Peer Effects and Rural Households’ Online Shopping Behavior: Evidence from China

Abstract

Amid the rapid expansion of the digital economy, online shopping has become increasingly common among rural households in China, yet the social interaction mechanisms driving such behavior remain insufficiently explored. This study examines the impact of peer effects on farmers’ online shopping behavior using data from the China Family Panel Studies (CFPS) covering the years 2014 to 2022. A Logit model is applied to estimate peer influence, and interaction terms are introduced to assess the moderating roles of land assets and social expenditures. The results reveal that peer behavior significantly increases the likelihood of rural households participating in online shopping, with the effect being particularly strong among low-income, less-educated households and those in western regions. Additionally, both land-rich households and those with higher social expenditures demonstrate greater responsiveness to peer influence. These findings highlight the importance of local social interaction in shaping rural online shopping behavior and provide theoretical and practical implications for digital inclusion and rural e-commerce strategies.

1. Introduction

Rural e-commerce, propelled by The Rural Vitalization Strategy and Plan for the Overall Layout of Building a Digital China, has emerged as a transformative force in reconfiguring the urban–rural economic nexus and stimulating domestic consumption [1]. Over the past decade, China has made substantial strides in rural infrastructure development, achieving near-universal access to electricity, telecommunications, cable television, and paved transportation networks across administrative villages. According to the 55th Statistical Report on China’s Internet Development published by the China Internet Network Information Center (CNNIC), as of December 2024, the national internet user base reached 1.108 billion, with an internet penetration rate of 78.6%; notably, 313 million of these users resided in rural areas, accounting for 28.2% of the total. Simultaneously, the hierarchical e-commerce and logistics system spanning counties, townships, and villages has evolved into a highly functional operational framework. According to the State Post Bureau, over 337,800 village-level logistics service stations have been established, ensuring effective last-mile delivery coverage; more than 100 million parcels now circulate daily between rural and urban areas, reflecting the growing institutionalization and normalization of digital consumption in rural settings. Distinct from many Western contexts, China’s e-commerce ecosystem is shaped by a unique interplay of institutional innovation and socio-cultural practices [2]. The proliferation of mobile payment systems, the emergence of livestream-based commerce, and the rapid development of village-level logistics have transformed online shopping into a socially embedded and routinized form of consumption. Rural consumers frequently rely on social recommendations and livestream demonstrations in place of in-person evaluation, while collective practices such as group purchasing and community-based parcel collection further underscore the institutionalized nature of digital consumption in rural society.

However, despite these infrastructural achievements, rural households continue to face substantial obstacles in accessing and utilizing e-commerce platforms, particularly when compared with their urban counterparts. These barriers include limited access to timely and reliable information, low levels of digital literacy, and a pronounced aversion to consumption-related risks [2]. Moreover, rural purchasing decisions are not made in isolation but are embedded in localized social networks and further constrained by household-specific resource endowments. Consequently, these realities signal the need to reorient academic inquiry from focusing solely on infrastructural provision toward exploring behavioral dynamics and the role of social interaction in facilitating inclusive digital participation.

In response to this shift, an emerging body of literature has increasingly emphasized the role of sociocultural factors in shaping economic behavior [3]. This perspective is particularly relevant in rural contexts, where ostensibly individual consumption behaviors, such as online shopping, are deeply situated within collective social structures. Through mechanisms such as social learning, normative pressures, and behavioral imitation, peer effects emerge as a critical driver of e-commerce [4]. Evidence from diverse domains lends support to this framework. For instance, in corporate finance, peer effects amplify exogenous shocks in capital structure, accounting for over 70% of variations in leverage [5]. Similarly, in green finance, peer dynamics have been shown to influence ESG disclosure behaviors [6]. In rural China, the dense web of interpersonal ties and strong localized trust provide fertile ground for such social spillovers. Empirical studies have linked peer effects to key rural behaviors, including land transfer [7], technology adoption [8], and climate adaptation [9,10]. In household consumption, social comparison and “face-based consumption competition” fuel imitation, especially those related to education and durable goods [11,12]. Nevertheless, despite the theoretical relevance of peer effects, two critical research gaps remain. First, the existing literature disproportionately focuses on urban populations or specific product categories [13,14,15]. Even when rural populations are considered, the focus tends to remain narrowly centered on agricultural technology adoption, thereby overlooking the broader landscape of rural E-commerce consumption [16,17,18,19]. Second, although prior research has examined how individual-level characteristics moderate peer effects, scant attention has been paid to the mediating influence of household asset configurations [20,21,22]. In fact, asset endowments, such as secure land tenure, can significantly reshape peer dynamics by influencing households’ risk preferences, information processing capabilities, and reliance on social capital [23,24,25,26].

Manski’s [27] theoretical framework provides a robust foundation for explicating peer effects in rural online shopping. In contexts characterized by uncertainty, farmers often infer behavioral norms by observing their peers and subsequently recalibrating their own expectations [28]. Simultaneously, informal norms and culturally embedded values can recalibrate individual preferences once widely diffused [29]. As a result, even seemingly private consumption choices, such as online shopping, are shaped by collective behavioral trends within villages. Importantly, the magnitude and direction of these peer effects are not uniform; rather, they vary depending on household-level factors, such as asset ownership and the strength of social ties. For instance, secure land tenure has been found to increase households’ risk tolerance and thereby promote consumption [30,31,32], while imperfect land transfer markets may impede liquidity and suppress consumption potential [33]. Similarly, personal relationship-based social expenditures, such as gift-giving, reinforce reciprocal trust and intensify peer effects through repeated social interaction. Despite their theoretical relevance, the moderating roles of household assets and social expenditures in peer-based consumption behavior remain insufficiently theorized. Prevailing studies often operationalize these variables as mere controls or assess only their linear impacts [34,35], overlooking the interaction between social dynamics, resource endowments, and behavioral responses.

