Abstract
Consumers decide what to purchase, under conditions of constraint (e.g., commodity price). According to behavioral economic demand, commodity purchase task (CPT) can measure hypothetical decisions about purchases under varied simulated policy conditions (e.g., introduction of new cigarette taxes, happy hour drinking specials). These tasks permit rapid data collection without sacrificing methodological rigor or the validity of conclusions reached. The CPT allows researchers to simulate new policies, to determine their relative risks and benefits, thus offering an opportunity to optimize prior to rollout. Behavioral outcomes related to consumer purchases also make the CPT data readily translatable to policymakers, including constituent health behavior. This article provides a background on CPTs, a review of literature related to policy-aimed CPTs, and a start on best practices for other behavioral scientists interested in applying CPT to inform public policy efforts. It also serves as a primer for policymakers seeking to evaluate usage of this tool.
Tweet
Behavioral economics provides novel insights into consumer decisions. Commodity purchase tasks (CPTs) can efficiently simulate varied policy impacts. CPT data provide insights such as maximum costs or expected expenditures related to policy decisions.
Key Points
Operant behavioral economics examines how organisms consume rewards and work for outcomes within constraints such as increasing costs or availability of substitutes.
The commodity purchase task (CPT) is a simulated purchasing scenario permitting flexibility in the variables of interest and a wide range of economic constraints.
Plotting demand curves quantifies how consumption decreases with increasing constraint.
The CPT is a rapid and safe approach to generating demand curves with human participants.
CPTs have assessed various socially important phenomena, all with direct policy implications, for example, cigarette purchase.
Data obtained provide a cost-efficient means to help policymakers determine the potential for risk and benefit under varied proposed policies to inform policy selection.
Behavioral Economics and Demand
Behavioral economics is broadly considered as an integration of microeconomic and behavioral psychology (Hursh, 1980, 1984). Under the umbrella of behavioral economics are two academic approaches. The first approach to behavioral economics is the field of behavioral finance—loosely speaking, this corresponds to the behavior of economics. That is, behavioral finance is concerned with how people make financial decisions under conditions of restraint, as well as how choice falls prey to cognitive biases (e.g., prospect theory; see Barberis & Thaler, 2003). The second approach to behavioral economics is operant in nature—that is, the study of the economics of behavior. In operant behavioral economics, the analysis centers on how goods (e.g., rewards) are consumed amid economic pressures such as the delay in receiving that good or effort required to obtain it (Hursh, 2014).
Operant behavioral economics has a rich experimental tradition, built upon decades of basic nonhuman research studies. The general thesis of operant behavioral economics—hereafter simply termed “behavioral economics” unless otherwise specified—is that consumption of rewards follows the Law of Demand. That is, organisms work to defend the consumption of a reward amid increases in that reward’s “unit price” (i.e., the amount of expenditure associated with accessing one unit of reward (Hursh et al., 1988) until that price becomes overwhelming and the organism stops consuming the reward.
Years of quantitative modeling refinement have shown that the demand curves generated in behavioral studies follow a predictable pattern in which consumption occurs at some desirable level at low prices until it begins to rapidly decrease toward zero consumption (Hursh & Silberberg, 2008). This level of consumption at a low price is often referred to as demand intensity (i.e., the level of consumption for which the organism works to defend), while the sensitivity to price is captured by other features of the demand curve such as the price where consumption begins to rapidly decrease (Pricemax or Pmax), the maximum level of expenditure (Outputmax or Omax), and the price of zero consumption (breakpoint).
The varying facets of demand derived from behavioral economics provide a comprehensive account of an organism’s motivation to access a reward. It is unsurprising, then, that this behavioral economic technology has provided novel insights into many meaningful behavioral phenomena (Reed et al., 2022). One area in which behavioral economics has proven exceptionally insightful is addiction and potential for misuse of novel drugs and drug formulations (Bickel et al., 2000). Early work in this area involved nonhuman animals responding in laboratory settings to access or self-administer drug rewards. Yet, despite the apparent benefits of behavioral economics to inform policy, its potential use for human behavior was limited by feasibility and ethical concerns. First, drug administration studies with humans were resource intensive and ripe with procedural concerns for participant safety. Second, drug reinforcers for laboratory use were pricey and also resource-intensive. Third, reaching stable responding in operant procedures took substantial time, which also meant the use of incentives to pay participants was high. Finally, obtaining a full demand curve within a human participant meant repeated exposure to drug consumption across many price manipulations, presenting significant ethical concerns.
