Adolescents were recruited from 2015–17 from high schools in West Virginia, Ohio, and Pennsylvania (n = 513) as well as an adolescent medicine clinic in north-central West Virginia (n = 54). The sample consisted of 567 English-speaking adolescents enrolled in high school and aged 14–18 years. Adolescents exhibiting significant cognitive deficits (identified by teachers or clinic staff) were excluded from participation.

Recruitment sites and the university Institutional Review Board approved study procedures. At schools, research staff described the study to students and asked adolescents to give invitation letters and consent forms to parents/legal guardians; consent forms included a box for participants to indicate if they did or did not want to participate. Researchers returned 2–7 days later to collect signed consent forms and administer questionnaires. In the adolescent medicine clinic, medical providers gave a brief study description to identified eligible adolescents. For those interested, research staff then provided details about participation and obtained informed consent (caregivers and adolescents ≥ 18 years old) and assent (adolescents < 18 years old).

Adolescents completed a questionnaire packet including but not limited to the measures described below. Youth who returned consent forms, regardless of whether they chose to participate, were entered into a lottery to win $20 gift cards. Participating youth were entered into a different lottery drawing to win $20 gift cards.

Analyses were conducted with IBM SPSS v24. Bartlett’s test of sphericity (Bartlett, 1954) and Kaiser-Meyer-Olkin (Kaiser, 1974) were used to determine if the correlation matrix was appropriate for factor analysis. A common factor analysis with a direct oblimin rotation was conducted to determine the latent factor structure of the EESA (Fabrigar et al., 1999). Empirical Kaiser Criterion keying (Braeken & van Assen, 2017), visual inspection of the scree plot, and theoretical relevance were used to determine the number of factors to retain. Descriptive statistics performed before the EFA indicated that individual items were non-normally distributed; thus, an unweighted least squares extraction method was used due to its tolerance of non-normality (Briggs & MacCallum, 2003). Items with factor loadings ≤ .40 were examined for removal. The final factor structure was chosen based on four criteria (Watkins, 2018): (a) achieving simple structure (secondary factor loadings < .30) (Thurstone, 1947); (b) at least three items loading onto each factor; (c) internal consistency ≥ .70 for each factor; and (d) each factor being theoretically meaningful.

Confirmatory factor analysis (CFA) was used to examine measurement invariance for youth by ECIG use category (never vs. ever/current use) following guidelines outlined by Putnick and Bornstein (2016). Configural invariance, which indicates that the number of factors and pattern of loadings is the same between the two groups, was supported if the configural model demonstrated adequate fit (Root Mean Square Error of Approximation [RMSEA] ≤ .08, Comparative Fit Index [CFI] ≥ .90, Standardized Root Mean Square Residual [SRMR] = .08). Metric, scalar, and residual invariance were examined by comparing the two-group model with all free parameters (configural model) to a model with constrained loadings (metric), loadings and intercepts (scalar), and loadings, intercepts, and residuals (residual). Non-significant chi-square and absence of change in model fit statistics (RMSEA, CFI) were evidence of invariance.

Convergent validity was assessed using univariate analyses and regression models involving variables conceptually related to ECIG use and ECIG expectancies. In three linear regression models, demographic characteristics (age, gender), cigarette use, personality (extraversion, conscientiousness, sensation seeking), and social factors (parental and peer ECIG use) were regressed onto each of the three EESA subscales. One-way ANOVA (with Tukey’s HSD post-hoc tests) was used to compare EESA subscales across levels of ECIG use (never users, lifetime ECIG use, past 30-day ECIG use). Then, four logistic regressions were used to predict intention to use ECIGs and the likelihood of ECIG use based on the EESA subscales, controlling for age, gender, and cigarette use. Models were run to predict intention to use ECIGs (yes vs. no), current ECIG versus nonuser, lifetime ECIG versus nonuser, and current ECIG versus lifetime ECIG use. Results were considered statistically significant at p < .05.

