The Anticipated Effects of Cannabis Scale (AECS): Initial Development and Validation of an Affect- and Valence-Based Expectancy Measure

Abstract

Prior research suggests that cannabis expectancies are related to cannabis misuse and problems. Although there are established measures of cannabis expectancies, existing measures have psychometric limitations and/or are lengthy. Existing measures typically have a two-factor structure of positive and negative expectancies, but recent conceptualizations of alcohol expectancies support a valence-(positive vs. negative) and arousal-based (high vs. low arousal) structure. Thus, the present study sought to test a similar structure for cannabis. Cannabis expectancy items underwent 2 preliminary studies, assessing item valance/arousal (n = 233) and relevance to cannabis (n = 124). A final pool of 76 items underwent exploratory factor analysis (n = 303), and remaining items underwent confirmatory factor analysis in a separate sample (n = 469). Lastly, an additional sample (n = 435) examined validity. Results suggested a 3-factor structure (general positive, high arousal negative, low arousal negative) for the 17-item Anticipated Effects of Cannabis Scale (AECS), which was invariant across cannabis use frequency, sex, and race/ethnicity. Positive expectancies were strongly associated with cannabis use, whereas low arousal negative expectancies were protective against cannabis frequency; high arousal negative expectancies were strongly associated with more negative consequences and dependence symptoms. In addition, the proposed interpretation of AECS test scores showed evidence of incremental validity relative to another abbreviated measure. The current study provides initial support for the AECS, a brief, psychometrically sound cannabis expectancies measure. The AECS captures the full range of cannabis effects and may be suited to test discrepancies between cannabis expectancies and subjective response. Additional research is needed to validate its structure and predictive utility.

Social learning theory (Bandura & Walters, 1977) posits that human behavior is determined by a constant interplay among cognitions, behaviors, and social environment. This interplay results in a form of reciprocal determinism, where both individual (i.e., cognitions and behaviors) and environmental influences affect one another and lead to learned associations about engaging in a specific behavior. In social learning, beliefs about appetitive or aversive effects of a behavior, namely outcome expectancies, are formed through experiential learning, either directly or through observing other’s behavior. Outcome expectancies have been identified as a particularly important predictor of substance use, with expectations for positive or negative reinforcement (e.g., becoming sociable, alleviating negative affect) guiding use behavior (e.g., Maisto et al., 1999).

Substance-specific outcome expectancies have been extensively studied, with much of the research focused on alcohol. Alcohol outcome expectancies are thought to influence drinking behavior both directly (Ham & Hope, 2003; Jones et al., 2001) and indirectly through drinking motivation (e.g., Cooper et al., 1995). Decades of research suggest that positive alcohol expectancies are linked to a family history of Alcohol Use Disorder (Mann et al., 1987; Waddell et al., 2020), earlier adolescent drinking and heavier drinking (e.g., Bekman et al., 2011; Cooper et al., 1988; Corbin et al., 2015; Ham & Hope, 2003; Jones et al., 2001; Lac & Luk, 2019), alcohol-related problems (e.g., Corbin et al., 2015; Pabst et al., 2014; Turrisi et al., 2000), development of alcohol use disorder (AUD; e.g., Brown et al., 1985; Cooper et al., 1988; Jones et al., 2001), and poor AUD treatment outcomes (e.g., Sebold et al., 2017). Less is known about negative alcohol expectancies, which inconsistently are linked with use behavior (e.g., Fromme et al., 1993; Greenfield et al., 2009; Lac & Luk, 2019; Lee et al., 1999; McMahon & Jones, 1993; Walther et al., 2019). However, recent conceptualizations of alcohol expectancies have moved beyond the dichotomy of positive versus negative expectancies, recognizing that positive and negative expectancies can be broken down further by level of arousal.

Research has shown that alcohol expectancies comprise distinct expectancies across the full affective space of valence by arousal, yielding four quadrants of effects including high arousal positive (HIGH+; e.g., “sociable”), high arousal negative (HIGH−; e.g., “aggressive”), low arousal positive (LOW+; e.g., “relaxed”), and low arousal negative (LOW−; e.g., “dizzy”) factors (Morean et al., 2012). HIGH+ alcohol expectancies are positively associated with drinking quantity, binge drinking and problems; HIGH − alcohol expectancies are associated with binge drinking and problems; and both LOW+ and LOW− alcohol expectancies are negatively associated with drinking indices; although LOW+ only are associated with less binge drinking (Morean et al., 2012, 2016). Thus, relations between expectancies and drinking behavior appear to differ based on both the valence and arousal of expectancies, and a better understanding of these unique effects may facilitate tailored prevention efforts targeting expectancies.

In contrast to the extensive literature on alcohol expectancies and their treatment implications (e.g., Darkes & Goldman, 1993), the literature on cannabis outcome expectancies is less developed. The Marijuana Effect Expectancy Questionnaire (MEEQ; Schafer & Brown, 1991; Aarons et al., 2001), one of the most commonly used cannabis expectancies measures, divides expectancies into positive (i.e., relaxation and tension reduction, social and sexual facilitation, perceptual and cognitive enhancement) and negative (i.e., cognitive and behavioral impairment, global negative effects and craving/physical effects) subscales. These positive and negative expectancy subscales relate to cannabis use, but relations differ for cannabis frequency/quantity versus dependence. Scores on the positive subscales of the MEEQ are related to elevated frequency and quantity of use (Bolles et al., 2014; Boden et al., 2013; Buckner, 2013; Buckner & Schmidt, 2008; Hayaki et al., 2010, 2011; de Dios et al., 2010; Frohe et al., 2018; Kristjansson et al., 2012; Neighbors et al., 2008; Schmits et al., 2016), whereas scores on the negative subscales of the MEEQ typically are related to less frequent cannabis use (Aarons et al., 2001; Buckner, 2013; Buckner & Schmidt, 2008; Hayaki et al., 2010, 2011; Kristjansson et al., 2012; Schmits et al., 2016; Vangsness et al., 2005). However, scores on the positive subscales of the MEEQ less consistently are linked with cannabis problems/risk behavior (i.e., driving while high; Arterberry et al., 2013; Aston et al., 2016), whereas scores on the negative subscale consistently are related to elevated cannabis dependence symptoms/problems (e.g., Arterberry et al., 2013; Buckner, 2013; Buckner & Schmidt, 2008; Foster et al., 2016; Hayaki et al., 2010, 2011; Schuster et al., 2019). Negative cannabis expectancies also are associated with perceived benefits from reducing cannabis use, above and beyond measures of cessation-specific expectancies (Metrik et al., 2017). Although research using the MEEQ has led to important insights regarding relations between expectancies and cannabis use outcomes, at 48 items, the MEEQ is relatively time-intensive to administer. To make the MEEQ shorter and more accessible, a brief six-item version (B-MEEQ) was validated (Torrealday et al., 2008), encompassing positive and negative subscales (three items each). The B-MEEQ shows similar associations with cannabis frequency and problems (Brackenbury et al., 2016; Torrealday et al., 2008), and lower B-MEEQ negative expectancies uniquely are associated with drinking and driving (King et al., 2020). The B-MEEQ’s brevity is conducive to its inclusion in large longitudinal study batteries (e.g., ABCD Study; Lisdahl et al., 2018).

