Effects of Visual Exposure to Natural (Vs. Built) Environments on Delay Discounting and Demand for Substances: Preliminary Evidence and Future Directions

Abstract

Substance use disorders (SUDs) affect millions of people and can be difficult to treat, warranting novel therapeutic approaches. Simultaneously, well-established findings demonstrate beneficial effects of nature exposure for mental health, including SUD comorbidities (e.g., anxiety), and emerging evidence suggests nature exposure could impact decision-making associated with harmful substance use. Across two within-subject experiments (Experiment 1: MTurk, n=170, Experiment 2: college students, laboratory setting, n=29), we evaluated the effects of visual exposure to natural versus built environment stimuli on delay discounting and demand for a substance among individuals who frequently use substances. Experiment 1 results showed that for subsets of participants, under certain conditions, visual exposure to natural compared to built environment stimuli reduced delay discounting and demand for cannabis, with medium effect sizes, but not for alcohol or cigarettes. Order effects emerged, demonstrating within-subject lower delay discounting following visual exposure to natural (vs. built) stimuli, when nature was presented second and lower demand for cannabis when presented first. Acknowledging sample size limitations, Experiment 2 results showed reduced demand for alcohol, with a large effect size, following visual exposure to natural (vs. built) stimuli, and no effect on delay discounting or demand for cannabis. Results contribute preliminary evidence of the beneficial effects of visual nature exposure on substance use-related decision-making outcomes among individuals who regularly use substances. We further discuss methodological challenges. Results hold implications for future research with clinical populations to develop and validate real nature prescription guidelines as prevention or adjunct to SUD treatment programs.

Millions of Americans live with substance use disorders (SUDs), which are considered chronic conditions and can be associated with physiological, behavioral, mental health, and social challenges. Treatment for SUDs includes psychosocial, behavioral, and pharmacological approaches. However, SUDs can be difficult to effectively treat, with approximate treatment dropout rates of 30% (Lappan et al., 2019) and long-term remission rates around 50% (Fleury et al., 2016). Many patients also report low satisfaction and high frustration with standard treatments (Evren et al., 2014; Granerud & Toft, 2015). Thus, the need for innovative and effective long-term therapeutic approaches remains critical.

A promising adjunct to typical SUD treatment options is a directed nature intervention with easily scalable potential for wide application. Research provides vast evidence for the beneficial effects of exposure to nature on physical and mental health (Alcock et al., 2014; Hartig et al., 2014; Twohig-Bennett & Jones, 2018), including depression and anxiety (Grassini, 2022), which are highly comorbid with substance use disorders across drug classes (Kelly & Daley, 2013, Onaemo et al., 2021). Exposure to natural environments improves affect and mood (McMahan & Estes, 2015; Norwood et al., 2019), reduces stress (Antonelli et al., 2019; Meredith et al., 2020), and improves cognitive performance (Ohly et al., 2016). While nature is recognized as beneficial for mental and physical health, and despite the common co-occurrence of poor mental health conditions and substance use (NIDA, 2022), research on the effect of nature exposure on substance use-related outcomes is sparse.

A small body of literature has shown associations between greater residential greenness and reduced substance use outcomes, for example, reduced craving for substances (Martin et al., 2019) and reduced binge drinking frequency and nicotine use (Wiley et al., 2022). When focused on the type of exposure, in contrast to exposure to residential greenness, only deliberately spending time in nature was found to be associated with reduced alcohol-related problems, mediated by reduced negative affect (Almog et al., 2022). With the scarcity of research on the effect of nature exposure on substance use-related outcomes, researchers have called for more research in this context (Wiley et al., 2020), highlighting suggested mechanisms in which exposure to nature may be relevant for prevention and treatment of substance use disorders such as improvements in affect, prosocial behaviors, and SUD relevant decision-making (Berry et al., 2021, 2020).

Two decision-making processes underlying problematic substance use and addictions within the behavioral economic framework are delay discounting and high valuation of a substance (Bickel, Johnson, et al., 2014). These processes are associated with substance use outcomes and independently offer unique predictive value for understanding substance use and other health-related behaviors (Strickland et al., 2017). Greater delay discounting, or a strong preference for immediate smaller rewards over larger but delayed rewards, is associated with substance use disorders, other behavioral addictions (Bickel et al., 2019, Bickel, Koffarnus, et al., 2014, Amlung et al., 2017), and psychiatric conditions such as depression (Amlug et al., 2019), and is manipulable experimentally (Rung & Madden, 2018; Scholten et al., 2019). High valuation of a reinforcer (i.e., the substance) is assessed by purchase tasks, and quantified as high demand for a substance, reflecting higher levels of motivation to purchase and consume a substance even when the cost (e.g., price) is high. The validated purchase tasks yield outcomes that are associated with real-world substance use outcomes such as frequency and quantity of use (Strickland, Campbell, et al., 2020), can predict treatment outcomes (Aston & Cassidy 2019), and are sensitive to acute manipulations (Acuff et al., 2020). Both delay discounting and demand may serve as important behavioral treatment targets for SUDs, and both may be reduced via natural environment exposure, although data in this area are lacking.

Preliminary experimental research supports the notion that natural environments, compared to more built environments may reduce delay discounting, although there is no corresponding research on the effects of natural environment exposure on substance demand. In a series of two laboratory and one field between-subjects experiments, van der Wal et al. (2013) found that delay discounting was reduced after exposure to natural compared to urban scenes among student and community adult samples. In the laboratory experiments participants viewed images for six minutes, and in the field experiment participants walked for five minutes in the respective more natural or more built environment before completing a delay discounting task and other measures. Similar results were found by Berry et al. (2014, 2015) in two between-subjects laboratory studies. Delay discounting was lower following an approximate 4-minute visual exposure (with additional repeated, brief visual exposures between each delay block of the discounting task) to natural environment images compared to built environment and control geometric shapes images (2014; see also Kao et al., 2019, for a study demonstrating reduced delay discounting and sugar consumption following visual exposure to natural versus built environment images).

While exposure to natural compared to built environments reduces delay discounting in different samples (i.e., college students, female university students who were interested in losing weight, general adult sample), there is no similar research among individuals who regularly use substances, who tend to exhibit greater delay discounting in general. Given that behavioral measures of delay discounting and demand are both associated with substance use in real-world contexts and are experimentally manipulable, in the present study we targeted both as potential behavioral markers with implications for treatment related to SUDs. Across two experiments (Experiment 1 conducted online; Experiment 2 conducted in a laboratory setting), we aimed to extend the aforementioned research regarding the effects of visual exposure to natural versus built environments on behavioral economic outcomes in several key domains: (a) test the generalizability of these effects to people who regularly use substances; (b) evaluate the effect of visual exposure to natural versus built environments on delay discounting using a within-subject design, and (c) evaluate the effect of visual exposure to natural versus built environments on demand for substances using a within-subject design. We hypothesized that delay discounting (Hypothesis 1) and demand for substances (Hypothesis 2) would be lower following visual exposure to natural compared to built environment stimuli.

Experiment 1

To investigate the effect of visual exposure to natural versus built environments on delay discounting and demand for either alcohol, cannabis, or cigarettes, Experiment 1 utilized a within-subject cross-over design in which each participant experienced both of the two conditions (hereafter Natural and Built), with a minimum of 5 days between the two sessions. This experiment was conducted with an online sample of people who regularly use substances. We also explored the role of order effects, which have been documented in previous natural versus built environment research (Stenfors et al., 2019), and interpreted according to the Attention Restoration Theory (Kaplan, 1995).

Experiment 1 Methods

We report how we determined our sample size, all data exclusions (if any), all manipulations, and all measures in the study.

Participants

Individuals who regularly use alcohol, cannabis, or cigarettes were recruited on the crowdsourcing platform Amazon Mechanical Turk (MTurk) from September 2021 to September 2022. Regular substance use was defined as use at least ten times in the past month for alcohol or cigarettes, and five times for cannabis (to allow the potential to capture a wider sample, given differences in legality and access across states). Participants were 18 years or older, residing in the U.S., and previously completed at least 100 other tasks on the platform, with an approval rate of at least 95% based on their MTurk account. Participants first completed a short independent screener, and only those who met the inclusion criteria of substance use frequency and demonstrated adequate English proficiency were able to access the main survey. We also aimed to recruit individuals who frequently use opioids, however, the opioid group sample size remained extremely small (n = 3) thus precluding analysis.

