Abstract
Cannabis legalization has expanded access to high-potency products (e.g., dab concentrates) that deliver large amounts of delta-9-tetrahydrocannabinol (THC). It is unclear whether cannabis consumers mitigate the effects of higher-potency products by consuming smaller amounts (i.e., self-titration). We investigated whether users adjust consumption amounts in response to potency when using cannabis flower and concentrates. Data come from four online surveys of U.S. cannabis consumers (N = 8158) reporting past-month flower and/or dab concentrate use. Participants reported their typical amounts (number of hits or grams) and potencies (%THC). We analyzed relationships between consumption amounts and %THC in three ways: (1) within-subject/between-product, (2) between-subject/between-product, (3) between-subject/within-product. To assess selection biases, we examined correlations among potency, use frequency, and age of initiation. Participants were approximately 52 % female, 84 % White, and 73 % reported daily cannabis use; average age was 42 years (SD = 16). In within-subject/between-product analyses, 59 %–92 % of participants used larger amounts of flower than concentrates. In between-subject/between-product analyses, flower-only consumers reported greater amounts than concentrate-only consumers (Median Hits: 8–14 vs. 5–8; Median Grams: 1–1.5 vs. 0.18–0.5). In between-subject/within-product analyses, using higher-THC versions of a product predicted larger amounts (Flower/Hits: rs = 0.12–0.19; Flower/Grams: rs = 0.07–0.29; Concentrate/Hits: rs = 0.09–0.14; Concentrate/Grams: rs = 0.19–0.25), more frequent use (rs = − 0.01–0.39), and earlier initiation (rs = − 0.06 to − 0.33). Results suggest that cannabis consumers self-titrate when switching between flower and concentrate product types, but more frequent consumers prefer higher-potency versions of a given product type and consume them in larger amounts.
Keywords: Cannabis, Self-titration, Amount, Potency, Concentrates, THC
1. Introduction
U.S. cannabis legalization has increased the availability of products with high concentrations of delta-9-tetrahydrocannabinol (THC; primary psychoactive component). THC has reinforcing effects in a dose-response fashion, and exposure to large amounts, both acutely and over extended periods, is associated with adverse health outcomes, including psychosis, cognitive impairment, and cannabis use disorder (CUD) (Cooper and Haney, 2009; Hall, 2015; Haney et al., 1997). New methods of administration that allow for rapid inhalation of large amounts of high-potency concentrates (e.g., “dabbing” concentrates) may exacerbate these risks (Bidwell et al., 2021; Pierre et al., 2016). However, the extent to which high-THC products will impact public health depends in large part on whether use of such products results in greater THC exposure among consumers. Some consumers are thought to adjust their intake based on product potency (i.e., %THC)—a behavior known as “self-titration” (Leung et al., 2021). If individuals consume smaller amounts of higher-potency cannabis to achieve a desired effect, then population-level THC exposure may remain stable.
Self-titration of nicotine and alcohol has been clearly documented (Huang and Riley, 2024); however, a recent systematic review reported that evidence for cannabis self-titration is mixed and highlighted the need for more research to clarify the relationship between amount (e.g., number of grams of product material) consumed and potency (i.e., % THC) (Leung et al., 2021). Understanding whether cannabis consumers self-titrate is essential for informing regulatory policies (e.g., potency caps, product labels, taxation strategies) and assessing the public health impact of high-THC products (e.g., CUD risk, acute adverse effects, and long-term mental health consequences) (Borodovsky and Budney, 2018).
In the present study, we test for evidence of cannabis self-titration by examining the relationship between cannabis potency (i.e., %THC) and amount consumed (i.e., number of grams or “hits” consumed) in three ways: (1) a “within-subject/between-product” analysis to test the hypothesis that individuals who consume both flower (i.e., lower-potency product) and concentrates (i.e., higher-potency product) consume greater amounts of flower than amounts of concentrates; (2) a “between-subject/between-product” analysis to test the hypothesis that the group-level amount of flower used by flower-only consumers is greater than the group-level amount of concentrates used by concentrate-only consumers; and (3) a “between-subject/within-product” analysis to test the hypothesis that those who consume lower-potency versions of a product type (e.g., those who use low-potency flower) consume greater amounts of that product than those who consume higher-potency versions of the same product type (e.g., those who use high-potency flower). To assess the replicability of the findings, we conducted these analyses using four datasets from four survey studies that used the Cannabis Exposure Index (CEI) (Borodovsky et al., 2023) to collect cannabis consumption data.
2. Methods
2.1. Recruitment
We analyzed four datasets collected from separate surveys conducted between May 2021 and October 2023. Participants were recruited via Meta (Facebook and Instagram) using cannabis-related advertisements and keyword targeting (Borodovsky, 2022; Borodovsky et al., 2018). We used geographic targeting to recruit individuals residing in the U.S. and demographic targeting to recruit those age 18 and older. No other demographic targeting or quotas were used. Individuals who clicked the advertisement were directed to a Qualtrics-hosted consent page. Those who declined consent or reported being under age 18 were automatically disqualified.