Although the MEEQ is the most widely used cannabis expectancy measure, it does not differentiate items based on arousal, which may contribute to differential prediction of cannabis outcomes, much like it does for alcohol use (e.g., Morean et al., 2012, 2013). In fact, most MEEQ subscales measure inherently low arousal effects, except for some items on the social/sexual facilitation and perceptual/cognitive enhancement subscales (both positive). In addition, the measure also asks solely about expectancies for smoked marijuana, which refers to the flower of the cannabis plant, failing to capture expected effects associated with a variety of more potent cannabis products (e.g., concentrates) and administration modes (e.g., edibles; e.g., Jones et al., 2018; Meier, 2017; Meier et al., 2019; Morean & Butler, 2019; Schauer et al., 2020). Although the B-MEEQ is a brief instrument at six items, it has notoriously poor internal consistency (e.g., Torrealday et al., 2008; Brackenbury et al., 2016) and uses quadruple barreled items (e.g., “marijuana makes it harder to think and do things; harder to concentrate or understand; slows you down when you move”). Lastly, the MEEQ was developed in a predominately nonusing sample of college students who had little direct experience with cannabis.

The Cannabis Expectancy Questionnaire (CEQ; Connor et al., 2011) is a more recent measure that addresses some limitations of the MEEQ. For example, the CEQ was developed and validated in a cannabis-using sample. In addition, the CEQ includes several items about anxiety and paranoia on the negative expectancy subscale, which were absent in the MEEQ despite their relation to cannabis use (e.g., D’Souza et al., 2004). The CEQ has a two-factor structure with positive and negative expectancies, and several studies have found that positive expectancies on the CEQ are related to cannabis use frequency, whereas negative expectancies are related to higher cannabis dependence severity, treatment seeking, and abstinence during treatment (Connor et al., 2011, 2014; Gullo et al., 2017; Papinczak et al., 2017, 2019), similar to the MEEQ.

Although the CEQ addresses some of the limitations of other measures like the MEEQ, it is similar in length to the full MEEQ (45 items), making it time-consuming to administer. The CEQ also uses the wording “smoking cannabis,” which limits the interpretation to just one mode of cannabis use (i.e., “smoking”). Most importantly, the CEQ examines expectancy valence (positive vs. negative) but it does not differentiate items based on arousal, despite including several high arousal items in both subscales. Both positive and negative expectancies may differ by level of arousal, so making this distinction may improve prediction of frequent and compulsive cannabis use. As previously mentioned, past conceptualizations of alcohol expectancies have benefited from a valence by arousal model, and there is reason to suspect that cannabis expectancies may as well. Research suggests that delta-9-tetrahydrocannabinol (THC; i.e., the primary psychoactive constituent of cannabis) increases physiological arousal (Wachtel et al., 2002), and subjective experiences of cannabis effects differ across heavier versus lighter users (e.g., Metrik et al., 2011), potentially due to tolerance to the drug’s acute effects. Thus, acute cannabis use may increase arousal, but the experience of such arousal may be predicated on one’s typical cannabis use and biological predispositions that may impact experiential learning. High arousal positive expectancies may confer risk for more frequent and heavier cannabis use due to their appetitive nature and vividness compared to lower arousal expectancies, whereas low arousal negative expectancies may be protective against more frequent and heavier cannabis use, much like for alcohol (Morean et al., 2012). Conversely, higher arousal negative expectancies may be associated with risk for cannabis problems and/or dependence symptoms, due to the stronger aversive nature and intensity of these effects.

Prior studies provide some support for this notion. For example, using the MEEQ, Aarons et al. (2001) found that heavier users endorsed more social/sexual facilitation and perceptual/cognitive enhancement expectancies compared to lighter users, though they found no group differences for relaxation expectancies. In addition, the items on the CEQ that load the highest for positive (i.e., “I have more self-confidence when smoking cannabis,” “I smoke cannabis to get full enjoyment out of life”) and negative (i.e., “Smoking cannabis makes me feel insecure,” “When I smoke cannabis I feel ‘panicky’”) expectancies generally are high arousal, potentially explaining their strong associations with cannabis use and problems, respectively. Despite the theoretical appeal of assessing cannabis expectancies based upon both arousal and valence, no measure to date allows for such delineation.

Therefore, the goal of the current study was to develop and validate a measure of cannabis expectancies that addresses the full valence by arousal affective space. The current study also addresses several limitations of past expectancy measures by (a) assessing a broader range of cannabis users from light to heavy, (b) specifying expectancies related to “cannabis” rather than “marijuana” and avoiding references to specific modes of use (e.g., smoked), and (c) using single items/short phrases to reduce time demands. In addition, short items/phrases are conducive to creating parallel measures of cannabis expectancies and subjective response, which is not possible with extant expectancy measures. Discrepancies between parallel alcohol expectancies (Morean et al., 2012) and in-the-moment subjective response to alcohol (Morean et al., 2013) are associated with drinking outcomes (Morean et al., 2015; Waddell et al., in press), and thus validating parallel measures for cannabis would provide the ability to test meaningful discrepancies. Lastly, the present study sought to use best practices in measurement development for a psychometrically sound questionnaire.

Methods Overview

The development and initial interpretation of test scores for the Anticipated Effects of Cannabis Scale (AECS) occurred over two measurement development phases and a subsequent validation phase, using five different samples. In the first phase of measurement development, a pool of 226 items was developed and then narrowed down based upon participant reports of cannabis relatedness (Sample 1) and valence/arousal (Sample 2). In the second phase of measurement development, an exploratory factor analysis was conducted on the selected item pool (Sample 3), subsequent confirmatory factor analyses were conducted in two separate samples (Samples 4 and 5), and measurement invariance (Samples 4 and 5) was tested. In the validation phase, we assessed the convergent, discriminant, concurrent, and incremental validity of the proposed interpretation of AECS test scores (Sample 5). We describe each sample and relevant analytic plan below. Becasue the item development and narrowing procedures (Measurement Development Phase 1) only used descriptive statistics, we outline the participants and procedures in the text and display descriptive statistics in Supplemental Tables S1 and S2 in the online supplemental material. For the samples used in Measurement Development Phase 2 and the Validation Phase, methods and results are displayed in typical fashion.

Measurement Development Phase 1: Item Development and Narrowing

Initial Item Pool

An initial pool of 226 items was derived from past cannabis expectancy measures (i.e., MEEQ, CEQ, MEICA; Alfonso & Dunn, 2007, MMBEQ; Linkovich-Kyle & Dunn, 2001), cannabis subjective response measures and studies (e.g., Haney et al., 1999; Metrik et al., 2012; Quinn et al., 2017), the marijuana motives measure (i.e., Simons et al., 1998), and alcohol expectancy measures (Brown et al., 1987; Fromme et al., 1993; Morean et al., 2012). All items were narrowed down to single words/short phrases to make them appropriate for the stem “I feel.” For example, “When I smoke cannabis I withdraw from others” was changed to “I feel . . . withdrawn.” The final item pool then was evaluated by eight cannabis expectancy/subjective response experts before undergoing pilot testing. Experts all had doctoral degrees in clinical psychology (and one medical doctor); were assistant, associate, or full professors; and had published extensively on alcohol/cannabis use and expectancies. Experts were recruited via e-mail and sent the initial list of words, asking for domains/effects that were not covered on the list. We created hypothesized quadrants of affective space (HIGH+, HIGH−, LOW+, LOW−) to facilitate item narrowing.

Sample 1: Establishing Cannabis Relatedness

Participants

A sample of young adults (N = 124) was recruited from Reddit, a social media platform with community subgroups (i.e., subreddits) focused on cannabis use. Jack T. Waddell posted surveys in subreddits, and community members were invited to participate in the survey and to be entered into a raffle to win one of two $25 gift cards. Participants were 46% female, and between the ages of 18–30 (M_age = 23.78, SD = 3.67). Eligible participants were selected based on past-year cannabis use and being 18 to 30 years of age. Most participants were heavy cannabis users, with over half (60.5%) reporting daily cannabis use (Supplemental Table S1 in the online supplemental material).

