1 Introduction

A shift in individual environmental behaviors towards a more sustainable way can have a significant impact on reducing greenhouse gas emissions (Whitmarsh et al. 2021). Research on the determinants of pro-environmental behavior adoption has been explored across various disciplines, including psychology, economics, sociology, and marketing (Wolske et al. 2020). Recent research has investigated the roles of peer effects on pro-environmental behavioral imitation, suggesting that individuals’ attitudes, values, or behaviors are influenced by the behaviors of members within a peer group (Sacerdote 2014; Nielsen et al. 2020). In some instances, these peer influences counteract certain benefits, such as cost, convenience, or efficiency (Nolan et al. 2008).

An important aspect of studying peer effect is the identification of peer groups (Bramoullé et al. 2009). Most of the existing literature on environmental behaviors defines peer groups in the sense of spatial proximity, such as all individuals within a city or geographic radius (Wolske et al. 2020). In contrast, economic geographers offer an alternative way to identify peer groups by defining relational forms of proximity (Boschma 2005). This approach also considers information flows occurring during long-distance social interactions between individuals who are relationally proximate despite being spatially dispersed (Fontes and Sousa 2016; Agrawal et al. 2008). Indeed, social interactions often occur beyond local geographic boundaries, such as in municipalities, especially in the Netherlands where human mobility is relatively high. For example, the average daily commute distance for Dutch residents in 2015 was 28.31 km (CBS 2016). In addition, municipalities in the Netherlands are typically small. Residents are thus usually not bounded by municipal geographic boundaries, which increases the possibility of long-distance social interactions.

The examination of the role of relational proximity in information dissemination and knowledge spillovers has a well-established history, particularly in the field of innovation diffusion (Lopolito et al. 2022; Fadly and Fontes 2019; Popp et al. 2011; Boschma and Frenken 2010). However, it has been largely ignored in the context of environmental behavior of individuals. While spatial proximity is often a good proxy for implicit channels of information transmission, neglecting the role of relational proximity provides limited insight into other potential drivers of knowledge spread over long distances (Caragliu and Nijkamp 2016). This study attempts to fill this gap by examining the peer effects stemming from both spatially and relationally proximate peers on environmental behavioral imitation.

Specifically, this study explores the individual environmental behaviors, proxied by energy and water consumption, of 53,590 individuals from 352 municipalities in the Netherlands in 2015. The study first defines spatially proximate peers as all other individuals in the same municipality. Next, going beyond the spatial dimension, this study employs intercity travel data to quantify the interconnectedness between municipalities. This measure represents the intensity of information flows between each pair of municipalities, and is used to identify relational peer effects between individuals living in these municipalities. The main results provide evidence for positive and significant peer effects in environmental behavioral imitation among both spatially proximate peers and relationally proximate peers. Finally, this study also examines the role of visibility of pro-environmental behaviors, as represented by the density of electric vehicle charging stations and cycling behaviors, in facilitating peer effects. The result shows that individuals living in municipalities with more prominent pro-environmental behaviors are more likely to adopt similar behaviors. This study mainly contributes to the literature on the role of peer effects in environmental behavioral imitation, particularly by highlighting the significance of relational peer effects.

The rest of this study is organized as follows. Section 2 offers a brief literature review on peer effects in environmental behavioral imitation, with a focus on relational peer effects. The hypotheses are then proposed. Section 3 presents the data and methodologies. Section 4 discusses the findings, and Sect. 5 concludes.

2 Literature and hypotheses

Peer effects usually refer to the influence of members of a peer group on an individual’s attitudes, values, or behaviors (Wolske et al. 2020; Sacerdote 2014; Rai and Henry 2016; Dietz et al. 2005). From a theoretical perspective, individuals can make rational decisions about environmental behaviors by directly evaluating each alternative. However, this can be a costly and time-consuming process. Another plausible alternative is to rely on information from others, especially those who have little information or prior experience on which to base expectations (Dubé et al. 2014; Rai and Henry 2016). In such cases, information from peers plays an important role in shaping individual perceptions, expectations, and ultimately their behaviors (Wilson and Dowlatabadi 2007; Henry and Vollan 2014). Such information flows are mediated by forms of direct advertising and social interactions (Bikhchandani et al. 1998; Babutsidze and Cowan 2014). Direct advertising emphasizes the role of visible information (Hamilton 2021), while social interactions highlight how information and knowledge are exchanged through interpersonal channels and influence behavioral changes (Noll et al. 2014). In either scenario, peer influence is reinforced and behavioral changes occur as a result of repeated exposure to certain information (Kiesling et al. 2012; Kosugi et al. 2019). This can be in an active way, where individuals learn spontaneously from others; or in a passive way, where individuals adjust their behaviors under the normative social influence (Bakker and Demerouti 2018; Boon-Falleur et al. 2022).