All CEI survey designs and data cleaning procedures included multiple data quality checks. To reduce bot and scam responses, no compensation was provided to participants. Automated metadata checks included exclusion of responses with reCAPTCHA scores below 0.5 and removal of submissions flagged by Qualtrics’ RelevantID system due to duplicate IP scores ≥75 or fraud scores ≥30, consistent with prior literature (Bonett et al., 2024; Donkin et al., 2025; Elenio et al., 2025; Rodriguez and Oppenheimer, 2024). The survey also included attention-check items (e.g., simple math problems) and bot honeypot questions. During data cleaning, we applied logic checks to identify implausible responses (e.g., age of cannabis onset had to be less than current age; 7-day frequency could not exceed 30-day frequency). Three datasets included in this analysis were also used in prior publications (Borodovsky et al., 2023, 2024, 2025) that provide additional details on the number of responses removed as part of these data quality procedures. These studies were approved by the Dartmouth Committee for the Protection of Human Subjects.
2.2. Inclusion criteria
In all four original studies, participants were eligible if they reported using cannabis at least once in the past month. The original purpose of this criterion was to ensure that infrequent past-month consumers (e.g., those who used once or twice in the past month) were present in our samples to enable testing of the CEI’s performance across a wider range of cannabis consumers. For the present study, retaining infrequent past-month consumers served a similar purpose by ensuring sufficient variability in use patterns needed to test for evidence of self-titration.
In this study, we further restricted the sample in three ways. First, participants must have reported smoking cannabis flower and/or dabbing concentrates at least once in the past month. Comparing these two specific product types provides a stronger test of the self-titration hypothesis because flower typically contains much lower %THC than concentrates, and dabbing involves rapid administration of concentrates. Second, participants must have reported consuming approximately the same total amount of cannabis on each day that they used. That is, even if participants used cannabis on non-consecutive days (e.g., Monday, Wednesday, and Saturday), they must have indicated that they typically used a consistent amount on each of those using days (e.g., 1.5 g on each of the three days). This restriction allowed us to minimize within- and between-person variability to provide the clearest possible answers about the presence or absence of self-titration behaviors. Third, we removed participants who reported cannabis amounts (e.g., number of hits) but did not report a corresponding potency. Additional details about missing data are provided in subsequent sections and Supplementary Tables 1 and 2
2.3. Measures
The Cannabis Exposure Index (CEI) is a survey instrument that captures detailed cannabis consumption patterns (Borodovsky et al., 2024; Budney et al., 2024; Walsh et al., 2023) and was used to collect data in all four survey datasets analyzed in this study. It measures past 7-day and 30-day consumption behaviors, including specific product types (e.g., flower, concentrates), administration methods (e.g., smoking, vaping), “amounts” (i.e., number of “hits” or number of grams) (Manthey et al., 2023), and potencies (i.e., %THC). All CEI surveys used a forced-choice design. Additional details about the key CEI variables analyzed in the present study are provided below.
2.4. Key variables
2.4.1. Product type (Flower and Dab Concentrate)
The two cannabis products and methods of administration analyzed in this study were smoking flower and dabbing concentrates. Flower products are lower potency (approx. 1–30 % THC) (Smart et al., 2017). Dab concentrates, in contrast, are highly potent cannabis extracts designed for rapid inhalation using a dab rig (approx. 40–100 % THC) (Bidwell et al., 2021; Smart et al., 2017). The large difference in potency between flower and dab concentrates makes these two products useful for testing the self-titration hypothesis.
2.4.2. Amount (number of “hits” or grams) consumed
In this study, the term “amount” refers to the amount of flower or concentrate product material that participants reported consuming. The CEI uses a unit-preference design that allows participants to report their typical consumption amounts per using day as either number of “hits” (i. e., puffs, tokes, inhalations) or number of grams. We analyzed number of hits and number of grams separately for flower and concentrates.
2.4.3. Potency
In this study, the term “potency” refers to the percentage of delta-9-tetrahydrocannabinol (%THC) in a cannabis product. The CEI collected potency data for each product type used by participants. Response options ranged from 1 to 30 % THC for flower and 40 to 100 % THC for concentrates. Participants reported the typical potency of the products they consumed on using days.
2.5. Analyses
2.5.1. Statistical methods
We conducted the same set of complete-case analyses on each of the four CEI datasets to assess the consistency of results across independent samples. We used nonparametric statistical methods, specifically Quantile Regression and Spearman Correlation, for three reasons. First, univariate distributions of cannabis use variables (e.g., hits, grams) were skewed and included outliers (see Supplementary Figure). Second, residuals from ordinary least squares regression models estimating cannabis use amounts as a function of potency violated assumptions of normality and homoscedasticity. Third, prior research suggests that while individuals may overestimate the absolute quantity of cannabis they use, rank-based comparisons of self-reported amounts can still meaningfully differentiate consumers (Prince et al., 2018). Nonparametric methods are well-suited to these conditions because they reduce the influence of outliers, accommodate heteroscedasticity, and make fewer distributional assumptions.
We did not adjust for covariates for three reasons. First, the within-subject/between-product design was a self-controlled comparison. Second, we already restricted the sample to minimize variability in key characteristics (e.g., restricted to only those who reported using the same amounts per using day). Third, unadjusted results suited our aim to provide simple, interpretable relationships that could serve as references for future research on this topic.
Finally, we did not apply formal alpha-level corrections for multiple comparisons and instead followed recommendations to “(1) describe what was done in a study; (2) report effect sizes, confidence intervals, and p values; and (3) let readers use their own judgment about the relative weight of the conclusions” (Althouse, 2016). Relatedly, we interpret p-values as continuous values that indicate the degree of compatibility between the data and the model assumptions (Greenland et al., 2016; Rothman, 1990). We believe this approach better reflects the scientific goals of the study by encouraging readers to focus on interpreting the overall pattern of evidence, rather than whether individual results cross significance thresholds.
