Behavioral precursors of nicotine product use trajectories among youth

Abstract

Introduction:

Preventing nicotine use among youth is a public health priority. Nicotine use emerges from complex relationships between numerous factors. This project used network analysis to model behavioral precursors of nicotine use (knowledge, attitudes, perceptions, intentions) among youth who had never used nicotine and determine which predicted future trajectories of use across multiple nicotine products.

Methods:

Data were from the Population Assessment of Tobacco and Health study (2013–2018; analyzed 2023–2024), youth ages 12–17, who reported never using nicotine at Wave 1. Network structure was determined for behavioral precursors at Wave 1 and central network nodes were identified (N=5087). Central nodes were then tested as predictors of trajectories of use across multiple nicotine products during Waves 1–4 (N=3851).

Results:

Central nodes of the Wave 1 network were harm perception, expectancy that tobacco would calm anger/reduce stress, and intention to try. Those with lower intent to try at Wave 1 had lower odds of being in an Experimentation or Use class vs a Nonuse class during Waves 1–4 (p<.0001). Those with more accurate harm perception had lower odds of being in an Experimentation vs Non-use class (p=.004). Those with positive expectancies had higher odds of being in an Experimentation vs Non-use or Use class (p=.04,.02).

Conclusions:

Findings suggest a network model of behavioral precursors of nicotine use that can be tested, including central nodes that predicted trajectories of use across multiple nicotine products, and therefore may be priority intervention targets.

Introduction

Despite a recent downward trend in youth nicotine product use, still nearly 1 in 4 U.S. high school students and 1 in 15 middle school students reported past month use of a nicotine product in 2020.^1,2 Furthermore, 35% of youth reported current use of more than one nicotine product,¹ and in some studies, youth reporting use of more than one product outnumbered those using a single nicotine product or device.^3,4 Any nicotine use among youth has immediate adverse health effects. Nicotine is highly addictive, and symptoms of dependence can occur soon after first puff.⁵ Nicotine use initiation during youth is linked to continued use in adulthood and many adverse health effects across the lifespan.^6,7 Use of more than one product may increase nicotine exposure and risk for addiction beyond that from a single product.^4,8 Indeed, youth who reported use of more than one product were less likely to intend to quit or perceive harm from nicotine use compared to those using a single product.⁴ Use of more than one product may also reduce the effectiveness of tobacco regulations targeted to a specific product.³ Thus, it is important to understand the patterns and predictors of trajectories of use across multiple nicotine products among youth.

Other tobacco products (other than cigarettes), such as e-cigarettes, have contributed to use of more than one product among youth. Up to 40% of youth who reported current use of more than one product reported use of products other than cigarettes.⁹ Demographic factors have been associated with using more than one product, including older age, male sex, and lower parental education,¹⁰ as have other substance use and mental health problems.¹¹ Using more than one product has been associated with a range of risk factors including sensation seeking,¹² perceived social norms and benefits,^4,13,14 tobacco marketing receptivity¹⁵ and exposure,¹⁶ lower harm perceptions,^4,13,14 menthol smoking,¹⁷ low positive affect,¹⁸ and even boredom relief.¹⁹ However, there is very limited research on factors that predict trajectories of use across multiple nicotine products among youth who have never used nicotine.

Potential behavioral precursors to nicotine use have been studied including knowledge, attitudes, intentions, and perceptions.²⁰ Theoretical frameworks for how behavior and intentions relate to nicotine use are frequently cited;²¹ however many studies are cross-sectional,²² there is limited use of population-based samples,²⁰ and studies are largely limited to cigarettes without considering other nicotine products and use of more than one product. Furthermore, only a few studies addressed stages of nicotine use such as “non-use,” “first puff,” “experimentation,” “regular use,” and “daily use.”²⁰ The large number of independent variables related to nicotine use among youth have presented a challenge to determining effective targets for intervention.^20,23 A better understanding of which variables predict trajectories of use across multiple nicotine products and how these are interrelated will help to prioritize targets for interventions.²⁴

