The assimilation affects research from three days’ social facilitation of eating by using whole network autocorrelation analysis plus multilevel linear modeling: a normative analysis based on prospective cohort study data in the UK

Abstract

Objective

This study aimed to evaluate the influence of social facilitation of eating on dietary assimilation effects using a multi-method approach integrating whole-network autocorrelation analysis (WNA) and multilevel linear modeling (MLM).

Methods

We analyzed data from 26 participants (13 friend pairs) comprising 468 observations across multiple social and solitary eating contexts. The study combined WNA of 70 dietary similarity matrices with multilevel modeling to examine both systemic behavioral patterns and individual-level effects. MLM specifically assessed hierarchical data structure and fixed effects of social context and meal type.

Results

Our analyses revealed that 31.1% of variance in eating behaviors originated from individual and dyad levels (ICC = 0.311), confirming the necessity of multilevel approaches. Network analysis identified"Friendship_close," "Friendship_howWell," and"Menu_VigEx" as primary factors influencing dietary assimilation during social eating, with effects varying significantly across consecutive days and mealtimes. MLM results showed strong meal-type effects (P < 0.001) but non-significant main effects of social context alone (P = 0.266), suggesting that social influence operates through behavioral coordination mechanisms rather than mean-level consumption changes. Notably, dietary similarity networks exhibited instability in assimilation effects during solitary eating, while social contexts promoted consistent behavioral patterning.

Conclusions

Social facilitation of eating enhances dietary assimilation through behavioral coordination patterns that vary under specific temporal and contextual conditions. The complementary findings from network analysis and MLM indicate that social influence manifests primarily in synchronized eating patterns rather than overall consumption increases. Leveraging these coordinated behavioral patterns offers promising avenues for promoting healthy dietary norms through social contexts, potentially mitigating excess energy intake while fostering lifestyle norms conducive to physical and mental health. Future research should employ integrated methodological approaches to further elucidate these complex social dynamics.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12963-025-00427-4.

Introduction

Social facilitation refers to the phenomenon wherein mere social presence amplifies individuals’ dominant behavioral responses within specific situational contexts [1]. This psychological mechanism extends to eating behavior through the social facilitation of consumption, first systematically documented through John de Castro’s pioneering diary studies. His research paradigm involved longitudinal tracking of participants’ dietary patterns over 7-day intervals, meticulously recording food intake quantities, meal locations, and dining companions [1]. Eating with others plays a role in reinforcing social connections, and food choices can be used to communicate and express ourselves to others [2]. Contemporary epidemiological data reveals that 77% of UK adults maintain familial meal-sharing traditions with weekly frequency [3], and a substantial proportion of people’s meals are eaten with others just like sharing a meal with friends, family, or work colleagues which is common in many areas of the world including the USA, Europe, and Thailand [4–7]. Numerous studies have consistently demonstrated that people ate much more food when they ate in company than when they ate alone, and social facilitation was the single most powerful influence on eating [8–13]. However, emerging evidence suggests moderating variables influencing this effect magnitude, particularly social attitudes about the stigma associated with appearing to eat excessively [12, 14]. Current methodological approaches in this research domain predominantly employ advanced statistical methodologies including multiple linear regression and multilevel modeling. Scholarly investigations typically concentrate on three primary mechanisms: (1) behavioral modeling through observational learning, (2) impression management strategies in dietary self-presentation, and (3) fundamental social facilitation processes in consumption patterns.

While the immediate effects of co-dining on consumption patterns are well-documented, critical questions persist regarding its longitudinal behavioral impacts and underlying mechanisms [15]. Current evolutionary psychology frameworks posit that this phenomenon constitutes an adaptive survival mechanism, potentially emerging from ancestral competition for scarce nutritional resources within ingroup dynamics [15]. Nevertheless, the broader biocultural implications of socially facilitated consumption remain insufficiently theorized, particularly concerning its role in modern public health challenges. There is a lot of evidence that obesity spreads through social networks [16], and social facilitation may be the underlying trigger mechanism, but little is known about how this role in promoting social facilitation to chronic diseases such as obesity and diabetes. Collectively, research confirms that the effect of social facilitation on people’s behaviors is a powerful phenomenon, and the consumption experience of loneliness is different from that of sharing with others. Unregulated amplification of socially reinforced eating behaviors may constitute a novel risk factor in syndemic disease spread, necessitating multidimensional analysis extending beyond mere caloric intake considerations.

The assimilation effect refers to the phenomenon that when individuals face social comparison information, their self-evaluation tends to be consistent with the comparison objective, that is, individuals improve their self-evaluation level in the upward social comparison, or reduce their self-evaluation level in the downward social comparison [17]. In this study, we refer to the phenomenon that the various dietary indicators of individuals in different states tend to be consistent in a limited range with other individuals as an assimilation effect, and hypothesize that the assimilation effects of various dietary indicators may be the underlying mechanism by which the social convenience of diet promotes the increase in energy intake. The main process of this study was as follows: Firstly, the autocorrelations between the five diet-related indicators and other indicators (including activity indicators, friend intimacy indicators, and other individual basic indicators) were analyzed in the two different conditions (“Social eating” or “Eating alone”). Then the autocorrelations between the above five dietary indicators and other indicators were further analyzed in the two different conditions for each day of the three consecutive days. Finally, a cross-sectional analysis of the autocorrelations between dietary indicators and other indicators of people under these two different conditions in three consecutive days of breakfast, lunch, and dinner was explored. This three-pronged analytical framework—incorporating horizontal (cross-meal), vertical (cross-day), and contextual (social vs. solitary) dimensions—aims to: (1) decouple the assimilative mechanisms driving socially facilitated consumption; (2) develop evidence-based guidelines for healthy commensality practices; and (3) identify methodological limitations inherent in current experimental paradigms within social dining research.

