Craving is impermanent and it matters: Investigating craving and cannabis use among young adults with problematic use interested in reducing use

Abstract

Rates of problematic cannabis use have nearly doubled over the last decade, and peak onset for cannabis use disorders occurs during young adulthood. Craving for cannabis is hypothesized to be an important factor that maintains cannabis use among people who desire to stop or reduce their use, including many young adults. Previous studies that used single timepoint assessment methods to demonstrate a link between craving and cannabis use have found mixed predictive utility of measurements. The impermanent, or time-varying nature of craving may be responsible for mixed findings, leading to inaccuracies in retrospective recall and greater difficulty measuring craving and detecting its association with cannabis use. The current study compared intensive longitudinal assessments and single timepoint assessments predicting cannabis use among young adults with problematic cannabis use who reported a desire to reduce their use. Participants (N=80) completed a baseline craving questionnaire and intensive longitudinal assessments of momentary craving and cannabis use up to four times per day for 14 days. Results suggested that averaged momentary craving predicted cannabis use above-and-beyond craving measured at baseline. An increase of one SD above the sample-mean for averaged momentary craving increased the probability of cannabis use by 367%, while a one SD increase in baseline craving was only associated with a 49% increase. Findings suggest that asking young adults who want to cut back on their cannabis use about their craving at a single timepoint may not be as clinically useful as tracking cravings repeatedly in near real-time and in ecologically valid contexts.

1. Introduction

Problematic cannabis use is a growing public health concern (Butterworth et al., 2014; Haberstick et al., 2014) and is linked to poorer mental health and other adverse consequences (Arria, Caldeira, Bugbee, Vincent, and O’Grady, 2015; Keith, Hart, McNeil, Silver, and Goodwin, 2015; Keith et al., 2015; Lyvers, Jamieson, and Thorberg, 2013; Phillips, Phillips, Lalonde, and Tormohlen, 2015). Epidemiology studies indicate that problematic cannabis use peaks during young adulthood, which has been identified as an important developmental period for early intervention (Caulkins et al., 2015; Farmer et al., 2015; Hasin et al., 2015). Craving is hypothesized to function as a core element in the development and maintenance of cannabis misuse, and there is a strong risk of relapse to problematic use associated with craving (Serre et al., 2015).

1.1. Craving for Cannabis and Models of Misuse

Craving is a core feature of cannabis use disorder (DSM-5, 2013) and there is growing evidence that craving is associated with an increased probability of subsequent cannabis use among young adults who frequently use cannabis (Buckner, Crosby, Silgado, Wonderlich, and Schmidt, 2012; Buckner et al., 2015; Buckner, Zvolensky, and Ecker, 2013; Phillips et al., 2015). The impact of craving on subsequent use may help explain why some young adults continue to use cannabis despite having the desire to stop or reduce their use (Buckner et al., 2015; Phillips et al., 2015). Craving, and situations that elicit craving, have long been hypothesized to function as a driving force in substance use disorders and relapse (Brewer et al., 2014; Witkiewitz and Marlatt, 2007). However, single timepoint measures of craving often have relatively modest associations with substance use, with many studies showing mixed results. For example, one systematic review noted that mixed results have “[called] into question the value of craving as a treatment target” (Wray et al., 2013). While mixed results have led some to question basic theory, others have begun to question whether mixed findings in the literature may be due to measures relying on single timepoint assessment of a phenomenon that is so time-varying (Kavanagh et al., 2013; Shiffman, 2009; Drummond, 2001; Merikle, 1999).

Ecological momentary assessment (EMA) can potentially improve the detection of links between craving and substance use. A recent systematic review of EMA focusing on craving and substance use pointed toward a strong predictive association between craving and use (Serre et al., 2015). Examination of the momentary relationship between craving and illicit substance use among young adults is limited, however, with a small number of studies investigating momentary craving and cannabis use (Phillips et al., 2015). In a study with college student cannabis users, a temporal association between craving and cannabis use was found, such that craving at one time point significantly predicted quantity of time spent using cannabis, and frequency of use reported at the subsequent time point (Phillips et al., 2015). Another study, with non-treatment-seeking adolescent cannabis users, found that momentary craving was associated with subsequent cannabis use (Buckner et al., 2015). Thus, a small but growing body of evidence indicates that momentary craving may be an important risk factor for young adults at risk of developing a cannabis use disorder, as it has for individuals with more severe substance use disorders (Donovan and Witkiewitz, 2012; Moore et al., 2014; Witkiewitz, Lustyk, et al., 2013). However, no studies using intensive longitudinal methods have investigated the association between craving and use among young adults with problem cannabis use who also desire to reduce their cannabis use.

