Early food intake modelling studies

The first laboratory-based modelling studies tested whether pairing a participant with an actor who was instructed to eat a lot v. very little food (a ‘confederate’) influenced how much a ‘real’ participant ate. The confederate appeared to be another participant and the food tended to be a high-energy snack food. In these studies, when the confederate ate a lot the participant also ate a lot and when the confederate ate very little then so did the participant⁽Reference Conger, Conger and Costanzo^9–Reference Goldman, Herman and Polivy¹²⁾. Some studies also included a control condition in which the participant ate alone, which enabled direct examination of whether the amount eaten by the confederate increases or decreases the amount of food eaten by the participants relative to what would be eaten normally. From this early literature both propositions are generally supported (see Conger et al. ⁽Reference Conger, Conger and Costanzo⁹⁾, Goldman et al. ⁽Reference Goldman, Herman and Polivy¹²⁾ or Rosenthal & Marx⁽Reference Rosenthal and Marx¹³⁾), although the results are not so clear in some studies⁽Reference Rosenthal and McSweeney¹⁴⁾.

A common feature of all these early studies was the use of a taste-test paradigm. The explanation given to participants for eating in the presence of someone else was that they were required to rate some foods together as part of a research study. The participant and confederate were left facing each other in a room with separate bowls of food and instructed to eat as much as they liked during the task. See Goldman et al. ⁽Reference Goldman, Herman and Polivy¹²⁾ or Rosenthal & Marx⁽Reference Rosenthal and Marx¹³⁾ for detailed descriptions of this method. A methodological weakness is that these settings could be argued to be somewhat artificial. It is possible that because the participants were in a strange setting and unsure of how to complete the taste test, they may have been encouraged to follow the eating behaviour of the confederate.

Recent food intake modelling studies

Two more recent studies⁽Reference Johnston^15, Reference Brunner¹⁶⁾ have replicated the modelling effect using the taste-test paradigm, but others have used more naturalistic settings, which may increase the validity of the findings. For example, in a study reported by Hermans et al. ⁽Reference Hermans, Engels and Larsen¹⁷⁾, participants ate together whilst watching television and were informed that snack food was provided to make the environment more natural. Similarly, the same researchers⁽Reference Hermans, Herman and Larsen¹⁸⁾ led participants to believe that their study was about the effect of breakfast on cognitive performance, so participants ate breakfast together in a naturalistic bar/restaurant setting⁽Reference Hermans, Herman and Larsen¹⁸⁾.

The effect of the presence of a high v. low intake confederate on snack food intake has been replicated many times, including two reports that participants in a high intake confederate condition ate twice as much as participants in a low intake condition (about 380 v. 190 kJ (90 v. 45 kcal) of candy, respectively)⁽Reference Hermans, Engels and Larsen^17, Reference Hermans, Larsen and Herman¹⁹⁾. Feeney et al. ⁽Reference Feeney, Polivy and Pliner²⁰⁾ conducted a study using a similar design and examined the effect of a low intake model v. no model on pizza snacking. When the confederate ate very little, food intake was 37 % lower than when the confederate was absent (a reduction of approximately 630 kJ (150 kcal)). Hermans et al. ⁽Reference Hermans, Larsen and Herman²¹⁾ also examined the modelling effect on intake of vegetable snacks and found that eating with a high intake model was associated with higher intake than eating with a low intake model.

Similar effects are observed for peer modelling in young people. Bevelander et al. ⁽Reference Bevelander, Anschutz and Engels²²⁾ examined the effect of a present model on the snack intake of normal-weight and overweight primary school children. The children completed a puzzle task concurrently with access to the snacks. Normal-weight and overweight participants modelled intake but the effect of a model eating a large amount of food was particularly pronounced in the overweight children. The modelling effect remained even when participants returned for a second session 2 d later and ate alone, which suggests that norm effects are long lasting.