Method

Data collection

The data was obtained through the UK Data Service, the nation’s primary government-funded repository for social science research infrastructure. Specifically, we analyzed from study 2 part of a project called “Eating Together: Effects of the Social Context of Eating on Appetite, 2018–2021” by Higgs, S et al., University of Birmingham [18].

Study 2 was a lab study in which Higgs, S et al. examined whether social facilitation effects on eating would be sustained across multiple meals and days, with 28 participants recruited. A within-subjects counter-balanced crossover design was used in Study 2 in which participants attended two phases of three consecutive weekdays each, separated by a 14-day washout period to mitigate carryover effects. In the social condition block, participants dined with a pre-assigned friend (also a study participant), whereas the solitary condition required individual dining. Phase order (Social vs. Alone) was randomized using a true random number generator (https://www.random.org/integers/) under researcher supervision to ensure allocation concealment [18].

Participants were eligible to take part if they met the following criteria which were listed in the study advertisement: (1) were aged over 18 years, (2) were occasional social eaters (1 to 3 meals per week), (3) had a self-reported BMI between 18 and 25 kg/m², (4) liked and were willing to eat the test foods, and (5) were willing to refrain from consuming calorie-containing food and drinks outside of those provided during the study [18]. Finally, a total of 26 participants (13 friend pairs, Participants with numbers 25 and 26 voluntarily exited midway through the study) took part in the “Social facilitation and longer-term energy intake” study.

After screening the original data, 18 indicators were included in our study. Two indicators are classified variables (Condition and Meal), and the rest are quantitative variables. Operational definitions and measurement specifications for all variables are systematically detailed in Table 1.

Table 1.

The basic information of the included indicators in this study

Code	Indicators	Description/operationalization	The stage in which the indicator is measured
-	Condition	Social context condition: In one phase, participants attended alone (alone condition) and in the other phase participants attended with a friend (social condition); 1="social”; 2="alone”	Before the laboratory experiment
-	Meal	Meal: 1="Breakfast”; 2="Lunch”; 3="Dinner”	Before the laboratory experiment
F1	TFEQ_R	Assess dietary restraint by using the Three-Factor Eating Questionnaire Revised 18-item version (TFEQ-18).	After dinner on day 3, phase 2
F2	TFEQ_un	Asess uncontrolled eating by using the Three-Factor Eating Questionnaire Revised 18-item version (TFEQ-18).	After dinner on day 3, phase 2
F3	TFEQ_em	Asess emotional eating by using the Three-Factor Eating Questionnaire Revised 18-item version (TFEQ-18).	After dinner on day 3, phase 2
F4	BMI	It is participants’ body mass index.	After dinner on day 3, phase 2
F5	Age	The number of years that a person has lived.	After dinner on day 3, phase 2
F6	Friendship_length (Month)	Participants stated how long they had known their friend with whom they had taken part in the study with “Friendship_length”.	After dinner on day 3, phase 2
F7	Friendship_close	Participants rated on 10-point scales how close they feel to their friend. (“Not very” and “Very” were assigned to values 1 and 10, respectively)	After dinner on day 3, phase 2
F8	Friendship_howWell	Participants rated on 10-point scales how well they know their friend. (“Not very” and “Very” were assigned to values 1 and 10, respectively)	After dinner on day 3, phase 2
F9	Liking_allmeals	Liking ratings were measured by the liking VAS, and then averaged across all foods consumed at each meal to conduct “Liking_av”. The liking VAS was anchored by “Didn’t like it at all” and “Liked it a lot” to the left and right of the scale, respectively.	Following the meal
F10	Mood_allmeals	It means “A positive mood score”. Mood assessments were taken using 100 mm Visual Analogue Scales (VAS). Each scale was anchored with “Not at all” on the left and “Extremely” on the right. A positive mood score was calculated by taking the average of scores assigned to the positive mood states and the inverse of scores assigned to the negative mood states.	Following the meal
F11	Time (Second)	The experimenter covertly recorded the duration of each meal in seconds “Time”.	After finishing each meal
F12	Appetite_pre	Assessments of hunger, fullness, and “liking” for the test foods were taken using 100-mm Visual Analogue Scales (VAS) anchors “Not at all” on the left and “Extremely” on the right. A composite “appetite” score was calculated by taking the mean rating assigned to the “hunger” VAS and the inverse rating assigned to the “fullness” VAS (100-fullness) [Before meal: Appetite_pre; After meal: AppetitePOST].	Before each meal
F13	Kcals	Calories consumed at each meal.	After finishing each meal
F14	Menu_LightEx (Minutes)	Before breakfast on each day (Over the past 24 h), participants were also asked to record the amount of time (in minutes) that they had spent engaging in light activities (e.g. household chores, slow walking).	Before breakfast on each day
F15	Menu_ModEx (Minutes)	Before breakfast on each day (Over the past 24 h), participants were also asked to record the amount of time (in minutes) that they had spent engaging in moderate activities (e.g. brisk walking).	Before breakfast on each day
F16	Menu_VigEx (Minutes)	Before breakfast on each day (Over the past 24 h), participants were also asked to record the amount of time (in minutes) that they had spent engaging in vigorous activities (e.g. running, spinning).	Before breakfast on each day

Constructing co-occurrence matrices and networks

In this study, 16 continuous variables were included in statistical inference except “Condition” and “Meal”. First of all, the 16 continuous variables were basically described, as shown in supplementary table S1. Then, according to the maximum and minimum values of each indicator, 468 values of the five diet-related indicators (i.e., values measured by 26 people at each meal of each day under different conditions, measurement repeated 18 times in total) were divided into 4 levels. Finally, matrices or social networks were created for diet-related indicators “Kcals”, “Appetite_pre”, “Mood_allmeals”, “Liking_allmeals” and “Time” according to different states (social or lonely, the first or second or third day of three consecutive days and breakfast, lunch and dinner were classified and combined, a total of 14 states), and they represented the “similarity” of each of the 26 participants to the other participants on a certain indicator. The numerical values in the matrices or social networks were the similarity indexes. If participant “P1” repeatedly measures indicator “A” for N times in a certain state, and at least half of the measured values (N ≥ n ≥ N/2) are at the same level as the value of participant “P2”, then the two are considered to be similar and their corresponding similarity index (S = n/N) is calculated. In this way, we got 14 matrices for each of the 5 diet-related indicators, with a total of 70 matrices. These matrices were subjected to autocorrelation analyses with the 15 remaining continuous variables using Mantel tests, enabling detection of state-dependent assimilation effects through comparison of matrix concordance patterns.