1.2. Current Study

The purpose of the current study was to investigate the association between craving and use over time using cross-sectional and intensive longitudinal methods for young adults with cannabis related problems interested in reducing their use. We investigated this question specifically among people who indicated an intention to control or limit their use and were having trouble being successful to focus on populations who may benefit most from cannabis-related interventions. We will refer to this population as “treatment-adjacent” due to their hazardous use, desire to reduce, and lack of current engagement in treatment. The current study also explores the association between different facets of craving and probability of subsequent use, and directly compares EMA versus cross-sectional measures of craving and their associations with subsequent cannabis use to help clarify the mixed results from previous studies. We hypothesized that both measures of craving would be positively associated with subsequent use, but that the EMA measures would more strongly predict subsequent cannabis use better than the single timepoint measure.

2. Material and methods

An intensive repeated measure design was used to investigate the association between craving and use for treatment-adjacent young adults (N=80). An initial baseline phase of data collection was followed by a two-week period of intensive longitudinal data collection using ecological momentary assessment methodology (Laird et al., 1992; Shiffman, 2009; Walls and Schafer, 2006).

2.1. Participants

Participants were recruited from a large public university in the Northwestern United States. Eligible participants were (1) between ages 18-29, (2) reported using cannabis an average of at least two days per week during the previous month (3) endorsed a Contemplation or Action stage of change (Rollnick et al., 1992), (4) had Cannabis Use Disorder Identification Test - Revised (CUDIT-R) scores that indicated an increased risk of cannabis related problems (Schultz et al., 2019), and (5) owned a smartphone.

2.2. Procedures and Retention

Interested participants completed a brief screening questionnaire to assess past month level of cannabis use, readiness to change and cannabis-related problems. Eligible participants were invited to complete a baseline assessment in the lab, which included single timepoint craving measurements. While in the lab, participants also downloaded the EMA application ‘mobile Ecological Momentary Assessment’ (mEMA: Spook, Paulussen, Kok, and van Empelen, 2013) to their phone, and received instructions on how to complete assessments. Following recommended guidelines for signal contingent EMA studies (Shiffman, 2009), surveys were sent at random points during three discrete four-hour time-windows (Morning: 10am-12pm, Afternoon: 3pm-5pm, Evening: 8pm-10pm), and participants had 2 hours to complete the survey.

2.3. Measures

2.3.1. Screening measures:

Age, sex, average days of cannabis use per week, cannabis-related problems, and readiness to change were all assessed during screening.

Cannabis-related problems.

Problems related to cannabis use were assessed using the Cannabis Use Disorder Identification Test – Revised (CUDIT-R, Adamson et al., 2010). The CUDIT-R is a brief 8-item measure of cannabis misuse during the previous 6 months with demonstrated internal validity, sensitivity, and specificity, as well as concurrent and discriminant validity. The internal consistency of the measure in the current sample (α=0.66) was below that reported in the original study (0.91; Adamson et al., 2010). However, it was similar to that obtained by a study of young adults recruited to participate in an online survey (0.66; Ramo et al., 2012). A cut off of 6 optimizes the sensitivity and specifity to distinguish between college students with and without harmful or problematic cannabis use (Schultz et al., 2019), and 13 or more indicates a possible cannabis use disorder (Adamson et al., 2010).

Readiness to change.

The Readiness to Change Questionnaire (RCQ: Rollnick et al., 1992) is a short, 12-item measure used to assess stages of change on three dimensions (pre-contemplation, contemplation, action) in relation to substance use. Summary scores for each of the dimensions range from −3 to 3, and the highest-scoring dimension determines stage of change. The internal consistency in the current sample (pre-contemplation, α =0.72; contemplation, α =0.73; action, α =0.75) was similar to that reported in the original psychometric study (α =0.73, α =0.80, α =0.85; Rollnick et al., 1992).

2.3.2. Baseline assessment measures:

Craving.