Procedure

The study protocol was approved by the Arizona State University institutional review board (IRB). Participants were asked to “Please rate each item on how related it is to cannabis use (i.e., whether people will feel each effect after using cannabis)” on a scale of 1 (not at all) to9 (extremely) for each of the 226 items. All items were shown in blocks of 20–30, which were displayed in random order to participants. Items with a relatedness rating of 3 or higher were retained. However, this method discarded several items that were hypothesized to be within the HIGH− and LOW− quadrants, which are potentially important but not often endorsed items (e.g., D’Souza et al., 2004). In addition, heavy using participants in this sample may have developed substantial tolerance to negative cannabis effects, such that removing these items would have missed important expectancies for lighter users. Thus, we relaxed the criteria of relatedness to 2.25 for HIGH− words and 2.5 for LOW− words, retaining additional face valid cannabis effects. This resulted in 164 items, with at least 25 in each hypothesized quadrant of affective space. Relatedness means for the 164 items are shown in Supplemental Table S2 in the online supplemental material.

Sample 2: Assessing Item Valence and Arousal

Participants

A sample of college students (N = 233) was recruited from the psychology subject pool at a southwestern university. Participants were invited to answer an online survey and received one credit toward fulfillment of their research requirement. Participants were 46% female and between the ages of 18–30 (M_age = 19.15, SD = 1.43). Eligible participants were selected based on past-year cannabis use and being 18 to 30 years of age. Participants were predominantly light to moderate cannabis users, with one fourth (25.3%) of participants reporting daily cannabis use (Supplemental Table S1 in the online supplemental material).

Procedure

The protocol was approved by the institution’s IRB. Participants were asked “how arousing do you think each word is” and “how positive do you think each word is” on a scale of 1 (not arousing) to 9 (extremely arousing) for arousal and a scale of 1 (negative) to 9 (positive) for valence. For each randomly displayed block of items, the Self-Assessment Manikin (SAM; Bradley & Lang, 1994), depicting sedation/stimulation and sadness/happiness, respectively, was used to aid in reporting. The SAM arousal faces start with very content and relaxed looking images and move toward a person who looks very aroused with lightning bolts surrounding them. The SAM valence faces range from a sad, frowning face to a happy, smiling face.

Item Narrowing Procedure

All 164 items from Pilot Study 2 were organized into affective space based on their reported means for valence and arousal. Items were narrowed in the following fashion. First, we plotted words based on their values for valence and arousal and analyzed words that had the same perceived meaning (e.g., slow, slowed down). If items had the same perceived meaning and were in close proximity to one another on the affective space grid, the item with a lower relatedness score was removed from the item pool. Second, we removed items that had multiple meanings (e.g., clear, normal). Third, we removed items that were confusing or had high reading levels. Reading level was determined by entering each item (e.g., “I feel calm”) into the Flesch Kincaid Grade Level Index; high reading level was defined as having a score of 10 or above (10th grade reading level). In each phase of item removal stated above, Jack T. Waddell and William R. Corbin met to determine items to be removed, and all decisions were double checked by Madeline H. Meier and Meghan E. Morean before final removal. After all narrowing steps, 76 items were retained for exploratory and confirmatory factor analyses (Supplemental Table S3 in the online supplemental material).

Measurement Development Phase 2: Factor Analysis

Sample 3: Exploratory Factor Analysis

Sample 3 Method

Participants.

A sample of college students (N = 303) was recruited from the psychology subject pool at a southwestern university. Participants were invited to complete an online survey and received half of a course credit that counted toward their six-credit research participation requirement. Participants were 43% female, and between the age of 18 and 30 (M_age = 19.26, SD = 1.55). Eligible participants were selected based on past-year cannabis use and being 18 to 30 years of age. Participants were predominantly light or moderate cannabis users, with about one fifth (21.7%) reporting daily cannabis use (Supplemental Table S1 in the online supplemental material).

Procedure.

The protocol was approved by the institution’s IRB. The procedures were identical to Pilot Sample 2. Participants were asked to “please rate the extent to which you expect to feel each of the following effects after using cannabis” for each of the 76 items using a 0 (not at all) to 10 (extremely) scale.

Data Analysis.

Exploratory factor analysis was used for preliminary investigation of the AECS factor structure. Because several items were nonnormally distributed, as is typical for expectancies for other substances (e.g., alcohol; Morean et al., 2012), maximum likelihood estimation with robust standard errors (MLR) was used to account for nonnormality and missing data were handled with full information maximum likelihood (FIML). Oblique rotation (direct oblimin) was used to allow for hypothesized factor correlations (e.g., HIGH+ and LOW+), as past research on alcohol suggests substantial correlations between expectancies across valence and arousal (Morean et al., 2012). To determine the optimal factor structure, we used a combination of the screeplot, interpretability of factors, number of items per factor, and theoretical fit, in that order (Jöreskog & Sörbom, 1989; Tabachnick & Fidell, 2001). After determining the best factor structure, items that had high cross-loadings (i.e., higher than .32) and/or low factor loadings (i.e., lower than .55) were removed (Comrey & Lee, 1992; Tabachnick & Fidell, 2001).

Sample 3 Results

The scree plot suggested three to four factors, and factor solutions above five factors were not interpretable (i.e., no common theme among items, few significant loadings). Thus, we evaluated five-factor, four-factor, and three-factor solutions.

In the five-factor solution, the fifth factor had no items that significantly loaded above a .45, which eliminated this factor structure as an option. Similarly, the fourth factor on the four-factor solution had only three items that loaded significantly without substantial cross-loadings, and there was no clear theme for the items in Factor 4. This left a three-factor solution with interpretable factors, each with at least four indicators loading above a .55. The three-factor solution was narrowed by removing indicators that had more than 10% overlapping variance with another factor (i.e., cross-loading above .32) or did not have at least 30% overlapping variance (i.e., loading greater than .55) within its primary factor. To ensure that the order of removing low loadings and cross-loadings did not alter the results, we removed items iteratively using two methods where (a) high cross-loadings were removed first and then low loadings were removed, and (b) low loadings were removed first and then high cross-loadings were removed. Items that were not consistent across both methods (i.e., “stoned,” “distracted,” “dry mouth,” “lost in thought”) were removed. This led to a three-factor structure with 38 items. To reduce time-burden for administering the measure, additional items were removed based on item loadings and content overlap with similar items (see Supplemental Table S3 in the online supplemental material for reasoning related to each item). After further item removal, a 27-item structure was retained (Supplementary Table S4).

The three-factor structure comprised two negative expectancy factors and one general positive expectancy factor. The negative factors differentiated high and low arousal items, however the positive factor did not. In addition, even when a four-factor solution was investigated, positive effects did not appear to differ as a function of arousal, as the few small loadings on the fourth factor did not indicate LOW+ effects (i.e., “high”, “stoned”, “thirsty”). In the final three-factor structure, HIGH− cannabis expectancies were significantly correlated with LOW− cannabis expectancies, r = .52, p < .001. Although statistically significant, the correlation between HIGH− and LOW− expectancies suggested correlated but distinct factors. Positive cannabis expectancies were not significantly correlated with either negative subscale.

It is worth noting that the current exploratory factor analysis (EFA) removed two items that were deemed, by experts, as particularly important for content validity (i.e., “relaxed” and “calm”). Therefore, we retained these items in the confirmatory factor analysis (CFA) stage.

Sample 4: CFA

Sample 4 Method

Participants.

A sample of young adults (N = 469) was recruited from cannabis subreddits. Jack T. Waddell posted surveys in subreddits, and community members were invited to participate in the survey and to be entered into a raffle to win one of two $50 gift cards. Participants were 71% male, and between the ages of 18–45 years (M_age = 29.52, SD = 8.92). This study included participants with an older average age to (a) obtain a larger sample of cannabis users and (b) validate the factor structure in an age-heterogeneous adult sample. Eligible participants were selected based on past-year cannabis use and being ages 18 to 45 years. Participants predominantly were heavy cannabis users, with over half (66.1%) of participants reporting daily cannabis use (see Supplementary Table S1 in the online supplemental material).