Procedures

All surveys were conducted using Qualtrics. Participants first chose their most frequently used substance (i.e., specified as nonmedical use). Based on their choice, they were presented with the tasks and questions relevant to their specified substance (either alcohol, cannabis, or cigarettes). Participants were given the instructions for the hypothetical monetary titrating amount delay discounting task, then exposed to 13 condition-specific images (~24 seconds each) of either natural or built environments for five minutes total before completing the delay discounting task (tasks and measures described in detail below). Next, they were given the instructions for a hypothetical purchase task of their specified substance and were asked three questions to verify comprehension. They were then presented with another six images (~20 seconds each) for two minutes total before completing the purchase task.

After the purchase task, participants responded to questions on their typical and very recent substance use, completed other scales and questions (e.g., nature relatedness, self-compassion, and empathy; not included in the present analysis and thus not reported), reported their demographics and had the opportunity to leave a general comment if they chose to. Participants who provided poor quality data or reported substance use that was less frequent than the inclusion criteria (i.e., did not match levels reported in the screener) were not invited to the second survey (more details on cleaning procedures below). Participants were able to see the second session survey a minimum of five days after the first session. The second survey was identical, except for the condition-specific visual stimuli that were presented. That is, in the first session, participants were randomized to view either the natural or built stimuli, and in the second session they viewed the stimuli they had not viewed previously. Participants who did not complete the second survey within 24 hours received a message the following day via MTurk, inviting them to complete it. Because our manipulation involved visual stimuli that require some level of immersion, the study was not available for completion on mobile phones. The median duration of the first session was 25.5 minutes, and 23.3 minutes for the second session. Participants were paid $0.15 for the screener, and $3.50 for each session for a total of $7.15 for study completion. All procedures were approved by the University of Florida Institutional Review Board under protocol #IRB201902033.

Visual Stimuli.

The visual stimuli (no experimenter-programmed audio) included images of either natural (e.g., lakes, mountains, vegetation) or built (e.g., buildings, bridges, roads) environments. To view the slideshow participants first clicked a box on the screen that opened a new tab (outside of Qualtrics). Participants were instructed to click another box on the screen when they were ready to view the images. Following that click the slideshow began and the images were displayed on the entire computer screen. When the slideshow was over, the tab closed automatically and returned the participants to the Qualtrics survey. The images and code were hosted on the university servers. To address limitations from past research where the natural condition presented beautiful natural scenes, and the urban condition presented unpleasant industrial environments, the built environment images were artistic and colorful capturing various architecture and urban scenery. The built environment and natural images were roughly matched on contour and color and were presented in the same order. The images used in these experiments are presented in the supplemental materials.

Data Quality Measures.

Following best practices in crowdsourcing research (e.g., Aguinis et al., 2021), we used several data quality measures to detect misrepresentation or inattention. First, we compared the reported age, sex, and race between the first and second surveys. Second, we used Qualtrics-generated timestamps before and after the slideshows to flag participants who did not watch the slideshows in full. Participants who continued the survey faster than the duration of the slideshow were not invited to complete the second survey and were excluded from analysis. Third, three attention checks were embedded in each session (total of six across the two sessions). Two attention checks were embedded in the delay discounting task (e.g., Would you prefer $0 now or $100 in a week) in which only one response is sensical. The third attention check was an instructional item that asked participants to remember a word for a later part of the survey. As attention may fluctuate during survey completion, and failing certain attention checks may not reflect inattention per se (Almog et al., 2023), we followed recommendations to treat performance on the attention checks as a continuum and allowed one error (Nichols & Edlund, 2020). Thus, failing two or more (out of six across the two sessions) was determined as poor-quality data resulted in exclusion from analyses.

Assessment of Recent Substance Use.

Lastly, as recent substance use might affect demand for substances in laboratory conditions (Amlung et al., 2015) and among remote samples (Almog et al., 2025), participants were asked about recent substance use. Those who reported using a substance within three hours before or during study participation in one or two of the sessions were excluded from the analysis as current intoxication may impact delay discounting and behavioral economic demand (see Almog et al., 2025, for secondary analysis of the influence of very recent substance use on discounting and demand independent of visual stimuli). However, we did not exclude participants who reported only nicotine cigarette smoking as this was expected to be prevalent in the naturalistic conditions of an online sample and is unlikely to impair decision-making in similar ways that recent alcohol or cannabis use might.

Measures

Delay Discounting.

Participants completed a hypothetical monetary titrating amount delay discounting task (Du et al., 2002; Rodzon et al., 2011). Participants were asked to choose between a smaller reward received immediately or a larger reward received with a delay (e.g., “Would you rather have $50 now or $100 in one month?”). Participants were presented with seven delays (1 week, 2 weeks, 1 month, 6 months, 1 year, 5 years, 25 years) with six questions in each delay block. In each block, the larger delayed reward was always $100, and the first smaller-immediate reward was $50, which changed according to the participant’s choice. The adjusted amount of the smaller-immediate reward was titrated by half from trial to trial (titrated by $25 after the first choice, $12.50 after the second, $6.25 after the third, etc.). The subsequent value of the immediate outcome following the final choice within each delay block served as the indifference point. Using the indifference points we calculated the ordinal area-under-the-curve value (Myerson et al., 2001, Borges et al., 2016) and used it as the delay discounting outcome in subsequent analyses. The delay discounting AUC (DD-AUC) values range from 0 to 1, where lower values represent greater delay discounting (reflecting preference for immediate smaller rewards, or more “impulsive” decision-making), and higher values represent lower discounting (reflecting preference for delayed larger rewards, or less “impulsive” decision-making).

Demand for a Substance.

Demand for the most frequently used substance was assessed with a state hypothetical drug purchase task – an alcohol purchase task (based on Murphy & MacKillop, 2006), a cannabis purchase task (based on Aston et al., 2021), and a cigarette purchase task (based on MacKillop et al., 2008). In the purchase task, participants were asked how many units of their substance (i.e., standard drinks for alcohol, grams for cannabis, number of cigarettes) they would purchase at that moment to use during the next five hours/one week/one day (for alcohol/cannabis/cigarettes respectively) with escalating prices, from free ($0) and up to $30/$60/$140, for alcohol/cannabis/cigarettes respectively. Replicating previous research, all vignettes asked participants to assume: (1) they can only purchase the substance from this source; (2) they cannot use any substance they already have at home; (3) they have their typical money/income when they make the decisions; (4) they did not use any substance before making these decisions; (5) they will consume all of the substance they choose to purchase within the designated timeframe, they cannot save it, share it, or sell it to others; and (6) the substance they purchase is similar in quality and strength to their typical substance.

A non-linear regression model (Koffarnus et al., 2015) was fit to the mean group purchase data to produce a curve and to the individual’s data to produce the demand indices (see Almog et al., 2025 for more details on the equation parameters). To analyze potential differences between the environments we used the observed intensity (i.e., consumption when the substance is free) and the derived rate of change of elasticity (i.e., alpha, reflecting sensitivity to price increases). Higher intensity and lower alpha values reflect greater demand, or higher valuation of the substance. These indices represent two dimensions of demand. Intensity reflects ‘amplitude’ of consumption, and rate of change of elasticity reflects ‘persistence’ of consumption (MacKillop et al., 2009). Intensity and rate of change of elasticity can show large effect sizes reflecting each dimension (Zvorsky et al., 2019), and are frequently the focus of studies evaluating demand (e.g., Bergeria et al., 2020, Greenwald et al., 2020, Strickland et al., 2019). Moreover, preselecting two demand indices (rather than all) aids in avoiding a Type I error.

Order of Experimental Conditions.

To evaluate a potential order effect, in which the order of the conditions might affect decision-making (i.e., natural stimuli in the first session and built stimuli in the second, and vice versa), a dichotomous variable was coded to reflect whether the participant viewed the natural stimuli in the first or second session.