2.5.2. Within-subject/between-product
We first conducted “within-subject/between-product” analyses of participants who reported using both flower (lower-potency product) and concentrate (higher-potency product) because this subgroup presumably provides the strongest test of the self-titration hypothesis. These analyses compared a participant’s amount of flower use to his or her amount of dab concentrate. We used an intercept-only quantile regression model with the within-person difference as the dependent variable (i.e., dependent variable is the result of subtracting concentrate amount from flower amount). The model intercept represents the estimated median within-subject difference in amount used between the two product types. We could not conduct within-subject/within-product analyses because participants reported only one potency value per product type.
2.5.3. Between-subject/between-product
The “between-subject/between-product” analysis examined differences in consumption amounts between participants who reported exclusively using flower (i.e., flower-only consumers) and participants who reported exclusively dabbing concentrates (i.e., concentrate-only consumers). This analysis compared the amounts of flower used by flower-only consumers to the amounts of concentrates used by concentrate-only consumers. We used quantile regression models with a binary independent variable (0 = flower, 1 = concentrate) to estimate two outcomes: (1) the difference in the median number of hits of the two product types, and (2) the difference in the median number of grams of the two product types.
2.5.4. Between-subject/within-product
We then conducted a “between-subject/within-product” analysis using Spearman correlations to assess whether participants who used low-potency versions of a product consumed larger amounts than those who used high-potency versions of the same product. For example, we compared the amounts of flower consumed by participants who used lower-potency flower (e.g., 10 % THC) to the amounts consumed by participants who used higher-potency flower (e.g., 30 % THC). Similarly, we compared the amounts of concentrates consumed by participants who used lower-potency concentrates (e.g., 40 % THC) to the amounts consumed by participants who used higher-potency concentrates (e.g., 80 % THC).
2.5.5. Post hoc selection bias analysis
More frequent cannabis use and younger age of onset are associated with increased tolerance to the subjective effects of cannabis (Colizzi and Bhattacharyya, 2018). Individuals with higher tolerance may prefer to use higher-potency products to achieve their desired intoxicating effect. If this preference exists, it could create a selection bias that leads to conflicting conclusions about the presence of a self-titration effect.
To explore this possibility, we used Spearman correlations to conduct post hoc tests of associations between self-reported potency and (1) past 30-day cannabis use frequency and (2) age of cannabis use onset. These analyses were conducted separately for flower and concentrates.
2.5.6. Sample size, subgroups, and missing data
The final, combined sample used for analyses in this study contained N = 8158 complete case respondents. Note, however, that different subsamples of complete cases were drawn from the combined sample and utilized for different analyses based on (1) participants’ unique product use profiles and (2) the subgroup definitions required by each type of analysis. For example, the within-subject/between-product analysis required participants to have used both flower and concentrates, which necessarily excludes those who used only one product. Conversely, the between-subject/between-product analysis included flower-only and concentrate-only users and excludes those who used both. Because potency is a critical variable in this study, participants with missing potency responses were excluded. Results from data cleaning for complete case analyses are provided in Supplementary Tables 1 and 2
3. Results
Table 1 describes participant demographics and cannabis use frequency across the four CEI survey datasets. Mean age ranged from 36 to 47 years (SD range: 15 to 16), and the percentage of female participants ranged from 45 % to 58 %. Most participants were White (range: 73 %–88 %). Bachelors-level education ranged from 18 % to 24 %, and full-time employment was the most commonly reported employment status across all samples (range: 45 %–54 %). Median past 30-day cannabis use frequency was 30 days in all studies; however, variability differed, with Study 3 showing the widest interquartile range (IQR: 12). Mean age of cannabis use onset ranged from 15 to 17 years (SD range: 4 to 6).
Table 1.
Demographics and past 30-day cannabis use frequency of four cannabis use survey datasets.
| Study | Age in years, M (SD) | % Female | Race (%)a | Educationb | Employmentb | Past 30-day use frequency, Median (IQR) | Age of first cannabis use, M (SD) |
|---|---|---|---|---|---|---|---|
| 1 (N = 2622) | 40 (15) | 58 % | Black: 4 % | No HS: 2 % | Full-time: 45 % | 30 (0) | 15 (4) |
| Asian: 1 % | HS: 48 % | Part-time: 11 % | |||||
| White: 83 % | Associate: 24 % | Disabled: 19 % | |||||
| Hispanic: 4 % | Bachelor: 18 % | Student: 6 % | |||||
| Other: 12 % | Graduate: 8 % | Unemployed: 9 % | |||||
| Retired: 9 % | |||||||
| 2 (N = 2836) | 47 (15) | 45 % | Black: 1 % | No HS: 2 % | Full-time: 45 % | 30 (0) | 16 (5) |
| Asian: 0.4 % | HS: 43 % | Part-time: 11 % | |||||
| White: 88 % | Associate: 23 % | Disabled: 16 % | |||||
| Hispanic: 6 % | Bachelor: 20 % | Student: 5 % | |||||
| Other: 6 % | Graduate: 12 % | Unemployed: 5 % | |||||
| Retired: 19 % | |||||||
| 3 (N = 1169) | 36 (15) | 49 % | Black: 5 % | No HS: 2 % | Full-time: 54 % | 30 (12) | 17 (6) |
| Asian: 1 % | HS: 48 % | Part-time: 11 % | |||||
| White: 73 % | Associate: 20 % | Disabled: 7 % | |||||
| Hispanic: 16 % | Bachelor: 23 % | Student: 15 % | |||||
| Other: 7 % | Graduate: 8 % | Unemployed: 5 % | |||||
| Retired: 7 % | |||||||
| 4 (N = 1531) | 42 (16) | 48 %c | Black: 3 % | No HS: 2 % | Full-time: 50 % | 30 (2) | 17 (5) |
| Asian: 1 % | HS: 40 % | Part-time: 12 % | |||||
| White: 80 % | Associate: 20 % | Disabled: 9 % | |||||
| Hispanic: 11 % | Bachelor: 24 % | Student: 8 % | |||||
| Other: 5 % | Graduate: 13 % | Unemployed: 6 % | |||||
| Retired: 14 % |
Other races include: American Indian/Alaskan Native and Pacific Islander, and Multiracial. Any Hispanic ethnicity (regardless of race) was reported as Hispanic.