Procedure.

This protocol was approved by the institution’s IRB. The procedure for survey responses and raffle participation was identical to Pilot Sample 1. Participants were asked to “please rate the extent to which you expect to feel each of the following effects after using cannabis” for the 29 items on a scale from 0 (not at all) to 10 (extremely). Participants also reported their biological sex and indicated how often they used cannabis on a scale from 0 (never) to 12 (more than once a day).

Data Analysis.

An initial CFA model fit the 29 items based on the factor structure identified in the EFA using Mplus Version 8 (Muthén & Muthén, 1998–2019). In the case of inadequate model fit, modification indices (MI) >20 were used to identify significant cross-loadings. After removing any cross-loading items, items with large covariances with other factors (MI > 20) and low factor loadings (<.45) were removed. Finally, remaining items with factor loadings <.45 were removed (Comrey & Lee, 1992; Tabachnick & Fidell, 2001). When deciding between two items, the item with the higher relatedness score from Pilot Study 1 was retained. The present study defined adequate model fit as root mean square error of approximation (RMSEA) values <.08, comparative fit index (CFI) and Tucker-Lewis index (TLI) indices close to .95, and standardized root mean square residual (SRMR) <.08 (Bentler, 1990; Browne & Cudeck, 1993; Hu & Bentler, 1999). Missing data were handled using FIML.

Sample 4 Results

The 29-item model provided poor fit to the data (RMSEA = .094, CFI = .72, TLI = .70, SRMR = .09). Thus, items that had large MI values indicating overlapping content were removed. The following item pairs all had MI values >20: “outgoing” and “sociable,” “like time is slowed” and “altered sense of time,” “relaxed” and “calm,” “nauseous” and “dizzy,” “inspired” and “creative,” “sociable” and “confident,” and “secure” and “in control.” After removing items “outgoing, altered sense of time, calm, nauseous, inspired, confident, and secure,” model fit still was still inadequate (RMSEA = .064, CFI = .86, TLI = .85, SRMR = .08). Thus, items that had large covariances with other factors and low loadings were removed. This resulted in the removal of “in control” (β = .40, MI covariance >20), and “relaxed” (β = .38, MI covariance >20). Next, we removed items with low factor loadings, including “dizzy,” “like time is slowed,” and “moody” (β < .45). The remaining 17-item factor structure (see Table 2 and Figure 1) provided adequate fit to the data (RMSEA = .052, CFI = .93, TLI = .92, SRMR = .06). All test scores had adequate internal consistency (ω = .76–.86).

Table 2.

Descriptive Statistics and Bivariate Correlations for Sample 5

Variable	M (SD)	1	2	3	4	5	6	7	8	9	10	11
1. AECS Pos	5.76 (2.08)	—	−.29**	−.16**	.72**	−.29**	.58**	−.29**	.33**	.16**	.07	.10*
2. AECS LOW−	5.67 (2.23)		—	.46**	−.33**	.54**	−.17**	.50**	−.26**	−.11*	.18**	.05
3. AECS HIGH−	3.36 (2.37)			—	−.29**	.66**	−.25**	.41**	−.18**	−.10*	.23**	.13**
4. CEQ Pos	3.14 (.62)				—	−.36**	.68**	−.32**	.40**	.23**	.09†	.11*
5. CEQ Neg	2.49 (.59)					—	−.33**	.66**	−.28**	−.08	.35**	.22**
6. B-MEEQ Pos	3.92 (.71)						—	−.20**	.32**	.18**	.15**	.15**
7. B-MEEQ Neg	3.26 (.60)							—	−.30**	−.10*	.25**	.16**
8. Frequency	7.58 (3.37)								—	.38**	.40**	.34**
9. Quantity	1.89 (1.38)									—	.29**	.22**
10. Consequences	10.57 (7.9)										—	.71**
11. Dependence	1.57 (2.02)											—

Note. AECS = Anticipated Effects of Cannabis Scale; CEQ = Cannabis Expectancy Questionnaire; B-MEEQ = Brief-Marijuana Effect Expectancy Questionnaire; Pos = positive subscale; LOW− = low arousal negative subscale; HIGH− = high arousal negative subscale.

^†p < .10.

^*p < .05.

^**p < .01.

Figure 1.

Anticipated Effects of Cannabis Scale Item Plot

Sample 5: CFA in a Separate Sample/Measurement Invariance

Sample 5 Methods

Participants.

College students (N = 435) were recruited from the psychology subject pool at a southwestern university. Participants were invited to answer an online survey and receive 1 credit toward fulfillment of their research requirement. Participants were 61% female, and between the ages of 18–30 years (M_age = 18.88, SD = 1.53). Eligible participants were selected based on past-year cannabis use and being ages 18 to 30 years. Participants predominantly were light to moderate cannabis users, with about one fifth (19.1%) of reporting daily cannabis use (see Supplementary Table S1 in the online supplemental material).

Procedure.

This protocol was approved by the institution’s IRB. The participant procedures were identical to Pilot Sample 2.

Measures.

Demographics.

Biological sex, race, and ethnicity were assessed.

AECS.

The 17-item AECS was administered to assess expectancies for high arousal negative (e.g., paranoid, like my heart is racing), low arousal negative (e.g., lazy, sluggish), and general positive (e.g., sociable, heightened senses) cannabis effects (Appendix). Participants were asked to “Please rate the extent to which you expect to feel each of the following effects after using cannabis” on a scale from 0 (not at all) to 10 (extremely).

B-MEEQ.

The B-MEEQ (Torrealday et al., 2008) is a six-item questionnaire assessing positive (i.e., relaxation and tension reduction, social and sexual facilitation, cognitive and perceptual enhancement) and negative (i.e., cognitive and behavioral impairment, global negative effects, craving, and physical effects) expectancies. Each item is measured on a scale from 1 (Strongly Disagree) to 5 (Strongly Agree). The positive subscale had adequate internal consistency (ω = .71) and the negative subscale, similar to other samples (e.g., Torrealday et al., 2008), had poor internal consistency (ω = .30) in the current sample.

CEQ.

The CEQ (Connor et al., 2011) is a 45-item questionnaire assessing positive (e.g., “I have more self-confidence when smoking cannabis,” “Smoking cannabis makes me feel excited”) and negative (e.g., “When I smoke cannabis I feel ‘panicky’,” “Smoking cannabis makes me more easily irritated”) expectancies. Each item is measured on a scale from 1 (strongly disagree) to 5 (strongly agree). Both subscales had excellent internal consistency (ω = .89 –.90).

Cannabis Use.

Cannabis use frequency was assessed by asking participants “How often do you use cannabis (marijuana, hash, or marijuana extracts or concentrates like rosin, hash oil/BHO/dab/wax/shatter)?” on a scale from 0 (never) to 12 (more than once a day). Cannabis use quantity was assessed by asking participants “How many cones/joints/pipes did you typically have on days when you were using cannabis? If you use cannabis in a form that isn’t cones/joints/pipes, please guess about how many cones/joints/ pipes you would typically use?”

Cannabis Consequences.

Negative cannabis consequences were assessed via an adapted version (i.e., “cannabis” instead of “marijuana”) of the Marijuana Consequences Questionnaire (MACQ; Simons et al., 2012). The MACQ is a 48-item questionnaire assessing a variety of negative consequences (e.g., “While using marijuana I have said or done embarrassing things,” “Because of my marijuana use, I have not eaten properly,” “I have driven a car when I was high”). Participants were asked if they had experienced each consequence (0 = no, 1 = yes) in the past year. All phrases that used the term marijuana were replaced with cannabis. We created a consequence count across subscales (ω = .92).

