Affect, Motives for Cannabis Use, Duration of Intoxication, and Cannabis Consequences: Cannabis Use Problem Severity as a Potential Moderator

Abstract

Background:

The multistage model of substance use proposes that reinforcement mechanisms and motives for use change as individuals develop substance use disorders, with positive reinforcement driving substance use when substance use disorder symptoms are absent/mild and negative reinforcement becoming dominant when substance use disorder symptoms are more severe. We aimed to determine whether event-level associations between affect, cannabis use, motives for use, and heaviness of use were moderated by cannabis use problem severity, consistent with the multistage model of substance use.

Methods:

We used EMA data (2 observations per day for 14 days) from 571 young adult females who regularly used cannabis. The sample included a majority sexual minority women and gender diverse individuals (77.6%).

Results:

Elevated anxious affect was concurrently associated with coping motives for cannabis use and longer duration of intoxication at the event-level but only among those with more severe CU problems. Further, positive affect was concurrently associated with a higher likelihood of endorsing enhancement motives for the full sample, but this association was stronger for those with fewer CU problems. However, many other associations between affect and cannabis use outcomes were not moderated by severity of CU problems.

Discussion:

Some findings were consistent with what would be expected based on the multistage model. However, results also suggest that positive reinforcement may continue to drive cannabis use among those with more severe CU problems which is not consistent with what would be expected based on the multistage model.

Introduction

Theoretical models posit that cannabis use (CU) is driven by the desire to decrease negative affect (e.g., self-medication model) and/or increase positive affect (e.g., motivational models; Cooper et al., 2016; Khantzian, 1997). Testing these models, ecological momentary assessment (EMA) studies have examined event-level associations between affect and CU (e.g., negative affect predicting CU later in the day), with mixed results (e.g., Buckner et al., 2015; Chakroun et al., 2010; Van Doren et al., 2024). Although some studies have linked negative affect with subsequent CU (Buckner et al., 2012; Buckner et al., 2015), others have not (Chakroun et al., 2010; Linden-Carmichael et al., 2022; Testa et al., 2019; Tournier et al., 2003; Van Doren et al., 2024). Some have even linked negative affect to a decreased likelihood of CU (Chakroun et al., 2010; Dora et al., 2023). Samples in which negative affect and CU have been linked have a higher proportion with CU disorders (63–87%; Buckner et al., 2012; Buckner et al., 2015) compared to those that did not (25–39%; Chakroun et al., 2010; Tournier et al., 2003). Thus, negative affect may only be associated with an increased likelihood of CU among individuals with CU disorders, consistent with the multistage model, which posits that substance use is driven by desire to reduce negative affect only among individuals with substance use disorders (SUDs).

Other aspects of event-level associations between affect and CU have received less attention, including effect of positive affect on CU and of affect on other CU outcomes (e.g., quantity, motives). Studies of positive affect have been mixed, with positive affect being associated with an increase (Buckner et al., 2015; Dora et al., 2023), decrease (Testa et al., 2019), or no change in likelihood of CU (Van Doren et al., 2024). We are aware of only one study examining associations between affect and CU quantity, which linked higher positive affect with heavier consumption, higher depressed affect with less consumption, and found no association between anxious affect and quantity (Chakroun et al., 2010). Similarly, we are aware of only one study examining associations between affect and motives for CU, linking negative affect with elevated coping and social, but not enhancement, motives (Buckner et al., 2015). Further, no prior studies have examined associations between affect and duration of intoxication, a measure of cannabis consumption that can be used across modes of CU (Fairlie et al., 2023; Waddell, et al., 2022; Walukevich-Dienst et al., 2023) and has been linked with higher CU quantity and more consequences (Calhoun et al., 2022). In the absence of a measure of CU quantity that can be used across modes, duration of intoxication can help to describe cannabis consumption events. Given few studies on these associations and mixed findings, there is a need for further research testing moderators that may explain mixed results.

