Grocery store interventions to change food purchasing behaviors: a systematic review of randomized controlled trials

1Food purchasing is a key antecedent of food consumption, and interventions in grocery stores are of interest to those trying to change food purchasing to promote health and those concerned with the marketing and sales of foods and drinks. The goals of each may differ but the types of interventions are similar. These include economic interventions, such as financial incentives and/or disincentives, environmental interventions, which could work at the conscious or unconscious level, and education, or combinations of the above. Evaluating the effectiveness of these interventions is complex. Testing such interventions in real grocery stores is not always feasible, and thus some of the aforementioned strategies have been evaluated within simulated (e.g., virtual) stores, with potential for different effects. In addition, interventions to change food purchasing may attenuate or exacerbate health disparities. Socioeconomically disadvantaged populations are more likely to suffer from nutrition-related morbidity, and there is some evidence to suggest that certain interventions, particularly those relying on executive functioning, may be more effective in more socioeconomically advantaged groups. On the other hand, disadvantaged groups may be more sensitive to economic interventions.

2Previous systematic reviews of grocery store interventions are either now outdated or more narrow in scope than ours [e.g., focus exclusively on interventions designed to promote health, or do not include price or labeling, or were conducted in specific populations], which restricts the ability of researchers and policy-makers to develop a comprehensive picture of the extant evidence. Here we focus on randomized controlled trials (RCTs) that evaluated the effectiveness of interventions implemented in grocery stores to change purchasing behavior and consumption, with no restrictions by intervention or population type. We also set out to examine how, if at all, effectiveness varied based on intervention components, setting (real compared with simulated) and socioeconomic status (SES).

3Our aim was to understand the effectiveness of interventions in grocery stores to aid the development of strategies to improve public health and to reduce inequalities, and to identify evidence gaps.

4A protocol was published in advance and is available in PROSPERO (https://www.crd.york.ac.uk/PROSPERO/; CRD42017068809). Methods for searching, screening, data extraction, and quality assessment followed those set out in the Cochrane handbook.

5We searched 13 electronic databases on 2 June 2017 using terms relating to grocery stores, food and nonalcoholic beverages, purchase and choice behaviors, and randomized controlled trials [see protocol for full strategy; Supplemental Table 1 for MEDLINE search strategy]. We also screened reference lists of included studies.

6We included RCTs when interventions were designed to change the purchase of any foods, nonalcoholic drinks, nutrients, energy, or products belonging to a defined dietary pattern or with defined dietary scores and when purchases of any of the above were reported at the individual or store level (our primary outcome). Our secondary outcome was participants’ consumption of the above items. To be included, interventions must have been implemented partly or completely in any online or physical grocery store, including simulated stores. Studies must have had a minimal control or a comparison between ≥2 interventions fulfilling the aforementioned criteria.

7Two reviewers independently screened studies for inclusion at title/abstract and full-text stage, extracted data with the use of a predefined and prepiloted data-extraction form, and assessed risk of bias with the use the Cochrane risk-of-bias tool, with discrepancies resolved by discussion or referral to a third reviewer. Data were extracted on: recruitment methods; inclusion/exclusion criteria; population; setting; intervention and comparator characteristics; outcomes; and whether these varied by socioeconomic status. When needed, we contacted authors for further information via email.

8We conducted a narrative synthesis of the data, tabulating results for our primary and secondary outcomes from the original study reports. When multiple time points were available, we chose that measured during or as close as possible to the end of the intervention. We did not conduct a meta-analysis due to substantial clinical heterogeneity with regards to the reported outcome, outcome measures, and study designs. We classified interventions into one of 4 categories, informed by existing literature: Economic interventions [any intervention including a price increase, decrease, or financial reward].Store environment changes [any intervention involving changes to the microenvironment, but not including economic interventions which are covered by (A), swaps which are covered by (C) or interventions based on product labeling or consumer education alone which are covered by (D)].Swap interventions, which offer consumers the opportunity to replace their usual food with a healthier alternative [but not including economic interventions, which are covered by (A)].Labeling and/or educational interventions [interventions involving product labeling and consumer education/information, but not economic or other store environment changes].

9Economic interventions [any intervention including a price increase, decrease, or financial reward].