Table 1 displays descriptive statistics for demographics, academic characteristics, and ECIG and cigarette use. Recruitment site (clinic vs. school), grade in school, and peer and parent ECIG use were associated with ECIG use.

A significant Bartlett’s test of sphericity (X²[903] = 18916.63, p < .001) indicated a non-random correlation matrix. The Kaiser-Meyer-Olkin statistic was 0.95, which is above the cut-off of .70 for sampling adequacy (Kaiser, 1974). Thus, the correlation matrix was deemed appropriate for factor analysis.

The Empirical Kaiser Criterion suggested a 7-factor solution was the appropriate number of factors to extract, while examination of the scree plot indicated a 3-factor solution would be most suitable; thus, 7- through 3-factor solutions were examined. A 7-factor solution was investigated first. When an adequate solution was not produced (i.e., not meeting all four criteria noted above), the factor analysis was conducted with one less factor and repeated until a final solution meeting all four criteria was found. The 7- and 6-factor solutions were rejected due to under-identified factors (i.e., < 3 items with loadings ≥ .40). The 3-, 4-, and 5-factor solutions included at least three items per factor but had items with secondary loadings ≥ .30. To address issues with cross-loadings, communalities and factor loadings for each factor solution were examined to potentially remove problematic items. An iterative approach was used when removing items that performed poorly; the EFA was re-run after removing an item to re-examine fit with the respective solution.

After reviewing the final scales, the 5- and 4-factor solutions were rejected because all items related to health costs (e.g., “Hurt your lungs”) had been removed. The relative health risks of ECIGs compared to cigarettes are consistently associated with ECIG use (Wills et al., 2015); removing these items may threaten the overall content validity of the measure. Additionally, the variance accounted for in the 5- (64.45%) and 4- (61.28%) and 3-factors solutions (57.92%) were comparable. Therefore, the 5- and 4-factor solutions did not bestow statistical advantages. Four items with cross-loadings ≥ .30 were removed from the 3-factor solution, resulting in a 39-item measure, but items related to health costs were retained. Some items related to affective benefits and weight/appetite control were removed, but the final measure still included items related to these constructs (e.g., “Feel less bored,” “Help you from overeating”). Therefore, the 3-factor solution was determined to have the best theoretical and statistical fit.

Factor loadings for the 3-factor solution (Table 2) ranged from 0.41 to 0.87, and were labeled as, “Costs,” “Affective/Weight Benefits,” and “Social Benefits.” The Costs subscale (21 items) included items that reflected perceived health, addiction, social, appearance, and financial consequences associated with using an ECIG. The Affective/Weight Benefits subscale (11 items) reflected mood control, boredom reduction, and weight control associated with using an ECIG. Finally, the Social Benefits subscale (7 items) pertained to possible social advancement and positive appearances associated with using an ECIG (e.g., look more attractive, gain respect from your friends). The Costs subscale was not strongly associated with Affective/Weight Benefits (r = 0.01) or Social Benefits (r = −0.10), though Affective/Weight Benefits and Social Benefits were strongly correlated (r = 0.50).

Results from the two-group CFA indicated that the EESA has configural invariance based on ECIG use (Χ² [1332] = 2935.79, p < .001; RMSEA = .07 [95%CI = .06-.07]; CFI = .90; SRMR = .08). Additionally, results satisfied metric invariance (Χ² (36) = 46.07, p = .12; ΔRMSEA = 0.003; ΔCFI=0.000) but not scalar invariance (Χ² (36) = 1334.37, p < .001; ΔRMSEA = 0.005; ΔCFI=0.085). Residual invariance was not examined as scalar invariance was not satisfied. Overall, this suggests that the factor structure and loadings of the EESA fit well for youth who had and had not use ECIG’s and therefore support comparison of the EESA across ECIG user groups.

Internal consistencies were excellent for the Costs (α = .97), Affective/Weight Benefits (α = .91), and Social Benefits (α = .88) subscales, and item-to-total correlations ranged from .49 to .84.