The Multistage Model of Substance Use

The multistage model of substance use posits that event-level factors contributing to substance use change as SUDs develop (Koob, 2013; Koob & Volkow, 2010). When SUD symptoms are absent or mild, substance use is theorized to be driven by desire to increase positive affect (i.e., positive reinforcement), with individuals using substances, using more heavily, and endorsing enhancement motives (i.e., desire to increase positive affect) for substance use when positive affect is elevated. When more SUD symptoms develop, negative reinforcement (i.e., desire to decrease negative affect) is theorized to become the dominant driver of substance use, with individuals using substances, using more heavily, and being more likely to endorse coping motives when negative affect is elevated. Numerous animal models and neuropsychological studies have provided support for the multistage model (Koob, 2013; Koob & Volkow, 2010; Kwako & Koob, 2017), but fewer have tested this theory in real world settings.

Studies have largely been consistent with the proposed negative reinforcement components of the multistage model. Cho et al. (2019) demonstrated that negative reinforcement was more strongly associated with heavier alcohol consumption among those with alcohol dependence compared to those without. In an EMA study of female college students, negative affect and tension reduction expectancies (i.e., expecting alcohol to increase relaxation) only predicted heavier consumption among those with more alcohol consequences (Dyar & Kaysen, 2023). Two EMA studies have tested the multistage model among sexual and gender minority individuals (SGM). Among young adult sexual minority women and gender diverse individuals (SMWGD), enacted stigma (a negative reinforcement risk factor) was concurrently and prospectively associated with a higher likelihood of drinking and CU, but only among individuals with more severe alcohol or cannabis use problems (Dyar et al., 2023). Among older adult SGM, a similar pattern was found but only for concurrent associations (Dyar et al., 2025). While studies have largely supported negative reinforcement components of the multistage model, the moderation of associations between affect, CU, and motives have not been examined.

Studies examining positive reinforcement components of the multistage model have produced more mixed results. Some findings indicate positive reinforcement factors (e.g., social/enhancement motives; positive affect) are associated with heavier drinking or CU for the entire sample, regardless of substance use problem severity (Cho et al., 2019; Dyar & Feinstein, 2025; Dyar & Kaysen, 2023). Others have even found that associations between positive reinforcement factors (i.e., social motives; positive affect) and drinking and/or CU are stronger for those with more severe alcohol or cannabis use problems – the opposite of what would be expected based on the multistage model (Dyar & Feinstein, 2025; Dyar & Kaysen, 2023). Consistent with the multistage model, one of these studies found that the association between sociability expectancies and drinking quantity was stronger among those with fewer alcohol consequences (Dyar & Kaysen, 2023). These largely negative results for positive reinforcement components of the multistage model may indicate that this component only applies to some dimensions of CU.

Focus on High-Risk Samples

Several studies testing the multistage model have focused on populations at high risk for SUDs: SGM and SMWGD (Dyar et al., 2025; Dyar & Kaysen, 2023; Dyar et al., 2023). The multistage model posits that the cross-sensitization of neural pathways involved in stress reactivity and substance withdrawal contributes to emotion dysregulation and interferes with problem-solving coping (Kwako & Koob, 2017; Lijffijt et al., 2014). In turn, these coping deficits increase use of substances to cope and contribute to development of SUDs (Kwako & Koob, 2017; Lijffijt et al., 2014). The added stressors experienced by SGM due to stigma (Meyer, 2003) may make them more susceptible to this cross-sensitization, increasing the likelihood that SGM with SUDs will respond as proposed by the multistage model. While testing this theory among populations that may be more susceptible to the patterns it describes is important, it is also necessary to determine whether the pattern of results replicates among cisgender, heterosexual samples.

The Current Study

The current study aimed to test hypotheses derived from the multistage model using EMA data from young adult SMWGD and heterosexual, cisgender women. This study focused on young adults (ages 18–25) as this is a period when CU disorders are elevated (Leung et al., 2020) and many CU disorders first develop (Solmi et al., 2022). We tested whether event-level associations between affect (anxious, depressed, positive) and CU outcomes (likelihood of CU, duration of intoxication, consequences, motives) were moderated by the CUDIT-R, a measure of CU problem severity associated with CU disorder diagnosis (Coelho et al., 2024; Myers et al., 2023). We expected these associations to be moderated by CU problem severity and display the following patterns.

1. When anxious and depressed affect were elevated, individuals with more severe CU problems would be more likely to use cannabis, report more coping motives for CU, be intoxicated for longer, and experience more consequences during that observation and the next. These associations would be weaker for those with less severe CU problems.

