Skip to main content
NCBI home page
Journal of Studies on Alcohol and Drugs logoLink to Journal of Studies on Alcohol and Drugs
. 2025 Aug 27;86(5):683–693. doi: 10.15288/jsad.24-00201

Adjusting for Measurement Bias in the MEEQ-B Across Sex and Race/Ethnicity in the ABCD Study

Tammy Chung a,,*, Shawn J Latendresse b, Nicole Kennelly a, Margret Z Powell b, Carolyn E Sartor a
PMCID: PMC12419513  PMID: 39630417

Abstract

Objective:

The primary aim of this study was to assess and adjust for measurement non-equivalence (bias) by sex, race/ethnicity, and co-occurring social identities (Sex × Race/Ethnicity) for the Marijuana Effect Expectancies Questionnaire–Brief (MEEQ-B) among Black, Latinx, and non-Latinx White youth. The second aim was to determine how group comparisons change after accounting for possible measurement bias.

Method:

Black, Latinx, and non-Latinx White youth from the Adolescent Brain and Cognitive Development Study Follow-up 3 (n = 8,982; mean age = 12.91, SD = 0.65; 47.28% female; 15.03% Black, 22.93% Latinx, 62.04% non-Latinx White) completed the MEEQ-B. Moderated nonlinear factor analysis (MNLFA) generated positive and negative expectancies factor scores accounting for non-equivalence. Analyses contrasted group differences by sex, race/ethnicity, and these co-occurring social identities using original (unadjusted) versus MNLFA-generated scores adjusted for measurement non-equivalence.

Results:

Measurement non-equivalence was observed for positive and negative expectancies across sex, race/ethnicity, and their co-occurring social identities. MNLFA revealed between-group differences at the factor and item level. Further, comparisons of original (unadjusted) and MNLFA-generated adjusted scores revealed that unadjusted scores underestimated or did not detect some group differences in positive expectancies identified using adjusted scores, and unadjusted scores underestimated how much lower negative expectancies were in Black and Latinx relative to non-Latinx White youth.

Conclusions:

Results highlight the need for caution when interpreting scores of a measure like the MEEQ-B that has not undergone measurement equivalence testing and demonstrate how failing to adjust for non-equivalence can result in biased estimates of positive and negative expectancies, particularly when used with diverse populations.

Cannabis outcome expectancies, or an individual's beliefs about the anticipated effects of cannabis use, robustly predict experimentation and escalation of cannabis use (Aarons et al., 2001; Montes et al., 2019). As a core precursor to cannabis use, cannabis expectancies develop either indirectly (for example, through observation of cannabis's effects) or through direct experience (Goldman et al., 1999). Given their role in the onset and progression of cannabis use, accurate measurement of cannabis expectancies across sex and racial/ethnic groups is crucial.

Adolescents' cannabis expectancies

Cannabis outcome expectancies measures, such as the Marijuana Effect Expectancies Questionnaire (MEEQ; Schafer & Brown, 1991), have generally identified two main factors: positive and negative effects (Torrealday et al., 2008). Positive cannabis expectancies include anticipated effects such as relaxation and tension reduction, social facilitation, and enhanced perception of certain sensory features (Torrealday et al., 2008). Negative cannabis expectancies include effects such as cognitive or behavioral impairment, negative physical effects on the body, and craving (Torrealday et al., 2008).

In support of expectancy theory (Goldman et al., 1999), adolescents' cannabis expectancies predict the onset and escalation of use (Montes et al., 2019). Specifically, greater positive cannabis expectancies have been associated with earlier onset of cannabis use (Montes et al., 2019) and more frequent cannabis use (Montes et al., 2019; Schafer & Brown, 1991). In comparison, greater negative cannabis expectancies have been associated with non-use (Aarons et al., 2001; Schafer & Brown, 1991) and lower levels of cannabis use (Aarons et al., 2001; Kristjansson et al., 2012; Torrealday et al., 2008).

Differences by sex and race/ethnicity in adolescents' cannabis expectancies

We are unaware of any investigations of sex differences in adolescents' cannabis expectancies and know of only limited research on differences by race/ethnicity. One of the few investigations by race/ethnicity was a cross-sectional study of 12- and 13-year-olds, which found that Latinx, compared with non-Latinx White, youth had higher positive cannabis expectancies and that non-Latinx White, compared with Black and Latinx, youth had higher negative expectancies (Shih et al., 2010). In contrast, a longitudinal study of adolescent girls (ages 12–17) found that Black young females reported higher negative expectancies than non-Latinx White females, but differences in positive expectancies between Black and non-Latinx White females were limited to higher expectancies at age 13 in Black females (Foster et al., 2018). Importantly, these studies did not use measurement equivalence methods before comparing group means to minimize possible measurement bias. Comparisons made in the context of measurement non-equivalence could inaccurately suggest apparent differences that reflect possible bias in measurement.

Measurement equivalence to examine co-occurring social identities of sex and race/ethnicity

Measurement equivalence (also known as measurement invariance) methods provide a way to compare constructs at the level of both the latent variable (factor mean and variance) and individual items (item intercepts and factor loadings) across social identities such as sex and race/ethnicity. The limited research on measurement non-equivalence for cannabis expectancies measures includes one study of young adults, which indicated measurement equivalence across sex, race/ethnicity, and frequency of cannabis use for the Anticipated Effects of Cannabis Scale (Waddell et al., 2021). Of note, equivalence regarding sex and race/ethnicity considered together (i.e., co-occurring social identities) was not examined in that study. Examining these social identities simultaneously while increasing analytic complexity could more effectively reduce measurement bias (Bauer et al., 2020). Thus, in this study, we tested measurement non-equivalence using moderated nonlinear factor analysis (MNLFA), which was used to derive factor scores that account for measurement non-equivalence attributable to sex, race/ethnicity, and co-occurring social identities (Gottfredson et al., 2019).

