Impact on alcohol selection and online purchasing of changing the proportion of available non-alcoholic versus alcoholic drinks: A randomised controlled trial

Study design

The study used a parallel-groups randomised controlled design. Individual participants were randomly allocated without stratification to one of three groups differing in the proportion (%) of non-alcoholic versus alcoholic drink options available for selection: Group 1: ‘25% non-alcoholic/75% alcoholic’; Group 2: ‘50% non-alcoholic/50% alcoholic’; Group 3: ‘75% non-alcoholic/25% alcoholic’.

Setting

The study was conducted online using simulated and real online supermarkets. First, participants completed a simulated supermarket selection task hosted on the Qualtrics online survey platform (see https://osf.io/2cy7t for example task images). The simulated supermarket was designed to look and function as similarly to the actual online supermarket as was possible within Qualtrics. Drinks were displayed in rows of four drink options and participants could add these to their basket, which displayed a total price after the selection had been made. Following this, participants were required to purchase drinks in Tesco online supermarket (Tesco.com), the largest national supermarket in the UK.

Participants

To be eligible for the study, participants had to be adults aged 18 years and over residing in England or Wales, who consumed beer or wine regularly (i.e., at least weekly), and purchased these drinks at least monthly from Tesco.com, with a minimum spend of £20. Participants had to be willing to complete a shop at Tesco.com following completion of the selection task, book a delivery or click-and-collect slot, and send proof of purchase (their receipt) to the research team. Similar proportions of males and females of a range of ages were recruited via Roots Research (https://rootsresearch.co.uk/), one of the largest research agencies in the UK, with a high-quality panel of over 350,000 participants. Recruitment occurred between March-July 2021.

Randomisation and masking

Randomised assignment of participants was completed via the default algorithm in Qualtrics with a ratio of 1:1:1. Participants were unaware of their group assignment throughout the study. The research team were blinded to allocation until participants had completed the primary outcome; the statistician completing the analysis was blinded to the allocation.

Intervention

All participants viewed a total range of 64 drink options. This comprised i) a range of beers, ciders, alcohol-free beer and cider alternatives, and soft drinks (32 options), and ii) a range of wines, alcohol-free wine alternatives, and soft drinks (32 options), modelled on the available range of products on Tesco.com. Initial scoping work found that Tesco.com proportions of alcoholic versus non-alcoholic options were roughly 25% non-alcoholic (360 options) and 75% alcoholic (1058 options). Alcohol-free beer, alcohol-free cider and alcohol-free wine options used in the task were matched as far as possible on brand and size characteristics with the alcohol options available online at Tesco.com. Additional alcoholic beer, cider and wine options were selected based on the leading brands of lager, ale, mild and stout, cider and wine [21–23] in Great Britain according to the number of consumers. Adult soft drinks were selected based on options that were commonly displayed next to alcoholic drinks in physical supermarkets and most likely to be consumed as a substitute for alcohol (meaning that children’s soft drinks, milk, and fruit juice were excluded). Participants viewed varying proportions of non-alcoholic and alcoholic drink options depending on their assignment: Group 1: ‘25% non-alcoholic/75% alcoholic’; Group 2: ‘50% non-alcoholic/50% alcoholic’; Group 3: ‘75% non-alcoholic/25% alcoholic’. The proportions used were based on previous food and alcohol studies [12,24]. Within each range of alcoholic drinks there were the same number of beer as wine options, and within each range of non-alcoholic drinks there were the same number of soft drinks as alcohol-free options. Participants were randomised to the order in which each subcategory (soft drinks; alcohol free; alcoholic) was presented within each of the beer and wine categories and the order of drinks within each subcategory was also randomised. Each drink option was displayed as an image, below which was a text description of the drink (identical to Tesco.com), the % alcohol by volume (ABV) for drinks containing alcohol, and its price.

Full details of the task, as well as the complete list of drinks are in the Supplementary Material S1. In the Typology of Interventions in Proximal Physical Micro‐Environments (TIPPME) [3], this is classified as an ‘Availability x Product’ intervention, while in a detailed conceptual framework specific to availability interventions [6], this is categorised as a ‘Relative Availability’ intervention.