Typical Substance Use.

To quantify typical substance use of the most frequently used substance, participants answered questions on frequency and quantity of substance use on a typical day of use. Participants reported the number of days in the past month they used their main group-affiliated substance, and the number of substance units they consumed on a typical day they used the substance (i.e., standard alcoholic drinks, grams of cannabis, cigarettes). To assist with the estimation, the alcohol and cannabis groups were presented with an image presenting different sizes of standard drinks or cannabis joints.

Demographics.

In the first session, participants were asked for their age in years, sex, race, ethnicity, education, and income.

Data Analysis

All analyses were conducted using Excel, SPSS (version 29.0.0.0), and GraphPad Prism (version 10.0.2). All distributions were checked for normality based on visual inspection and skewness/kurtosis. Distributions with skewness values equal or higher than |2| or kurtosis values equal or higher than |4| were determined as nonnormal (Kim, 2013). When the main variables (i.e., DD-AUC, demand intensity, demand alpha) had non-normal distribution values were log-transformed to allow for parametric tests. We report the analyses results including effect size with 95% confidence intervals, to quantify the primary within-subject effect of visual exposure to natural (vs. built) stimuli.

Characterizing the Overall Sample and Subsamples.

The full sample, the two order-based groups (i.e., Natural first or Built first), and the three substance-based groups (i.e., alcohol, cannabis, cigarette) were described for demographic and typical substance use characteristics. To assess baseline differences in demographics and typical substance use across the order-based groups, we used parametric or nonparametric (Mann-Whitney) t-tests, based on normality of distributions, and Fisher’s Exact test/chi-square tests for categorical variables.

Delay Discounting.

To characterize and assess data quality, discounting data were evaluated for systematicity based on commonly used criteria (Johnson & Bickel, 2008). In general, systematic discounting data would be expected to exhibit a reduction in indifference points with increased delays. Per the task used in this study, the first criterion (i.e., Bounce) flagged an observation as nonsystematic if there was an increase greater than $20, from one indifference point to the next. The second criterion (i.e., Trend) flagged an observation as nonsystematic if the reduction from the first to last indifference points was smaller than $10. A titration monetary task, such as the task used in the present study, can yield between 16 and 21% nonsystematic discounting data (Smith et al., 2018). In the present dataset, of 342 observations (across the two conditions), 14 (4.1%) were nonsystematic (11 violated the first “Bounce” criterion, and 3 violated the second “Trend” criterion). Based on the low rate of nonsystematic data, and the use of a manipulation that was aimed to influence discounting (Stein et al., 2016) all participants were included in the delay discounting analyses. DD-AUC values in the two conditions were normally distributed.

To test the differences in DD-AUC between conditions (Natural vs. Built), a mixed model (between-within subjects) analysis of variance (ANOVA), with Bonferroni-adjusted post hoc pairwise comparison tests, was used to assess the main effect of environmental condition (i.e., Natural vs. Built) and the interaction between environmental condition and order of condition presentation. The dependent variable was the DD-AUC, the within-subject factor was the environmental condition (Natural and Built), and the between-subject factor was the order of the conditions (i.e., those who viewed the natural stimuli first vs. those who viewed the built stimuli first). To verify the effect emerged regardless of the most frequently used substance, we explored the analysis in each substance-based group separately. To detect a small to medium effect size with power of 0.8 and error probability of 0.05 a power analysis required a minimum sample of 34 to 51 participants for each 2X2 mixed ANOVA. To address the remote MTurk limitations and reach the minimum sample needed for each substance-based group, we aimed to obtain a sample of at least 150 participants (for three substance-based groups).

Demand for a Substance.

Demand data were analyzed for the three substance-based groups separately (i.e., alcohol, cannabis, cigarette groups). Demand data were examined for systematicity following commonly used criteria (Stein et al., 2015). The three criteria for identifying nonsystematic demand data, per the used tasks, are: (a) Trend; if there was no global reduction in consumption due to increasing price, (b) Bounce; if there was more than one local increase in consumption (of at least 25% of the maximal consumption when the substance is free) with increasing price, and (c) Reversal from Zero: if there was a non-zero response following two zero consumption responses across consecutive increasing prices. Nonsystematic data, which may indicate lack of understanding or attention to the task, were excluded from the demand analysis. Participant data were excluded if the maximal consumption was generally unrealistic (e.g., 100 alcoholic beverages in 5 hours) or if the within-subject demand data across conditions appeared unrealistic, defined as a difference larger than 8x across conditions at the same price (e.g., consumption at no cost increased from 5 to 99 grams of cannabis across sessions).

A two-stage approach was used for analysis. First, the exponentiated model (Koffarnus et al., 2015), which is frequently used and can accommodate data points of zero consumption was fit to the individual purchase task data to yield an individual alpha (rate of change of elasticity) value for each participant. Second, the observed intensity and derived alpha values were used in subsequent between-within ANOVA (after transformation if needed) with Bonferroni-adjusted post hoc pairwise comparison tests, to assess the main effect of environmental condition (Natural vs. Built) and any potential order effect.

Experiment 1 Results

Participant and Data Characteristics

Of 353 participants who completed the first survey, 272 were invited to the second survey, of whom 226 completed both sessions. Fifty-six participants were excluded from analysis due to: using a substance before or during participation in at least one of the sessions (n = 49), poor data quality or error (n = 4), and participants whose most frequently used substance was opioids (n = 3). See Figure 1 for study participation flowchart. Of the included 170 participants, 15 participants failed only one attention check (out of six across the two sessions), and none failed more than one. All reported consistent age, sex, and race across sessions, and provided good quality data overall. Thus, all 170 participants were included in the analysis. The substance-based groups included 64 participants in the alcohol group, 56 participants in the cannabis group, and 50 participants in the cigarette group. Participant demographic and typical substance use characteristics of the full sample are presented in Table 1. The full sample (N = 170, 56.5% female) had a mean age of 41.1, was mostly white (87.6%), and non-Hispanic (91.8%). Sample characteristics of the substance-based groups are presented in Table 2. In the alcohol group, the mean number of drinking days in the past month was 20.6 days (SD = 7.10) with a mean number of alcoholic drinks per drinking day of 4.3 (SD = 3.48). In the cannabis group, the mean number of days of cannabis use in the past month was 22.7 days (SD = 8.10), with a mean number of grams per cannabis-using day of 1.1 (SD = 1.32). In the cigarette group, the mean number of smoking days in the past month was 29.0 days (SD = 3.69) with a mean number of cigarettes per smoking day of 17.5 (SD = 10.17).

Figure 1.

Experiment 1 Study Participation Flowchart

Table 1.

Experiment 1: Sample Characteristics of the Full Sample and Order-Based Groups.

	Full sample N = 170	Natural First N = 82	Built First N = 88	p
Age in years, Median (Q1 – Q3)	40.5 (32–49.25)	42 (33–50)	40 (30.25–49)	.569
Sex, Female, n (%)	96 (56.5)	47 (57.3)	49 (55.7)	.878
Race, n (%)				.020
American Indian/Alaska Native	1 (0.6)	0 (0.0)	1 (1.1)
Asian	5 (2.9)	1 (1.2)	4 (4.5)
Black/African American	8 (4.7)	1 (1.2)	7 (8.0)
White	149 (87.6)	77 (93.9)	73 (83.0)
Other/Mixed	7 (4.1)	3 (3.7)	3 (3.4)
Ethnicity, n (%)				1.000
Hispanic/LatinX	14 (8.2)	7 (8.5)	7 (8.0)
Education, n (%)				.584
Less than high school	1 (0.6)	1 (1.2)	0 (0.0)
High school graduate	16 (9.4)	10 (12.2)	6 (6.8)
Some college (no degree)	47 (27.6)	24 (29.3)	23 (26.1)
Associate degree (2 year)	17 (10.0)	7 (8.5)	10 (11.4)
Bachelor’s degree (4 year)	61 (35.9)	29 (35.4)	32 (36.4)
Master’s degree	20 (11.8)	7 (8.5)	13 (14.8)
Doctoral degree	4 (2.4)	2 (2.4)	2 (2.3)
Professional degree (MD, JD)	4 (2.4)	2 (2.4)	2 (2.3)
Income, n (%)				.099
$25K or less	42 (24.7)	22 (26.8)	20 (22.7)
$26–50K	44 (25.9)	27 (32.9)	17 (19.3)
$51–75K	36 (21.2)	12 (14.6)	24 (27.3)
$76–100K	23 (13.5)	9 (11.0)	14 (15.9)
$101–125K	8 (4.7)	4 (4.9)	4 (4.5)
$126–150K	6 (3.5)	5 (6.1)	1 (1.1)
$151K or more	6 (3.5)	2 (2.4)	4 (4.5)
Prefer not to say	5 (2.9)	1 (1.2)	4 (4.5)
Days of substance use in past month, Median (Q1 – Q3)	28 (16.75 – 30)	30 (20 – 30)	26 (15 – 30)	.186
Standardized units of substance per day of use, Median (Q1 – Q3)		−0.08 (−.66 - .55)	−.22 (−.65 - .21)	.327