Studies 1 and 3 have participants missing data for Education and Employment status. Study 1: missing 2.1 % of Education responses, missing 0.6 % Employment status responses. Study 3: missing 1.4 % of Education and Employment Status responses.
Study 4 reports of sex contained 6 % indicating a non-binary/other gender.
Within-subject/between-product comparisons (Table 2)
Table 2.
Median within-subject differences in flower amounts (hits or grams) and dab concentrates amounts (hits or grams) among those who used both types of productsa.
| Study | % reporting flower hits > concentrate hits | Median (IQR) w/in-subject diff (flower hits minus concentrate hits) | Quantile Regression DV: w/in-subject diff in hits |
% reporting flower grams > concentrate grams | Median (IQR) w/in-subject diff (flower grams minus concentrate grams) | Quantile Regression DV: w/in-subject diff in grams |
|---|---|---|---|---|---|---|
| 1 | 59 % (Total N = 297) |
2 (−2, 8) | β0: 2 (95% CI: 1.22, 2.78) p < 0.001 |
NAb | NAb | NAb |
| 2 | 75 % (Total N = 63) |
6 (0, 14) | β0: 6 (95% CI: 2.65, 9.35) p = 0.001 |
89 % (Total N = 71) |
1.4 (0.4, 3.5) | β0: 1.38 (95% CI: 0.63, 2.12) p < 0.001 |
| 3 | 88 % (Total N = 56) |
8 (4, 15.5) | β0: 8 (95% CI: 4.57, 11.43) p < 0.001 |
92 % (Total N = 49) |
1.1 (0.4, 2.5) | β0: 1.13 (95% CI: 0.48, 1.77) p = 0.001 |
| 4 | 78 % (Total N = 68) |
6 (1.5, 26) | β0: 6 (95% CI: 0.10, 12.10) p = 0.054 |
90 % (Total N = 68) |
1.3 (0.3, 3.4) | β0: 1.31 (95% CI: 0.55, 2.08) p = 0.001 |
Note: DV = “Dependent Variable”.
Participants in the within-subject/between-product analysis were limited to those who reported consumption amounts for both flower and concentrates using the same unit (either hits or grams). For example, participants who reported hits of flower and hits of concentrates, or grams of flower and grams of concentrates, were included in the analysis. Participants who reported hits of flower and grams of concentrates (or vice versa) were excluded due to non-comparable units.
Participants in Study 1 were only permitted to report concentrate amounts using the hits unit.
Among participants who consumed both flower and concentrates, the majority (range: 59 %–92 %) reported consuming larger amounts of flower (i.e., lower-potency product) than of concentrates (i.e., higher-potency product) across all four studies. Quantile regression consistently estimated positive medians of the within-subject differences in amount consumed (i.e., participants typically used more flower than concentrate), and p-values incompatible with the null hypothesis of no difference in amount consumed between product types.
Between-subject/between-product comparisons (Table 3)
Table 3.
Group median amount (hits or grams) of flower use among flower-only consumers vs. group median amount of concentrate use among concentrate-only consumersa.
| Study | Median (IQR) # of HITS | Quantile Regression DV: # of hits IV: product type (flower vs. concentrate) |
Median (IQR) # of GRAMS | Quantile Regression DV: # of grams IV: product type (flower vs. concentrate) |
||
|---|---|---|---|---|---|---|
| Flower | Dab Concentrate | Flower | Dab Concentrate | |||
| 1 | 8 (5, 13) (n = 496) |
6 (3, 12) (n = 43) |
β1 = −2 (95% CI: −4.62, 0.62) p = 0.134 |
1 (0.5, 2) (n = 1744) |
N/Ab | N/Ab |
| 2 | 8 (4, 12) (n = 1161) |
5 (3, 8) (n = 33) |
β1 = −3 (95% CI: −4.92, −1.08) p = 0.002 |
1.5 (0.9, 3) (n = 815) |
0.5 (0.3, 1) (n = 19) |
β1 = −1 (95% CI: −1.22, −0.78) p < 0.001 |
| 3 | 8 (5, 14) (n = 520) |
6 (3, 10) (n = 25) |
β1 = −2 (95% CI: −4.54, 0.54) p = 0.122 |
1.1 (0.5, 2.5) (n = 397) |
0.18 (0.1, 0.5) (n = 24) |
β1 = −0.88 (95% CI: −1.11, −0.64) p < 0.001 |
| 4c | 14 (7, 26) (n = 798) |
8 (5, 13) (n = 12) |
β1 = −6 (95% CI: −11.56, −0.44) p = 0.034 |
1 (0.3, 2) (n = 788) |
0.3 (0.2, 0.6) (n = 12) |
β1 = −0.75 (95% CI: −0.88, −0.62) p < 0.001 |
Note: DV = “Dependent Variable”; IV = “Independent Variable”.