10Store environment changes [any intervention involving changes to the microenvironment, but not including economic interventions which are covered by (A), swaps which are covered by (C) or interventions based on product labeling or consumer education alone which are covered by (D)].

11Swap interventions, which offer consumers the opportunity to replace their usual food with a healthier alternative [but not including economic interventions, which are covered by (A)].

12Labeling and/or educational interventions [interventions involving product labeling and consumer education/information, but not economic or other store environment changes].

13We present results separately for real and simulated settings.

14In addition to our narrative synthesis, we also employed an exploratory crisp-set qualitative comparative analysis (QCA) to identify combinations of intervention components associated with statistically significant changes (P < 0.05) in the desired direction for at least one of the foods targeted by the intervention. We used 5 variables in our QCA. The first 4 were modelled on the groups above but were not mutually exclusive, namely whether or not the intervention involved an economic component (as per group A) or changes to the store environment (as per B) or swaps (as per C) or consumer education/information (as per D). The fourth variable was whether or not the intervention was based in a real grocery store. We only included comparisons between eligible interventions and minimal controls. We excluded configurations that originated from multiple similar interventions tested in one single study. QCA is a method that aims to identify variables present when an intervention is effective. The analyses were conducted with the use of fsQCA software, with a consistency threshold of 0.75 above and a frequency threshold of 2. Accordingly, a combination of intervention characteristics was defined as being associated with “significant changes in the desired direction for at least one of the foods targeted by the intervention” when ≥75% of all interventions with this combination, and ≥2 such interventions, were associated with the aforementioned outcome. Though prespecified in our protocol, this was an exploratory analysis used to augment the narrative review.

15Excluding duplicates, 1466 references were retrieved from database searches, with 1 additional paper from screening reference lists. We assessed the full text of 135 studies, 100 of which were excluded, most commonly because the study was not an RCT or did not measure purchasing behavior (Supplemental Figure 1). We included 35 studies, representing 55 references.

16The key characteristics of included studies are summarized in Table 1 and described below, with more detail in Supplemental Tables 1 and 2.

17Twenty-two studies randomized at the individual level. The remainder randomized at the store or community level, 10 of which did not report the number of participants/customers included. Across those studies that reported it, the number of participants included in this review was 20,156. One study was conducted in children (mean age 11 y). When reported, mean age across studies in adults ranged from 29 to 52 y (median 42 y). Nine studies reported BMI (in kg/m2); when reported, means ranged from 25.8 to 30.2. In the 23 studies that reported gender, all were predominantly female (range 55–100%, median 81%). In the 15 studies that reported ethnicity, 11 had a majority of white/Caucasian participants. Twelve studies had inclusion criteria or recruitment settings that specifically targeted people from socioeconomically disadvantaged groups. When reported, the remainder (11 studies) included people predominantly of middle or high SES.

18Eighteen studies were conducted in the United States, 6 in the Netherlands, 3 in Australia and New Zealand, and 1 each in Canada, China, France, Sweden, and the United Kingdom. Across those studies that reported store number, 807 stores were included. Nineteen studies described study area: 10 were conducted in urban/metropolitan settings; 2 in suburban settings; 3 across mixed settings; and 4 in rural areas.

19Table 1 and Supplemental Table 1 contain more detail.

20Twenty-seven of the studies consisted of interventions conducted in functioning grocery stores that existed outside of the research context: 21 exclusively in physical supermarkets; 3 exclusively in convenience/corner stores; 2 in supermarkets and convenience/corner stores; and 1 in an online supermarket. The remaining 8 studies were conducted in simulated supermarkets. Overall, the 35 included studies represented 89 intervention arms and 28 control arms (no intervention) that met our inclusion criteria, with 57 intervention versus control comparisons. The vast majority of interventions (81 of 89) were implemented solely in the store environment.

21Thirty-one of 35 studies aimed to promote health, whereas 2 aimed to increase store profit, 1 aimed to increase the volume of food purchased, and 1 aimed to increase sales of a specific (nonhealth-related) product. Intervention length ranged from a one-off shopping trip to 2 y. Interventions typically consisted of multiple components, which are summarized below (for more detail see Table 1 and Supplemental Table 2).

