Understanding marijuana use initiation vs. frequency of use on risk for hallucinogen use: testing for mediating mechanisms underlying the gateway hypothesis

ABSTRACT

The gateway hypothesis posits that marijuana use initiation may lead to neurological changes predisposing individuals to escalation to harder drug use later on. However, there is limited research, which has examined marijuana as a gateway drug leading to hallucinogen use and examination of mechanisms linking the two drug use behaviors. This study examined marijuana use as a predictor of hallucinogen use and tested for mediation via social, affective, and cognitive mechanisms. The Pathways to Desistance data were analyzed. This is a longitudinal dataset comprised a sample of 1,354 justice-involved youth. Path analyses were used to test for direct and indirect effects of interest. Findings indicated that elevated marijuana use frequency prior to baseline was a significant predictor of increased log-odds of later hallucinogen use. None of sensation-seeking, impulse control, nor deviant peer association significantly mediated this relationship. Ever having used marijuana prior to baseline was not significantly related to later hallucinogen use risk and no mediation effects were identified for this independent variable. Results indicated that higher frequency marijuana use may result in gateway effects for later hallucinogen use. Both marijuana use variables were also statistically significant predictors of increased sensation-seeking, diminished impulse control, and greater deviant peer association.

KEYWORDS:

• Gateway Hypothesis
• hallucinogens
• marijuana
• justice-Involved Youth

Introduction

Marijuana has long been predicted to be a ‘gateway drug’ whose use may increase risk for an individual to initiate the use of other illicit drugs (Kandel & Faust, 1975; Kandel, Yamaguchi, & Chen, 1992). Indeed, research has indicated that marijuana use does generally precede the use of illicit drugs in terms of temporal order of initiation (Cappelli, Ames, Xie, Pike, & Stacy, 2021; Kandel, Yamaguchi, & Chen, 1992; Keyes, Rutherford, & Miech, 2019). While there have been a range of mechanisms posited to explain this relationship, there remains limited research on focused on identifying these mechanisms linking marijuana use to later hallucinogen use initiation (e.g.: psilocybin mushrooms, LSD/acid, ayahuasca, etc.). This omission is problematic given the potential legal and long-term health consequences of hallucinogen use (Martinotti et al., 2018; Salas-Wright et al., 2021). Considering these health and abuse issues, identifying risk factors underlying initiation of use of this class of drugs characterized by euphoria, hallucination, and substantial alterations in thought, mood, and perception should be a paramount concern for criminal justice and public health researchers. Another key concern here relates to potential differences that marijuana use initiation vs. high frequency marijuana use may have for gateway effects on hallucinogen use. The present study sought to address these gaps in the literature by examining social, affective, and cognitive mechanisms linking these different attributes of marijuana use to risk for later hallucinogen use among a sample of justice-involved youth (JIY). Figure 1 provides a visual representation of the proposed mediation model.

Figure 1. Mediation Model linking marijuana use to hallucinogen Use.

The gateway hypothesis

As noted above, the gateway hypothesis broadly posits that marijuana use a gateway drug that may increase risk for the use of ‘hard drugs’ later on (Kandel & Faust, 1975; Kandel, Yamaguchi, & Chen, 1992). Some research has indicated that marijuana use may impact cognitive development, thus, priming users for later drug use and increasing risk for the behaviors, whereas other studies have focused on social mechanisms by which this may occur (Anderson, Rizzo, Block, Pearlson, & O’Leary, 2010; Jacobus & F Tapert, 2014; Lisdahl, Gilbart, Wright, & Shollenbarger, 2013; Mayet, Legleye, Beck, Falissard, & Chau, 2016; Nkansah-Amankra, 2020). Evidence overall for the gateway hypothesis overall also remains mixed (Kononoff et al., 2018; Melas et al., 2018; O’Brien, Comment, Liang, & Anthony, 2012). For example, Kononoff et al. (2018) found that exposure to a cannabinoid receptor agonist was associated with lower risk of cocaine self-administration later in life among rats. That said, Melas et al. (2018) found that cannabinoid exposure in rats may indeed result in sensitization to cocaine, but only among adolescent rats. While the obvious limitation of the highlighted research is that these studies utilized non-human samples, but it does shed some doubt on the posited mechanisms by which marijuana should act as a gateway drug. The other issue here is that studies have yet to examine these same mechanisms for hallucinogens specifically.

While evidence for the gateway hypothesis has been mixed overall, there is some limited evidence that does indicate that marijuana use does generally temporally precede initiation of hallucinogen use. For example, Mayet, Legleye, Beck, Falissard, and Chau (2016) observed that hallucinogen use initiation occurred an average of about 2.5 years after average marijuana use initiation and this study also identified a temporally ordered pathway from marijuana use to other illicit drug use in a manner consistent with the postulations of the gateway hypothesis. However, the gateway pattern identified here collapsed all other illicit drug use into a single variable, making it impossible to disentangle the specific gateway effects of marijuana on hallucinogen use risk. Such collapsing of other illicit drugs into a single category here is relatively commonplace in the literature testing the gateway hypothesis (Delgado-Lobete et al., 2020; Gallegos, Zaring-Hinkle, Wang, & Bray, 2021; Sabia, Dave, Alotaibi, & Rees, 2021). While small N’s of hallucinogen users makes this practice a reasonable choice, this has resulted in a dearth of research which has solely examined hallucinogen use within the gateway hypothesis framework, indicating a need for additional research on the specific causal pathway of interest here. That said, there still remains reason to believe that marijuana use may act as a gateway drug in this regard. Quednow et al. (2022), for example, reports similar findings in terms of age of onset for marijuana compared to hallucinogens also.

