Presentation Matters: Effects of Cigarette Purchase Task Design on Systematic Data and Purchasing Behavior

Abstract

Aim:

Hypothetical purchase tasks are effective tools for evaluating participants’ demand for substances. The present study evaluated the effect of task presentation on producing unsystematic data and purchasing behavior in a sample of individuals who smoke cigarettes.

Methods:

Participants (n=365) were recruited from Amazon Mechanical Turk and randomly assigned to complete two of three hypothetical purchase task presentations: List (prices on one page in an increasing order), Ascending (one price per page in an increasing order), or Random (one price per page in a random order). We evaluated outcomes using a mixed model regression with a random effect for participants.

Results:

We observed a significant effect of task presentation on passing the criterion assessing consistency in effects of contiguous prices (i.e., Bounce; (X²(2) = 13.31, p=0.001). A significant effect of task presentation on Trend or Reversals from Zero was not observed. For purchasing behavior, we observed a significant effect of task presentation on R² (X²(2) = 17.89, p<0.001), BP₁ (X²(2) = 13.64, p=0.001), ln(α) (X²(2) = 332.94, p<0.001) and ln(O_max) (X²(2) = 20.26, p<0.001); we did not observe a significant effect of task presentation on ln(Q₀) or ln(P_max).

Conclusions:

We recommend against using the Random hypothetical purchase task presentation to avoid unsystematic data. While the List and Ascending presentations do not differ across unsystematic criteria or purchasing behavior, the List presentation may be preferred due to participant experience.

Introduction:

In behavioral economics, a demand curve characterizes the change of consumption of a commodity as a function of price paid or work performed. This analytical approach quantifies individual differences in terms of commodity valuation and has been widely applied in addiction research to evaluate factors influencing consumption and the effect of interventions (Acuff et al., 2020).

Hypothetical purchase tasks (HPTs) are one of the behavioral assessments developed to construct the demand curve (Jacobs & Bickel, 1999). HPTs have the advantage of studying a wide range of prices without constraining time or resources. Additionally, these tasks can be self-administered and easily modified to investigate different commodities. Importantly, drug demand measured by HPTs is reliable and significantly correlated with real purchases and consumption (González-Roz et al., 2019; Murphy et al., 2009; Schwartz et al., 2021; Wilson et al., 2016; Zvorsky et al., 2019). As such, HPTs represent an effective, flexible, and practical approach to examining demand in addiction research.

The implementation of HPTs typically involves asking participants to purchase a drug across a range of monetary prices, assuming that the drugs need to be consumed within a predefined period (e.g., 24 hours), cannot be saved or given away, and are unavailable from any other source (Stein et al., 2015). Purchase data are then fitted by a mathematical equation (e.g., the exponentiated model proposed by Koffarnus et al., 2015) to derive the demand measures or purchase indices, which include intensity (Q₀; level of consumption without cost or other constraints), elasticity (α; sensitivity of consumption to price increases), price maximum (P_max; price at which an individual transition from elastic to inelastic), and output maximum (O_max; maximum response output/expenditure at P_max). Measures of observed data are also evaluated, including breakpoint (BP₁; highest price at which an individual persists to purchase).

Interpretation of the results could be limited when unsystematic responses to HPTs are present. To reduce research bias, Stein et al. (2015) developed three criteria to identify potentially unsystematic responses, assuming a global, price-dependent reduction in consumption and consistency in purchasing across prices could be observed in the demand for any commodity. Specifically, a demand curve may be deemed unsystematic if: (1) the reduction in consumption from the first to the last price imposed is minimal (Trend), (2) the increase in consumption with increment in price is frequent (Bounce), and/or (3) a resurgence of consumption at a given price after consumption falls to zero at lower prices is observed (Reversals from Zero).

