Estimating the effect of moving meat-free products to the meat aisle on sales of meat and meat-free products: A non-randomised controlled intervention study in a large UK supermarket chain

Background Reducing meat consumption could bring health and environmental benefits, but there is little research to date on effective interventions to achieve this. A non-randomised controlled intervention study was used to evaluate whether prominent positioning of meat-free products in the meat aisle was associated with a change in weekly mean sales of meat and meat-free products. Methods and findings Weekly sales data were obtained from 108 stores: 20 intervention stores that moved a selection of 26 meat-free products into a newly created meat-free bay within the meat aisle and 88 matched control stores. The primary outcome analysis used a hierarchical negative binomial model to compare changes in weekly sales (units) of meat products sold in intervention versus control stores during the main intervention period (Phase I: February 2019 to April 2019). Interrupted time series analysis was also used to evaluate the effects of the Phase I intervention. Moreover, 8 of the 20 stores enhanced the intervention from August 2019 onwards (Phase II intervention) by adding a second bay of meat-free products into the meat aisle, which was evaluated following the same analytical methods. During the Phase I intervention, sales of meat products (units/store/week) decreased in intervention (approximately −6%) and control stores (−5%) without significant differences (incidence rate ratio [IRR] 1.01 [95% CI 0.95–1.07]. Sales of meat-free products increased significantly more in the intervention (+31%) compared to the control stores (+6%; IRR 1.43 [95% CI 1.30–1.57]), mostly due to increased sales of meat-free burgers, mince, and sausages. Consistent results were observed in interrupted time series analyses where the effect of the Phase II intervention was significant in intervention versus control stores. Conclusions Prominent positioning of meat-free products into the meat aisle in a supermarket was not effective in reducing sales of meat products, but successfully increased sales of meat-free alternatives in the longer term. A preregistered protocol (https://osf.io/qmz3a/) was completed and fully available before data analysis.


Study design
A non-randomised controlled before and after study comparing sales/purchasing data in active trial stores delivering the intervention and a matched sample (1:5) of control stores.
The intervention has been carried out in 2 phases: -Phase I -Positioning of meat-free products in meat aisle The intervention consists of repositioning meat-free products in the meat aisle from the week commencing January 27 th to April 21 st 2019 (12 weeks). A total of 20 trial stores implemented this intervention over the specified period, while 88 matching control stores carried out business as usual.
Although the supermarket pre-specified that the intervention was going to finish in the w/c April 28 th 2019, the trial stores maintained the meat-free aisle until the end of the study period (w/c December 1 st 2019). For the main analyses, we will consider the intervention period to run for 12 weeks from w/c February 3 rd to w/c April 21 st 2019.
-Phase II -Addition of extra bay for meat-free products in meat aisle The phase II intervention consists of the addition of another bay for meat-free products in the meat aisle from the week commencing July 28 th 2019 with no pre-defined end date for the trial. A total of 8 stores from the above trial 20 stores were selected to incorporate the extra unit. The remaining 12 trial stores continued with the phase I intervention, while the 88 matching control stores continued carrying out business as usual. The 8 stores were selected by the retailer based on the number of meat-free products in stock. Of the 20 original intervention stores, 8 had sufficient meat-free products to fill a second bay. Other stores did not have enough for a complete bay and mixing meat and meat-free products in the same bay was seen as unacceptable.
Matching was carried out by the commercial partner and full details of the matching algorithm are not available to the research team. Although matching of 5 control stores to 1 intervention store was sought, in practice some matching groups have fewer than 5 controls. Some stores will be allocated to a single matching group by selecting the matching group with fewest control stores before the data analysis.

Store characteristics
We have the following store characteristics which related to the surrounding population which may be included as covariates in the analysis:  Affluence: coded as More; Average; Less  Age group: coded as binary (Older vs Younger)  Ethnicity: coded as White; Asian; Other  Area: coded as More Urban; Urban; Less Urban

Outcome measures
Primary outcome Two measures of sales are available -number of units sold and sales revenue (£) for the product categories of interest. Our primary outcome will be on units sold: -Average total weekly sales (Units) of mince, burgers, meatballs and sausages (meat) Average weekly sales (units and £) will be calculated as the average of weekly sales for the relevant study period i.e. either the comparator or intervention period All outcome measures are based on sales of the 26 stock keeping units (SKUs) that have been moved into the MFP aisle and on the meat product equivalents.