Beyond the research indicating the average difference in age of initiation of use between marijuana and hallucinogens, other research has also indicated that very few individuals report their initial drug use to be hallucinogens (1.19%) compared to that of marijuana (68.5%) (Lin et al., 2020). Again though, research on the temporally ordered pathway running from marijuana use to hallucinogen use initiation remains understudied and in need of focused research which establishes the temporal ordering between the two substances, as well as identifies mechanisms underlying use.

While marijuana use initiation is a key component of the gateway hypothesis, there remains a dearth of research which has examined potential gateway effect of high frequency marijuana use. It may be that initiation is only one aspect of this gateway phenomenon and increased use of marijuana may actually have a greater gateway effect. Indeed, prior research has indicated that marijuana use is associated with a range of different outcomes related to dysfunctional cognitive development (Frolli et al., 2021; Ho et al., 2022; Paige & Colder, 2020), a key set of mechanisms posited to underpin the gateway effect. Elevated frequency of use may also necessitate greater engagement with black market networks in order to facilitate chronic use. In this way, more frequent users may become more embedded in these networks and have increased association with deviant peers that may result in increased risk for trying harder drugs like hallucinogens. For these reasons, it is necessary to not only examine initiation of marijuana use, but also to examine how elevated frequency of use may impact hallucinogen use risk. Beyond this, there also remains a dearth of research examining mediators of the relationship between marijuana use and hallucinogen use. This study sought to examine several mediating mechanisms that may explain this relationship so that limited resources may be more effectively targeted towards mechanisms that will have the greatest impact on limiting the potential gateway effects of marijuana.

Social, affective, and cognitive mediation of gateway effects

While cognitive, affective, and peer pathways have been posited in the past, specifying these distinct pathways clearly may help to make sense of the manner in which marijuana use influences risk for later hallucinogen use. In terms of cognition, the dual systems model may provide a useful framework for understanding the gateway effects of interest. This framework focuses on the dual development of impulse control and sensation-seeking during adolescence and emerging adulthood (Steinberg, 2010; Steinberg et al., 2008). Sensation-seeking refers to the drive to seek out novel, thrilling, and exciting experiences; such as substance use. Impulse control refers to the capacity of an individual to stop and consider the potential consequences of an action before engagement. While impulse control provides the necessary capacity to limit an individual’s sensation-seeking drive towards behaviors like substance use, the relevant regions of the brain governing impulse control often are some of the last to reach full maturity (Spencer-Smith & Anderson, 2009). The relevant regions of the brain governing sensation-seeking are instead develop rapidly following puberty and result in novel surges of dopamine in response to exciting and thrilling stimuli (Doremus-Fitzwater, Varlinskaya, & Spear, 2010), thus reinforcing the drive to seek out such experiences. As such, adolescents generally develop the drive to seek out behaviors like hallucinogen use long before they fully develop the capacity to stop and consider the potential risks involved with the behavior. These mechanisms are posited to be the reason that risk for engagement in antisocial behavior tends to peak during this period of the life-course (Steinberg, 2010; Steinberg et al., 2008).

Research has indicated that initiation of marijuana use during adolescence may result in diminished impulse control and may also result in dysfunctional development of reward systems associated with dopamine (Jacobus & F Tapert, 2014; Lubman, Cheetham, & Yücel, 2015; Renard et al., 2017; Shen, 2020). For the latter, this may result in increased sensation-seeking, as individuals develop a stronger drive toward risky behaviors like hallucinogen use in order to flood the brain with dopamine. Because both elevated sensation-seeking and diminished impulse control are risk factors for hallucinogen use (García-Marchena et al., 2018; O’Connor, Aston-Jones, & James, 2021; Wojciechowski, 2021), these constructs would seem to provide potential mechanisms mediating the relationship between marijuana use initiation of marijuana use and later initiation of hallucinogen use. While initiation of marijuana use may certainly result in this sort of dysfunctional cognitive development, it begs the question of what the effects of heavier use may have on these processes also. More frequent marijuana use during adolescence has been demonstrated to have a large effect on cognitive functioning. Frolli et al. (2021) found that chronic marijuana users demonstrated greater cognitive dysfunction compared to occasional users and abstinent controls. This then may indicate that more frequent marijuana use during this period of the life-course may have an outsized effect on cognition, resulting in greater risk for hallucinogen use do to the corresponding larger impact on impulse control and sensation-seeking.

Beyond cognition, research has also indicated that chronic marijuana use may have an impact on mental health (Guttmannova et al., 2017; Memedovich, Dowsett, Spackman, Noseworthy, & Clement, 2018; Thompson, Merrin, Ames, & Leadbeater, 2018). Some research has suggested that chronic marijuana use may exacerbate depression and related mental health issues, though this link remains inconclusive (Memedovich, Dowsett, Spackman, Noseworthy, & Clement, 2018). That said, depression has also been found to be associated with increased hallucinogen use risk (Grant, Lust, & Chamberlain, 2019), though it should be noted that hallucinogens are also being tested for their efficacy in treating such conditions also. It may be that even if cognitive factors do not provide a gateway mechanism that marijuana use may impact emotional/affective symptoms and development of these symptoms. This would seem to provide a potential alternative domain through which elevated marijuana use during adolescence functions to impact hallucinogen use risk through changes in brain functioning.