Cannabis Dependence.

Cannabis dependence was assessed with the five-item Severity of Dependence Scale for Cannabis (SDS; Martin et al., 2006). Each item has a scale of 0 to 4. For ease of interpretation, all scores were added to create a cannabis dependence severity sum (ω = .78).

Data Analysis.

Measurement Model.

A three-factor CFA model was fit to data from Sample 4. FIML was used to handle missing data.

Measurement Invariance.

After the final factor structure was obtained, multiple group CFA models tested for measurement invariance across cannabis use frequency, participant sex, and race/ethnicity in Sample 5.

Measurement invariance was tested in four steps. First, a model investigating the pattern of loadings (configural invariance) was fit across groups. The highest loading in each group was set to 1.0 and factor means within groups were set to 0. If this model had adequate fit and all items loaded significantly onto their retrospective factor, configural invariance was achieved. Next, this model was compared to a model that constrained all factor loadings to be equivalent across groups (metric variance), which was then compared to a model that constrained all factor loadings and intercepts to be equal, while allowing factor means to freely vary across groups (scalar invariance). Lastly, the scalar model was compared to a model that constrained all factor loadings, intercepts, and variances to be equal across groups (strict invariance). Change in model fit was evaluated using criteria outlined by Chen (2007), which suggests metric invariance is present if the decrement in model fit from configural to metric does not exceed SRMR ≥ .030, RMSEA ≥ .015, or CFI ≥ −.01. Scalar and strict invariance is demonstrated if there is not a change in CFI ≥ −.01 that is accompanied by a change in SRMR ≥ .010 or RMSEA ≥ .015. MLR estimation was used to address nonnormality in the data. Missing data related to racial/ethnic and sex invariance (grouping variable) were casewise deleted and there were no missing data on cannabis use frequency.

Sample 5 Results: CFA and Measurement Invariance

Measurement Model.

The 17-item structure (see Table 1) fit the data well (RMSEA = .062, CFI = .939, TLI = .929, SRMR = .058), and all items loaded significantly onto their respective factor.

Table 1.

Confirmatory Factor Structure of the AECS Subscales

	Sample 4			Sample 5
Subscale	Beta	SE	β	Beta	SE	β
High arousal negative	ω = .862			ω = .912
Paranoid	1.00		.81	1.00		.87
Panicked	.76	.05	.77	.91	.03	.90
Fearful	.78	.05	.75	.78	.04	.80
Anxious	.89	.05	.79	.93	.03	.88
Suspicious	.76	.06	.66	.79	.05	.62
Like my heart is racing	.57	.07	.47	.67	.04	.74
Low arousal negative	ω = .845			ω = .873
Sluggish	1.00		.84	1.00		.90
Drowsy	.59	.06	.52	.68	.04	.68
Out of it	.95	.05	.77	.93	.04	.81
Slow	.95	.06	.77	.82	.05	.73
Lazy	.90	.06	.71	.70	.05	.67
General positive	ω = .764			ω = .812
Excited	1.00		.65	1.00		.64
Heightened senses	.72	.09	.50	.81	.10	.52
Sociable	.78	.08	.55	.89	.07	.57
Creative	.83	.09	.65	1.22	.13	.79
Wise	.97	.10	.64	1.12	.11	.71
In tune with nature	.86	.12	.55	1.18	.14	.66

Note. AECS = Anticipated Effects of Cannabis Scale; Beta = unstandardized beta; β = standardized beta; ω = McDonald’s omega.

Measurement Invariance.

Measurement invariance (see Table 3) first was assessed across cannabis use frequency. Cannabis frequency was operationalized as monthly or less (n = 230) versus weekly or more (n = 205).

Table 3.

Measurement Invariance in Validation Sample

	Gender (1 = female, 2 = male)				Use level (1 = monthly, 2 = weekly or more)				Race/ethnicity (1 = minority, 2 = non-Hispanic white)
Variable	CFI	TLI	RMSEA	SRMR	CFI	TLI	RMSEA	SRMR	CFI	TLI	RMSEA	SRMR
Configural invariance	.933	.921	.066	.065	.937	.927	.062	.062	.936	.925	.063	.066
Metric invariance
Overall model fit	.933	.925	.064	.069	.935	.929	.061	.068	.938	.931	.06	.068
Change from configural	.000	.004	−.002	.004	−.002	.002	−.001	.005	.002	.006	−.003	.002
Scalar invariance
Overall model fit	.924	.92	.066	.070	.924	.921	.064	.072	.934	.931	.06	.069
Change from metric	−.009	−.005	.002	.001	−.011	−.008	−.003	.004	.004	.001	.000	.001
Strict invariance
Overall model fit	.926	.928	.063	.072	.917	.918	.066	.077	.937	.931	.06	.069
Change from scalar	.002	.008	−.003	.002	−.007	−.003	.002	.005	.003	.000	.000	.000

Note. CFI = comparative fit index; TLI = Tucker-Lewis index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual.

To establish configural invariance, a two-group CFA model was estimated. The highest factor loading across each factor was set to 1.0 and the factor means within groups were set to zero. The resulting model provided adequate fit to the data, and all items loaded significantly onto their respective factors (see Table 3). To establish metric invariance, the factor loadings were constrained to be equal across groups. This model did not show a significant decrement in model fit compared to the configural model. Next, to establish scalar invariance, we constrained item intercepts to be equal across group, while allowing the factor means to freely vary. This model did not show a significant decrement in fit compared to the metric invariance model. Lastly, to establish strict invariance, we added constraints on item variances while allowing factor means to freely vary. This model did not show a significant decrement in fit compared to the scalar invariant model (see Table 3 for all model parameters). Thus, the proposed interpretation of the factor structure was invariant across cannabis use frequency (monthly or less vs. weekly or more).

The model was also tested for measurement invariance across participant sex and race/ethnicity. For simplicity and a low N for some racial/ethnic subgroups, we created a dummy variable where non-Hispanic White participants (N = 255) were compared to ethnic/racial minority participants (N = 162). The proposed interpretation of the AECS factor structure evidenced strict invariance across both biological sex and race/ethnicity (see Table 3).

Tests of measurement variance in Sample 4 across cannabis use frequency (multiple times a day vs. not; daily vs. nondaily) biological sex, and race/ethnicity also showed evidence of strict measurement invariance (see Supplemental Table S5 in the online supplemental material).

Measurement Validation Phase: Initial Interpretation of AECS Test Scores

Sample 5: Validation

All analyses examining validity of the proposed interpretation of test scores were conducted in Sample 5.

Validation Analyses

Before conducting analyses, all dependent variable distributions were analyzed for outliers and winsorized by replacing values outside of 3 SDs from the mean with the next highest value within 3 SD (Tabachnick & Fidell, 2001). Less than 1% of cases on three cannabis outcomes (2 cases for quantity, 5 cases for consequences, 7 cases for dependence) were winsorized. All models used MLR estimation to address any remaining nonnormality in the data and FIML was used to handle missing data in all validation models.

Bivariate correlations between the AECS, B-MEEQ, and CEQ subscales were examined to assess evidence for convergent and discriminant validity of the proposed interpretation of AECS test scores. Correlations were considered small (r = .1–.3), moderate (r = .3–.5) or large (r = .5+). We also tested discriminant validity of the proposed interpretation of AECS test scores, the B-MEEQ, and the CEQ positive subscales using CFA. We did this by comparing a model where the AECS and B-MEEQ were distinct factors to a model where both loaded onto a single factor. The same procedure was repeated for the CEQ. Chi-square difference testing (correcting for the use of the MLR scaling correction factor; Satorra & Bentler, 2001) was used to identify any significant differences. A significant decrement in model fit would signify that the two constructs (i.e., AECS positive vs. B-MEEQ positive, AECS positive vs. CEQ positive) operate significantly better as separate factors rather than a single “positive expectancies” construct. This procedure was not conducted for negative subscales, as the AECS clearly differentiates negative expectancies based on arousal whereas the CEQ and B-MEEQ do not.

