnaire (B-CEOA; Ham et al., 2013). Participants were given the prompt, “If I were under the influence of alcohol” followed by 15 alcohol expectancy items (e.g., it would be easier to talk to people, I would feel dizzy, I would take risks). Response options were (1) disagree, (2) slightly disagree, (3) slightly agree, and (4) agree. The B-CEOA has been used with college students and adolescents and has shown good internal consistency (Ham et al., 2005, 2013, 2019). Positive (seven items; α = .78) and negative (six items; α = .78) expectancy subscales were created based on the two-factor structure reported by Ham and colleagues (2013).

Data analyses

In Aim 1, multilevel EFA was used to determine the daily- and person-level factor structures of the expanded daily expectancies measure. Multilevel EFA models were fitted in Mplus Version 8.7 (Muthén & Muthén, 1998–2017). Given the ordered categorical nature of the response options, the expectancy items were treated as ordinal. A robust weighted least squares estimator using a diagonal weight matrix was used, and oblique rotation (GEOMIN) was used to obtain factor loadings and factor correlations. The number of factors extracted at each level ranged from 1 to 10 across models. Deciding how many factors to retain was guided by consideration of (a) results of parallel analysis, which statistically determines the break in scree plots (Horn, 1965; Longman et al., 1989); (b) simple structure (Thurstone, 1947); (c) factor interpretability; and (d) theory. Parallel analyses were performed separately at each level using the psych (Revelle, 2024) package in R 4.3.2 (R Core Team, 2023). Items were considered to load on a factor if factor loadings were greater than or equal to .40 (Yong & Pearce, 2013).

In Aim 2, multilevel composite reliability estimates were calculated using the multilevelTools (Wiley, 2020) R package. Composite reliability estimates, ω (omega), are conceptually similar to Cronbach's α as they represent the proportion of a measure's estimated true score variance to its estimated total variance (Geldhof et al., 2014; McNeish, 2018). Composite reliability estimates are calculated based on factor loadings from confirmatory factor analysis (CFA) models and provide more precise reliability estimates than those provided by α (Geldhof et al., 2014; McNeish, 2018). When extended to the multilevel context, multilevel CFA allows for measurement model parameters to be estimated at each level (i.e., the daily and person levels) and, in turn, for reliability to be estimated at each level (Geldhof et al., 2014).

In Aim 3, linear multilevel models fit with the lme4 (Bates et al., 2015) R package were used to test associations between person-level scores on an established expectancy measure completed at baseline (i.e., B-CEOA) and daily-level scores on the expanded daily expectancies measure. Convergent validity was assessed by determining whether the positive B-CEOA subscale was positively associated with the positive daily expectancy subscale and whether the negative B-CEOA subscale was positively associated with the negative daily expectancy subscale. Models also included cross-valence associations (e.g., positive B-CEOA subscale predicting negative daily expectancies), as there were positive correlations between the subscales of each measure. All subscale scores were standardized in these models.

For all three aims, two sets of sensitivity analyses were conducted. First, the analytic sample was stratified by sex, and separate analyses were conducted for the female (n = 6,138 days, 128 females) and male (n = 3,434 days, 73 males) subsamples. Second, analyses were conducted when limiting the analytic sample to days without prior alcohol consumption (based on responses to the item, “Have you consumed any alcohol today?”; n = 8,561 days, 201 participants).

Results

Descriptive statistics

Table 1 presents descriptive statistics for all 27 alcohol expectancy items. Generally, items representing positive alcohol outcomes (e.g., feel more relaxed) had higher average scores than items representing negative outcomes (e.g., pass out or faint suddenly), meaning that participants tended to rate positive outcomes as being more likely to occur. The ICCs of the expectancy items ranged between .39 and .65, meaning that 39%–65% of the total variability in alcohol expectancies was between persons and the remaining variability was within persons across days.