The price sequence in HPTs is a potential source of unsystematic responses. Ascending order has been previously recommended (Kaplan et al., 2018). This methodological approach is supported by evidence showing that while differences in alcohol purchase indices between an ascending order and randomized sequence are small, unsystematic data is higher in the randomized sequence compared to an ascending order (Salzer et al., 2021), however, in this study only alcohol purchases were investigated. Another previous investigation that measured demand for cigarette puffs by consumption under different response requirements showed that progressive ratio schedules (i.e., ascending response requirements) and fixed ratio random sequences produced similar demand indices (Giordano et al., 2001). Similarly, a study in rats did not find demand elasticity differences for the mu-opioid receptor agonist remifentanil between an ascending and a mixed order of fixed ratio schedules (Maguire et al., 2020). Thus, further investigation of the influence of price presentations on the purchase indices is needed to examine the robustness and the generalizability of the previous findings.

The present study aimed to test the effect of HPT’s price presentation on rates of unsystematic data and purchasing behavior. Additionally, we investigated whether presenting prices on single or separate pages would change unsystematic data and purchase indices. Daily cigarette smokers completed hypothetical cigarette purchase tasks that differed in presentation: List (increasing prices listed on one page), Ascending (one price per page in an ascending order), or Random (one price per page in a random order). Stein et al.’s (2015) criteria were used to identify unsystematic responses. We hypothesized that the: (1) Random presentation of the task would produce the highest rates of unsystematic data across all three criteria (i.e., Trend, Bounce, and Reversals from Zero), and (2) task presentations would produce differences in purchasing behavior. Specifically, the α and O_max would be lower in Random than the other two presentations, consistent with Salzer et al. (2021).

Methods:

We report how we determined our sample size, all data exclusions (if any), all manipulations, and all study measures. The data, study materials, and analysis code are available upon request by emailing the corresponding author. This study’s design and analyses were not pre-registered. This study was approved by the Virginia Polytechnic Institute and State University Institutional Review Board.

Participants

Participants (n=365) were recruited from April-May 2022 via Amazon Mechanical Turk. Participants were screened and eligible to participate if they: (1) were daily cigarette smokers (i.e., smoked at least 20 cigarettes per day), (2) provided a consistent number of cigarettes smoked per day through a numeric entry and the categorical Fagerstrom Test for Cigarette Dependence (FTCD; Fagerstrom, 2012) question (e.g., 22 cigarettes per day and the choice 21–30 cigarettes per day), (3) lived in the United States, (4) were 21 years old or older, (5) had not previously received compensation from Virginia Tech, and (6) had a ≥ 99% acceptance rate on previously completed studies. Participants were also required to complete a CAPTCHA.

Procedure

Participants completed an approximately 10-minute online survey administered through Qualtrics survey software (Qualtrics, Provo, UT). The survey included HPTs for cigarettes, demographic questions, and the FTCD to measure cigarette dependence. Participants were compensated $2.00 for completion of the survey.

Hypothetical Cigarette Purchase Tasks

Participants were instructed to imagine a typical cigarette smoking day and assume: (1) their normal cigarettes are the cigarettes available for purchase, (2) their income/savings are the same as now, (3) they have no access to any tobacco products other than the task offered cigarettes, and (4) that they consume the cigarettes during that day (i.e., they cannot save/stockpile them). The prices investigated were: $0.00, $0.03, $0.06, $0.12, $0.25, $0.50, $1.00, $2.00, $4.00, $8.00, $16.00, $32.00, $64.00 (Athamneh et al., 2019, 2021; Stein et al., 2018). Three different presentations of the HPT were explored. The List presentation showed all prices on one page in an increasing order. The Ascending presentation showed one price per page in an increasing order. The Random presentation showed one price per page in a random order. Participants were randomly assigned two of the three presentations. The order of the presentations was also random.

Evaluation of Systematic Responding

Participant HPT data was evaluated for systematic responding based on the three criteria proposed by Stein et al. (2015): Bounce, Trend, and Reversal from Zero (defined above).

Sample size calculation

An a priori power analysis was performed using a between-subject design to conservatively identify the number of participants needed for the study. With an estimate of difference in proportion failing a criterion between two presentations of 0.2 (p1 = 0.2, p2=0.4; similar to that reported in Stein et al. (2015)), an alpha of 0.01 to adjust for multiple comparisons, and a power of 95%, we required 193 participants to complete each presentation. Using our incomplete block design (Montgomery 2017), we extended this to the three presentations resulting in at least 579 completed presentations from 290 participants. We sought to recruit at least 300 participants total to account for some missing data through the completion of the study. Note participants completed two out of the three presentations to balance power and participant burden.