Study aims

This study aimed to assess measurement equivalence by sex, race/ethnicity, and their co-occurring social identities in the Marijuana Effect Expectancies Questionnaire–Brief (MEEQ-B; Torrealday et al., 2008) among Black, Latinx, and non-Latinx White youth and to generate scores adjusted for measurement non-equivalence, using a national sample of youth. To identify potentially inaccurate conclusions regarding differences by sex, race/ethnicity, and their co-occurring social identities (Sex × Race/Ethnicity) in cannabis expectancies as a function of measurement bias, we contrasted mean group differences using original (unadjusted) scores and scores adjusted for measurement non-equivalence. The study addressed two research questions: (1) What is the nature of possible biases in measurement by sex, race/ethnicity, and their co-occurring social identities and (2) how do results of group comparisons change after accounting for possible measurement bias associated with these identities?

Method

Sample and procedures

The Adolescent Brain Cognitive Development (ABCD) Study is an ongoing multisite longitudinal study of adolescent health and cognitive development in the United States [https://abcdstudy.org]. Study design and ascertainment are detailed in prior publications (e.g., Garavan et al., 2018). In brief, 21 sites recruited youth age 9 or 10 and their primary caregiver from 2016 to 2018. Enrollment data from the National Center for Education Statistics and data from the U.S. Census Bureau's American Community Survey were used to derive ascertainment targets, applying probability sampling to target schools within catchment areas. A centralized institutional review board approved all study protocols. Written informed consent from a caregiver and assent from the child were obtained at enrollment.

Surveys were administered in person with the child and caregiver, querying demographics (e.g., primary caregiver education level, race/ethnicity) and health information, including a wide range of substance use–related factors. Data for the current study were drawn from Follow-up Year 3, release 5 (Mage = 12.91, SD = 0.65, range: 11.40–14.75 years old). Given our interest in measurement equivalence across Black, Latinx, and non-Latinx White youth, we used data from these groups, categorized as follows by the ABCD Study, using caregiver-reported youth race (Black/African American, non-Latinx White) and ethnicity (Hispanic/Latinx). All youth identified as Latinx/Hispanic were categorized as Latinx. Thus, “Black” refers to non-Latinx Black individuals, and non-Latinx White refers to individuals who did not identify as Latinx/Hispanic and identified as White. Analyses focused on these three racial/ethnic groups (Black, Latinx, non-Latinx White), the largest represented in ABCD, because there were insufficient numbers representing other racial/ethnic groups for analysis. The caregiver reported on the youth's sex assigned at birth, which was available for the full sample.

At baseline, there were 10,377 Black, Latinx, and non-Latinx White youth, with 87% completing Follow-up 3 (n = 9,029). The ABCD Study website provides details on attrition over follow-up. Youth who reported having heard of any cannabis product were administered the MEEQ-B. Youth did not need to report cannabis use to be included in the analysis (1.07% reported any lifetime cannabis use at Follow-up 3). The analytic sample included 8,982 youth with no missing values on individual MEEQ-B items at Follow-up 3: 47.28% female, 52.72% male, 15.03% Black, 22.93% Latinx, and 62.04% non-Latinx White. Regarding total household income, 25.2% of caregivers reported less than U.S.$50,000; 27.1% reported $50,000–99,000; 40.2% reported more than $100,000; and income was missing for 7.5% of the sample. For caregiver's highest level of education, 4.5% reported less than a high school degree, 11.4% reported graduating high school, 28.8% reported some college, 29.1% graduated college, 26.0% had postgraduate education, and highest education was missing for 1.8% of the sample.

Measures

Marijuana Effect Expectancies Questionnaire–Brief. The MEEQ-B (Torrealday et al., 2008) consists of two scales: positive expectancies (three items; Table 1) and negative expectancies (three items; Table 2). Items are rated from 1 (disagree strongly) to 5 (agree strongly). See Supplemental Table 6 for MEEQ-B item means. (Supplemental material appears as an online-only addendum to this article on the journal's website.) MEEQ-B original scores for positive expectancies are the sum of the three positive expectancy items; for negative expectancies, they are the sum of the three negative expectancy items. The MEEQ-B has similar associations with cannabis use frequency and related problems as the full MEEQ (Brackenbury et al., 2016; Torrealday et al., 2008). Internal consistency reliability (omega, total; ωt) in the analytic sample for positive expectancies was as follows: total sample ωt = .84; Black male ωt = .83, Black female ωt = .83, Latinx male ωt = .86, Latinx female ωt = .86, non-Latinx White male ωt = .82, non-Latinx White female ωt = .84. Internal consistency reliability for negative expectancies was as follows: total sample ωt = .76; Black male ωt = .76, Black female ωt = .74, Latinx male ωt = .75, Latinx female ωt = .72, non-Latinx White male ωt = .76, non-Latinx White female ωt = .75.

Table 1.

Parameter estimates from the final moderated nonlinear factor analysis model: Positive expectancies scale

graphic file with name jsad.24-00201tbl1.jpg

Reference Parameter Sex Estimate (SE) Race/Ethnicity 1 Estimate (SE) Race/Ethnicity 2 Estimate (SE) Sex × Race/Ethnicity 1 Estimate (SE) Sex × Race/Ethnicity 2 Estimate (SE)
M -0.01 (0.02) 0.13 (0.03)*** 0.06 (0.04) -0.02 (0.03) -0.04 (0.04)
Variance -0.03 (0.03) -0.15 (0.05)** -0.05 (0.07) -0.00 (0.04) -0.14 (0.07)*
MEEQB-2. Marijuana helps a person relax and feel less tense (helps a person unwind and feel calm)
 Intercept – – – – –
 Loading – – – – –
MEEQB-3. Marijuana helps people get along better with others (talk more; feel more romantic)
 Intercept – -0.28 (0.07)*** -0.08 (0.10) – –
 Loading – -0.15 (0.09) -0.06 (0.15) – –
MEEQB-4. Marijuana makes people feel more creative and perceive things differently (music sounds different; things are more interesting)
 Intercept 0.19 (0.03)*** – – – –
 Loading – – – – –
Open in a new tab

Notes: Parameter coding: Sex (female = 1, male = -1), race/ethnicity 1 (Black = -1/3, Latinx = -1/3, non-Latinx White = 2/3), race/ethnicity 2 (Black = -1/2, Latinx = 1/2, non-Latinx White = 0), Sex × Race/Ethnicity 1 (Black female = -1/3, Black male = 1/3, Latinx female = -1/3, Latinx male = 1/3, non-Latinx White female = 2/3, non-Latinx White male = -2/3), Sex × Race/Ethnicity 2 (Black female = -1/2, Black male = 1/2, Latinx female = 1/2, Latinx male = -1/2, non-Latinx White female = 0, non-Latinx White male = 0). MEEQB = Marijuana Effect Expectancies Questionnaire–Brief.