These analyses excluded individuals who reported both smoking flower and dabbing concentrate products so that the same individual did not contribute multiple data points. “Hits” and Grams are self-reported typical amounts used per using day.
Participants in Study 1 were only permitted to report concentrate amounts using the hits unit.
Participants in this study were asked to report their amount for a given product two times: once using the hits unit, and once using the grams unit. For example, a participant who only smoked flower would be asked to report his/her flower amount using both units (hits and grams). Consequently, the subgroup size differences between hits and grams for a given product are similar or the same.
Table 3 compares the amount of flower used by flower-only consumers to the amount of concentrates used by concentrate-only consumers. Across the studies, the median number of hits and grams consumed per using day was larger among flower-only users than among concentrate-only users. This pattern was supported by quantile regression results, which consistently indicated negative β1 coefficients and p-values that were largely incompatible with the null hypothesis of no difference in amount consumed between product types.
Between-subject/within-product comparisons (Table 4)
Table 4.
Between-subject/within-product Spearman correlations (rs) of self-reported product potency (%THC) and consumption amount (hits or grams)a.
| Study | FLOWER | DAB CONCENTRATE | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Correlation between: number of hits & %THC | Correlation between: number of grams & %THC | Correlation between: number of hits & %THC | Correlation between: number of grams & %THC | |||||||||
| rs | p | n | rs | p | n | rs | p | n | rs | p | n | |
| 1 | 0.16 | <0.001 | 814 | 0.07 | 0.002 | 1744 | 0.09 | 0.25 | 178 | N/Ab | N/Ab | N/Ab |
| 2 | 0.19 | <0.001 | 1323 | 0.08 | 0.004 | 1197 | 0.12 | 0.02 | 422 | 0.25 | <0.001 | 173 |
| 3 | 0.12 | 0.005 | 588 | 0.29 | <0.001 | 530 | 0.09 | 0.25 | 158 | 0.21 | 0.06 | 80 |
| 4 | 0.18 | <0.001 | 1367 | 0.18 | <0.001 | 1356 | 0.14 | 0.21 | 77 | 0.19 | 0.10 | 76 |
“Hits” and grams are self-reported typical amounts used per using day; Participants who used both products (flower and concentrates) were included in these analyses; sample sizes correspond to “Reported Potency” column in supplementary data of missing data patterns.
Participants in Study 1 were only permitted to report concentrate amounts using the hits unit.
We observed either positive correlations or no correlation between potency and amount for both flower and concentrates across all four studies. For flower, positive associations were relatively consistent across studies, and many p-values indicated incompatibility with the null hypothesis of no association. Concentrates showed a similar pattern in the direction of correlations but with generally weaker magnitudes and p-values that were more compatible with the null.
Associations Between Cannabis Potency, Age of Onset, and Frequency of Use (Table 5)
Table 5.
Past month frequency of use and age of cannabis initiation Spearman (rs) correlations with self-reported product potencya.
| Study | Correlation between: Age of initiationb & %THC of Flower | Correlation between: Age of initiationb & %THC of Concentrate | Correlation between: Number of days usedb & %THC of Flower | Correlation between: Number of days usedb & %THC of Concentrate | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| rs | p | n | rs | p | n | rs | p | n | rs | p | n | |
| 1 | −0.09 | <0.001 | 2558 | −0.12 | 0.11 | 178 | 0.14 | <0.001 | 2558 | 0.05 | 0.49 | 178 |
| 2 | −0.11 | <0.001 | 2520 | −0.06 | 0.13 | 595 | 0.19 | <0.001 | 2520 | 0.10 | 0.017 | 595 |
| 3 | −0.27 | <0.001 | 1118 | −0.33 | <0.001 | 238 | 0.39 | <0.001 | 1118 | 0.32 | <0.001 | 238 |
| 4 | −0.13 | <0.001 | 1382 | −0.13 | 0.24 | 77 | 0.20 | <0.001 | 1382 | −0.01 | 0.96 | 77 |
Sample sizes are larger in these analyses relative to Tables 3 and 4 analyses because Table 5 analyses combine those who reported amounts using the hits unit and those who reported amounts using the grams unit into a single group. For example, in Table 4, Study 1, there are n = 814 participants who reported flower hits, and n = 1744 participants who reported flower grams; 814 + 1744 = 2558 participants reported in Study 1 in this table. Note that, participants who used both products (flower and concentrates) were included in these analyses. Therefore, the totals in study 4 in Table 4 will not match the 1382 reported in Table 5 because the design of study 4 required all participants to report their consumption amounts twice: once using the hits unit, and again using the grams unit.
Age of initiation and number of days used in the past 30 days refer to use of any form of cannabis (not just flower or dab concentrates).
To investigate potential selection bias mechanisms underlying the positive correlations between amount and potency observed in Table 4 results, we conducted post hoc tests to determine whether participants’ age of cannabis initiation and past-month frequency of use were associated with the potency of products they reported using. Across all four studies, participants who used higher-potency flower products tended to report more frequent cannabis use and earlier age of onset. These results suggest that individuals with a greater likelihood of tolerance to the effects of cannabis (i.e., frequent consumers and those who began consuming at an earlier age) may be more likely to use higher-potency versions of a given product type.
