Study design and data source
In cooperation with the Consumer Goods Forum (CGF) Collaboration for Healthier Lives (CHL, https://www.theconsumergoodsforum.com/health-wellness/healthier-lives/), data was collected from one major UK retailer, who tested three sets of promotional activity in 2021 with the aim of encouraging healthier and more sustainable purchases. The retail partner developed the promotions and implemented them in stores. Each intervention varied in duration, extent of promotional activity and number of stores, but all three involved price promotion and two involved prominent positioning interventions. This study tested the impact of the combined promotional activity (described below), as it is not possible to isolate the impact of any one promotional strategy. Data was made available to researchers through cooperation with CGF as part of an independent evaluation of these trials.
Promoted products (determined by the retailer) for each of the three natural experiments included the following products: (experiment 1) no added sugar (NAS) plant-based milks, (experiment 2) vegan products during a vegan January event (i.e. Veganuary), and (experiment 3) seasonal fruits and vegetables. These cover a range of possible healthy and/or sustainable foods products that could be promoted, albeit by no means a complete list.
Aggregate data was provided to the independent evaluation team on sales of selected categories (units and value) from 2018 to 2021. The protocol was pre-registered at https://osf.io/9rkde/. R 4.1.3 was used for all statistical analyses.
Interventions and analyses
No added sugar (NAS) plant-based milks
Intervention
The NAS plant-based milk intervention ran for three weeks in May 2021 and involved price promotions, banners, aisle fins and social media encouraging the purchasing of a brand of NAS plant-based milk. Price reduction stickers were placed near the products, coupons were located near the checkout till, and online there were banners for the promoted products during the period of promotion. Positioning of milks remained unchanged under this promotion. The price promotion involved offering selected NAS plant-based milks, most commonly priced from £1.30 to £1.70, for £1 during the promotional period. This included eight different NAS plant-based milk products offered on promotion (with one store offering only 7). The intervention was applied in all stores, with the retailer providing data from a sample of 199 stores across the UK representing a range of customer demographics and regions for analysis, based on in-house classifications for store affluence, store size, and region.
Data
Sales data, units and value (£), of promoted plant-based milks and all plant-based milks, aggregated at the weekly level, were used as primary outcomes. To analyse the impact on all NAS plant-based milks, those milks were identified based on terms ‘no added sugar’, ‘unsweetened’, and ‘light’ in the product name. Nutritional information for those with ‘light’ in the name was checked and inclusion in analyses was limited to only those that had no added sugars. This totalled 30 unique NAS plant-based milk products, with stores offering between 19 and 29 of these products during the study period, commonly priced between £1 and £1.70. We also analysed sales of all plant-based milk products, of which there were 130, with stores offering between 57 and 125 different products over the study period, and products commonly priced between £1 and £1.70.
Demographic data available included retailer-defined classifications for the predominant age and household structure of shoppers in store (“store age”) and included middle aged adult, young adult, retired, families with pre-school children, families with school children, and elderly. For purposes of analysis, retired and elderly were combined into one group. Families with pre-school children and the families with school children were also grouped. Store affluence, was classified as price sensitive, mid-market, upmarket, and super-upmarket, interpreted as increasing SEP of the predominant shoppers. Postcode data was also provided, which allowed for analysis by Index of Multiple Deprivation (IMD), available from the UK government website providing data from 2019 [26]. For analysis, IMD deciles were grouped into the following categories: Low IMD (deciles 1–3; most deprived), Medium IMD (deciles 4–7), High IMD (deciles 8–10; least deprived).
Analysis
The primary method of analysis for all three trials was interrupted time-series (ITS). Newey-West standard errors with lag 4 were applied, determined prior to analysis following similar previous studies [11, 13]. This was to account for autocorrelation and heteroskedasticity in the data, and was set prior to analyses so that there would be consistency across all trials. ITS models were run on data up to the end of the intervention, with the interruption occurring at the start of the intervention and running into the fourth week. Multivariable hierarchical mixed-effects models were used to adjust for demographic characteristics. For the mixed-effects models, the full time period of available data up until the end of the intervention was used, but each model included a term for the intervention time period. Where sales value (in £) was the outcome, linear mixed-effects models were used, and where units sold was the outcome, negative binomial models were used with results reported as an incidence rate ratio (IRR). Where the intervention was shown to be significant, an additional mixed-effects model, either linear mixed-effects or negative binomial depending on the outcome being tested, was run including an interaction term between affluence and intervention.