*

p < .05;

**

p < .01;

***

p < .001.

Table 2.

Parameter estimates from the final moderated nonlinear factor analysis model: Negative expectancies scale

graphic file with name jsad.24-00201tbl2.jpg

Reference Parameter Sex Estimate (SE) Race/Ethnicity 1 Estimate (SE) Race/Ethnicity 2 Estimate (SE) Sex × Race/Ethnicity 1 Estimate (SE) Sex × Race/Ethnicity 2 Estimate (SE)
M -0.03 (0.01)* 0.31 (0.03)*** 0.12 (0.05)* – –
Variance -0.12 (0.02)*** -0.28 (0.05)*** -0.32 (0.08)*** – –
MEEQB-1. Marijuana makes it harder to think and do things.
 Intercept 0.05 (0.03) -0.01 (0.05) 0.18 (0.09)* -0.12 (0.05)* -0.02 (0.08)
 Loading – – – – –
MEEQB-5. Marijuana generally has bad effects on a person.
 Intercept – – – – –
 Loading – – – – –
MEEQB-6. Marijuana has bad effects on a person's body and gives people cravings.
 Intercept – -0.10 (0.07) -0.17 (0.11) – –
 Loading – -0.04 (0.09) -0.33 (0.14)* – –
Open in a new tab

Notes: Parameter coding: Sex (female = 1, male = -1), race/ethnicity 1 (Black = -1/3, Latinx = -1/3, non-Latinx White = 2/3), race/ethnicity 2 (Black = -1/2, Latinx = 1/2, non-Latinx White = 0), Sex × Race/Ethnicity 1 (Black female = -1/3, Black male = 1/3, Latinx female = -1/3, Latinx male = 1/3, non-Latinx White female = 2/3, non-Latinx White male = -2/3), Sex × Race/Ethnicity 2 (Black female = -1/2, Black male = 1/2, Latinx female = 1/2, Latinx male = -1/2, non-Latinx White female = 0, non-Latinx White male = 0). MEEQB = Marijuana Effect Expectancies Questionnaire–Brief.

*

p < .05;

***

p < .001.

Coding of sex and race/ethnicity. Sex, race/ethnicity, and sex by race/ethnicity classifications were included as categorical covariates. For use in MNLFA (see below), these variables were converted into a set of one or more orthogonal contrast-coded predictors using Helmert coding. Sex was coded as a single contrast (female = 1, male = -1). Race/ethnicity was recoded into two contrast codes: (1) non-Latinx White versus Latinx and Black youth (Black = -1/3, Latinx = -1/3, non-Latinx White = 2/3) and (2) Latinx versus Black youth (Black = -1/2, Latinx = 1/2, non-Latinx White = 0). Two contrast codes reflecting the products of the sex code with each of the two race/ethnicity codes operationalized comparisons for the co-occurring identities: Sex × Race/Ethnicity 1 (Black female = -1/3, Black male = 1/3, Latinx female = -1/3, Latinx male = 1/3, non-Latinx White female = 2/3, non-Latinx White male = -2/3), and Sex × Race/Ethnicity 2 (Black female = -1/2, Black male = 1/2, Latinx female = 1/2, Latinx male = -1/2, non-Latinx White female = 0, non-Latinx White male = 0).

Analytic plan

Use of sample weights and missing data. In keeping with standard practice, weighting variables were not used in measurement equivalence analyses (multiple-group confirmatory factor analysis [CFA]) and MNLFA (Research Question 1) because these analyses aimed to identify bias in response patterns in the sample. Sample weights were applied in analyses involving between-group comparisons (Research Question 2), consistent with the aim of accounting for ascertainment strategy, i.e., generalization to the larger population (Heeringa & Berglund, 2020). There were no missing data on MEEQ-B items after excluding youth who did not complete the MEEQ-B at Follow-up 3.

Research Question 1: Testing and adjusting for possible biases in measurement

Measurement equivalence to test for possible bias. Measurement equivalence analyses were conducted separately for MEEQ-B positive and negative expectancies because they represent two separate scales. The aim of the measurement equivalence analyses was to first test for possible bias in measurement and then adjust for potential bias in the MEEQ-B as the measure is used, that is, as two separate scales. The presence and extent of measurement equivalence of MEEQ-B scales across six discrete identities (Black females, Black males, Latinx females, Latinx males, non-Latinx White females, and non-Latinx White males) was assessed by comparing three standard, incrementally constrained models of measurement equivalence using multiple-group CFA in Mplus version 8.11 (Muthén & Muthén, 2017). Following prior recommendations, model testing started with the least and progressed to the most stringent model (Widaman & Reise, 1997). Specifically, configural invariance tested equivalent model structure (i.e., dimensions and indicators), metric invariance tested the equivalence of factor loadings, and scalar invariance tested the equivalence of item intercepts.