Researchers also argue that the observed associations in the relationship between marijuana use and later drug use are actually due to social mechanisms. Rather than having a distinct causal impact on harder drug use through cognition or affect, this branch of science argues that this relationship is potentially due to involvement with peers who may facilitate other drug use through influence and by offering channels through which an individual may actually access these drugs (Gallegos, Zaring-Hinkle, Wang, & Bray, 2021; Goode, 1970; Tarter, Vanyukov, Kirisci, Reynolds, & Clark, 2006). If this is the case, then youth who have initiated marijuana use may be at increased risk for progression to later hallucinogen simply by virtue of having increased capacity to access the drug through their known associates, whereas an individual without such associates may lack this capacity. While limited research has been conducted in this specific area, Wilcox, Wagner, and Anthony (2002) did find that individuals who initiated marijuana use were at greater risk for hallucinogen use via the mechanism of having the opportunity to use; a finding that is consistent with this aspect of increased access. Beyond access, it is also well established that having peers involved in antisocial behavior presents elevated risk for one’s own drug use through mechanisms of differential association, socialization, and influence (Hahlbeck & Vito, 2022; Kruis, Seo, & Kim, 2020; Rocheleau, Vito, & Intravia, 2020). This is consistent with Akers’ (Akers, 1973) social learning theory which posits that exposure to definitions favorable toward criminal behaviors like these and social reinforcement and peer pressure from such individuals may also function as a means of increasing drug use risk. Prior research has indicated some support for the mechanisms of social learning theory (Archer & Flexon, 2021; Orak, Yildiz, Solakoglu, Aydogdu, & Aydiner, 2020; Solakoglu & Yuksek, 2020). So, alternative to the gateway hypothesis positing that marijuana use may result in alterations to brain functioning that places youth at greater risk for progression to hallucinogen use, this alternative mechanism posits that marijuana use may facilitate greater access to hallucinogens and/or expose them to peers who may pressure or reinforce such progression to hallucinogen use.

Justice-involved youth as a priority population

Beyond the actual processes involved with the gateway hypothesis and identifying mechanisms underpinning a potential relationship between marijuana use and hallucinogen use, it is also important to note the relevance of examining these processes among JIY specifically. JIY are a population that are at-risk for substance use and exposure to risk factors for substance use (Vaughn, Freedenthal, Jenson, & Howard, 2007). JIY also demonstrate elevated risk for substance use disorder (Robertson, Dill, Husain, & Undesser, 2004), indicating the need for early intervention to interrupt this progression. With these youth under criminal justice supervision during their adolescent years, this provides a point of intervention that could aid in reducing these risks. This indicates the need for identification of risk factors for hallucinogen use and determination of what mechanisms may underpin a potential gateway relationship. This would facilitate more effective targeting of finite system resources to be directed at programming meant to address significant mediators of this relationship among JIY who have initiated marijuana use. In doing so, this could help to reduce later substance use-related problems and relieve pressure from a strained system.

Apart from the predicted relationships discussed above, it is also important to note the relevance of controlling for additional factors in order to ensure that risk of bias in estimation of relationships of interest is mitigated. Demographic differences in substance use risk have been noted in the literature. Men/boys tend to report increased risk for substance use and substance use disorder compared to women/girls (McHugh, Votaw, Sugarman, & Greenfield, 2018; Young et al., 2002). Racial differences present mixed evidence in regards to substance use issue prevalence rates, particularly depending on the data source used (Mereish & Bradford, 2014; Mitchell & Caudy, 2015). Further, research on social class as a predictor of substance use problems also remains equivocal, with some research demonstrating a negative relationship, other research indicating a positive relationship, and other research indicating no relationship (Galea, Nandi, & Vlahov, 2004). Substance use risk also tends to decline as individuals age into and through adulthood (Schulenberg et al., 2020; Vasilenko, Evans-Polce, & Lanza, 2017), making age a pertinent control variable here as well. Prior research on the Pathways to Desistance data specifically also notes that controlling for waves with valid data is relevant for understanding outcomes of interest as well (Kijowski & Wilson, 2022). Exposure to violence is also a risk factor for substance use (Earnshaw et al., 2017; Menard, Covey, & Franzese, 2015), as individuals may turn to drugs and alcohol as a means to cope with emerging mental health symptoms and impact on cognition. Considering that depression and dual systems constructs are examined as mediators here, it is important to control for exposure to violence in order to ensure that marijuana use may be causing changes in depressive symptoms and cognitive variables.

The present study sought to better understand relationships of interest by examining mediating mechanisms running from marijuana use to later hallucinogen use and testing the following hypotheses:

Hypothesis 1:

Ever using marijuana and higher frequency of lifetime marijuana use prior to baseline will be associated with increased odds of hallucinogen use at wave three.

Hypothesis 2:

Sensation-seeking, impulse control, and deviant peer association at wave two will mediate the relationships between baseline marijuana use and wave three hallucinogen use

Methods

Data

This study utilized data from the first three waves of the Pathways to Desistance dataset (Mulvey, 2000–2010). The first three waves were used here because these waves had the least missing data while also having the capacity to establish temporal ordering between independent, dependent, and mediating variables of interest. Observation periods between waves were six months in length. The Pathways to Desistance data are comprised of the responses of 1,354 JIY who had recently been adjudicated for a serious offense just prior to baseline measurements. Participants were recruited from Maricopa County, Arizona and Philadelphia, Pennsylvania between 2000 and 2003. The entire study period lasted from 2000–2010. Serious offenses, which qualified participants for inclusion in the study consisted of all felony offenses and also misdemeanor weapons charges and sexual assault charges. Of all qualified JIY who were approached regarding their interest in the study, 32% declined the opportunity to take part. A cap on the total number of male drug offenders included in the sample was also applied in order to maintain heterogeneity in these constructs at baseline. This capped the total number of these offenders at 15% of the baseline sample. Attrition rates for the present study peaked at wave three, with 93% of the original sample still providing data.