Against this backdrop, this study draws on five waves of the China Family Panel Studies (CFPS) from 2014 to 2022 to construct a village-level indicator of peer effects. Using Probit and moderation models, we empirically examine the influence of peer behavior on rural households’ online shopping decisions, as well as the underlying mechanisms and heterogeneity. Furthermore, to address potential endogeneity, we use the logarithm of the average total consumption expenditure of other households in the same village as an instrumental variable. This paper makes three key contributions. First, it bridges the theoretical gap between rural social networks and online shopping by contextualizing peer effects within the digital economy. Second, it incorporates land assets and social expenditures into the moderation analysis, offering novel insights into the resource–behavior nexus. Third, the findings offer policy insights for tailoring rural e-commerce strategies to local conditions, enhancing the effectiveness of digital inclusion efforts.

2. Theoretical Analyses and Research Hypotheses

2.1. Peer Effects

The concept of peer effects originated in social psychology and is aptly encapsulated by the traditional Chinese proverb, “one who stays near vermilion gets stained red, and one who stays near ink gets stained black.” This metaphor underscores the propensity for individual behavior to be shaped by group dynamics. While widely acknowledged in societal contexts, peer effects have also garnered substantial attention in economic domains, where they shape decisions across firms, organizations, and institutional settings. Within corporate and governmental environments, peer influence is reflected in strategic planning, capital allocation, managerial behavior, and risk-taking tendencies [36,37,38]. For example, firms often emulate their industry peers when adopting financialization strategies [39]. Building on this, another study reveals that in highly competitive markets, particularly for firms in weaker positions, peer effects function as crucial external catalysts for innovation [40]. Peer influence is also evident in dividend policy formulation, where the adoption of cash dividend schemes often triggers emulation among firms [41].

At the individual level, scholarly investigations into peer effects have primarily concentrated on four key domains: education, consumption, labor, and employment [42,43,44,45]. In the labor sector, empirical evidence from the China Migrants Dynamic Survey (CMDS) indicates that a one-percentage-point increase in the incidence of overwork within a district leads to a 0.73 percentage-point rise in the probability of individual overwork, underscoring the regulatory function of peer effects in labor governance [46]. Turning to the educational domain, systematically leveraging peer dynamics among school-aged children can significantly enhance their non-cognitive competencies [47]. In the realm of consumption, the “Catching up with Joneses” or “Keeping up with the Joneses” paradigm has been widely adopted to explain heterogeneity in consumption preferences and spending behavior—that is, individuals tend to mimic the consumption behaviors of their peers to maintain perceived social parity, and they estimate that peer effects account for 19.8% of the variation in household consumption expenditure [11]. In addition, further evidence is provided that peer influence exerts a significant positive impact on urban households’ investment in children’s extracurricular education [12].

Within the domain of agricultural decision-making, peer effects have been shown to exert a demonstrative influence on farmers’ production and investment behaviors, particularly among younger cohorts in rural communities [48]. Extending this line of inquiry, significant peer-driven dynamics are also found in farmland transfer decisions among rural households identifies [7,49]. The behavioral mechanism underlying these effects stems from farmers’ propensity to emulate coping strategies adopted by socioeconomically similar peers. These imitation tendencies are particularly salient within cooperatives and community-based organizations, where behavioral reinforcement from proximate actors often acts as a critical catalyst for individual decision-making [8]. Building on this foundation, empirical evidence using micro-level survey data from 497 maize-producing households (2014–2018) is provided, integrated with national meteorological drought monitoring data, to demonstrate that peer effects significantly enhance farmers’ adaptive capacity in response to drought-related stress [9]. Furthermore, recent studies have corroborated the role of peer influence in encouraging farmers’ participation in environmentally sustainable production practices [10], thereby reinforcing its relevance across diverse agricultural contexts. Despite the growing recognition of peer effects in shaping production-related decisions, relatively little scholarly attention has been directed toward their influence on rural consumption behavior. Given that production activities are often a behavioral extension of farmers’ underlying consumption preferences, the role of peer effects in shaping household consumption patterns merits further empirical exploration.

2.2. Peer Effects on Rural Households’ Online Shopping Behavior

Peer effects fundamentally reflect the importance of social interaction in shaping rural households’ consumption decisions. Rather than being purely individualistic, these decisions are embedded in the social context of village life, where individual behavior is significantly influenced by the consumption patterns of surrounding community members [50,51]. Prior research suggests that group behavior can influence individual choices through mechanisms such as information exchange, behavioral imitation, and social normative pressure, thereby producing notable spillover effects [52]. In rural China, tightly knit social networks and frequent interpersonal interactions amplify these dynamics, establishing a robust social foundation for household consumption behavior. When rural consumers encounter e-commerce for the first time, they may face challenges such as a lack of experience, limited digital literacy, or low trust in online platforms. In such contexts, peer effects play a critical role in reducing uncertainty and encouraging initial trial behavior. On one hand, social learning theory posits that farmers acquire relevant consumption information—such as product quality, pricing, and after-sales service—by observing their neighbors’ online shopping experiences, and this process reduces information asymmetries and decision-making costs, thereby enhancing farmers’ willingness to engage in e-commerce activities [53,54,55]. On the other hand, the rural-specific phenomenon of “face-based consumption competition” further incentivizes households to conform to prevailing consumption norms. Such conformity is driven not only by the desire for social inclusion but also by the avoidance of potential reputational loss or normative sanctions associated with non-participation [56,57]. In addition, the diffusion of innovations theory suggests that new behaviors spread progressively through social networks [58], often guided by opinion leaders and early adopters—implying that once some households begin to adopt online shopping, others are likely to follow. This view is further supported by informational cascade theory, which posits that in environments with limited access to reliable information—as is often the case in rural China—individuals tend to imitate the actions of others and base their decisions on observable peer behavior [59]. Based on this reasoning, the first hypothesis is proposed:

H1: 

Peer effects significantly promote rural households’ participation in online shopping consumption.

However, the magnitude of peer effects is heterogeneous across rural households and is conditioned by their underlying resource endowments. Land assets, as one of the most critical forms of rural capital, serve risk-buffering and income-stabilizing functions [60]. Secure land tenure enhances economic resilience and increases households’ capacity to tolerate risk, thereby amplifying their susceptibility to social influence. This effect becomes particularly salient when land-rich households observe neighboring peers deriving tangible benefits from online shopping, which in turn motivates behavioral emulation in pursuit of similar utility gains or social approval [61]. Thus, land assets are expected to strengthen the influence of peer effects. In such cases, households with greater land assets, due to their stronger resource base, are more capable of emulating peer behaviors and are better positioned to absorb potential risks and realize consumption returns. Accordingly, the second hypothesis is formulated:

H2: 

Land assets positively moderate the relationship between peer effects and rural households’ online shopping behavior; greater land holdings correspond to stronger peer effects.

Social expenditure serves as a direct indicator of the extent to which rural households are embedded within traditional acquaintance-based social networks. In rural China’s customary social structure, such expenditures transcend mere cultural or ceremonial practices; they also function as mechanisms to reinforce neighborly ties and sustain social capital [62]. Higher levels of social expenditure often reflect more active participation in communal affairs and denser interpersonal interactions, which, in turn, intensify the operation of peer effect channels such as information exchange, behavioral mimicry, and normative conformity. Specifically, frequent gift exchanges and ritual-based interactions not only deepen emotional bonds but also facilitate sustained interpersonal communication among neighbors. These practices enhance mutual trust and familiarity, thereby increasing individuals’ attentiveness and responsiveness to the opinions, preferences, and consumption behaviors of those around them. Such heightened social sensitivity renders rural residents more prone to normative conformity in their decision-making, including in digital consumption domains such as online shopping. Furthermore, these expenditures are often motivated by social comparison and face-saving concerns, prompting individuals to consume in ways that affirm their social standing within the community [63]. Taken together, these mechanisms suggest that social expenditure could intensify the behavioral impact of peer effect in rural households’ online shopping. Accordingly, this leads to the third hypothesis:

H3: 

Social expenditure positively moderates the relationship between peer influence and rural households’ online shopping behavior; the greater the social expenditure, the stronger the reinforcing effect of peer influence.

In summary, the theoretical framework proposed in this study not only clarifies the mechanisms by which peer effects shape rural households’ online shopping decisions, but also delineates how land assets and social expenditure act as distinct moderating factors within this behavioral process. By integrating social interaction mechanisms with resource heterogeneity, this framework offers a more nuanced understanding of online shopping behavior in rural settings [61,64]. Such an integrative perspective provides a robust conceptual foundation for the empirical analyses that follow.

3. Materials and Methods

3.1. Data Source

The empirical analysis in this study is based on data from the China Family Panel Studies (CFPS), a nationally representative longitudinal survey initiated in 2011. As of the time of this study, seven waves of follow-up surveys have been completed, covering the period from 2014 to 2022. The CFPS spans 25 provinces across China and includes over 34,000 households. It provides rich and detailed information on household demographics, consumption behavior, income levels, and other socio-economic attributes, making it a highly valuable source for empirical research on household behavior. While the use of secondary survey data inevitably limits flexibility in selecting explanatory variables, the CFPS remains one of the most authoritative and reliable datasets for analyzing household decision-making in China. To test the proposed hypotheses, this study utilizes panel data from all five waves of the CFPS between 2014 and 2022, as the CFPS questionnaire only began to include items related to the dependent variable of interest starting in 2014. Several data preprocessing steps were undertaken to ensure consistency and analytical reliability. First, each wave of data was cleaned to address inconsistencies and ensure data quality. Second, the regional-, household-, and individual-level datasets were merged within each wave to construct comprehensive annual cross-sectional files. Third, we identified and retained households that were consistently tracked across the selected waves to construct a balanced panel. Finally, observations with missing or abnormal values were excluded from the sample. Following these steps, the final dataset comprises 8538 valid household-level observations, with a specific focus on rural households.

3.2. Variable Selection

3.2.1. Explained Variable

The dependent variable is online shopping behavior, measured as a binary indicator. It is based on the CFPS question asking whether the household had engaged in any online shopping over the past year. If the household responded affirmatively, the variable is coded as 1; otherwise, it is coded as 0.

3.2.2. Explanatory Variable

The core explanatory variable is peer effects, which are calculated as follows:

$$Pee{r}_{it}={\displaystyle \frac{{{\displaystyle \sum }}_{j\ne i}^{N_c}Onlin{e}_{jt}}{N_c-1}}$$

(1)

Specifically, peer effects are measured as the average online shopping behavior of all other households in the same community and year, excluding the focal household. This variable captures the intensity of group-level influence within localized social networks.