Aim 1: Multilevel factor structure

Multilevel EFA was performed on all 27 expectancy items. Considerations of parallel analysis results, simple structure, factor interpretability, and theory resulted in two factors being retained at each level, broadly representing positive and negative expectancies. Factor loadings for this two-factor multilevel EFA are shown in Table 2. Most items loaded cleanly (factor loadings ≥ .40 on a single factor) on either the positive or negative factor, with the same items loading on the positive and negative factors at each level. Thirteen items loaded on the negative factor and not on the positive factor. All seven negative items in the previous measure also loaded on the negative factor here. Six new items loaded on the negative factor: get into a serious fight, pass out or faint suddenly, have a blackout, drunk text/dial, post something on social media that you would not normally post, and damage property on purpose. Ten items loaded on the positive factor and not on the negative factor. All six positive items in the previous measure also loaded on the positive factor here. Four new items loaded on the positive factor: feel sexier than usual, feel more attractive, forget about your worries, and have more desire for sex. Four items (all newly added) either did not cleanly load on either factor or loaded on both factors: be unable to study, be overly emotional, not get up and do what you were supposed to do, and spend too much money. These four items were removed from the final EFA and subsequent analyses. The positive and negative factors were positively correlated at both the daily (r = .49, p < .05) and person (r = .47, p < .05) levels.

Table 2.

Factor loadings from the final two-factor multilevel exploratory factor analysis of daily alcohol expectancy items

Expectancy item	Within person		Between person
	Positive	Negative	Positive	Negative
Feel more relaxed	.78	-.25	.89	-.36
Be more sociable	.83	-.08	.88	.00
Be in a better mood	.85	-.20	1.00	-.21
Feel a buzz	.70	.07	.76	-.01
Feel more energetic	.70	.03	.81	.16
Express your feelings more easily	.65	.13	.83	.18
Feel sexier than usuala	.61	.18	.68	.28
Feel more attractivea	.65	.13	.72	.22
Forget about your worriesa	.58	.12	.71	.20
Have more desire for sexa	.60	.15	.60	.33
Have a hangover	.25	.50	.21	.68
Become aggressive	-.06	.70	.02	.88
Feel nauseated or vomit	.02	.69	.09	.76
Hurt or injure yourself by accident	.01	.71	-.02	.94
Not remembering what you did while drinking	.16	.67	.15	.86
Be rude or obnoxious	.10	.66	.08	.79
Do something that embarrassed you	.28	.55	.20	.81
Get into a serious fighta	-.20	.82	-.09	.89
Pass out or faint suddenlya	-.18	.82	-.09	.94
Have a blackouta	.02	.81	-.01	.92
Drunk text/diala	.23	.54	.22	.74
Post something on social media that you wouldn't normally post (e.g., Snapchat)a	.21	.58	.12	.79
Damage property on purposea	-.12	.74	-.06	.90

Notes: N = 9,233 days nested within 201 individuals. All residual variance estimates are positive. Four items did not load on either factor singly and were omitted from the final model: be unable to study, be overly emotional, not get up and do what you were supposed to do, and spend too much money. Correlation between within-person factors: .49, p < .05. Correlation between between-person factors: .47, p < .05.

^a Newly added expectancy item.

Sensitivity analyses in which multilevel EFA was performed on subsamples (a) stratified by sex (Supplemental Table A) and (b) limited to days without prior alcohol consumption (Supplemental Table B) produced findings that were very similar to those of the main analyses using the full sample. (Supplemental material appears as an online-only addendum to this article on the journal's website.) The primary difference was that four of the newly added items (feel sexier than usual, feel more attractive, forget about your worries, and have more desire for sex) loaded on both the positive and negative factor (rather than just the positive factor) at the person level in the male subsample. The within-person factor structure and loadings in the male subsample and the factor structure and loadings at both levels in the subsamples limited to females and days without prior alcohol consumption produced results that were very similar to the main findings.