Note. Percentages refer to order-based groups (i.e., columns) and may not add to 100 due to rounding. To describe the order-based groups with one variable reflecting typical quantity of substance use across the three substances, the number of substance units (i.e., drinks, grams, cigarettes) used on a typical day of use was standardized within each group before merged together. Age and days of substance use in past month had normal distributions across the groups and were compared with a parametric t-test. Standardized units of substance per day of use had non-normal distributions and were assessed with a Mann-Whitney test. To meet the chi square test assumptions of minimum of 80% of cells with minimum counts, Race was truncated to White or Non-White, four Education categories were truncated to two (High school graduate and less than high school; Doctoral degree and Professional degree (MD, JD)), and the three high income categories were truncated (101K and higher). Sex, Race, and Ethnicity were assessed with Fisher’s exact tests.

Table 2.

Experiment 1: Sample Characteristics of the Substance-Based Groups.

Variable	Alcohol group n = 64	Cannabis group n = 56	Cigarette group n = 50
Age in years, Median (Q1--Q3)	41 (31–49)	35 (29–48.75)	43 (37–51)
Sex, Female, n (%)	32 (50.0)	29 (51.8)	35 (70.0)
Race, n (%)
American Indian/Alaska Native	0 (0.0)	1 (1.8)	0 (0.0)
Asian	3 (4.7)	2 (3.6)	0 (0.0)
Black/African American	2 (3.1)	2 (3.6)	4 (8.0)
White	56 (87.5)	48 (85.7)	45 (90.0)
Other/Mixed	3 (4.7)	3 (5.4)	1 (2.0)
Ethnicity, n (%)
Hispanic/LatinX	2 (3.1)	9 (16.1)	3 (6.0)
Education, n (%)
Less than high school	0 (0.0)	0 (0.0)	1 (2.0)
High school graduate	3 (4.7)	4 (7.1)	9 (18.0)
Some college (no degree)	11 (17.2)	18 (32.1)	18 (36.0)
Associate degree (2 year)	4 (6.3)	4 (7.1)	9 (18.0)
Bachelor’s degree (4 year)	31 (48.4)	21 (37.5)	9 (18.0)
Master’s degree	11 (17.2)	6 (10.7)	3 (6.0)
Doctoral degree	2 (3.1)	2 (3.6)	0 (0.0)
Professional degree (MD, JD)	2 (3.1)	1 (1.8)	1 (2.0)
Income, n (%)
$25K or less	9 (14.1)	19 (33.9)	14 (28.0)
$26–50K	15 (23.4)	15 (26.8)	14 (28.0)
$51–75K	12 (18.8)	16 (28.6)	8 (16.0)
$76–100K	11 (17.2)	3 (5.4)	9 (18.0)
$101–125K	5 (7.8)	2 (3.6)	1 (2.0)
$126–150K	5 (7.8)	1 (1.8)	0 (0.0)
$151K or more	4 (6.3)	0 (0.0)	2 (4.0)
Prefer not to say	3 (4.7)	0 (0.0)	2 (4.0)
Days of substance use in past month, Median (Q1--Q3)	20 (15–27.75)	26 (15–30)	30 (30–30)
Units of substance per day of use, Median (Q1--Q3)	3 (2–5)	.88 (.25–1.0)	15 (10–20.5)

Note. Percentages refer to substance-based groups (i.e., columns) and may not add up to 100 due to rounding. Units of substance per day of use = alcoholic drinks in the alcohol group, grams of cannabis in the cannabis group, and cigarettes in the cigarette group.

Participants were randomized to the Natural or Built condition in the first session. Participants either viewed the natural environment stimuli first and built second (n = 82, 48.2%) or built environment stimuli first and natural second (n = 88, 51.8%). Demographics and typical substance use of the order-based groups are presented in Table 1. The order-based groups in the full sample (and overall in the alcohol, cannabis, and cigarette groups separately) did not differ based on age, sex, ethnicity, education, income, the number of substance-using days in the past month, or the number of substance units used in a day of use. The group who viewed the built environment stimuli first had significantly more non-white participants than the group who viewed the natural stimuli first (18.2% vs. 6.1%, respectively).

The subset of participants who did not return to, or did not complete the second survey (n = 46) was significantly younger than the analyzed sample (M = 34.2, SD = 9.0, and M = 41.1, SD = 11.6, respectively), but did not differ on any other demographic characteristic, number of days of substance use in the past month, or allocation to experimental condition (see Table S1 in the supplemental material).

Delay Discounting

Figure 2 presents the DD-AUC results for the full sample. A between-within ANOVA using DD-AUC values as the dependent variable showed a significant interaction between condition and order, F(1,168) = 7.96, p = .005, η²p = .05. Post hoc comparisons showed that DD-AUC in the Natural condition was higher than in the Built condition only for those who viewed the natural stimuli in the second session, with a medium effect size, M_(Natural) = .68, SE = .02, M_(Built) = .64, SE = .02, p < .001, d = .42, 95% CI (.20, .63), with no differences between conditions for those who viewed the natural stimuli in the first session, M_(Natural) = .61, SE = .02, M_(Built) = .61, SE = .02, p = .687, d = −.04, 95% CI (−.26, .18). Results and conclusions remained similar after excluding 11 participants that violated the first criterion of systematicity (Johnson & Bickel, 2008). Overall, similar patterns in DD-AUC remained across the cannabis and alcohol groups when analyzed separately but were less evident in the cigarette group. Figure S1 presents the delay discounting results of each of the substance-based groups.

Figure 2.

Experiment 1: Delay Discounting of the Full Sample

Note. Delay discounting area-under-the-curve (DD-AUC) between-within ANOVA results (N = 170). Main effect of condition within-subject of the full sample (panel A) suggests lower discounting following visual exposure to natural environments. The interaction between order and condition (panel B) suggests lower discounting following visual exposure to natural environments only when the natural stimuli were presented in the second session. Bars and error bars represent the mean ordinal DD-AUC and the standard error of the mean.

*p < .05, ***p < .001

Substance Demand

Means and standard deviations of all outcomes of the full sample and of order-based groups are presented in Table 3.

Table 3.

Experiment 1: Means and Standard Deviations of the Full Sample and Order-Based Groups.

	Full Sample					Natural first, Built second					Built first, Natural second
Environmental stimuli		Natural		Built			Natural		Built			Natural		Built
Variable	n	M	SD	M	SD	n	M	SD	M	SD	n	M	SD	M	SD
DD-AUC	170	0.64	0.20	0.63	0.20	82	0.60	0.19	0.61	0.20	88	0.68	0.20	0.64	0.20
Cannabis intensity	40	28.46	31.77	30.46	33.50	20	26.98	32.80	34.13	35.77	20	29.95	31.49	26.80	31.56
Cannabis alpha	40	0.0016	0.0021	0.0016	0.0024	20	0.0013	0.0008	0.0011	0.0008	20	0.0020	0.0028	0.0021	0.0032
Alcohol intensity	60	6.97	4.10	7.40	4.93	29	6.62	3.94	7.62	5.16	31	7.29	4.28	7.19	4.79
Alcohol alpha	60	0.0094	0.0157	0.0087	0.0214	29	0.0099	0.0158	0.0118	0.0302	31	0.0089	0.0159	0.0057	0.0053
Cigarette intensity	50	26.94	19.07	24.76	16.13	25	31.40	23.46	28.20	19.82	25	22.48	12.26	21.32	10.66
Cigarette alpha	50	0.0063	0.0039	0.0069	0.0082	25	0.0051	0.0029	0.0050	0.0027	25	0.0075	0.0043	0.0089	0.0110

Demand for Cannabis.