22Of the 89 interventions, 43 were economic interventions (group A). Of these, 13 involved price increases, 35 involved price decreases, and 1 involved financial rewards that were received post-shopping. Eighteen also involved advertising or signage in-store, 17 also provided education or information to consumers, and 2 each also involved in-store taste testing and changes to item stocking levels.

23A further 30 interventions involved changes to the store environment with no economic components (group B). Twenty-one involved signage, 5 altered item placement, and 13 involved other changes, including partitioned grocery carts, and providing convenience stores with additional refrigerated units for produce. Twenty-seven also provided education or information to consumers.

24A further 6 interventions involved suggested swaps (group C), either as a standalone intervention or along with additional educational components.

25The remaining 10 interventions consisted of consumer education or information, or product labeling, without any additional economic or store environment changes (group D). Six of these evaluated different forms of product labeling; the remainder provided educational information (typically in the form of print leaflets) to participants.

26Our primary outcome was purchasing behavior. Twenty-nine studies measured purchasing during the intervention and for the remaining 6 we used immediately postintervention data. Twenty-five studies reported purchases at the individual level and 11 at the store level. Outcome was measured objectively (typically via sales data or transaction data) in all studies except 5 in which a self-reported measure was used (Supplemental Table 2). Five also reported on consumption as an outcome, typically via dietary questionnaires.

27Ten studies were judged to be at low risk of bias across all domains assessed (i.e., at low risk of bias overall) and 14 at high risk of bias in ≥1 of the domains assessed (i.e., at high risk of bias overall). The remaining studies are considered at unclear risk of bias. Table 2 lists judgments by domain for individual studies.

28The effects of the interventions on our primary and secondary outcomes are summarized by group. No studies reported statistically significant results for our primary or secondary outcomes that were in the direction opposite from that intended. Tables 3 and 4 contain numeric data (when available) for the results presented below; Supplemental Table 3 provides further data on relevant outcomes at our primary time point.

29Twelve studies tested economic interventions in real store environments; 11 applied discounts on target items at time of purchase and 1 provided store vouchers after purchase. Four also reported consumption as an outcome.

30All but 1 of the 9 studies comparing price decreases with control detected a statistically significant increase in purchases for ≥1 of the target items (see Table 3); none of the studies that decreased prices of healthy food reported increases in purchases of unhealthy items. Both Anderson et al. and Ni Mhurchu et al., reporting studies that aimed to increase healthy food purchases, found a statistically significant increase in purchase of target items and a statistically significant decrease in purchases of fat. Three studies aiming to increase purchases across a range of items only found differences for one of the products measured; Ball et al. (targeting fruit, vegetables, and beverages) only detected a statistically significant increase in vegetable purchases but no differences in consumption; Geliebter et al. (also targeting fruit, vegetables, and beverages) found an increase in purchasing of fruit and vegetables, but no differences in consumption and no differences in beverage purchases; and Budd et al. only detected a statistically significant increase in healthier snack foods, despite also targeting beverages, vegetables, and whole-wheat bread. In Kristal et al., a 50-cent coupon for fruit and vegetables affected neither purchase nor consumption. In contrast, in Waterlander et al. a much larger, 50% discount led to a statistically significant increase in fruit and vegetable purchase and consumption. Ma et al., reporting a study that aimed to decrease salt consumption through subsidy of a salt substitute, found a statistically significant increase in purchase of the salt substitute. In Phipps et al., financial rewards postpurchase led to a statistically significant increase in fruit and vegetable purchase. Smith et al. aimed to increase overall food purchase in food-insecure households and detected a statistically significant increase in food expenditure through provision of vouchers.

31In Brimblecombe et al., both study arms received discounts on fruit, vegetables, and water, but one arm also included in-store posters, activity sheets, taste testing, and cooking demonstrations, and there was no control arm. The arm with the added components purchased significantly more vegetables, though no statistically significant differences in purchases were observed for the other target items. Dhar et al., reporting a study that aimed to increase store profit, measured a statistically significant increase in 1 of the 2 target items with additional signage, over and above discounts alone (which also were associated with an increase).