2.1 Spatial peer effects

A growing body of empirical studies provides evidence for spatial peer effects in environmental behavioral imitation. Of particular relevance are studies that find peer effects in the diffusion of household renewable energy technologies such as residential solar photovoltaics, electric vehicles, and water heaters (Bollinger and Gillingham 2012; Graziano et al. 2019; Graziano and Gillingham 2015; Irwin 2021; Zhang et al. 2023; Kucher et al. 2021). Most of these studies define peer groups based on spatial proximity, using aggregated data within pre-defined spatial bands or existing geographic administrative divisions. For example, Graziano and Gillingham (2015) study spatial patterns of solar photovoltaic adoption in Connecticut, USA. They define peers as the cumulative number of solar photovoltaic installations within a 0.5-, 1- and 4-mile radius of each observation point. Similarly, Rode and Weber (2016) study solar photovoltaic adoption in Germany where they identify peers as installations within 10, 4, 1, and 0.5 km around each grid point.

This study defines spatial peer groups based on municipal geographical administrative divisions. In this case, the spatial scale is essential. Babutsidze and Chai (2018) identify regional clusters of pro-environmental behaviors by examining survey data from 3096 respondents across 250 three-digit regions and 53 two-digit regions in Australia. Richter (2013) investigates the social effects of solar photovoltaic diffusion within geographically defined peer groups of 2239 zip code districts in the UK. Dharshing (2017) studies solar photovoltaic adoption across 402 German regions, while a similar study by Balta-Ozkan et al. (2021) focuses on 378 UK regions. Despite of different spatial scales, these studies suggest positive and significant peer effects relating to pro-environmental behavior imitation due to spatial proximity.

Consistent with previous studies, by examining survey data of 53,590 respondents at a significantly finer spatial scale, this study hypothesizes that individuals who are geographically proximate tend to adopt similar environmental behaviors. It is expected that individual environmental behaviors, proxied by energy and water consumption, are positively correlated with the behaviors of peers from the same municipality.

2.2 Relational peer effects

In addition to spatial proximity, researchers explore the role of non-spatial proximity in facilitating information dissemination by identifying different types of non-spatial spaces. They argue that geographic proximity is not necessary during knowledge spillovers (Agrawal et al. 2008; Boschma 2005). Indeed, the dissemination of information is not constrained to a geographic unit, but can also be spread over long distances across geographic units due to relational proximity (Caragliu and Nijkamp 2016; Noseleit 2018). Nevertheless, although the role of relational proximity in information dissemination is well established in innovation diffusion literature, little is known about relational peer effects in the context of environmental behavioral imitation.

The process of innovation diffusion is similar to environmental behavioral imitation in several ways. The first aspect is about learning from peers, where individuals observe and learn from the experiences of others. Interactions between early adopters and late adopters facilitate information dissemination about the innovation or behavior, reducing the associated uncertainties and risks, and thus enhancing positive peer effects (Smith and Urpelainen 2014). Moreover, during innovation diffusion, competition motivates agents to adopt innovations in order to gain competitive advantages (Berry and Baybeck 2005). Similarly, during environmental behavioral imitation, social normative beliefs motivate individuals to adopt behaviors that are generally accepted in their peer groups (Miller and Prentice 2016). This helps individuals to avoid discrepancies in social norms and obtain group identities (Henry and Dietz 2012).

There are a few studies using small-scale social network data to explore relational peer effects on environmental behaviors. For example, Babutsidze and Cowan (2014) investigate the effects of face-to-face interactions with close neighbors on pro-environmental behavior adoption. They find that information spreads through stable social networks and generates behavioral clusters. Geiger et al. (2019) study how social network structures predict pro-environmental behaviors by examining pro-environmental behavioral diffusion across friendship connections. Using data from 67 participants in a small religious community, they find that respondents who are socially active with each other exhibit greater similarities in environmental behaviors than those who are not. Nevertheless, the complexity of constructing actual networks with large samples has led to limited studies exploring relational peer effects. Most studies are typically based on small-scale data within certain groups and tend to neglect long-distance social interactions. To gain a deeper understanding of relational peer effects on environmental behavioral imitation, further research with large samples on a larger geographic scope is necessary.