All data used in the present study were collected via participant self-report through interviews with the research team. Interviews were conducted in locations that were convenient for participants (e.g. participants’ homes, criminal justice facilities, libraries, etc.). The research team provided each participant with a laptop computer during interview sessions. Participants then used these laptops to manually input responses to survey prompts. The present study was exempt from full ethical review because it entailed secondary data analysis of a previously approved dataset that is publicly available. Stata code is available upon request. The Pathways to Desistance dataset is available via the Inter-University Consortium for Political and Social Research. This study was not preregistered. All adult participants provided informed consent for the study and minor participants provided assent with parent/guardian consent obtained.

Measures

Hallucinogen use

The main dependent variable examined in this study was hallucinogen use at wave three. This was coded as a binary variable which delineated participants who reported any hallucinogen use from those who reported no hallucinogen use during this observation period (0 = No; 1 = Yes).

Marijuana use

The key set of independent variables examined in analyses pertained to marijuana use prior to baseline. Two variables were used in this study. The first was binary and delineated participants who reported ever using marijuana prior to baseline from those who did not (0 = No; 1 = Yes). The second variable was ordinal and asked participants to report how frequently they had ever used marijuana during the period of time of most frequent use during their lifetimes (0= Not at all/not used in lifetime; 1 = 1–2 times; 2 = Less than 1× per month; 3=Once per month; 4 = 2–3 times per month; 5=Once per week; 6 = 2–3 times per week; 7 = 4–5 times per week; 8 = Everyday).

Deviant peer association

One of the key mediating constructs examined in this study was deviant peer association measured at wave two. This construct was measured using a scale adapted from the Rochester Youth Development Study (Thornberry, Lizotte, Krohn, Farnworth, & Jang, 1994). A series of ordinal items asked participants to indicate the general number of their friends who influenced them to engage in seven different antisocial acts (e.g. During the last six months how many of your friends have suggested that you should sell drugs?). A mean score was then computed from these ordinal items in order to provide every participant with a single deviant peer association score at wave two.

Impulse control

Another key mediating construct examined in this study was impulse control at wave two. Impulse control was measured using the Weinberger Adjustment Inventory (Weinberger & Schwartz, 1990). This instrument utilized a series of ordinal items, which asked participants to indicate the degree to which they agreed or disagreed with statements describing their attitudes and behaviors pertaining to impulse control (e.g. I say the first thing that comes into my mind without thinking enough about it). Seven of the eight individual ordinal items were then reverse coded so that higher scores corresponded to higher levels of impulse control. A mean score was also computed from these individual ordinal items so that each participant had a single impulse control score at wave two.

Sensation-seeking

Another mediating construct examined in this study was sensation-seeking at wave two using the Youth Psychopathic Traits Inventory (Andershed, Kerr, Stattin, & Levander, 2002). Sensation-seeking was assessed using a series of ordinal items that asked participants to rate the degree to which they agreed or disagreed with statements asking about their attitudes and behaviors pertaining to sensation-seeking (e.g. I like to be where exciting things happen). An index score ranging from 4 to 20 was then computed from these individual sensation-seeking item scores so that every participant was provided with a single sensation-seeking score at wave two.

Depression

Depression was also examined as a mediator. The Brief Symptom Inventory was used to measure depression at wave two (Derogatis & Melisaratos, 1983). This instrument utilized a series of ordinal items which asked participants to indicate the degree to which they had been bothered by various depressive symptoms during the previous week (e.g. Feeling no interest in things). A mean score was then computed from the individual item scores so that all participants had a single depression score at wave two.

Control variables

Beyond the main variables analyzed, a number of control variables were also included in analyses in order to mitigate bias in estimation of effects of interest. The first of these control variables was biological sex. Biological sex was coded as a binary variable at baseline which delineated male and female participants into distinct categories (0 = Male; 1 = Female).

Another control variable included in analyses was race at baseline. Race was measured at baseline as a nominal variable with four categories: Black, Hispanic, White, and Other Race. Dummy variables corresponding to each race category were then computed (e.g. 1 = Black; 0 = All other participants). The dummy variable corresponding to White participants was then omitted from analyses. This was done in order to provide a reference group for interpreting race coefficients in comparison to.

Socioeconomic status (SES) was also controlled for in analyses. Hollingshead’s (1957) two-factor index of social position was utilized to measure SES at baseline. This instrument measured SES as a weighted score comprised participants’ parents’ educational attainment and occupational prestige scores. If both parents were able to provide baseline data, then a mean score was computed so that all participants had one SES score at baseline.

Exposure to violence was also included as a control variable. The Exposure to Violence Inventory was used to measure this construct at wave two (Selner-Ohagan, Kindlon, Buka, Raudenbush, & Earls, 1998). This instrument measured the presence/absence of two distinct forms of exposure to violence: witnessed violence and direct victimization. A binary variable was then used in analyses which delineated participants by whether or not they reported experiencing either form of exposure to violence during the wave two observation period (0 = No; 1 = Yes).

Another control variable included in analyses was age at wave three. Age was measured in single-digit year intervals.

It was important to control for the number of follow-up waves that participants provided valid data for. A variable which provided a count of the number of waves for which each participant provided valid data was included as a control variable.

Wave three observation period length also needed to be controlled for in analyses because longer observation periods necessarily meant that there was more time with which participants could have used hallucinogens. While all participants’ wave three observation periods were about six months in length, there was some degree of variance in length in terms of the exact number of days. For this reason, a variable which provided a count of the exact number of days in each participant’s wave three observation period was included in analyses.