2. When positive affect was elevated, individuals with less severe CU problems would be more likely to use cannabis, report more enhancement motives for CU, be intoxicated for longer, and experience more consequences during that observation and the next. These associations would be weaker for those with more severe CU problems.

3. We also explored whether results were similar for SMWGD and cisgender, heterosexual women.

Methods

Participants and Procedures

The current analyses used data from a longitudinal study of substance use among a sample of young adult females, including sexual minority and heterosexual cisgender women and gender diverse sexual minorities assigned female at birth. Participants were recruited via online advertisements on social media (e.g., Facebook) between January and December 2024. The study included a baseline assessment (day 0) and a 14-day EMA study (days 1–14). During the EMA period, participants completed one survey in the morning (8:00am-1:00pm in their time zone) and one in the evening (6:00pm-12:00am in their time zone) via REDCap. The study received IRB approval at Ohio State University.

Eligible participants were U.S. residents, aged 18–25, assigned female at birth, identified as a sexual minority and a woman or as gender diverse (e.g., nonbinary) or identified as heterosexual and as a woman, and used cannabis at least four times in the past month at baseline.¹ Participants were paid up to $140: $30 for baseline, $2.50 for each EMA survey, and $10 bonus for each 6 surveys completed in a row. Participants who screened eligible based on their eligibility survey were sent a text message by study team members to verify eligibility. Participants were asked to text demographic information (i.e., age, state of residence, email address), which was cross-checked with their responses in the eligibility survey.

653 participants completed the baseline assessment. The analytic sample included all individuals who reported CU at least once during the EMA (n =571). The sample was diverse in race/ethnicity, with 49.4% identifying with at least one marginalized racial/ethnic identity. Participants included 128 heterosexual cisgender women (22.4%), 291 sexual minority cisgender women (50.9%), and 152 (26.6%) gender diverse sexual minorities. See Table 1.

Table 1.

Demographics of Analytic Sample at Baseline (N = 571)

Demographic Variable	n	%
Sexual Identity
Lesbian/Gay	170	29.8%
Bisexual	139	24.3%
Pansexual	34	6.0%
Queer	100	17.5%
Heterosexual	128	22.4%
Race/Ethnicitya
White	367	64.3%
Black	108	18.9%
Latine	84	14.7%
Asian	83	14.5%
Other Race/Ethnicity	30	5.3%
Gender Identity
Cisgender Woman	419	73.4%
Gender Minority	152	26.6%
Age (M, SD)	22.67 (1.98)

^aPercentages add up to more than 100% because participants could select multiple racial/ethnic identities.

Measures

EMA Measures

Affect was assessed using four items from the Profile of Mood States Short Form’s depressed (discouraged; sad) and anxious subscales (anxious; on edge) (Cranford et al., 2006; Curran et al., 1995) and two items from the Positive and Negative Affect Scale’s happy subscale (cheerful, happy; Ebesutani et al., 2012). Participants were asked to indicate how much they felt each emotion since the last survey (0 [not at all] to 4 [extremely]). Affect was assessed prior to other items to avoid order effects. Internal consistency was estimated using procedures recommended by Nezlek (2017) and was acceptable (.56–.67).

CU was assessed by asking participants, “which of the following have you used since the last survey?” Participants could select alcohol and/or marijuana or “none of the above.” A binary variable was created indicating no CU (0) or CU (1).

CU Outcomes.

Duration of Intoxication was measured by asking “How many hours were you high?” Responses were provided in increments of one hour from 0 to 12+ hours. This item was adapted from the CU Inventory and participants were asked to answer it thinking about their use since the last survey (Cuttler & Spradlin, 2017).

CU Motives.

Participants were asked whether they used cannabis to cope (3 items; e.g., “to reduce my anxiety”); and/or to enhance positive emotions (1 item; e.g., “to feel good or have fun”). Response options were no (0) and yes (1). Participants could endorse multiple motives. Items were adapted from Patrick et al. (2019). Consistent with Dyar et al. (2022), three coping motive items were used to create a single binary variable (0 = endorsement of no coping motives, 1 = endorsement of at least one coping motive) so the scale matched that for enhancement motives (0 = enhancement item not endorsed; 1 = enhancement item endorsed).

