The effects of nudges on purchases, food choice, and energy intake or content of purchases in real-life food purchasing environments: a systematic review and evidence synthesis

Background Adults with a low socioeconomic position (SEP) are more likely to engage in unhealthy diets as compared to adults with high SEP. However, individual-level educational interventions aiming to improve food choices have shown limited effectiveness in adults with low SEP. Environmental-level interventions such as nudging strategies however, may be more likely to benefit low SEP groups. We aimed to review the evidence for the effectiveness of nudges as classified according to interventions in proximal physical micro-environments typology (TIPPME) to promote healthy purchases, food choice, or affecting energy intake or content of purchases, within real-life food purchasing environments. Second, we aimed to investigate the potentially moderating role of SEP. Methods We systematically searched PubMed, EMBASE, and PsycINFO until 31 January 2018. Studies were considered eligible for inclusion when they i) complied with TIPPME intervention definitions; ii) studied actual purchases, food choice, or energy intake or content of purchases, iii) and were situated in real-life food purchasing environments. Risk of bias was assessed using a quality assessment tool and evidence was synthesized using harvest plots. Results From the 9210 references identified, 75 studies were included. Studies were generally of weak to moderate quality. The most frequently studied nudges were information (56%), mixed (24%), and position nudges (13%). Harvest plots showed modest tendencies towards beneficial effects on outcomes for information and position nudges. Less evidence was available for other TIPPME nudging interventions for which the harvest plots did not show compelling patterns. Only six studies evaluated the effects of nudges across levels of SEP (e.g., educational level, food security status, job type). Although there were some indications that nudges were more effective in low SEP groups, the limited amount of evidence and different proxies of SEP used warrant caution in the interpretation of findings. Conclusions Information and position nudges may contribute to improving population dietary behaviours. Evidence investigating the moderating role of SEP was limited, although some studies reported greater effects in low SEP subgroups. We conclude that more high-quality studies obtaining detailed data on participant’s SEP are needed. Registration This systematic review is registered in the PROSPERO database (CRD42018086983).


Introduction
An unhealthy diet is one of the major risk factors for non-communicable diseases (NCDs), such as type 2 diabetes and cardiovascular disease [1]. Adults with a low socioeconomic position (SEP) in particular are at high risk for NCDs, as they are more likely to engage in unhealthy diets as compared to adults with high SEP [2]. Despite this, individual-level educational interventions that aim to improve healthy food choices have shown to have limited effectiveness in adults with low SEP and may increase health inequalities [3]. This may partly be attributed to the fact that these interventions often necessitate access to various resources (e.g., knowledge, skills, social networks) which may be more limited in low SEP groups [4,5]. Alternatively, environmental-level interventions are more likely to benefit adults with low SEP and reduce health inequalities [3], because they rely to a lesser extent on an individual's access to resources but rather create healthy opportunities for all.
The rationale underlying such environmental-level interventions is rooted in dual process models of human behaviour, which conceptualize the regulation of human behaviour into two main cognitive processes: 1) an unconscious, fast, and automatic cognitive process, and 2) a conscious, slow, and more effortful cognitive process [6]. Whereas individual-level educational interventions tap into the conscious and effortful processesby for example providing nutrition knowledge to target populationsenvironmental interventions make use of environmental cues or heuristics that subconsciously guide food-decision making [7], thus requiring limited amounts of cognitive resources.
Nudging has been proposed as a promising environmental intervention strategy for modifying food choices. The term 'nudge' was originally coined by Thaler and Sunstein in 2008 and defined as: 'Any aspect of the choice architecture that alters people's behaviour in a predictable way, without forbidding any options or significantly changing their economic incentives' (p.6) [8]. Nudging became popular as it opposed the reigning idea that humans are rational actors who constantly seek opportunities that maximize their utility. Instead, it acknowledges that people's ability to make rational decisions is limited by cognitive boundaries, biases and habits, leading people to make choices not compatible with their long-term goals [9]. Nudges make use of the same principles that cause flawed decision-making, to steer people towards choices that serve them in their own interest. When applied to modifying diets, this means that nudges make healthy choices more easy, by for example making them more salient, without constraining choice for unhealthy alternatives [9]. So far, numerous nudging studies have been performed describing a wide range of interventions, for example placing healthier foods at convenient and visible locations in supermarkets (e.g., position nudge) or making healthy foods salient through the use of signage (e.g., information nudge). To establish more conceptual clarity regarding nudging interventions and to facilitate evidence synthesis, the typology of interventions in proximal physical micro-environments (TIPPME) was introduced, distinguishing six distinct nudging interventions types: availability, position, functionality, presentation, size, and information [10].
The multiple systematic reviews and meta-analyses on the effectiveness of TIPPME nudging interventions in modifying food choices or consumption [11][12][13] mainly focused on availability and position nudges [12,13] or specific foods [11], and studies were primarily conducted in laboratory settings. Only one of these systematic review addressed the question whether the effects of nudging interventions are moderated by SEP, for which indications were found [13]. Therefore, insights are lacking on the effectiveness of other TIPPME intervention types in real-life food purchasing environments, and the moderating role of SEP.
In the present systematic review, our first aim is to review the evidence for the effectiveness of nudges as classified according to the TIPPME typology in promoting healthy purchases, food choice, or affecting energy intake or content of purchases within real-life food purchasing environments among adult populations. Second, we aimed to investigate the potentially moderating role of SEP.