It was also necessary to control for the amount of time that participants spent in secured facilities with no community access during the wave three observation period (e.g. jail, prison, psychiatric hospital, etc.). This was because spending time in such facilities would necessarily impact participants’ access to hallucinogens during the time in question. This variable was operationalized as a proportion ranging from 0 to 1 for each participant, with higher scores indicating more of the wave three observation period spent in secured facilities (e.g. .37 = 37% of the observation period spent in secured facilities).

Analytic strategy

The analyses for the present study utilized a series of path analysis models. This strategy was chosen because of its capacity to test for direct and indirect effects of interest and determine whether or not identified dual systems and peer variables provided significant mediation of the relationships between marijuana use and hallucinogen use variables. This entailed the decomposition of each of the indirect effects of interest to determine the magnitude and significance of each of these pathways running from the independent to dependent variables of interest. The generalized version of this method was chosen because of analysis of the binary dependent variable. As such, logistic regression models were used to assess relationships of interest within the path analysis framework. Two models were estimated for each independent variable of interest. Model 1 assessed the direct effect of each marijuana variable on hallucinogen use risk, net of all control covariates. Model 2 then included all the proposed mediating pathways through impulse control, sensation-seeking, and deviant peer association in order to determine the robustness of the direct effects of marijuana use at baseline. Because this study was focused on gateway effects of marijuana, this necessarily meant that participants who had ever used hallucinogens prior to baseline needed to be excluded from analyses. With these participants excluded, a final sample size of 983 participants was utilized in analyses. Listwise deletion was used to manage missing data because multiple imputation was not compatible with the gsem function in Stata used to carry out these analyses. Coefficients are interpreted as the predicted change in the log-odds of hallucinogen use given a one-unit increase in a given independent variable of interest, net of all control covariates.

The next phase of analyses extended the path analyses to determine whether any or all of the proposed pathways running through dual systems and peer variables provided significant mediation of the relationships between marijuana use and hallucinogen use variables. In order to calculate statistical significance of these pathways, they needed to be transformed into scalars and standard errors need to be computed for each. While the delta method can be used to compute said standard errors, this can result in non-normally distributed standard errors that may lead to biased estimation of indirect effects of interest. As such, the Preacher and Hayes (2008) method of computing these standard errors was utilized. This entails the use of a bootstrap resampling process that utilizes random sampling methods to compute normally distributed standard errors and alleviates concerns of bias in determining statistical significance. A total of 500 resamples were carried out for these analyses.

Results

Table 1 provides descriptive statistics for all variables examined in analyses. A total of 22 participants reported hallucinogen use during the wave three observation period (2.39.%). The sample was comprised mainly of male JIY (Male = 86.88%; Female = 13.12%). The plurality of participants in this sample were Black (49.24%), followed by Hispanic participants (33.57%), then White participants (13.02%), and then Other Race participants (4.17%). The average age of participants during the wave three observation period was 16.966 years old.

Table 1. Descriptive statistics.

	Mean/Proportion	Standard Deviation	Minimum	Maximum
Hallucinogen Use Wave 3 No = 97.61% (N = 897) Yes = 2.39% (N = 22)
Ever Used Marijuana Prior to Baseline No = 20.35% (N = 200) Yes = 79.65% (N = 783)
Most Ever Used Marijuana Prior to Baseline	5.661	3.356	1	9
Impulse Control at Wave 2	3.192	.922	1	5
Sensation-Seeking at Wave 2	13.029	3.379	4	20
Depression at Wave 2	.432	.620	0	3.67
Deviant Peer Association at Wave 2	.883	.861	0	4
Gender Male = 86.88% (N = 854) Female = 13.12% (N = 129)
Race Black = 49.24% (N = 484) Hispanic = 33.57% (N = 330) White = 13.02% (N = 128) Other Race = 4.17% (N = 41)
Socioeconomic Status at Baseline	51.967	12.470	11	77
Exposure to Violence at Wave 2 No = 45.47% (N = 417) Yes = 54.53% (N = 500)
Number of Waves with Valid Data	8.977	1.998	0	10
Age at Wave 3	16.966	1.155	15	19
Days in Wave 3 Observation Period	175.747	30.533	61	245
Proportion of Wave 3 Spent in Secure Facilities	.423	.437	0	1

Table 2 provides Model 1 and Model 2 results for the binary marijuana use independent variable; whereas Table 3 provides these same results for the marijuana frequency independent variable.

Table 2. Generalized structural equation modeling logistic regression of ever having use marijuana prior to baseline on hallucinogen risk at wave 3 (N = 917).