This study employs behavioral similarity network analysis based on systems science principles, aiming to detect naturally emerging behavioral coordination patterns under shared environmental conditions. Unlike traditional social network analysis, we do not assume pre-existing social connections but test whether environmental factors can systematically shape behavioral similarity structures.

Whole network analysis (wna): the main part of the social network analysis

Social network analysis (SNA) is an important method to investigate “relational data” by utilizing mathematics and graph theory. It can describe the characteristics of the whole network, the relationship among individuals, and the position of the individual in the network [19]. SNA can be divided into two types:

1. Macro-level network analysis (Whole Network Analysis, WNA): Focuses on structural properties of complete networks,including:

• Density and centrality metrics (e.g., betweenness, eigenvector centrality).

• Subgroup detection through community structure algorithms.

• Quadratic assignment procedure (QAP) for relational hypothesis testing.

• Exponential random graph models (ERGMs) for network formation mechanisms.

WNA’s principal strength lies in identifying structural isomorphism – the correspondence between emergent network configurations and collective behavioral outcomes – while quantifying how nodal embeddedness influences system-level phenomena [19].

• 2.Ego-centric network analysis: Examines local ties surrounding focal actors, emphasizing:

• Alter composition and tie strength heterogeneity.

• Brokerage opportunities in structural holes.

• Social capital accumulation mechanisms.

The present study prioritizes WNA to elucidate how global network configurations mediate assimilation effects in dietary behaviors.

Test of the relationship between continuous variables and social networks (matrixes)

In this study, Moran’s I and Geary’s C indices were used to test whether there is an autocorrelation between the “attribute variables of the distance measurement level” and the “network distance between them”. These two indices, which originally came from geography and later emerged in spatial statistics, measure whether the similarity in geographical features of any two places is related to the spatial distance between them [20]. Geary’s C compares pairs of data, while Moran’s I focuses on covariation.

Moran’s I was proposed by Moran in 1948. It reflects the similarity degree of adjacent unit attribute values in space and is generally used to test whether there is clustering in space. Its calculation formula is as follows: [21].

where is the attribute value of region i, C is the spatial weight matrix, C_ij represents the degree of influence between space units i and j, and is the mean value of each x. The value range of Moran’s I is between − 1 and 1. The closer it is to −1, the greater the difference between units; the closer it is to 1, the closer the relationship between units; the closer it is to 0, the irrelevant between units. The formula for Geary’s C is: [21].

Geary’s C is usually between 1 and 2. If Geary’s C is between 0 and 1, it indicates positive spatial autocorrelation. If Geary’s C is between 1 and 2, it indicates negative spatial autocorrelation. This study shifts its focus from “space” to social networks. Our core research question formalizes as: Do topological proximity (network distance) and attribute similarity (behavioral metrics) exhibit statistically significant covariation in social dining contexts?

Multilevel linear modeling as a complementary approach

In order to provide methodological triangulation, we also conducted Multilevel Linear Modeling (MLM) analyses. Critically, the MLM analyses utilized the exact same data foundation of 468 observations as the social network analysis, ensuring direct comparability of results between the two methodological approaches.

1. Data Foundation Consistency

• MLM analyses were based on the same 26 participants, 13 dyads, and 468 observations as the social network analysis.

• Utilized the same five core dietary variables: calorie intake (Kcals), pre-meal appetite (Appetite_pre), mood state (Mood_allmeals), food liking (Liking_allmeals), and meal duration (Time).

• Employed the same 14 state classifications (2 social contexts × 3 days × 3 meals, excluding non-existent combinations).

• Maintained identical data screening and processing procedures as the social network analysis.

Complementary Methodological Design

• 2.MLM and social network analysis examine research hypotheses from different yet complementary perspectives:

• MLM: Tests average effects at the individual level, quantifying fixed effects of social context and meal type on calorie intake.

• Social Network Analysis: Reveals behavioral similarity structures at the system level, detecting state-dependent assimilation patterns.

• Data Foundation Consistency: Ensures observed differences stem from methodological perspectives rather than data foundation discrepancies.

• 3.Model Specification

We constructed a series of MLM models:

1. Null model: Kcals ~ 1 + (1 | DyadID/PersonID).

2. Main effects model: Kcals ~ Condition + Meal + (1 | DyadID/PersonID).

3. Full model: Kcals ~ Condition + Meal + Appetite_pre + Mood_allmeals + Liking_allmeals + Time + (1 | DyadID/PersonID).

All models used restricted maximum likelihood (REML) estimation, with model comparisons conducted via likelihood ratio tests. Statistical analyses were performed using the lme4 package (version 1.1–35) in R_4.5.1.

Results

Basic description of the population characteristics and various indicators

Pairs of friends were recruited via social media and poster advertisements which were placed around the University of Birmingham campus. As shown in Supplementary Table 2, none of the 16 continuous variables conforms to the normal distribution.

As is shown in Fig. 1; Table 2, premeal appetite and liking rates for food of the study group during social eating did not change significantly compared with that during eating alone, while energy intake during social eating was higher than that during eating alone, but not as significant as positive mood and meal duration. On three consecutive days of social eating, the group spends more time on various activities on the second and third days than on the first day.

Fig. 1.