3.2.3. Moderating Variable

The moderating variables include land assets and financial assets. Land assets refer to the estimated market value of the land owned by the household (in RMB). Social expenditure refers to the total annual amount spent by the household on social obligations, such as weddings and funerals. This variable captures the degree of the household’s social embeddedness and participation in traditional interpersonal networks. Both variables are log-transformed to reduce the potential impact of heteroskedasticity and extreme values.

3.2.4. Control Variable

Individual and household-level control variables are selected with reference to similar research [65], encompassing both demographic and economic characteristics. At the individual level, the control variables include age (measured in years), gender (a binary variable, where 1 = male and 0 = female), and years of schooling (measured by the total number of years of formal education). At the household level, per capita household income is calculated by dividing total household income by household size and expressed as a natural logarithm to reduce the effect of heteroskedasticity. Family size is measured by the number of permanent family members. In addition, two demographic indicators are included: the elder dependency ratio, defined as the ratio of family members aged 65 and older to members of working age, and the child dependency ratio, defined as the ratio of family members aged 14 and younger to members of working age. The descriptive statistics for the variables are shown in

Table 1. Descriptive statistics of variables.

Variable	Code	Average Value	Standard Deviation	Minimum Value	Maximum Values	Observation
Explained Variable
Online Shopping	OS	0.38	0.283	0	1	8538
Explanatory Variable
Peer Effect	Peer	0.471	0.499	0	1	8538
Moderating Variable
Land Asset	Land	6.938	4.803	0	15.43	8538
Social Expenditure	Soc	7.43	2.134	0	11.85	8538
Control Variable
Age	Age	41.301	11.571	15	84	8538
Education	Educ	0.566	0.496	0	1	8538
Gender	Gender	8.477	3.901	0	19	8538
Income	Income	9.645	0.943	0	14.8	8538
Household Size	Fml	4.139	1.855	1	15	8538
Elder Dependency Ratio	Elder	0.154	0.343	0	3	8538
Child Dependency Ratio	Child	0.378	0.489	0	5	8538

.

3.3. Model Specifications

3.3.1. Benchmark Model

Based on the results of the Hausman test, and given that the dependent variable in this study is binary (whether a household engages in online shopping), nonlinear estimation is more appropriate than a linear probability model. Therefore, this study employs a Logit model with year and province fixed effects to estimate the impact of peer effects on farmers’ online shopping behavior. To account for temporal variation and nationwide trends in online shopping adoption, all regression models include year fixed effects (i. year). This ensures that the estimated peer effects are net of any time-specific shocks or macro-level changes between 2014 and 2022. The model is specified as follows:

$$O{S}_{it}=\alpha +\beta Pee{r}_{it}+\gamma Contro{l}_{it}+{\mu }_i+{ϵ}_{it}$$

(2)

where $O{S}_{it}$ denotes the binary outcome indicating whether household i engaged in online shopping in year t; $Pee{r}_{it}$ is the core explanatory variable capturing peer effects at the village level; $Control{s}_{it}$ is a vector of control variables; ${\mu }_i$ represents unobserved household-specific random effects; and ${ϵ}_{it}$ is the idiosyncratic error term.

3.3.2. Intermediary Effect Model Specification

To assess the potential moderating roles of land and financial assets, the baseline model is extended by incorporating interaction terms between peer effects and each type of asset. The augmented model is formulated as follows:

$$O{S}_{it}=\alpha +{\beta }_{1}Pee{r}_{it}+{\beta }_2\left(Lan{d}_{it}\times Pee{r}_{it}\right)+\gamma Contro{l}_{it}+{\mu }_i+{ϵ}_{it}$$

(3)

$$O{S}_{it}=\alpha +{\beta }_{1}Pee{r}_{it}+{\beta }_2\left(So{c}_{it}\times Pee{r}_{it}\right)+\gamma Control{s}_{it}+{\mu }_i+{ϵ}_{it}$$

(4)

where ${\mathrm{Land}}_{\mathrm{it}}$ and $So{c}_{it}$ refer to land assets and financial assets, respectively, both included in logarithmic form to reduce heteroskedasticity. The interaction terms $Lan{d}_{it}\times Pee{r}_{it}$ and $So{c}_{it}\times Pee{r}_{it}$ are used to identify the respective moderating effects of the two types of assets on the relationship between peer influence and online shopping behavior.

4. Empirical Analysis

4.1. Benchmark Regression Results

Table 2. Results of Baseline Regression Model.