Procedure

Participants were initially provided with an information sheet, instructions, and a link to the study via email. Participants were told the study was investigating “Adult drink preferences in England and Wales” and were not made aware of the study aim. Participant instructions outlined the stages of the study in detail, i.e., that participants were required to select the drinks for their next shop from Tesco.com in a simulated online supermarket (Stage 1), then to immediately go to Tesco.com to book a delivery slot and add these drinks to their shopping basket (Stage 2), and finally to send their receipt to the study team on their delivery or collection day (Stage 3). Once they had started the study task, participants were again presented with this information and provided consent. Participants were randomised and in a simulated online supermarket environment replicating Tesco.com (within Qualtrics) they were shown the available drink selection. They chose all the drinks they wanted to purchase in their next online shop at Tesco.com. They were then shown their total drink selection and price, and given the opportunity to amend their selection before continuing. Participants then completed demographic and drinking behaviour measures.

After completing the simulated online supermarket task, participants were automatically sent an email detailing their selection. They were prompted to open this email and given further instructions for completing purchasing, alongside a direct link to Tesco.com. Participants placed their selected drinks in their Tesco.com shopping basket, along with any other items, booked their delivery or collection slot, and confirmed this within 48 hours. They were sent a reminder email on their delivery/collection day and requested to send an itemised receipt to the research team within 48 hours. Up to two follow-up reminders were sent, two and four days later. Purchases were recorded from receipts, including any additional drink purchases. Substitutions by the participant or by Tesco that were explained (e.g., not in stock) were marked as the original drink they attempted to purchase. Participants were debriefed via email and reimbursed £25(~$35) for their time taking part in the study (but not the drinks they purchased).

Outcome measures

Statistical analysis

Analyses were pre-registered in a detailed statistical analysis plan (https://osf.io/4yuca).

All participants who completed the selection task were included in the primary outcome analysis. Participants who failed to complete the selection task and those whose responses were flagged as incomplete or suspicious - e.g., those that forged data (i.e., submitted fake receipts) or selected an unrealistically large number (with a cut-off of >100 drinks) of drinks that were not purchased – were excluded (see Figure 1 for details by group). The criteria used to exclude data were not pre-registered but were defined and applied prior to data analysis, while researchers were unaware of group allocation. The data included participants that did not select any drinks, as this was still a valid choice. The distribution of the primary outcome was highly skewed and zero inflated, and therefore a hurdle model was used for analysis, fitting i) a binary logistic model (part 1) to the zero and non-zero outcomes and ii) a truncated negative binomial model (part 2) to just the positive values [29,30]. The model results for the positive values are therefore based only on participants who selected at least one drink containing alcohol (see Supplementary S2), with non-integer variables rounded to integer values before hurdle model analysis. The marginal effect estimates, with 95% CI, are presented in Table 4.

• Download figure
• Open in new tab

Figure 1. Participant flowchart

For most secondary outcomes, hurdle models were repeated as per the primary outcome model. Model results for the binary outcomes (part 1 of the model) and the positive values (part 2, i.e., based on values above zero) are reported in Table 3. Marginal effect estimates for all secondary outcomes, with 95% CI, are presented in Table 4. The p-values for part 2 of the model and the change in marginal effect estimates (with the associated percentage reduction) are reported in the Results. For additional purchasing outcomes, negative binomial regression was required due to the skewed data. For the proportion outcomes (i.e., percentage of total drinks selected, and purchased, that were alcoholic), a beta binomial regression was used to model the proportion using the counts of relevant drinks selected out of the count of all drinks selected and this could accommodate the bimodal distribution observed. For these outcomes only, due to the nature of the model, any participants who did not select any drink (as appropriate for the outcome) were excluded.

Two per-protocol analyses were pre-specified, in which the primary outcome analysis was repeated for (i) participants who purchased what they selected, either with or without additional drinks (per-protocol analysis 1); (ii) participants who purchased exactly what they selected and purchased no additional drinks (per-protocol analysis 2).