Numerical distribution of 16 quantitative indicators: Violin plot and Scatter plot (including median line and interquartile range lines)

Table 2.

The basic description and normality test results of 16 continuous variables

Code	Indcators	Condition	Mean	Std. deviation	Median	Interquartile range	95% Confidence interval for mean		Tests of normality
							Lower bound	Upper nound	Statistic	P
F1	TFEQ_R	-	14.135	3.959	14.000	5.500	12.371	15.669	0.159	0.102
F2	TFEQ_un	-	22.920	4.741	22.000	9.000	20.960	24.960	0.170	0.061
F3	TFEQ_em	-	7.150	2.395	7.500	4.000	6.060	8.020	0.137	0.200
F4	BMI	-	22.953	2.885	22.430	3.655	21.762	24.144	0.115	0.200
F5	Age	-	20.810	2.772	20.000	2.000	19.520	21.680	0.351	0.000
F6	Friendship_length (Month)	-	21.080	29.417	12.000	8.000	9.160	33.880	0.427	0.000
F7	Friendship_close	-	8.000	1.789	8.000	1.000	7.410	8.830	0.272	0.000
F8	Friendship_howWell	-	8.080	1.647	8.000	1.000	7.430	8.810	0.311	0.000
F9	Liking_allmeals	Social	74.760	15.234	77.750	20.000	72.631	76.699	0.091	0.000
		Alone	73.089	16.251	76.000	19.520	70.804	75.252	0.106	0.000
F10	Mood_allmeals	Social	80.556	12.292	81.670	18.170	78.888	82.193	0.095	0.000
		Alone	55.733	13.126	55.335	19.585	54.121	57.679	0.049	0.200
F11	Time (Second)	Social	1428.282	483.738	1380.540	659.980	1373.453	1501.302	0.057	0.077
		Alone	862.928	298.504	780.550	360.138	822.075	902.403	0.121	0.000
F12	Appetite_pre	Social	74.033	17.622	76.000	26.500	71.924	76.592	0.094	0.000
		Alone	74.416	14.907	76.500	20.000	72.578	76.579	0.081	0.002
F13	Kcals	Social	806.325	285.340	798.155	480.980	767.576	843.539	0.066	0.022
		Alone	752.264	275.958	742.645	380.673	713.867	787.579	0.044	0.200
F14	Menu_LightEx (Minutes)	Social	54.740	59.604	40.000	40.000	42.020	70.480	0.281	0.000
		Alone	68.920	92.865	40.000	63.000	47.360	90.970	0.255	0.000
F15	Menu_ModEx (Minutes)	Social	48.910	47.678	35.000	50.000	35.800	58.640	0.212	0.000
		Alone	55.180	60.454	40.000	40.000	41.700	70.030	0.243	0.000
F16	Menu_VigEx (Minutes)	Social	17.260	22.562	15.000	30.000	11.760	22.410	0.261	0.000
		Alone	19.330	30.105	0.000	30.000	11.510	24.440	0.335	0.000

Test results of the relationship between three consecutive days’ total dietary similarity matrices and continuous variables under social or alone condition

As for energy intake: The test results showed that under social condition, the more similar the energy intake of people was, the more similar their performance in “BMI”(G: 0.737), “Friendship_length”(M: 0.044), “Friendship_close”(G: 0.713, M: 0.063) and “Friendship_howWell”(G: 0.063, M: 0.063) were; Under alone condition, the more similar the energy intake of people was, the more similar meal duration(G: 0.685, M: 0.685) was, but the more different the time spent on light activities last day(M: −0.112) was. About premeal appetite, liking rates for food and positive mood: Under social condition, people with more similar premeal appetite, liking rates for food and positive mood had more similar energy intake, spend more similar time on vigorous activities, and had the more similar degrees about how well they know about the meal partner and about how close they feel to the meal partner, but were more different in the “TFEQ_un” and “Friendship_length” under alone condition. The more similar the premeal appetite was, whether socially or alone, the more similar their liking rates for food and positive mood were.

As for the meal duration: The results showed that whether under social or alone condition, the more similar the meal duration was, the more similar the liking rates for food (Social: [G: 0.621, M: 0.240], Alone: [M: 0.122]) and positive mood (Social: [G: 0.330, M: 0.592], Alone: [G: 0.182, M: 0.456]) were, and this trend is more significant under social condition. In the social condition, the more similar the meal duration was, the more similar the degree about how well they know about the meal partner (G: 0.626, M: 0.087), the degree about how close they feel to the meal partner (G: 0.593, M: 0.143), the energy intake (G: 0.667) and the length of time spent in vigorous activities (G: 0.795) were. Under alone condition, it reflected the more similar the restriction degree of food intake (G: 0.842), but the more different the number of months of acquaintance with the meal partner (G: 1.241, M: −1.241) and the degree about lacking control of food intake (M: −0.122) were.

For more details, see Fig. 2 and Table S1 (Supplementary file).

Fig. 2.

Geary’s C and Moran’s I of autocorrelation analysis between three days’ total similarity matrices and continuous variables under social or alone condition. The circle represents the values of the two exponents in the social condition, and the triangle represents the values of the Moran’s I and Geary’s C in the alone condition. A solid circle or solid triangle indicates the value of Geary’s C, a hollow circle or hollow triangle indicates the value of Moran’s I, and a red circle or red triangle indicates that P ≤ 0.05

Test results of the relationship between the dietary similarity matrices and continuous variables for each of three consecutive days under the social or alone condition

After the autocorrelation analysis of the diet-related indicators’ similarity matrices which included a total of three consecutive days’ data under social or alone conditions, we longitudinally split each similarity matrix containing three consecutive days of data into the similarity matrix of each day and then performed autocorrelation analysis on it. We can see that the autocorrelation analysis results of each day’s similarity matrix and other indicators were not exactly the same for three consecutive days. On the whole, in the three consecutive days, the test of the relationship between the similarity matrix of dietary indicators on the second day and other continuity variables showed that the number of statistically significant continuity variables was the largest, followed by the first day and the third day.