Variable	OS
	(1)	(2)	(3)	(4)
Peer	0.300 ***	0.752 ***	0.473 ***	0.352 ***
	(0.077)	(0.086)	(0.089)	(0.092)
Age		−0.049 ***	−0.060 ***	−0.061 ***
		(0.002)	(0.002)	(0.003)
Gender		−0.295 ***	−0.291 ***	−0.288 ***
		(0.048)	(0.049)	(0.051)
Educ		0.086 ***	0.092 ***	0.090 ***
		(0.007)	(0.007)	(0.007)
Income		0.403 ***	0.335 ***	0.283 ***
		(0.031)	(0.031)	(0.032)
Fml		0.005	0.004	0.001
		(0.015)	(0.015)	(0.015)
Elder		0.175 **	0.106	0.102
		(0.072)	(0.074)	(0.075)
Child		0.328 ***	0.211 ***	0.200 ***
		(0.055)	(0.057)	(0.058)
Cons	−0.230 ***	−3.017 ***	−2.822 ***	−1.866 ***
	(0.036)	(0.323)	(0.333)	(0.515)
Control Variable	NO	YES	YES	YES
Province Fixed	NO	NO	NO	YES
Year Fixed	NO	NO	YES	YES
Pseudo R2	0.0013	0.1155	0.1429	0.1513
Observations	8538	8538	8538	8538

Note: The standard errors are displayed in parentheses, with **, and *** denoting significance at the 5% and 1% levels, respectively.

presents the baseline regression results. Columns (1) through (4) sequentially report estimates from models without control variables, with control variables, with time fixed effects, and with both time and provincial fixed effects. The results consistently demonstrate a significant positive impact of peer effects on rural household online shopping behavior. In the baseline Logit model reported in Column (1), the coefficient of the peer effects variable is 0.300. After adding individual and household-level control variables in Column (2), the coefficient increases markedly to 0.752, suggesting that omitting relevant covariates may lead to a substantial underestimation of peer effects. When controlling for time fixed effects in Column (3), the coefficient declines to 0.473, and further drops to 0.352 after accounting for both time and provincial fixed effects in Column (4). Despite these adjustments, the peer effects remain statistically significant, thereby confirming the robustness of the relationship. Among the control variables, the household head’s gender and income show expected signs. Female-headed households are more likely to engage in online shopping, which may reflect greater online shopping involvement among women in rural areas. Higher household income is also positively associated with online shopping participation, indicating that economic resources play a facilitating role in online shopping behavior. The effect of the elderly dependency ratio is only marginally significant, potentially due to variations in consumption structure across different household types.

4.2. Endogeneity Test

When examining the impact of peer effects on farmers’ online shopping behavior, it is important to note that the peer effect is essentially an aggregate explanatory variable, constructed as the average behavior of other farmers within the same village and year. Given the mutually influential nature of individual behavior, a potential issue of reverse causality may arise, that is, one household’s online shopping behavior could in turn affect the behavior of other households in the same village. In addition, there may be omitted variable bias or unobserved shared social and cultural contexts, which could correlate the explanatory variable with the error term, thereby introducing endogeneity [66]. To address these concerns, this study adopts a Two-Stage Residual Inclusion (2SRI) method, using the logarithm of the average total consumption expenditure of other households in the same village as an instrumental variable [67]. On the one hand, this variable captures the general consumption environment and exposure to market behavior within the village, and is thus strongly correlated with peer effects. On the other hand, as a village-level aggregate that excludes the focal household, it is unlikely to directly influence that household’s own online shopping behavior once individual-level covariates (e.g., income, education, and age) are controlled for. This satisfies both the instrument relevance and exclusion restriction assumptions [68].

In the first-stage regression, the instrumental variable is found to be significantly associated with the peer effect variable, and the resulting residual is statistically significant. In the second-stage Logit estimation, as shown in

Table 3. Results of Endogeneity Analysis.

Variable	OS
Peer	0.609 **
	(0.242)
Cons	−1.816 ***
	(0.517)
Control Variable	YES
Province Fixed	YES
Year Fixed	YES
Pseudo R2	0.1514
Observations	8538

Note: The standard errors are displayed in parentheses, with **, and *** denoting significance at the 5% and 1% levels, respectively.

, the inclusion of the residual term confirms the robustness of the results: the coefficient on peer effect remains positive and statistically significant at the 1% level (coefficient = 0.609, p = 0.009). Furthermore, the McFadden pseudo-R² of the model is 0.1514, indicating a reasonably good model fit. These findings affirm the robustness of the estimated relationship between peer effects and rural households’ online shopping behavior, even after accounting for potential endogeneity.

4.3. Robustness Check

The results from the benchmark regressions and the endogeneity test provide consistent support for Hypothesis 1, confirming the positive influence of peer effects on rural households’ online shopping behavior. Nevertheless, concerns may still arise regarding potential estimation biases stemming from sample selection, variable specification, and model choice. To ensure the robustness of our findings, we conducted several robustness checks, as presented in

Table 4. Results of Robustness Check.

Variable	OS
	(1)	(2)	(3)	(4)
Peer	0.545 ***	0.363 ***		0.649 ***
	(0.061)	(0.092)		(0.150)
Peer median			0.682 ***
			(0.052)
Control Variable	YES	YES	YES	YES
Province Fixed	YES	YES	YES	YES
Year Fixed	YES	YES	YES	YES
Pseudo R2		0.1539	0.1676	0.1418
Observations	8538	8538	8538	5805

Note: The standard errors are displayed in parentheses, with *** denoting significance at the 1% level, respectively.

.

First, to evaluate the sensitivity of the results to distributional assumptions, we re-estimate the baseline model using a panel Probit specification, as shown in Column (1) of Table 4. While the Logit model assumes a logistic distribution for the error term, the Probit model assumes a standard normal distribution. The estimated coefficient of Peer is 0.545 and remains positive and statistically significant at the 1% level, which is consistent with the baseline Logit results. Although the magnitude is slightly smaller, this difference reflects the inherent scale variation between the Logit and Probit models and does not compromise the substantive conclusion. These results confirm that the positive influence of peer effects on rural online shopping behavior is robust to alternative model specifications.