Three linear regressions were used to predict the EESA subscales (Table 3). Higher perceived costs of ECIG use were associated with identifying as male, higher levels of conscientiousness, lower levels of sensation seeking, and having fewer peers that tried ECIGs. Higher perceived Affective/Weight Benefits were associated with having smoked a cigarette at least once in their lifetime, lower levels of conscientiousness, higher levels of sensation seeking, and having more peers that tried ECIGs. Finally, higher perceived Social Benefits were associated with lower levels of conscientiousness and having more peers that tried ECIGs.

Table 1 displays descriptive statistics for the EESA subscales, conscientiousness, extraversion, sensation-seeking, peer ECIG use, and parental ECIG use. In results from one-way ANOVA’s there were statistically significant differences in EESA subscales by ECIG use status (Table 1). Post-hoc tests revealed that all groups differed significantly on Costs (nonuser > lifetime user > current user) and Affective/Weight Benefits (current user > lifetime user > nonuser). Both current and lifetime user groups had higher Social Benefits scores than nonusers, but differences were not statistically significant.

Logistic regressions (Table 4) were used to predict intention to use ECIGs and ECIG use group based on EESA subscale scores. Higher perceived Costs was associated with reduced odds of intending to use and ECIG or being a lifetime ECIG user compared to nonuser, while higher perceived Affective/Weight Benefits was associated with increased odds of intending to use ECIGs or being a lifetime ECIG user compared to nonuser. The same pattern of results for Costs and Affective/Wiehgt Benefits was found for predicting current ECIG use compared to nonusers and for predicting current ECIG use compared to lifetime ECIG use.

The SESA (Hine et al., 2007) has eight first-order subscales, split evenly between benefits (Affect Control, Social Benefits, Boredom Reduction, Weight Control) and costs (Appearance-Presentation Costs, Health Costs, Social Costs, Addiction), as well as two second-order factors (overall Benefits and Costs). The EESA factors more closely mirror these latter SESA factors, with two subscales for benefits (Affective/Weight Benefits and Social Benefits) and one for costs (combines appearance/presentation, health, social, and addiction factors). Also similar to that observed for the SESA (Hine et al., 2007), the EESA subscales had statistically significant associations with ECIG use and peer ECIG use. This overall pattern of results suggests that ECIG and cigarette expectancies are different yet related, as noted elsewhere (Morean & L’Insalata, 2017).

The EESA possesses similarities with other measures of ECIG expectancies. The EESA Social Benefits and Affective/Weight Benefits subscales mirror the Social Enhancement (“Become more popular”), Positive Social Expectancies (“ECIGs help me socialize”), Affect Regulation (“Feel less bored”), and Positive Internal Expectancies (“Feel more relaxed”) subscales identified in young adults (Correa et al., 2019; Pokhrel et al., 2014) and adolescents (Cristello et al., 2020), as well as the Social Influences (“Be more popular”) and Enjoyment (“Like it”) subscales identified in adolescents and young adults (Barker et al., 2019; Cristello et al., 2020) using other measures. The consistent finding of subscales defined by social factors is not surprising; users often tout that ECIGs facilitate their interactions with others, are permissible in many locations, and engender less stigma than cigarette smoking (Soule et al., 2017; Soule et al., 2016). Some users also report that ECIGs help to relieve unpleasant states (e.g., stress, boredom) and to control their appetite (Harrell et al., 2019; Morean et al., 2019), similar to that for cigarette smoking (Mathew et al., 2014). Thus, social, affective, and weight benefits appear to be salient ECIG expectancies for both populations (Barker et al., 2019; Correa et al., 2019; Cristello et al., 2020; Pokhrel et al., 2014).