Craving at baseline was assessed using the Marijuana Craving Questionnaire (MCQ: Heishman, Singleton, and Liguori, 2001). The MCQ is a 12-item self-report measure of four craving-related factors: compulsivity (an inability to control use, α=0.67), emotionality (use to produce relief from withdrawal or negative affect, α=0.76), expectancy (expectation of a positive outcome due to use, α=0.60), and purposefulness (the intention or plan to generate a positive outcome from use, α=0.80). Internal consistencies for the MCQ subscales were similar to those reported in the original psychometric study for the scale (α=0.82, α=0.78, α=0.55, α=0.68; Heishman et al., 2001).

2.3.3. Ecological momentary assessment items:

Momentary craving was assessed with two items drawn from existing craving measures to assess thoughts about cannabis and urges or desire to use cannabis (Flannery, Volpicelli, and Pettinati, 1999). Momentary thoughts were assessed using the following item: “How much were you thinking about using marijuana or how it would make you feel?” Momentary urges were assessed with the item: “When you wanted to use marijuana the most, how strongly did you want to use?” Responses were on an 11-point visual sliding scale from 0 or “None” to 10 or “Most ever.”

Cannabis use was assessed with one EMA item, “Since the last survey, approximately how many times did you use marijuana?” Use within the subsequent 0-4 hour period, and reported on the same survey period was considered “proximal”, and use reported on the subsequent survey (up to 9 hours later) was considered “distal”. Due to the relatively extended period of time that elapsed between the previous night and the following morning, all morning cannabis use assessments (809 cases, or 21.7% of total cases) were excluded from the distal use variable. All cases where participants indicated that their craving had reached its peak after using cannabis (197 cases, or 5.3% of total cases) were excluded from the proximal use variable. For the proximal use outcome, including only cases where respondents had reported use after their peak craving enabled the most time-sensitive analysis possible and did not require time-lagging to retain temporal precedence.

2.4. Data Analytic Plan

Multilevel modeling (MLM) was used to explore the associations between craving and use while accounting for nesting of repeated measures within participants (Snijders and Bosker, 2011). MLMs were fitted using the glmer function in the lme4 package (Bates et al., 2015) in R (R Core Team, 2013) using a logistic response function to model the binary outcomes reflecting cannabis use (1) or non-use (0).

The prospective associations between craving and use were investigated using the four subscales from the MCQ measured at baseline and the two EMA craving predictors. These predictors predicted proximal (not time-lagged) and distal (time-lagged) cannabis use outcomes in two separate MLMs. Estimates of mean levels of craving for each individual (“person-mean” or between-subjects effect) and momentary deviations from each person’s mean level (“momentary” or within-subjects effect) were each computed and entered as predictors. The momentary craving items were used to create two variables: person-mean levels of craving for each individual (between-subjects effect) and momentary deviations from each person’s mean level (within-subjects effect). All craving predictors were standardized, such that between-subjects variables were centered on the population mean and on a population-standard deviation scale, and within-subjects variables were centered by person-mean and on a person-standard deviation scale.

Person-mean craving was included in the final model to clarify the different elements of the momentary data. The person-mean variable captures a person’s average level of craving relative to the sample average and allows for a more direct comparison between EMA variables and single timepoint variables. Momentary craving was included in the final model to examine how momentary fluctuations in craving predict use. Time was entered as a categorical predictor to control for the effect of time of day on probability of use. The four baseline craving measures (MCQ subscales) were centered at the sample mean and included in the final model to investigate the predictive utility of single timepoint craving assessment relative to EMA.

Six participants were excluded from the sample used for final analyses (N=80) because they did not complete any EMA surveys due to technological difficulties. Survey completion rates were examined in relation to all study variables to detect potential systematic reasons for missing data (Snijders and Bosker, 2011). No study variables were significant predictors of missingness. Rates of missing data in the current sample were similar to those reported in other EMA studies evaluating substance use in young adults (Buckner et al., 2012; Phillips, Phillips, Lalonde, and Dykema, 2014). Of the 4,816 EMA surveys across all participants, 78% were completed. Maximum likelihood estimation allowed our models to include all available observations even when some missing data was present, while also reducing bias and increasing precision of parameter estimates (Hallgren and Witkiewitz, 2013).

MLMs included fixed effects for momentary (level-1) and person-level (level-2) craving measures. Random effects were included for subject-level intercepts and the two momentary craving predictors, which allows the intercept and slopes of these effects to vary between participants (i.e., different participants having different associations between momentary craving and cannabis use).