32Six studies tested economic interventions in simulated environments, and all detected a statistically significant effect on at least one of the measured outcomes. However, they varied in whether they impacted total energy purchased. In Epstein et al., a study that aimed to increase nutrient quality and decrease energy, both subsidies (12.5% or 25% for healthier items) and taxes (12.5% or 25% for unhealthy items) led to an increase in purchases in the healthier items compared to control, but neither altered total energy purchased, whereas in Nederkoorn et al., a 50% tax on high-energy foods significantly decreased total energy purchased compared with control. Waterlander et al. used a factorial design to compare various discounts (none, 25% or 50% on healthier foods) and various taxes (5%, 10%, or 15% on unhealthier foods); participants receiving a 50% discount purchased significantly more healthy foods, but also purchased significantly more energy, which was the same in Waterlander et al.. There were no significant effects of the different price increases and no significant interactions. Waterlander et al. compared a 25% discount on fruit and vegetables with no discount; the discount led to statistically significantly greater fruit and vegetable purchases with no differences in purchases in other food categories. In Waterlander et al., a 19% tax on sugar-sweetened beverages led to significantly fewer purchases of these beverages.

33Eleven studies, all in real stores, tested interventions altering the store environment that did not involve an economic component. None reported consumption as an outcome.

34Eight studies in real stores compared a range of interventions with control, with mixed results. Of the 3 studies manipulating item availability (e.g., changing item stocking) among other components, 2 detected an effect. Dreze et al. compared 2 different interventions with control: the intervention that involved changes to placement only (e.g., changes to where items are located in store, referred to by the authors as “space to movement”) resulted in changes opposite to the intended effect, whereas the intervention that also manipulated availability led to a statistically significant increase in 4 of the 5 categories targeted. In Ma et al., stocking of a salt substitute and consumer education led to a statistically significant increase in salt substitute sales compared with control (no salt substitute). However, Lent et al., which targeted students from grades 4 to 6 and involved a multicomponent intervention including item availability, other store changes, and other components delivered in school, failed to detect an effect for any of the outcomes measured.

35The remaining 5 studies employed a range of different store environment changes without altering item availability. Two of the 5 detected a statistically significant effect: Foster et al. used advertising, signage, item placement, and taste testing to promote healthier items in 5 product categories and detected a statistically significant increase for ≥1 product in favor of the intervention in each category, and Thorndike et al. increased fruit and vegetable purchases through increasing visibility and quality of fresh produce. In contrast, in Jeffery et al. signage, recipes, and brochures had no impact on the purchase of low-fat foods, and in Budd et al. no significant difference was found for any of the outcomes measured when comparing an arm with advertising, signage, changes in shelf-height, taste testing, and consumer education with control. Finally, Russo et al. tested 12 different types of signage; no significant differences were found when compared with each other or with a no-signage control.

36In Elofsson et al., climate-related store signage increased sales of climate-certified milk compared with signage without climate-related information. In Milliron et al., a nutrition-based intervention involving shelf-tags, educational leaflets, and an information session by an in-store dietitian led to statistically significant improvements in 2 of the 6 purchasing outcomes compared with shelf-tags alone. Finally, Wansink et al. partitioned shopping carts, indicating a target proportion of the cart for produce, led to increased purchase fruits and vegetables, with the effect greatest when flyers highlighting nutritional benefits were distributed as opposed to flyers promoting cost savings.

37Three studies tested interventions that involved swaps, 2 in real environments and 1 in a simulated online grocery store. The 2 studies in real environments detected statistically significant effects: Huang et al. automatically suggested swaps in an online supermarket and observed a statistically significant decrease in saturated fat purchased; and in Winnett et al., a nutrition-based in-store computer kiosk in which participants entered their intended purchases and swaps were suggested to promote healthier choices led to statistically significant differences in favor of the intervention for 3 of the 7 target categories (the remaining 4 were not reported). In contrast, Forwood et al. tested suggested swaps in a simulated online supermarket with a focus on reductions in energy density and did not detect any significant differences in purchasing.

