Online grocery store
An online grocery store (NUSMart Online Grocery Store) was developed for testing these hypotheses (https://nusmart.duke-nus.edu.sg). At the time of the trial, NUSMart contained over 3200 food and beverage products commonly purchased at local supermarkets in Singapore. The web store was designed to mirror actual web-based grocery stores in Singapore, such as RedMart Online Supermarket (https://redmart.com) or Fairprice Online (https://fairprice.com.sg), in both look and feel. It contained products across major food & beverage categories, including:
All products included pictures of the items, current retail price and product descriptions. NUSMart operated similar to other on-line grocery stores with a cart that filled as consumers shopped and the ability to add and remove products and review purchases before hitting the checkout button.
Participants and procedures
There were no face-to-face visits between the study team and participants as all study-related procedures were conducted online. A power calculation revealed that at least 140 participants were required to detect differences between any two arms assuming an effect size of 0.3 (a relatively small effect) or larger, for the proportion of the basket represented by Lower Calorie products (primary outcome). The calculation assumed a two-tailed test, three comparisons, power of 0.9, alpha of 0.05, and a crossover design. Based on these inputs and an assumed attrition rate of 20%, our target sample was 168.
Participants were recruited from existing users of the online grocery store, RedMart, via Facebook advertisements. At the point of study, RedMart was the largest online grocery store in Singapore and thus had the largest customer base. Facebook advertisements were chosen as the recruitment platform as it was RedMart’s main platform for online communication with their consumers. Recruiting existing online grocery shoppers ensured that participants would be comfortable with online shopping. Prospective participants were directed from RedMart’s Facebook page to the study website (https://nusmart.duke-nus.edu.sg) and asked to complete an online screening questionnaire to determine their eligibility for the study. Potential participants were eligible only if they were 21 years of age or above, the primary grocery shopper for their household, and a registered RedMart shopper.
Potential participants who were both interested and eligible were then asked to complete: 1) a registration form containing name, delivery address, National Registration Identity Card number, mobile number and email address; 2) an online consent form; and 3) the baseline questionnaire. Upon completion of the registration form, the website created the participant account and unique participant identification number for use throughout the study.
We designed a simple ‘Lower Calorie’ directive logo (see Fig. 1) [25] to test the hypotheses. We made it primarily green in color as green labels have been shown to increase perceived healthfulness of foods [26] and included a smiley face nested within the label to act as a signal that the label is referring to a positive choice. The final label was thus both simple and directive for ease of comprehension [25].
Arm 1 was the Control condition, which did not display the label on any products. Arm 2 displayed the label on the 20% of products that were lowest in calories per serving within each product category (full category list available in Supplemental Table 1). Arm 3 displayed the label on the 20% of all products that were lowest in calories per serving. Prior to conducting the analysis, we standardized the serving size by using the mean serving size within each subcategory. This standardization ensured that similar products were compared equally as serving sizes can be arbitrarily set by the manufacturers [27]. The labels were displayed below the product images (See Fig. 2).
Using a crossover design, all participants were exposed once to the three shopping conditions (1xControl, 1xWithin-category, 1xAcross-category) in random order (see Supplemental Table 2). Each participant was randomly assigned at baseline to 1 of 6 groups that varied in sequence of shopping conditions and which shopping tasks resulted in an actual food delivery. Participants were asked to shop once a week for a total of three weeks and were told that at least one and up to all three of their grocery orders would actually be purchased using their credit card. The result was only revealed to them after they hit the checkout button, which led to their weekly shop being recorded for inclusion in the study. The positive probability of having to purchase and receive the chosen products increased the chance that the purchases were an accurate reflection of participants’ actual shopping patterns. For each shop, there was a minimum spend of SGD50 and a maximum spend of SGD250. A minimum spend ensured that participants completed a typical weekly grocery order. A maximum spend was intended to make the study more manageable. Participants were informed in the consent form before enrollment, and a pop-up message appeared on-screen if they attempted to checkout a cart below or above the minimum and maximum values.
The grocery orders that needed to be fulfilled were ordered and delivered in partnership with RedMart. Following each shopping task, participants completed a brief survey to assess their mood and hunger level. ‘Mood’ took the values 1–5 where 1 was ‘very happy’ and 5 was ‘very unhappy’. ‘Hunger’ took the values 1–10, where 1 was ‘not at all hungry’ and 10 was ‘extremely hungry’. Participants who completed all study elements were rewarded with SGD75 worth of RedMart electronic vouchers.
Outcome measures
The primary outcome is the proportion of the basket represented by Lower Calorie products as this is the most direct test of the influence of the labels. Since it is possible that consumers could respond to the labels by purchasing more or a different distribution of both labeled and unlabeled products and therefore not reduce their net caloric purchases, we also included the following secondary outcomes:
Calories purchased per dollar spent (in kcal per $)
Total Spending ($)
Total calories purchased (in kcal)
Calories per serving (in kcal/serving)
Statistics
To test our hypotheses, individual-fixed effects regressions using a first difference approach were estimated. Each participant generated two observations, with each dependent variable being the difference in the outcome for each treatment condition (Within-category and Across-category) relative to the Control condition. For secondary outcomes, executing this approach was straightforward. However, in the case of the primary outcome (proportion of labeled products purchased), even though there is only one Control condition, we needed to identify two proportions, one for the products that would have been labeled in the Within-category arm and one for the products that would have been labelled in the Across-category arm. We therefore generated two proportions for each Control shop and subtracted each treatment condition from its corresponding proportion. The regression specification was then estimated as follows:
$$ \Delta Outcome\ relative\ to\ Control=\alpha +{\beta}_A Across+{\epsilon}_i $$
(1)
where subscript i = participant and ‘Across’ is a dummy variable equal to one if the order is placed under the Across-category arm and zero otherwise. The following hypotheses were then tested:
α > 0 tests whether the outcome is greater in the Within-category arm relative to Control.
α + βA > 0 tests whether the outcome is greater in the Across-category arm relative to Control.
βA > 0 tests whether the outcome is greater in the Across-category arm relative to the Within-category arm.
We then tested whether the impact of the label is moderated by mood and level of hunger at the time of purchase using the following equation:
$$ \Delta Outcome\ relative\ to\ Control=\alpha +{\beta}_A Across+{\beta}_2 Moderator\ Dummy+{\beta}_3 Moderator\ Dummy\ x\ Across+{\epsilon}_i $$
(2)
The ‘Moderator Dummy’ takes on the value = 1 for values of mood and hunger that lie above the median, and 0 for values that lie at or below the median. The median for ‘Mood’ was 2 and the median for ‘Hunger’ was 3. For ease of presentation, we defined above the median as hungry or unhappy. We then tested the following hypotheses:
β2 < 0 tests whether being hungry or unhappy at the time of purchase will diminish the impact of labeling in Within-category arm relative to Control.
β2 + β3 < 0 tests whether being hungry or unhappy at the time of purchase will diminish the impact of labeling in Across-category arm relative to Control.
We conducted the analyses on total baskets and separately for beverages only. For all regressions, standard errors were clustered at the participant level to account for correlation within individuals across shopping tasks. All regressions were run in Stata Version 15.2 (Stata Corp LP, College Station, TX).