Criteria used to determine model fit. Individual models were evaluated as having adequate fit using the following criteria: ratio of χ2 to degrees of freedom ≤ 2 (Wheaton et al., 1977), root mean square error of approximation (RMSEA) < .06 (Bentler, 1990; Brown, 2015; Browne & Cudeck, 1993; Hu & Bentler, 1999), and the comparative fit index (CFI) > .95 (Bentler, 1990; Hu & Bentler, 1999). Likewise, nested comparisons were evaluated using scaled chi-square difference tests (Satorra & Bentler, 2001), where the adequacy of the more constrained model is reflected by a nonsignificant change in chi-square. However, because this chi-square difference test is biased toward large samples, a pair of alternative tests, based on RMSEA and CFI, have been recommended (Brown, 2015; Kline, 2011). Regarding the former, a more constrained model can be adopted if overlap is seen in the 90% confidence intervals for RMSEA, where non-overlapping confidence intervals suggest significant decrements in model fit (Wang & Russell, 2005). For the latter, a difference in CFI values of .01 or less indicates a nonsignificant decrement in model fit, such that the more constrained model can be adopted (Cheung & Rensvold, 2002).

Adjusting for measurement non-equivalence using MNLFA. Given evidence of inadequate cross-group equivalence using multiple-group CFA, we adjusted for measurement non-equivalence with separate analyses of the positive and negative expectancies scales using MNLFA (Bauer, 2017). MNLFA was conducted in R (Version 4.2.2) and Mplus (Version 8.11) using the semi-automated MNLFA R package developed for scale scoring (aMNLFA, Version 1.1.2; Cole et al., 2021; Gottfredson et al., 2019). The package generated a series of Mplus input files collectively used to determine the extent to which the covariates of sex, race/ethnicity, and the Sex × Race/Ethnicity interactions moderated parameters defining the structural (i.e., factor mean and variance) and/or measurement (i.e., item intercepts and factor loadings) models for the two MEEQ-B scales.

MNLFA involved three stages: testing an initial model, simultaneous testing of covariates in a conditional model, and generating a final model, separately for positive and negative expectancies scales. Initial MNLFA models separately assessed the impact of each covariate (e.g., sex, race/ethnicity) on the mean and variance of the latent positive and negative expectancies factors and individual item intercepts and factor loadings while holding all other intercepts and loadings invariant. For subsequent models, all sets of orthogonal or dependent contrast-coded covariates remained linked; that is, the retention of any covariate due to significance resulted in the retention of the other(s). All significant (and linked) effects of sex, race/ethnicity, and Sex × Race/Ethnicity interactions exceeding a threshold of p < .10 within initial mean and variance impact models or a significance thresh-old of p ≤ .05 for factor loadings and item intercepts in the item-specific measurement equivalence models were selected for simultaneous testing in a conditional model. Following convergence of the conditional model, all significant (p < .05) and linked effects of covariates on the mean and variance of the latent factor were selected for inclusion in a final model. Results from the initial models and the subsequent simultaneous models are in Supplemental Tables 2 and 3 for the positive expectancies scale and Supplemental Tables 4 and 5 for the negative expectancies scale.

From the conditional model (simultaneous testing of covariates), significant (and linked) covariate effects were subjected to sequential trimming via the adoption of a 5% false discovery rate (Benjamini & Hochberg, 1995), applied first to loadings and then to intercepts, to guard against inflation in type I error. This retained collection of parameters defined a final version of the model from which bias-adjusted factor scores for each scale were derived.

Research Question 2: Comparing original mean scores versus MNLFA-adjusted factor scores

MEEQ-B scale scores were derived using the original (unadjusted) item values, using the ABCD scoring procedure of taking the sum of the items for each scale. After applying the ABCD weighting variable to account for ascertainment strategy and accounting for clustering within families and sites, pairwise comparisons of mean scale scores were tested across sex, race/ethnicity, and co-occurring identities. Parallel analyses were conducted using the MNLFA-adjusted positive and negative expectancies factor scores. A 5% false discovery rate was applied for analyses in Research Question 2. Hedges' g effect sizes for the comparisons are interpreted as small = .2, medium = .5, and large = .8 (Hedges & Olkin, 1985).

Results

Research Question 1: Testing and adjusting for possible biases in measurement

Identifying measurement non-equivalence. The multiple-group CFAs suggested an inadequate degree of measurement equivalence based on model fit indices (see analysis plan for model fit criteria) for the two MEEQ-B scales (Supplemental Table 1), supporting the next step of adjusting for bias by conducting MNLFA.

Adjusting for bias in the positive expectancies scale using MNLFA. One of the MEEQ-B positive expectancies scale items (MEEQB-2: Cannabis helps a person relax and feel less tense [helps a person unwind and feel calm]) was fully equivalent with respect to sex, race/ethnicity, and Sex × Race/Ethnicity (Table 1). Significant differences in the contrast-coded predictors representing sex and race/ethnicity persisted across the three modeling stages (initial, simultaneous, final) for the other two items and for the factor mean and variance of the latent positive expectancies construct. Both the mean level and variance of the factor score were higher for Black and Latinx youth relative to non-Latinx White youth. Furthermore, among Black and Latinx youth, there were sex differences in variance: higher for Black females than Black males but lower for Latinx females than Latinx males.

The item intercept, a parameter representing the average extent to which respondents agreed or disagreed with the behavior/characteristic represented by the item at a given level of the latent variable (i.e., positive expectancies), showed group differences for MEEQB-3 (“Marijuana helps people get along better with others [talk more; feel more romantic]”) and MEEQB-4 (“Marijuana makes people feel more creative and perceive things differently [music sounds different; things are more interesting]”). For MEEQB-3, item intercepts were higher for Black and Latinx youth than non-Latinx White youth. For MEEQB-4, item intercepts were higher for females versus males. Factor loadings, which indicate how well the item represents the underlying construct, did not vary across groups for any positive expectancy item.

Adjusting for bias in the negative expectancies scale using MNLFA. One of the MEEQ-B negative expectancies scale items (MEEQB-5: “Marijuana generally has bad effects on a person”) was fully equivalent with respect to sex, race/ethnicity, and Sex × Race/Ethnicity. Significant differences were observed for the other two negative expectancy items, as well as for the factor mean (Table 2). The negative expectancies factor mean was higher for males than females. The negative expectancies factor mean was highest for non-Latinx White youth, followed by Latinx youth, and lowest for Black youth.