However, innovative translations of laboratory procedures led to simulated purchasing tasks (Aston & Cassidy, 2019; e.g., Jacobs & Bickel, 1999), which leverage humans’ verbal capacities to recount their own behavior and project how they would consume under hypothetical pricing scenarios. The advent of these commodity purchase tasks (CPTs; Roma et al., 2016) has proven to be a watershed moment in use of behavioral economics in policy development and refinement (Hursh & Roma, 2013; Roma et al., 2017).
Prototypical Commodity Purchase Tasks
Decades of research refining the CPT have resulted in a general prototype (Roma et al., 2016). Participants first read carefully worded instructions designed to help control important features underlying consumption decisions such as the openness of economy (i.e., whether there is access to the reward outside the simulated purchasing scenario), budget/supply limits (i.e., how much money is available to use in purchasing, as well as the duration of the consummatory period), and availability of substitutable goods (i.e., whether other similarly functioning drugs are available). A vignette may accompany these instructions to help provide a purchasing scenario that is relevant to the question at hand (e.g., an alcohol purchase task might include instructions such as: “imagine you are at a bar with your friends from 9pm to 2am to see a band…”). Finally, participants respond to a series of prices—typically ascending—associated with the target commodity and are asked to indicate how much of the commodity they would purchase at each of those prices.
The prototypical CPT shares many commonalities with other survey-based methods in economics—notably, stated-preference, discrete-choice surveys (e.g., Carson, 2000). Essentially, these are self-reported choices between two alternatives. Stated preference surveys are commonly used to simulate policies by presenting respondents with numerous discrete choice scenarios. For example, “would you prefer: 1) Policy A that decreases a health risk by 35% through a 25% increase in excise taxation, or 2) Policy B that decreases a health risk by 45% through a 50% increase in excise taxation?”
Another variation of the stated preference survey uses contingent valuation or “willingness to pay” methods, such as asking: “Are you willing to pay an extra $55 in taxes throughout the year to improve children’s access to nutritious food in school cafeterias?” These methods typically entail an informative scenario to fully spell out the parameters of the simulated choice event to help control for factors that might influence responding (e.g., time of the year or access to other goods). Respondents are randomly assigned to different choice conditions—often assessing three or more variables (e.g., different policy ideas), with each variable featuring two or more value levels (e.g., health impacts, costs). However, any single respondent will only see a handful of these combinations—not all variables, and not all value levels of those variables—such that there is no complete demand profile within any single respondent. Put another way, some respondents respond to some choice options to determine the impact of policies for all respondents. This increases response efficiency, minimizes fatigue, and makes it less likely that the respondent would provide responses anchored to or biased by their other responses to similarly worded scenarios. However, this approach also requires a large number of respondents to generate meaningful data for each scenario variation and can only provide estimates at the group level.
Unlike stated preference surveys, all CPT respondents experience all levels of the relevant variable values (i.e., respond to the entire range of unit prices), permitting a demand model to be fit to each individual respondent’s dataset (see discussion by Kaplan, Franck, et al., 2021). In this approach, researchers obtain demand variables for every respondent. Researchers may then compare averaged demand variables (across respondents) between specific groups of people in the sample (e.g., respondents of different demographics) and/or how respondents’ demand variables vary between different simulated policies. This method thus informs how a potential policy may impact all respondents across all possible policy conditions under study. CPTs offer a novel lens through which we can better understand individuals’ reward valuation, beyond that achieved in stated preference surveys due to the comprehensive, but still rapid, nature of these tasks.
Advantages for Policy Insights
A common criticism of the CPT is the use of the hypothetical report. Nonetheless, decades of work indicate that these tasks are valid and reliable, corresponding well with actual purchasing and consumption behavior. Simulated procedures—like the CPT—are of particular importance in instances when it is infeasible or potentially unethical to expose participants to actual consummatory events. In the special case of policy development, researchers often lack the luxury or resources to fully test policy ideas in the field prior to community-wide deployment. Moreover, rolling out certain experimental policies without any advanced data on efficacy (e.g., happy hour drinking laws, drug regulation) could be dangerous to the constituencies affected. The CPT provides a safe yet rigorous approach to simulating such policies to circumvent feasibility or ethical concerns.
In a recent paper, we proposed that behavioral economic approaches to simulating policy fit well into a broader framework of policy development using behavioral science (Reed et al., 2022). First, there are foundational research studies using human, nonhuman, and/or clinical laboratory experiments to flesh out behavioral mechanisms that may serve as means to influence behavior via regulations. Second, post hoc or archival analyses of real-world—that is, undisturbed by scientific experimentation—econometric data may help support the conceptual grounding and external validity of the experimental research, with respect to its potential use to inform policy. Such data serve as proof-of-concept support warranting further nuanced experimentation, using simulated tasks, such as CPT.