On energy intake: On day three, the more similar the energy intake was, the more similar the meal duration (Social: [G: 0.690, M: 0.180], Alone: [G: 0.777, M: 0.096]) were, but this was not shown on day one or day two; The first two days’ analysis showed that under social condition, people with similar energy intakes had a similar degree about how close they feel to the meal partner (Day 1: [G: 0.772], Day 2: [G: 0.718, M: 0.112]), but only the third day of the analysis showed that people with similar energy intakes had a similar number of months of acquaintance with meal partner (M: 0.061).

For premeal appetite, liking rates for food, and positive emotions of eating: the daily autocorrelation regression results for these three matrices were very similar. All results showed that on the second day of the social condition, the more similar these three indicators were, the more similar the degree of how well they know the meal partner (In turn, [M: 0.07], [M: 0.07], [M: 0.07]), the degree about how close they feel to the meal partner [In turn, [G: 0.805, M: 0.108], [G: 0.805, M: 0.108], [G: 0.805, M: 0.108]] and the number of months in acquaintance with friends was, but the results on the first day showed that the number of months in acquaintance with friends was more different under alone condition. The results showed that people with similar premeal appetites (For more details, see the supplementary Table S2) and similar liking rates for food (For more details, see the supplementary Table S2) had similar positive moods on any given day, socially or alone. Only on the third day of the social condition, the more similar these three indicators were, the more similar their energy intake was (In turn, [G: 0.750], [G: 0.75], [G: 0.494, M: 0.052]).

Regarding meal duration: Regardless of the day, social condition, or alone condition, the positive mood of people with similar meal duration was more similar (for more details please see supplementary Table S2); The results of the first two days of the analysis showed that under social condition, people with the more similar meal duration had the more similar degree about how close they feel to the meal partner (Day1: [G: 0.598, M: 0.121], Day 2: [G: 0.805, M: 0.108]).

For more details, see Fig. 3 and Table S2 (Supplementary file).

Fig. 3.

Geary’s C and Moran’s I of autocorrelation analysis between similarity matrices of each of three consecutive days and continuous variables under social or alone condition. The circle represents the values of the two exponents in the social condition, and the triangle represents the values of the Moran’s I and Geary’s C in the alone condition. A solid circle or solid triangle indicates the value of Geary’s C, a hollow circle or hollow triangle indicates the value of Moran’s I, and a red circle or red triangle indicates that P ≤ 0.05

Test results of the relationship between dietary similarity matrices and continuous variables for breakfast, lunch or dinner for three consecutive days under social or alone condition

Based on the above analysis, we horizontally split each similarity matrix containing three consecutive days of data into the similarity matrix of breakfast, lunch, and dinner, and then performed autocorrelation analysis on it. We can see that the similarity matrix of each dietary indicator had different results in the autocorrelation analysis of three meals a day.

Regarding energy intake: The more similar the energy intake at lunch, whether social or alone, the more similar the meal duration (Social: [G: 0.734, M: 0.177], Alone: [G: 0.727, M: 0.098]); At breakfast, under social condition, the more similar the energy intake, the more similar the degree about how close they feel to the meal partner (G: 0.590, M: 0.166) and the degree about how well they know the meal partner (G: 0.642, M: 0.099).

For premeal appetite, liking rates for food, and positive mood: the results of autocorrelation analysis for the similarity matrix of these three indicators were very similar. Under the social condition, the more similar the value ranges of these three indicators were, the more different the values of “TFEQ_R”, “TFEQ_un” and “TFEQ_em” were (See supplementary Table S3 for more details). No matter for breakfast, lunch, or dinner, the more similar the value ranges of these three indicators were, the more similar the values of “Friendship_close” and “Friendship_howWell” were (See supplementary Table S3 for more details), and the more similar the value ranges of “Liking_allmeals” and “Appetite_pre” were, the more similar the value of “Mood_allmeals” was (See supplementary Table S3 for more details). However, only under social condition, the more similar the value range of indicator “Mood_allmeals” was, the more similar the value of indicator " Liking_allmeals” (Breakfast: [G: 0.683, M: 0.272], Lunch: [G: 0.746, M: 0.138], Dinner: [G: 0.573, M: 0.158]) was.

In terms of meal duration: whether it was breakfast, lunch, dinner, social or alone, the more similar the meal duration was, the more similar the value of “Mood_allmeals” was (See supplementary Table S3 for more details); No matter for breakfast, lunch or dinner, the more similar the meal duration was under social condition, the more similar the values in “Friendship_close” (Breakfast: [M: 0.128], Lunch: [G: 0.571, M: 0.127], Dinner: [G: 0.592, M: 0.084]), “Friendship_howWell” (Breakfast: [M: 0.100], Lunch: [G: 0.572, M: 0.063], Dinner: [G: 0.609, M: 0.069]) and “Liking_allmeals” (Breakfast: [G: 0.683, M: 0.272], Lunch: [G: 0.746, M: 0.138], Dinner: [G: 0.573, M: 0.158]) were.

For more details, see Fig. 4 and Table S3 (Supplementary file).

Fig. 4.

Geary’s C and Moran’s I of autocorrelation analysis between similarity matrices of breakfast, lunch or dinner of three consecutive days and continuous variables under social or alone condition. The circle represents the values of the two exponents in the social condition, and the triangle represents the values of Moran’s I and Geary’s C in the alone condition. A solid circle or solid triangle indicates the value of Geary’s C, a hollow circle or hollow triangle indicates the value of Moran’s I, and a red circle or red triangle indicates that P ≤ 0.05

Effects of social context and meal type on caloric intake: a multilevel modeling analysis

The model demonstrated good explanatory power, with fixed effects accounting for 50.4% of variance (marginal R² = 0.504) and the full model explaining 65.8% of variance (conditional R² = 0.658). The intraclass correlation coefficient (ICC = 0.311) indicated that 31.1% of the variance in caloric intake was attributable to individual and dyad-level factors, confirming the appropriateness of the multilevel modeling approach.Meal type effects were highly significant, with calorie intake at lunch and dinner significantly higher than breakfast.