One way to examine the relationship among behavioral precursors is using network analysis, an approach used to identify and analyze patterns of statistical association in multivariate data.²⁵ Network analysis estimates interrelationships among a set of observed variables to determine which are more strongly associated with each other and most influential on other variables in the set. Thus, unlike factor analysis, the network approach does not seek to identify latent variables that are used to explain, but rather uses partial correlation to evaluate mutual relationships among observed variables.²⁶ Network analysis has been recently formalized in psychology.²⁵ In network analysis, network structure consists of nodes, representing variables in a dataset, and edges, representing pairwise conditional association between variables in the dataset, while conditioning on the remaining variables.²⁵ Computational models suggest that the activation of one node in a strongly connected network can lead to the activation of other connected nodes–a cascading effect. Network structure can be examined for interpretable features of the network,²⁷ such as nodes with the highest centrality are the most strongly connected to other nodes in the network. Activating highly central nodes has a greater potential to spread activation throughout the network than activating peripheral nodes, suggesting potential intervention targets.²⁸ Using network analysis to model potential behavioral precursors of trajectories of use across multiple nicotine products and identify central variables may help to identify and prioritize prevention and policy intervention targets.

This study used network analysis and statistical models to better understand the patterns and predictors of trajectories of use across multiple nicotine products in a U.S. nationally representative youth sample from the Population Assessment of Tobacco and Health (PATH) study. A network model was first determined using PATH Wave 1 items assessing behavioral precursors of nicotine use (knowledge, attitudes, perceptions, intentions) among youth who reported never using nicotine products. Network structure was examined to identify the most central nodes or precursors, which were then tested as predictors of previously defined longitudinal trajectories of frequency of use across multiple nicotine products from Waves 1–4.²⁹

Methods

Study Population

Data were from Waves 1–4 of the PATH study (Appendix Figure 1).³⁰ N=13651 youth (ages 12–17 years) were surveyed at Wave 1 (09/12/13–12/14/14) and then yearly for 3 years (Wave 2, 10/23/14–10/30/15; Wave 3, 10/19/15–10/23/16; Wave 4, 12/01/16–01/03/18). Youth were included in the network analysis if all of the questions of interest about precursors of nicotine use were applicable at Wave 1 (N=5087), i.e., they had seen/heard of but never used [cigarette, e-cigarette, cigar, smokeless tobacco]. Predicting trajectories of use across multiple nicotine products included the subset of youth who had longitudinal weights at Wave 4 (N=3851). These weights were assigned by PATH to Wave 4 respondents in the Wave 1 cohort who also participated in Waves 2 and 3, and account for variable probabilities of selection and attrition. Because publicly available, deidentified PATH data were used in the study, IRB approval was not required.

Measures

For behavioral precursors of nicotine use, PATH Wave 1 included fourteen variables that measured a behavioral precursor (knowledge, attitudes, perceptions, intentions). See Table 1 for the full PATH items, tobacco products assessed, and rating scales, and the labels used in the network model in Figure 1. Raw PATH items were used and were averaged across nicotine products (cigarettes, e-cigarettes, smokeless tobacco, and cigars). All youth included in the network analysis answered all fourteen items. PATH data can be downloaded at: https://www.icpsr.umich.edu/web/NAHDAP/studies/36498.

Table 1.

Behavioral precursors of nicotine use from PATH Wave 1.

PATH item	Node Fig 1	Tobacco productsa	Scale
Ever been curious about using [tobacco product]	Curious	cigarettes, e-cigarettes, cigars, smokeless tobacco	1=very curious 4=not at all curious
Think you will try [tobacco product] soon	Intent	cigarettes, e-cigarettes, cigars, smokeless tobacco	1=definitely yes 4=definitely not
Would use [tobacco product] if one of your best friends offered you one	IntentFriend	cigarettes, e-cigarettes, cigars, smokeless tobacco	1=definitely yes 4=definitely not
Likeliness of someone becoming addicted to [tobacco product]	Addict	cigarettes, e-cigarettes, cigars, smokeless tobacco	1=very unlikely 5=very likely
How much do you think people harm themselves when they use [tobacco product]	Harm	cigarettes, e-cigarettes, cigars, smokeless tobacco	1=no harm 4=a lot of harm
How much you think people harm themselves when they use [tobacco product] every day	HarmEvrDay	cigarettes, smokeless tobacco	1=no harm 4=a lot of harm
How much you think people harm themselves when they use [tobacco product] some days but not every day	HarmSmDay	cigarettes, smokeless tobacco	1=no harm 4=a lot of harm
How long do you think someone has to use [tobacco product] before it harms their health?	TimeHarmb	cigarettes, e-cigarettes, smokeless tobacco, other kinds of smoked tobacco combined	1=it will never harm their health 2=less than a year … 6=20 years or more
I think I would enjoy using tobacco	EnjoyTob	-	1=strongly agree 4=strongly disagree
Using tobacco would be energizing	EnergyTob	-	1=strongly agree 4=strongly disagree
Using tobacco would help me reduce or handle stress	StressTob	-	1=strongly agree 4=strongly disagree
Using tobacco would help me calm down when I am angry	CalmTob	-	1=strongly agree 4=strongly disagree
Using tobacco would help me control my weight	WeightTob	-	1=strongly agree 4=strongly disagree
Using tobacco would help me feel more comfortable at parties	PartyTob	-	1=strongly agree 4=strongly disagree