To do this, this study uses large-scale intercity travel data to quantify the relational proximity between individuals based on the intensity of human flows between municipalities. Cities interact with each other through various types of flows, such as information, money, goods, and people (Hall and Hay 1980). When cities are intensely interrelated by these flows, information and knowledge can be transmitted from one to another readily (Perry-Smith and Shalley 2003). While different types of flows can be the potential channels for information dissemination, this study concentrates on human flows considering that people are the actual executors of social interactions. Specifically, relationally proximate peers are identified as all the respondents from the most connected municipalities. Due to the greater likelihood of social interactions, it is hypothesized that individual environmental behaviors, proxied by energy and water consumption, are positively correlated with the behaviors of individuals from the most connected municipalities.

2.3 Visible pro-environmental behaviors

A deeper question here is what reinforces peer effects during environmental behavioral imitation. Some studies attribute one of the reasons to observational learning upon peers’ visible pro-environmental behaviors. Observational learning is the acquisition of attitudes, values, and ways of thinking and behaving through the observation of examples provided by others (Kaufmann et al. 2021). Social learning theory suggests that individuals learn about the best courses of actions by observing the choices of their peers (Golman et al. 2016; Bandura 1977). Empirically, Rode and Weber (2016) investigate the spatiotemporal diffusion of solar photovoltaic adoption in Germany, suggesting highly localized imitative solar photovoltaic adoption behaviors. They attribute the peer effects to the visibility of solar photovoltaic to passers-by, which enables behavioral imitation without direct verbal interactions. Babutsidze and Chai (2018) examine pro-environmental behavioral imitation across Australia, providing evidence of spatial clustering of visible pro-environmental behaviors. Visible mitigation actions of peers increase individuals’ concerns about climate change, and further lead to a greater possibility of taking similar mitigation actions (Beattie et al. 2019).

This study examines the extent to which visible pro-environmental behaviors, proxied by EV charging stations and cycling behaviors, enhance peer effects among spatially proximate peers. It is hypothesized that individuals living in municipalities with higher densities of EV charging stations and cycling behaviors perform more pro-environmental behaviors due to their greater likelihood of observing peers’ pro-environmental behaviors. It is notable that individuals who frequently commute long distances are also more likely to observe pro-environmental behaviors such as solar photovoltaics along the way. Likewise, social interactions that disseminate information also exist among spatially proximate peers. Nevertheless, the main purpose of this study is to demonstrate the existence of spatial and relational peer effects, as shown in Hypothesis 1 and Hypothesis 2, rather than to explain the underlying mechanisms. The purpose of Hypothesis 3 is to provide empirical evidence for the literature that emphasizes the role of visibility in pro-environmental behavior imitation.

3 Data and methods

This section discusses the data sources, the variables constructed, and the empirical model used to analyze the spatial and relational peer effects on environmental behavior imitation among individuals in the Netherlands. Multiple databases are used to construct the variables. Table 1 summarizes the variables used in this study.

Table 1 Variable definitions and data sources
Full size table

3.1 Individual environmental behaviors

The dependent variable is individual environmental behaviors, proxied by annual energy (electricity and natural gas) and water consumption per capita. It is worth mentioning why consumption data is chosen as a proxy for environmental behaviors. A widely used strategy for quantifying environmental behaviors is to calculate self-reported behaviors by asking questions such as ‘Did you use less electricity last year?’. One potential concern with this proxy strategy is that respondents tend to confuse ‘wanted to do (values)’ with ‘actually did (actions)’ because they do not remember precisely what has been actually done. They provide answers based on their values and vague memories. Various studies suggest that there is a discrepancy between stated beliefs and actual behaviors, the so-called ‘value-action gap’ (Flynn et al. 2009). As a result, these self-reported data are often criticized for their limited reliability (Tiefenbeck 2014). Moreover, individual differences in consumption inertia, deep habits and lifestyle may contribute to the heterogeneity of environmental behavioral types (Babutsidze and Chai 2018). The actual consumption data provides us with deeper insights into individuals’ daily behaviors and lifestyles. It is the result of daily behaviors that reflect the aggregated impacts of multiple behavioral decisions rather than a single action.