A number of other recent snack food studies have assessed food intake modelling using dyad designs, whereby two ‘real’ participants are left alone to consume high-energy sweet snack food and the main analysis of interest is the correlation between dyad members' food intake. In line with the results from confederate studies, the intake of one member of the dyad is closely rated to their partner's intake⁽Reference Herman, Koenig-Nobert and Peterson^23, Reference Robinson, Tobias and Shaw²⁴⁾. Salvy et al. ⁽Reference Salvy, Jarrin and Paluch²⁵⁾ reported that female and mixed-sex dyads exhibited significant modelling but there was no modelling in male dyads. Salvy et al. ⁽Reference Salvy, Romero and Paluch^26, Reference Salvy, Howard and Read²⁷⁾ have also conducted a number of dyad studies in normal-weight and overweight children. A good predictor of how much one dyad member eats is the amount eaten by the other dyad member⁽Reference Salvy, Romero and Paluch^26, Reference Salvy, Howard and Read²⁷⁾.

Intake modelling during main meals

Few studies have examined food intake modelling during main meals. Hermans et al. ⁽Reference Hermans, Herman and Larsen¹⁸⁾ examined modelling effects on breakfast consumption and found that the presence of a low intake confederate did not cause participants to eat less than they did in the presence of a high intake confederate. This finding was interpreted as evidence that meal-modelling effects may be less powerful than snack intake modelling. It is possible that individuals have pre-existing conceptions about how much they should eat for main meals but are more uncertain about snacking. The only other mealtime food intake modelling study we are aware of is by the same authors. Hermans et al. ⁽Reference Hermans, Larsen and Herman²⁸⁾ also tested modelling during an evening meal. Pairing participants with a high intake model resulted in greater meal intake by approximately 310 kJ (75 kcal) than pairing with a low intake model. Although statistically significant, this 15 % difference in food intake is smaller than that usually observed in snacking studies. Yet, there has been no comparison of modelling effects in snacks v. main meals in the same study, so it is difficult to draw conclusions regarding differences between the two meal types.

What drives intake modelling: high or low eaters?

Recent studies do not provide strong support for the suggestion that a high intake confederate increases food intake relative to eating alone, but the effect of a low intake confederate appears to be more robust⁽Reference Hermans, Herman and Larsen^18, Reference Hermans, Larsen and Herman¹⁹⁾. It may be that food intake modelling effects are driven primarily by self-presentation concerns and avoidance of being seen as a ‘greedy’ eater in front of a partner eating little or nothing. More direct of testing of the proposition that high eating partners liberate individuals from restraining their intake and thus may contribute to the spread of obesity though social networks is required⁽Reference Christakis and Fowler²⁹⁾.

Modelling of food choices

Robinson & Higgs⁽Reference Robinson and Higgs³⁰⁾ tested whether a confederate choosing predominantly high-energy food items from a lunch buffet influenced the food selection of participants. Relative to a choosing-alone condition, and a condition in which a confederate chose predominantly low-energy food items (including vegetables), the presence of the high-energy confederate resulted in participants choosing and consuming up to 40 % lower amount of vegetables. Two studies by McFerran et al. ⁽Reference McFerran, Dahl and Fitzsimons³¹⁾ provide evidence that the amount of food individuals choose to serve themselves can be influenced by confederates choosing large and small amounts of food. McFerran et al. ⁽Reference McFerran, Dahl and Fitzsimons³¹⁾ also manipulated the body size of confederates in their studies. Both slim and overweight confederates influenced food selections. Compared with a choosing-alone control condition, the presence of a slim confederate who selected a large amount of food increased the amount of snack food selected by the participants by 70 and 241 %. The presence of an overweight confederate who selected a large amount increased participant selections by 25 and 168 %. Finally, outside of the laboratory, Guarino et al. ⁽Reference Guarino, Fridrich and Sitton³²⁾ reported findings from an observational study in a canteen. Being accompanied by others who chose dessert increased the likelihood of the participant also choosing to eat dessert. It should be noted that not all studies examining modelling of food choices have found significant effects. Pliner & Mann⁽Reference Pliner and Mann³³⁾ found little evidence that participants would choose a low-energy and unpalatable cookie over a higher-energy palatable cookie after learning that other participants had chosen the former. Thus, although in the majority of studies the modelling effect is observed, there are likely to be some barriers that prevent modelling (for example, when the available food in question is disliked).