Methods
The protocol for the present systematic review was registered in the PROSPERO database (registration number: CRD42018086983). A systematic literature search was conducted in accordance with the guidelines in the Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement (www.prisma-statement. org) (Additional file 1).

Data sources and searches
In order to maximize the yield of our search, we adopted an elaborate search strategy including general nudging terms (e.g., nudging and choice architecture) as well as more specific nudging terms (e.g., signage) according the TIPPME typology ( Table 1). Types of nudges considered in other categorizations were evaluated on their applicability to the current review [14,15]. As a result, the search strategy was further extended by adding the default nudge, which we defined as follows: 'to provide a standard food option for which no active choice needs to be made'.
For the search queries, search terms for the (type of) nudging intervention, outcome, and setting were combined using Boolean operators and were limited to title and abstract. The search strategies for each of the databases can be found in Additional file 2. We systematically searched the databases PubMed, EMBASE, and PsycINFO until 31 January 2018. Additionally, references included in existing reviews were included for screening [11,12,16].

Study selection
Titles, abstracts, and full-text articles retrieved from database searches were screened for eligibility in duplicate by a team of five researchers (MH, FdB, IS, JWJB, FR). Studies were included if they: 1) involved a manipulation of the food purchasing environment, in such a way that the availability, position, functionality, presentation, size, and/or information of products (e.g., foods), related objects (e.g., shelfs), or the wider environment (e.g., supermarket) was altered; 2) examined the effects on actual food purchases, energy intake or energy content of purchases, or food choice; 3) were situated in a food purchasing environment where people purchase food or meals on a regular basis; 4) were conducted among adult populations; 5) were originally published articles and were written in English language.
Studies were excluded if they: 1) did not report the effects of the nudges separately from other non-nudge interventions, such as pricing interventions; 2) studied the effects of nudges on behavioural intent; 3) were performed in settings in which people do not purchase food or meals on a regular basis (e.g., sit-down restaurants); 4) changed the intrinsic characteristics of foods (e.g., dietary composition); 5) examined the effects of mandatory legislation.
Inconsistencies in eligibility judgements were resolved by discussion among two reviewers (MH and IS) and if consensus could not be reached, inconsistencies were resolved by discussion with a third reviewer (JWJB, FR, or FdB). After this process was completed, titles, abstracts, and full-text articles retrieved from the reference lists of existing reviews were screened for eligibility by MH. A 10% subsample of the studies retrieved from the reference lists was checked by a second reviewer (IS), which revealed no inconsistencies in eligibility judgements.

Quality assessment
Risk of bias was assessed using the Quality Assessment Tool for Quantitative Studies [17], as this tool was specifically designed to critically appraise public health interventions and encompassed a wide range of research designs, including non-randomized designs. This tool evaluates the risk of bias with regard to selection of study participants, study design, confounding variables, blinding, data collection methods, and withdrawals and drop-outs. Each domain can be attributed a weak, Table 1 Overview of nudging interventions in TIPPME as defined by Hollands et al. [10] Intervention type

Availability
To add or remove (some or all) products or objects to increase, decrease, or alter their range, variety, or number

Position
To alter the position, proximity, or accessibility of products or objects Functionality To alter the functionality or design of products or objects to change how they work, or guide or constrain how people use or physically interact with them Presentation To alter visual, tactile, auditory or olfactory properties of products, objects or stimuli

Size
To alter size or shape of products or objects Information Add, remove, or change words, symbols, numbers or pictures that convey information about the product or object or its use moderate or strong quality score. Articles were considered of i) strong quality if no domains were rated as weak; ii) moderate quality if only one domain was rated as weak; 3) weak if at least two domains were rated as weak. Quality assessment was conducted in duplicate by a team of five researchers (MH, FdB, IS, JWJB, FR). Inconsistencies were resolved by discussion with a third reviewer.