The consumption data are from The Netherlands Housing Survey 2015 (BZK and CBS 2016). It is a nationwide individual-level survey that is carried out every three years. The Netherlands Housing Survey 2015 provides information on household energy and water consumption, household composition, building characteristics, individual socioeconomic status, and neighborhood environment. One of the unique benefits of using this database is that registration data from other official agencies are linked to each respondent, such as the Dutch Tax Authorities, energy suppliers, and Municipal Personal Records Database. The Netherlands Housing Survey 2015 contains data from more than 62,000 respondents from 399 municipalities. After removing respondents with incomplete data such as age, income, education level, or building ages, a total of 53,590 final observations are retained. Since the consumption data in The Netherlands Housing Survey are at the household level, the total expenditure on household energy and water consumption is divided by the total number of household members to obtain the expenditure per capita.

The geographic information of the respondents is publicly available at the municipal level. Notably, for administrative reasons, less populous municipalities in the Netherlands have been merging over years. The number of municipalities reduced from 399 in 2015 to 355 in 2019. Since the 2015 data for the other variables are only available based on 2019 administrative boundaries, data collected based on 2015 administrative borders (e.g., The Netherlands Housing Survey 2015) are coordinated to reflect the administrative boundaries as of 2019. After excluding the three municipalities with no respondents, the 53,590 are distributed across 352 municipalities. The average size of the municipalities is 117 square kilometers, with 152 respondents per municipality. This represents a reasonably small spatial unit and a sufficient sample compared to similar studies. Table 2 presents the counts of the spatial distribution of respondents in these 352 municipalities. The municipalities with the fewest respondents (7) are Ameland and Rozendaal, while the municipalities with the most respondents are The Hague (1800) and Amsterdam (1777). Individual utility expenditure data are aggregated at the municipal level to explore spatial heterogeneity. As shown in Fig. 1, individuals with high utility consumption appear to be concentrated in the southeast and northeast of the country.

Table 2 Spatial distribution of respondents across municipalities
Full size table
Fig. 1
figure 1

Average Energy and Water Consumption Per Municipality in Euros in the Netherlands, 2015 (Data source: The Netherlands Housing Survey 2015)

Full size image

3.2 Spatial peer effects

A crucial part of this study is the construction of an index that reflects the spatial and relational relationship between individuals as well as the environmental behaviors of the peer groups. Following Babutsidze and Chai (2018), spatially proximate peers are defined as all the other respondents living in the same municipality as the respondent. After the spatial peer groups are identified, their environmental behaviors are measured by their average annual energy and water consumption. This can be statistically written as

$${SC}_{j}^{a}=\frac{\sum {x}_{j}- {x}_{a}}{n-1}$$
(1)

where \({SC}_{j}^{a}\) is the spatial peer effects of individual \(a\). It is measured by the average consumption of the spatially proximate peers relative to each individual \(a\) in municipality \(j\), \(n\) is the number of respondents living in municipality \(j\), and \(x\) is the consumption of each respondent. This index is unique to each respondent, as the respondent’s own consumption is excluded in the calculation.

3.3 Relational peer effects

Similarly, in order to quantify the relational peer effects, it is necessary to first identify the relationally proximate peer groups. To this end, this study employs intercity travel data to calculate the interconnectedness between every two municipalities as a proxy for information flows between two municipalities. Next, all the individuals from these most connected municipalities are then defined as relationally proximate peers of the individuals from the given municipality. The more frequent intercity travel, the greater opportunities for information and knowledge exchange between municipalities, and therefore the more likely that people are influenced by residents of those highly connected municipalities (Derudder and Witlox 2005; Derudder et al. 2007).

The data for intercity travel are from The Dutch National Travel Survey 2015 (CBS and RWS 2016). A total of 37,351 individuals participated in this survey and were asked to document all the trips that occurred on that day. In total 115,987 trips were reported, of which 34,842 (nearly 30%) were intercity trips. The geographic information of these intercity trips is of particular interest, as it allows us to measure the interconnectedness between each and every pair of municipalities. Specifically, the municipal geographic information about the origins and destinations of these trips is imported into Gephi for the construction of the intercity travel network as shown in Fig. 2.

