Social influence effects on food valuation generalize based on conceptual similarity

Abstract

Opinions of others influence behavior and decision-making, with important consequences for health. An unaddressed question is whether and how social influence can generalize across different situations or decisions. From a learning perspective, generalization is the transfer of previously acquired information to new stimuli and can be based on both perceptual and conceptual similarity. Here, we test whether social influence generalizes to new choices based on shared conceptual features, such as the healthiness and tastiness of different food items. We conducted three studies (total N = 468), in which healthy participants rated how much they would like to eat different food items and were subsequently presented with the ratings of several other people (‘social ratings’). Unbeknownst to our participants, they were randomly assigned to social ratings that either reflected a mainly health-driven valuation of food items (‘Health group’) or to social ratings that reflected a taste-driven valuation of food items (‘Taste group’). The results in all three studies showed that participants' food ratings became more influenced by healthiness in the ‘Health group’ than in the ‘Taste group’. In one study, these effects further transferred to food choices in a naturalistic supermarket task. Our findings provide experimental evidence of generalization of social influence effects based on inferred social health norms. Futures studies could test conceptual generalization of other types of social and non-social learning and characterize the brain mechanism underlying these effects.

1. Introduction

Human behavior is strongly influenced by social and cultural context. A prominent example are food choices and eating behavior. Some of these social norms are relatively stable and strict. We are likely used to eat beef if we have grown up in Argentina, but we would be repelled by the idea if we were Hindu. Other social norms may be more subtle, evolve over time, and vary between different subcultures or even smaller groups. For example, even if we usually like to eat steak with fries, we might choose something different if we go to the restaurant with a new group of friends that are health-conscious and/or concerned about climate change. Social norms can be defined as “the implicit or explicit rules a group has for the acceptable behaviors, values, and beliefs of its members” (Aronson et al., 2010, p. 258). Whereas explicit norms reflect conscious consideration of how the group norms, implicit norms develop through exposure to other people's reactions or evaluations (Yoshida et al., 2012). This framework invites the question: How do humans learn about and adapt their behavior to implicit social norms?

Many experimental and observational studies show that food choices are influenced by the observed behavior of other people (Higgs, 2015; Robinson, 2015; Stok et al., 2016). For example, when people eat together with someone eating a large amount, people are likely to consume more than when they eat alone (Cruwys et al., 2015; Vartanian et al., 2015). Food preferences are influenced by the ratings of even anonymous peers (Nook & Zaki, 2015), and even non-human primates have been shown to choose food in line with observed group norms (Van de Waal, Borgeaud, & Whiten, 2013). Social influence effects are thus an important factor when making food choices and may interact with other factors known to be important, such as health and taste considerations (Rangel, 2013; Sobal et al., 2006; Steptoe et al., 1995). Epidemiological and social network studies have shown that a person's chance of becoming obese increases with the number of social contacts who are obese (Christakis & Fowler, 2007), and that social influence is the strongest driver of changes in diet (Eker et al., 2019). Observational studies have analyzed large-scale purchasing data to show that colleagues who eat together, choose similarly healthy food in workplace restaurant (Levy et al., 2021), and that over time, students and staff on university campuses are influenced in the healthiness of food and beverage purchases by the persons they regularly eat with (Gligorić et al., 2021). Using ecological momentary assessment, a recent study suggests that social cues influence eating behavior via changes in momentary social norms about the appropriateness of snacking behavior (Schüz et al., 2018), suggesting that perception of norms may change even on a relatively short time scale. Such effects are important, because, over time, even small social and cultural influences on food choices and other health-related behaviors such as exercising may have important consequences for health.

An important yet unaddressed question is whether and how social influence may generalize across different situations and across different foods that share common features. Previous studies have typically tested whether people's behavior is influenced by specific observation of others' behavior, such as whether they become more likely to choose the salad over the steak if others have chosen the salad. However, social and cultural influences likely operate in more general and abstract ways. For instance, we might not choose the same salad but something else that is also in line with the inferred values and norms of the group—such as an equally healthy vegetable soup. Here, we test the idea that social influence can generalize across food choices based on abstract conceptual dimensions such as healthiness and tastiness of food items.

From a learning perspective, generalization can be defined as the transfer of previously acquired information to new stimuli and situations, based on the similarity between the original and a new situation (Guttman & Kalish, 1956). Generalization was already noticed by Pavlov, 2010 in his classic conditioning experiments with dogs where he associated food with the sound of a bell and noticed that the animal's reaction generalized to other similar sounds. Subsequently, other scientists have obtained empirical "stimulus generalization gradients", which relate the probability, speed, or strength of a learned response to the difference between the test and training stimuli (Guttman & Kalish, 1956; Shepard, 1987). Generalization does not necessarily reflect a lack of discrimination between learned and new stimuli. Instead, generalization can be seen as an active process of associating a stimulus to a more general underlying pattern or rule, which is adaptive in a world where situations are often similar but rarely identical (Shepard, 1987). Accordingly, generalization follows both perceptual similarity gradients as well as conceptual relationships (Dunsmoor & Murphy, 2015). Thus, generalization allows people to learn not only about specific stimuli but to infer more general rules and apply them to novel contexts to adapt (Behrens et al., 2018).