Second, to mitigate the influence of extreme values, which potentially arise from measurement errors or outlier consumption behavior, we applied a 1% winsorization to all continuous control variables. Column (2) reports the results after trimming values below the 1st and above the 99th percentiles. The coefficient of Peer increases to 0.363 and remains statistically significant at the 1% level, suggesting that the removal of extreme values enhances the precision and reliability of the estimates. This finding implies that outliers may obscure the true strength of peer effects and that a trimmed sample provides a clearer view of underlying behavioral patterns.

Third, we assess the robustness of the peer effect measurement by replacing the mean-based construction with a median-based alternative. While the median may not fully capture central tendency in highly skewed distributions, its resistance to outliers can enhance the reliability of group-level behavioral indicators. As shown in Column (3), the coefficient of Peer median is 0.682, statistically significant at the 1% level, and directionally consistent with the baseline estimate. This indicates that our findings are not sensitive to the specific method used to operationalize peer effects.

Lastly, to address the potential bias introduced by small-sample communities, we excluded villages with fewer than five observations from the analysis. Small communities may exhibit fragmented social networks or lack internal cohesion, which could distort the measurement of peer effects. After removing these observations, as shown in Column (4), the coefficient of Peer increases to 0.649 and remains significant at the 1% level. This adjustment improves the sample’s representativeness and the model’s ability to capture typical peer dynamics, thereby reinforcing the explanatory power of the peer effect mechanism in shaping rural online shopping behavior.

4.4. Moderating Effects Analysis

Social expenditures, referring to the monetary outlays that rural households allocate for social events, such as weddings, funerals, and festive gatherings, serve as an important proxy for social embeddedness and the intensity of interpersonal interactions [69,70]. Households with higher levels of social spending are typically more deeply embedded in local social networks, maintain more frequent contact with neighbors, and actively participate in village affairs. More critically, such households tend to place greater value on social reputation and collective identity, making them more susceptible to peer pressure and more likely to engage in imitative consumption behavior.

Table 5. Results of the moderating effect model.

Variable	OS
	(1)	(2)
	Land	Soc
Peer × Soc	0.029 ***
	(0.009)
Peer × Land		0.072 ***
		(0.011)
Control Variable	YES	YES
Province Fixed	YES	YES
Year Fixed	YES	YES
Pseudo R2	0.1509	0.1443
Observations	8538	8538

Note: The standard errors are displayed in parentheses, with *** denoting significance at the 1% level, respectively.

presents the moderating effects of household asset variables. In Column (1), the interaction term between peer effects and social expenditures yields a coefficient of 0.072, which is statistically significant. This result indicates that higher social spending significantly amplifies the influence of peer effects on online shopping decisions. In essence, households that are more socially engaged exhibit a heightened sensitivity to social cues, likely due to their stronger desire for affiliation and conformity, which in turn reinforces their tendency to emulate the consumption behavior of others in the community.

Land assets, in contrast, are fundamental productive resources in rural households and play a vital role in providing economic security and mitigating financial risks [30,31]. A greater volume of land assets generally reflects higher household wealth and stronger expectations for future returns, making such households more inclined to replicate behaviors observed among their peers, particularly when such behaviors are perceived to be beneficial [32,33]. From the perspective of information economics, land-rich households possess not only a more stable economic base but also a greater ability to absorb potential losses. This reduces the perceived uncertainty associated with online shopping and increases their openness to peer-driven behavioral shifts. As reported in Column (2) of Table 5, the interaction term between land assets and peer effects is 0.029 and statistically significant at the 1% level. This positive interaction suggests that the strength of peer influence intensifies with increasing land holdings, possibly because such households are better positioned to act on observed social information or because they engage more frequently in productive consumption patterns influenced by peers.

4.5. Heterogeneity Analysis

4.5.1. Income Heterogeneity

Income level is a key determinant of digital adoption in rural areas. Households with different income levels may differ in their affordability and motivation to engage in online consumption, which can affect their sensitivity to peer influence. Therefore, we examine whether the peer effect varies across income groups.

Table 6. Results of Income Heterogeneity Analysis.

Variable	OS
	Low-Income	Middle-Income	High-Income
Peer	0.420 **	0.222	0.384 **
	(0.167)	(0.167)	(0.153)
Age	−0.060 ***	−0.065 ***	−0.062 ***
	(0.005)	(0.005)	(0.004)
Gender	−0.232 **	−0.262 ***	−0.357 ***
	(0.090)	(0.089)	(0.090)
Educ	0.098 ***	0.078 ***	0.082 ***
	(0.012)	(0.013)	(0.013)
Income	0.161 ***	0.540 ***	0.267 ***
	(0.051)	(0.159)	(0.093)
Fml	-0.039	0.034	0.055 *
	(0.026)	(0.027)	(0.029)
Elder	0.047	0.169	0.123
	(0.120)	(0.128)	(0.154)
Child	0.209 **	0.257 **	0.164
	(0.089)	(0.101)	(0.125)
Cons	−1.398	−4.214 **	−1.447
	(1.347)	(1.779)	(1.107)
Province Fixed	YES	YES	YES
Year Fixed	YES	YES	YES
Pseudo R2	0.1463	0.1388	0.1477
Observations	2894	2764	2826

Note: The standard errors are displayed in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% levels, respectively.

reports the heterogeneous effects of peer influence on rural households’ online shopping behavior across different income groups. Based on income terciles, the sample is divided into three groups: low-income, middle-income, and high-income households. The results reveal that peer effects are most pronounced among low-income households, with a coefficient of 0.420. In contrast, the effect is statistically insignificant for middle-income households and only marginally significant for high-income households. These findings suggest that lower-income households are more reliant on peer behavior for information and decision-making, likely due to limited access to external resources and a heightened dependence on community norms. High-income households, while generally more autonomous in consumption decisions, may still be influenced by social comparison or conspicuous consumption within their reference group. Middle-income households appear to occupy a transitional position, exhibiting a relatively weaker tendency to conform to peer behavior. Additionally, the estimated coefficients of control variables, such as age, education, income, and child dependency ratio, align with theoretical expectations, further validating the model’s explanatory power.