4. Discussion
This study tested the cannabis self-titration hypothesis using within-subject and between-subject analyses of self-reported amounts (number of hits or grams) consumed and potencies (%THC) of flower and dab concentrate products across four datasets. Results from two analyses suggest that cannabis consumers do self-titrate: those who use both flower (i.e., lower potency product) and dab concentrates (i.e., higher potency product) tend to use greater amounts of flower than concentrates, and the median amount of flower used among flower-only consumers is consistently larger than the median amount of concentrates used among concentrate-only consumers. In contrast, we also observed positive correlations between potency and amount consumed within a given product type (e.g., lower amounts of low-THC flower and higher amounts of high-THC flower), which contradicts the self-titration hypothesis. However, these positive correlations may be partially explained by selection effects, as our post hoc analyses showed that those who use cannabis more frequently and began using at an earlier age (i.e., individuals who likely have greater tolerance) tend to use higher-potency versions of products. This selection effect could be responsible for inducing the positive association between potency and amount that we observed in the within-product analysis. These results have several implications for cannabis research, regulatory strategies, and public health. First, they emphasize the importance of considering both within-product and between-product analyses when studying consumption patterns. Second, they support the notion that potency caps or taxation strategies aimed at reducing high-potency product use could benefit from considering how consumers adjust their consumption to avoid undermining the intended effects of such policies. Third, accounting for self-titration behaviors could help improve model-based estimates of THC exposure and intoxication hours (Caulkins, 2017) when quantifying the population risks of CUD or mental health outcomes associated with access to high-potency products (Borodovsky and Budney, 2018).
Although the within-subject/between-product and between-subject/between-product analyses supported the self-titration hypothesis, results from the between-subject/within-product analysis did not. We found that when comparing within a product type (e.g., amount of low-potency flower vs. amount of high-potency flower), there was either no relationship or a positive relationship between amount and potency. Stated differently, we found that those who used lower potency versions of a product also consumed smaller amounts of that product, whereas those who used higher potency versions of a product consumed larger amounts of that product. One possible explanation for this counterintuitive result is that those with greater tolerance to the effects of cannabis (e.g., those who consume cannabis more frequently and began using at a younger age) are also more likely to prefer high-potency products.
Fig. 1 provides a hypothesized explanation and summary of our results using simulated cannabis amount and potency data. The figure demonstrates how it is possible that between-product and within-product analyses of the same data can lead to conflicting conclusions about cannabis self-titration. The black dashed line corresponds to our between-subject/between-product results (i.e., negative association between potency and amount). In contrast, the positive slopes within the flower data cluster and within the concentrate data cluster correspond to our between-subject/within-product results (i.e., positive correlations between potency and amount). More frequent and more experienced cannabis consumers may use higher-potency products in larger amounts because of their higher tolerance, which, in turn, creates a selection bias that induces positive within-product correlations between potency and amount. This dynamic is a form of Simpson’s Paradox (Hernan et al., 2011; Simpson, 1951) and similar patterns of results have been documented in the literature. For example, Prince and Conner found that collapsing potency data across product types obscured associations with health outcomes, whereas analyzing potency within specific product types revealed associations with health outcomes (Prince and Conner, 2019).
Fig. 1.
Presents simulated data that summarize the hypothesized relationships between cannabis potency and amount consumed based on findings from the empirical results. Within each product type (flower or concentrates), a between-subject analysis would indicate that higher potency corresponds to larger amounts consumed (positive trends within each cluster), which is explained, presumably by a selection effect whereby those with greater tolerance use large amount of higher-potency products. However, when comparing data between the two product types, the overall trend (black dashed line) shows a negative relationship between potency and amount consumed. This is an example of Simpson’s Paradox such that within-product associations yield different conclusions than between-product associations.
This study has several limitations. We recruited participants using online advertisements, which likely oversampled an idiosyncratic subgroup of cannabis consumers and explains why many participants were daily consumers. Additionally, our samples were composed of individuals who reported cannabis use at least once in the past month and reported using the same amount of cannabis on using days, which may limit generalizability. Our analyses also required separating participants into distinct subgroups based on their product use patterns. For example, the within-subject/between-product analysis included only individuals who used both flower and concentrates, while the between-subject/between-product analysis included only individuals who used a single product type. As a result, the subgroups analyzed were non-overlapping and may have differed in clinically meaningful ways (e.g., CUD severity, use motives). These differences may limit the comparability of results across analyses. Relatedly, this study involved a large number of design and analysis choices (e.g., inclusion criteria, variable coding, modeling choices, etc). These procedural choices can affect the results and should be considered when interpreting the findings (Gelman and Loken, 2013; Steegen et al., 2016).
Another important limitation is that all data were self-reported, which introduces measurement error into both the amount and potency variables. For potency, participants living in states with legal cannabis may have relied on labeled product information, while those in states without legalization may have estimated potency based on less reliable or unavailable product information. For amount, prior research has shown that cannabis users tend to systematically overestimate amounts, and the degree of overestimation can vary by product type (Prince et al., 2018). Additionally, although we excluded prefilled concentrate cartridge vape pens to isolate dab concentrate use, some misclassification may have occurred due to inconsistent product terminology, misunderstanding, or lack of attention. Finally, the “same amount” survey item assessed consistency in overall total cannabis use per using day but did not assess whether individuals used the same amount of each specific product type.