Fig. 2
figure 2

Intercity Network in the Netherlands, 2015 (Data source: The Dutch National Travel Survey 2015)

Full size image

In this network structure, each municipality is a node, and the trips between every two municipalities are edges. Mathematically, the network \(\mathrm{G }(D,E)\) consists of a set of municipalities \(N=\left|d\right|\) and a set of linkages \(E=\left|\varepsilon \right|\). Each matrix cell of \(D\) is represented by \(d(i)={\sum }_{i}^{m}{(d}_{ij}*{w}_{ij})\), where \({d}_{ij}=1\) if there is a link between municipality \(i\) and municipality \(j\), \({w}_{ij}\) is the number of links between municipality \(i\) and municipality \(j\), and m is the total number of municipalities in the network. Each matrix cell of \(E\) is presented by \({\varepsilon }_{ij}={w}_{ij}\), where \({w}_{ij}\) is the number of links between municipality \(i\) and municipality \(j\). As shown in Fig. 2, the sizes of the nodes represent the centralities of the municipalities, and the thicknesses of the edges indicate the travel frequencies between every two municipalities.

Based on the strength of the edges \(E\), the eight municipalities with the most frequent intercity travel to a given municipality are defined as its most connected municipalities. Individuals from these eight municipalities are therefore defined as the relationally proximate peers of the respondents in the given municipality. The average consumption of these individuals is then calculated to quantify the relational peer effects relative to each individual \(a\) in municipality \(j\). Statistically, it can be written as

$${RC}_{j}^{a}=\frac{\sum {x}_{i}}{\sum {n}_{i}}\, \left(i\ne j\right),$$
(2)

where \({RC}_{j}^{a}\) represents the relational peer effects of individual \(a\), \(\sum {x}_{i}\) denotes the total consumption of peers in the eight municipalities that are most closely connected to municipality \(j\), and \(\sum {n}_{i}\) denotes the total number of respondents in these eight municipalities. This index is the same for all respondents in the same municipality.

3.4 Visible pro-environmental behaviors

This study also examines the role of the visibility of pro-environmental behaviors in environmental behavioral imitation. Following Vandenbulcke et al. (2011), Kaufmann et al. (2021) and Babutsidze and Chai (2018), two behaviors are used to measure the extent to which the respondents are able to observe pro-environmental behaviors, namely the EV charging station installations and cycling. EV charging station reflects EV ownership, which reflects the pro-environmental behaviors of the municipality’s residents (Morton et al. 2018). Many empirical studies confirm that the visibility of solar panels can create peer effects, leading to increased adoption of solar panels or other related behaviors due to inter-behavioral spillover effects. Similarly, EV charging stations and cycling are highly visible, and the process of charging an EV can be easily observed. It is thus expected that observing peers engaging in pro-environmental behaviors can trigger similar actions in others, as reflected in consumption data that presents the aggregated outcomes of their behavioral decisions. Specifically, EV charging station density is calculated by dividing the total number of EV charging stations in a given municipality by its population. The data on EV charging stations is from The Ministry of Infrastructure and Water Management of the Netherlands (RWS 2016). Figure 3 represents the EV charging station density per municipality in 2015. Overall, the EV charging station density in urbanized areas, such as the Randstad area, is higher than that in the peripheral areas. Finally, cycling behavior density is the ratio of the population using bicycles as the transportation mode for trips shorter than 7.5 km to the total population in a given municipality. The data on bicycle usage is from The Dutch National Institute for Public Health and the Environment (RIVM 2016). As shown in Fig. 4, municipalities in the south of the Netherlands score lower.

Fig. 3
figure 3

EV Charging Station Density Per Municipality in the Netherlands, 2015 (Data source: RWS 2016)

Full size image
Fig. 4
figure 4

Cycling Density Per Municipality in the Netherlands, 2015 (Data source: RIVM 2016)

Full size image

3.5 Individual and contextual effects

There are three potential explanations for behavioral similarities among individuals living in the same area, namely, individual effects, contextual effects, and peer effects. Individual effects arise from shared individual characteristics, such as similar socioeconomic status, while contextual effects arise from exposure to a similar natural and social environment such as clusters of wealthy people. The challenge lies in distinguishing whether individual behavioral choices are derived from peer effects or omitted unobserved variables (Manski 1993, 1999; Halleck Vega and Elhorst 2015). One modeling strategy to tackle this challenge, as suggested by Irwin (2021) and Zhang et al. (2023), is to carefully control for individual and contextual effects in order to mitigate bias when estimating peer effects. Therefore, this study incorporates a range of individual and contextual variables that may affect energy and water consumption. Specifically, the individual-level variables include respondents’ age, gender, educational level, household income, household composition, and dwelling age. Additionally, municipality characteristics are captured by migrant ratio, housing value, home ownership, and address density. The individual-level data are obtained from external registered databases linked to The Netherlands Housing Survey 2015, while the municipal-level data are sourced from the Central Bureau of Statistics of the Netherlands.