Interestingly, items related to affect and weight control loaded onto the same subscale, in contrast to other work (e.g., Cristello et al., 2020). Smoking to manage stress has been associated with decreased food intake and weight loss (Meule et al. 2018). Therefore, it may be that when adolescents perceive benefits in affect control they also perceive benefits in weight control, resulting in these items loading onto the same subscale. It should also be noted that items assessing positive (feel good) and negative (relax, stress relief) affect reinforcement loaded onto the same scale, which is congruent with other ECIG expectancies scales (Correa et al., 2019; Cristello et al., 2020). The pattern of results herein may be because only one item assessed positive affect reinforcement. However, future research should examine whether adolescents differentiate between positive and negative affective benefits.

Nonetheless, the EESA differs from ECIG expectancy measures in the number of costs subscales. Only a single cost subscale was identified for the EESA, consistent with two expectancy measures validated with adults (Correa et al., 2019; Morean & L’Insalata, 2017). However, other work has identified additional cost subscales (health, addiction, smoker association, social costs) in adolescents (Barker et al., 2019; Cristello et al., 2020) or adults (Pokhrel et al., 2014). Cross-study differences may be due to the types of items included in the measure. The EESA, for example, excludes items related to sensory experience (e.g., taste, smell, hand-to-mouth movement) and to cigarette smoking (e.g., ECIG use for cigarette cessation or harm reduction; look like a cigarette smoker). Differences also may be due to the tobacco use history of those sampled; rates of lifetime (18.1%) and current (7.8%) cigarette use were low relative to those in previous research (Barker et al., 2019; Pokhrel et al., 2014). Future qualitative work may help refine the items for the costs subscale on the EESA and other expectancy measures.

Importantly, results support previous work in that the Costs and Affective/Weight Benefits EESA subscales were associated with ECIG use, peer ECIG use, and sensation-seeking in the expected directions (Barker et al., 2019; Correa et al., 2019; Pokhrel et al., 2014). The Social Benefits EESA subscale was associated with ECIG use in univariate but not multivariable analyses. It may be that other variables (e.g., expectancies, concurrent cigarette use) are stronger predictors of ECIG use than Social Benefits. The average score on the Social Benefits subscale also was low relative to scores on the other EESA subscales, possibly impacting findings. Finally, results expand the literature by demonstrating associations between ECIG expectancies and conscientiousness, a risk factor for smoking initiation and progression (Kubicka et al., 2001; Terracciano et al., 2008; Zvolensky et al., 2015). Consistent with the literature on personality and risk-taking during adolescence (Gullone & Moore, 2000), less conscientious adolescents may perceive using ECIGs as less risky.

Strengths of the current study include the adaptation of a previously validated measure for cigarette smoking expectancies, an approach used by others (Pokhrel et al., 2014), and a relatively large sample. We also used the recommended best practices for EFA to promote statistical validity and examined convergent validity with a variety of psychological and social variables, in addition to ECIG use. Limitations include the cross-sectional study design; predominantly white, rural/suburban, and academically successful sample; inability to assess discriminant validity; and limited assessment of ECIG use (e.g., quantity of use). Additionally, data were collected prior to the widespread use of pod/salt devices (e.g., Juul), and expectancies regarding these devices may be different. Future research should use nationally representative samples who use more current devices, examine discriminant validity, and further examination of measurement invariance for other sociodemographic and ECIF use groups (never vs. ever vs. current use). Confirmatory factor analysis for the EESA is also needed using a different sample.

The EESA is a valid and reliable measure of adolescents’ expectancies of ECIG use. While the EESA used the same items as the SESA, albeit referencing ECIG rather than cigarette use, the factor structure is different from that reported by Hine and colleagues (Hine et al., 2007). This suggests that the EESA more accurately measures adolescents’ expectancies of using an ECIG than the SESA’s original factors. The EESA can help researchers obtain valid assessments of adolescents’ beliefs around ECIG use and how these expectancies relate to ECIG use, which can inform messaging used in prevention and cessation interventions (e.g., highlighting the potential costs of ECIG use).