Of 56 participants in the cannabis group, 16 participants were excluded from the demand analysis for providing nonsystematic demand data in at least one of the sessions (n = 9; six failed the Trend criterion, and three others failed the Bounce criterion presenting three or more increases in consumption with increasing price), unrealistic data across conditions (i.e., difference of more than 8x between conditions at first or last price, n = 5), or reporting confusion following task completion (n = 2). The final cannabis group for the demand analysis included 40 participants. The exponentiated model (with k parameter set to 2.5) provided good fit to mean group data, R²_(Natural) > .99, R²_(Built) > .99; RMSE_(Natural) = .75, RMSE_(Built) = .68, and individual data (Q1 - Q3), R²_(Natural) = .90 - .97, R²_(Built) = .90 - .97; RMSE_(Natural) = .61 – 2.76, RMSE_(Built) = .56 – 3.01. Figure 3 depicts mean within-subject demand curves across conditions (panel A) and of the two order-based subgroups (panels B and C).

Figure 3.

Experiment 1: Demand for Cannabis, Alcohol, and Cigarettes, Within-Subject and Order Effect

Note. Cannabis/alcohol/cigarette consumption as a function of price for both natural and built conditions. The x-axis (logarithmic) represents the price of a single cannabis gram/alcoholic drink/cigarette. The y-axis represents the number of cannabis grams/alcoholic drinks/cigarettes purchased. Symbols represent mean observed consumption. Curves represent best fit of the exponentiated model to the data points (Koffarnus et al., 2015). Panels A, B, and C present demand for cannabis. Panel A depicts mean group within-subject demand data (n = 40). Panel B depicts the demand curves for the subgroup of participants who experienced the Natural condition in the first session (n = 20). Panel C depicts the demand curves for the subgroup of participants who experienced the Natural condition in the second session (n = 20). Panels D, E, and F present demand for alcohol. Panel D depicts mean group within-subject demand data (n = 60). Panel E depicts the demand curves for the subgroup of participants who experienced the Natural condition in the first session (n = 29). Panel F depicts the demand curves for the subgroup of participants who experienced the Natural condition in the second session (n = 31). Panels G, H, and I present demand for cigarettes. Panel G depicts mean group within-subject demand data (n = 50). Panel H depicts the demand curves for the subgroup of participants who experienced the Natural condition in the first session (n = 25). Panel I depicts the demand curves for the subgroup of participants who experienced the Natural condition in the second session (n = 25). Some of the error bars are hidden by the symbols.

Intensity.

Figure 4 depicts the mean non-transformed within-subject cannabis demand indices (panels A-D) of the two cannabis subgroups based on the order of the conditions. A between-within ANOVA with observed intensity (i.e., consumption when the substance was free) values revealed a significant interaction between environmental condition and order, F(1,38) = 8.00, p = .007, η²p = .17. Post hoc pairwise comparisons showed that intensity in the Natural condition was lower (i.e., lower demand) compared to the Built condition only for the subgroup of participants who viewed the natural stimuli in the first session (n = 20), with medium effect size, M_(Natural) = 26.98, SE = 7.19, M_(Built) = 34.13, SE = 7.54, p = .008, d = −.55, 95% CI (−1.02, −.07), but not for those who viewed the built stimuli first and the natural stimuli second (n = 20), M_(Natural) = 29.95, SE = 7.19, M_(Built) = 26.80, SE = 7.54, p = .229, d = .32, 95% CI (−.14, .76).

Figure 4.

Experiment 1: Cannabis and Alcohol Demand Indices, Within-Subject and Order Effect

Note. Non-transformed mean demand indices of cannabis (panels A-D) and alcohol (panels E-H) within-subject and in order-based subgroups. Intensity = consumption when substance is free. Alpha = rate of change of elasticity, indicating sensitivity to price increases. Significance is based on Bonferroni-adjusted pairwise comparisons from the between-within ANOVAs. Error bars represent the standard error of the mean.

*p < .05 **p < .01

Rate of Change of Elasticity.

Similarly, a between-within ANOVA was conducted with log-transformed alpha values (i.e., rate of change of elasticity), and revealed a significant interaction between environmental condition and order, F(1,38) = 9.68, p = .004, η²p = .20. Post hoc pairwise comparisons showed that rate of change of elasticity in the Natural condition was higher (i.e., participants were more sensitive to price increases), compared to the Built condition, only for the subgroup of participants who viewed the natural stimuli in the first session, with a medium effect size, M_(Natural) = −3.04, SE = .10, M_(Built) = −3.14, SE = .10, p = .013, d = .60, 95% CI (.12, 1.07), but not for those who viewed the built stimuli in the first session, M_(Natural) = −2.95, SE = .10, M_(Built) = −2.88, SE = .10, p = .080, d = −.39, 95% CI (−.84, .07).

Demand for Alcohol.

Of 64 participants in the alcohol group, four participants were excluded. One participant provided nonsystematic demand data, violating the second criterion “Bounce” (with 2 or more increases in consumption that are higher than 25% of maximal consumption). Two participants provided unrealistic consumption of 30 drinks or higher over five hours, and one provided unrealistic data across conditions (i.e., difference of more than 8x between conditions at first or last price). The final alcohol group in the demand analysis included 60 participants. The exponentiated model (with k parameter set to 1.8) provided excellent fit to mean group demand, R²_(Natural) > .99, R²_(Built) > .99; RMSE_(Natural) = .15, RMSE_(Built) = .10, and individual data (Q1 - Q3), R²_(Natural) = .86 - .96, R²_(Built) = .86 - .95; RMSE_(Natural) = .34 - .85, RMSE_(Built) = .39 - .86. Figure 3 depicts mean within-subjects alcohol demand data across conditions (panel D), and of the two order-based subgroups (panels E and F).

Intensity.

Figure 4 depicts the mean non-transformed within-subject demand indices (panels E-H) of the two alcohol subgroups based on the order of the conditions. A between-within ANOVA with observed intensity values (i.e., consumption when the substance is free) revealed no interaction between environmental condition and order, F(1,58) = 2.68, p = .107, η²p = .04, and no main effect of environmental condition, F(1,58) = 1.82, p = .183, η²p = .03.

Rate of Change of Elasticity.

A between-within ANOVA with log-transformed alpha values (i.e., regression derived rate of change of elasticity) revealed no interaction between environmental condition and order, F(1,58) = .01, p = .920, η²p = .00, and no main effect of environmental condition, F(1,58) = 1.91, p = .172, η²p = .03.

Demand for Cigarettes.

All 50 participants in the cigarette group were included in the analysis. The exponentiated model (with k parameter set to 2.5) provided excellent fit to mean group demand, R²_(Natural) > .99, R²_(Built) > .99; RMSE_(Natural) = .52, RMSE_(Built) = .68, and individual data (Q1 - Q3) R²_(Natural) = .96 - .98, R²_(Built) = .96 - .98; RMSE_(Natural) = 1.09 – 2.17, RMSE_(Built) = 1.01 – 2.13. Figure 3 depicts mean within-subjects alcohol demand data across conditions (panel G) and of the two order-based subgroups (panels H and I).

Intensity.

A between-within ANOVA with log-transformed intensity (i.e., consumption when the substance is free) values revealed no interaction between environmental condition and order, F(1,48) = .20, p = .660, η²p = .00, and no main effect of condition, F(1,48) = 1.90, p = .175, η²p = .04.

Rate of Change of Elasticity.

A between-within ANOVA with log-transformed alpha (i.e., rate of change of elasticity) values revealed no interaction between environmental condition and order, F(1,48) = .06, p = .816, η²p = .00, and no main effect of condition, F(1,48) = .38, p = .543, η²p = .01.