Data extraction
Data extraction was performed by one researcher (MH) using a predefined data extraction form, and conducted in duplicate for a subsample of the included studies (n = 8), which showed high levels of agreement. Data was extracted on the type of nudge (including nudge description), country, study design, study size, intervention duration, SEP, setting, study outcomes, outcome assessment, and main findings.

Data synthesis
For the tabulation of study characteristics and main findings, nudges were classified using the TIPPME intervention typology (MH & FdB) into either one of the following intervention types: availability, position, functionality, presentation, size or information. On the basis of the quality assessment, study design was categorized into before-after studies (both within-and between-subjects), controlled trials, or randomized controlled trials. Intervention duration was defined as the duration for which the nudge was implemented and categorized according to the following categories: ≤ 1 week; > 1 week & ≤ 1 month; 1 < month(s) ≤ 6; 6 < months ≤12 and > 1 year. Study size could pertain to amount of purchases and/or transactions, number of customers, or number of stores. Study outcomes could pertain to purchases, energy intake or energy content of purchases or food choice. Outcome assessment was categorized as either one or a combination of the following: point-of-sale system, observer-reported, computer-generated response, digital photographic method, food weighing, hand counts, questionnaires, dietary recall, and records of inventory movement. Lastly, we report SEP characteristics for each study based on descriptive characteristics for proxies of SEP reported in the baseline table or intext (e.g., educational level, job type). Besides the tabulation of study characteristics and main findings, we visualized the main findings and study characteristics of studies within each of the TIPPME categories in harvest plots [18]. The harvest plot groups studies according to their intervention effect (positive/ negative or no effect) in a matrix, and allows to further incorporate relevant study information by varying characteristics of the matrix, including bar length, width, and colour, and by adding rows to the matrix. As such, harvest plots provide a qualitative summary to the reader by enabling them to visually appraise the most prominent patterns in the matrix, and judge study characteristics and study quality.
For the present review, the matrix comprises three columns representing the intervention effect (increase, no change, or decrease) and three rows comprising the types of outcomes (purchases, energy intake or energy content of purchases or food choice). Studies were plotted in the matrix based on the direction of the association that was reported for each outcome (e.g., if a nudge is associated with higher purchases, this study was plotted in the 'increase' column). Each study was plotted in the matrix using bars, with a study reference number below the bar corresponding to the tabulation of the study characteristics and main findings in Table 2. If studies assessed multiple outcomes, studies appear in the matrix for each outcome denoted by an additional letter (e.g., 1a, 1b). The bars were further modified to represent several relevant study characteristics. More specifically, high bars represent RCTs and controlled trials and low bars represent before-after study designs; narrow bars indicate shorter study duration and increasing width indicates longer study duration; red bars indicate unhealthy foods, blue bars indicate healthy foods, and white bars indicate calorie intake or content of purchases. Lastly, settings as retrieved from the data extraction were categorized into cafeterias (denoted by letter C) and supermarkets and small food stores (denoted by letter S).
We were not able to visuzalize nine studies in harvest plots, due to outcomes that were difficult to categorize on relative healthiness (e.g., targeted foods for which insufficient information was available to determine this); the absence of formal statistical analysis or the use of a factorial design. These studies can be found in Additional file 3.

Results
From the 9210 references identified from the database searches and reference list screening, 224 were eligible for full-text review, and 68 references were included in the narrative synthesis of findings. The 68 references comprised 75 studies (Fig. 1).

