Extra-Experimental Scarcity Impacts Hypothetical Operant Demand: A Natural SARS-CoV-2 Experiment

Abstract

Behavioral economic demand models quantify the extent to which an organism defends its consumption of a commodity. Commodity purchase tasks permit humans a quick yet psychometrically sound approach to assessing commodity demand for various retail products. Operant behavioral economic literature suggests economy type (open vs closed) can significantly alter demand, yet this effect is largely undocumented in the commodity purchase task literature. In this study, we leveraged the market pressures for retail goods (hand lotion and sanitizer; paper towels and toilet paper; soda and water) resulting from SARS-CoV-2 into a natural experiment comparing within-subject demand across two time-points during the pandemic using a crowdsourced approach. Results suggest that hypothetical commodity purchase tasks are sensitive to extra-experimental market pressures (e.g., scarcity due to the closing of economies), adding additional confidence to the self-report nature of purchase task responding and providing further construct validity to these approaches.

1. Introduction

Operant demand analyses quantify the extent to which an organism defends consumption of a constrained reinforcer, where constraint most often comes in the form of unit price manipulations (Hursh, 1984). The primary dependent measure in demand analyses is the rate of change in elasticity (i.e., the change in consumption with each change in unit price; the α parameter in demand models; Gilroy et al., 2020). The α parameter is sensitive to price and other contextual variables, such as the availability of reinforcer substitutes or complements and open or closed economies (Kearns, 2019).

The extant literature on contextual variables modulating the α parameter in demand models is relegated mainly to nonhuman operant experiments. In human demand studies, however, the contemporary approach to data collection is through commodity purchase tasks that include hypothetical purchasing scenarios (Roma et al., 2016). While these purchase tasks may be convenient conduits to translating basic concepts from the operant lab to everyday human issues (Roma et al., 2017), our understanding of contextual influences are limited to only a handful of experimental manipulations, and mainly focused on addictive commodities (e.g., alcohol, cigarettes; Acuff et al., 2020). Regardless of commodity type, we are aware of no research that explicitly evaluates whether extra-experimental changes in economy type (i.e., open/closed) affect hypothetical purchasing on these tasks. Such economy types can be either open (i.e., the target commodity in the study is available outside the target response contingencies) or closed (i.e., the target commodity in the study is only available through response contingencies). Research targeting the influence of extra-experimental influences on simulated responses would further ground these hypothetical tasks in the concepts detailed in basic operant laboratory studies.

The SARS-CoV-2 pandemic presented a natural experiment on contextual impacts on demand because retail goods worldwide became scarce and effectively shifted relatively open economies towards the closed end of the spectrum for commodities such as toilet paper and hand sanitizer (Kirk and Rifkin, 2020). The current study capitalized on the extra-experimental scarcity of products by examining within-subject changes in hypothetical demand for relevant commodities at two time points over the pandemic: March 2020 and March 2021. We hypothesized that the purchase tasks would show construct validity by capturing relatively inelastic demand during the beginning of the pandemic, in which there were surges in consumer demand for relevant commodities, compared to demand one year later in which marketplace scarcity was less of a consumer issue.

2. Methods

2.1. Participants

In March 2020, we recruited an initial sample of 227 participants on Amazon Mechanical Turk to complete our Human Intelligence Task (HIT) survey for $1.00 compensation—only Workers in the United States with a HIT approval rate greater than 95% for at least 100 approved HITS could access the survey link. A total of 73 participants were initially excluded based on failing a qualitative item test or displaying a non-US Internet Protocol address. The 154 participants with complete datasets in March 2020 (Time 1 [T1]) were invited to repeat the study in March 2021 (Time 2 [T2]). The final sample comprised 75 participants who completed all methods across T1 and T2, was 52% female (45.3% male; 2.7% other), 84% White (5.3% Asian, 4.0% Hispanic, 2.7% African American, 2.7% Other, 1.3% Native American) and lived across 27 different states. Most participants (58.7%) reported household incomes above $30,000USD.

2.2. Study Tasks

All purchase tasks were designed per best practices guidelines as suggested by Kaplan et al. (2018), Reed et al,. (2020) and Roma et al. (2016). We identified pairs of commodities that we anticipated as related. In each case, the first commodity of the following pairs was hypothesized to be insensitive to extra-experimental availability due to COVID-19 consumer impacts (i.e., a “control”), while the second was anecdotally scarce in retail outlets—relative to the former—during the pandemic: 1) 12 oz pump bottles of hand lotion and hand sanitizer, 2) rolls of paper towels and toilet paper, and 3) 20 oz bottles of soda and water. Participants completed each commodity purchase task in isolation.