For all outcomes, for the co-primary comparisons of primary interest (using the ‘25% non-alcoholic’ group as the reference group), a 5%/2 adjustment to the interpretation threshold for statistical significance was made. For the third comparison of tertiary interest (where ‘75% non-alcoholic’ and ‘50% non-alcoholic’ groups were compared), a simplistic 5%/3 adjustment was made rather than using methods which may not report all p-values (e.g., Benjamini-Hochberg, Holm-Bonferroni). These additional tests were calculated by refitting the same model with different reference categories.

Sample characteristics

Figure 1 shows the flow of participants. In total, 737 participants were randomised, 640 of whom completed the selection task. 607 participants were included in the primary outcome analysis. The primary analysis dataset was 59.7% female and the mean age was 37.8 years (SD = 11.4; range: 18-76). Groups were well balanced on all characteristics (Table 1).

Primary outcome

Raw primary outcome data are presented in Table 2, modelled estimates for each part of the hurdle model in Table 3 and the overall marginal effect estimates in Table 4.

In the first part of the hurdle model, a greater proportion of participants in the ‘75% non-alcoholic’ group did not select any alcohol (27/206 [13.1%]) compared to the ‘25% non-alcoholic’ group (7/207 [3.4%]; 95% CI -2.09. -0.63; p < 0.001); there was no evidence of a difference between the ‘50% non-alcoholic’ (14/194 [7.2%]) and the ‘75% non-alcoholic’ group (95% CI 0.10, 1.34; p = 0.022, given the adjusted threshold of p = 0.0167) or between the ‘50% non-alcoholic group’ and the ‘25% non-alcoholic’ group (95% CI -1.44, 0.17; p = 0.121). In the second part of a hurdle model in participants (559/607) selecting any drinks containing alcohol, the ‘75% non-alcoholic’ group selected fewer alcohol units compared to the ‘50% non-alcoholic’ (95% CI -0.44, -0.14; p < 0.001) and ‘25% non-alcoholic’ (95% CI-0.54, -0.24; p < 0.001) groups, with no evidence of a difference between the ‘50% non-alcoholic’ and ‘25% non-alcoholic’ groups (95% CI -0.24, 0.05; p = 0.178). Overall, across all participants, 17.46 units (95% CI 15.24, 19.68) were selected in the ‘75% non-alcoholic’ group; 25.51 units (95% CI 22.60, 28.43) in the ‘50% non-alcoholic’ group; and 29.40 units (95% CI 26.39, 32.42) in the ‘25% non-alcoholic’ group. This corresponds to 8.1 fewer units (32% reduction) in the ‘75% non-alcoholic’ group compared to the ‘50% non-alcoholic’ group, and 11.9 fewer alcohol units compared to the ‘25% non-alcoholic’ group (a 41% reduction); 3.9 fewer units (13% reduction) were selected in the ‘50% non-alcoholic’ group than in the ‘25% non-alcoholic’ group.

Secondary outcomes

Raw secondary outcome data are presented in Table 2, modelled estimates for each part of the hurdle model in Table 3 and the overall marginal effect estimates in Table 4. For purchasing outcomes, of the 640 participants who completed the selection task, 422 (66%) went on to purchase drinks from Tesco.com. Attrition from selection to purchasing stages was very similar across the three randomised groups (with 136, 141, and 145 completing purchasing).

Results for all secondary selection and purchasing outcomes demonstrated a wholly consistent pattern of results with amounts and proportions of alcohol selected and purchased consistently lowest in the ‘75% non-alcoholic’ group, although not always significantly so.

Per-protocol analyses

Of the 422 participants who purchased drinks, 344 participants purchased all the drinks they had selected in the selection task and 78 participants had one or more missing drinks. Of the 344 participants that purchased all the drinks they selected, 182 purchased no additional drinks.