Descriptive characteristics of included studies
Of the 75 retrieved studies, 42 studies were categorized as studying information nudges, ten studies were categorized as studying position nudges, 18 studies were categorized as studying mixed nudging interventions, two studies were categorized as studying size nudges, two studies were categorized as studying a functionality nudge, and one study was categorized as studying a presentation nudge. No studies were categorized as studying          Asterix indicates statistical significance (p < 0.05); ∞ Main findings are aggregated across two consecutive intervention phases for which data on statistical significance was not available default or availability nudges. Given the vast amount of information nudges identified, we further categorized these groups of interventions into the following categories: information nudges using symbols (n = 15); information nudges providing nutrition information (n = 13); and information nudges using signage (n = 14). Studies most often employed a before-after design (56%), followed by a controlled trial design (32%) and randomized controlled trial design (12%). Only 19% of studies had an intervention duration longer than 6 months, and studies were most often situated in cafeterias (55%), followed by supermarkets (25%) and small food stores (16%).

Effects of nudging by TIPPME category Information nudges using symbols
The harvest plot for information nudges using symbols is shown in Fig. 2 and study characteristics and main findings are presented in Table 2. Eight studies received a moderate quality rating, four received a weak quality rating, and three received a strong quality rating. Studies examining information nudges via symbols generally highlighted healthy or unhealthy foods using symbols such as star-ratings and promotional logos. The effects of information nudges using symbols were most often studied in association to purchasing outcomes. Overall, in mainly cafeteria settings, identifying healthy food items through the use of symbols generally did not affect purchases of those  9]. Concluding, the effects of highlighting healthy and unhealthy foods through the use of symbols in supermarket, small food store, and cafeteria settings were heterogeneous but showed a modest tendency towards no effects on studied outcomes.

Information nudges providing nutrition information
The harvest plot of information nudges providing nutrition information is shown in Fig. 3 and study characteristics and main findings are presented in Table 2. Three studies could not be visualized in the harvest plots and are presented in Additional file 3. Seven studies received a moderate quality rating, five studies received a weak quality rating, and one study received a strong quality rating. Studies examining information nudges providing nutrition information usually did so by providing nutritional labels at the point-of-choice. The effects of nutrition information nudges were most often studied in relation to purchases as the outcome as well as energy intake or energy content of purchases. Some studies provide evidence that the provision of nutrition information in food purchasing environments increases purchases of or choice for healthy items [1a, 7a, 7b, 8a, 10a], decreases purchases of unhealthy items [1b, 10b], and similarly, decreases energy intake or energy content of purchases [1c, 2a, 2b, 3, 4], although one study observed increased energy intake [5]. Contrary, other studies found no effects on purchases of healthy or unhealthy items [7c, 9a, 9b], or on energy intake or content of purchases [6] or food choice [8b]. Concluding, the effects of providing nutrition information in supermarket, small food store and cafeteria settings were heterogeneous but showed a modest tendency towards beneficial effects on studied outcomes.

Information nudges using signage
The harvest plot of information nudges using signage is shown in Fig. 4 and study characteristics and main findings are presented in Table 2. Two studies could not be visualized in the harvest plots and are presented in Additional file 3. Eight studies received a moderate quality rating, three studies received a weak quality rating, and three studies received a strong quality rating. Studies examining information nudges using signage generally displayed posters with health prompts, social norms, or health primes. The effects of signage nudges were generally evaluated on purchasing outcomes and studies were primarily conducted within cafeteria settings. Signage was associated with increased purchases of healthy items in several studies [2b, 2c, 3, 5a, 6, 7a, 7b, 7c, 9a, 10, 11], increased choice for healthy food [4] and with decreased purchases of unhealthy items [1a]. Contrary, also no change in purchases of healthy or unhealthy [1b, 2a, 2d, 5b, 8a, 8b, 9b, 12] items were observed. Concluding, effects for information nudges using signage in supermarket, small food store, and cafeteria settings were heterogeneous but showed a modest tendency towards beneficial effects on studied outcomes.

