We used a three-dimensional (3D) web-based virtual supermarket (the SN VirtuMart) to test the efficacy of nudging, pricing and combined nudging and pricing strategies. This experimental study was part of the ‘Sustainable Prevention of Cardiometabolic Risk through Nudging Health Behaviors’ (Supreme Nudge) project [21]. This randomized trial (NTR7293) was registered in the Dutch trial registry.
The virtual supermarket was designed to simulate a real-life shopping experience by imitating a typical Dutch supermarket (i.e., layout of the shelves, colors, products and product prices). The virtual supermarket included almost 1200 unique name-brand and budget-brand products categorized into 12 food groups. The quantity and variety of products was such that participants with a variety of household sizes and budgets were able to do their weekly shopping in the virtual supermarket. For example, the proportion of healthy products within the SN VirtuMart was comparable to the proportion found in real-life supermarkets. In the SN VirtuMart, 19% (221 out of 1175) of products were considered to be healthy (as based on the Dutch dietary guidelines [22]) compared to 16% of products in Dutch supermarkets that are healthy in terms of being fresh, unprocessed or lightly processed foods [23]. Common marketing, branding and promotion techniques as well as sounds and background noise were used in all shopping conditions to simulate real-life supermarket shopping experiences. The virtual supermarket software was pilot-tested prior to the study, but no formal usability test was conducted. The virtual supermarket was designed by co-author L.N. van der Laan (Van der Laan LN, Papies EK, Ly A, Smeets PA: How health goal priming promotes healthy food choice: a virtual reality fMRI study, submitted). The study design and procedures were approved by the Medical Ethics Review Committee of VU University Medical Centre (OHRP: IRB00002911). More information regarding the SN VirtuMart program and selection of foods and beverages can be found in the Additional Material.
Study design
This study used a mixed randomized experimental study design consisting of five study conditions (within-subject design) and three study arms (between-subject design). Participants were randomized into one of the three study arms (25% price increases, 25% price discounts, or 25% price increases and discounts) and within these arms exposed to five study conditions (control, nudging, pricing, price salience and price salience with nudging). A 25% price change was chosen based on the finding that at least a 20% price change is needed to result in significant effects on population health [24] and based on discussions with a Dutch supermarket chain regarding what price changes would be feasible in real-world supermarkets [25]. The order in which the participants received the five study conditions was also randomized. Table 1 displays the study design.
The control condition represented regular supermarket price promotions and product placement. The nudge condition included salience nudges to promote high-fiber products, frozen vegetables and low-fat dairy products. Salience nudges are nudges that draw individual’s attention towards a particular option through for example the use of arrows or frames (Additional Material). The price condition represented discounts in the prices of healthy products and/or increases in the prices of unhealthy products, depending on study arm. In the price salience condition, price increases and discounts were implemented and communicated to participants. Price increases were communicated to participants by showing a newspaper announcement of an unhealthy food tax of 25% on their screen, before they entered the supermarket. Price discounts were communicated with price promotion signs within the virtual supermarket. The last shopping condition was a combination of the nudge and price salience condition.
Participants randomized to the price increases arm were exposed to 25% price increases on 37% of the available unhealthy products (n = 356/955) (i.e., pizza, white bread, confectionary, sugary drinks, high-fat and/or high-sugar dairy products, salted nuts and sweet bread spreads). Participants randomized to the price discounts arm were exposed to 25% discounts on 89% of healthy products (n = 195/220) available within the virtual supermarket (i.e., all fresh and frozen fruits and vegetables, canned vegetables, high-fiber bread and bread alternatives, whole-wheat pasta and rice, low-fat and low-sugar dairy products, fish, unsalted nuts, water and tea). Participants randomized to the price increases and discounts arm were exposed to both the price increases and decreases. Products were considered to be healthy based on the Dutch dietary guidelines [22] and unhealthy products were all products not recommended by these guidelines.
Randomization and masking
Two online block randomizer generators were used to allocate participants equally to the arms and to determine the order of the conditions. Participants were masked to the nature of the conditions they were assigned to and were not aware that the study aimed to evaluate the effect of nudging and pricing strategies. Instead, they were told the study was about shopping behaviors in general. The online registration, randomization procedure and analysis of the data were all conducted by the research team.
Participants and recruitment
Details on the recruitment of participants are described elsewhere [17]. Briefly, the aim was to include an approximately equal distribution of low and high SEP individuals. Participants were recruited from the general Dutch population through targeted Facebook advertisements and home-delivered flyers in selected low-SEP neighborhoods throughout the Netherlands. The advertisements and flyers directed participants to the registration website. Upon signing online informed consent, participants received a questionnaire assessing eligibility criteria and socio-demographic characteristics. Inclusion criteria were: being an adult (18 years or older); being the main household shopper; being able to read in Dutch; and to have an email address. Participants were excluded if another household member was already participating in the study, or if they did not have a computer or laptop. Participants who met the inclusion criteria and completed a training task of ‘buying’ five specific products were included in the study.