Chi-squared tests indicated that there was no evidence against assuming equal attrition occurred. Exploratory analyses indicated that attrition was greater amongst participants with higher baseline alcohol purchasing, but regression using an interaction term suggested this did not bias the comparisons between groups, as there was no evidence of an effect at the usual threshold for interaction terms (p = 0.01) (Supplementary Material S4). See Table 5 for model results.

In participants (n = 344) who completed purchasing of the drinks they had selected, either with or without additional drinks, those assigned to the ‘75% non-alcoholic’ group selected fewer alcohol units than the ‘50% non-alcoholic’ group (95% CI -0.48, -0.14; p < 0.001, overall difference between marginal effect estimates of 6.9 drinks, 29% reduction); the ‘75% non-alcoholic’ group also selected fewer alcohol units than those in the ‘25% non-alcoholic’ group (95% CI -0.50, -0.16; p < 0.001, overall difference between marginal effect estimates of 8.4 drinks, equivalent to a 34% reduction). There was no evidence of a difference between the ‘50% non-alcoholic’ and the ‘25% non-alcoholic’ groups. For purchasing, when including additional drinks from study categories only, fewer alcohol units were purchased in the ‘75% non-alcoholic’ group compared to the ‘50% non-alcoholic’ group (95% CI -0.52, -0.16; p < 0.001, overall difference between marginal effect estimates of 8.5 drinks, 30% reduction), and in the ‘75% non-alcoholic’ group compared to the ‘25% non-alcoholic’ group (95% CI - 0.43, -0.07; p = 0.006 overall difference between marginal effect estimates of 6.6 drinks, 25% reduction). There was no evidence of a difference between the ‘25% non-alcoholic’ and ‘50% non-alcoholic’ groups, and no evidence of a difference between groups for purchasing when including all additional drinks.

In participants (n = 182) who completed purchasing of only the drinks they had selected with no additional drinks, those assigned to the ‘75% non-alcoholic’ group selected and purchased fewer alcohol units than did those in the ‘25% non-alcoholic’ group (95% CI -0.53, -0.08; p = 0.009, overall difference between marginal effect estimates of 7.5 drinks, 31% reduction). There was no evidence of a difference for the other comparisons.

Full results for the additional outcomes can be found in Supplementary Material S5.

Strengths and limitations

To our knowledge, this study is the first randomised controlled trial using a naturalistic setting to estimate the impact of increasing the proportion of non-alcoholic drinks. Meaningful selection and actual purchasing outcomes were assessed, with participants able to complete their typical online shop, including selecting and purchasing multiple options from a wide range of drinks.

The study had some limitations. First, while the primary selection outcome was assessed in the context of intention to subsequently purchase, and was minimally affected by attrition, there was substantial drop-out between selection and actual purchasing outcomes. However, attrition between groups was very similar by study condition, and there was sufficient power to detect effects despite this; as there is an absence of studies that look at purchasing of alcohol in this setting, effect sizes could not be anticipated, but large effects on purchasing were observed. While substantial attrition is expected in studies of this nature because of time between selection and purchasing, more generally it may be hard to avoid for any measure of unconstrained purchasing in a real-world online supermarket. Although we are not aware of other directly comparable studies in this context, more generally, ‘cart abandonment’ – where people do not purchase items they put in their shopping cart – is common in online (including supermarket) shopping contexts [31]. Future studies may be able address this through more intensive initial screening or follow-up of participants, or by forcing participants to immediately complete their online shop. However, such processes would arguably be less naturalistic, and including only the most motivated participants risks including a less representative sample.