To assess evidence of concurrent validity of the proposed interpretation of AECS test scores, cannabis use frequency, quantity, negative consequences, and dependence symptoms (separate analyses) were regressed on each AECS subscale, covarying age and biological sex. Because of the count nature of three of the outcomes (i.e., quantity, consequences, dependence), overdispersion was investigated using the AER package in R (Kleiber et al., 2020), which tests for overdispersion using the Cameron and Trivedi (1990) method. Cannabis quantity (z = 1.75, p = .04), cannabis consequences (z = 11.3, p < .001) and cannabis dependence (z = 7.47, p < .001) all showed significant overdispersion, so a negative binomial distribution was used for these outcomes.

Incremental validity of the proposed interpretation of AECS test scores was examined using a similar series of regression models, but with the addition of established expectancy measures as independent variables. Incremental validity of the proposed interpretation of AECS test scores was assessed above and beyond the B-MEEQ (results comparing AECS and the CEQ are shown in Supplemental Table S7 in the online supplemental material). The B-MEEQ was entered into a regression with participant sex and age, and a subsequent model added the AECS as a predictor. Adjusted R² estimates only are reported for cannabis frequency since such estimates are not available with negative binomial distributions.

Sample 5 Results: Preliminary Validity

Convergent and Discriminant Validity

All AECS subscales significantly correlated with other measures assessing similar constructs (See Table 2). The AECS positive subscale correlated strongly with the CEQ and B-MEEQ positive subscales, r = .58–.72, p < .001. The AECS LOW− subscale correlated strongly with the CEQ and B-MEEQ negative subscales, r = .50–.54, p < .001. The AECS HIGH− subscale correlated strongly with the CEQ negative subscale, r = .66, p < .001 and moderately with the B-MEEQ negative subscale, r = .41, p < .001. In addition, the AECS positive subscale was inversely correlated with both the AECS LOW−, r = −.16, p = .001 and HIGH− subscales, r = −.29, p < .001, and also inversely correlated with the CEQ and B-MEEQ negative subscale, r = −.29, p < .001. The AECS LOW− and HIGH− subscales were inversely correlated with both the CEQ (r = −.33, −.29, p < .001, respectively) and B-MEEQ (r = −.17, −.25, p < .001, respectively) positive subscales.

Discriminant validity of the proposed interpretation of the AECS positive test score was tested within a CFA framework as described previously. The model specifying the B-MEEQ and AECS positive items as separate factors, χ²(26) = 150.09, p < .001, showed a significant improvement in model fit when compared with a model specifying items as a single factor, χ²(27) = 187.98, p < .001, Δχ²(1) = 22.25, p < .001. The model specifying the CEQ and AECS positive subscales as separate factors, χ²(251) = 1340.86, p < .001, also showed a significant improvement in model fit when compared with a model specifying the two as a single factor, χ²(252) = 1412.66, p < .001, Δχ²(1) = 35.07, p < .001.

Concurrent Validity

Table 3 shows findings from the model assessing evidence of concurrent validity of the proposed interpretation of AECS test scores. As demonstrated in Table 4, AECS subscales were associated with cannabis outcomes as theoretically expected. AECS positive expectancies were associated with more frequent cannabis use, greater cannabis use quantity, more consequences, and dependence severity. AECS LOW− expectancies were associated with less frequent cannabis use and more negative consequences but were not associated with cannabis quantity or dependence severity, above and beyond the other subscales. AECS HIGH− expectancies were associated with more cannabis consequences and dependence severity but were not associated with cannabis frequency or quantity.

Table 4.

Concurrent Validity of the AECS Subscales

Subscale	ΔR2	Beta	SE	β	Sig.
Cannabis frequency
Sex	.05 (Step 1)	−1.33**	.34	−.19	<.001
Age		−.13	.13	−.06	.32
Race		.63	.33	.09	.055
AECS Pos	.18 (Step 2)	.45**	.08	.28	<.001
AECS Lo Neg		−.18*	.08	−.12	.031
AECS Hi Neg		−.12	.07	−.08	.08
Cannabis quantity
Sex		−.27**	.07	−.70	<.001
Age		−.02	.02	−.13	.51
Race		.04	.07	.10	.59
AECS Pos		.05**	.02	.50	.007
AECS Lo Neg		−.01	.02	−.17	.48
AECS Hi Neg		−.02	.02	−.20	.32
Cannabis negative consequences
Sex		−.21 **	.07**	−.44	.002
Age		.01	.02	.01	.06
Race		.11	.07	.23	.14
AECS Pos		.06**	.02**	.49	.002
AECS Lo Neg		.05*	.02*	.46	.012
AECS Hi Neg		.06**	.02**	.56	.001
Cannabis dependence symptoms
Sex		−.42**	.13	−.62	.001
Age		.02	.04	.08	.70
Race		.12	.13	.18	.36
AECS Pos		.08*	.03	.47	.016
AECS Lo Neg		.01	.03	.09	.69
AECS Hi Neg		.09**	.03	.63	.002

Note. AECS = Anticipated Effects of Cannabis Scale; Pos = positive subscale; Lo Neg = low arousal negative subscale; Hi Neg = high arousal negative subscale; Race (0 = racial/ethnic minority, 1 = non-Hispanic white). N = 435. Cannabis frequency used a normal distribution, cannabis quantity, negative consequences, and dependence symptoms used a negative binomial distribution. Because count models do not provide R² estimates, we only provide R² for frequency. Beta = unstandardized beta; β = standardized beta.

^*p < .05.

^**p < .01.

Incremental Validity

Table 5 shows findings from the model assessing evidence of incremental validity of the proposed interpretation of AECS test scores. Because the B-MEEQ is the only other brief measure of cannabis expectancies, we focused our incremental analyses on this measure. When the B-MEEQ and the AECS were entered in the same model, AECS positive expectancies were associated with more frequent use above and beyond the B-MEEQ. In addition, AECS HIGH− expectancies were associated with more negative consequences and dependence symptoms above and beyond the B-MEEQ. Notably, although the AECS predicted unique variance above and beyond the B-MEEQ, the B-MEEQ subscales also predicted unique variance above and beyond the AECS. Incremental validity of the proposed interpretation of AECS test scores relative to the CEQ is shown in Supplemental Table S6 in the online supplemental material. The AECS subscales were not associated with cannabis use measures above and beyond the CEQ, which was expected given the greater length of the CEQ.

Table 5.