What factors influence modelling?

The exact mechanisms underlying modelling effects are not entirely clear, although we suggest that they are driven mainly by social desirability. There have been two main suggestions as to why the eating behaviour of present diners influences food consumption: (1) appropriateness norms; (2) ingratiation concerns. Although these two explanations are likely to be closely linked, they produce distinct hypotheses that should help us understand when and how individuals will influence each other's eating behaviour.

Modelling for appropriateness

Herman et al. ⁽Reference Herman, Roth and Polivy⁸⁾ propose a normative model, which suggests that the amount eaten by a confederate constitutes an appropriate amount of food to eat that should not be exceeded. Participants avoid eating more than this amount so as not to appear greedy. In support, individuals do not like to eat more than other diners because there are negative stereotypes regarding overeating⁽Reference Vartanian, Herman and Polivy³⁴⁾. The appropriateness account also suggests that modelling should be more likely in eating contexts that promote uncertainty about how much to eat. As noted, during main meals, for which individuals have clear scripts about what they normally eat, modelling effects are not particularly strong⁽Reference Hermans, Herman and Larsen^18, Reference Hermans, Larsen and Herman²¹⁾. On the other hand, in settings where there is more uncertainty over appropriate eating behaviour, such as snacking on candy whilst watching television or during the taste-test paradigm, modelling effects are robust⁽Reference Johnston^15, Reference Hermans, Larsen and Herman²¹⁾. One implication of the appropriateness account is that modelling should be less strong when eating in an unfamiliar situation or with a dissimilar dining companion (for example, a short slim women eating with a heavy man). This is because the eating behaviour of a dissimilar eating companion would not be deemed ‘appropriate’ to model. Indeed, similarity between a participant and confederate in terms of body size promotes modelling⁽Reference Hermans, Larsen and Herman¹⁹⁾, although dissimilarity does not necessarily eliminate it⁽Reference McFerran, Dahl and Fitzsimons³¹⁾, suggesting that additional accounts are needed to explain modelling effects.

Modelling for ingratiation

Modelling effects may be explained in part because diners eat the same amounts as their partner to be liked (‘imitation is the sincerest form of flattery’). Hermans et al. ⁽Reference Hermans, Engels and Larsen¹⁷⁾ tested this proposition indirectly by instructing a confederate to act in a friendly v. unsociable manner. In the condition in which the confederate was friendly there was no evidence of modelling. Thus, when there was little desire to be liked (because the confederate was already accepting), the modelling effect was removed. Conversely, when the confederate was not sociable, a modelling effect was observed. We recently tested whether making individuals feel socially accepted before eating with a confederate would reduce the similarity in intake between the participant and the confederate. In line with the findings of Hermans et al. ⁽Reference Hermans, Engels and Larsen¹⁷⁾ we found that there was less modelling in a condition that encouraged social acceptance than in a neutral control condition⁽Reference Robinson, Tobias and Shaw²⁴⁾. Furthermore, the matching of intake in dyads is reported to be moderated by the desire to be liked by others and the tendency to consider the viewpoint of others⁽Reference Robinson, Tobias and Shaw²⁴⁾, as well as the extent to which individuals express themselves in social situations⁽Reference Brunner³⁵⁾.

The ingratiation account of modelling suggests that the extent to which an individual is keen to be liked will increase the likelihood that they copy their dining partner⁽Reference Robinson, Tobias and Shaw²⁴⁾. Similarly, eating situations in which individuals are keen to make a good impression (amongst new acquaintances, for example) should also increase the likelihood of modelling. What the ingratiation account predicts for modelling amongst friends or spouses is less clear. One hypothesis could be a reduced modelling effect, because partners would not feel the need to portray a positive image for those they are already close to. Conversely, if there were a wish to maintain a bond and be viewed positively by friends, modelling effects would be expected. Few studies have examined modelling in friends v. strangers and the evidence is mixed, with some studies finding a similar modelling effect in friends and strangers and others finding stronger modelling in friends than strangers⁽Reference Salvy, Jarrin and Paluch^25, Reference Salvy, Vartanian and Coelho³⁶⁾.