Position nudges
The harvest plot for position nudges is shown in Fig. 5 and study characteristics and main findings are presented in Table 2. Eight studies received a moderate quality rating and two received a weak quality rating. Studies examining position nudges generally manipulated proximity to healthy and unhealthy foods (e.g., decreasing proximity to healthy foods and increasing proximity to unhealthy foods). The effects of position nudges were most often studied in relation to purchasing outcomes. Overall, it can be concluded that in small food stores and cafeterias, increasing or decreasing the accessibility or visibility of healthy and unhealthy foods, respectively, showed increased purchases of healthy foods and decreased choice for unhealthy foods [1a, 2a, 3,5,6,9]. However, other studies conducted in larger purchasing contexts such as supermarkets showed no effects on healthy food purchases [8, 10a]. Moreover, purchases of relocated unhealthy snacks (e.g., snacks that were relocated to more distant locations as a consequence of making healthy foods more accessible) [1b, 10b], energy intake [2b], or food choice [4] were not affected in both small and larger purchasing contexts. Lastly, one study showed counterintuitive findings, with increased and decreased purchases of unhealthy and healthy items, respectively, when healthy items had been made more accessible [7a, 7b]. Concluding, the effects of altering the proximity of healthy and unhealthy foods showed a modest tendency towards beneficial effects on outcomes in primarily smaller food purchasing environments, but not in larger food purchasing environments.

Mixed nudging interventions
Several studies were identified that studied a combination of TIPPME intervention categories, which we phrased 'mixed nudging interventions'. The harvest plot for mixed nudging intervention is shown in Fig. 6 and study characteristics and main findings are presented in Table 2. Four studies could not be visualized in the harvest plots and are presented in Additional file 3. Eight studies received a moderate quality rating, eight studies received a weak quality rating, and two studies received a strong quality rating. The effects of mixed intervention nudges were most often studied in relation to purchasing outcomes in cafeteria or supermarket settings. Moreover, studies were often characterized by high quality study designs (e.g., RCTs and controlled trials). As for the effects of mixed nudging interventions on the outcomes studied, mixed nudging interventions generally did not affect purchases of healthy items [1a, 2b, 2d, 3c, 3d, 3e, 4a, 5b, 5e, 9, 11, 13a, 14] or unhealthy items [1b, 2a, 2c, 4b, 13b], or energy intake or  6]. Contrary, some studies observed increased purchases of healthier items [3a, 3b, 5a, 5c, 5f, 7a, 10a, 12a], decreased purchases of unhealthy items [7b, 10b, 12b], and decreased calorie content of purchases [8]. Also some counterintuitive findings were observed, with mixed nudging interventions being associated with increased purchases of unhealthy items [2e] and decreased purchases of healthy items [2f, 5d]. Concluding, the effects mixed nudging interventions in supermarket, small food store, and cafeteria settings were heterogeneous but showed a modest tendency towards no changes in studied outcomes.

Availability, size, functionality, and presentation nudges
Two studies were categorized as size nudges [74,75]. In these studies, increasing the portion size of an entrée [74] and decreasing the portion size of sausages [75], was associated with increased energy intake and decreased meat purchases, respectively. Two studies described the effects of a functionality nudge [76]. In these studies, arrows on supermarket floors indicating the location of fresh fruits and vegetables were associated with increased fruit and vegetable purchasing. One study was categorized as a presentation nudge, during which participants were provided with a healthy or unhealthy sample and subsequent purchases in a supermarket were monitored [77]. The study showed that the consumption of a healthy sample was associated with increased subsequent healthy purchases.

Evidence for differential effects across SEP
Six studies evaluated the effects of nudges across levels of SEP, for which several indicators were used including educational level, food security, job type, and income. In subgroup analyses, there were modest indications that nudgesincluding signage, mixed nudging interventions, and position nudgeswere significantly more effective among people with a lower educational level [44], in people with food insecurity [63], or in people on a food assistance program [59], respectively. Similarly, in two other mixed nudging intervention studies which used traffic-light labelling and accessibility changes, the extent to which red and green-labelled purchases were affected by the intervention differed in magnitude across job type in subgroup analyses [71] and the effect of the intervention on red-labelled purchases was significantly modified by job type, but not for overall purchases [67]. However, no evident pattern in purchasing differences across job types could be discerned, as job types could not be clearly classified by SEP. In another study which examined the effect of an information nudge providing nutrition information on calorie intake, no significant effect modification by income or educational level was observed [37].