Cannabis Consequences were measured only during morning assessments (and asked about the past 24 hours) by using eight items from two measures of marijuana consequences (Lee et al., 2021; Simons et al., 2012). Participants were asked which “of the following things happened to you as a result of your marijuana use yesterday?” (e.g., “I felt dizzy or sick”). A sum of consequences was calculated.

Baseline Measure

Cannabis use problem severity was assessed using the Cannabis Use Disorder Identification Test Revised (CUDIT-R; Adamson et al., 2010). The CUDIT-R includes 8 items rated on different scales. For example, the item “How often during the past 6 months did you fail to do what was normally expected from you because of using marijuana?” was rated from 0 (never) to 4 (daily or almost daily). Total scores ranged from 0 to 32 (a = .78) and were used in analyses. Prior research indicates that scores of 10+ indicate probable DSM-5 cannabis use disorder (Bonn-Miller et al., 2016) and 13+ indicate probable DSM-IV cannabis dependence (Adamson et al., 2010).

Analytic Plan

Analyses were conducted in Mplus 8.11. There were 12,696 completed surveys from 571 participants. The median individual-level completion rate was 89.3% (M = 76.7%, SD = 28.0%). Within completed surveys, <1% of data were missing and were handled using Bayesian methods (Asparouhov & Muthén, 2010). Analyses of CU outcomes (e.g., duration of intoxication, motives, consequences) included observations when CU was reported (5,321 observations).

Bayesian multilevel structural equation modeling (MSEM) with non-informative priors was used.² Latent variables were only used to separate within and between-person versions of composite variables, and composite scores (not individual items) were used in model estimation. We used Markov Chain Monte Carlo (MCMC) algorithms to generate a series of 10,000 random draws from the multivariate posterior distribution of our sample for each model. Trace plots and the Gelman-Rubin potential scaling reduction (PSR) were used to determine whether convergence was achieved (Depaoli & Clifton, 2015; Muthen, 2010). Bayesian credible intervals were used to determine statistical significance.

First, we examined within- (i.e., event-level) and between-person associations between affect (i.e., anxious, depressed, happy) and CU. In concurrent models, affect during one observation predicted CU during the same observation. In prospective models, affect during one observation (t) predicted CU during the next observation (t+1) and an autocorrelation for CU (AR1 structure) was included (which effectively controls for CU at t). Baseline CU problem severity (i.e., CUDIT-R scores) was entered as a between-person predictor of the event-level association between affect and CU (i.e., a cross-level interaction). This cross-level interaction was used to determine whether associations between affect and CU outcomes varied based on CU problem severity. The Johnson-Neyman technique was also used to determine at which values of CUDIT-R the associations between affect and CU outcomes became significant. Simple slopes were calculated using model constraints and loop functions. For significant cross-level interactions, predicted probabilities for binary outcomes were calculated using recommendations described by McNeish et al. (2024). A probit link was used for binary outcomes as logistic regression is not available in Bayesian MSEM. Probit regression coefficients represent the variance shared by the predictor and the latent continuous response variables underlying each binary observed item (Agresti, 2003). In all models, we controlled for day of EMA survey period and assessment type (weekend/weekday [weekends included Friday, Saturday, and Sunday]; morning/evening) at the within-person level. Within-person associations among affect and CU variables as well as autocorrelations were allowed to vary across individuals. Age, sexual identity, sex/gender identity, and race/ethnicity were included as covariates at the between-person level. See Figures 1 and 2 for a visual representation of the main paths of interest included in these analyses.

Figure 1.

Visual depiction of concurrent multilevel model. Lines with dots at the within-person level indicate random slopes. Lines without dots indicate fixed effects. The line between CUDIT-R and the within-person association between affect and cannabis use outcome represents the cross-level interaction. While associations between anxious, depressed, and positive affect were estimated in the same models, we present the association for only one affect variable in the figure for clarity. Covariates and residual variances are not presented for simplicity. Covariates included: within-person (day of EMA period, weekend vs weekday, and morning vs. evening survey); between-person (race/ethnicity, sexual identity, gender identity, age).

Figure 2.