Work on unconscious mimicry may also support the modelling for ingratiation account. Individuals automatically mimic the gestures and hand movements of others⁽Reference Chartrand and Bargh^37, Reference Chartrand and van Baaren³⁸⁾. One suggestion why individuals mimic each other's eating behaviour is that it is an unconscious attempt to make the other individual like them because mimicry eases social interactions⁽Reference Chartrand and Bargh^37, Reference Chartrand and van Baaren³⁸⁾. Hermans et al. ⁽Reference Hermans, Lichtwarck-Aschoff and Bevelander³⁹⁾ examined this proposition by analysing video data from a confederate modelling study. They found that in 10 s time-gaps after a confederate had picked up a piece of food, the likelihood that a participant would also pick up food was increased. Sip taking during social alcohol drinking follows a similar pattern of results⁽Reference Larsen, Engels and Souren⁴⁰⁾ and self-report measures of social desirability predict alcohol modelling⁽Reference Caudill and Kong⁴¹⁾.

Modelling for reasons unrelated to appropriateness or ingratiation?

Unconscious mimicry may occur in the absence of ingratiation attempts. It has been reported that viewing another individual performing an action (for example, grasping for food) activates a similar reaction in an individual's motor system⁽Reference Rizzolatti and Craighero^42, Reference Iacoboni, Woods and Brass⁴³⁾. Moreover, it has been suggested that such mimicry could occur regardless of salient goals or intentions⁽Reference Cook, Bird and Lünser⁴⁴⁾. If mimicry underlies modelling, then modelling of food picking would be predicted to occur immediately after the action of a high or low intake model. Hermans et al. ⁽Reference Hermans, Lichtwarck-Aschoff and Bevelander³⁹⁾ reported that modelling was more likely within 10 s of a model picking up food, but, as far as we are aware, there have been no investigations of modelling within shorter time frames.

Beliefs about the types and amount of food that other individuals eat have profound effects on eating behaviour, even when those other individuals are not present, suggesting that the effects cannot be driven solely by ingratiation⁽Reference Tanner, Ferraro and Chartrand^45, Reference Roth, Herman and Polivy⁴⁶⁾. Roth et al. ⁽Reference Roth, Herman and Polivy⁴⁶⁾ and Pliner & Mann⁽Reference Pliner and Mann³³⁾ examined the effect of providing bogus information about what non-present others ate in an experiment. This design is commonly referred to as the ‘remote-confederate’ design. Before giving participants the opportunity to snack on a high-fat and -sugar snack food (cookies), the researchers exposed participants to fictional information about what other individuals had eaten during earlier sessions. When participants believed that other individuals in the study had eaten large amounts of the snack, their own intake increased substantially⁽Reference Pliner and Mann^33, Reference Roth, Herman and Polivy⁴⁶⁾. In a similar study by Feeney et al. ⁽Reference Feeney, Polivy and Pliner²⁰⁾, when participants were led to believe that the norm was to eat very little, they more than halved their intake of pizza and ate 840 kJ (200 kcal) less than participants in a no-norm control. Using a remote-confederate design Robinson et al. ⁽Reference Robinson, Benwell and Higgs⁴⁷⁾ found bidirectional effects of norm information: participants in a high-norm condition ate significantly more (approximately 40 % or 310 kJ (75 kcal)) than in a no-norm condition, whereas participants in a low-norm condition ate significantly less than the no-norm group (approximately 40 %).