Secondary outcomes
Of major importance is re-assessment of the primary outcome using an alternative measure (£) as well as changes in sales of meat-free alternatives: -Average total weekly sales (£) of mince, burgers, meatballs and sausages (meat) -Average total weekly sales (Units) of meat free alternative mince, burgers, meatballs and sausages -Average total weekly sales (£) of meat free alternative mince, burgers, meatballs and sausages Additionally, when an effect is identified on one of the above four outcomes, we will further stratify by type of product: -Average weekly sales (Units and £) per store of mince (meat) -Average weekly sales (Units and £) per store of burgers (meat) -Average weekly sales (Units and £) per store of meatballs (meat) -Average weekly sales (Units and £) per store of sausages (meat) -Average weekly sales (Units and £) per store of meat-free alternative mince -Average weekly sales (Units and £) per store of meat-free alternative burgers -Average weekly sales (Units and £) per store of meat-free alternative meatballs -Average weekly sales (Units and £) per store of meat-free alternative sausages -Average weekly sales (Units and £) per store of other meat-free items (e.g. tofu, falafel) Control outcomes -Average weekly sales (Units and £) per store of fish -Average weekly sales (Units and £) per store of non-dairy milk -Average weekly sales (Units and £) per store of vegetables -Average weekly sales (Units and £) per store of fruit -Average weekly sales (Units and £) per store of personal-care products

Sample size considerations
Original: This is a pilot study. No statistical power calculations have been conducted and we consider the findings to be of a exploratory nature. Data from this pilot study will be used to inform the design of future intervention evaluation studies.
Revised: The study was initially conceived as a pilot study with the view of estimating the effect sizes for a future trial. However the retailer did not continue their plans to conduct a larger trial. Therefore following the conduct and evaluation of a natural experiment, a power analysis was not conducted and the final number of trial stores with matched control stores was chosen by the retailer.

Descriptive analysis of pre-intervention data
We will tabulate store characteristics by intervention vs control group, and test for differences with t-test (continuous variables) or chi-squared test (categorical variables);  Affluence;  Age group;  Ethnicity;  Area;  "Veganuary" effect (e.g. higher sales of meat-free products in January 2018 and 2019, see below);  Average sales of <outcome> per week in 12 weeks before baseline.
Normality assumptions for continuous variables will be checked graphically (quantile plots) to determine whether alternative methods (log transformation or if necessary non-parametric methods) are required.
We will compute store-level measures designed to capture propensity to reduce meat purchase and increase purchases of meat-free alternatives: in particular, change in sales of meat and meat-free alternatives in 'Veganuary' relative to sales in other periods.
The data required to conduct these preliminary analyses are summarised in annex A.

Analysis strategy
The purpose of the analysis is to describe changes in sales volume of meat products and meat-free alternatives and to test for differences in sales volume between the intervention and control stores, adjusted for differences in sales volume during the pre-intervention comparator period. The analysis will also adjust for store characteristics found to be potential confounders in the preliminary analysis but we will also explore doing subgroup analyses by store characteristics.
Prior to analysing the primary and secondary outcome data, we will run the analysis for the control outcome measures to check that sales of these two products do not differ between intervention and control stores, after adjusting for pre-intervention sales and store characteristics.
The main analysis will be conducted using all intervention and control stores. Outliers will be investigated and excluded if necessary in a sensitivity analysis. If the intervention and control stores are found not to be comparable with regard to time trends, we may additionally perform a subgroup analysis based on stores considered to be comparable.
All statistical tests will be conducted at a 5% significance level, but interpretation of findings for secondary outcomes will be based on consistency of findings in recognition of the risk of false positives associated with multiple testing.
Statistical methods for primary outcome Analyses will exclude the first week (to allow time for the intervention to be implemented). The primary analysis will compare weekly units sold in intervention to control stores during the phase I intervention period (w/c 3rd February 2019 to w/c 21 April 2019) using a hierarchical negative binomial model, with fixed effect adjustment for store affluence, store age group, store ethnicity, store area and average units sold per week in 12-week pre-intervention period (w/c 2 September 2018 to w/c 18 November 2018) and a random effects term for matching group.

Statistical methods for other outcomes
Analyses of other outcomes will proceed as above except that the model will be hierarchical Normal for sales (£) outcomes.

Secondary analyses
For both phase I and II trials, the impact of the interventions will be also evaluated using aggregated store level sales data and interrupted time series analysis will be conducted to evaluate changes in weekly sales associated with the interventions using data from w/c 1 st August 2017 to w/c 1 st December 2019. We will compute a graphical representation of the outcomes with p-values. For evaluation of the Phase I intervention, we will plot (a) the average sales of meat and meat-free products across all 20 intervention stores and (b) the average sales of meat and meat-free products across all 88 control stores, together with fitted linear trend lines before and after the Phase I intervention began in the w/c 3rd February 2019. To assess whether differences visible in this graph are statistically significant between intervention and control stores, we will use a difference-indifference approach, calculating the mean difference in sales (units or £) at each week between intervention and control group, and testing whether this time series of differences changed after intervention time using a linear regression model. We will use a Chow-type test for any difference (in either intercept or slope, or both) after vs. before, and Newey-West standard errors with lag 2 to allow for autocorrelation in the time series. We will use the same approach to evaluate the Phase II intervention but only using data from the 20 intervention stores (12 Phase I and eight Phase II stores) from the 3rd of February 2019 to 1st December 2019; with fitted linear trend lines before and after the Phase II intervention began in the week commencing 28th July 2019.

Sensitivity analysis
If time trends appear to differ between the intervention and control stores, we may define a subgroup of stores which we consider to have comparable sales time trends, and we will perform a re-analysis of the main outcomes.
Sensitivity analyses of the ITS models will test the effect of controlling for seasonality.