Compared to breakfast, lunch consumption was associated with 370.88 kcal higher intake (95% CI: 318.68–423.08.68.08, P < 0.001), while dinner consumption showed an even greater increase of 491.58 kcal (95% CI: 439.38–543.78.38.78, P < 0.001).

By utilizing the same data foundation, MLM results are directly comparable with social network analysis findings. MLM showed that the main effect of social context was not statistically significant (β = −29.60, SE = 26.56, P = 0.266), while social network analysis revealed structural patterns of behavioral similarity. This discrepancy reflects the different levels examined by each method: MLM tests mean differences, while network analysis detects coordination patterns.

For more details, see Fig. 5; Table 3.

Fig. 5.

Comprehensive diagnostic plots for the multilevel linear model examining social context effects on caloric intake. Top-left: Residuals versus fitted values assessing homoscedasticity. Top-right: Q-Q plot evaluating normality of residuals. Bottom-left: Normality of random effects at dyad and individual levels. Bottom-right: Multicollinearity assessment showing variance inflation factors (VIFs) for fixed effects

Table 3.

The results of multilevel linear modeling

Predictors	Null model			Main effects model			Interaction model
	Estimates	CI	P	Estimates	CI	P	Estimates	CI		P
(Intercept)	779.29	722.21–836.38	< 0.001	531.07	468.29–593.85	< 0.001	518.84	452.55–585.13	< 0.001
Condition [alone]				−54.06	−60.32	< 0.001	−29.6	−104.41	0.266
Meal [Lunch]				363.42	326.49–400.36	< 0.001	370.88	318.68–423.08	< 0.001
Meal [Dinner]				462.34	425.41–499.28	< 0.001	491.58	439.38–543.78	< 0.001
Condition [alone] × Meal [Lunch]							−14.92	−147.65	0.691
Condition [alone] × Meal [Dinner]							−58.48	–132.30–15.34	0.12
Random effects
σ 2	69981.75			27557.25			27518.56
τ 00	2029.55 PersonlD: DyadID			4386.46PersonlD: DyadID			4388.61PersonlD: DyadID
	8012.30 DyadID			8012.36DyadID			8012.36DyadID
ICC	0.13			0.31			0.31
N	26 PersonID			26PersonID			26PersonID
	13 DyadID			13DyadID			13DyadlD
Observations	468			468			468
MarginalR2/conditionalR2	0.000/0.125			0.502/0.657			0.504/0.658

Discussion

This study provides comprehensive evidence to quantify assimilation effects in dietary indicators across three consecutive days, comparing social and solitary meal contexts through whole network autocorrelation analysis and multilevel linear modeling using the same data foundation. The analysis identified distinct promotive and inhibitory factors influencing these effects, while the MLM analysis revealed significant effects of meal type but no significant main effect of social context on caloric intake. Crucially, assimilation effects were significantly stronger during social eating than during solitary consumption. These findings suggest that strategically targeting modifiable factors could mitigate chronic diseases associated with the social facilitation of eating, while also harnessing its potential for therapeutic applications in psychological disorders.

Specifically, when dietary indicators of any two individuals fell within the same range, this assimilation was associated with similar values in three key variables: familiarity with the meal partner, perceived closeness to the partner, and the amount of vigorous activity undertaken the preceding day. Furthermore, similarity in energy intake between individuals correlated with similarity in premeal appetite, food liking ratings, positive mood state, and meal duration. The distinct nature of these associations within social eating contexts implies the presence of specific influencing factors and social facilitation effects that enhance dietary indicator assimilation. Conversely, during solitary eating, similarity in energy intake was primarily linked to comparable meal duration. Additionally, if other dietary indicators of two individuals fall within the same range, it may result from differences in their scores on uncontrolled eating and the duration of acquaintance with their friends who participated in the study simultaneously. This meant that these two indicators inhibited the assimilation effects of the dietary indicators.

Nevertheless, social eating consistently promotes assimilation effects across various dietary indicators over three consecutive days, whereas the influence of exercise remains inconsistent. Crucially, if dietary indicators the two participants who eat together align within the same range, this is still largely associated with how well they know their meal partner and how close they feel to them. This partner-driven effect peaked on Day 2 but became nearly imperceptible by Day 3. This attenuation likely reflects the diminishing impact of the meal partner on assimilation after sustained interaction. Supporting this, research suggests socially consumed food may be less satiating than food eaten alone [22]. Notably, by Day 3 of sustained social eating with the same partner, dietary alignment between any two individuals was primarily associated with the similarity in their “Liking allmeals” and “Mood allmeals” — mirroring patterns observed when eating alone. In summary, while social eating initially promotes assimilation driven by partner familiarity, this effect wanes over consecutive days. Concurrently, the influence of individual food preferences and mood on dietary alignment strengthens, eventually resembling solitary eating patterns by Day 3.