Visual depiction of prospective multilevel model. Lines with dots at the within-person level indicate random slopes. Lines without dots indicate fixed effects. The line between CUDIT-R and the within-person association between affect and cannabis use outcome represents the cross-level interaction. While associations between anxious, depressed, and positive affect were estimated in the same models, we present the association for only one affect variable in the figure for clarity. Covariates and residual variances are not presented for simplicity. Covariates included: within-person (day of EMA period, weekend vs weekday, and morning vs. evening survey); between-person (race/ethnicity, sexual identity, gender identity, age).

The second set of models followed the same pattern described above but focused on concurrent and prospective effects of affect on motives for and outcomes of CU (i.e., duration of intoxication, consequences) during observations when CU was reported. As consequences were only assessed during the morning survey and asked about experiences over the last 24 hours, affect during the prior evening survey (t-1) was used for concurrent associations and affect as reported during the prior morning survey (t-2) was used for prospective associations with consequences. CU problem severity (i.e., CUDIT-R scores) was entered as between-person predictors of the event-level association between affect and cannabis motives and outcomes.

Sensitivity Analyses

We conducted sensitivity analyses in which analyses were stratified for SMWGD and heterosexual, cisgender women to determine if similar patterns were identified for these two groups. Given that power was low for three-way interactions, we conducted stratified analyses by comparing the significance and strength of associations among the main variables of interest to determine whether the pattern of results was similar or different for these two groups. As the CUDIT-R includes items about two related but distinct constructs: frequency and consequences of CU, we also conducted sensitivity analyses in which only items capturing CU problems were included in the CUDIT-R score to determine if this changed the results.

Results

Participants used cannabis on 42% of observations. Approximately 32.2% of participants were below the CUDIT-R threshold for a probable CU disorder (scores < 10), while 67.8% met criteria for probable CU disorder (38.7% meeting criteria for cannabis dependence/severe CU disorder). Table 2 provides means, standard deviations, and intraclass correlations.

Table 2.

Means, Variances, and Intraclass Correlations

	Mean	Standard Deviation	Range	Intraclass Correlation
Anxious Affect	1.32	1.13	0–4	.41
Depressed Affect	.99	1.05	0–4	.42
Positive Affect	2.02	1.04	0–4	.41
Cannabis Use (CU)	.42	.49	0–1	.28
Coping Motives for CU	.79	.41	0–1	.23
Enhancement Motives for CU	.40	.49	0–1	.18
Duration Cannabis Intoxication	3.06	1.94	0–12	.43
Cannabis Consequences	.51	.89	0–7	.44
Baseline CU Problem Severity (CUDIT-R)	13.01	6.18	1–32	-

CUDIT-R = Cannabis Use Disorder Identification Test – Revised.

Moderation of concurrent event-level associations by CU problem severity

CU problem severity was a significant predictor of three concurrent event-level associations: between anxious affect and coping motives, anxious affect and duration of intoxication, and positive affect and enhancement motives (Table 3), indicating that these associations varied depending on an individual’s level of CU problem severity. Associations between anxious affect, coping motives, and duration of intoxication became positive and significant only for those with more severe CU problems, consistent with hypotheses. Experiencing more anxious affect than usual was significantly associated with a higher likelihood of using cannabis to cope only among individuals with CUDIT-R scores of 14 or higher. Simple slopes are presented in Table 5 and Figure 3. Experiencing more anxious affect than usual was associated with being intoxicated for less time among individuals with lower CUDIT-R scores (7 or below), was not associated with duration of intoxication for individuals with moderate CUDIT-R scores (8–18), and was associated with being intoxicated for longer among those with very high CUDIT-R scores (19+).

Table 3.

Within-person concurrent direct effect estimates and moderation

			Anxious Affect			Depressed Affect			Positive Affect
Model	Predictor	Outcome	b	95% CI	p	b	95% CI	p	b	95% CI	p
Cannabis Use	Affect	Cannabis Use	−.03	−.12, .07	.61	.05	−.06, .15	.37	.21	.11, .31	< .001
	CUDIT-R	Affect → Cannabis Use	.004	−.002, .01	.23	.001	−.01, .01	.87	−.002	−.01, .005	.62
Coping Motives	Affect	Coping Motives	−.12	−.32, .07	.22	.14	−.11, .39	.25	−	−	−
	CUDIT-R	Affect → Coping Motives	.01	.002, .03	.02	.01	−.01, .02	.36	−	−	−
Enhancement Motives	Affect	Enhancement Motives	-	-	-	-	-	-	.58	.39, .77	< .001
	CUDIT-R	Affect → Enhancement Motives	-	-	-	-	-	-	−.01	−.02, −.0001	.04
Duration Intoxication	Affect	Duration Intoxication	−.22	−.38, −.05	.01	.09	−.09, .27	.31	.19	.01, .37	.04
	CUDIT-R	Affect → Duration Intoxication	.02	.01, .03	.004	−.01	−.02, .01	.32	−.001	−.01, .01	.84
Consequences	Affect	Consequences	−.03	−.13, .06	.48	.08	−.03, .20	.16	−.03	−.16, .10	.65
	CUDIT-R	Affect → Consequences	.01	−.001, .01	.10	−.001	−.01, .01	.78	.003	−.01, .01	.43