The item intercept for MEEQB-1 (“Marijuana makes it harder to think and do things”) was higher for Latinx compared with Black youth and differed by sex within racial/ethnic group. Whereas for Black and Latinx youth, the MEEQB-1 item intercept was higher for females than males, among non-Latinx White youth it was higher for males. For MEEQB-6 (“Marijuana has bad effects on a person's body and gives people cravings”), the factor loading varied by race/ethnicity. The MEEQB-6 factor loading was higher for Latinx than for Black youth.

Research Question 2: Comparing original mean scores versus MNLFA-adjusted factor scores

Results of pairwise testing of group differences by sex, race/ethnicity, and Sex × Race/Ethnicity in original (unadjusted) and MNLFA-adjusted MEEQ-B scale scores are reported in Table 3 (positive expectancies) and Table 4 (negative expectancies). Means of unadjusted scores are in the top row of Tables 3 and 4, and MNLFA-adjusted scores are in the left column. Hedges' g effect sizes for each comparison are reported above the diagonal for unadjusted scores and below the diagonal for MNLFA-adjusted scores. For a given comparison, the larger of the two effect sizes (unadjusted or adjusted score) is printed in bold; the smaller is printed in regular type.

Table 3.

Group sizes and means for positive expectancies scale sum (unadjusted) scores and adjusted factor scores, with effect sizes (Hedges' g) for group differences and internal consistency reliability (Omega)

graphic file with name jsad.24-00201tbl3.jpg

M (SD) Sex Race/ethnicity Co-occurring identities
F 7.67 (2.78) M 7.64 (2.85) B 7.44 (2.86) L 7.57 (2.93) W 7.74 (2.76) BF 7.53 (2.89) BM 7.34 (2.83) LF 7.54 (2.85) LM 7.60 (3.01) WF 7.76 (2.72) WM 7.72 (2.79)
F .03 (.81) n = 4,248 n.s.
M .05 (.85) n.s. n = 4,735
B -.07 (.87) n = 1,350 n.s. .109
L -.01 (.88) n.s. n = 2,061 n.s.
W .09 (.80) .198 .120 n = 5,572
BF -.06 (.89) n = 681 n.s. n.s. n.s. n.s. n.s.
BM -.09 (.85) n.s. n = 669 n.s. n.s. .151 .136
LF -.03 (.83) n.s. n.s. n = 976 n.s. n.s. n.s.
LM .01 (.93) n.s. n.s. n.s. n = 1,085 n.s. n.s.
WF .08 (.78) .165 .209 .138 n.s. n = 2,591 n.s.
WM .10 (.82) .187 .230 .161 .104 n.s. n = 2,981
Open in a new tab

Notes: Group-specific sample sizes are shown on the diagonal, with corresponding means and standard deviations under group labels atop the upper diagonal for MEEQ-B sum scores, and beside the left adjacent group labels below the diagonal for moderated nonlinear factor analysis (MNLFA) adjusted MEEQB factor scores; off-diagonal coefficients within the three boxes reflect Hedges' g for significant group differences in MEEQ sum scores (above diagonal) and adjusted factor scores (below diagonal) given a false discovery rate of 5%. Regarding absolute between-group differences using the two scoring methods, larger effects are printed in bold typeface. F = female; M = male; B = Black; L = Latinx; W = non-Latinx White; BF = Black female; BM = Black male; LF = Latinx female; LM = Latinx male; WF = non-Latinx White female; WM = non-Latinx White male; n.s. = nonsignificant.

Table 4.

Group sizes and means for negative expectancies scale sum (unadjusted) scores and adjusted factor scores, with effect sizes (Hedges' g) for group differences

graphic file with name jsad.24-00201tbl4.jpg

M (SD) Sex Race/ethnicity Co-occurring identities
F 11.9 (2.36) M 11.9 (2.49) B 11.1 (2.78) L 11.6 (2.51) W 12.2 (2.25) BF 11.2 (2.69) BM 11.0 (2.87) LF 11.6 (2.38) LM 11.6 (2.62) WF 12.1 (2.21) WM 12.1 (2.28)
F .06 (.74) n = 4,248 n.s.
M .12 (.85) .065 n = 4,735
B -.17 (.94) n = 1,350 .176 .450
L -.05 (.81) .147 n = 2,061 .259
W .21 (.73) .490 .336 n = 5,572
BF -.17 (.87) n = 681 n.s. .132 .130 .475 .411
BM -.18 (1.01) n.s. n = 669 .224 .217 .471 .508
LF -.08 (.74) .112 .121 n = 976 n.s. .366 .276
LM -.02 (.87) .169 .174 n.s. n = 1,085 .265 .264
WF .18 (.68) .390 .487 .253 .240 n = 2,591 n.s.
WM .23 (.77) .504 .503 .406 .314 .074 n = 2,981
Open in a new tab

Notes: Group-specific sample sizes are shown on the diagonal, with corresponding means and standard deviations under group labels atop the upper diagonal for MEEQB sum scores, and beside the left adjacent group labels below the diagonal for moderated nonlinear factor analysis (MNLFA)-adjusted MEEQB factor scores; off-diagonal coefficients within the three boxes reflect Hedges' g for significant group differences in MEEQ sum scores (above diagonal) and adjusted factor scores (below diagonal) given a false discovery rate of 5%. Regarding absolute between-group differences using the two scoring methods, larger effects are printed in bold typeface. F = female; M = male; B = Black; L = Latinx; W = non-Latinx White; BF = Black female; BM = Black male; LF = Latinx female; LM = Latinx male; WF = non-Latinx White female; WM = non-Latinx White male; n.s. = nonsignificant.