4.5.2. Region Heterogeneity

Educational attainment shapes digital literacy and the ability to process external information, which may influence how households perceive and respond to peer behaviors. Accordingly, we explore the heterogeneity of peer effects across different education levels.

Table 7. Results of Region Heterogeneity Analysis.

Variable	OS
	Eastern	Central	Western
Peer	0.191	0.332 *	0.502 ***
	(0.174)	(0.170)	(0.146)
Age	−0.073 ***	−0.063 ***	−0.056 ***
	(0.006)	(0.005)	(0.004)
Gender	−0.213 **	−0.284 ***	−0.362 ***
	(0.109)	(0.092)	(0.077)
Educ	0.074 ***	0.100 ***	0.098 ***
	(0.016)	(0.015)	(0.010)
Income	0.300 ***	0.197 ***	0.299 ***
	(0.067)	(0.057)	(0.047)
Fml	0.076 **	−0.022	−0.022
	(0.031)	(0.028)	(0.023)
Elder	−0.199	0.196	0.113
	(0.170)	(0.148)	(0.104)
Child	0.089	−0.018	0.323 ***
	(0.120)	(0.107)	(0.088)
Cons	−1.653 **	−1.276 **	−3.110 ***
	(0.825)	(0.600)	(0.522)
Province Fixed	YES	YES	YES
Year Fixed	YES	YES	YES
Pseudo R2	0.1719	0.1387	0.1418
Observations	1976	2454	3888

Note: The standard errors are displayed in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% levels, respectively.

reports the regional heterogeneity in the influence of peer effects on rural household online shopping behavior. Following the regional classification by the National Bureau of Statistics, the sample is divided into eastern, central, and western regions based on the households’ provincial locations. The results reveal that the peer effect is strongest and statistically significant in western China (coefficient = 0.502, p < 0.01), followed by a weaker but marginally significant effect in central China (coefficient = 0.332, p < 0.10). In contrast, the effect in eastern China is positive but not statistically significant. These findings suggest that rural households in less economically developed western regions tend to rely more heavily on social networks when making consumption decisions, possibly due to limited access to formal information channels and lower levels of digital literacy. In contrast, households in the eastern region, where digital infrastructure is more developed and individual decision-making capacity is higher, are less susceptible to peer influence. For control variables, the estimated coefficients generally align with theoretical expectations: education and income levels exhibit stable positive effects on online shopping participation, while age and gender show negative associations. Notably, in western regions, the child dependency ratio has a significant positive effect, which highlights the distinctive role of household structure in shaping consumption behavior across different geographic contexts.

4.5.3. Education Heterogeneity

Rural areas in China exhibit substantial variation in infrastructure, economic development, and internet accessibility across regions. To capture potential spatial disparities in peer influence on digital behavior, we conducted subgroup analyses by eastern, central, and western regions. The sample is categorized based on the level of educational attainment, with Columns (1) through (3) of

Table 8. Results of Education Heterogeneity Analysis.

Variable	OS
	Lower Secondary or Below	Upper Secondary	Tertiary Education
Peer	0.603 ***	0.320 ***	−0.037
	(0.174)	(0.118)	(0.318)
Age	−0.069 ***	−0.062 ***	−0.066 ***
	(0.005)	(0.003)	(0.011)
Gender	−0.143	−0.328 ***	−0.635 ***
	(0.091)	(0.067)	(0.200)
Educ	0.065 ***	0.073 ***	0.465 ***
	(0.018)	(0.024)	(0.178)
Income	0.239 ***	0.315 ***	0.223 **
	(0.055)	(0.042)	(0.112)
Fml	−0.017	−0.002	0.100 *
	(0.027)	(0.020)	(0.056)
Elder	−0.044	0.206 **	0.067
	(0.129)	(0.099)	(0.293)
Child	0.165	0.232 ***	0.024
	(0.101)	(0.077)	(0.210)
Cons	−1.956	−2.321 ***	−5.512 *
	(1.778)	(0.665)	(3.066)
Province Fixed	YES	YES	YES
Year Fixed	YES	YES	YES
Pseudo R2	2811	4949	769
Observations	0.1244	0.1311	0.1594

Note: The standard errors are displayed in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% levels, respectively.

representing households whose heads have completed junior secondary education or below, senior secondary education, and tertiary education, respectively. The results show that Peer is most pronounced among the junior secondary or below group, with a coefficient of 0.603, followed by the senior secondary group (0.320), while it is statistically insignificant in the tertiary education group. This pattern suggests that individuals with lower educational attainment are more reliant on social imitation, whereas those with higher education levels exhibit greater autonomy in decision-making. Notably, the direct effect of education on online shopping behavior remains significantly positive, which further underscores the independent promotive role of educational attainment in facilitating online consumption.