Covariates included: within-person (day of EMA period, weekend vs weekday, and morning vs. evening survey); between-person (race/ethnicity, sexual identity, gender identity, age). Associations among variables at the between-person level were also estimated but are not included in tables. See supplementary materials for tables with all parameters. Significant effects are presented in bold. CUDIT-R = Cannabis Use Disorder Identification Test – Revised.

Table 5.

Within-Person Simple Slopes for Significant Interactions

		CUDIT-R 5		CUDIT-R 10		CUDIT-R 15		CUDIT-R 20		CUDIT-R 25
Predictor	Outcome	b	95% CI	b	95% CI	b	95% CI	b	95% CI	b	95% CI
Anxious Affect	Coping Motives (Unlagged)	−.05	−.19, .09	.02	−.07, .12	.10	.02, .17	.17	.07, .27	.24	.09, .40
	Duration Intoxication (Unlagged)	−.14	−.26, −.02	−.06	−.14, .02	.02	−.04, .08	.10	.01, .18	.18	.05, .30
Positive Affect	Enhancement Motives (Unlagged)	.52	.38, .66	.46	.36, .56	.40	.33, .47	.34	.25, .43	.28	.15, .41
	Duration Intoxication (Lagged)	−.06	−.19, .07	.01	−.08, .10	.08	.02, .15	.16	.06, .25	.23	.09, .37
Depressed Affect	Coping Motives (Lagged)	.17	−.02, .36	.08	−.05, .21	−.01	−.11, .09	−.10	−.22, .03	−.19	−.37, −.01
	Consequences (Lagged)	−.08	−.17, −.003	−.04	−.09, .01	−.003	−.05, .05	.04	−.03, .11	.08	−.02, .18

Simple slopes were calculated at different values of CUDIT-R for models with significant cross-level interactions. A significant simple slope indicated that within-person affect was significantly associated with the outcome among individuals with that score on the CUDIT-R. Simple slopes may not differ significantly from one another. See Supplementary Materials section on marginal effects to see marginal effects, which test whether effects of affect on the outcome differ from one another at different levels of CUDIT-R. CUDIT-R = Cannabis Use Disorder Identification Test – Revised. Significant effects are presented in bold.

Figure 3.

Within-person simple slopes at different levels of CUDIT-R for significant interactions. See Table 5 for simple slopes and their significance. CUDIT-R = Cannabis Use Disorder Identification Test – Revised.

CU problem severity also predicted the within-person association between positive affect and enhancement motives. Experiencing more positive affect than usual was associated with a higher likelihood of using cannabis for enhancement reasons at all CUDIT-R scores, but this association was strongest at lower CUDIT-R scores, consistent with hypotheses.

No associations between affect and cannabis consequences were predicted by CU problem severity and neither were any associations between depressed affect and CU, motives, or outcomes. Only two main effects of affect were statistically significant outside of the context of significant moderation effects. Experiencing more positive affect than usual was associated with a higher likelihood of CU and being intoxicated for longer during the same observation among all individuals regardless of CU problem severity.

Moderation of prospective event-level associations by CU problem severity

CU problem severity predicted three prospective event-level associations : depressed affect predicting coping motives and consequences and positive affect predicting duration of intoxication (Table 4). The pattern for the association between depressed affect and coping motives did not match hypotheses. Experiencing more depressed affect than usual was associated with a lower likelihood of using cannabis to cope during the next observation among individuals with CUDIT-R scores of 24 and above. This association was not significant for those with lower CUDIT-R scores.