It is also important to note that individuals may use different products for different purposes, such as recreational use in social settings versus medical symptom relief in solitary settings. Differences in motivation and context could influence the amount consumed independent of potency or tolerance and, therefore, represent a potential alternative explanation for the observed patterns of use. Our surveys did not include standardized measures of cannabis use motives or contexts (e.g., Marijuana Motives Questionnaire) (Lee et al., 2009). Without these measures, we could not directly evaluate the contribution of motivational or contextual factors. Future studies could incorporate motive and context assessments to better distinguish whether potency–amount relationships reflect self-titration behavior vs. situational differences in consumption patterns. We were also not able to include measures of cannabis use disorder severity (e.g., Cannabis Use Disorders Identification Test) (Adamson and Sellman, 2003) in this study, which would have allowed us to test whether clinical severity influenced potency preferences or consumption patterns. Future work could examine whether self-titration modifies the association between high-potency product use and CUD severity to clarify whether titration reduces the clinical risks commonly linked to high-THC products.
Finally, the study design did not allow for within-subject comparisons of titration across potency levels within the same product type. Each participant reported a single typical potency for each product used, which prevented direct comparison of, for example, the number of hits used with low-THC flower versus high-THC flower by the same individual.
5. Conclusion
This study examined cannabis self-titration by using survey data to explore the relationship between product potency and consumption between and within product types. Understanding self-titration is critical for developing evidence-based regulatory strategies, such as potency caps and labeling requirements, and for crafting public health messaging to reduce harms from high-THC products. The findings also highlight the importance of methodological rigor in cannabis research by illustrating how different analytical approaches can lead to varying conclusions about population cannabis use patterns.
Supplementary Material
Acknowledgments
Funding for this study was provided by grants from the National Institute on Drug Abuse (NIDA) R01DA050032, T32-DA037202, P30-DA037202. The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. In the writing of this manuscript, the author(s) solely developed the initial content and ideas. Author JTB employed ChatGPT and Grammarly to help refine grammar, sentence structure, and word choice. The author(s) reviewed and retain full responsibility for the manuscript’s content.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.org/10.1016/j.jpsychires.2025.09.014.
Footnotes
Declaration of competing interest
All authors of this manuscript have no conflicts of interest to report.
References
- Adamson SJ, Sellman JD, 2003. A prototype screening instrument for cannabis use disorder: the Cannabis Use Disorders Identification Test (CUDIT) in an alcohol-dependent clinical sample. Drug Alcohol Rev. 22, 309–315. 10.1080/0959523031000154454. [DOI] [PubMed] [Google Scholar]
- Althouse AD, 2016. Adjust for multiple comparisons? It’s not that simple. Ann. Thorac. Surg 101, 1644–1645. 10.1016/j.athoracsur.2015.11.024. [DOI] [PubMed] [Google Scholar]
- Bidwell LC, Martin-Willett R, Karoly HC, 2021. Advancing the science on cannabis concentrates and behavioural health. Drug Alcohol Rev. 40, 900–913. 10.1111/dar.13281. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bonett S, Lin W, Topper PS, Wolfe J, Golinkoff J, Deshpande A, Villarruel A, Bauermeister J, 2024. Assessing and improving data integrity in web-based surveys: comparison of fraud detection systems in a COVID-19 study. JMIR form. Res 8, e47091. 10.2196/47091. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Borodovsky JT, 2022. Generalizability and representativeness: considerations for internet-based research on substance use behaviors. Exp. Clin. Psychopharmacol 30, 466–477. 10.1037/pha0000581. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Borodovsky JT, Budney AJ, 2018. Cannabis regulatory science: risk-benefit considerations for mental disorders. Int. Rev. Psychiatr 30, 183–202. 10.1080/09540261.2018.1454406. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Borodovsky JT, Hasin DS, Shmulewitz D, Walsh C, Livne O, Aharonovich E, Struble CA, Habib MI, Budney AJ, 2024. Typical hits, grams, or joints: evaluating cannabis survey measurement strategies for quantifying consumption. Cannabis Cannabinoid Res. 9, 646–658. 10.1089/can.2022.0237. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Borodovsky JT, Hasin DS, Wall M, Struble CA, Habib MI, Livne O, Liu J, Chen L, Aharonovich E, Budney AJ, 2025. Quantity of delta-9-tetrahydrocannabinol consumption and cannabis use disorder among daily cannabis consumers. Addiction 120, 676–685. 10.1111/add.16700. [DOI] [PubMed] [Google Scholar]
- Borodovsky JT, Marsch LA, Budney AJ, 2018. Studying cannabis use behaviors with Facebook and web surveys: methods and insights. JMIR Publ. Health Surveill 4, e48. 10.2196/publichealth.9408. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Borodovsky JT, Struble CA, Habib MI, Hasin DS, Shmulewitz D, Walsh C, Livne O, Aharonovich E, Budney AJ, 2023. Exploring survey methods for measuring consumption quantities of cannabis flower and concentrate products. Am. J. Drug Alcohol Abuse 49, 733–745. 10.1080/00952990.2023.2246635. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Budney AJ, Borodovsky JT, Struble CA, Habib MI, Shmulewitz D, Livne O, Aharonovich E, Walsh C, Cuttler C, Hasin DS, 2024. Estimating THC consumption from smoked and vaped cannabis products in an online survey of adults who use cannabis. Cannabis Cannabinoid Res. 9, 688–698. 10.1089/can.2022.0238. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Caulkins JP, 2017. Recognizing and regulating cannabis as a temptation good. Int. J. Drug Pol 42, 50–56. 10.1016/j.drugpo.2017.01.012. [DOI] [PubMed] [Google Scholar]