38Five studies tested interventions that involved only provision of consumer education/information or product labelling. All studies in this group included a no-intervention control. One measured consumption. Three studies tested the provision of consumer education in real stores, with mixed effects. In Achabal et al., printed materials did not increase produce sales. In Ball et al., newsletters and a supermarket tour did not change produce purchased, though self-reported vegetable consumption was statistically significantly higher in the intervention group. In Winnett et al., an educational intervention primarily delivered in the home but including a supermarket visit with study staff reduced simple carbohydrate purchases, but did not alter the other 7 purchasing measures.

39The remaining 2 studies evaluated nutritional labelling. In Ni Mhurchu et al. there was no significant difference in healthiness of packaged food purchases when an app to show either traffic light labels or health star labels were compared with control in a physical supermarket. In an online experiment, Ducrot et al. found that 3 of the 4 labels (5-color nutrition label, green tick, and multiple traffic lights) led to statistically significant benefits for ≥1 of the measured outcomes compared with control; labels based on guideline daily amounts did not show an effect.

40Although many studies adjusted results by SES, only 6 presented analyses testing if results differed by SES (Supplemental Table 4). In 4 studies of in-store interventions, including labeling, in either real stores or simulated environments, there was some evidence of greater benefits for less-deprived groups. In the 2 studies of nutrition labels, Ni Mhurchu et al. (real) found significant interactions by income with control more effective than traffic light or health star labels for low-income participants, and Ducrot et al. (simulated) found that the effect of labels, though still present, was smaller in low-income participants. In the 2 studies of swaps in an online supermarket, Huang et al. (real) found no significant difference in effect by education, employment, or income, whereas Forwood et al. (simulated) found that less-deprived participants were more likely to accept swaps. However, in the 2 studies of price decreases that analyzed results by SES (1 real, 1 simulated), the intervention effects did not differ by income, education, or budget.

41Results from our exploratory QCA are presented in Supplemental Tables 5 and 6. In summary, results pointed to the effectiveness of economic interventions regardless of setting and of environmental interventions and swap interventions in real-life settings (Supplemental Table 5). The 4 configurations associated with statistically significant changes in the desired direction for ≥1 of the foods targeted by the intervention were as follows: 1) economic interventions in real and simulated grocery stores (without education, environmental components, or swaps); 2) economic interventions in real grocery stores (without environmental components or swaps, and with/without education components; there were no studies in simulated environments which would have fitted this description); 3) environmental interventions in real grocery stores (without swaps or economic or environmental components); and 4) swaps with education in real grocery stores (without environmental or economic components). These configurations covered 85% of the effective interventions included in the QCA. When we tested the inverse—namely which configurations, if any, were not associated with statistically significant changes in the desired direction for at least one of the foods targeted by the intervention—the only intervention configuration that emerged was education in real grocery stores without economic components or swaps (Supplemental Table 6).

42This review includes 35 studies and, to the best of our knowledge, is the first to synthesize evidence from RCTs in grocery stores across a wide range of intervention types. The vast majority of studies (29 out of 35) aimed to improve health and we interpret the remainder in the context of their lessons for public health strategies. Overall, economic interventions showed the most promise, with 8 of the 9 studies in real store environments and all 6 studies in simulated environments detecting a statistically significant effect. The effects of these interventions appeared to be enhanced by additional promotional activity. Swap interventions appeared promising in real grocery stores, but only 2 studies tested them in this context. Interventions that altered the store environment showed mixed effects. In interventions that consisted solely of consumer education, findings were positive in simulated environments but for the most part no effect was detected in real grocery stores. The very limited data available suggested that the effects of economic interventions did not differ by SES, whereas studies of other in-store interventions presented evidence of both positive and negative impacts.

43Although this review included 35 studies, with >20,000 participants across >800 grocery stores, important gaps in the evidence remain. All but one study was conducted in a high-income country, though interventions are also required in middle- and low-income countries, where grocery store shopping is on the rise and predicted increases in diet-related disease are the greatest. In addition, due to the practical limitations of testing such interventions in RCTs, we found no interventions in real settings testing the effect of price increases. Only 5 studies measured consumption as well as purchasing, and these found mixed results. Research suggests that objectively documented household food purchases yield a reasonably accurate estimate of overall diet quality, but some caution must remain in interpreting purchasing as a proxy for consumption, particularly in regard to intake of specific nutrients. More studies are also needed to test whether the impact of interventions varies with SES, so as to avoid widening existing health disparities.