Main findings
In the present review, we aimed to assess the evidence for the effectiveness of nudges as classified according to TIPPME in promoting healthy purchases, food choice, or affecting energy content of purchases or intake within real-life food purchasing environments. Additionally, we aimed to investigate whether the effects of nudges are moderated by SEP. We observed that the evidence to date predominately focused on the effectiveness of information nudges (56%) and position nudges (13%), while less evidence is available on the effectiveness of other types of TIPPME nudging interventions. We also observed that studies often investigated short-term outcomes, with 81% of studies having an intervention duration shorter than 6 months. Also, the studies often relied on non-randomized designs and were most often conducted in cafeteria or supermarket settings.
The harvest plots showed that for information and position nudges modest tendencies towards beneficial effects on studied outcomes were present. Finally, we found indications that the effects of nudges may be moderated by SEP, showing larger effects among low SEP individuals. However, evidence was limited in quantity and the use of different measures of SEP hampered comparison of the evidence. Overall, studies were generally considered of moderate or weak quality, raising concerns about potential bias and warranting caution in the interpretation of the results.
Findings from the present review are in line with previous literature. Similar to the present study, a scoping review conducted by Hollands et al. concluded that most studies focused on information nudges [78]. The effectiveness of information nudges is however debated, as they deviate from the original definition of nudging, by relying partly on cognitive processing. One previous meta-analysis of field studies by Cadario and Chandon explored the effectiveness of nudges, using their own categorization of cognitive nudges, affective nudges and behavioural nudges. They concluded that cognitive nudges were least effective in affecting selection and consumption outcomes [79], observing a small effect size of d = 0.12, supporting the argument that information nudges are 'sub-optimal'. In the present review, we observed that information nudgeslargely overlapping with the definition of cognitive nudges by Cadario and Chandonpositively affected outcomes, but we could not compare the magnitude of effects to other TIPPME nudges given the inability to meta-analyse findings. Further evidence that information nudges work, even though considered 'sub-optimal' in terms of how they operate on a psychological level, comes from two recent systematic reviews and meta-analyses of nutritional package and/or point-of-purchase labelling in primarily supermarkets, cafeterias, and restaurants, showing statistically significant average decreases of 6.6 and 7.8% in energy intake, respectively [80,81], although for the latter review the quality of evidence was rated as low.
We also observed a tendency towards healthier purchasing in smaller food purchasing contexts for position nudges. Although evidence is tentative and qualitative in nature, this finding is in line with multiple systematic reviews that examined the effects of position nudges on consumption and selection; choice, sales or servings; or on sales and consumption in primarily laboratory settings [13], school settings [11], and a range of microenvironments including cafeteria and laboratory settings [12], respectively. However, all reviews highlight that effects are generally small in magnitude and that the quality of evidence is considered to be low.
Finally, we observed that the effects of nudges may differ by SEP, with limited studies observing somewhat stronger effects in low SEP populations. Only one other systematic review and meta-analysis that examined the effectiveness of availability and proximity nudges systematically assessed whether the effects of these interventions were potentially modified by SEP, and found that effect sizes for position nudges were larger among studies conducted among populations with low deprivation status, as compared to studies conducted among populations from both high and low deprivation status [13]. For availability nudges, insufficient data was available to assess whether intervention effects were modified by SEP. An important reason for why evidence is limited in the present review, may be due to the fact that it is challenging to obtain detailed information on SEP in studies conducted in real-life food purchasing environments, as there is often less active engagement with the research population. For example, studies often monitor purchases following a nudging intervention, without consent or active participation of customers.