Positive expectancies: Original mean scores versus MNLFA-adjusted factor scores. Mean positive expectancies scores computed using original scoring did not differ by sex (Table 3). Pairwise comparisons using original scoring by race/ethnicity and Sex × Race/Ethnicity revealed few significant differences: specifically, higher mean positive expectancies for non-Latinx White relative to Black youth and non-Latinx White versus Black males. By contrast, pairwise comparisons of positive expectancies scores adjusted using MNLFA for non-equivalence revealed the same differences by race/ethnicity using original scoring, but with slightly larger effect sizes and several additional between-group differences (small effects) not detected using original scoring: higher mean scores for non-Latinx White versus Latinx youth, non-Latinx White versus Black females, non-Latinx White versus Latinx females, non-Latinx White versus Black males, and non-Latinx White versus Latinx males.

Negative expectancies: Original mean scores versus MNLFA-adjusted factor scores. The pattern of results for the between-group comparisons of negative expectancies scores was quite different from the positive expectancies scale scores (Table 4). Most notably, all group differences across race/ethnicity (including sex-specific comparisons) were significant when using original scores, with non-Latinx White youth having the highest mean scores, followed by Latinx youth and Black youth with the lowest mean scores. Group differences using MNLFA-adjusted scores revealed the same pattern of results, with one additional comparison (non-Latinx White males higher than non-Latinx White females) emerging as significant.

For negative expectancies, all race/ethnicity differences identified using original scoring were also observed using MNLFA-adjusted scores. Compared with the original scoring, effect sizes for MNFLA-adjusted scores were slightly larger for the contrast between Latinx and non-Latinx White youth and between Black and non-Latinx White youth. Race/ethnicity differences among females remained significant, but effect sizes decreased slightly using MNLFA-adjusted scores for all comparisons. Group differences for Black versus Latinx males and Black versus non-Latinx White males remained significant using MNLFA-adjusted scores. However, effect sizes decreased slightly, whereas the effect size for the comparison of Latinx and non-Latinx White males increased slightly.

Discussion

Drawing on a national sample of Black, Latinx, and non-Latinx White pre-to early adolescent females and males, this study demonstrated the importance of applying measurement equivalence methods that consider not only sex and race/ethnicity but also co-occurring social identities, and more specifically their implications for understanding cannabis expectancies in youth. As a crucial step after identifying bias in measurement, MNLFA was used to generate scores that account for non-equivalence due to sex, race/ethnicity, and co-occurring social identities in MEEQ-B positive and negative expectancies scales. MNLFA-adjusted scores allowed a more valid comparison of relative mean expectancies scores across six subgroups: Black females, Black males, Latinx females, Latinx males, non-Latinx White females, and non-Latinx White males. Measurement non-equivalence impacted comparisons of MEEQ-B positive and negative expectancies scores across these identities, thus potentially leading to inaccurate interpretation of results in the absence of adjusting for measurement non-equivalence.

Factor-level differences by sex, race/ethnicity, and co-occurring social identities

Using MNLFA, both positive and negative expectancies factors evidenced differences in the mean and variance of scores as a function of sex, race/ethnicity, and/or co-occurring social identities. The greater factor score variance for Black and Latinx youth, relative to non-Latinx White youth, warrants comment. It suggests greater heterogeneity among Black and Latinx youth in anticipated positive effects at this early age, underscoring the importance of attending to variability within racial/ethnic groups (Montgomery et al., 2022).

Item-level differences by sex, race/ethnicity, and co-occurring social identities

At the item level, only two of the MEEQ-B's six items (one positive expectancy item and one negative expectancy item) were fully equivalent across sex, race/ethnicity, and co-occurring social identities. The positive expectancy item, “Marijuana helps a person relax and feel less tense (helps a person unwind and feel calm),” reflects cannabis's anticipated effects on general mood. The negative item, “Marijuana generally has bad effects on a person,” captures negative perceptions of cannabis's effects broadly. These global expected effects may be acquired without personal experience with cannabis use, consistent with the very limited use in the pre-adolescent age range sampled. Prior longitudinal research with youth in this age range, starting as young as age 10 (Kristjansson et al., 2012) or 12 (Foster et al., 2018), also found that youth had formed cannabis expectancies despite limited experience with cannabis use.

With regard to specific MEEQ-B items, the nature and direction of non-equivalence findings at the item level varied. For example, the item intercept was higher for Black and Latinx relative to non-Latinx White youth on one positive expectancy item (MEEQB-3) but higher for females than males on another (MEEQB-4); for one negative expectancy item (MEEQB-1), sex differences in intercepts differed across racial/ethnic groups. The absence of a clear pattern suggests that there is no single source of measurement bias to address. Rather, the complexity of the findings underscores the importance of conducting measurement equivalence testing and adjusting for potential bias. The core advantage of MNLFA over other methods for testing for measurement equivalence is that MNLFA can generate adjusted scores to minimize measurement bias without the need to develop a new measure (Gottfredson et al., 2019).

Differences in means derived from unadjusted versus adjusted cannabis expectancies scores

We found extensive evidence that adjusting for measurement bias in the MEEQ-B with respect to sex, race/ethnicity, and co-occurring social identities changed estimated group differences in both positive and negative cannabis expectancies. Five group differences in positive expectancies that were undetected using unadjusted scores emerged using adjusted scores: higher positive expectancies for non-Latinx White relative to Latinx youth as well as multiple within-sex differences across racial/ethnic groups. The between-group differences in mean adjusted expectancies scores show some correspondence to slightly higher lifetime cannabis use by eighth grade among non-Latinx White youth relative to Black and Latinx youth (Miech et al., 2024), suggesting the possible role of peer norms in the development of expectancies and cannabis use (Aarons et al., 2001; Schafer & Brown, 1991).

Group mean comparisons using original (unadjusted) scores for the negative expectancies scale indicated that nearly all racial/ethnic group comparisons, across and within sex, were significant (non-Latinx White youth reported the highest average scores, followed by Latinx youth, then Black youth). In contrast to positive expectancies scores, only one new significant difference emerged after adjusting for measurement equivalence. However, the effect size changed for all previously observed group differences. Unadjusted scores underestimated how much lower negative expectancies were in Black and Latinx relative to non-Latinx White youth and overestimated how much lower they were in Black youth relative to Latinx youth. Of note, negative expectancies scores were still high in this sample of pre-to early adolescents, as expected in this developmental period and consistent with their limited cannabis use at this age (1.07% reported lifetime cannabis use).