Aim 2: Reliability of the resulting subscales

Overall, the four expectancy subscales had good to excellent reliability. The person-level reliability estimates for the positive and negative expectancy subscales were ω = .958 (95% CI [.949, .967]) and ω = .964 (95% CI [.956, .972]), respectively, indicating strong person-level reliability according to Taber (2018). The daily-level reliability estimates for the positive and negative expectancy subscales were ω = .879 (95% CI [.876, .883]) and ω = .845 (95% CI [.841, .850]), respectively, indicating good daily-level reliability. Sensitivity analyses indicated that reliability estimates in subsamples limited to females and males separately (Supplemental Table C) and days without prior alcohol consumption (Supplemental Table D) were very similar to reliability estimates in the full sample.

Aim 3: Initial test of convergent validity

Associations between baseline measurements of the positive and negative B-CEOA subscales and the expanded daily alcohol expectancy subscales in linear multilevel models provided initial evidence of convergent validity. The positive B-CEOA expectancy subscale was positively associated with the positive daily expectancy subscale, and the negative B-CEOA expectancy subscale was positively associated with the negative daily expectancy subscale (Table 3). More specifically, a 1-SD increase on the positive B-CEOA subscale was associated with a 0.20-SD increase on participants' mean positive subscale score on our expanded daily expectancies measure. Similarly, a 1-SD increase on the negative B-CEOA subscale was associated with a 0.18-SD increase on participants' mean negative subscale score on our expanded daily expectancies measure. These associations were independent of cross-valence associations (e.g., negative baseline expectancies predicting positive daily expectancies). Regarding the cross-valence associations, the positive B-CEOA expectancy subscale was positively associated with the expanded daily negative expectancy subscale, and the negative BCEOA expectancy subscale was not associated with the expanded daily positive expectancies subscale.

Table 3.

Multilevel models testing convergent validity of expectancy subscales

	Outcome: Daily positive expectancies NDays = 9,230 NPersons = 201			Outcome: Daily negative expectancies NDays = 9,218 NPersons = 201
Predictor	β	SE	p	β	SE	p
Intercept	.03	.05	.48	.06	.05	.29
B-CEOA positive subscale	.20	.05	<.01	.17	.06	<.01
B-CEOA negative subscale	.06	.05	.22	.18	.06	<.01

Notes: All subscales were standardized. B-CEOA = Brief Comprehensive Effects of Alcohol questionnaire (Ham et al., 2013) measured at baseline.

Sensitivity analyses indicated that the results of this initial test of convergent validity among subsamples limited to females (Supplemental Table E) and days without prior alcohol consumption (Supplemental Table F) were similar to findings among the full sample. In the subsample limited to males (Supplemental Table E), neither the positive nor negative B-CEOA subscale was significantly associated with mean daily negative expectancy subscale scores. As in the main findings, the positive (but not negative) B-CEOA subscale was positively associated with mean daily positive expectancy subscale scores.

Discussion

The current study evaluated the psychometric properties of an expanded daily alcohol expectancies measure using daily data collected from a sample of college students. This measure is intended to be used in daily process studies of alcohol use, expectancies, and consequences and in interventions aimed at curbing alcohol use and its resulting consequences by altering expectancies. This measure is suitable for use among 2- and 4-year college students who drink alcohol regularly and occasionally in heavy quantities (e.g., 4+/5+ drinks on an occasion for females/males) and who experience alcohol-related negative consequences. Findings showed that the final 23-item measure had reasonably sound psychometric properties. As hypothesized, multilevel EFA findings suggested that expectancy items loaded on two factors (positive and negative) at both the daily and person levels. The resulting four subscales all had good to strong composite reliability. Associations of these subscales with an established, traitlike expectancy measure provided initial evidence of convergent validity.

Within-person variability in expectancies

As with our previous work (Lee et al., 2015), an important implication of the current findings is that there was considerable within-person variability (35%–61%) across days in all expectancy items. Given that most work on alcohol expectancies has treated expectancies as stable, traitlike constructs that do not vary within individuals (e.g., Leigh & Stacy, 2004; Patrick et al., 2010; Waddell et al., 2022), our work continues to show that expectancies vary both within and between individuals and highlights the need for psychometrically sound measures of daily expectancies.