Statistical analysis

Demographic characteristics are summarized using mean (standard deviation) and frequency (percentage) where appropriate. The systematic criteria and demand indices for each HPT were evaluated using the R package beezdemand (Kaplan et al., 2019). To evaluate the rates of total criteria passed and passing each individual criterion, we used a logistic mixed model regression with pass/fail as the outcome measure, HPT presentation, and order (a design effect) as the fixed independent variables, and a random effect for participants. This analysis was implemented using the R package lme4 (Bates et al., 2015). To evaluate demand indices, a linear mixed model regression was performed with demand indices as the outcome measure, HPT presentation, and order (a design effect) as the fixed independent variables, and a random effect for participants using the package nlme in R (Pinheiro et al., n.d. version 3.1–145). Derived demand indices were natural log transformed to normalize the data and stabilize variance. Significance of HPT presentation was determined using a Type III ANOVA. Pairwise contrasts were adjusted for multiple comparison using the Tukey method (emmeans; Russell 2019). Significance was defined as p<0.01. Analyses were conducted using R version 3.5.1 (Team, 2018).

Results:

Demographics

Demographics of the full sample can be found in Table 1 (by task pair see Supplementary Table 1). The average age of participants was 38.93 years (±12.04), and the majority of the sample was male (59.5%), White (93.7%), and Non-Hispanic (92.6%). Participants reported smoking an average of 25.63 (±8.29) cigarettes per day and scored an average FTCD score of 6.16 (±1.94), indicating a moderate level of dependence.

Table 1.

Demographics of all participants (n = 365).

Demographics	Frequency (%) / Mean (SD)
Age	38.93 (12.04)
Gender = Male	217 (59.5%)
Race
American Indian or Alaskan Native	5 (1.4)
Asian	4 (1.1)
Black or African American	11 (3.0)
White	342 (93.7)
Multiple Races	3 (0.8)
Ethnicity = Not Hispanic or Latino (%)	338 (92.6)
Marital Status
Single (Never married)	25 (6.8)
Married	328 (89.9)
Living with significant other and sharing financial resources	5 (1.4)
Divorced	5 (1.4)
Widowed	2 (0.5)
Education
High school or less	22 (6.0)
Some college or college degree	225 (61.6)
Some graduate school or graduate degree	118 (32.3)
Family Income
$24,999 or less	49 (13.4)
$25,000 to $99,999	282 (77.3)
$100,000 or more	34 (9.3)
Cigarettes per day	25.63 (8.29)
FTCD	6.16 (1.94)

Systematic Criteria

The total number of criteria passed was evaluated using a logistic mixed model binary regression. Passing was defined as individuals passing all three criteria and failing was defined as individuals failing one or more criteria. We did not observe a significant effect of presentation (X²(2)=4.09, p=0.129) or order (X²(5)=8.47, p=0.132) on total criteria passed.

The proportion of participants passing each unsystematic criterion (Stein et al., 2015) are reported in Figure 1 (A–C). The type III ANOVA for Bounce indicated a significant effect of presentation (X²(2)=13.31, p=0.001) but not of order (X²(5)=12.90, p=0.024). Contrast comparisons using tukey-adjusted estimated marginal means indicated that List was significantly higher than Random (Z=3.65, p<0.001). Ascending, however, was not significantly different from List (Z=−1.78, p=0.176) or Random (Z=1.87, p=0.148). We did not observe a significant effect of presentation (X²(2)=0.25, p=0.881) or order (X²(5)=0.42, p=0.995) on the proportion of individuals passing Reversals from Zero. Similarly, a significant effect for passing the Trend criterion based on presentation (X²(2)=0.52, p=0.773) or order (X²(5)=5.62, p=0.345) was not observed.

Figure 1.

Proportion of individuals passing the A) Bounce, B) Trend, and C) Reversals from Zero criterion from Stein et al. (2015; N=730 tasks). Mean D) R² for model fit, E) Breakpoint (BP₁), F) intensity (ln(Q₀)), G) elasticity (ln(α)), H) price maximum (ln(P_max)), and I) output maximum (ln(O_max)) per group for hypothetical purchase tasks (N = 358 tasks passing all three Stein et al. (2015) criteria). Error bars represent standard error.