Table 3 provides summary statistics for the main variables used in this study. The sample for this study is 53,590 individuals from 352 municipalities in the Netherlands. The dependent variable, individual environmental behaviors, is measured by energy and water consumption per capita. On average, the annual energy and water consumption expenditure per capita of the respondents in 2015 is 936.45 Euros. The three main independent variables are namely spatial peer effects, relational peer effects, and visible pro-environmental behaviors. The spatial peer effects examine the extent to which individuals are influenced by the environmental behaviors of their spatially proximate peers, i.e., respondents who live in the same municipality. The relational peer effects examine the extent to which individuals are influenced by the environmental behaviors of their relationally proximate peers, i.e. respondents in the municipalities with which they interact most frequently. Finally, two highly visible pro-environmental behaviors, i.e. EV charging station installations and bicycle usage, are used to examine the role of visibility in environmental behavioral imitation.

Table 3 Summary statistics
Full size table

3.6 Model specification

To investigate the spatial and relational peer effects in environmental behavior imitation, an ordinary least square (OLS) regression is performed. Specifically, the following model is estimated:

$${Y}_{i}=\alpha +{\beta }_{1}SC+{\beta }_{2}RC+{\beta }_{3}IPB+\beta {X}_{i}+\varepsilon ,$$
(3)

where \({Y}_{i}\) is the dependent variable for individual \(i\), proxied by individual energy and water consumption expenditure. For individuals from multi-person households, it is calculated by dividing total household consumption by the number of household members. \(\alpha \) is the intercept, \(\beta \) is the vector of coefficients for the independent variables, and \(\varepsilon \) is the error term. \(SC\) is the peer effects from spatially proximate peers derived in Eq. (1), \(RC\) is peer effects from the relationally proximate peers derived in Eq. (2), \(IPB\) is the visible pro-environmental behaviors in a given municipality, and \({X}_{i}\) includes a range of individual and contextual variables that may be associated to energy-related consumption.

There are three expectations regarding the expected signs of the coefficients of the main independent variables. First, it is argued that spatial proximity is likely to play a role in transmitting information and knowledge. Therefore, the coefficient of spatial peer effects \({\beta }_{1}\) is expected to be positive. Second, it is believed that information is not only spread not only within bounded space but also over long distances through social interactions. Thus, a positive coefficient of the relational peer effects \({\beta }_{2}\) is expected. Finally, the third hypothesis relates to visible pro-environmental behaviors. A negative coefficient \({\beta }_{3}\) is expected, as individuals who are more likely to observe pro-environmental behaviors will also adopt similar behaviors and therefore consume less.

4 Results and discussion

4.1 Main results

Table 4 presents the coefficient estimates of the OLS. Regarding the spatial peer effects, the result shows evidence of environmental behavioral imitation among spatially proximate peers, which supports Hypothesis 1. As indicated in the second row in Table 4, the parameter estimate for the energy and water consumption of spatially proximate peers (\(\beta \) = 0.234) is found to be positively significantly correlated with individual consumption at the 1% level. It suggests that individuals imitate the environmental behaviors of their peers residing in the same municipality. The result is consistent with the literature emphasizing the effects of spatial proximity in environmental behavioral imitation (Bollinger and Gillingham 2012; Kim et al. 2018; Babutsidze and Chai 2018).

Table 4 Main results (N = 53,590)
Full size table

Regarding the relational peer effects, the parameter estimate (\(\beta \) = 0.238) is found to be positive and significant at the 1% level. This supports Hypothesis 2 which assumes peer effects among relationally proximate peers. The frequent human flows between municipalities imply that there is a high level of social interaction among individuals in these municipalities. This allows knowledge and information such as tax incentives, rebates, and the profitability of certain environmental behaviors to be more widely disseminated among these individuals. Information exchange enhances active peer effects through social learning. This finding is consistent with a large literature that emphasizes the critical role of relational proximity in the diffusion of innovations. Notably, the peer effects that arise during long-distance commuting can also be due to a greater likelihood of observing environmental behaviors, such as rooftop solar panels along the journey. Although the mechanisms are not distinguished, this does not affect the document of the presence and magnitude of relational peer effects.