	Model 1			Model 2
	Coefficient	p-Value	95% Confidence Interval	Coefficient	p-Value	95% Confidence Interval
Ever Used Marijuana Prior to Baseline (0=No; 1=Yes)	1.085	.159	−.425 2.594	.629	.445	−.987 2.246
Impulse Control (Wave 2)	—	—	—	−.381	.329	−1.146 .383
Sensation-Seeking (Wave 2)	—	—	—	.054	.603	−.150.258
Depression (Wave 2)	—	—	—	.376	.375	−.455 1.207
Deviant Peer Association (Wave 2)	—	—	—	.463	.180	−.213 1.138
Gender (0=Male; 1=Female)	.126	.848	−1.162 1.415	−1.085	.348	−3.350 1.180
Race (reference=White) Black Hispanic Other Race	−1.473 −1.342 −13.692	.010 .035 .981	−2.592–.354 −2.590–.093 −1166.293 1138.908	−1.168 −1.358 −12.524	.109 .093 .973	−2.958 .262 −2.940.225 −727.533 702.486
Socioeconomic Status at Baseline	.006	.766	−.035 .048	−.003	.910	−.049 .044
Exposure to Violence (0=No; 1=Yes) (Wave 2)	.358	.454	−.579 1.295	−.513	.419	−1.759 .733
Age	−.171	.406	−.574 .232	−.288	.264	−.794 .217
Days in Observation Period	>-.001	.992	−.015 .015	.004	.697	−.014 .022
Proportion of Observation Period Spent in Secure Facilities	.045	.935	−1.026 1.116	−.724	.320	−2.150.702
Number of Waves with Valid Data	.001	.996	−.324 .326	.038	.880	−.452 .527
Constant	−1.349	.758	−9.939 7.240	2.155	.727	−9.935 14.246
Marijuana Use→Impulse Control	—	—	—	−.186	.013	−.332–.040
Marijuana Use→Sensation-Seeking	—	—	—	1.371	<.001	.798 1.944
Marijuana Use→Depression				.092	.098	−.017 .201
Marijuana Use→Deviant Peer Association	—	—	—	.389	<.001	.253 .525

Table 3. Generalized structural equation modeling logistic regression of marijuana use frequency prior to baseline on hallucinogen risk at wave 3 (N = 917).

	Model 1			Model 2
	Coefficient	p-Value	95% Confidence Interval	Coefficient	p-Value	95% Confidence Interval
Highest Frequency Ever Used Marijuana Prior to Baseline	.178	.031	.016 .341	.090	.362	−.103 .282
Impulse Control (Wave 2)	—	—	—	−.348	.360	−1.124 .409
Sensation-Seeking (Wave 2)	—	—	—	.053	.612	−.152 .258
Depression (Wave 2)	—	—	—	.382	.364	−.442 1.206
Deviant Peer Association (Wave 2)	—	—	—	.407	.256	−.296 1.111
Gender (0=Male; 1=Female)	−.027	.968	−1.343 1.289	−1.156	.322	−3.445 1.132
Race (reference=White) Black Hispanic Other Race	−1.580 −1.471 −15.029	.006 .025 .990	−2.717–.442 −2.754–.188 −2337.652 2307.594	−1.220 −1.390 −12.594	.096 .089 .972	−2.655 .215 −2.993 .212 −721.553 626.366
Socioeconomic Status at Baseline	.006	.774	−.035 .048	−.004	.872	−.050.043
Exposure to Violence (0=No; 1=Yes) (Wave 2)	.322	.505	−.624 1.267	−.483	.448	−1.732 .766
Age	−.218	.301	−.631 .195	−.309	.236	−.819 .202
Days in Observation Period	.001	.947	−.015 .016	.004	.698	−.015 .022
Proportion of Observation Period Spent in Secure Facilities	−.051	.927	−1.150 1.047	−.776	.294	−2.226 .674
Number of Waves with Valid Data	.015	.928	−.312 .343	.049	.843	−.440.538
Constant	−.660	.882	−9.369 8.049	2.555	.679	−9.536 14.645
Marijuana Use→Impulse Control	—	—-	—	−.035	<.001	−.052–.017
Marijuana Use→Sensation-Seeking	—	—	—	.180	<.001	.111 .248
Marijuana Use→Depression	—	—	—	.009	.162	−.004 .022
Marijuana Use→Deviant Peer Association	—	—	—	.070	<.001	.054 .086

Model 1 results for the binary marijuana use independent variable indicated that initiation of marijuana use prior to baseline was not a significant predictor of hallucinogen use at follow-up (Coefficient = 1.085; p < .159). Only the race variables indicated a significant effect on hallucinogen use in this mode, with Black and Hispanic participants reporting lower risk for hallucinogen use than White participants. Model 2 results for the binary marijuana use independent variable indicated little difference in effects upon inclusion of the hypothesized mediators. Marijuana use remained a nonsignificant predictor of hallucinogen use (Coefficient = .629; p < .445). No control variables nor any of the hypothesized mediators were significant predictors of hallucinogen use in this model Marijuana use was a significant predictor of greater sensation-seeking, lower impulse control, and greater deviant peer association in this model also (Sensation-seeking coefficient = 1.371, p < .001; Impulse control coefficient = −.186, p < .013; Deviant peer association coefficient = .389, p < .001), but not depression (Depression coefficient = .092, p < .098).

Results for Model 1 utilizing the marijuana use frequency measure indicated that having ever used marijuana at higher frequencies at some point prior to baseline was associated with increased log-odds of hallucinogen use at wave three (Coefficient=.178; p < .031). None of the proposed control variables included in this model other than race variables were significant predictors of hallucinogen use at wave three, with Black and Hispanic participants reporting lower risk for hallucinogen use than White participants. Model 2 results pertaining to the analyses for the marijuana use frequency independent variable indicated that inclusion of the proposed mediating pathways resulted in the relationship between baseline lifetime marijuana use frequency and wave three hallucinogen use risk to be attenuated to non-significance (Coefficient=.090; p < .362). None of deviant peer association, sensation-seeking, nor impulse control exerted significant direct effects on hallucinogen use risk in this model. None of the control variables in the model were significant predictors of hallucinogen use at wave three in this model either. Greater lifetime marijuana use frequency also predicted greater sensation-seeking, lower impulse control, and greater deviant peer association in this model (Sensation-seeking coefficient = .070, p < .001; Impulse control coefficient=−.035, p < .001; Deviant peer association coefficient = .180, p < .001), but not depression (Depression coefficient = .009, p < .162).