Study findings show that failing to adjust for measurement non-equivalence can result in biased estimates of constructs and invalid comparisons across groups, particularly for marginalized populations (Han et al., 2019). Although this is not the first study to assess measurement equivalence of a cannabis expectancies measure across sex and race/ethnicity (see Waddell et al.'s [2021] study of the Anticipated Effects of Cannabis Scale with young adults), it is the first known to consider co-occurring social identities and the first to examine pre-to early adolescents. Emerging research highlights the importance of considering co-occurring social identities to better capture potential sources of measurement bias and improve valid assessment in diverse samples (Bauer et al., 2020).

Study limitations

Several study limitations need to be considered. First, inferences cannot be made about measurement equivalence at other ages or for racial/ethnic groups not included in this study. Second, as noted in the sample description, sex assigned at birth was used. Third, there is substantial heterogeneity within the broadly defined racial/ethnic categories “Black,” “Latinx,” and “non-Latinx White” that merit further exploration. Fourth, as substance use shapes expectancies (Smith et al., 1995), it will be important to incorporate cannabis use into analyses of expectancies in older samples, when cannabis use is less limited. Fifth, although meaningful in terms of conclusions drawn, the majority of group mean differences represented small effects. Sixth, MEEQ-Bspecific limitations warrant consideration, including the limited specificity of certain effects due to combining multiple effects into single items for brevity, querying exclusively smoking cannabis, and the absence of assessing arousal or value of a given effect.

Conclusions

Results emphasize the need for caution when interpreting the scores of a measure such as the MEEQ-B that has not undergone testing and adjustment for measurement equivalence (Han et al., 2019). When we used observed (unadjusted) scores, both underestimation and overestimation of differences between groups were identified. The use of MNLFA methods to address measurement equivalence has implications for improving the cultural inclusiveness of assessment measures through accurate measurement and valid group comparisons by accounting for possible sources of systematic bias (Han et al., 2019).

Footnotes

This research was supported by grants from the National Institute on Minority Health and Health Disparities (R01MD016922) and National Institute on Drug Abuse (U01 DA056472-01A1). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health (NIH). Data used in the preparation of this article were obtained from the Adolescent Brain Cognitive Development SM (ABCD) Study (https://abcdstudy.org), held in the NIMH Data Archive (NDA). This is a multisite, longitudinal study designed to recruit more than 10,000 children ages 9–10 and follow them over 10 years into early adulthood. The ABCD Study is supported by the NIH and additional federal partners under award numbers U01DA041048, U01DA050989, U01DA051016, U01DA041022, U01DA051018, U01DA051037, U01DA050987, U01DA041174, U01DA041106, U01DA041117, U01DA041028, U01DA041134, U01DA050988, U01DA051039, U01DA041156, U01DA041025, U01DA041120, U01DA051038, U01DA041148, U01DA041093, U01DA041089, U24DA041123, U24DA041147. A full list of supporters is available at https://abcdstudy.org/federal-partners.html. A listing of participating sites and a complete listing of the study investigators can be found at https://abcd-study.org/consortium_members. ABCD consortium investigators designed and implemented the study and/or provided data but did not necessarily participate in the analysis or writing of this report. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or ABCD consortium investigators.