There are several aspects that explain why peer effects affect environmental behavioral imitation, among both spatially and relationally proximate peers. First, individuals are concerned about their social status, which is partly determined by peer predispositions. These predispositions, in turn, are signaled by individual actions (Messick 1999). As a result, individuals tend to follow a single, homogeneous standard of behavior in order to maintain their desired group identities and gain peer conformity (Contractor and DeChurch 2014; Hogg and Reid 2006). Second, descriptive social norms suggest that people do what most people do to avoid social disapproval. In a social environment, even small deviations from social norms can undermine social status and generate harsh accusations of non-conformed activities (Ioannides and Topa 2010). Third, social learning theory suggests that the behaviors of others can serve as a signal of the quality of a public good due to the lack of financial resources, know-how, confidence, integrity, or incomplete information (Vesterlund 2003).

The role of visibility of pro-environmental behaviors in environmental behavioral imitation is also investigated. The coefficient estimates of − 0.019 for EV charging station density and − 0.044 for cycling behavior density are negatively and statistically significant at the 1% level, supporting Hypothesis 3. These findings suggest that the higher the density of EV charging stations and cycling behaviors in a given municipality, the more likely respondents in the same municipality adopt similar pro-environmental behaviors. This study does not examine imitations between identical environmental behaviors since environmentalism is a result of multiple daily behavioral decisions rather than a single action. Pro-social behavior theory suggests that people make pro-social adjustments to their behaviors in reference to the behaviors of their peers in order to display their social identities, self-images, and reputations. In the environmental context, people may perceive their peers’ pro-environmental behaviors as signals of their environmental awareness. As a result, they imitate their peers to demonstrate their environmentalism and not be divorced from society. In addition, social learning theory implies that people also consider the behaviors of their peers as a reference for the best choice. Where information is incomplete, public decisions reduce the uncertainties and risks associated with certain actions. The visibility of pro-environmental behaviors, which conveys information without verbal communication, more directly enhances peer effects and further lead to the imitation of environmental behaviors.

Turning to the individual-level control variables, income is positively correlated with energy and water consumption. This can be attributed to several reasons. First, time efficiency and convenience, rather than cost, are the primary considerations of higher income groups. Certain energy conservation behaviors are found to be time-consuming and inconvenient. Second, high-income groups are more likely to engage in conspicuous consumption, such as fireplaces and other luxury appliances with high energy consumption. In terms of educational level, the results indicate that people with higher educational levels consume more energy and water. This is somewhat curious as it is expected that people with higher education will be more pro-environmental due to the sustainability education they receive from higher education institutions.

The result also indicates that residential energy and water consumption increases with the ages of the respondents. As Estiri and Zagheni (2019) point out, the increase in energy consumption over the life course can be attributed to the growing size of houses, luxurious energy-consuming household appliances, and the accumulation of wealth. Moreover, the result shows that individuals from multi-person households consume less energy and water than those from single-person households. One potential explanation is that some home energy is shared and consumed per space rather than per inhabitant. Compared to males, females are found to be more likely to engage in pro-environmental behaviors. Finally, energy and water consumption are found to be increased with the age of the buildings. This is to be expected because of the use of more sustainable and environmental-friendly construction techniques in the new buildings.

Regarding the municipality-level control variables, it is found that immigration ratio may impact energy consumption. This could be due to cultural differences resulting in varying living habits, such as cooking and preferences for indoor temperature, which can lead to variations in energy usage among households (Winkler and Matarrita-Cascante 2020). Moreover, respondents from municipalities with high housing values tend to have higher energy-related consumption. This is possibly because these areas tend to have a concentration of more affluent households who can afford larger homes that require more energy to operate (Bao and Li 2020; Glaeser and Gyourko 2018). Additionally, homeownership rates are found to be positively correlated with energy usage, possibly due to the fact that owner-occupied homes tend to be larger, and homeowners tend to engage in more energy-intensive activities such as landscaping (Brounen et al. 2012). Finally, respondents from high-density municipalities have lower energy-related consumption, possibly because these areas consist mostly of apartments that use less energy (Ng 2009).