There is also research indicating that food intake modelling can occur in adults and children as a result of watching a non-present individual's eating behaviour⁽Reference Tanner, Ferraro and Chartrand⁴⁵⁾. Romero et al. ⁽Reference Romero, Epstein and Salvy⁴⁸⁾ examined whether viewing a video of another child selecting and eating a small or large number of cookies influenced selection and consumption of cookies in children aged 8–12 years who were participating in the same study. Healthy-weight and overweight girls were recruited. More cookies were eaten when watching another child select and eat a large amount of cookies than when watching another child eat a small amount (a 389 kJ (93 kcal) difference between conditions). The interaction between participant weight status and amount chosen by the video model was only marginally significant, but the effect was particularly pronounced in overweight children: a 682 kJ (163 kcal) difference between conditions for the overweight children v. a 184 kJ (44 kcal) difference for the healthy-weight children. The finding that a model can influence intake when participants believe that they will never meet that individual and when they believe that their eating behaviour is not being monitored is of interest because it suggests that social desirability concerns are unlikely to motivate mimicry. Although this could be perceived as evidence that automatic mimicry is occurring, the design of such studies could accommodate a further possibility: that a model's eating behaviour provides information about how most other individuals behave and creates a perceived eating norm.

Manipulating context-specific perceived eating norms

Burger et al. ⁽Reference Burger, Bell and Harvey⁶¹⁾ report two studies in which participants made a food choice and were provided with information suggesting that other individuals had chosen a low-energy ‘light’ snack bar or high-energy ‘full-fat’ snack bar. In study 1, the energy difference between the two snack bar types was up to 710 kJ (170 kcal). Both studies found that the participants who were led to believe that the norm was to select the high-energy snack bar were far more likely to select a high-energy snack bar than participants led to believe that the norm was to select a low-energy snack bar. In study 1, the high-energy-choice norm resulted in 28 % more of participants selected the higher-energy option and in study 2 the increase was 35 %⁽Reference Burger, Bell and Harvey⁶¹⁾.

Mollen et al. ⁽Reference Mollen, Rimal and Ruiter⁶²⁾ have also shown that norms guide food choices in a field experiment. Participants were exposed to social-norm messages in a food court. Participants were more likely to choose ‘healthier’ food options after exposure to a norm message emphasising healthy eating, than after exposure to a no-norm control message. Similarly, Salmon et al. ⁽Reference Salmon, Fennis and de Ridder⁶³⁾ have reported that individuals are more likely to choose a healthy food option after it was suggested to them that this choice was endorsed by the majority of other individuals. This effect was more pronounced for individuals lacking self-control. The authors suggested that the norms provided a form of decision-making heuristic (or ‘behavioural guide’). Moving outside of the laboratory, Prinsen et al. ⁽Reference Prinsen, de Ridder and de Vet⁶⁴⁾ have also shown that healthy food choices are more likely when aspects of the environment imply that it is the norm to choose this option.

The duration of perceived norm effects on intake has been less well investigated. Feeney et al. (JR Feeney, P Pliner, J Polivy and PC Herman, unpublished results) used a repeated-measures design in which participants were exposed to a high- or low-intake norm before eating and then returned for three further sessions on consecutive days. There was no eating-norm information present in these later sessions. A significant eating-norm effect on pizza intake was observed during the first session. The difference in intake between the low- and high-intake-norm conditions was approximately 1260 kJ (300 kcal), confirming the strong effect of perceived eating norms on food intake. Significant differences were also observed for the following two sessions when the eating-norm information was not present. Although the results were in the expected direction, by the third session the difference in intake between norm conditions was sizeable (710 kJ (170 kcal)), but no longer statistically significant. One reason why the influence of a perceived eating norm decreased by the final session might be that repeated daily consumption of the test food resulted in some participants habituating to the food, weakening the influence of a high-intake norm⁽Reference Epstein, Temple and Roemmich⁶⁵⁾. Nevertheless, these findings indicate that perceived eating norms can have both immediate and persistent influences on food intake.