Incremental Validity of the AECS Subscales Relative to the B-MEEQ

	Cannabis frequency					Cannabis quantity				Cannabis negative consequences				Cannabis dependence symptoms
Variable	ΔR2	Beta	SE	β	Sig.	Beta	SE	β	Sig.	Beta	SE	β	Sig.	Beta	SE	β	Sig.
Sex	.19 (Step 1)	−1.20**	.31	−.17	<.001	−.28**	.07	−.69	<.001	−.21**	.07	−.35	.004	−.42**	.13	−.52	.001
Age		−.14	.12	−.06	.23	−.02	.02	−.17	.36	−.01	.02	−.05	.68	.01	.04	.01	.97
Race		.84*	.30	.12	.005	.04	.07	.09	.63	.07	.07	.12	.33	.09	.13	.12	.48
B-MEEQ Pos		1.27**	.18	.27	<.001	.18**	.05	.63	<.001	.26**	.05	.63	<.001	.34**	.10	.61	<.001
B-MEEQ Neg		−1.33**	.24	−.24	<.001	−.06	.06	−.20	.32	.40**	.06	.81	<.001	.45**	.10	.70	<.001
B-MEEQ Pos	.21 (Step 2)	.86**	.24	.18	<.001	.14**	.06	.49	.012	.27**	.06	.59	<.001	.32*	.12	.51	.011
B-MEEQ Neg		−.96**	.28	−.17	<.001	−.01	.07	−.03	.88	.30**	.07	.54	<.001	.42**	.13	.56	.002
AECS Pos		.24*	.09	.15	.008	.02	.02	.18	.41	.02	.02	.13	.34	.04	.04	.18	.34
AECS Lo Neg		−.13	.08	−.09	.12	−.02	.02	−.19	.40	.01	.02	.11	.45	−.03	.04	−.16	.39
AECS Hi Neg		.01	.07	.01	.92	−.01	.02	−.09	.65	.06**	.02	.42	.001	.08**	.03	.44	.009

Note. AECS = Anticipated Effects of Cannabis Scale; B-MEEQ = Brief-Marijuana Effect Expectancy Questionnaire; Pos = positive subscale; Neg = negative subscale; Lo Neg = low arousal negative subscale; Hi Neg = high arousal negative subscale; Race (0 = Racial/Ethnic Minority, 1 = Non-Hispanic White). N = 435. Cannabis frequency used a normal distribution, cannabis quantity, negative consequences, and dependence symptoms used a negative binomial distribution. Because count models do not provide R² estimates, we only provide R² for frequency. Beta = unstandardized beta; β = STDXY standardized beta.

^*p < .05.

^**p < .01.

Discussion

The current study developed and provided evidence of preliminary validation of the proposed interpretation of test scores for a novel measure of cannabis expectancies (AECS), with items spanning the full affective space across valence (positive vs. negative) and arousal (sedating vs. stimulating). The AECS used a comprehensive and up-to-date list of cannabis effects, samples of light to heavy cannabis users, single words/short phrases, and rigorous psychometric analyses. Study findings suggested that negative expectancies diverged by level of arousal, with seven items representing high arousal (e.g., anxious, paranoid) and five items representing low arousal (e.g., slow, lazy) negative effects. In contrast, positive expectancies (e.g., sociable, excited, wise, in tune with nature) did not break down by level of arousal. The proposed interpretation of test structure was fully invariant across frequency of cannabis use, sex, and race/ethnicity and was associated with variability in cannabis outcomes above and beyond the only other brief cannabis expectancy measure. Integration of findings with prior research and discussion of implications and future directions is provided below.

The current study’s focus on arousal, particularly in relation to negative expectancies, sets it apart from past cannabis expectancy measures. Although the CEQ added high arousal items not included in the MEEQ, it includes a single negative expectancy factor that does not differentiate low from high arousal items. Both exploratory and confirmatory analyses suggested separate negative factors based upon arousal, each associated with variance in cannabis use/problems above and beyond the other. More specifically, low arousal negative expectancies were associated with less frequent cannabis use yet more cannabis negative consequences, and high arousal negative expectancies were associated with both more cannabis consequences and dependence symptoms. The arousal-based negative expectancy findings may explain the generally poor reliability of the negative expectancy subscale of the MEEQ. In addition, they may help reconcile conflicting findings from past studies using different scales (e.g., Aarons et al., 2001; Connor et al., 2011; Kristjansson et al., 2012). For example, although samples differed in previous use and age, Aarons et al. (2001) found that negative expectancies in adolescents (using the MEEQ) were protective against cannabis use, whereas Connor et al. (2011) found that negative expectancies in a treatment sample (using the CEQ) were associated with cannabis dependence severity. Thus, MEEQ negative effects may be more indicative of the low arousal negative expectancies from the AECS (e.g., slow), whereas CEQ negative effects may be more indicative of the high arousal negative expectancies on the AECS (e.g., panicked, paranoid). Alternatively stated, MEEQ negative expectancies may be less intense/arousing negative effects, as they were protective in novice cannabis users, whereas CEQ negative expectancies may be more intense/arousing negative effects, as they were associated with problems/dependence in heavy cannabis users. Indeed, at the bivariate level, the HIGH− subscale was the stronger correlate of CEQ negative expectancies, whereas the LOW− was the stronger correlate of B-MEEQ negative expectancies. It is possible that lower arousal negative expectancies may deter one from using, whereas higher arousal negative expectancies may be associated with problems related to cannabis dependence symptoms for those who decide to use despite these expectancies. This is in line with alcohol research, suggesting that more persistent use may be associated with stronger negative expectancies (e.g., McMahon & Jones, 1993). Another explanation could be that someone who expects high arousal and intense effects, but still decides to use, has personality traits associated with problem use (e.g., impulsivity) and may be at risk for psychosis (Vadhan et al., 2017, 2019). Regardless, the present study suggests that differentiating negative expectancies based on arousal is important in accounting for differential cannabis outcomes.

In contrast to negative expectancies, positive expectancies did not differ by level of arousal. In both exploratory and confirmatory models, low arousal positive expectancies (e.g., relaxed, wise) did not load onto a separate factor from high arousal positive expectancies (e.g., sociable, excited). In fact, positive expectancies generally were rated as higher arousal compared to negative expectancies and showed less variation in arousal ratings compared with negative expectancies (see Figure 1). Thus, it is possible that cannabis users do not differentiate positive effects based upon arousal in the same way they do for negative effects. Most hypothesized “low arousal positive” effects seemed to be higher in arousal in our data. Arousal means for words such as “relaxed” and “calm,” which one would not expect to be even moderately arousing, were rated as above average on arousal (e.g., 5.3/9, 5.2/9, respectively). This is interesting as cannabis users rated anticipated negative low arousal effects such as “slow” and “sluggish” as low in arousal (3.7/9, 3.6/9, respectively). The lack of arousal differences for positive expectancies is important for prevention/intervention efforts geared toward challenging unrealistic positive expectancies. Although no cannabis expectancy intervention challenge has been tested to date, past experimental research showed that expectancy manipulations during cannabis administration (i.e., being told you’re receiving cannabis) led to perceived changes in subjective intoxication and other subjective and behavioral effects (Metrik et al., 2009, 2012). These studies suggest that subjective experiences may be due, at least in part, to expectancies. In contrast to the alcohol literature, which suggests that challenging high arousal positive expectancies may drive reductions in drinking (e.g., Morean et al., 2015), the present study suggests that challenging general positive expectancies, rather than high arousal positive expectancies, may be most relevant to target/challenge for cannabis users.

In addition to demonstrating a novel three-factor structure based on valence and arousal, the proposed interpretation of AECS test scores provided evidence for convergent, discriminant, and concurrent validity. The proposed three-factor interpretation of AECS test scores also showed evidence for incremental validity above and beyond the B-MEEQ, such that the AECS positive subscale was associated with frequency of cannabis use, and the AECS high arousal negative subscale was associated with negative cannabis consequences and dependence symptoms. In addition, standardized estimates of AECS positive and B-MEEQ positive subscales predicting frequency of use (b = .27 and b = .28) were nearly identical, and AECS high arousal negative and B-MEEQ negative estimates predicting dependence symptoms were similar (AECS b = .63, MEEQ b = .70). This suggests that the AECS could act as a stand-alone expectancy questionnaire, as it is associated with cannabis use/problems similarly to and incrementally above the only other brief measure. At the same time, the B-MEEQ was associated with unique variance above and beyond the AECS. However, the internal consistency of the negative B-MEEQ subscale has been consistently low in past research, as it was in our data (e.g., Torrealday et al., 2008). In addition, the B-MEEQ uses quadruple-barreled items (e.g., “marijuana makes it harder to think and do things (harder to concentrate or understand); slows you down when you move”), making it difficult to know which aspect of the statement participants expect versus do not expect. Given these limitations, the AECS provides an alternative and psychometrically sound way to assess expectancies briefly. It is worth noting, though, that the AECS did not predict variance above and beyond the CEQ. This was expected, as the CEQ is three times the length of the AECS and asks for much more detailed information. Therefore, in studies well-suited to more time-consuming measures, the CEQ should be considered as well.