- Colizzi M, Bhattacharyya S, 2018. Cannabis use and the development of tolerance: a systematic review of human evidence. Neurosci. Biobehav. Rev 93, 1–25. 10.1016/j.neubiorev.2018.07.014. [DOI] [PubMed] [Google Scholar]
- Cooper ZD, Haney M, 2009. Actions of delta-9-tetrahydrocannabinol in cannabis: relation to use, abuse, dependence. Int. Rev. Psychiatr 21, 104–112. 10.1080/09540260902782752. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Donkin L, Henry N, Kercher A, Pedersen M, Wilson H, Chan AHY, 2025. Effective recruitment or bot attack? The challenge of internet-based research surveys and recommendations to reduce risk and improve robustness. Interact. J. Med. Res 14, e60548. 10.2196/60548. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Elenio EG, Tovar A, San Soucie J, Vadiveloo MK, 2025. Population recruitment strategies in the age of bots: insights from the what is on your plate study. Curr. Dev. Nutr 9, 107442. 10.1016/j.cdnut.2025.107442. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Gelman A, Loken E, 2013. The garden of forking paths: why multiple comparisons can be a problem, Even when there is no ‘Fishing Expedition’ or ‘p-Hacking’ and the research hypothesis was posited ahead of time. Dep. Stat. Columbia Univ https://sites.stat.columbia.edu/gelman/research/unpublished/forking.pdf. [Google Scholar]
- Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, Altman DG, 2016. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. Eur. J. Epidemiol 31, 337–350. 10.1007/s10654-016-0149-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hall W, 2015. What has research over the past two decades revealed about the adverse health effects of recreational cannabis use? Addiction 110, 19–35. 10.1111/add.12703. [DOI] [PubMed] [Google Scholar]
- Haney M, Comer SD, Ward AS, Foltin RW, Fischman MW, 1997. Factors influencing marijuana self-administration by humans. Behav. Pharmacol 8, 101–112. 10.1097/00008877-199702000-00009. [DOI] [PubMed] [Google Scholar]
- Hernan MA, Clayton D, Keiding N, 2011. The Simpson’s paradox unraveled. Int. J. Epidemiol 40, 780–785. 10.1093/ije/dyr041. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Huang S, Riley AL, 2024. Drug discrimination learning: interoceptive stimulus control of behavior and its implications for regulated and dysregulated drug intake. Pharmacol. Biochem. Behav 244, 173848. 10.1016/j.pbb.2024.173848. [DOI] [PubMed] [Google Scholar]
- Lee CM, Neighbors C, Hendershot CS, Grossbard JR, 2009. Development and preliminary validation of a comprehensive marijuana motives questionnaire. J. Stud. Alcohol Drugs 70, 279–287. 10.15288/jsad.2009.70.279. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Leung J, Stjepanovic D, Dawson D, Hall WD, 2021. Do cannabis users reduce their THC dosages when using more potent cannabis products? A review. Front. Psychiatr 12, 630602. 10.3389/fpsyt.2021.630602. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Manthey J, Pons-Cabrera MT, Rosenkranz M, Lopez-Pelayo H, 2023. Measuring cannabis quantities in online surveys: a rapid review and proposals for ways forward. Int. J. Methods Psychiatr. Res 32, e1971. 10.1002/mpr.1971. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Pierre JM, Gandal M, Son M, 2016. Cannabis-induced psychosis associated with high potency “wax dabs.”. Schizophr. Res 172, 211–212. 10.1016/j.schres.2016.01.056. [DOI] [PubMed] [Google Scholar]
- Prince MA, Conner BT, 2019. Examining links between cannabis potency and mental and physical health outcomes. Behav. Res. Ther 115, 111–120. 10.1016/j.brat.2018.11.008. [DOI] [PubMed] [Google Scholar]
- Prince MA, Conner BT, Pearson MR, 2018. Quantifying cannabis: a field study of marijuana quantity estimation. Psychol. Addict. Behav 32, 426–433. 10.1037/adb0000370. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rodriguez C, Oppenheimer DM, 2024. Creating a Bot-tleneck for malicious AI: psychological methods for bot detection. Behav. Res. Methods 56, 6258–6275. 10.3758/s13428-024-02357-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rothman KJ, 1990. No adjustments are needed for multiple comparisons. Epidemiol. Camb. Mass 1, 43–46. [PubMed] [Google Scholar]
- Simpson EH, 1951. The interpretation of interaction in contingency tables. J. R. Stat. Soc. Ser. B Methodol 13, 238–241. [Google Scholar]
- Smart R, Caulkins JP, Kilmer B, Davenport S, Midgette G, 2017. Variation in cannabis potency and prices in a newly legal market: evidence from 30 million cannabis sales in Washington state. Addiction 112, 2167–2177. 10.1111/add.13886. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Steegen S, Tuerlinckx F, Gelman A, Vanpaemel W, 2016. Increasing transparency through a multiverse analysis. Perspect. Psychol. Sci 11, 702–712. 10.1177/1745691616658637. [DOI] [PubMed] [Google Scholar]
- Walsh CA, Struble CA, Aharonovich E, Shmulewitz D, Borodovsky J, Habib MI, Budney A, Livne O, Hasin DS, 2023. Evaluating cannabis exposure in survey items: insights, strategies, and remaining challenges identified from cognitive interviewing. Drug Alcohol Depend. Rep 7, 100161. 10.1016/j.dadr.2023.100161. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.