Extensions of the original measure

The expectancies measure evaluated here expands the original measure by adding 10 items. Although this increases participant burden, having a comprehensive and up-to-date expectancies measure may be of greater interest to researchers whose primary interests involve expectancies, including those developing expectancy-related interventions. Adding items pertaining to technology (e.g., drunk texting/dialing), sexuality (e.g., feel sexier than usual), and severe physical consequences (e.g., blacking out) is important given findings published since the validation of the original measure showing that expectancies in these domains are related to alcohol use and consequences (e.g., Florimbio et al., 2018; Miller et al., 2020) and to evaluations of negative consequences (e.g., Merrill et al., 2019). The additional items also expand the potential options for item-level analyses (e.g., associations between alcohol–sex expectancies and sexting). The expanded measure is not intended to replace the original measure. Rather, it is intended to provide a more comprehensive daily expectancies measure for researchers whose primary interests are related to expectancies. The original measure may be better suited for studies in which expectancies are not the primary focus and/or where minimizing participant burden is a priority.

When analyses were conducted on subsamples limited to females, males, and days without prior alcohol use, findings were generally similar to those of the main analyses conducted using the full sample. Regarding sex, although measurement invariance was not tested, findings indicated that the measure performed relatively consistently between females and males. Findings among the male subsample diverged from findings among the full sample in a few instances. This may have been because of some expectancies (e.g., feel sexier than usual) being less salient and/or less clearly having a positive or negative valence for males than females. Diverging findings could have also been attributable to lower power and/or less representativeness in the male subsample, as the female subsample contained nearly twice as many participants. The second set of sensitivity analyses that were limited to days without prior alcohol consumption were nearly identical to the findings among the full sample, indicating that the main findings were not significantly biased by days in which prior alcohol consumption had already occurred.

Limitations

This work had several limitations. First, the sample was limited to 2-year and 4-year college students, so it is unclear how well these findings generalize to young adults who are not college students. Second, the sample used here was fairly high-risk regarding alcohol use (by design), as all participants reported heavy episodic drinking in the past 2 weeks at screening. Therefore, it is unclear to what extent findings generalize to students who drink in lesser quantities.

Future directions

Although these findings provide initial evidence that this expanded daily expectancies measure is psychometrically sound, there are several important next steps. First, evidence of other types of validity is needed to fully validate this measure. For instance, future work should aim to demonstrate the predictive validity of the measure by testing daily associations between scores on this expectancies measure and alcohol use and use-related consequences. Second, it will be important for future work to demonstrate that this measure can adequately detect change in expectancies over time (e.g., as a result of intervention). Third, future work should assess whether this measure has incremental validity over the shorter, previously developed expectancies measure. Evidence of incremental validity would indicate that the expanded measure is more favorable for measuring daily alcohol expectancies among young adult drinkers than the shorter measure and that the expanded measure could be used in place of the shorter measure when it could be accommodated into a study's protocol. Until incremental validity is established, both measures should be considered equally viable and psychometrically sound options for measuring daily expectancies among young adult drinkers. Fourth, confirmatory factor analyses should be conducted in other samples to verify the factor structure and demonstrate that it generalizes across samples. Fifth, it would be of interest to examine whether there is measurement invariance across sex, race/ethnicity, and student status. Sixth, regularization approaches could be used to evaluate differential item functioning.

Conclusions

Research continues to show that alcohol expectancies vary both across days and between young adults. As alcohol use research increasingly uses daily designs, it is important for expectancy measures to be designed for daily administration and to be capable of measuring expectancies well at both the daily and person levels. The expanded 23-item daily alcohol expectancies measure examined here demonstrated reasonably sound psychometric properties that suggest it is appropriate for use in daily process studies of alcohol use as well as in daily interventions focused on alcohol expectancies.

Conflict-of-Interest Statement

The authors have no conflicts of interest to report.