^{a -}Scores were averaged across the tobacco products to create an overall score. Cronbach alpha was >0.6 for all variables except harmEvrDay (0.48).

^{b -}TimeHarm rescaled in the analysis such that higher scores were better (1=it will never harm their health, 2=20 years or more, …, to 6=less than a year)

Figure 1.

Estimated network of potential behavioral precursors to nicotine use.

Estimated network, with thicker lines representing stronger relationships, or partial correlations. Solid lines represent positive relationships while dashed lines represent negative relationships. The network appears to have three clusters: one cluster related to intent, with the nodes 1) Curious –curious about using tobacco; 2) Intent - will try tobacco soon; 3) IntentFriend - use tobacco if offered by a friend; one cluster related to harm perceptions, with the nodes 4) Harm – how much people harm themselves when using tobacco; 5) Addict - likeliness of becoming addicted to tobacco; 6) HarmEvrDay - how much people harm themselves using tobacco every day; 7) HarmSmDay - how much people harm themselves using tobacco some days; 8) TimeHarm – how long someone had to use tobacco before it harms their health; and one related to nicotine use expectancies, with the nodes 9) EnjoyTob – would enjoy tobacco; 10) EnergyTob – tobacco would be energizing; 11) StressTob – tobacco would reduce/handle stress; 12) CalmTob – tobacco would calm them down when angry; 13) WeightTob – tobacco would help with weight control; 14) partyTob –tobacco would help them feel more comfortable at parties.

Regarding use of nicotine products, participants indicated the number of days in the past 30-days in which they used each: cigarettes, e-cigarettes, traditional cigars, cigarillos, filtered cigars, and smokeless tobacco. The maximum days they used traditional cigars, cigarillos, and filtered cigars were combined to an overall cigar variable. A description of how trajectory classes were derived from these variables has been given previously.²⁹ Trajectory classes included nonuse, experimentation, increasing cigarette/cigar use, increasing e-cigarette/cigarette use, stable smokeless tobacco/cigarette use, and decreasing cigarette/e-cigarette/cigar use. Based on low sample sizes in some trajectories, the current analysis used three trajectories: Nonuse, characterized by a consistent low average use of all tobacco products (0–0.03 days in the past 30 days) across waves; Experimentation, characterized by an average of <1.2 day of use of all tobacco products in the past 30 days across waves; and Use, characterized by >10 days of use of a tobacco product at some timepoint and >2.7 days of use of another tobacco product (0.01–17.65 days in the past 30 days) at some timepoint. For more information on trajectories including sample sizes, see Appendix Table 1.

Covariates included age (12–14, 15–17 years old), sex (male/female), race/ethnicity (Hispanic, non-Hispanic white, non-Hispanic Black, and non-Hispanic other, with the last two categories combined in an Other category due to small sample size), and parental education (more than high school vs not), which were self-reported in the PATH study.

Statistical analysis

Analyses were conducted in 2023–2024. Network analysis was performed in R (version 4.2.2) and is reported according to current guidelines.³¹ The network was estimated using the ggmModSelect method in the bootnet package (version 1.5) based on Spearman correlations of pairwise complete observations.³² This unregularized method was chosen over LASSO because the data is low-dimensional (14 nodes) with high sample size.³³ The method consists of model selection on unregularized models using Bayesian information criterion.³² The estimated network was plotted using the qgraph package (version 1.9.3). The nodes in the network are variables and the edges represent connections between two nodes. Strength of connections between two nodes are measured as edge-weights using partial correlation coefficients, reflecting strength of association corrected for other nodes in the network. To identify the key nodes in the network, the two most reliable measures were used (centrality indices³⁴): strength, the sum of the absolute value of the weights connected to a node; and expected influence, the sum of the weights connected to a node and a measure of positive connectivity (i.e., positive partial correlation) in networks with both positive/negative (i.e., negative partial correlation) edges.³⁵ Network accuracy was evaluated by bootstrap³¹ with 1000 samples to test for differences between pairs of centrality indices and edge-weights (using the routine in the bootnet package).³⁶

To test network stability to dropped samples,³¹ correlation stability coefficients were computed for strength, expected influence, and edges, as the maximum proportion of cases that can be dropped to retain a correlation of .7 between original measures and measures in reduced random subsets in at least 95% of the subsets. This approach assesses whether the same ordering of edges/centralities is maintained across random samples of the data. Values above .5 are preferred.³⁶ The code for fitting the network is available in the Appendix.