Unique Insights from the CPT
One-Time or Low-Rate Decisions
The gold standard of behavioral study of socially important behavior is direct observation of a response under differing experimental conditions to the point of stable behavior (Sidman, 1960). This standard is certainly ideal and should be used when possible. But not all socially important behaviors are repeatable for an organism. For example, receiving an immunization shot is often a discrete behavior that occurs in low rates across an organism’s lifetime (e.g., once or twice, annually). Analyzing this behavior is not amenable to the procedures typically used in experimental behavioral science (e.g., pressing a lever for a set number of times to earn access to an immunization). In these cases, policymakers must identify ways to evoke a single response (e.g., receiving a vaccination) to achieve a public health need. Analyzing archival data with statistical models may identify potential variables that can optimize efficacy—but this is limited to those variables with an existing market history (i.e., novel ideas may have no relevant dataset to permit archival analysis). Field studies could attempt to roll out different marketing campaigns across different public health departments and use group designs to test hypotheses—but this approach is expensive, time-consuming, and potentially dangerous if quick answers are needed in a public health crisis such as a virus outbreak.
The CPT approach, on the other hand, can permit wide-scale data collection in a short time to test a range of potential policies in simulated policy vignettes that thereby pose a limited risk. Examples of using CPTs to experimentally simulate policy approaches for low-rate decisions include SARS-CoV-2 vaccination and precautionary decisions (Strickland et al., 2022), underscoring the rapid nature of data collection that can address immediate health needs. A key example of this used scenarios reflecting real-world efforts to develop vaccination (i.e., operation “warp speed”) to evaluate how these various conditions impacted vaccine uptake (Hursh et al., 2020; Strickland et al., 2022). Approaches that framed vaccine safety (i.e., positive frames) greatly increased vaccine uptake as compared to those that emphasized possible side effects (i.e., negative frames). Such insights could have a long-standing benefit for mobilizing the masses during periods of heightened viral transmission and how public health campaigns should message education about novel preventive health measures.
Illicit Drug Use
Following in line with historic precedence, much success has been had with CPTs in pharmacological evaluation, particularly in keeping pace with the shifting national and global marketplace. The development of legal cannabis (marijuana) markets means policymakers must consider shifting consumers away from illegally obtained cannabis markets. In a finding with direct implication for this ongoing deliberation, the CPT identified prices at which a licit market was able to outcompete an illicit market product (Amlung et al. 2019)—recent market research found that the simulated market data accurately predicted purchasing effects in Canada’s transition to legalized cannabis (Wadsworth et al., 2022). Other research has shown how formulation changes can alter the abuse potential of novel drug formulations (Schwartz et al. 2019). These studies have shown that respondents with no history of pill tampering reported relatively comparable consumption of both standard and abuse-deterrent opioid pill formulations, while those respondents with a self-reported history of modified pill administration (e.g., crushing for insufflation) reported lower consumption of abuse-resistant formulations as compared to the standard formulation. Together, these studies offer novel insights into potential policy design with plausible likelihood to curb illegal and/or excessive drug consumption.
Non-Illicit Drug Use
The CPT application has also successfully modeled choice and captured policy-relevant outcomes for commodities in the licit drug market (e.g., alcohol and tobacco). For instance, intention to purchase and consume alcoholic beverages varies across commonly used happy hour frames (Kaplan & Reed, 2018). In theory, deals claiming a “buy one get one” (BOGO) purchasing opportunity and those pushing “half price” beverages are of comparable economic value and should yield similar rates of consumption. A large crowdsourced sample viewed a well-vetted bar purchasing vignette (see Kaplan et al., 2018) and reported in a between-group comparison the number of drinks they would purchase and consume if presented with one of two happy hour frames. Participants reported relatively greater intended alcohol consumption in the BOGO condition, particularly at lower per drink prices, a finding with potential to create inroads toward curbing excessive alcohol consumption during peak times and inform policy limits on potentially dangerous marketing strategies.