Although the current study used rigorous methods to make determinations about item content, we were unable to retain items with strong content validity. First, “pain relief” did not make it past the EFA stage. Pain relief is a popular and known cannabis effect for individuals using medical marijuana, and “pain relief” is one of the highest loading positive expectancy items on the MCEQ (Morean & Butler, 2019). Because the majority of participants in our study were young adults, who may not be using for medicinal, pain reducing purposes, item endorsement was low. In a medical cannabis using population, this item may have functioned differently. In addition to pain relief, both “relaxed” and “calm,” did not fit well into any of the EFAs or CFAs tested across samples. At the EFA stage, each had large cross-loadings with the negative factors, and at the CFA stage, these items had low loadings on a single factor and significant cross-loading (i.e., high MI values). This could be due to limited variability in endorsement. For example, on the 0–10 scale, over 90% of participants scored above the midpoint of “relaxed.” In addition, at the CFA stage, “relaxed” loaded significantly onto both the low arousal (positively) and high arousal negative (negatively) subscales, while also not loading onto the positive subscale. Therefore, this item could not be retained. Notably, “relaxed” also was dropped in item development for the CEQ (Connor et al., 2011). Considering that neither the CEQ or AECS were able to retain “relaxed,” it may be the case that past questionnaires were able to differentiate this item based on largely noncannabis using samples. Future research should continue to investigate how negative reinforcement (i.e., pain relief) and relaxation/tension reduction fit into expectancy and subjective response measures for cannabis.

Despite the novel measurement properties of the AECS, and evidence of reliability and validity from the proposed interpretation of AECS test scores, the findings must be interpreted in light of limitations. All data were self-reported. Self-reported substance use generally is agreed to be reliable and valid (e.g., Vitale et al., 2006), although it is subject to potential bias in reporting. Relatedly, another limitation was our measure of cannabis quantity. Cannabis quantity is difficult to assess (Gray et al., 2009; Hindocha et al., 2018), particularly given the proliferation of cannabis products (e.g., edibles, vapes), and there is, as yet, no standardized means of assessing it (Hindocha et al., 2018; Lorenzetti et al., 2016). In addition, cannabis outcome expectancies (i.e., anticipated effects or consequences from using cannabis) may not fully correspond to the subjective effects or behavioral consequences experienced following cannabis administration. Therefore, future research using placebo-cannabis administration that experimentally controls for expectancy effects (i.e., the effect of believing that one is using cannabis rather than placebo) is needed to determine if the factor structure identified in the current study fits reports of subjective experiences following cannabis administration (e.g., Metrik et al., 2009). The current study also relied on cross-sectional data collected online. Although we had attention checks to identify participants who were answering quickly and carelessly, it is possible that participants were not completely honest in their responses. Further longitudinal replication of relations to cannabis use outcomes is needed.

The generalizability of the findings also is limited to cannabis users, and findings might not generalize to cannabis-naïve populations. Other studies of cannabis expectancies, such as those using the B-MEEQ/MEEQ, have included both nonusers and users. However, a limitation of those studies is that cannabis use overall was relatively infrequent. On the contrary, we focused on cannabis users to ensure participants had more experience with cannabis. Nonetheless, much of the alcohol literature suggests that alcohol expectancies may be formed before use initiates (e.g., Miller et al., 1990), so it may be important to capture information on expectancies prior to onset of use. Future research should confirm the factor structure of the AECS in a sample that includes nonusers. Further, we did not examine whether the factor structure of the AECS differed across cannabis product types (e.g., flower, concentrates, edibles) or methods of cannabis administration (e.g., smoke, vape, edible). Some research has suggested that product type or method of administration might impact cannabis effects (Cavazos-Rehg et al., 2018; Morean & Butler, 2019; Spindle et al., 2018). Therefore, future research might consider testing measurement invariance of the AECS across type of cannabis and method of administration.

Although the current study has limitations, there also are important strengths to consider. First, we used a very rigorous statistical design that included an exploratory analysis stage, a confirmatory analysis stage, and examination of measurement invariance to establish a strong factor structure across cannabis frequency, sex, and race/ethnicity. Although these methods entailed narrowing items substantially, the purpose was to create a short, easily administered measure. The initial development of our measure is an important starting point that suggests the potential of considering both valence and arousal in future cannabis expectancy research. Future research is needed to evaluate the factor structure of the AECS across clinical and nonusing samples to ensure it is suitable to use for all populations.

In sum, the present study developed and provided initial evidence for validity of the proposed interpretation of AECS test scores, using innovative measurement and multiple, heterogeneous samples. The AECS brought together a comprehensive list of short-phrased items and obtained the most up-to-date relatedness, valence, and arousal norms on cannabis, which is much less stigmatized today than when other questionnaires were developed. The new measure also was constructed in a way that lends itself to direct comparisons with in-the-moment cannabis subjective response items, which currently are being assessed via cannabis administration. Research suggests significant discrepancies between alcohol expectancies and subjective response (e.g., Morean et al., 2015), and these discrepancies have direct treatment implications (i.e., the value of challenging distorted expectancies as a component of efficacious interventions). In fact, one of the primary purposes of the current study was to create a measure that could test such discrepancies for cannabis use to inform prevention and intervention efforts. Therefore, the AECS may ultimately facilitate efforts to improve cannabis treatment efficacy, which is critical given the rising rates of cannabis use. Taken together, this new measure offers a novel way to assess cannabis expectancies, demonstrates the ability to account for variance in cannabis outcomes similarly and incrementally relative to other short measures, and provides a foundation upon which to build a parallel and directly comparable measure of subjective response.

Public Significance Statement.

The current study outlines the development and preliminary validation of test scores for a novel cannabis expectancies measure, suggesting a three-factor structure covering valence and arousal (i.e., high arousal negative, low arousal negative, general positive). Differentiating negative expectancies by level of arousal may have important implications for cannabis use motivation and interventions designed to address these motivations.

Appendix. The Anticipated Effects of Cannabis Scale

Instructions:

The following is a list of experiences that people may have when using cannabis. Please note, we want to know what experiences you expect when using cannabis, not what you are currently experiencing.

On a scale of 1–10 please rate the extent to which you expect to feel each of the following effects after using cannabis.

Effect	Not at all		Moderately							Extremely
1. Creative	0	1	2	3	4	5	6	7	8	9	10
2. Heightened senses	0	1	2	3	4	5	6	7	8	9	10
3. Drowsy	0	1	2	3	4	5	6	7	8	9	10
4. Like my heart is racing	0	1	2	3	4	5	6	7	8	9	10
5. Out of it	0	1	2	3	4	5	6	7	8	9	10
6. Sluggish	0	1	2	3	4	5	6	7	8	9	10
7. Panicked	0	1	2	3	4	5	6	7	8	9	10
8. Sociable	0	1	2	3	4	5	6	7	8	9	10
9. Fearful	0	1	2	3	4	5	6	7	8	9	10
10. Slow	0	1	2	3	4	5	6	7	8	9	10
11. Excited	0	1	2	3	4	5	6	7	8	9	10
12. Paranoid	0	1	2	3	4	5	6	7	8	9	10
13. Lazy	0	1	2	3	4	5	6	7	8	9	10
14. Wise	0	1	2	3	4	5	6	7	8	9	10
15. In tune with nature	0	1	2	3	4	5	6	7	8	9	10
16. Anxious	0	1	2	3	4	5	6	7	8	9	10
17. Suspicious	0	1	2	3	4	5	6	7	8	9	10