Experimentally manipulating non-context-specific eating norms

Other studies have examined whether perceived eating norms that are not context specific produce similar effects on the types and amounts of food that individuals eat. We recently conducted two studies in which university students viewed flyers and posters about the daily fruit and vegetable intake of other students⁽Reference Robinson, Fleming and Higgs⁶⁶⁾. Participants either saw materials emphasising the health benefits of eating fruit and vegetables or information about fruit and vegetable eating norms. We used real survey data to suggest that other individuals (students from the same university) eat more portions of fruit and vegetables on a daily basis than might be expected. In an apparently unrelated later part of their visit to the laboratory, participants selected and consumed a lunch (study 1) or a snack (study 2). In study 1, after being exposed to an eating-norm message about vegetable intake, usually low consumers of vegetables ate more vegetables than after exposure to health message. Compared with the participants shown information about the health benefits of eating vegetables, they more than doubled their intake to 67 g, which is approximately one recommended daily serving. There was also a proportional increase in the percentage of the meal derived from vegetables. There was no effect of the manipulation on habitual high consumers of vegetables and this is probably because they were already adhering to the presented eating norm and did not feel motivated to change their behaviour⁽Reference Robinson, Fleming and Higgs⁶⁶⁾. In study 2, the eating-norm information that was provided emphasised that other students eat more portions of fruits and vegetables than might be expected. Food choice and intake during a snack were later measured. A similar pattern of results was observed as for study 1. Usually high consumers of fruits and vegetables consumed a relatively large amount of fruits and vegetables regardless of seeing eating-norm or health information. In usually low consumers of fruits and vegetables, a three-fold increase was observed after the social v. the health information and this approximated to consuming an extra recommended daily serving. The increase in fruit and vegetable intake was also associated with a decrease in consumption of high-energy snack foods (to accommodate the additional fruit and vegetables), resulting in the eating-norm message condition consuming 318 fewer kJ (76 fewer kcal) than the health information control condition. There was no effect of a message stating that other individuals approve of eating fruit and vegetables and we turn to the significance of this finding shortly.

Three other studies have examined whether information about eating norms influences fruit and vegetable intake. Croker et al. ⁽Reference Croker, Whitaker and Cooke⁶⁷⁾ exposed participants to survey information suggesting that UK residents have high intentions to consume their five portions of fruits and vegetables per d. In participants with low intentions at baseline (males), exposure to this information was associated with greater intended fruit and vegetable consumption than exposure to control messages. Stok et al. ⁽Reference Stok, de Ridder and de Vet⁶⁸⁾ also examined the effect of exposing participants to eating norms about fruit intake on intentions to eating fruit. Participants were either led to believe that other students eat a lot of fruit or very little. Intentions to consume fruit during the following week were then measured. Participants exposed to a high fruit intake norm intended to consume significantly more portions of fruit during the following week than participants exposed to the low-fruit-intake-norm message. Following up these results in a second study, the authors tested whether norm messages increased self-reported fruit intake. Participants were aware that the study was designed to increase fruit consumption. Participants were given some dietary advice and then kept a food diary for 1 week. During this time they were exposed intermittently to messages emphasising that others eat a lot or very little fruit. The norm reference group was either other students or the national population. Although there was no effect of a national population norm, students exposed to high fruit intake norms about other students self-reported that they ate on average four more portions of fruit than students who were exposed to low-fruit-intake-norms messages. This is a moderate effect but it should be noted that participants were already attempting to increase their fruit intake, so an important implication is that norms may influence the individuals who are already motivated to change their diet.

Using perceived eating norms to reduce intake of ‘unhealthy’ foods

Perceived eating-norm messages have also been shown to reduce the selection and intake of unhealthy foods. In a recent study, we exposed participants either to a message that most university students try to limit the amount of ‘junk food’ they eat or to a control message. We then examined the effect of this manipulation on snack choice and intake⁽Reference Robinson, Harris and Thomas⁶⁹⁾. Junk food was defined as food that is high in energy with little nutritional value. Exposure to the eating-norm message resulted in a 28 % reduction in the weight (g) of high-energy snack foods eaten (chocolate, cookies and crisps) relative to the control message (a reduction of about 250 kJ (60 kcal))⁽Reference Robinson, Harris and Thomas⁶⁹⁾. Berger & Rand⁽Reference Berger and Rand⁷⁰⁾ and Berger & Heath⁽Reference Berger and Heath⁷¹⁾ examined whether eating norms about an undesirable social group that individuals would not want to associate with influenced food choices and intake. Exposing participants to an eating-norm message suggesting that an undesirable group regularly ate ‘junk food’ resulted in participants reducing their ‘junk food’ intake, in particular energy from fat⁽Reference Berger and Rand^70, Reference Berger and Heath⁷¹⁾.