Extensions of the path analysis models using the Preacher and Hayes (2008) bootstrap resampling method to compute standard errors and significance levels were used to determine whether significant mediation was observed for any of the pathways running from the marijuana use independent variables of interest to hallucinogen use risk at wave three. The previous analyses indicated that the binary marijuana use variable was not a significant predictor of wave three hallucinogen use at wave three in any of the models, but mediation effects were still tested for here since such examinations remain a relevant endeavor even when a significant IV→DV relationship has not been observed (Shrout & Bolger, 2002b). There were no significant mediation effects observed for the binary marijuana use variable pathways (Depression indirect effect = .035, Standard error = 35.607, p < .999, 95% confidence interval = −69.754—69.823; Deviant peer association indirect effect = .180, Standard error = 32.962, p < .996, 95% confidence interval = −64.424—64.784; Impulse control indirect effect = .071, Standard error = 87.434, p < .999, 95% confidence interval = −171.297—171.438; Sensation-seeking indirect effect = .074, Standard error = 101.615, p < .999, 95% confidence interval = −199.088—199.236; Total indirect effect = .360, Standard error = 82.750, p < .997, 95% confidence interval = −161.828—162.547). Mediation analyses for the marijuana use frequency models indicated that none of the hypothesized pathways provided significant mediation of the relationship between lifetime marijuana use frequency and hallucinogen use either (Depression indirect effect = .004, Standard error = .020, p < .859, 95% confidence interval=−.036—.043; Deviant peer association indirect effect = .028, Standard error = .103, p < .783, 95% confidence interval=−.174—.231; Impulse control indirect effect = .012, Standard error = .027, p < .643, 95% confidence interval=−.040—.065; Sensation-seeking indirect effect = .010, Standard error = .129, p < .941, 95% confidence interval=−.243—.262; Total indirect effect = .054, Standard error = .048, p < .257, 95% confidence interval=−.039—.147). This lack of mediation occurred despite the fact that inclusion of mediating pathways in Model 2 for this independent variable indicated that about 50% of the relationship was accounted for by these cognitive and social variables. As such, deeper examination of these analyses in order to determine why significant mediation was not observed were undertaken. These analyses entailed examination of each individual mediating pathway in the model as the sole mediator of the marijuana use-hallucinogen use relationship. None of the mediators fulfilled the criteria for significant mediation. None of the mediators exerted significant direct effects on hallucinogen use risk, meaning that they would not mediate this relationship through shared variance with marijuana use. The deviant peer association pathway appeared to account for the vast majority of the indirect effect of marijuana use on hallucinogen use, but again, because sensation-seeking itself did not exert a significant direct effect on hallucinogen use, this was likely the reason for a lack of observed mediation. So, despite the fact that inclusion of all mediators resulted in a large degree of attenuation of the effect of marijuana use frequency on hallucinogen use risk, explanations for this relationship could not be identified with confidence.

Discussion

This study provided a unique examination of the mechanisms underpinning the relationship between marijuana use and hallucinogen use in a test of the gateway hypothesis and popular alternative conceptualizations. While marijuana use initiation was not a significant predictor of risk for later hallucinogen use, lifetime marijuana use at elevated frequency did predict increased odds of later hallucinogen use. Interestingly, none of the cognitive, affective, nor social mediators of this relationship significantly accounted for this relationship. In this way, while high frequency marijuana use presented a risk factor for later hallucinogen use, the reasons why this relationship exists remain unexplained by gateway hypothesis and popular alternative hypotheses. There are a number of relevant implications of these findings for public health and criminal justice professionals focused on reducing risk for hallucinogen use among JIY.

As noted above, only the frequency of marijuana use predicted increased odds of hallucinogen use among JIY, but not just initiation of use during adolescence. These findings further suggest that the gateway effects of marijuana are more complex than simply initiating use. Instead, this indicates that the potential impact of marijuana may only become relevant once the level of use reaches heavier levels. This suggests that screening for marijuana use may need to extend beyond simply identifying if there is a history of any past use. Identification of youth who are or were chronic marijuana users at some point may provide more utility for diversion into treatment programming to prevent progression to hallucinogen use. In doing so, the juvenile justice system may have greater capacity for reducing the prevalence of hallucinogen use among this subpopulation of youth. That said, there remains the issue of determining the specific form of programming that may be best apt to address these concerns.

While this study established a link between heavy marijuana use during adolescence and elevated risk for later hallucinogen use, analyses were unable to identify a mediating mechanism that helped to explain this relationship. While social, affective, and cognitive mechanisms were examined in this regard, none of the proposed mechanisms significantly mediated the relationship between marijuana use and hallucinogen use. As such, identification of the best way to actually reduce risk of progression to hallucinogen use among this population remains in doubt. It may be that other cognitive, affective, or social mechanisms that were left unexamined by this study may play a role here. Additionally, it may be that more distinct neurological mechanisms that are difficult to measure using psychometric instruments may have been relevant here. It may have been that dopamine availability may have been impacted by marijuana use, leading to individuals being driven to seek out additional novel and stimulating experiences like hallucinogen use in order to trigger dopamine release. Indeed, prior research has indicated that heavy marijuana use may impact dopamine response to medication meant to address low dopamine availability (Volkow et al., 2014). The implications here would be the potential for the development of novel medications that address this issue with dopamine availability that may function to interrupt gateway effects if this does provide a mediating mechanism. While sensation-seeking can be relevant for understanding dopamine from a psychological perspective, the relationship between this psychological construct and its neurological underpinnings are complex and the psychological data may not have adequately captured this. That said, a full examination of such mechanisms was beyond the scope of this study due to a lack of data on such measures. Future research should seek to examine the relevance of these alternative mechanisms for understanding gateway effects of marijuana use among JIY.