Sample size
The sample size calculation was based on previous literature [12, 26]. We determined that a sample of 150 participants (i.e., 50 participants in each pricing arm) completing all five conditions would be adequate to detect, among others, a target difference of 135 (SD: 370) grams of vegetables per week between the control condition and nudging condition (level of significance 0.05, power > 0.90). Larger differences were expected for the pricing conditions compared to the nudging condition. Also, we expected the salient pricing strategies to be larger than the non-salient pricing strategies and that the combination of price increases and discounts would be larger than the single pricing strategies. Lastly, the largest difference was expected when combining all strategies (i.e., nudges and pricing strategies and price increases and discounts). The same standard deviation of 370 was used for all conditions. In order to be able to have enough power to stratify the results by low and high SEP, we aimed to include double the sample, leading to a total sample of 300 participants who completed all five shops. We aimed to oversample participants and monitor their drop-out so that we would end up with 300 participants. More information regarding the sample size calculation has been reported elsewhere [17].
Purchasing task
Participants were asked to perform five shops in the virtual supermarket over five consecutive weeks. During each virtual supermarket visit, participants were asked to do their regular weekly household groceries, using a virtual budget. Participants’ shopping budgets (eight categories) were based on self-reported actual grocery shopping budgets and they were only able to leave the supermarket at the check-out if they had spent between 50 and 125% of their budget [17].
Outcome measures
The primary outcome was the percentage of healthy products based on the healthy and unhealthy purchases in grams per week. This outcome measure was chosen as pricing and nudging strategies aim to increase the proportion of healthy purchases through increasing healthy purchases and/or decreasing unhealthy purchases. The total grams of healthy and unhealthy products purchased per week were used as secondary outcome measures. These secondary outcome measures can help explain whether the percentage of healthy purchases changed due to; 1) a change in healthy purchases (which is expected in the cases of price discounts and nudges), 2) a change in unhealthy purchases (which is expected in the case of price increases), 3) or a change in both healthy and unhealthy purchases (which is expected when combining price increases and discounts). In the original trial registry, the primary outcomes included the percentage of healthy purchases as well as the purchases of healthy and unhealthy foods. However, given that price increases mostly affect the purchases of unhealthy foods and price discounts the purchases of healthy foods, it would not be very informative to calculate the purchases of healthy and unhealthy foods for the overall sample. These were therefore classified as secondary outcome measures.
To investigate if participants spent more money in the virtual supermarket, especially when exposed to price increases, we also investigated the total amount spent in the virtual supermarket per week in Euros (tertiary outcome). Sensitivity analyses were conducted with the number of healthy and unhealthy products purchased per week and the percentage healthy products based on the number of healthy and unhealthy products purchased.
Covariates
The baseline questionnaire asked participants for their age, sex, educational attainment, household net monthly income, usual weekly budget spent on groceries, weight and height. Self-reported height and weight were used to calculate body mass index (BMI - kg/m2). BMI was dichotomized according at overweight status (BMI ≥ 25 kg/m2).
Educational level and income were used as two separate proxies for SEP because they assess different aspects of SEP. Educational level was categorized into two groups: low educational level included those who completed primary education, intermediate vocational education and higher secondary education, and high educational level included those who completed higher vocational education or university. In order to adjust household income for household size, we used the OECD-modified equivalence scale [27]. After this adjustment, low income was defined as a value equal to or below the median individual income of €1743 per month and high income was defined as all values above this median individual income. The average monthly gross income in the Netherlands in 2017 was €2667 [28].
Statistical analyses
Descriptive statistics for socio-demographic variables and the outcome variables were reported using percentages, means and standard deviations or medians and interquartile distances in case of non-normality. Mean changes from the control condition were analyzed using a maximum likelihood-based repeated measures approach including a random intercept for participants to account for the clustering of shops within participants. An exchangeable covariance structure was used to model the within-participant errors. The research aim regarding the independent and combined effect of nudging and several pricing strategies was assessed using a linear mixed model with the percentage of healthy purchases as the only outcome (aim 1). Only for the single nudging condition, a linear mixed model was used to investigate the effect of nudges on the secondary outcome measures (i.e., healthy and unhealthy purchases) without stratifying for the pricing arms. The differential effects of price increases and discounts (aim 2) was assessed by stratifying the linear mixed model by the three pricing arms for both the primary (percentage of healthy purchases) and secondary outcome measures (amount of healthy and unhealthy purchases in grams). In order to investigate the third aim regarding the added value of salient price increases, the reference group included the pricing only condition instead of the control condition. To investigate effect modification by SEP (aim 4), the analyses were stratified for SEP indicators (i.e. educational level and income separately). For the within-subject analyses, participants that completed at least two shops contributed to the analyses. Whether there were statistical differences between SEP strata was examined by comparing the model without interactions with the model with interactions between conditions and SEP using the likelihood ratio test, separately for the three pricing arms. Additionally, the total amount spent in the virtual supermarket for each experimental condition compared to the control condition was analyzed and can be found in the additional material. Furthermore, sex differences were investigated because males and females may react differently to price changes due to differences in competing factors such as perceived quality, price, taste and habit [29] (Additional Material). Lastly, in order to determine whether a possible order or learning effect occurred due to the study design, we investigated the effect of the intervention period (ranging from week 1 to week 5) on the percentage of healthy purchases. Also, we compared the unadjusted beta coefficients to the period adjusted beta coefficients of the effect of the experimental conditions on the percentage of healthy purchases (Additional Material).
Analyses were conducted in STATA version 14.1 and the absence of zero in the 95% confidence interval or a p-value of 0.05 or smaller was regarded as a statistically significant effect. We did not adjust for multiple testing given the fact that we used a single primary outcome, and the findings from the secondary outcomes were used to explain the primary outcome findings.