Experiment 1 Discussion

Visual exposure to natural (vs. built) environments reduced delay discounting, supporting Hypothesis 1. The effect of visual exposure to natural stimuli on delay discounting was most apparent when the natural stimuli were presented second, with a medium effect size. Hypothesis 2 (i.e., visual exposure to natural versus built environments would reduce substance demand) was partially supported, and mostly among participants who most frequently use cannabis. Interestingly, an order effect also emerged in the demand data. However, demand for cannabis appeared to be lower in the Natural condition when the natural stimuli were presented in the first session, with medium effect size. No effect was observed for demand for alcohol or cigarettes, suggesting not all populations of individuals who use various substances respond similarly to visual exposure to natural as opposed to built environments, at least among MTurk participants. To address the MTurk limitations of remote participation with less controlled conditions and professional survey responders, Experiment 2 was conducted to replicate Experiment 1 among a sample of university students in a laboratory setting.

Experiment 2

Because cigarette smoking is less prevalent among university students, Experiment 2 focused on students who reported regular use of alcohol or cannabis. Anticipating a potential larger effect in-laboratory compared to the remote crowdsourcing sample where the exposure manipulation (e.g., screen size) and assessment conditions are better controlled, increased power of within-subjects designs (compared to previous between-subject designs, Berry et al., 2015, 2014), and balancing resources, an analyzed minimum sample of 25 participants was targeted. Power calculations required a sample of 27 to detect a medium effect with power of 0.80, and error probability of 0.05 when using a t-test. Similar sample sizes (n = ~30) have also been used in other laboratory studies that investigated novel manipulations on decision making processes (Hudson et al., 2024, Luo & Monterosso, 2014, Guan et al., 2015). Overall, all measures and analyses were similar to those employed in Experiment 1 with two exceptions. Because the sample was substantially smaller, we inferentially tested only the main effect of environmental stimuli. Acknowledging the sample may be underpowered in general, and more so for the demand for substances analyses, we focused on effect sizes, which can inform future studies. Second, the cannabis purchase task was modified to a smaller unit of cannabis (i.e., hits rather than grams) to accommodate potentially lower rates of cannabis use among students compared to online adult populations.

Experiment 2 Methods

Participants and Procedures

Participants were 18 years or older, undergraduate or graduate students who reported using alcohol or cannabis at least three times in the past month. Flyers across campus, course announcements, lab registry, and word of mouth were used for recruitment. Main data collection began in September 2023 and ended in March 2024. Participants first completed a short independent screener online, and only those who met inclusion criteria of substance use frequency were invited by email to participate. All procedures of the study were similar to Experiment 1, except the location of study completion, and compensation. Participants arrived at our laboratory space twice, one week apart, and completed the questionnaires and behavioral economic tasks in a quiet room, with no access to their phones, using a desktop computer with a 24-inch monitor. Participants were paid a total of $10.00 (Amazon gift card) upon study completion. All procedures were approved by the University of Florida Institutional Review Board under the same protocol of Experiment 1 #IRB201902033.

Measures

In general, all measures with the exception of the cannabis purchase task were identical to Experiment 1 (including the delay discounting task and alcohol purchase task).

Demand for Cannabis.

The cannabis purchase task used hits of cannabis as the unit of purchase rather than grams (based on the Marijuana Purchase Task, Aston et al., 2015). Participants were asked to indicate the number of hits they would purchase at that moment to use over a week, across eighteen prices (from $0 to $20 per hit).

Data Analysis

All data preparation procedures were identical to Experiment 1, using ordinal AUC (Borges et al., 2016) for delay discounting and the exponentiated model (Koffarnus et al., 2015) for the demand data. However, unlike in Experiment 1, due to the smaller sample of Experiment 2, only within-subject differences in DD-AUC, demand intensity, and demand alpha were examined using paired parametric (with data transformation if needed to achieve normality of distribution) or non-parametric t-tests. Due to the small sample size, we anticipated reduced power for the separated alcohol and cannabis demand analyses, and thus focused our findings on effect size and 95% confidence intervals of the effect size for parametric t-test, which can be used to inform future research.

Experiment 2 Results

Participants

A total of 31 participants were recruited from which 30 completed both study sessions. One participant was excluded from analysis due to cannabis use within 3 hours of participation. Replicating Experiment 1, recent nicotine use was not exclusionary. Table 4 presents the demographics and mean typical substance use of the analyzed sample (n = 29, 58.6% female) and the two substance-based groups. On average, participants were 21.1 years old (SD = 2.14), mostly undergraduate students (93.1%), White (79.3%), and non-Hispanic (62.1%). The alcohol group included 16 participants with a mean of 7.6 days (SD = 3.29) of alcohol use in the past month and a mean of 3.7 alcoholic drinks per day of use (SD = 2.37). The cannabis group included 13 participants with a mean of 21.8 days (SD = 10.27) of cannabis use in the past month and a mean of 0.42 grams per day of use (SD = .30). Of six attention checks across the two sessions, only one participant failed one check embedded in the delay discounting data, thus all participant data were included in the analyses.

Table 4.

Experiment 2. Student Sample Characteristics

Variable	Full sample N = 29	Alcohol group n = 16	Cannabis group n = 13
Age (years), Median (Q1--Q3)	21 (20–22)	21 (20.25–22)	21 (19–22)
Female, n (%)	17 (58.6)	9 (56.3)	8 (61.5)
Race, n (%)
Asian	1 (3.4)	1 (6.25)	0 (0.0)
Black	2 (6.9)	2 (12.5)	0 (0)
White	23 (79.3)	12 (75.0)	11 (84.6)
Mixed/Other	3 (10.4)	1 (6.25)	2 (15.4)
Ethnicity, n (%)
Hispanic	11 (37.9)	8 (50.0)	3 (23.1)
Level of Education, n (%)
Undergraduate Student	27 (93.1)	15 (93.8)	12 (92.3)
PhD Student	2 (6.9)	1 (6.2)	1 (7.7)
Days of substance use in past month, Median (Q1--Q3)	10 (5–26)	7 (5–9.75)	27 (11.50–30)
Units of substance per day, Median (Q1-- Q3)		3 (2.5–4.75)	0.3 (0.15–0.75)

Note. Percentages refer to columns. Units of substance per day of use = alcoholic drinks in alcohol group, and grams of cannabis in cannabis group.

Delay Discounting

Three delay discounting observations in the first session were nonsystematic, violating the Bounce criterion (Johnson & Bickel, 2008). Following Experiment 1 procedures, all observations were included in the DD-AUC analysis. DD-AUC values in both conditions were normally distributed. There were no differences in DD-AUC within-subject across the Natural versus Built conditions, M_(Natural) = .66 (SD = .17), M_(Built) = .67 (SD = .15), t (28) = −.50, p = .618, d = −.09, 95% CI (−.46, .27). Results remained non-significant even when excluding the three participants who provided nonsystematic data (Johnson & Bickel, 2008).

Demand for Cannabis

The cannabis group for the demand analysis included 13 participants. All cannabis purchase data were systematic. Figure 5 (panel A) depicts mean within-subjects demand curves across conditions. The exponentiated model (with k parameter set to 2.6) provided good fit to mean group demand, R²_(Natural) > .99, R²_(Built) > .99; RMSE_(Natural) = 1.52, RMSE_(Built) = 1.11 and to individual data (Q1-Q3), R²_(Natural) = .94 - .97, R²_(Built) = .94 - .98; RMSE_(Natural) = 1.20 – 4.35, RMSE_(Built) = 1.07 – 4.19. Figure 5 depicts the mean demand intensity (Panel B) and rate of change of elasticity (i.e., alpha, Panel C) in the two conditions. There were no differences within-subject across the Natural versus Built conditions stimuli in log transformed intensity, M_(Natural) = 1.54 (SD = .34), M_(Built) = 1.52 (SD = .32), t (12) = .68, p = .510, d = .19, 95% CI (−.36, .73). No differences were found in rate of change of elasticity, M_(Natural) = .0030 (SD = .0020), M_(Built) = .0030 (SD = .0018), t(12) = .01, p = .992, d = .00, 95% CI (−.54, .55).