References

  1. Aarons G. A., Brown S. A., Stice E., Coe M. T. Psychometric evaluation of the marijuana and stimulant effect expectancy questionnaires for adolescents. Addictive Behaviors. 2001;26(2):219–236. doi: 10.1016/S0306-4603(00)00103-9. [DOI] [PubMed] [Google Scholar]
  2. Bauer D. J. A more general model for testing measurement invariance and differential item functioning. Psychological Methods. 2017;22(3):507–526. doi: 10.1037/met0000077. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bauer D. J., Belzak W. C. M., Cole V. T. Simplifying the assessment of measurement invariance over multiple background variables: Using regularized moderated nonlinear factor analysis to detect differential item functioning. Structural Equation Modeling: A Multidisciplinary Journal. 2020;27(1):43–55. doi: 10.1080/10705511.2019.1642754. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Benjamini Y., Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B (Methodological) 1995;57(1):289–300. doi: 10.1111/j.2517-6161.1995.tb02031.x. [DOI] [Google Scholar]
  5. Bentler P. M. Comparative fit indexes in structural models. Psychological Bulletin. 1990;107(2):238–246. doi: 10.1037/0033-2909.107.2.238. [DOI] [PubMed] [Google Scholar]
  6. Brackenbury L. M., Ladd B. O., Anderson K. G. Marijuana use/cessation expectancies and marijuana use in college students. The American Journal of Drug and Alcohol Abuse. 2016;42(1):25–31. doi: 10.3109/00952990.2015.1105242. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Brown T. A. Confirmatory factor analysis for applied research. 2nd edition. Guilford Press; 2015. [Google Scholar]
  8. Browne M. W., Cudeck R. Alternative ways of assessing model fit. In: Bollen K. A., Long J. S., editors. Testing structural equation models. Sage; 1993. pp. 136–162. [Google Scholar]
  9. Cheung G. W., Rensvold R. B. Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling: A Multidisciplinary Journal. 2002;9(2):233–255. doi: 10.1207/S15328007SEM0902_5. [DOI] [Google Scholar]
  10. Cole V. T., Gottfredson N. C., Giordano M., Janssen T. Automated fitting of moderated nonlinear factor analysis (MNLFA) through the Mplus program. 2021 R package version 1.0. [Google Scholar]
  11. Foster D. W., Ye F., Chung T., Hipwell A. E., Sartor C. E. Longitudinal associations between marijuana-related cognitions and marijuana use in African-American and European-American girls from early to late adolescence. Psychology of Addictive Behaviors. 2018;32(1):104–114. doi: 10.1037/adb0000323. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Garavan H., Bartsch H., Conway K., Decastro A., Goldstein R. Z., Heeringa S., Jernigan T., Potter A., Thompson W., Zahs D. Recruiting the ABCD sample: Design considerations and procedures. Developmental Cognitive Neuroscience. 2018;32:16–22. doi: 10.1016/j.dcn.2018.04.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Goldman M. S., Del Boca F. K., Darkes J. Alcohol expectancy theory: The application of cognitive neuroscience. In: Leonard K., Blane H., editors. Psychological theories of drinking and alcoholism. 2nd ed. Guilford Press; 1999. pp. 203–246. [Google Scholar]
  14. Gottfredson N. C., Cole V. T., Giordano M. L., Bauer D. J., Hussong A. M., Ennett S. T. Simplifying the implementation of modern scale scoring methods with an automated R package: Automated moderated nonlinear factor analysis (aMNLFA) Addictive Behaviors. 2019;94:65–73. doi: 10.1016/j.addbeh.2018.10.031. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Han K., Colarelli S. M., Weed N. C. Methodological and statistical advances in the consideration of cultural diversity in assessment: A critical review of group classification and measurement invariance testing. Psychological Assessment. 2019;31(12):1481–1496. doi: 10.1037/pas0000731. [DOI] [PubMed] [Google Scholar]
  16. Hedges L. V., Olkin I. Statistical methods for meta-analysis. Academic Press; 1985. [Google Scholar]
  17. Heeringa S. G., Berglund P. A. A guide for population-based analysis of the Adolescent Brain Cognitive Development (ABCD) Study baseline data. bioRxiv. 2020 doi: 10.1101/2020.02.10.942011. [DOI] [Google Scholar]
  18. Hu L.-t., Bentler P. M. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal. 1999;6(1):1–55. doi: 10.1080/10705519909540118. [DOI] [Google Scholar]
  19. Kline R. B. Principles and practice of structural equation modeling. 3rd ed. Guilford Press; 2011. [Google Scholar]
  20. Kristjansson S. D., Agrawal A., Lynskey M. T., Chassin L. A. Marijuana expectancies and relationships with adolescent and adult marijuana use. Drug and Alcohol Dependence. 2012;126(1–2):102–110. doi: 10.1016/j.drugalcdep.2012.04.024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Miech R. A., Johnston L. D., Patrick M. E., O'Malley P. M. Monitoring the Future national survey results on drug use 1975–2023: Overview and detailed results for secondary school students. Ann Arbor, MI: Institute for Social Research; 2024. [Google Scholar]
  22. Montes K. S., Witkiewitz K., Pearson M. R., Leventhal A. M. Alcohol, tobacco, and marijuana expectancies as predictors of substance use initiation in adolescence: A longitudinal examination. Psychology of Addictive Behaviors. 2019;33(1):26–34. doi: 10.1037/adb0000422. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Montgomery L., Dixon S., Mantey D. S. Racial and ethnic differences in cannabis use and cannabis use disorder: Implications for researchers. Current Addiction Reports. 2022;9(1):14–22. doi: 10.1007/s40429-021-00404-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Muthén L. K., Muthén B. O. Mplus user's guide. 8th edition. Muthén & Muthén; 2017. [Google Scholar]
  25. Satorra A., Bentler P. M. A scaled difference chi-square test statistic for moment structure analysis. Psychometrika. 2001;66(4):507–514. doi: 10.1007/bf02296192. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Schafer J., Brown S. A. Marijuana and cocaine effect expectancies and drug use patterns. Journal of Consulting and Clinical Psychology. 1991;59(4):558–565. doi: 10.1037/0022-006x.59.4.558. [DOI] [PubMed] [Google Scholar]
  27. Shih R. A., Miles J. N. V., Tucker J. S., Zhou A. J., D'Amico E. J. Racial/ethnic differences in adolescent substance use: Mediation by individual, family, and school factors. Journal of Studies on Alcohol and Drugs. 2010;71(5):640–651. doi: 10.15288/jsad.2010.71.640. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Smith G. T., Goldman M. S., Greenbaum P. E., Christiansen B. A. Expectancy for social facilitation from drinking: The divergent paths of high-expectancy and low-expectancy adolescents. Journal of Abnormal Psychology. 1995;104(1):32–40. doi: 10.1037/0021-843x.104.1.32. [DOI] [PubMed] [Google Scholar]
  29. Torrealday O., Stein L. A. R., Barnett N., Golembeske C., Lebeau R., Colby S. M., Monti P. M. Validation of the Marijuana Effect Expectancy Questionnaire–Brief. Journal of Child & Adolescent Substance Abuse. 2008;17(4):1–17. doi: 10.1080/15470650802231861. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Waddell J. T., Corbin W. R., Meier M. H., Morean M. E., Metrik J. The Anticipated Effects of Cannabis Scale (AECS): Initial development and validation of an affect- and valence-based expectancy measure. Psychological Assessment. 2021;33(2):180–194. doi: 10.1037/pas0000881. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Wang M., Russell S.S. Measurement equivalence of the Job Descriptive Index across Chinese and American workers: Results from confirmatory factor analysis and item response theory. Educational and Psychological Measurement. 2005;65:709–732. doi: 10.1177/0013164404272494. [DOI] [Google Scholar]
  32. Wheaton B., Muthén B., Alwin D. F., Summers G. F. Assessing reliability and stability in panel models. Sociological Methodology. 1977;8:84–136. doi: 10.2307/270754. [DOI] [Google Scholar]
  33. Widaman K. F., Reise S. P. Exploring the measurement invariance of psychological instruments: Applications in the substance use domain. In: Bryant K. J., Windle M., West S. G., editors. The science of prevention: Methodological advances from alcohol and substance abuse research. American Psychological Association; 1997. pp. 281–324. [DOI] [Google Scholar]

Articles from Journal of Studies on Alcohol and Drugs are provided here courtesy of Rutgers University. Center of Alcohol Studies

RESOURCES