The most central network nodes were identified and tested as predictors of latent trajectories of use across multiple nicotine products. For nodes that were clustered together and measured theoretically similar constructs, an average was taken to avoid multicollinearity. A multivariable multinomial logistic model (in PROC SURVEYLOGISTIC, SAS 9.4) examined the association between predictors (central nodes) and classes, with individuals assigned to their most likely class. The model controlled for sociodemographics and used weights to account for the PATH complex design. Our analyses tested whether central nodes are significant predictors of nicotine use. A sensitivity analysis was conducted to analyze the significant predictors selected by using backward elimination on all variables and is reported in the Appendix.

Results

The estimated network (Figure 1) appears to have 3 main clusters: one cluster containing nodes related to intent to try tobacco, one cluster related to tobacco use expectations, and another cluster related to tobacco harm perceptions. Figure 2 presents the strength of nodes as z-scores. The strongest five nodes were: harm perception (harm), expectations about calm (calmTob), expectations about reducing stress (stressTob), intent to try from friend (intentFriend), and intent to try soon (intent). Bootstrapped difference tests for strength (Appendix Figure 2) indicated that harm perceptions (harm), the strongest node (node strength=1.20), was significantly stronger than all but one node in the network, expectations about calm (calmTob). The next strongest nodes: expectations about calm (calmTob), expectations about reducing stress (stressTob), and intent to try from friend (intentFriend), were significantly more central than about 9–10 other nodes. The expected influence plots/results are not presented as they did not differ from strength.

Figure 2.

Strength of network nodes.

Strength of nodes as standardized z-scores, ordered by value. The meaning of the nodes is the same as in Figure 1.

Bootstrapped difference tests indicated that many edges differ from one another (Appendix Figure 3). The strongest edge was between intent to try soon (intent) and intent to try from friend (intentFriend); followed by expectations about reducing stress (stressTob) and expectations about calm (calmTob); harm when used every day (harmEvrDay) and harm when used some days (harmSmDay); addictiveness (addict) and harm perception (harm); and harm perception (harm) and harm when used every day (harmEvrDay). These edges differed from almost all other edges. The network correlation stability coefficients were all appropriate (.75), meaning up to 75% of the observations could be removed and network measures would stay stable.

Based on the network model, the following predictors of trajectories of use across multiple nicotine products were examined: 1) harm perception, using the most central node (harm); 2) intent to try tobacco, an average of two strong, highly connected nodes, intent to try soon (intent) and intent to try from friend (intentFriend); 3) tobacco use expectations about calm/stress, an average of two strong, highly connected nodes, expectations about calm (calmTob) and expectations about reducing stress (stressTob); and 4) an overall measure reflecting other tobacco use expectations, an average of expectations about energy (energyTob), enjoyment (enjoyTob), weight control (weightTob), and comfort at parties (partyTob).

For trajectories, growth patterns, and sizes, see Appendix Table 1. For participant descriptive statistics overall and by class, see Appendix Table 2. Table 2 presents the results from the multinomial model predicting class membership for associations of central nodes with latent trajectories of use across multiple nicotine products. Youth with more accurate harm perceptions (i.e., youth who thought that tobacco products cause more harm) at Wave 1 had 30% lower odds of being in the Experimentation class vs the Nonuse class during Waves 1–4 (AOR=0.70, 95% CI 0.54–0.89, p=.004). Those with lower intent of using at Wave 1 (i.e. higher values on the variable scale) had lower odds of being in the Experimentation class and Use class vs the Nonuse class (both p<.0001). Those who thought tobacco would help them with calm/stress had higher odds of being in the Experimentation class versus Nonuse (p=0.04) and Use class (p=0.02).

Table 2.