The commodity purchase task contained the following instruction: “Imagine yourself today. Imagine you are out of [commodity] and have an opportunity to purchase [commodity] from one store close to your home. No other stores in your region have [commodity] in stock. The following questions ask how many [commodity units] you would buy at various prices right now, considering today’s market for and availability of [commodity].” A generic image of the commodity appeared below the instructional text, followed by the following bulleted assumptions: a) “Your preferred [commodity] brand is the only [commodity] available.” b) “You have NO ACCESS to [commodity] other than the [commodity unit] available at these prices.” c) Any [commodity unit] you buy today is for your own household use.” d) “You have the same income and savings as you do right now.” Participants indicated the number of commodity units they would purchase at each of the following prices: $0.50, $1.00, $1.50, $2.00, $3.00, $4.00, $5.00, $6.00, $8.00, $10.00, $15.00, $20.00, $30.00, $50.00, $75.00, $100.00.

We administered the study using the Qualtrics XM survey platform. Participants first read an information statement indicating KU Lawrence IRB approval (study #00141983). Upon agreeing to continue, participants completed the commodity purchase tasks. We randomized the sequence of commodity pairs across participants and randomized the order of the commodity within those pairs.

2.3. Data Analysis

We used nonlinear mixed-effects models (Kaplan et al., 2021) to examine measures of intensity (Q₀) and change in elasticity (α) for each of the commodity pairs (lotion, hand sanitizer; paper towels, toilet paper; soda, water) across the two time-points (T1, T2). Nonlinear models, in general, typically require reasonable starting values to solve for the unknown coefficients in the model fitting process. A three-step approach was used to identify these starting values (Rzeszutek et al. 2022). We first fit a fixed-effects only regression model to the pooled data leaving Q₀, α, and k as parameters to be solved. We then used these model estimates as starting values in a fixed-effects only generalized nonlinear model. Here we specified a commodity by time interaction on both Q₀ and α and specified a global, shared k (to be solved). These model estimates served as highly informed starting values in the nonlinear mixed-effects model step. Due to difficulties with estimating a reasonable k value in two mixed-effects models, we used the k value estimate from the generalized nonlinear model as a constant in this step. In this mixed-effects step, we specified: 1) fixed effects of Q₀ and α with a commodity by time interaction; 2) random effects of Q₀ and α associated with the individual; and 3) random effects of Q₀ and α associated with the commodity by time interaction. Random effects were specified with independent covariance matrices at the individual and condition levels, and the mixed-effects models were solved via full maximum likelihood estimation. Fixed effect estimates with associated 95% confidence intervals are reported below.

3. Results

3.1. Lotion and Hand Sanitizer

Results of the mixed effects model for lotion and hand sanitizer revealed a significant time by commodity interaction associated with Q₀ (F[1,4718] = 11.59, p = .0007). Estimated lotion Q₀ at T1 was 9.78 [7.72, 12.4] and increased to 12.5 [9.56, 16.4] at T2, whereas sanitizer Q₀ was similar at T1 17.1 [13.3, 21.9] and T2 17.4 [13.0, 23.2].

The model also indicated a significant time by commodity interaction associated with α (F[1,4718] = 14.2152, p = .0002). Estimated lotion α at T1 was 0.0137 [0.0104, 0.0180] and showed little change at T2: 0.0151 [0.0106, 0.0215] at T2. Sanitizer α was estimated at T1 0.00776 [0.00589, 0.0102] and higher at T2: 0.0136 [0.00915, 0.0201].

3.2. Paper Towels and Toilet Paper

Results of the mixed effects model for paper towels and toilet paper did not reveal a significant time by commodity effect for either Q₀ (F[1,4718] = 0.1713, p = .6790) nor α (F[1,4718] = 0.3738, p = .5410). We subsequently fit a simpler model with only main effects and no interaction; however, a likelihood ratio test suggested the model with the interaction term provided a better fit to the data (χΔ²(4) = 692.23, p < 0.0001). Because the interpretation of the results did not fundamentally change, we report the estimated marginal means of the interaction model here .We observed a significant main effect of commodity (Q₀: F[1,4718] = 15.5527, p = 0.0001; α: F[1,4718] = 33.6095, p < 0.0001), but not time (Q₀: F[1,4718] = 0.9605, p = 0.3271; α: F[1,4718] = 1.4316, p < 0.2316) for both Q₀ and α.

Estimated marginal means of the full interaction model indicated paper towel Q₀s at 26 [21.3, 31.7] and 28.7 [22.4, 36.8] at T1 and T2, respectively, whereas toilet paper Q₀s tended to be higher at 35.8 [30.0, 42.7] and 37.8 [29.1, 49.0] at T1 and T2, respectively. For values of α, the model estimated paper towels at 0.00881 [0.00689, 0.0113] and 0.0101 [0.00731, 0.0140] at T1 and T2, respectively. Likewise, the model estimated toilet paper as relatively more inelastic at 0.00574 [0.00456, 0.00723] and 0.00696 [0.00492, 0.00984].