A growing list of research efforts has captured similar modulating variables in an ongoing venture to limit cigarette abuse liability, all of which have made use of the CPT in some fashion. The concurrent availability of alternative sources of nicotine (e.g., nicotine gum) has reduced consumption of the combustible cigarette (e.g., Johnson et al., 2004). Examination of cigarette purchasing in high-resolution (i.e., relatively greater price density) reveals the impact of pricing frames where requiring prices to be at a whole- and half-dollar prices makes consumption more elastic (i.e., flexible and sensitive to price increases; e.g., MacKillop et al., 2012); an effect also observed with alcohol purchasing (see Salzer et al., 2019). Still, other research has drawn into question enacted policy efforts, as in the effect of current graphic warning labels to curb cigarette purchases (e.g., Pacek et al., 2019). Capping nicotine content is another prospective policy effort to influence cigarette consumption that has been informed by ongoing attention via CPT administration (e.g., Kaplan, Koffarnus, et al., 2021). A defining feature of the CPT, then, is an applicability sufficiently dynamic to examine multiple facets of purchasing within a single drug commodity.
Other Commodities of Potential Public Health Harm
Perhaps furthest from the traditional installments of behavioral economics, CPT application has been used to explore a variety of substances and commodities of atypical yet health-relevant administration. To date, these studies spread a wide gamut and offer scalable, policy-relevant data poised to inform environmental design. One study evaluated the use of the CPT for ultraviolet indoor showing elevated demand for a month of unlimited bed use by recent ultraviolet indoor tanning users as compared to nonusers and informing policy decisions surrounding these subscription models (Reed et al., 2016). Further, a CPT evaluating gambling behavior could distinguish between those with and without a history of disordered gambling behavior, suggesting the diagnostic utility of these procedures (Weinstock et al., 2016). Such modeling of intention to purchase a commodity among experienced users has potential to inform pricing mandates, such as fixing costs at or above a value most likely to deter nonrecent users from reengaging in a socially undesirable behavior.
In the domain of risky sexual decision-making, CPT administrations have garnered awareness for varied policy-relevant change agents. For example, CPT research on pornography access has guided parallels to other addictive behaviors and how policies used to promote drug abstinence may be similarly applied to reduce undesirable pornography use (Mulhauser et al., 2018). Other research has shown how cost can impact decisions to engage in protected sexual activities (e.g., use of condoms in casual sexual encounters). A synthesis of findings (see Dolan et al., 2020; Harsin et al., 2021; Strickland et al., 2020) underscores the importance of low-cost, easily accessed condoms to maximize the likelihood of timely and appropriate use. These findings show how ongoing CPT applications can show how substantive public health benefits can be reached by a keener understanding of seemingly small environmental changes (e.g., reducing the burden on obtaining condoms in commercial spaces).
Best Practice for CPT Development
Several best practice approaches exist for CPT development. Researchers have explicitly proposed many of these components (e.g., Roma et al., 2016), whereas others are a product of unspoken consensus. This section lays out what we believe to be starting points for best practice in the construction and analysis of a CPT and subsequent data.
Assumptions
In considering the simulated nature of CPT administration, it is critical to properly flesh out a vignette through which participants envision their decision making. A key aspect of this is the series of assumptions within which respondents make their choice. Some of these are straightforward. For instance, respondents should respond as if they themselves would be making said purchases. Budgets should be imagined as the respondent’s ongoing fiscal status. Purchased commodities should be for engagement within the outlined time period and should not be viable to stockpile for future, more gradual, or shared engagement. Other assumptions might be unique to the commodity itself and require more consideration of real-world constraints. Ingesting greater volumes of a drug with a carryover effect (e.g., alcohol-induced hangover) could have implications for next-day responsibilities, and so indicating whether those responsibilities should be considered could indirectly impact reported purchasing (see Berman & Martinetti, 2017; Skidmore & Murphy, 2011). Arriving at the most appropriate set of assumptions may require testing within the specific condition evaluated but is essential for keeping responses consistent and focused.
Pricing
Price sequencing involves two primary considerations: (1) the relative density of prices across the full range of values assessed, and (2) the order in which said prices are presented. As it comes to the first of these considerations, the guiding notion is that a relatively higher density of pricing around the suspected value in which consumption shifts from high and stable to rapidly decreasing. One can readily assume that, at relatively low prices, demand will remain unchanging, whereas at very high prices, demand should be at near-zero rates. It is this middle range, then, particularly around market price, that should be assessed at greatest density. Roma et al. (2016) provide a general set of guidelines for developing a novel price sequence (see also Reed et al., 2014). With respect to consideration over price sequencing, replicated research indicates little difference in quality or reliability of responding when prices are presented in ascending, descending, or randomized sequences (e.g., Salzer et al., 2021). Although the field would surely benefit from more research to arrive at a definite recommendation, current thinking lends to employing a price sequence that most logically fits with the commodity and framing in question.