Strengths and limitations
Some limitations of the present review need to be addressed. First, given the substantial heterogeneity in Fig. 6 Harvest plot for mixed nudging interventions study characteristics and incomplete study reporting, it was not possible to quantify the effects of the TIPPME intervention types using conventional meta-analyses techniques. An important reason for the heterogeneous study characteristics and study findings may relate to the focus on real-life purchasing contexts which are naturally less controlled environments as compared to laboratory settings. Additionally it may be due to our studied outcomes which were heterogeneous in terms of the types of foods that were targeted with nudging strategies. However, the use of harvest plots offers a visually appealing way to summarize the study information and study findings. This approach is preferable over a narrative analysis of study findings, as information is more easily digested by the reader and also less prone to bias, as studies are plotted in a systematic way [82]. Second, very few studies assessed dietary intake as outcome of nudging interventions. Alternatively, energy content of purchases was often calculated as a proxy of energy intake. Therefore, the majority of evidence is based on the evaluation of food purchases. As nudging is often suggested as a potentially important strategy in battling the obesity epidemic, it is crucial to evaluate its effects on more proximal health parameters, such as dietary intake, as well. Third, we adopted a broad search strategy, including general nudging terms (e.g., nudging and choice architecture) as well as more specific nudging terms (e.g., signage) according the TIPPME typology. As a result of this search strategy, studies were included that did not clearly indicate to test a nudge, but did comply with nudging definitions laid out by the TIPPME typology. As these studies provided little theoretical background of the intervention under study, there was often limited information available to categorize the study according to TIPPME. For example, studies we categorized as information nudges based on the TIPPME definition, may partly rely on cognitive processing, and therefore may not satisfy the criteria for nudging. Finally, the majority of studies received a moderate to weak quality rating. Major quality issues related to the study design, which was often not randomized, which consequently raised concerns about potential for confounding. Concerns about the quality of nudging studies have also been highlighted in previous reviews [11][12][13].
Strengths of the current review include that it used an extensive search strategy, not only using 'nudging' and 'choice architecture' as search terms, but adding specific nudging intervention types as search terms as well. Indeed, a previous systematic review investigating the effectiveness of nudging strategies only included studies if they were specified as such by the original authors, resulting in only thirteen eligible publications [83]. Additionally, the present review builds upon the TIPPME typology which was the result of an extensive scoping review, and therefore provides a useful conceptual framework for structuring the evidence base. However, we acknowledge that categorizations remain broad and may be susceptible to different interpretations, and further enhancement of conceptual clarity is needed.

Implications for improved methods
Given the limitations of the evidence base addressed in this review, we provide several suggestions for improved methods. First, given the level of heterogeneity in study characteristics there is an urgent need for harmonization of methods in nudging studies to facilitate evidence accumulation. It is therefore important to establish common measures to asses SEP, such as composite measures combining both income, education, and job status [84]. Additionally, adherence to reporting standards such as Journal Article Reporting Standards (JARS) as laid out by the American Psychological Association would improve study reporting and therefore enhance evidence synthesis. Moreover, the field of psychological and behavioural science has been scrutinized for its inability to replicate some of its findings [85]. For example, a recent pre-registered study found no association between plate size and food consumption, which contrasted with earlier findings [86]. Therefore, efforts such as preregistration of study protocols which allow replication are warranted to further advance the field of (nutrition) nudging [87].

Implications for future research and practice
From the present evidence, we highlight the following knowledge gaps present in nudging literature. First, future studies should focus on studying the effectiveness of non-information nudges (e.g., availability, position, functionality, or sizing nudges) in real-world settings. Second, given the limited available data on potential moderators of nudging effectiveness in real-world settings, the use of loyalty cards containing customer's personal information would be a valuable contribution to the existing literature, allowing to examine the role of potential moderators such as age, sex, and SEP. Third, nudging studies often only targeted limited food categories, which does not justify complex food environments in which multiple other food choices are made. Moreover, it is difficult to make inferences about what changes in purchases of a selected number of foods actually constitutes in terms of an individual's health. Therefore, future nudging studies that use loyalty cards, could nudge a wider array of food products and estimate changes in overall dietary quality on an individual level. Fourth, as the included literature in the present study mainly studied short-term effects, future studies should consider including a longer follow-up, as this long-term effectiveness is crucial to assess potential public health impact. Lastly, the present review highlights the viability of conducting nudging interventions in real-life purchasing contexts. Consequently, local policy makers or owners of local food stores could be encouraged to implement nudging interventions at local level. From a policy perspective, it is also of importance to consider the ethical aspects of nudging, which have been outlined previously [88].

Conclusion
This systematic review was the first to examine the effectiveness of nudging interventions on purchases, energy intake or content of purchases, and food choice in real-life food purchasing environments, using an elaborate search strategy drawing upon the TIPPME framework. We showed that evidence mainly focuses information and position nudges, while less evidence is available on the effectiveness of other TIPPME intervention types. We qualitatively demonstrated that information and position nudges might be effective in improving outcomes, especially purchasing outcomes, and that SEP may be a moderator for the effectiveness of nudges. However, evidence is limited and difficult to compare. More high-quality studies focusing on non-information nudges and examining long-term effectiveness in real-life food purchasing environments and obtaining detailed data on participant's SEP are needed.
Additional file 2. Search strategy for bibliographic databases.
Additional file 3. Studies not appropriate for visualisation in harvest plots.