3. Results

Means and standard deviations for all study variables are available in Table 1. Participants were 41% women and the mean age was 19.67 (SD=2.20) years old. Cannabis use occurred 3.76 (SD=1.74) days per week on average. The average CUDIT-R score was 12.79 (SD=4.56), with 43% of the sample at or above the 13-point threshold for probable Cannabis Use Disorder (Adamson et al., 2010). On average, the sample endorsed the Action stage of change (M=0.92, SD=3.61) more strongly than the Contemplation stage (M=0.07, SD=3.48) or the Precontemplation stage (M=−.58, SD=3.10). The range for both momentary urges and momentary thoughts was 0-10 and the median was 1 for both variables.

Table 1.

Sample size, means, and standard deviations.

Variable	n	M	SD	1	2	3	4	5	6
Proximal use	3728	0.28	0.45
Distal use	2868	0.32	0.47
Momentary urges	3700	1.92	2.44
Momentary thoughts	3702	1.91	2.33
1. Person-mean urges	80	1.91	1.33
2. Person-mean thoughts	80	1.90	1.23	.95***
3. Compulsivity	86	1.68	0.94	.29**	.26*
4. Emotionality	86	3.06	1.51	.31**	.27*	.42***
5. Expectancy	86	3.70	1.47	.47***	.45***	.26*	.72***
6. Purposefulness	86	3.89	1.70	.44***	.41***	.31**	.64***	.67***
7. Probability of use	80	0.25	0.22	.59***	.53***	.19	.15	.22*	.36***

Note. Momentary craving (urges and thoughts) and use were assessed using EMA. In total, 78% of EMA surveys were completed. Person-mean scores were Level 2 variables calculated from momentary craving EMA data, standardized within-person. Variables 3-6 were Level 2 variables measured at baseline using the Marijuana Craving Questionnaire. Probability of use was calculated by averaging proximal use scores for each participant from EMA data.

^*= p < .05,

^**= p < .01,

^***= p < .001.

Correlations between the proportion of EMA assessments with any cannabis use (probability of cannabis use) and the six person-level craving measures are shown in Table 1. Person-mean urges and person-mean thoughts (both measured using EMA) were strongly associated with the proportion of assessments with cannabis use (r=.59 and r=.53, ps<.001). The purposefulness and expectancy subscales of the baseline MCQ measure were also positively associated with cannabis use (r=.36, p<.001, and r=.22, p<.05 respectively). The compulsivity and emotionality subscales of the baseline MCQ measure were not significantly associated with probability of cannabis use.

3.1. Proximal Cannabis Use

As hypothesized, each of the two momentary craving predictors was uniquely and positively associated with proximal cannabis use, controlling for time of day, person-level craving measures, and the other momentary craving predictor (see Table 2). Two of the person-level craving measures also significantly predicted proximal use: person-mean urges measured via EMA and baseline purposefulness measured via the MCQ. For the momentary predictors, odds ratios (ORs) represent the increased odds of cannabis use given a one person SD increase in the predictor above a person’s mean. For person-level measures, ORs represent the change in probability of cannabis use given a one-SD increase above the sample mean.

Table 2.

Final multilevel model results, clustered by participant, predicting proximal use.

Predictor	b	SE	p	OR	95% CI
(Intercept)	−3.05	0.26	<.001	0.05***	[0.03, 0.08]
Afternoon	0.18	0.22	0.409	1.20	[0.78, 1.84]
Evening	0.99	0.22	<.001	2.68***	[1.76, 4.10]
Night	2.18	0.21	<.001	8.83***	[5.86, 13.30]
Momentary urges	1.21	0.14	<.001	3.34***	[2.55, 4.37]
Person-mean urges	1.54	0.47	0.001	4.67**	[1.85, 11.80]
Momentary thoughts	0.51	0.12	<.001	1.66***	[1.32, 2.11]
Person-mean thoughts	−0.78	0.50	0.121	0.46	[0.17, 1.23]
Compulsivity	0.02	0.21	0.942	1.02	[0.67, 1.55]
Emotionality	−0.17	0.20	0.401	0.85	[0.58, 1.25]
Expectancy	−0.25	0.21	0.237	0.78	[0.52, 1.18]
Purposefulness	0.40	0.15	0.008	1.49**	[1.11, 2.01]

Note. Proximal use was cannabis use that was reported at the same survey prompt, but as having occurred after craving was strongest. Time of day was treated as a categorical variable, with Morning as the comparator. Craving-related urges and thoughts were assessed using a single item on each EMA survey. Momentary variables were Level 1 momentary reports. Person-mean variables were Level 2 variables created using the aggregated mean of each person on the momentary item.