Adjusted odds ratios (AOR) from multivariable multinomial model predicting class membership (N=3807)

Predictor	Experimentation vs non-use		Use vs non-use		Use vs. Experimentation
Effect	AOR (95% CI)	p	AOR (95% CI)	p	AOR (95% CI)	p
Age 15–17 vs 12–14 years old	1.24 (0.94, 1.63)	0.13	1.37 (1, 1.87)	0.05	0.91 (0.61, 1.34)	0.62
Race/ethnicity
Hispanic vs non-Hispanic white	0.61 (0.43, 0.86)	0.006	0.59 (0.38, 0.93)	0.02	1.02 (0.59, 1.77)	0.94
Other vs non-Hispanic white	0.7 (0.48, 1.01)	0.06	0.48 (0.32, 0.73)	0.001	1.45 (0.84, 2.5)	0.18
Sex female vs male	0.87 (0.67, 1.14)	0.30	0.56 (0.43, 0.72)	<.0001	1.57 (1.1, 2.23)	0.01
Parental education more than high school (vs not)	0.89 (0.66, 1.19)	0.43	0.7 (0.5, 0.98)	0.04	1.26 (0.83, 1.92)	0.27
How much people harm themselves when they use tobacco (1=no harm, 4=a lot of harm)	0.7 (0.54, 0.89)	0.004	0.83 (0.59, 1.17)	0.27	0.84 (0.56, 1.26)	0.39
Intent to try soon/if friend offered tobacco product (1=definitely yes, 4=definitely not)	0.23 (0.16, 0.34)	<.0001	0.17 (0.1, 0.28)	<.0001	1.41 (0.84, 2.37)	0.19
Using tobacco would help stress/calm (1=strongly agree, 4=strongly disagree)	0.77 (0.61, 0.99)	0.04	1.23 (0.88, 1.72)	0.22	0.63 (0.42, 0.94)	0.02
Using tobacco would help energy/enjoyment/at parties/with weight (1=strongly agree, 4=strongly disagree)	0.97 (0.64, 1.49)	0.90	0.69 (0.42, 1.14)	0.14	1.4 (0.79, 2.5)	0.25

Note: N=44 subjects with missing data on at least one of the predictors were excluded from the model. Boldface indicates statistical significance (p<0.05).

Discussion

This study used network analysis to assess the interrelationships among behavioral precursors of nicotine use (knowledge, attitudes, perceptions, and intentions) in youth who reported never using nicotine, identify key predictors, and test whether these were associated with previously defined trajectories of use across multiple nicotine products²⁹ in the PATH study. Behavioral precursors found to be central to the network at Wave 1 included harm perception, intention to try, and expectancies that nicotine use would help with calm/stress. These central nodes predicted trajectories of nicotine product use across Waves 1–4, whereas a combined variable representing other nicotine use expectancies that were less central to the network was not found to be predictive. Findings extend prior work by identifying predictors of trajectories of nicotine product use in a longitudinal, U.S. nationally representative sample.

The most central node in the Wave 1 network was harm perception, or how much people harm themselves when they use a nicotine product. This suggests that harm perception is a central behavioral precursor to nicotine use among youth who had never used nicotine, as this node was most strongly connected to other network nodes. Additionally, youth who had never used nicotine and had more accurate harm perceptions at Wave 1 had lower odds of being in an Experimentation vs Nonuse class across Waves 1–4, although harm perceptions did not distinguish between Use vs. Nonuse or Experimentation vs. Use. A related study examined harm perception and nicotine use among youth across Waves 1–2 using a series of logistic regression models,³⁷ finding that perceptions of nicotine use harm/addictiveness increased from Wave 1 to 2, and that “low” and “medium” vs. “high” perceived harmfulness at Wave 1 related to an increased probability of use at Wave 2. That study looked at each nicotine product separately. A strength of the current study was consideration of use of more than one nicotine product, which acknowledges that youth are experimenting with and initiating use of more than one product.² This finding supports addressing harm perceptions to prevent youth from experimenting with nicotine products.