Effect of norm types

It should be noted that some studies have assessed the predictive power of perceived eating norms on dietary behaviour but found no effect⁽Reference Povey, Conner and Sparks⁷²⁾. This may be because the conceptualisation of what constitutes a perceived eating norm varies across studies. A descriptive norm refers to beliefs about how others behave, whereas an injunctive norm refers to beliefs about approval of behaviours. Injunctive norms are not always predictive of behaviour. Conner et al. ⁽Reference Conner, Norman and Bell⁷³⁾ found no evidence of the predictive power of injunctive norms, whereas another study⁽Reference Cook, Kerr and Moore⁷⁴⁾ found that injunctive norms predicted intentions towards consuming GM food. Similar inconsistencies are observed in studies examining whether injunctive norms predict habitual fruit and vegetable intake⁽Reference Guillaumie, Godin and Vézina-Im⁷⁵⁾. This pattern of results may suggest that perceptions about what other individuals are actually eating are a stronger predictor of eating behaviour than perceptions of what other individuals think should be eaten. This perhaps comes as no surprise, but additional aspects of three of the studies we reviewed earlier support this proposition. Lally et al. ⁽Reference Lally, Bartle and Wardle⁵⁵⁾ found that perceptions of peers' attitudes towards consuming different food types did not predict intake, whereas perceptions of what peers were eating were predictive. Mollen et al. ⁽Reference Mollen, Rimal and Ruiter⁶²⁾ too found no evidence that an injunctive norm increased healthier food intake and we have also found that a message about approval of fruit and vegetable consumption did not affect eating behaviour⁽Reference Robinson, Fleming and Higgs⁶⁶⁾.

Applied relevance: perceived eating-norm messages

In line with socio-cognitive models of health behaviour and norm theories in social psychology⁽Reference Rivis and Sheeran^52, Reference Lapinski and Rimal^77, Reference Schultz, Nolan and Cialdini⁸⁴⁾, perceived eating norms appear to influence behaviour by acting as a form of behavioural guide. Thus, public health messages could emphasise the healthy eating habits of other individuals or minimise the perception that other individuals are eating unhealthily. How best to formulate these messages should be driven by our understanding of the underlying mechanism(s). For example, messages outlining the behaviour of others (rather than the attitudes or intentions of others) would be expected to be most effective because this information may serve as the best adaptive cue. In addition, messages about individuals for whom a social connection is lacking may be easily dismissed as an inappropriate guide. Furthermore, if a norm suggests implicitly that a significant proportion of individuals choose not to follow that norm (for example, ‘did you know that 38 % of people consume five portions of fruit and vegetables a day’ also can be interpreted as 62 % of people choose not to), this could have the opposite of the desired effect and encourage those already adhering to the presented norm to eat less healthily.

The majority of studies to date have examined context-specific norms (‘what other people do in this experiment’) and the effect sizes observed are substantial. Thus, context-relevant point of choice or purchase social-norm messages may be hypothesised to be particularly effective in changing behaviour (‘most people in this canteen have vegetables with their lunch’). Context-specific norms can presumably produce a clear behavioural guide that reduces ambivalence over whether or not individuals view themselves as adhering to or rejecting a norm. In contrast, a non-context-specific-norm message (‘most people eat three portions of vegetables on a daily basis’) may not commit consumers to act immediately on exposure to the norm message. If the norm is not tied to a specific context or setting, individuals could choose not to eat vegetables during their lunch without rejecting the norm.