^*= p < .05,

^**= p < .01,

^***= p < .001.

For example, when all other predictors are held constant, a one SD increase in momentary urges above an individual’s mean level of momentary urges was associated with a 3.34-fold increase in the odds of cannabis use within the same EMA period (OR=3.34, 95% CI [2.55, 4.37]). Similarly, when all other predictors are held constant, a one SD increase in cannabis-related thoughts above an individual’s mean level of cannabis-related thoughts was associated with a 1.65-fold increase in odds of cannabis use during the next four hours (OR=1.66, 95% CI [1.32, 2.11]). For the Level 2 person-mean urges predictor, a one population-SD increase above the population-mean was associated with a 4.67-fold increase in odds of cannabis use (OR=4.67, 95% CI [1.85, 11.80]).

3.2. Distal Cannabis Use

Only person-mean urges derived from EMA were associated with distal cannabis use (see Table 3). When all other predictors were held constant, a one SD increase in person-mean urges above the sample mean was associated with a 3.60-fold increase in odds of cannabis use (OR=3.60, 95% CI [1.87, 6.95]).

Table 3.

Final multilevel model results, clustered by participant, predicting distal use.

Predictor	b	SE	p	OR	95% CI
(Intercept)	−1.98	0.18	<.001	0.14***	[0.10, 0.20]
Afternoon	0.95	0.15	<.001	2.58***	[1.91, 3.49]
Evening	1.92	0.16	<.001	6.79***	[4.98, 9.24]
Momentary urges	0.13	0.09	0.16	1.14	[0.95, 1.37]
Person-mean urges	1.28	0.34	<.001	3.60***	[1.87, 6.95]
Momentary thoughts	0.12	0.09	0.182	1.12	[0.95, 1.33]
Person-mean thoughts	−0.69	0.35	0.050	0.50 *	[0.25, 1.00]
Compulsivity	0.06	0.16	0.713	1.06	[0.78, 1.44]
Emotionality	−0.02	0.15	0.900	0.98	[0.74, 1.31]
Expectancy	−0.22	0.15	0.152	0.81	[0.60, 1.08]
Purposefulness	0.11	0.12	0.351	1.11	[0.89, 1.40]

Note. Distal use was cannabis use that was reported at the subsequent survey prompt. Time of day was treated as a categorical variable, with Morning as the comparator. Craving-related urges and thoughts were assessed using a single item on each EMA survey. Momentary variables were Level 1 momentary reports. Person-mean variables were Level 2 variables created using the aggregated mean of each person on the momentary item.

^*= p < .05,

^**=p < .01,

^***= p < .001.

4. Discussion

4.1. Summary of Current Findings

Evidence from the current study demonstrates that craving is an experience that is positively associated with subsequent cannabis use behavior for treatment-adjacent young adult cannabis users who desire to reduce their use of cannabis. Although craving has consistently played a fundamental role in theories that purport to predict substance use behavior, attempts to measure this phenomenon have not consistently demonstrated predictive validity. The utility of single timepoint baseline craving measures as predictors of subsequent substance use behavior has been in doubt, with systematic reviews reporting mixed results (Wray et al., 2013). Baseline craving and momentary thoughts and urges were associated with increased odds of subsequent cannabis use among individuals with a desire to cut down or quit using during a crucial period of development.

Findings from the current study substantiate concerns regarding the relative utility of single timepoint baseline measurements of craving. Zero-order correlations suggested that the momentary assessments were more strongly associated with use than the baseline measurement. In the MLMs, when all craving predictors were included in a single model, momentary measurement of craving provided unique predictive utility over and above baseline craving measures. The positive association between craving and use in the MLMs suggested that both momentary craving and a person’s average momentary craving are robust predictors of proximal use while controlling for baseline craving. That momentary craving was not associated with distal use serves to underscore the important role of temporal proximity in the assessment of craving and use.