5. Discussion

Drawing on panel data from the 2014–2022 China Family Panel Studies (CFPS), this study identifies robust peer effects in rural households’ online shopping behaviors and further investigates the heterogeneity of these effects across asset endowments, income levels, educational attainment, and regional characteristics. These findings extend the applicability of social interaction theory to the domain of digital consumption and provide micro-level empirical evidence for understanding behavioral structures in the context of digital transformation in rural areas of developing countries.

The results demonstrate that rural households’ participation in online shopping is significantly influenced by the consumption behaviors of their neighbors—a pattern consistent with the theoretical foundations of social learning [71] and imitative behavior [27]. In rural contexts characterized by severe information asymmetries and a lack of institutional signals, individuals often rely on peer consumption as a source of indirect experience, serving both as behavioral reference points and as mechanisms for reducing perceived risk. This form of “group-based indirect information transmission” has been extensively validated in various domains, including agricultural technology adoption [72,73] and financial decision-making [74]. Our findings extend this mechanism into the digital consumption sphere, reaffirming Granovetter’s [75] notion of “embedded economic behavior” by illustrating that even seemingly individualistic decisions such as online shopping are deeply shaped by local social structures.

Further analysis highlights the moderating role of asset structure in shaping peer effects. Land assets, as fundamental productive resources in rural households, not only provide long-term economic security but also serve as indicators of future income expectations. Households with larger landholdings typically have stronger economic capacity and a greater ability to bear potential consumption risks. This enables them to respond more actively to peer-driven behavioral signals, particularly when such behaviors are perceived as beneficial or efficient. This finding is consistent with theories of productive consumption and risk absorption [76,77], highlighting that land-rich households may not simply imitate others out of dependence, but do so strategically to optimize consumption under peer-referenced norms. In addition, high levels of expenditure on social obligations, which reflect the extent of one’s embeddedness in social networks, suggest that in highly reciprocal communities, behavioral norms and imitation pressures exert even stronger influences on individual decision-making [74]. Peer effects, therefore, not only function as information conduits but also embody complex behavioral mechanisms shaped by collective norms and social signals.

In terms of heterogeneity, the analysis indicates that peer effects are most pronounced among low-income households. This pattern supports Rogers’ [58] “critical threshold” theory in innovation diffusion, which posits that resource-constrained individuals are more likely to delay adoption and mimic others under uncertainty. Educational heterogeneity also plays a substantial role: the peer effect diminishes markedly with increasing education levels and becomes statistically insignificant among those with college degrees or above. This finding reaffirms the ability of education to enhance information processing and digital literacy, thereby reducing reliance on social cues. Regionally, peer effects are most salient in western China, followed by the central region, and are statistically insignificant in the eastern region. These spatial differences may be attributed to disparities in information accessibility, social density, and behavioral conformity. Consistent with Aker and Ksoll [78], areas with limited communication infrastructure and underdeveloped networks tend to exhibit stronger peer influences due to constrained individual information access, reinforcing the role of social referencing.

The empirical findings have significant implications for policy. Current rural e-commerce policies predominantly emphasize infrastructure and platform development, while neglecting the social structural constraints embedded in behavioral diffusion. The results suggest that policymakers should target populations with lower educational attainment, moderate income levels, and strong social ties as pivotal actors in the diffusion of online shopping. Building demonstration mechanisms within villages, such as selecting “digital consumption model households”, could catalyze behavioral diffusion through social learning. For rural households with dominant landholdings, we should therefore focus on increasing the productive potential and liquidity of land resources—for example, by optimizing land transfer systems, to further encourage informed participation in online shopping. Meanwhile, for households with high levels of social expenditures, integrating online shopping behaviors into trust-based social evaluations could promote positive peer externalities within rural communities. In addition, attention should be paid to improving farmers’ digital literacy through targeted training programs, which can enhance their understanding of digital platforms, reduce psychological barriers, and increase their willingness to participate. Local governments may also incorporate online shopping behavior into existing grassroots governance systems—such as village credit records or collective incentive mechanisms—to provide explicit rewards for early adopters.

Despite this study’s methodological rigor, several limitations remain. First, peer effects are measured using village-level averages, which fail to capture the nuances of specific social networks, emotional proximity, and information diffusion pathways. Future research should incorporate social network surveys, mobile communication data, or digital trace data to better delineate the boundaries of behavioral influence. Second, while this study focuses on the binary outcome of online shopping participation, it does not explore dimensions such as purchase frequency, product categories, or platform preferences. Future studies could employ platform-level operational data to develop structural behavioral models. Third, the interactions between algorithmic recommendations, social media exposure, and offline peer networks represent a promising avenue for future inquiry [79,80]. Developing a hybrid online–offline behavioral model would offer novel insights into the mechanisms underlying online shopping decisions among rural households.

6. Conclusions

Overall, this study identifies significant and robust peer effects in rural households’ online shopping behavior, with notable heterogeneity across income levels, education, and regional contexts. Theoretically, the findings extend social interaction theory into the domain of digital consumption and highlight the critical role of community-based behavioral dynamics in shaping technology adoption. Practically, this study provides new policy insights by emphasizing the roles of peer effects, land assets, and social expenditures—for example, through mechanisms such as the establishment of model households and targeted information campaigns, particularly for moderate-income, low-education groups embedded in strong social networks. Additionally, the dependent variable in this study focuses solely on whether households participate in online shopping, without further exploring dimensions such as purchase frequency or product category. Future research should incorporate social network data and platform-level behavioral traces to more accurately capture the mechanisms of influence and the evolving interplay between online and offline channels in rural digital transformation.