Second, although the setting was as naturalistic as was feasible and actual purchasing outcomes were measured, the process involved two stages. Drinks were initially selected within a simulated online supermarket, before purchasing was completed in an actual online supermarket (albeit with the visual presentation of the former modelled on the latter). The principal purpose of including a measure of purchasing in the actual online supermarket was to validate and strengthen our primary outcome of selection, rather than to measure purchasing behaviour in a separate context. However, this meant that additional drink options were available in the real online supermarket, and participants could not be prevented from buying these if they wished to. As a result, the clearest effects on purchasing behaviour were in participants that followed the protocol as instructed and only purchased what they selected in the simulated supermarket where the intervention was implemented. To avoid this, the intervention would ideally have been implemented entirely within a real online supermarket. However, to our knowledge, this is the first study of an availability intervention to make use of such a setting (albeit in conjunction with a simulated supermarket component), although simulated retail settings, both online and physical, are commonly used in similar intervention studies [32-36]. This represents the most robust design used to date and could provide a useful method through which to assess interventions without requiring complex collaboration with commercial retailers, although further research is needed to assess its external validity. Finally, while participants were largely representative of Tesco.com shoppers [37] they were mostly of higher socioeconomic position. The generalisability of these findings to disadvantaged populations therefore needs consideration, particularly as buying alcohol-free drinks is more likely to occur in less socially and materially deprived households [18].

Implications for research and policy

This study suggests that increasing the available non-alcoholic, and reducing the available alcoholic options has the potential to meaningfully reduce selection and purchasing of alcohol. Although there was some evidence of a reduction in alcohol selected and purchased when half of the options available were non-alcoholic, effects were only consistently observed when non-alcoholic drinks became the majority. Currently, supermarkets typically stock a wider range of alcoholic than non-alcoholic alternatives to alcohol, and these results suggest that if non-alcoholic options were to become the majority instead, we might expect to see substantial reductions in alcohol purchasing. As it is yet to be seen if such major changes in ranges of drinks are feasible in real-world settings, these findings are most reasonably interpreted as proof of principle, rather than able to directly inform policy options. It is plausible that this situation could rapidly change, however. The increase in popularity of alcohol-free drinks is relatively recent, with the global market growing substantially in the last four years, and in the UK it is forecast to continue to increase [15]. This recent increase in the popularity of alcohol-free drinks has led to the emergence of drinking settings reflecting this, such as an alcohol-free off licence in London [38]. In food retail contexts there have been substantial changes seen in healthier or more sustainable ranges - such as the introduction of 50% plant-based menus [39] and the requirement to provide at least 50% healthier options in healthcare settings in Scotland [40] - suggesting that shifts of such magnitude are possible. However, before policy recommendations are made, a robust evidence base suggesting potential effectiveness is required [5], which this study provides an initial step towards. Future studies should investigate the impact of smaller and more granular alterations in proportions of non-alcoholic drinks, and in a wider range of field settings, to establish how such interventions could be used. Given the relatively large effects observed in this study, subtler interventions could elicit smaller effects that would nonetheless remain meaningful for population health, especially when considering the inherent potential for scalability across retail settings.

This intervention simultaneously increased the number of non-alcoholic drinks and decreased the number of alcoholic drinks whilst the overall number of drinks remained constant. It is unclear whether the effect is predominantly driven by one or the combination of these changes. Further studies are needed to disentangle this and investigate potential mechanisms more broadly; noting that there is some preliminary exploration of possible mechanisms in food contexts [6,41,42]. Importantly, the overall number of drinks that participants selected and purchased remained similar between groups, suggesting that effects were a result of shifting, rather than necessarily restricting, choices. This implies overall drink sales and potentially revenues may be relatively unchanged if such an intervention were to be implemented, albeit dependent on non-alcoholic drink pricing. Increasing non-alcoholic drink availability could also ultimately lead to a greater range of alcohol-free drinks and soft drinks being manufactured, further increasing their popularity in synergistic fashion [18], and many alcohol companies have already committed to this [17]. It is important to note that because many alcohol-free alternatives are marketed by the alcohol industry and there is no regulation on the often-exaggerated health claims that are made about these drinks [43]. Such industry involvement has potential harms and should be monitored closely [44–47]. In addition, although some of the non-alcoholic drink options in the current study contained no sugar and were generally lower in calories than the alcoholic options (an average of 64 calories per non-alcoholic drink versus 233 calories per alcoholic drink), many soft drinks and alcohol-free alternatives still contain large amounts of sugar and calories. Given the health risks associated with sugary drink consumption [47], continued regulation and policies to reduce sugar content and consumption from both alcoholic and non-alcoholic drinks is needed to mitigate these risks.