44There are also questions about the applicability of this evidence. The interventions that appear most effective—namely, those manipulating price and those suggesting tailored swaps based on an individual's shopping list—may also be some of the most difficult to implement. Although some individual studies of environmental interventions showed promise, particularly regarding availability, these need replication before widespread implementation given mixed results across the body of evidence. This review also raises questions about the external validity of findings from simulated grocery stores, particularly with regard to educational interventions, which appear effective in simulated grocery stores and ineffective in real grocery stores. The lack of effect in real stores may be due to a greater lag between exposure to the message and enacting the behavior, or the presence of other competing information and cues that may influence purchasing decisions. Lastly, it is questionable whether findings from the 4 nonhealth-based interventions are directly translatable to public health interventions in a grocery store setting.

45Although other systematic reviews overlap in scope with ours, to the best of our knowledge this review is the most comprehensive and up-to-date in a field that has seen a recent upsurge in research. The most similar recent review to ours did not include pricing and product labeling and ran its searches in 2015; hence, it contains only 11 RCTs compared with our 35. A 2013 review of grocery-store based interventions to promote health contains only 6 RCTs. In addition, we are the first review in this area to use QCA, a technique that augmented our narrative synthesis and was particularly valuable for exploring the variation in results between real and simulated settings.

46Multiple reviews in this area have, as their headline conclusion, stated that more research is needed. Those reviews that drew conclusions on effectiveness are summarized below. Cameron et al. noted that shelf-labeling interventions appeared promising when evaluating on a wide range of study types, but when restricted to RCTs, as in the present analysis, interventions of this type showed mixed results. The review by Adam et al., which was limited to only obesity-related interventions, found that interventions combining price, information, and easy access and availability to health foods appeared promising but would need to be carefully implemented; findings on price and access were similar to ours. The review by Thow et al., which evaluated taxes and subsidies only, was consistent with our review in concluding such interventions are likely to be effective in altering purchasing behavior.

47This review uses gold standard methods, as set out by Cochrane, to minimize bias. Restricting to RCTs minimizes confounding, making us more confident in our results than some previous reviews, but arguably also restricts the nature of interventions that our review is able to evaluate. However, despite restricting our studies to RCTs, we judged 14 of the 35 studies to be at high risk of bias, and, despite searching trial databases and conference abstracts, cannot rule out the possibility of publication bias.

48The scope of our review means that we are able to provide a broad appraisal of the evidence available for grocery store interventions across a range of settings, aims, and population groups, and to compare different intervention types. However, this inevitably brings with it heterogeneity and a large volume of data. When planning this review, we made a number of pragmatic decisions to deal with this heterogeneity. First, we chose to focus on results during or immediately postintervention and therefore are only able to draw conclusions regarding the effects of these interventions whilst implemented. Second, to synthesize the data we had to categorize it by intervention type; for some interventions (e.g., economic, education-only), this classification was relatively straightforward, but we are aware that other researchers may have defined some interventions differently, particularly labeling and swaps. In the absence of a clear consensus on how to group these types of interventions, we were guided by the existing literature. We also acknowledge that most studies were conducted in higher-income countries, limiting the generalizability to low- and middle-income countries, which may be the target of future studies aiming to improve diet as these countries undergo economic transition.

49The explorative QCA represented a novel and empirically driven approach to help categorize and identify patterns in the studies we included. We recommend caution when interpreting the results of this analysis, as some studies measured the interventions’ impact on the purchase of multiple foods, thus inflating the probability of finding significant effects by chance.

50Finally, our scope limits the amount of information we are able to present for each study. Our hope is that researchers and public health professionals aiming to explore more granular questions can use the data contained in this review, including the Supplementary data, as a starting point; interested readers are encouraged to contact the authors for further data.

51This review draws upon the best available evidence from RCTs and in doing so highlights the range of opportunities to change purchasing behaviors in grocery stores. Although the changes detected in purchasing were often small, given the scale of poor diet as a public health issue and the key role of grocery stores in shaping food consumption at a population level, our findings suggest interventions implemented in these settings—particularly ones that manipulate price, suggest swaps, and perhaps manipulate item availability—may play an important role in a multifaceted public health approach to reducing diet-related disease.