Supermarkets represent an important environment in which to intervene to improve population-level dietary behaviors. RCTs provide high quality data in support of causal inference, however even RCTs have limitations, particularly in the context of complex, real-world, population-level nutrition interventions [15, 16]. In the SHELf RCT, individuals in the price reduction group were approximately twice as likely to report having increased proportionate purchasing of FV in study supermarkets from baseline to post-intervention, relative to all other groups. Qualitative comments from a prior process evaluation substantiated these findings, and suggested that these differential changes in FV purchasing were related to the financial incentive provided by the 20% price reductions [12]. SHELf is the first study, to our knowledge, to formally assess the patterning of proportionate food purchasing within the context of a supermarket-based RCT, albeit on the basis of self-reported data.
While RCTs are undoubtedly the gold standard of research design, collecting data pre- and post-intervention, without measuring what happens in between, risks misstating the true impact of interventions on outcomes [15]. The current analysis was motivated by responses provided during a prior process evaluation of the SHELf RCT [12], and supported those earlier findings. When food purchasing data are not available from all sources, differential changes in proportionate food purchasing between randomized groups, such as those documented here, can make it difficult to discern the true impacts of nutrition interventions. This is because it becomes unclear to what extent measured changes in food purchasing are real, or merely represent shifts in food purchasing from other stores into study-associated venues, while overall purchasing remains constant. Notably, the Healthy Incentives Pilot rebate program also found evidence to suggest participants randomized to receive rebates may have partially shifted their FV purchases into participating retailers to maximize rebate earnings [17].
The current findings appear to be the result of two factors operating in tandem. First, the intervention and comparison conditions were somewhat asymmetric in terms of their perceived benefits to participants. Participants who received price reductions had a strong financial incentive to displace their FV purchasing from other stores into study supermarkets, whereas those in the control and skill-building groups did not [12]. Second, purchasing data were only collected within Coles supermarkets, and proportionate FV purchasing was low at Coles at all three time points (i.e. less than half of participants reported purchasing ≥50% of their FV at Coles, with only one-quarter purchasing ≥75% of their FV at Coles at all three time points). It is likely to be the co-occurrence of these circumstances that proved problematic. For instance, had complete purchasing data for participants been available for all venues in which they shopped, any asymmetries in the nature of the intervention and comparison conditions would have been relatively immaterial. Conversely, had the price reduction groups not had a strong incentive to displace their food purchasing, our inability to measure purchasing in all venues should not have posed a substantial limitation because randomization should have ensured similar proportions of participants changed their food purchasing behaviors over time.
It is not clear why a similar displacement of proportionate FV purchasing into study supermarkets was not evident in the group that received a combination of price reductions and skill-building. This result is, however, consistent with the different food purchasing and consumption behaviors exhibited by the price reduction and combined groups throughout the SHELf study [7]. This finding additionally suggests that while differential displacement of food purchasing may occur in some supermarket-based interventions, it is not necessarily inevitable, and should therefore be quantified and not assumed. It is also unclear why the change in self-reported proportionate FV purchasing in the price reduction group was not reflected in a similar shift in this group’s total grocery purchasing, however we expect this may be because FV purchases comprised a relatively small proportion of participants’ total food purchasing (i.e. participants purchased on average just 2.6 kg/week of FV) [7]. Finally, although 35% of those in the price reduction group reduced FV purchasing at Coles from T2 to T3 compared to an average of 22-25% in other groups, this decline was not statistically significant.
Study limitations and strengths
The current findings derive from self-reported data, however they are consistent with economic theories of human behavior, and are supported by regression analyses demonstrating directional correspondence between change in reported proportional, and objectively quantified FV purchasing (data not shown), as well as by qualitative comments provided during a prior process evaluation [12]. Moreover, identical questions were asked at all three time points, there was little missing data, and there is no reason to expect differential reporting of proportionate food purchasing over time.
Implications
The challenges associated with quantifying intervention-associated change in food purchasing described herein may be difficult to address given that modern environments offer ubiquitous access to food, and that incentives to alter food purchasing behaviors are inherent within the design of some supermarket-based interventions. The degree to which these and other potential biases arise will likely vary across studies according to factors such as the specific nature of each intervention, study location, and the participant population. Given that economic incentives are increasingly being deployed in an attempt to improve lifestyle behaviors, the current findings may also have broader relevance to other types of studies.
It is not possible to quantify to what extent shifts in food purchasing affected outcomes in the SHELf study. However, correspondence between purchasing and intake measures within the price reduction group provides reassurance that study-associated increases in fruit purchasing were not solely the result of displacement of proportionate fruit purchasing into study supermarkets. Moreover, that measured fruit purchasing increased in the combined group, a group in which significant displacement of proportionate FV purchasing was not reported is also instructive, as are qualitative comments from participants indicating that price discounts led them to purchase more FV overall. Authors of similar supermarket-based price interventions have employed sensitivity analyses [8], statistical adjustment [10], triangulation of measures [10, 11], and participant pledges to shop exclusively within study supermarkets [11] to avoid potential biases associated with displacement of food purchasing. The effectiveness of such strategies remains uncertain, however.
Study findings suggest that measurement of total food purchasing may be essential. These data could be obtained by asking participants to scan the bar codes of packaged foods brought into the home (e.g. analogous to the Neilsen Homescan Panel [18]) and to photograph unpackaged items, or by supplementing electronic purchasing data from study supermarkets with receipts from foods purchased elsewhere. However, just as individuals underreport food intake, households may also underreport their supermarket shopping [19, 20]. These additional measures might also deter study participation due to increased participant burden, although in one study participants deemed collection of food receipts a useful tool in helping them to purchase healthier foods [21]. Development of new, technology-based integrated data collection methods may simultaneously improve the validity of the data collected, while reducing participant burden.