Table 4.

Within-person prospective direct effect estimates and moderation

			Anxious Affect			Depressed Affect			Positive Affect
Model	Predictor	Outcome	b	95% CI	p	b	95% CI	p	b	95% CI	p
Cannabis Use	Affect	Cannabis Use	−.02	−.13, .09	.66	.05	−.08, .17	.43	−.03	−.14, .08	.65
	CUDIT-R	Affect → Cannabis Use	−.001	−.01, .01	.79	−.003	−.01, .01	.53	−.001	−.01, .01	.89
Coping Motives	Affect	Coping Motives	.10	−.15, .33	.43	.26	.00, .52	.05	-	-	-
	CUDIT-R	Affect → Coping Motives	−.001	−.02, .01	.86	−.02	−.03, −.002	.03	-	-	-
Enhancement Motives	Affect	Enhancement Motives	-	-	-	-	-	-	.06	−.12, .25	.51
	CUDIT-R	Affect → Enhancement Motives	-	-	-	-	-	-	−.004	−.02, .01	.51
Duration Intoxication	Affect	Duration Intoxication	−.02	−.21, .18	.85	.05	−.16, .24	.62	−.13	−.32, .06	.17
	CUDIT-R	Affect → Duration Intoxication	.003	−.01, .02	.67	−.002	−.01, .01	.72	.01	.002, .03	.02
Consequences	Affect	Consequences	.01	−.13, .14	.95	−.13	−.24, −.01	.04	−.04	−.16, .07	.48
	CUDIT-R	Affect → Consequences	.0001	−.01, .01	.93	.01	.001, .02	.046	.004	−.004, .01	.29

Covariates included: within-person (day of EMA period, weekend vs weekday, and morning vs. evening survey); between-person (race/ethnicity, sexual identity, gender identity, age). Associations among variables at the between-person level were also estimated but are not included in tables. See supplementary materials for tables with all parameters. Significant effects are presented in bold. CUDIT-R = Cannabis Use Disorder Identification Test – Revised.

Inconsistent with hypotheses, experiencing more positive affect than usual was associated with being intoxicated for longer during the next observation among individuals with CUDIT-R scores of 14 or higher, while this association was not significant for those with lower CUDIT-R scores. The association between depressed affect and cannabis consequences followed the expected pattern, with depressed affect being associated with experiencing fewer consequences among those with CUDIT-R scores of 6 or lower, while this association was non-significant for those with higher CUDIT-R scores.

Sensitivity Analyses

We conducted sensitivity analyses in which quantity and frequency items were excluded from CUDIT-R scores. The moderation analyses for these symptom-only scores followed the same pattern as those for the full CUDIT-R scores.

Analyses stratified by SGM status produced largely the same pattern of results as those from the full sample. Due to the drop in sample size, several moderations that were significant in the full sample became marginally significant in analyses of SMWGD and nonsignificant for heterosexual women. However, effects were largely of the same magnitude for heterosexual women and SMWGD, indicating that this non-significance was a result of lower power to detect effects among the smaller subsample of heterosexual women.

Discussion

The current study tested hypotheses derived from the multistage model among SMWGD and heterosexual cisgender women. Results provided partial support for hypotheses, with elevated anxious affect concurrently associated with coping motives and longer intoxication but only among those with more severe CU problems. Further, positive affect was concurrently more strongly associated with enhancement motives among those with less severe CU problems. However, many other associations between affect and CU outcomes were not moderated by CU problem severity or displayed patterns of moderation that differed from those hypothesized.

Concurrent event-level associations between anxious affect and cannabis outcomes were most consistently moderated by CU problem severity. When anxious affect was elevated, individuals with more severe CU problems were more likely to use cannabis to cope and were intoxicated for longer. These associations were negative or nonsignificant for those with less severe CU problems. These findings suggest that associations between anxious affect and CU outcomes may depend on the proportion of individuals in a sample with more severe CU problems.

Two prospective associations between depressed affect and CU outcomes were moderated by CU problem severity, displaying divergent patterns. Partially consistent with hypotheses, elevated depressed affect was associated with fewer consequences among individuals with less severe CU problems and this association was non-significant for those with more severe CU problems. Although this association approached significance at high CU problem severity, it did not reach significance. This pattern is similar to that observed for associations between anxious affect and CU outcomes. The other significant moderation involving depressed affect displayed the opposite pattern. Depressed affect only predicted an increased likelihood of coping motives during the next observation among individuals with less severe CU problems. This counterintuitive finding should be examined in future studies to determine whether it is replicated.