Another important finding of this study pertained to the first portions of the hypothesized mediating pathways examined in these analyses; as both marijuana use independent variables were significant predictors of greater deviant peer association, diminished impulse control, and elevated sensation-seeking. Neither marijuana use variable was a significant predictor of depression. These significant findings are consistent with the extant literature indicating that marijuana use during adolescence may lead to dysfunctional cognitive development and affiliation with antisocial peers (Gallegos, Zaring-Hinkle, Wang, & Bray, 2021; Goode, 1970; Jacobus & F Tapert, 2014; Lubman, Cheetham, & Yücel, 2015; Renard et al., 2017; Shen, 2020; Tarter, Vanyukov, Kirisci, Reynolds, & Clark, 2006). Despite the fact that none of these pathways provided significant mediation in these analyses, these findings still indicate the potential for additional iatrogenic outcomes associated with marijuana use during adolescence. The dysfunctional cognitive development associated with the dual systems constructs and the elevated levels of deviant peer association observed among marijuana using JIY are concerning in their own right given that prior research has indicated that all of these constructs are risk factors associated with a range of other antisocial behaviors (Gildner, Kirwan, Pickett, & Parkhill, 2021; Haeny et al., 2020; Karras, Csillik, & Delhomme, 2023; Lin et al., 2020; Wojciechowski, 2018). While the specific mediation effects of interest to this study were not observed, these significant relationships on the front end of the mediating pathways raise concerns about marijuana use leading to gateway effects to other types of behaviors operating through these pathways. Indeed, recent research has indicated that gateway effects related to marijuana use may not be limited to substance use behaviors. For example, Wojciechowski (2022) observed that marijuana use operated through dual systems imbalance to increase risk for violent offending among JIY. Examination of other outcomes in this regard was beyond the scope of this study, but these significant relationships indicate the need for continued research examination other gateway effects associated with marijuana use. Doing so may provide a more nuanced understanding of how marijuana use during adolescence impacts social and psychological domains that can increase risk for a variety of behavioral outcomes.

A final issue to consider here is the low baseline rate of hallucinogen use in the sample utilized here despite the high-risk nature of this JIY sample. This is likely due in large part to exclusion of participants who reported pre-baseline use of hallucinogens in order to ensure that gateway effects related to lifetime marijuana use could be isolated. While only 2.39% (N = 22) of the sample used here reported hallucinogen use at wave three, prior to excluding pre-baseline lifetime hallucinogen users, this prevalence was 5.08% (N = 64). This is much more consistent with the elevated rates of substance use that would be expected among this JIY sample. This likely reduced statistical power and inhibited the capacity of these analyses to identify significant mediation effects. The wide 95% confidence intervals for the indirect effects provide evidence supporting this explanation. This indicates the need for future research to re-examine these mediation effects with a more robust JIY sample that is not impacted by the need to exclude individuals with pre-baseline lifetime use from the analyses. Doing so would allow for robustness testing of these findings and the potential that any or all of the mediators examined here may provide a mechanism for understanding whether or not the gateway hypothesis is explained.

Despite this study providing a unique examination of gateway effects linking marijuana use to hallucinogen use, there remain a number of relevant limitations. First, generalizability of these findings may be a concern. The Pathways to Desistance sample is comprised only of JIY, a group of adolescents who are at elevated risk for substance use (McCuish et al., 2017). As such, these relationships may not be applicable to the general population of adolescents. Relatedly, these data were collected via purposive sampling. This nonprobability sampling strategy then may mean that the sample is not generalizable to the population from which it is drawn. The cap applied to the number of male drug offenders included in the sample exacerbates this concern. As such, there is a need to re-examine these processes using a general population sample of same-age youth collected via probability sampling methods in order to determine the robustness of these findings. Another limitation pertains to the hallucinogen use measure used in analyses. The number of participants reporting having used hallucinogen during the wave three observation period was relatively small and may have contributed to the null mediation effects observed in analyses. Firth logistic regression may have been a good option for addressing this power problem, but this form of regression is not available within the path analysis framework in Stata. As such, formal mediation analyses could not be conducted using both methods. This indicates the need for continued research on this topic that is less limited in terms of power in order to test for mediation effects in a more robust manner. Another limitation of this study pertains to the inability to control for earlier levels of mediating variables. Because of the lifetime measures of marijuana use at baseline utilized in this study, it is impossible to control for pre-use levels of mediating variables. As such, it is difficult to say with certainty that marijuana use actually caused changes in the values of these mediators or if these levels were established prior to marijuana use. This could help to explain the nonsignificant mediation effects observed in this study and most definitely indicates the need for additional research on this topic that is able to control for pre-marijuana use levels of mediators in order to increase the validity of causal inference.

While these limitations temper the findings of this study, there remain importance to these findings. Only high frequency marijuana use was found to predict increased risk for hallucinogen use later on, but not any marijuana use. The former finding is consistent with the gateway hypothesis, whereas the latter is not. That said, none of the proposed mediators indicated that they were significant in providing mechanisms linking the two substance use behaviors. This indicates the need for continued research on the gateway hypothesis in this regard to better understand why marijuana use may lead to greater risk for hallucinogen use, particularly high frequency marijuana use. Doing so could facilitate the design and implementation of more effective treatment programming for JIY in order to reduce risk for progression to use of this class of drugs following high frequency marijuana use.

Disclosure statement

No potential conflict of interest was reported by the author(s).