3.3. Soda and Water

We were able to model k as a global shared value successfully, and a likelihood ratio test indicated this model was preferred over a model with k as a constant (χΔ²(1) = 1011.353, p < 0.0001), This model revealed a significant commodity by time interaction associated with α (F[1,4717] = 3.9445, p = 0.0471), but not Q₀ (F[1,4717] = 0.4207, p = 0.5166). Soda α increased from 0.00691 [0.00515, 0.00927] at T1 to 0.00888 [0.00626, 0.0126] at T2. Water α showed a relatively greater change increase from 0.00307 [0.00232, 0.00406] at T1 to 0.00496 [0.00336, 0.00730] at T2. As can also be seen, water α, regardless of time, was lower than soda, indicating a relatively greater valuation. For Q₀, soda was estimated at 55.0 [40.7, 74.5] at T1 and 29.4 [19.9, 43.4] at T2, whereas water was estimated at 78.8 [60.2, 103.0] at T1 and 46.1 [30.4, 69.8] at T2.

Discussion

The current study leveraged changes in marketplace scarcity over one year of the SARS-CoV-2 pandemic to evaluate the sensitivity of the hypothetical purchase task to variations in economic conditions expected to impact commodity valuation. We found that commodities like hand sanitizer relevant to the SARS-CoV-2 pandemic and thus scarce during its early months showed expected changes in price sensitivity over the one year examined. Specifically, consumption of these commodities was less price sensitive early in the pandemic aligning with greater valuation under conditions of marketplace scarcity. These findings support the construct validity of the purchase task procedure by demonstrating that hypothetical purchasing was sensitive to and aligned with extra-experimental economic conditions.

These findings are consistent with prior work using simulated vignettes to evaluate the sensitivity of the purchase task procedure to hypothetical changes in scarcity. For example, the presentation of vignettes in which participants are asked to consider a future negative context (e.g., loss of job and income; significant hurricane damage) results in increased demand intensity for fast food as well as increased preference for immediate rewards in hypothetical delay discounting procedures (Mellis et al., 2018; Snider et al., 2020; Sze et al., 2017; but see Stein et for failed replication of the demand effect). These findings align with the idea that resource scarcity and environmental instability promote a tendency to focus on meeting immediate needs at the expense of long-term goals and do so using these simulated procedures. A significant limitation, however, is that we did have permission to collect information on employment, COVID-19 infection, or vaccination status for this study; thus, we missed an opportunity to better quantify the relation between negative contexts and increased demand.

Methodologically, the consistency between the effects observed in these hypothetical arrangements and the changes in demand observed in this study support the notion that such tasks may act as an appropriate substitute to study complex—if not impossible—to manipulate contexts to inform policy forecasting (see Reed, Gelino, & Strickland, 2022; Reed, Strickland, et al., 2022). As in the COVID-19 pandemic, some behavioral repertoires or contexts may be of interest for behavioral scientists to study but impossible to directly observe due to ethical or practical reasons (e.g., experimental exposure to poverty or household loss is not ethical). Using simulated procedures provides a safe and valid approach to model potential public health events and the impact of introduced policies that may mitigate or exacerbate harms (Hursh, 1991; Reed, Strickland, et al., 2022).

We a priori selected commodities we expected to be sensitive to or not sensitive to market disruptions in the SARS-CoV-2 pandemic (e.g., hand sanitizer as sensitive and hand lotion as not sensitive). While our findings were generally consistent with these predictions, in the case of soda and bottled water, both commodities showed relevant changes under the differing conditions of scarcity. The primary aim of this project was also to conduct a proof-of-concept within-subject evaluation of hypothetical demand under extra-experimental periods of scarcity. Future research on the consumer demand of these commodities’ price elasticities should seek to evaluate the reasons why such commodities may be differentially impacted by scarcity—such questions were outside the scope of our analysis. Finally, an additional consideration is using only two time-points and the lack of pre- SARS-CoV-2 data. Ideally, our findings would have demonstrated a baseline, disruption, and return to baseline. Nonetheless, the correspondence between our results and prior research using purely hypothetical arrangements supports the broader demonstration of construct validity for these procedures.

The current study contributes to a broader literature developing and validating the hypothetical purchase task for public policy research. Our findings show how simulated procedures may accurately correspond with extra-experimental variations in product scarcity, replicating prior work using simulated marketplace disruptions. These findings collectively support the continued use of the commodity purchase task, specifically, and behavioral economics, broadly, as a tool for evaluating potential public policy decisions.

Fig.1.

Demand for Commodity Pairs Amidst the Start of the SARS-CoV-2 Pandemic (March, 2020; T1; solid lines) and 1 Year Later (March, 2021; T2; dashed lines).

Highlights.

• Reinforcer scarcity modulates operant demand

• The SARS-CoV-2 pandemic presented a natural experiment on scarcity

• Crowdsourced data shows effects of real-world scarcity on hypothetical demand