Response Types
Administration of CPTs has evolved from paper-and-pencil to more technologically savvy approaches, but in any format, particular response topography should draw some consideration. Across the literature, CPT completions have taken form of free-response text boxes, visual analog scales, and binary responses (e.g., would/would not purchase). All have produced prototypical demand data that adequately model choice. Yet there are some nuances to consider. Free response formats permit respondents to indicate a raw quantity or probability value for each corresponding price, often sacrificing boundaries or entry constraints (i.e., respondents are typically free to enter any value, even the seemingly nonsensical) for wide applicability. Responses logged on a visual analog scale are subject to more overt constraints (i.e., no value outside that of the sliding scale can be logged) but are less applicable for “quantity purchased” style tasks (as opposed to tasks wherein probability of purchase is the primary dependent variable). Dichotomous responding offers the least variability for participant responding but is more likely to force orderly responding and may thusly be well-suited for novel or atypical task administration.
Analyses
Data produced via responding on the CPT should be first analyzed for systematicity—that is, whether the data conform to expected patterns (e.g., decreases in consumption as price increases, no sudden increases in consumption following previous decreases, and no spontaneous increases in consumption at prices higher than those that previously returning zero/suppressed consumption; see Stein et al., 2015). Once data are readied for analysis, they are typically fit using a nonlinear model; demand metrics of interest can then be derived and assessed. The literature offers various models that may be better suited pending the characteristics of the unique data set (see Koffarnus et al., 2022). The complexities of model selection are beyond the scope of this paper. In the stead of a lengthier discussion, we note that data analysis may be approached via a number of freely available analytic resources. For example, Gilroy et al. (2018) offer a web-based platform to partially automate model fitting. Similarly, Kaplan et al. (2019) curate an open-source statistical package to facilitate operant demand analysis. Finally, more recent advances in analysis propose mixed-effects modeling that retains the theoretical model of demand while simultaneously permitting inclusion of nonsystematic data and analysis of fixed and random effects (see Kaplan, Franck, et al., 2021). We encourage authors to explore these options and their respective authors for further guidance.
Conclusion
Behavioral economics emerged as a conceptual framework for studying consumer decisions and bridging the disciplines of finance, economics, and psychology. Early behavioral economic work on nonhuman drug administration showcased the potential robustness of this approach to quantifying behavior and understanding the effects of experimental contexts on willingness to emit responses for particular consequences within the operant demand framework. The relevance of these operant demand findings for understanding addiction and other substance use became readily apparent, paving the way for translational approaches to studying demand in simulated contexts. The CPT specifically permits a safe, reliable and valid, conceptually systematic, and rigorous approach to understanding the consumption of goods across differing simulated contexts such as varying public policies under consideration.
The unique data produced by the CPT are readily translatable to policymakers, given the focus on economic concepts and their impacts on human behavior. Thus, policy-relevant work has historically dominated the CPT literature, providing novel behavioral economic insights for topics including but not limited to: (a) excise taxation impacts on consumer spending (e.g., Epstein et al., 2010; MacKillop et al., 2012; Reed et al., 2016), (b) drug dosing and packaging (e.g., Kaplan, Koffarnus, et al., 2021; Pacek et al., 2019; Schwartz et al., 2019), (c) vaccine framing and costs (e.g., Hursh et al., 2020; Strickland et al., 2022), and (d) drug consumption regulations and legalization (e.g., Amlung et al., 2019; Kaplan & Reed, 2018). In each case, behavioral economists can rapidly obtain such insights using simulated methods, providing safe and efficient means of addressing emerging public health crises and concerns.
Despite these advances, important work remains. The next step is more widespread adoption of the CPT in policy spaces to inform a research-to-policy feedback loop (Reed et al. 2022). This incorporation of behavioral economics and CPT data stands to directly benefit policymaking given that data obtained via the CPT provide a cost-efficient means to determine the potential for risk and benefit under varied proposed policies and ultimately inform policy selection grounded in empirical data rather than strictly theoretical grounds. We hope that the starting points provided in this paper help others continue to translate behavioral science into policy and fill these research gaps.
Footnotes
Acknowledgments
The authors acknowledge the incredible support and insights on behavioral economic implications for policy provided to us by Steven Hursh, Peter Roma, David Jarmolowicz, Thomas Critchfield, William Stoops, Brent Kaplan, and Matthew Johnson. This manuscript is dedicated to the memory of our colleague, David Jarmolowicz.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the University of Kansas, National Institute on Drug Abuse (grant number KU GRF 2133083, R03 DA054098, T32 DA007209).