Here, we tested whether similar conceptual generalization mechanisms could apply to social influence effects. Recent brain imaging studies suggest parallels between social influence and reinforcement learning, demonstrating that agreeing with others evokes activity in reward-related brain areas (Campbell-Meiklejohn et al., 2010; Klucharev et al., 2009; Zaki et al., 2011). Disagreeing with others is associated with activity in brain areas related to error processing and cognitive conflict (Klucharev et al., 2009; Koban et al., 2014), leading to subsequent adjustments in behavior and decision-making to reduce this kind of social conflict (Klucharev et al., 2009). If social influence is based on partially similar mechanisms as reinforcement learning, then it may also generalize based on perceptual and conceptual features.

Food choices naturally lend themselves for testing this idea. As other types of decisions, valuation of food items can be described as a function of different attributes and the importance (measured as beta weights) of these attributes for any given individual. Two of the most important attributes when evaluating food are its tastiness and its healthiness (Rangel, 2013) and the importance of these attributes varies substantially across individuals and situations (Hare et al., 2011; Schmidt et al., 2018). We hypothesized that social influence on food choices can generalize by shifting how much people take into account health and taste attributes. Thus, we predicted that, during observation of other people's food ratings, people implicitly infer underlying social norms regarding the importance of health and taste, which then influence their own food choices. In other words, if we observe that other people prefer broccoli and plain yoghurt over chocolate bars and sodas, we could infer that they are influenced by the healthiness of food items, and we might also shift our preferences to more healthy food options, even if this concerns new food items (in this example, other foods than broccoli or plain yoghurt).

For this purpose, we developed a novel social influence task in which participants rated how much they would like to eat different food items, first during a baseline phase without social information, then during a social influence phase (Fig. 1A). During the social influence phase, they were presented with the ratings of several other people (‘social ratings’) after their own food rating. For each food item, two sets of social ratings were selected to create two different experimental groups (Fig. 1B). One group of participants (the Health group) was presented with social ratings that reflected a preference for healthy food items (i.e., ratings were selected based on high weights of health and low weights of taste). The other group was presented with social ratings that reflected a low weighting of health and a high weighting of taste aspects of the food (the Taste group). If participants generalize from the social ratings, then participants in the Health group should increase the importance (or weight) of health attributes compared to the Taste group in the social influence phase (Fig. 1C). Participants in the Taste group in contrast should increase the importance of taste attributes compared to the Health group. We tested these hypotheses in three different studies, with different food stimuli and across two different countries. Further, in order to investigate the possible transfer of generalized social influence to another task, we implemented an additional online supermarket task (Lai et al., 2020) in Study 2 and 3, providing a measure of participants' food preferences closer to how they make food choices in the real world (Fig. 1A). We expected that participants in the Health group would still be more guided by food healthiness than the Taste group and thus buy healthier food products in the supermarket (Fig. 1C).

Fig. 1. Experimental design and hypotheses. A) Task design. During the food choice task (left and middle), participants had to rate different food items, which varied on healthiness and tastiness (healthi and tastei) as rated by an independent sample of participants. Brighter versus darker colored bars illustrate less to more healthy (light to darker blue) and less versus more tasty (light to darker red) food items (order randomized across participants). During the baseline phase (left), participants were presented with 16 food items in total without receiving any social feedback. During the social influence phase, participants were presented with 64–80 food items. For each food item, they rated how much they would like to eat it and were then presented with what we told them were the ratings of several other individuals. In Studies 2–3 and following the food rating task, participants performed a naturalistic online supermarket task. They were instructed to select several foods and drinks, as if they had to buy groceries for next days' breakfast and snacks. B) Selection of social ratings in the two experimental conditions in all three studies. Participants were randomly assigned to one of two experimental conditions of the food rating task. In the Health group, the mean social value (SV) of each food item i was computed with a high beta weight for the item's healthiness (healthi) and a low weight for its tastiness (tastei). In contrast, in the Taste group, the mean social value (SV) of each food item i was computed with a high beta weight for the item's tastiness (tastei) and a low weight for it's healthiness (healthi). Several ratings were randomly drawn from a distribution around the mean SV to ensure plausible and variable social ratings. Thus, as illustrated on the right side, in the Health group, healthy items such as vegetables or fruit would be followed by relatively high ratings, whereas unhealthy but tasty items such as chocolates or sweets would be followed by relatively low ratings. In the Taste group, this relationship was the other way around. C) Hypotheses. We expected that participants in the Health group would over time increase how much healthiness of food items influenced their food ratings as influenced by increased beta weights for healthi, whereas participants in the Taste group would show lower influence of health considerations in their choices during the social influence phase. The opposite pattern was expected for the influence of tastiness on food ratings (beta weights of tastei). We further expected that this effect would transfer to a second, independent supermarket task, such that participants in the Health group would select healthier food items (as measured by a higher objective health star point score) than participants in the Taste group. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

2. Methods

2.1. Participants

During the first 16 trials of the food rating task (baseline phase), all participants rated the same 16 food items (presented in randomized order) without any social information, allowing us to control for participants’ habitual weighting of health and taste considerations. Then, during the social influence phase (80 trials in Studies 1 and 3, 64 trials in Study 2), participants were presented with the ratings of several other participants following their own rating.Participants were randomly assigned to one of two experimental groups using a pseudo-random number generator of Qualtrics. In the Health group, healthier but less tasty food items were associated with higher average preference ratings of others ('social ratings'), whereas unhealthy but tasty items were associated with lower social ratings. In the Taste group, tastier but less healthy food items received higher preference ratings from others than healthier and less tasty items (see also Fig. 1B). See below for more details on the selection of social ratings for the different food stimuli.

Click to read more