Analysis of Demand Data

To evaluate purchasing behavior, we examined the differences of demand data across tasks that passed all three systematic criteria (total = 378; ascending: 123 [53.95%], list: 136 [54.40%], and random: 119 [47.22%]). The R² of the demand curve model fit had a significant effect of presentation (X²(2)=17.89, p<0.001; Figure 1D) but not of order (X²(5)=9.33, p=0.097). Contrast comparisons using Tukey-adjusted estimated marginal means indicated that Random was significantly lower than List (t(153)=−4.08, p<0.001) and Ascending (t(153)=−2.99, p=0.009); Ascending was not significantly different from List (t(153)=−0.89, p=0.648).

The empirical measure of Breakpoint₁ (BP₁) had a significant effect of presentation (X²(2)=13.64, p=0.001; Figure 1E) but not of order (X²(5)=10.66, p=0.059). Contrast comparisons using tukey-adjusted estimated marginal means indicated that List was significantly lower than Random (t(154)=−3.37, p=0.003). Ascending was not significantly different from List (t(154)=−2.88, p=0.013) or Random (t(154)=−0.35, p=0.934). Note 220 tasks (58.2%) did not suppress purchases at $64, the highest price.

Intensity, ln(Q_o), did not have a significant effect of presentation (X²(2)=0.57, p=0.751; Figure 1F) or order (X²(5)=3.19, p=0.671). Elasticity (ln(α)) had a significant effect of presentation (X²(2)=324.94, p<0.001; Figure 1G) but not of order (X²(5)=11.37, p=0.045). Contrast comparisons using tukey-adjusted estimated marginal means indicated that Random was significantly lower (less sensitive to price) than List (t(154)=15.73, p<0.001) and Ascending (t(154)=15.41, p<0.001). Ascending was not significantly different from List (t(154)=0.58, p=0.829).

Further, ln(P_max) did not have a significant effect of presentation (X²(2)=4.99, p=0.083; Figure 1H) or order (X²(5) = 13.33, p=0.021). We observed a significant effect of presentation (X²(2) = 20.26, p<0.001; Figure 1I) but not of order (X²(5)=11.37, p=0.045) on ln(O_max). Contrast comparisons using tukey-adjusted estimated marginal means indicated that Random was significantly higher than List (t(154) =4.44, p<0.001). Ascending was not significantly different from List (t(154)=1.39, p=0.347) or Random (t(154)=2.83, p =0.015).

Discussion:

We found a significant effect of task presentation on the proportion of individuals passing the Bounce criterion. Contrast comparisons indicate that the Random presentation produced significantly more failures of the Bounce criterion than the List presentation. These results are consistent with our hypothesis that Random would produce the greatest amount of failures of the unsystematic criteria and with previous reports of greater unsystematic data among random presentation of prices in an alcohol purchase task (Salzer et al., 2021). In terms of unsystematic data, these findings suggest that local effects, but not global effects, of price are sensitive to task presentation. For example, when comparing larger differences in prices (e.g., $0 and $64), significant differences in the rates of passing the systematic criterion (i.e., Trend) are not seen. However, comparing prices with smaller differences between contiguous prices (e.g., $1 and $2), a significant effect of task presentation with Random producing more unsystematic data than List is found. Finally, local effects in Reversals from Zero were not observed, which could reflect the low rates of failing this criterion across all presentations. This could be because the highest price selected for the purchase tasks failed to suppress purchasing in most participants (observed in analysis of BP₁). The differences we report regarding nonsystematic data (e.g., Bounce) may speak to the feasibility of conducting these studies, both online and in person in regards to resources spent. Additionally, the differences reported here may inform design of future research as increased systematic data may aid our ability to interpret data since nonsystematic data interpretation may not reflect valuation.