Applied relevance: positive social image and healthy eating

Tarrant & Butler⁽Reference Tarrant and Butler⁸²⁾ and Berger & Rand⁽Reference Berger and Rand⁷⁰⁾ suggest that social influence on eating behaviour can occur because of aspirations regarding affiliation or identity. In short, individuals will conform to the eating behaviour of others because they wish to portray a positive image⁽Reference Burger, Bell and Harvey^61, Reference Mollen, Rimal and Ruiter⁶²⁾. Making use of this knowledge to encourage adolescents to make healthier dietary choices could be fruitful. Adolescence is a period of life in which desires to be liked and ‘fit in’ to peer groups are high⁽Reference Steinberg^85, Reference Mwamwenda⁸⁶⁾. One possibility is to extend traditional healthy eating interventions and use campaigns led by aspirational figures, such as children in older school years, or more specifically, adolescents who are perceived as the ‘cool’ in the crowd. In a similar vein, school-based interventions that use media to encourage adolescents to associate healthy eating behaviours with desirable social groups warrant further investigation⁽Reference Horne, Tapper and Lowe^87, Reference Lowe, Horne and Hardman⁸⁸⁾.

Moving away from working directly with adolescents, another possibility is to embed healthy eating habits within social identity. For example, university students tend to base their sense of self on their position as a student of their specific institute. Similarly, sports fans can have a strong sense of affiliation to their respective teams. Thus, approaches that aim to pair identities with healthy eating could motivate individuals to eat more healthily in order to maintain identity. Pairing healthier eating with desirable groups rather than pairing unhealthy eating with undesirable groups may be more feasible and less offensive to the social group deemed undesirable.

Future research

In the present review we have drawn attention to the important influence of social context on eating behaviour, as well as highlighting underlying mechanisms that may be relevant for nutritional interventions. However, a number of questions remain unanswered. Changes to societal norms have been proposed as one mechanism by which overeating and obesity have become commonplace⁽Reference Pachucki, Jacques and Christakis⁸⁹⁾. Although the literature reviewed here suggests that eating norms can shape food choice and intake, how norms are transferred through large social groups remains unclear. Social proximity may predict the norms that individuals will follow and would include the normative behaviour of siblings, parents, spouses, friends and work colleagues. Hence, it will be critical to understand the factors that determine whose norms we follow and why.

We will also benefit from a better understanding of how broader eating norms make an impact on behaviour. Information about other individuals in the UK can motivate intentions to eat more healthily⁽Reference Croker, Whitaker and Cooke⁶⁷⁾. Similarly, it has been reported that making a national identity more salient results in changes to dietary intentions⁽Reference Tarrant and Butler⁸²⁾. If eating-norm messages or interventions prove to be feasible, broad messages would have the potential to reach larger groups of individuals in need of intervention. It will be important to examine the effectiveness of different types of message and the underlying mechanisms.

Better understanding of the drivers of eating in social situations is also required. There are several potential underlying mechanisms: ingratiation concerns; appropriateness; updating of perceived eating norms; unconscious mimicry. Picking apart these mechanisms is not easy but in future it would be informative to test their relative mediating roles. For example, we have reported recently that social information influences how much an individual believes they like a food⁽Reference Robinson and Higgs⁹⁰⁾. Thus, it could be that some social influence effects on behaviour occur in part because social information changes beliefs about the self and individuals then change their behaviour to maintain a sense of consistency between their self-concept and behaviour.

A major limitation of much of the social influence research to date has been an emphasis on sampling young adult healthy-weight females. Only a small number of studies have examined modelling in males⁽Reference Salvy, Jarrin and Paluch^25, Reference Hermans, Herman and Larsen⁹¹⁾. Moreover, as the population weight norms are changing, if social influence inspired interventions for dietary behaviours are to be effective, we will also need to understand whether overweight and obese individuals respond in a similar way to eating norms as healthy-weight individuals. One prediction is that obese individuals may categorise themselves and their eating as ‘different’ from a general population, so they may disregard norms about the eating behaviour of most other individuals. Alternatively, there has long been a suggestion that obese individuals are more prone to external eating cues⁽Reference Schachter and Gross⁹²⁾ and one might predict enhanced responsiveness to information about perceived norms. Although weight gain is likely to be under social influence⁽Reference Christakis and Fowler²⁹⁾, further evidence is required that social norms about eating contribute to unhealthy dietary practices and the development of obesity.