Anxious and depressed affect were differentially associated with cannabis outcomes, with anxious affect playing a more central role in concurrent CU while depressed affect may be more important to subsequent CU. This may be due to differences in arousal level associated with anxious versus depressed affect. Anxious affect is a high arousal negative affective state that results in a heightened drive to reduce anxiety, which may contribute to shorter lengths of time between anxious affect and using cannabis to reduce anxiety (Clark & Watson, 1991; Heller & Nitschke, 1998). On the other hand, depressed affect is a low arousal negative affective state that is characterized by lower drive to reduce depressed affect, which may help to explain the longer lag between depressed affect and CU (Clark & Watson, 1991; Heller & Nitschke, 1998). Notably, many studies have examined negative affect as a whole (Linden-Carmichael et al., 2022; Van Doren et al., 2024), combining depressed and anxious affect. Given our different findings for anxious and depressed affect, examining negative affect as a monolith may be problematic.

Similarly, two associations involving positive affect were moderated by CU problem severity – those between positive affect and concurrent enhancement motives and prospective duration of intoxication. Consistent with the multistage model, elevated positive affect was more strongly associated with enhancement motives among those with less severe CU problems; however, positive affect was associated with enhancement motives for the entire sample. In the inverse pattern, positive affect was associated with longer intoxication during the next observation for individuals with more severe CU problems, while this association was not significant for individuals with less severe CU problems. Two concurrent associations between positive affect and cannabis outcomes were significant for the full sample, including positive affect predicting a higher likelihood of CU and a longer intoxication. Overall, this suggests that positive reinforcement remains an important contributor to CU among those with more severe CU problems. These findings are largely consistent with other studies, which have also found that positive reinforcement risk factors are consistently associated with substance use regardless of substance use problem severity (Cho et al., 2019; Dyar & Feinstein, 2025; Dyar & Kaysen, 2023).

Many prospective associations between affect and cannabis outcomes were nonsignificant and not moderated by CU problem severity. This is likely a result of the long lag between observations (~12 hours). We are unable to determine the directionality of significant concurrent effects. Therefore, concurrent associations may reflect affect predicting CU and other outcomes and/or CU contributing to changes in affect. As CU is known to produce changes in affect (Cuttler & Spradlin, 2017; Metrik et al., 2011), it is likely that our concurrent effects include a combination of both. Future research with shorter lags between observations should determine whether our concurrent findings replicate prospectively.

Limitations

The current study should be considered in light of its limitations. Given that the sample was comprised of 18–25 year old females who used cannabis regularly and lived in the US, it is unclear whether findings will generalize to males, other age groups, lower frequencies of CU, or other countries. Second, the multistage model focuses on changes in patterns of CU and related problems over time rather than individual differences. However, because this manuscript did not include repeated assessments of CUDIT-R, we were unable to examined changes in the severity of CU problems. Third, the current study examined only main effects of positive and negative affect. However, positive and negative affect do not occur in isolation, so future research should consider affect profiles at the daily level more holistically via interactions between positive and negative affect or latent profile analyses. Further, these analyses focused on a subset of the dimensions of affect that have been linked with CU (Emery et al., 2023). Fourth, this study asked about affect since the last survey, rather than focusing on momentary affect. Participants may have varied in how accurate they were in reporting affect over such a relatively long period. Fifth, this study examined CU problem severity rather than CU disorder symptomology and as a result provides an indirect test of the multistage model. Finally, although sensitivity analyses suggest that the pattern of results were similar for SMWGD and cisgender, heterosexual women, we did not statistically compare results for these two groups due to low power for the three-way interactions required to do so.

Conclusion

The current study found some support for hypotheses derived from the multistage model. Consistent with hypotheses, anxious affect was associated with coping motives and being intoxicated for longer among individuals with more severe CU problems and positive affect was more strongly associated with enhancement motives among those with less severe CU problems. However, findings also suggest that positive reinforcement may continue to drive CU among those with more severe CU problems, which is not consistent with the multistage model.