We observed significant differences in the R², ln(α), and ln(O_max) between task presentations. Consistent with our hypothesis, Random produced a significantly lower R² than List and Ascending; List and Ascending were not significantly different. We observed that the ln(α) of Random was significantly lower than both List and Ascending (i.e., more inelastic), consistent with Salzer et al. (2021). Also consistent with these results, ln(O_max) was significantly higher in Random than in List, indicating a higher expenditure at price Pmax (Kaplan et al., 2019). Collectively, one explanation for these results is that participants are able to review (i.e., List) their response for the immediately previous price when they are presented in this manner. Participants may not recall their choices at adjacent prices (i.e., immediately before or after the current price) in Random.

Moreover, we report that BP₁ was significantly lower in List compared to Random. One explanation to describe these differences is that presenting all of the prices together may inform participants’ global decision-making during the task. For example, seeing all prices on one page may make participants more conscious of the increases in price and their hypothetical purchasing at these increasing prices. Additionally, the list of prices could inform them of the highest price observed in the task, whereas, in the presentations with one price per page they are not aware of the price ceiling. The reported significant differences in model fits (e.g., R2) and demand indices (e.g., ln(α), ln(O_max), BP₁) may represent procedural variations that may alter noise and measurement error. By pairing specific procedural variations with other methodological improvements (e.g., modeling, participant sampling), we may better be able to optimize purchase tasks to answer specific research questions. The present report, however, does not indicate how the differences in parameter estimates associate with other smoking related behaviors and outcomes.

We did not observe significant differences in ln(Q₀) or ln(P_max), suggesting that purchasing behavior for these indices are not influenced by task presentation. These two parameters are consistent with previous literature examining the effects of progressive ratios versus random sequences of responding in human self-administration studies (Giordano et al., 2001). However, the differences between task presentation in Ascending (i.e., progressive ratio schedule) versus the Random (i.e., a random presentation of fixed ratios) for the demand indice ln(α), are inconsistent with this report. The results presented here are also inconsistent with literature involving the presentation order of fixed ratio among self-administration studies in rats, which report no significant differences in elasticity between presentation order (Maguire et al., 2020). Differences in the designs of these studies should be important considerations when evaluating the discrepancies. First, Giordano et al. (2001) and Maguire (2020) examined self-administration among humans and rats over a defined time period instead of hypothetical purchases for 24 hours. Second, neither Giordano et al. (2001) or Maguire (2020) had a group comparable to the List presentation in our report due to the nature of the self-administration paradigm, making conclusions drawn from comparisons with this group difficult to interpret. Further investigations to elucidate the inconsistencies in these results are warranted.

Data collected online poses limitations, especially regarding data quality (Chmielewski & Kucker, 2020; Mellis & Bickel, 2020). However, to mitigate concerns about data quality on MTurk, participants were required to pass a CAPTCHA and consistency check in the screener to be eligible for the main study. The participants in the present study were mostly male, White, and Non-Hispanic. Future investigations in more diverse populations are warranted. The tasks presented were hypothetical in nature, which may represent a threat to external validity. However, there have been several reports that hypothetical tobacco purchasing is correlated with real-world use (González-Roz et al., 2019; Zvorsky et al., 2019), supporting the validity of the tasks. The present purchase tasks failed to suppress purchasing to zero in a majority of the participants, which may have affected rates of passing Reversals from Zero and the empirical BP₁. Future studies should evaluate higher prices to ensure purchasing suppression and therefore elucidate the full demand curve. Finally, we did not include a fourth group that examined the presentation of prices in a random order on one page. Although we are not aware of any studies implementing this presentation, future examinations could include this group as a comparison to better tease apart contributions of price order and presentation order.

Conclusion:

To avoid unsystematic data, we recommend against using the Random presentation for HPTs. While both the List and Ascending task presentations did not differ in data quality (i.e., systematic data and most indices of purchasing behavior), the List presentation may be preferred over the Ascending presentation due to participant experience. Specifically, the List presentation is contained to one page, which not only may limit potential burden of progressing through multiple survey pages, but also enables participants to compare their answers across prices without having to move backwards in the survey.

Public health significance statement:

The present study evaluated the presentation of hypothetical purchase tasks on unsystematic data and purchasing behavior. We report that purchase task presentation influences the rate of systematic data and certain indices of demand for cigarettes in a cigarette purchase task.

Data availability statement

Readers are encouraged to email wkbickel@vtc.vt.edu to obtain more data for this study.