Figure 5.

Experiment 2: Demand for Cannabis and Alcohol Within-Subject

Note. Panel A depicts within-subject (n = 13) mean cannabis purchasing data as a function of price in the two conditions. The x-axis (logarithmic) represents the price of a single cannabis hit. The y-axis represents the number of hits purchased. Circles and bars represent the mean and standard error of the mean of the observed purchasing values (cannabis hits). Panels B and C depict mean non-transformed within-subject intensity (consumption when free; B) and rate of change of elasticity (alpha; C). Analysis revealed no differences in intensity or alpha within subjects across conditions. Panel D depicts within-subject (n = 16) mean alcohol purchasing data as a function of price in the two conditions. The x-axis (logarithmic) represents the price of a single alcoholic drink. The y-axis represents the number of alcoholic drinks purchased. Circles and bars represent the mean and standard error of the mean of the observed purchasing values (alcoholic drinks). Curves represent the best fit of the exponentiated model to the data (Koffarnus et al., 2015). Panels E and F depict within-subject intensity (consumption when free; E) and rate of change of elasticity (alpha; F). Significantly lower intensity and higher alpha values were revealed after viewing the natural stimuli.

*p < .05, **p < .01

Demand for Alcohol

All participants in the alcohol group (n = 16) provided systematic purchase task data. Figure 5 (panel D) depicts mean within-subjects demand curves across conditions. The exponentiated model (with k parameter set to 1.9) provided excellent fit to mean group demand, R²_(Natural) > .99, R²_(Built) = .98; RMSE_(Natural) = .13, RMSE_(Built) = .42, and individual data (Q1 - Q3) R²_(Natural) = .88 - .95, R²_(Built) = .92 - .94; RMSE_(Natural) = .40 - .64, RMSE_(Built) = .44 - .85. Figure 5 depicts the mean demand intensity (i.e., consumption when the substance is free, Panel E) and rate of change of elasticity (i.e., alpha, Panel F) in the two conditions. A paired t-test with log transformed intensity values revealed significantly lower intensity, with a large effect size, following visual exposure to the natural compared to built stimuli, M_(Natural) = .76 (SD = .25), M_(Built) = .84 (SD = .24), t (15) = −3.47, p = .003, d = −.87, 95% CI (−1.44, −.28). The distribution of alpha values could not be normalized, hence we used a nonparametric paired t-test that revealed significantly higher alpha (greater sensitivity to price increases), with large effect size, following visual exposure to the natural stimuli, Mdn (Q1-Q3) Natural = 0.0057(.0035 - .0106), Built = .0041(.0030 - .0068), Wilcoxon = 28.00, p = .039, r = −.51.

Experiment 2 Discussion

Experiment 2 showed reduced demand for alcohol following exposure to natural relative to built stimuli, with a large effect size, supporting Hypothesis 2. In contrast to Experiment 1 (and Hypotheses 1 and 2), the delay discounting and demand for cannabis analyses yielded null results with minimal to no effect sizes, suggesting no effect of visual exposure to natural compared to built environment stimuli.

General Discussion

The two experiments together, including an online sample (Experiment 1), and a university student sample in a laboratory setting (Experiment 2), provide preliminary evidence for the effect of visual exposure to nature to potentially influence decision-making that is related to cannabis and alcohol use among individuals who regularly use these substances, although null effects were also observed. Both delay discounting and demand processes are considered important facets of substance use disorders (Bickel, Johnson, et al., 2014), and both were attenuated by visual exposure to natural versus built stimuli in some cases, for some substances, under certain conditions. Supporting the hypotheses, the overall combined results suggest that visual exposure to nature compared to built-urban environments, may lead to reduced delay discounting (Experiment 1), reduced demand for cannabis (Experiment 1), and reduced demand for alcohol (Experiment 2). In the online MTurk sample (Experiment 1), the within-subject repeated measures results revealed an order effect in which nature’s effect was enhanced depending on the task and whether it was experienced in the first or second session. As the sample sizes and power were reduced when the substance specific groups were analyzed separately, and potential unique effects may exist across substances, more research is needed on each substance with larger samples to tease apart substance-specific nature effects among various populations.

Our findings of reduced delay discounting following visual exposure to natural compared to built environment stimuli (Experiment 1) replicated past research (Berry et al., 2015, 2014; van der Wal et al., 2013), while extending the results to individuals who regularly use substances in a within-subject design. Additionally, the apparent order effect suggested that the effect on delay discounting was enhanced when the natural condition was experienced in the second session. These results align with Stenfors et al.’s study (2019) who found an additional improvement in cognitive performance in the second session, only for those who walked in a natural (compared to urban) environment.

In Experiment 2 the effect on delay discounting was not replicated. The null findings may have resulted from sample size and characteristics, or nature-related factors. In Experiment 2, compared to Experiment 1, the sample was younger (mean age 21.1 and 41.1, respectively). The effect of natural and built environments may differ among different age groups (similar to differences in nature connectedness, Hughes et al., 2019). Different age groups may also experience typical screen time differently and thus may be more or less sensitive to the visual stimuli on the computer screen. The effect of visual exposure to nature on delay discounting may also differ across different levels of typical substance use. The student sample (and especially the alcohol group) reported lower typical substance use than the MTurk sample. The nature effect may be more apparent among heavier substance users who tend to exhibit greater delay discounting in general, or among those with higher baseline levels of discounting independent of substance use.

In Experiment 1, demand for cannabis was lower when the natural condition was experienced first (in contrast to the reductions in discounting when the natural condition was experienced second). Delay discounting and demand represent different constructs and may not necessarily respond uniformly to the same presentation order. Still, it is unclear whether visual exposure to natural environments reduced demand for cannabis/alcohol when it was novel in the first session, and/or attenuated an increase in demand in the second session. It is possible that in the second session participants were more fatigued and bored, and given that boredom might increase sensitivity to rewards (Milyavskaya et al., 2019), the built condition yielded an increase in demand, whereas the visual exposure to natural environment attenuated it. Especially when the sample in Experiment 1 included experienced MTurk Workers who complete many surveys for income purposes. In Experiment 2, conducted in controlled laboratory conditions with university students potentially more naïve to survey completion (as some explicitly shared), viewing natural versus built stimuli reduced alcohol demand regardless of condition order. These data may implicate several potential mechanisms at play including cognitive restoration, reduction of mental fatigue/boredom (Berman et al., 2008, Stenfors et al., 2019), and/or reduction in sensitivity to rewards with viewing natural environments, and these may not be mutually exclusive. Future research should include theory-driven measures (e.g., fatigue and restoration) to further understand potential underlying processes related to the beneficial effects of exposure to nature on substance use related decision making.

Although preliminary evidence with a small sample, in Experiment 2 alcohol demand was reduced when viewing natural relative to built stimuli, with a large effect size. The student sample was younger and reported less frequent alcohol use than the MTurk sample. These differences may suggest nature is beneficial for sustaining lower levels of consumption, however more research is needed to form strong conclusions. On the other hand, cannabis demand did not differ across conditions in Experiment 2 (contrary to Experiment 1, in which cannabis demand was reduced with visual exposure to natural relative to built environments), even though the typical frequency of use was similar across the samples. Although null effect of visual exposure of nature via images in younger cannabis using populations is a possible explanation, the inconsistency in demand for cannabis between Experiments 1 and 2 may have resulted from methodological issues and inconsistent tasks across the experiments.

Assessing demand for cannabis is more challenging than other substances (Aston et al., 2021), and the cannabis purchase task was slightly altered from Experiment 1 to 2, introducing additional challenges. In Experiment 2 the cannabis purchase task asked about hits (rather than grams as in Experiment 1), and although all purchase data were systematic, the results of Experiment 1 were not replicated. Moreover, based on the RMSE values and visual inspection of the demand curves, the cannabis hits task yielded the poorest fit compared to all other tasks. Cannabis units are not standardized like alcoholic drinks or nicotine cigarettes, and the task may be harder to complete with unclear estimations of use, or may cause confusion, as several participants explicitly commented. It is possible that asking the participants about purchasing hits of cannabis (rather than grams), which is not commonly the unit of purchase but the unit of consumption made the estimation more complex. Assessing cannabis consumption in survey studies is unstandardized and challenging in general (Borodovsky et al., 2024), and so is finding the optimal task units in cannabis purchase tasks specifically (Aston & Cassidy, 2019). For example, in an e-cigarette vaping purchase task participants reported difficulty in estimating use (typical use or purchasing in the task) in units of puffs/hits (Cassidy et al., 2017). In the same manner, it is possible that an alcohol purchase task that would ask about purchasing sips of alcohol would yield different results than the typical task that presents a drink as the unit of purchase. These inconsistencies highlight the complexity of separating manipulation effects from the methodological challenges with unstandardized cannabis units.