While two subscales of the MCQ (Expectancies and Purposefulness) were associated with probability of use, the strength of the correlation was smaller than the EMA measures. In the final model predicting use, only one of the four subscales of the MCQ (Purposefulness) provided unique additive utility. The other subscales (Expectancies, Compulsivity, Emotionality) were all positively correlated with the EMA measures of craving, but appear not to add anything unique when predicting use. It seems likely that the MCQ is capturing some elements of the experience of craving, but that these elements are not predictive of use. Thus, when investigating mechanisms of action and treatment targets, EMA may be more useful than the MCQ.

The application of intensive longitudinal assessment methods such as EMA more directly maps onto theories that describe craving as a dynamic and time-varying experience, as opposed to an event that can be accurately recalled and reported on during a single timepoint baseline assessment (Buckner et al., 2015a; Phillips et al., 2015; Serre et al., 2015). The strong positive association between craving and proximal cannabis use confirms the importance of using assessment methods that can capture the dynamic and time-varying aspects of craving in order to better understand how craving impacts cannabis use, cessation, and relapse.

In the current study, urges to use cannabis were a stronger predictor of subsequent use than thoughts. Given this relative difference, if investigators intend to examine craving as a predictor of use through a single EMA item, then it is recommended that momentary urges be assessed. While momentary thoughts did provide additional unique predictive utility in the proximal model, person-mean thoughts was negatively associated with distal use. Participants who reported higher mean levels of thoughts about using relative to the sample mean had a lower probability of using cannabis. It may be that limited availability for some participants, or some other third variable, led to an increase in thoughts about use and a decrease in actual use. However, it is unclear why decreased availability, or another third variable, would have increased thoughts about using but not urges to use. This was an unexpected finding, and deserves further research.

4.2. Clinical Implications

Findings from the current study provide evidence that craving is an important target for treatment-adjacent young adult cannabis users. Clinicians may gain more clinically relevant data from clients by encouraging frequent self-monitoring of cravings for cannabis (i.e., multiple times per day) as opposed to single timepoint assessments. Such momentary measures of craving could provide valid and useful clinical data to inform measurement-based care and reduce individual clients’ likelihood of using cannabis. Ecological momentary assessment for measurement-based care may help to more accurately gauge the effectiveness of interventions that aim to reduce craving directly or reduce the impact of craving on cannabis use (e.g., through craving management skills). Importantly, clinicians and clients may obtain more valid treatment-related information if clients answer fewer craving-related questions more frequently than if they answer longer, psychometrically-validated craving-related questionnaires less frequently. The development of methods to intervene in the moment when craving is high may enhance treatment effects and increase likelihood of achieving a person’s goals for treatment.

4.3. Limitations

There are important limitations to acknowledge when interpreting the results. Firstly, data from the current study has limited generalizability outside of college student cannabis users aged 19-29 who are interested in reducing their use. Research with other populations is needed to increase generalizability.

Secondly, substance use research using EMA frequently faces difficulties with noncompliance and missing data (Shiffman, 2009), and the current study did not evade these challenges. While appropriate steps were taken to identify missing data patterns and evaluate the likelihood that data were missing at random (Little and Rubin, 1987), missingness limits the strength of the conclusions that can be drawn from results.

Thirdly, there was some information loss in the proximal outcome by excluding cases where craving peaked following use. Respondents did occasionally experience peak craving after using, and the model predicting proximal use did not account for this phenomenon. However, the model predicting distal use did not have this limitation, and the pattern of results was similar to the proximal use outcome. Thus, it appears unlikely that our decision to exclude such cases biased the results, and doing so allowed us to retain temporal precedence for the proximal use outcome.

5. Conclusions

Findings from the current study suggest that craving is a powerful predictor of cannabis use among treatment adjacent young adults. The strength of the association between craving and use appears to be highly sensitive to the timing with which craving and use are measured. When targeting craving in treatment contexts, it would behoove researchers and clinicians to enhance ecological and temporal validity by measuring craving in a manner that is consistent with the dynamic and time-varying manner in which people often experience it. Efforts to evaluate the effectiveness of treatments that target craving may benefit from applying intensive longitudinal assessment methods to enhance temporal resolution and capture a more valid predictor of substance use. By increasing the validity of craving measurements, we can improve our understanding of how craving is related to and has an impact on the development, maintenance, and recurrence of substance use disorders.

Highlights.

• Cravings were positively associated with probability of cannabis use

• EMA measurements of craving were uniquely associated with probability of use

• EMA measurement of craving may be useful for clinicians and researchers