Veganuary
Intervention
The Veganuary intervention ran for four weeks in January 2021. Price promotions, banners and social media were deployed in store to encourage purchasing of plant-based foods across a range of food categories. The retailer also utilised television advertisements, radio partnerships, influencers on social media, email, and online advertisements to further draw attention to the ongoing promotion. In stores, there were smart screens with information about the promotion, recipe cards, recommendations next to vegan products of how they could be used, and large signs hanging overhead to increase visibility of the promotion. Promoted items were placed at the end of aisles, otherwise product positioning was not targeted. Promotions included price reductions mostly ranging from £0.20 to £1, with a sticker highlighting the price reduction and an edge of shelf placard drawing attention to the Veganuary promotional campaign. During Veganuary, most promoted products cost between £1 and £2, with an average price of £1.73, compared to a base price range from £1.6 to £2.75, with an average price of £2.29. The intervention was applied in all retailer stores, with analysis focused a priori on 96 larger stores which implemented the most promotional activity, for example using smart screens to advertise the promotions in store and including substantial end-of-aisle activity to promote products.
Data
Primary outcome measures for analysis were weekly sales data, units and value (£), for the foods promoted in the Veganuary intervention. Promoted foods were considered to be plant-based food products and alternatives which offered a direct substitute for animal-based products. For example, plant-based ready meals, plant-based meat, non-dairy milks, etc. We did not include products such as canned vegetables that were also promoted under this campaign to boost vegetable intake. Wider categories were also analysed where compensatory behaviours may be anticipated. For example, meat alternative products were promoted as part of Veganuary, so it was hypothesised that effects could be seen in the wider meat and meat alternative categories as purchasing behaviours shifted. Meat and meat alternatives were classified based on buyer area categories provided by the retailer. Dairy and dairy alternatives were classified based on key retailer-provided categories, supplemented by identifying products and brands through keywords related to the appropriate category (See Appendix for list of identifying terms). For this analysis, data was provided from January 2018 to December 2021. However, due to data limitations in product labelling, data before April 2019 could not be used when product names were required for keyword searches to determine identification (i.e. dairy/dairy alternative and ready meal analyses). Demographic data on store age and store affluence were provided by the retailer, with the same groupings as for NAS plant-based milk, but with no postcode data.
Analysis
ITS analysis with lag 4 was used for primary analysis. Regular peaks and valleys were noted in the Veganuary data, so dummy variables for the month of January and the last week of the year were applied to the model. Following the intervention period, additional data were made available to enable ITS models to be run from April 2019 to December 2021, with the interruption set at the start of the intervention in January 2021. To adjust for demographic characteristics, multivariable hierarchical mixed-effects models were also applied. The full time period of available data up until the end of the intervention was used, but each model included a term for the intervention time period. Where sales value (in £) was the outcome, linear mixed-effects models were used, and where units sold was the outcome, negative binomial models were used with results reported as an incidence rate ratio (IRR). Where the intervention was shown to be significant, an additional mixed-effects model, either linear mixed-effects or negative binomial depending on the outcome being tested, was run including an interaction term between affluence and intervention.
Seasonal fruit
Intervention
The seasonal fruit intervention ran for approximately 13 weeks, from May to August 2021, with price promotions, tastings and messaging to encourage purchase of seasonal summer fruits. There were 18 different fruit products and one variety of lettuce offered, with all stores in the trial offering all 19 products. Price promotion stickers were prominently placed next to the products on the shelf, and hanging placards were used to draw attention to the aisle location where the promotion was taking place. The price promotions included: multibuy deals (e.g. 2 for £3) that allowed for a mix and match of products, boxes of fruit for a discounted price (usually £3 or £4), or products offered with a fixed 25% discount. During the study period, there was some change in the products included in the price promotion, however, all products were included in the prominent location at the end of the aisle and in the placards drawing attention to the products. The intervention was applied in 100 stores selected by the retailer based on capacity to implement the interventions, with an equivalent number of stores as the control group, matched using proprietary analytics based on customer demographics, store size, stock, and sales.
Data
Sales data, units and value (£), of promoted summer fruits and all fresh fruits (i.e. all fruit offered in the store throughout the study period) were used as primary outcomes, aggregated at the weekly level. Demographic data on store age and store affluence were provided by the retailer, with the same groupings as for NAS plant-based milk, but with no postcode data.
ITS analysis was used with Newey-West standard errors and lag 4 applied. Two approaches (difference-in-difference and ratio) were explored to assess how intervention stores compared to control before and during the intervention. A t-test was used to compare the units sold and revenue (in £) between control and intervention store groups. Given the discrepancy in sales between intervention and control stores (with sales in intervention stores approximately double those in control stores), the ratio method was prioritised. ITS models were run on data up to the end of the intervention, with an interruption occurring at the start of the intervention.
Analysis
Multivariable hierarchical mixed-effects models were used to adjust for demographic characteristics and analyse differential effects by demographics, with linear mixed models (for sales value in £) and negative binomial models (for sales value in units). Results from negative binomial models were reported as the incidence rate ratio (IRR) or the percentage change from this value. The mixed model compared intervention and control stores during the intervention period, with a fixed effect adjustment, which took an average of the weekly sales (in £ or units depending on the model), for the pre-intervention baseline period.