Finally, these results suggest that delay discounting and cigarette demand among cigarette smokers were less influenced by viewing natural versus built environments. It is possible that visual nature effects interact differently with different substances, on a pharmacological level, but also on a behavioral or social level. For instance, cigarette smoking is prohibited or socially unaccepted in many urban environments (e.g., near buildings), whereas in natural environments smoking is more acceptable and common. Thus, in contrast to the hypothesis, visual exposure to natural scenery might have less of an effect on cigarettes compared to other substances. Second, most of the cigarette group reported smoking a cigarette before or during one or two of the surveys. In a secondary analysis, Almog et al. (2025) compared data from participants who used a substance before or during survey completion to data from participants who did not use any substance in proximity to participation. Results showed that very recent substance use, including smoking cigarettes, was associated with higher demand and craving for substances. Additionally, the authors suggested that proximate substance use might possibly interact with or eliminate experimental manipulations. It is possible that proximate nicotine use, with possible higher state craving among the majority of the cigarette group overshadowed the effect of visual exposure to nature, especially in terms of demand for cigarettes. More research is needed with this population to understand how recent smoking or deprivation affects demand for cigarettes, and how nature versus built environments interact with acute smoking or deprivation.

The present experiments had several limitations. First, there was no pre-test measurement of discounting and demand without any manipulation involved, to serve as baseline control. Although Berry et al. (2014) found no differences between a built environment condition relative to a geometric shapes control condition, we did not assume that the built environment serves as a control condition. The two environments were chosen as two potential real-world environments that individuals might experience in their day-to-day lives. It is possible that participants were negatively affected by the built environment stimuli rather than positively affected by the natural stimuli. However, in contrast to past research that used unpleasant industrial environments for the built condition, in the present study, the built condition included artistic urban images capturing architecture and beautiful urban landscapes, roughly matching the nature images in colors and contours. Nevertheless, an additional measurement to establish baseline discounting and demand levels could have served as a useful control to assist with the interpretation of the results, and especially the observed order effect. Second, the sample sizes of the substance-based groups were relatively small. This became especially apparent when reducing the overall sample into substance specific subgroups based on order of conditions to examine the order effect. A repeated measures effect in the behavioral economic literature is not frequently the focus or often not accounted for per se. Thus, we did not hypothesize an order effect a priori, resulting in somewhat smaller possibly underpowered samples. A secondary analysis study (Stenfors et al. 2019) of the effect of nature on cognitive performance revealed a similar order effect when reviewing data from multiple studies and hundreds of participants. It is possible that substantially larger samples for each group are needed to reveal this effect consistently in all substance-based groups.

Third, the present study had methodological limitations that should be addressed and advanced in future research. For example, it is unclear how the hypothetical purchase task interacts with the manipulation in repeated measures in terms of anchoring effects, washout periods, as well as participant considerations across repeated measures. For example, as one participant commented: “This time I specifically reminded myself that I wasn’t getting fancy unique cocktails and adjusted my answers accordingly”. Beyond being familiar with the task (which could be achieved with a practice session), are people more, or less, thoughtful in subsequent measurements? And how does it affect choice? As another participant commented:

“…the part about how much money you would spend was really interesting to me the second time around. It really made me think critically about how much cannabis I truly would buy, and at what price I would genuinely stop buying. Honestly, I’d be interested in seeing the results of a round three of this part of the study because I think you consider the question differently over time.”

Moreover, our study combined state purchase tasks, repeated measures, remote participation, and a visual manipulation to reduce consumption – a combination of factors that required a combination of traditional (e.g., systematicity of consumption data) and specific data-informed approaches to prepare the data. To determine reasonable differences across repeated measures that may be related to the manipulation, we identified a cutoff (of 8x difference) for what could be considered unreasonable differences in consumption across repeated measures that may be related to inconsistent data quality across sessions. Lastly, we acknowledge common concerns related to remote MTurk samples. Although we used Qualtrics timestamps to gauge engagement with the slideshows and excluded participants who continued the survey before the slideshow’s duration ended, we were unable to ensure all participants watched the slideshows in full or at all. Refining the methodological components will enable better research tools to use in repeated measures studies in general and a better understanding of the nature effect specifically.

Critically, research is needed with larger and more diverse samples (as our samples were mostly white with two gender identities) that expands the understanding of nature’s effect on specific substance-using populations (e.g., based on severity of substance use or type of substance used). Moreover, although our MTurk sample reported daily (or almost daily) substance use we did not assess potential SUD or polysubstance use, warranting more research specifically with clinical populations. Future research should also focus on clinical populations with comorbid depression and/or anxiety and evaluate whether and how individual differences such as preference for environment, area where the individual grew up, or nature-relatedness moderate the effects. Expanding the present study, future research should also evaluate the effect of exposure to real nature in substance using populations, as the effects may be larger with more immersive environments, compared to the controlled short visual manipulation in our study The effect of nature may also differ across repeated exposures, and novelty and/or when it is experienced in certain psychological states (e.g., fatigue). Novelty can be achieved by visiting different natural spaces or by walking (versus sitting) in the natural space that entails changing scenery.

Our findings showed varied effect sizes ranging from medium to large effects in some individuals, which were within-subject, and any reduction in consumption of the most frequently used substance could be meaningful on a public health level. Additionally, the effects emerged after a short and limited visual manipulation on a computer screen. Longitudinal research with repeated exposures to real nature will allow for evaluation of whether the effects on decision-making processes are larger and whether they will translate into real-world substance use outcomes (e.g., reductions in substance consumption). Alternatively, the effect of exposure to nature could be evaluated as an adjunct component in validated cognitive-behavioral and/or pharmacological therapy treatments for substance use disorders. Exposure to nature could enhance other therapies and/or improve program adherence and participation. Exposure to nature for therapeutic purposes is becoming increasingly popular (e.g., Shinrin Yoku “forest bathing” in Japan) and nature prescription is a recognized prescription practice by healthcare providers in most states in the U.S. (i.e., known as “nature prescription”, Nguyen et al., 2023). A more comprehensive understanding with additional scientific evidence could inform the development of enhanced and individualized nature prescription protocols, as nature appears to simultaneously benefit multiple aspects of health and well-being. Future research in this vein could also inform city planning and nature accessibility efforts.

In conclusion, our study provides preliminary evidence that visual exposure to natural versus built environments appears to influence aspects of decision-making processes that are associated with substance use disorders for some individuals in a potentially beneficial way. The present study extended previous research to cautiously suggest that visual exposure to nature appears to reduce delay discounting and may reduce demand for cannabis (Experiment 1) and alcohol (Experiment 2) among individuals who frequently use these substances. However, to form strong conclusions, more research is needed, especially testing the effect of exposure to real nature on substance use outcomes among individuals with SUDs, while untangling methodological challenges. With future extensions to clinical populations, exposure to real nature, and repeated exposures, this line of research could inform nature prescription or educational programs more generally. These lines of inquiry might also inform SUD treatment programs specifically, as exposure to nature may be a beneficial, accessible, low-cost adjunct with potential to interact with multiple elements of substance use disorders (e.g., mental health, delay discounting, substance demand).

Public Significance Statement.

This study suggests that visual exposure to natural versus built environments may affect decision-making processes associated with substance use disorders, specifically reducing delay discounting and behavioral economic demand for cannabis and alcohol. Future research with repeated exposure to real nature holds implications for developing nature prescription protocols as an adjunct, low-cost, and scalable component in substance-use disorder prevention and treatment programs.