Central network nodes additionally included expectancies that nicotine use would help youth reduce stress and calm them down. Outcome expectancies, or the believed consequences of one’s behavior, are thought to influence behavior.³⁸ Positive expectancies have been associated with youth susceptibility to cigarette smoking,³⁹ cigarette smoking behavior,^40,41 e-cigarette susceptibility and use,⁴² and polytobacco use,⁴³ although studies are limited, in particular for other nicotine products. As in this study, expectancies were related to stress and relaxation,³⁹ and slimness.^44,45 Here, youth never-users who had positive expectancies that nicotine use would help calm anger/reduce stress at Wave 1 had higher odds of being in an Experimentation vs Use or Nonuse class across Waves 1–4, although expectancies did not distinguish between Use vs. Nonuse. Similarly, a Texas youth cohort study demonstrated that, for youth who had never used nicotine, baseline expectations of stress relief were related to susceptibility and use at 6 months for cigarettes, e-cigarettes, cigars, smokeless tobacco, and hookah.⁴⁶ The current findings extend to a generalizable U.S. youth sample, and to associations with trajectories of use across multiple nicotine products. Thus, interventions need to offer youth alternative solutions for reducing stress that do not include nicotine products,⁴⁶ in particular because nicotine use exacerbates stress among youth.⁴⁷ Treatments such as cognitive behavioral therapy⁴⁸ and mindfulness⁴⁹ offer alternative strategies to cope with stress. Additionally, tobacco advertising appeals to youth by implying relief of negative affect (stress/anxiety), therefore findings suggest policy targets to combat these marketing practices.⁵⁰

Finally, central network nodes also included intentions to use nicotine products. Intention to use has been commonly related to subsequent use of nicotine products among youth.^51,52 Three items about intentions are often combined to measure susceptibility, including intentions to try soon, intentions to try from a friend, and curiosity about trying, with susceptibility defined as the “absence of a firm resolve” to never use nicotine products.⁵³ In studies of cigarette smoking,⁵⁴ susceptibility identified 40–80% of adolescent who never smoked but tried cigarettes in the next 3–4 years^55,56 and 40–80% of adolescent who never smoked but became established cigarette smokers 4–6 years later.^56,57 Susceptibility has also been associated with youth initiation of cigar, e-cigarette, hookah, and smokeless tobacco use.^51,52,58,59 Another study found that youth reporting use of more than one nicotine product were more likely to smoke a cigarette if offered by a friend.⁴ The current network analysis included the three susceptibility items, of which intentions to try soon and if offered from a friend were most central to the network. These were combined and it was determined that youth who had never used nicotine and had lower intentions to use at Wave 1 had lower odds of being in an Experimentation or Use class vs. a Nonuse class during Waves 1–4, although intentions did not distinguish between Use vs. Nonuse. A recent study found that susceptibility at Wave 1 was related to onset of exclusive nicotine use, exclusive alcohol use, and polysubstance use vs. never substance use at Wave 2.⁶⁰ Future work could test whether central nodes related to nicotine use intentions are associated with onset of other substance use.

The network approach identified central behavioral precursors to youth nicotine use and demonstrated the relatedness of these variables in a network model that can be tested. Previous network studies have shown that if a variable is highly central, a change in that variable is strongly related to a change in the overall network.³⁵ The implication is that by targeting central behavioral precursors, an intervention might impact nicotine use risk overall. It might also be possible to target central nodes in order to impact connected nodes, although a recent review found that no studies in the network literature have yet targeted individual nodes in order to measure the impact on connected nodes.³⁵ Either way, network structure might be informative, for example, if it is found that changing harm perceptions doesn’t change expectancies, this might indicate that targeting both variables is warranted.

Limitations

This study had several limitations. First, only behavioral precursors included in the PATH study could be tested. Other factors of interest include subjective norms and perceived behavioral control. Second, there are limitations to network analysis, including how to interpret centrality indices, and variability in the importance of centrality based on network structure and inter-node interaction.⁶⁴ Causality cannot be established, so it is currently unknown whether targeting central nodes can improve prevention. Nevertheless, central nodes were strongly associated with trajectories of nicotine use, suggesting policy and clinical relevance. Third, trajectories were defined based on most likely fit, but individuals may not perfectly follow the observed pattern.²⁹ Fourth, sample size was limited for some trajectories,²⁹ which were collapsed for testing prediction. Fifth, all the different types of nicotine products used by youth (e.g., hookah, pipe, etc.) were not modelled, typically because rates of use were too low to model.²⁹ Finally, future studies could examine network structure over time, including considering how the network changes in the tobacco regulatory environment and with emerging nicotine products.

Conclusions

Findings provide evidence that nicotine use harm perceptions, expectancies, and intentions to try are central behavior precursors among youth who reported never using nicotine and predict different longitudinal trajectories of use across multiple nicotine products among a U.S. nationally representative youth sample. Findings suggest that targeting these precursors may be meaningful to prevention and policy strategies.