McAlpine et al. [22] discovered that eating socially reduces hunger comparably to eating alone, despite promoting larger meal sizes. Our study further indicates that hunger exerted only a minor influence on the assimilation effect of energy intake. Specifically, only socially consumed lunches promoted energy intake assimilation. The assimilation effects promoted by social facilitation of eating occurred whether it was breakfast, lunch, or dinner. When people eat socially, if any two people’s dietary indicators (excluding energy intake indicators) were in the same value range at breakfast, lunch, or dinner, it was still related to the similar degree about how well they know the meal partner and how close they feel to the meal partner. We also discovered that the indicators “TFEQ R,” “TFEQ un,” and “TFEQ em” inhibited or promoted the assimilation effects of dietary indicators at breakfast and dinner, but not at lunch. This finding aligns with the hypothesis that individuals with higher levels of dietary restraint may be more susceptible to social facilitation effects, as social eating might similarly impair cognitive controls over intake. Vartanian and colleagues [23] reported that restrained eaters were more responsive than unrestrained eaters to high-intake social norms: restrained eaters were more likely to model this high intake norm while eating with a confederate who ate a big amount than did unrestrained eaters. Several studies, however, have found no evidence that dietary restraint moderates social facilitation of eating [24–26]. Consistent with these null findings, our study found that food restriction did not promote the assimilation effect of energy intake, although it did promote assimilation effects for other diet-related indicators. These contrasting results regarding the role of dietary restraint—particularly its apparent lack of influence on social facilitation of energy intake compared to other external influences—highlight the need for further research to clarify the underlying mechanisms.

Evidence from the eating behavior literature indicates that humans exhibit limited ability to compensate for changes in energy intake [27]. For example, participants who skipped breakfast failed to adequately compensate for the resulting energy deficit at subsequent meals [28]. Similarly, participants overfed by 133% of their usual caloric intake over two weeks did not subsequently reduce consumption below baseline levels in the following weeks [29]. In contrast, our study observed that people engaged in social eating had a temporary increase in the amount of time they exercised, and time spent on vigorous activities the previous day promoted the assimilation effects of dietary indicators at some meals of the day. However, this effect was inconsistent, likely due to limited activity data and the sporadic nature of human behavior. Given the established failure to fully compensate for excess caloric intake, sustained social eating may contribute to a net increase in energy consumption and potential weight gain. Consequently, long-term studies investigating the effects of social eating on energy balance hold significant implications for dietary interventions. Importantly, since social eating may enhance overall well-being and foster social cohesion [4], completely avoiding social meals is neither practical nor advisable.

To mitigate the impact of social eating on total calorie intake, strategies could include encouraging individuals to intentionally consume smaller meals either before or after a social meal, enhancing awareness of dietary management, and fostering consistent and healthy exercise habits. Physical inactivity is a major modifiable risk factor for non-communicable diseases and mental health conditions [30]. The WHO Global Action Plan on Physical Activity 2018–30 outlines evidence-based policies to increase population activity levels [30]. Increased physical activity effectively reduces the incidence and prevalence of these conditions, thereby lowering healthcare costs and improving well-being [31]. A new study shows that the total cost of physical inactivity globally would be approximately INT$ 520 billion over an 11-year period (2020–30) if global levels of physical activity are not increased [32]. Therefore, promoting awareness of the benefits of regular exercise may help offset excess energy intake during social occasions while concurrently reducing the risk of non-communicable diseases. Leveraging the assimilation effect inherent in social eating contexts presents a promising approach to fostering the healthy lifestyle.

Especially, the analysis revealed breakfast as the meal period most influenced by social facilitation dynamics, where interpersonal familiarity and perceived closeness exerted their strongest effects on dietary assimilation. While John de Castro persists that increased intake during social meals stems from extended meal duration [33] — a view supported by observed positive associations between group size, food intake, and meal duration [24, 34] Despite these findings, other research has demonstrated that longer meal duration is neither necessary nor sufficient for social facilitation to occur [13, 35]. Instead, they tend to attribute people’s food intake to internal factors such as hunger or a “liking” for the food, despite despite weak correlations between these factors and actual food intake [36].

Moreover, our analysis revealed that for three consecutive days of either social eating or eating alone, the assimilating effect of meal duration on energy intake manifested only on the third day or at lunch. Conversely, the assimilating effect of energy intake on meal duration was observed exclusively on the third day or during dinner in social eating contexts. Specifically, if two individuals exhibited similar energy intake levels on the third day, this could be attributed to their comparable time spent on light activities the previous day and similar meal durations. Additionally, the findings indicated that regardless of whether eating alone or socially, the energy intake of any two individuals within the same value range correlated with a similar meal duration at lunch but not at breakfast or dinner. One plausible explanation is that on day 3, the impacts of internal factors, such as meal duration and positive mood, began to outweigh the effects of external factors, such as the lunch companion factor, or the availability of free food in the lab. In addition, psychological stress caused by unfinished tasks or limited time for planning may hinder individuals from fully enjoying their lunch or engaging with companions, thereby diminishing the effects of social eating.

Consistent with prior research, individuals who ate with others consumed significantly larger meals compared to those who ate alone. Notably, evidence for the social facilitation of eating was found for meals consumed with and without alcohol, throughout the week and on the weekends, and for all meal kinds (breakfast, lunch, supper, and snacks) [37, 38]. In contrast, we found that the factors behind the social convenience of eating at different meals of the day were not the same, especially the social convenience of eating at lunch. What’s more noteworthy is that, if any two people were in the same energy intake range on the social diet, it might be related to their similar BMIs, and this feature only showed up on the third day of the three consecutive days’ diet or at dinner. When this was investigated further, it was discovered that if the individuals’ BMIs were normal, and any two people were in the same range for all four dietary indicators other than energy intake during social eating, it could be due to their similar energy intake, and more importantly, this trait was only reflected on the third day of the three consecutive days’ diet or at dinner.

Social dining enhances hedonic experiences, correlating with increased consumption. Empirical evidence indicates commensality events associate with elevated positive affect [32] and proportionally expanded meal sizes during positive emotional states [39]. Food tastes better when eaten socially is further evidenced. In one study, chocolate was rated as more palatable when it was consumed with another person [40]. Notwithstanding these findings, after controlling for group size, one self-report study discovered that positive mood ratings did not predict additional variation in food intake [33]. Consistent with this, our analysis indicates positive mood does not modulate energy intake assimilation, suggesting it is not the primary mechanism driving socially facilitated consumption. However, whether eating socially or alone, positive mood promoted assimilative effects on pre-meal appetite, liking rates for food, and meal duration, regardless of any meal of the day or any day of the three days. We can make reasonable use of these assimilation effects to actively guide the establishment of a healthy dietary norm and a social environment to enjoy food happily and alleviate the symptoms of appetite loss or anorexia caused by various psychological diseases in society. Nevertheless, experimental research is needed to clarify any causal associations between the presence of other people, meal enjoyment, and food intake, under controlled settings.

Social facilitation of eating is driven by culturally agreed norms to eat more when eating with other people. Several studies indicate that people underestimate how much their food intake is impacted by other people’s [36]. If social eating contributes to overeating over the longer term, then a further concern is that misattributing overeating and weight gain to internal factors (such as a lack of self-control), rather than to external/social factors, may cause people to experience self-blame for overeating and increase internalized weight stigma amongst people with overweight [15]. Robust evidence confirms social norms have a strong influence on people’s decisions about what and how much to eat [41]. The assimilation effect of social eating, which indirectly reflects the social dietary norm, further supports this view. While chronic disease risks from sustained energy surplus warrant concern, the mental health benefits of social dining demand equal consideration. Multiple studies have found there are strong associations between having multimorbidity and poorer mental and physical health outcomes [42, 43], and a direct association between multiple physical chronic illnesses and subsequent common mental disorders [44–46]. Instead of just looking at the negative effects of the social facilitation of eating through increased energy intake on various chronic diseases, we have to recognize the positive effects of the social facilitation of eating on improving people’s positive emotions and therefore their mental health. Social eating can be a natural health intervention based on social norms of healthy eating and regular exercise. In fact, there is growing evidence that nature-based health interventions can transform both mental and physical health [47, 48]. Strategically harnessing its assimilation effects may foster societal physical and mental health through norm-based promotion of healthy eating and exercise practices.

The discrepant findings between MLM and WNA carry important theoretical implications: While MLM did not detect a significant main effect of social context, WNA revealed significant behavioral similarity patterns. Using the same 468 observations ensured that result differences stemmed from methodological perspectives rather than data foundation discrepancies. This design allows us to confidently conclude that social context influences manifest more through structural patterns of behavioral coordination than through mean differences at the individual level (tested by MLM). The combined results from both methods suggest that social influence on eating behavior operates primarily through coordination mechanisms rather than average level differences. Participants exhibited greater behavioral similarity in social contexts, but this coordination does not necessarily translate into significant changes in overall intake levels.

This study provides a novel perspective on the social facilitation of eating by examining its role in promoting dietary assimilation effects. Only female participants were recruited to reduce error variance related to sex/gender differences in amounts consumed, and because women eating with men is associated with reduced intake due to impression management concerns [30]. Moving beyond conventional methods limited by aggregate data, we employed whole-network autocorrelation analysis to elucidate interdependencies between interpersonal relationship matrices and multidimensional dietary parameters. Given that social eating constitutes a form of interpersonal interaction, this analytical approach represents a novel and feasible strategy for investigating assimilation within this context. Nevertheless, several limitations warrant consideration. First, reliance on experimental data inherently restricts the generalizability of findings to the full spectrum of real-world social versus solitary eating scenarios. Second, while the study revealed that assimilation effects varied with the number of consecutive social meals and differed across daily meals (breakfast, lunch, dinner), the experimental design—notably involving repeated meals with the same gender-matched partner over three consecutive days—deviates significantly from typical social eating patterns (e.g., with family or romantic partners). This design limitation likely impacted the expression of social facilitation effects and complicates inferences about continuous social eating’s impact on dietary indicators. Third, although this study confirmed the existence of assimilation effects driven by social facilitation for various dietary indicators, it could not quantify the magnitude of this facilitation (“promotion force”) nor the specific quantitative contributions of individual influencing factors under social conditions. Finally, while a transient increase in exercise duration was observed among social eaters, the underlying cause—whether compensatory for higher energy intake or driven by elevated positive mood—remains unclear. Moreover, the limited data on physical activity precludes conclusions about how varying activity levels during social eating might promote or inhibit dietary assimilation effects; this constitutes a critical avenue for further research.

Conclusions

This study provides compelling evidence that social facilitation of eating significantly promotes assimilation effects across various dietary indicators, with the strength of this influence varying systematically across daily meals and between social contexts. The apparent contradiction from the results of MLM with our network findings carries important theoretical implications: social influence appears to operate primarily through behavioral coordination mechanisms rather than simple mean-level changes.

The convergence of evidence from both methodological approaches suggests that assimilation effects represent a form of emergent social dietary norms, reflecting the practical implementation of social norms within specific populations. Crucially, social facilitation of eating promotes these effects through subtle behavioral synchronization that may not manifest as overall consumption increases but rather as coordinated patterning across individuals. Moving forward, we recommend that future research: (1) employ multidimensional assessment frameworks integrating network analysis with traditional statistical methods; (2) prioritize experimental designs that enhance ecological validity while maintaining methodological rigor; and (3) explore practical applications of these findings to promote healthy eating norms within social contexts. By strategically leveraging social facilitation mechanisms, public health initiatives can cultivate positive dietary assimilation effects while advocating for balanced nutrition and regular physical activity patterns that propagate through social networks.

Author contributions

QW conceived the study together with LW and ZM. QW, LW, and XL provided input on the study process. LW and XL extracted the data, assisting in manuscript drafting under QW’s guidance. QW, XL, and YZ performed the analyses. ZM and YC provided expert input over the review process and final drafting. The authors read and approved the final manuscript.

Funding

Our work was supported by the National Natural Science Foundation of China (project no.71673126), the Key Project of Guangdong Institute of Health Economics, China (project no.2022-WJZD03).

Data availability

The data supporting this study’s findings are openly available in [UK Data Service] at https://reshare.ukdataservice.ac.uk/854464/, reference number [22].

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.