4.2 Robustness checks

In order to check the validity of the results, three robustness checks are performed. The first two namely observe heterogeneous peer effects among individuals from different income levels (M1–M3)Footnote 1 and educational backgrounds (M4–M6). The motivation is that social class may influence people’s pro-social behaviors. Socioeconomic status and educational level are associated with people’s freedom over economic resources, control over life outcomes, attention to the needs of others, and access to social interaction opportunities (Piff et al. 2010). These differences may result in different peer effects among people of different socioeconomic statuses. Specifically, respondents are classified into three groups based on their income, each quantile containing 17,864 respondents. The average annual household income for the low-income group, the middle-income group, and the high-income group is 9205.57 euros, 21,514.57 euros, and 60,066.73 euros, respectively. Regarding educational attainment, 11,853 (22.12%), 23,002 (42.92%), and 18,738 (34.97%) of the respondents are assigned to the low-education group (under secondary school), the middle-education group (secondary school), and the high-education group (above secondary school). Finally, the third robustness check (M7) distinguishes the purposes of intercity travel that is used to identify relational peers, as different travel purposes may result in disparities in the types and effectiveness of information exchange. Specifically, intercity travel for work-related purposes is used to identify relational peers, given that work-related networks tend to comprise individuals with similar social status and background. Out of 34,842 intercity trips, 20,973 (60.19%) are for work-related purposes.

The results are presented in Table 5. Regarding the three variables of primary concern, the results are mostly consistent with the baseline model, except for the insignificant but still negative estimate of cycling density for the high-educated group. Nevertheless, the result of EV charging stations, another indicator of peer visible pro-environmental behaviors, remains consistent among the high-educated group.

Table 5 Robustness checks
Full size table

5 Conclusion

Relational proximity is crucial in information transmission and knowledge spillover, yet little is understood about its impacts in the context of environmental behaviors. To fill this gap, this study examines spatial and relational peer effects on environmental behavioral imitation, using microdata from 53,590 residents in 352 municipalities in the Netherlands. First, the study confirms the existence of environmental behavioral imitation among both spatially and relationally proximate peers. It underlines the importance of social interaction and observation in fostering such effects, whether through active social learning or passive reinforcement of social norms. The findings provide empirical evidence for social learning theory and normative social influence theory. A key contribution of this study is the demonstration of peer effects among peers who are relationally proximate, adding to the literature on peer effects that primarily focuses on spatial proximity. The Intercity travel data used to identify relational peer groups essentially implies the dissemination of information and knowledge among them. Finally, the study also highlights the significance of the visibility of pro-environmental behaviors in enhancing peer effects among spatially proximate peers.

The findings of this study can provide insights and inform discussions among policymakers who seek to leverage peer influence to promote pro-environmental behaviors. First, social learning fosters peer effects and environmental behavioral imitation. Therefore, policies can create opportunities to facilitate experience and knowledge sharing among peers and stakeholders, potentially through community-based projects. One example is the Deventer Energy Strategy 2020–2050 in the Dutch municipality of Deventer. This plan promotes collaboration between local governments, energy companies, community groups, and residents to jointly invest in renewable energy projects and share the profits. Another local initiative that stimulates peer-to-peer collaboration is Buurkracht (Power of Neighbors). It aims to support citizens to engage with their neighbors in activities that promote energy efficiency and renewable energy use such as collectively purchasing solar panels and insulation materials. Buurkracht provides local communities with tools and resources, such as access to energy experts and information on energy-saving techniques, products, and services.

Additionally, such community-based projects not only facilitate the knowledge flows among peers and stakeholders but also arise and shape social norms. For instance, the Solar Panel Village project in the municipality of Amersfoort pools community resources to install solar panels on a shared roof, creating a visible example of peers’ pro-environmental practices. The presence of solar panels in community services as a visual reminder of the choices of peers, fostering a sense of collective responsibility. This is also consistent with the finding that the visibility of pro-environmental behaviors plays a role in promoting behavioral imitation. Third, the finding also highlights relational peer effects, suggesting that knowledge can be transmitted over long distances. With successful practices from community projects in certain regions, policymakers could develop strategies to facilitate the flow of knowledge into regions with limited experience. One such example is the Green City Network project in Germany which brings German cities together to exchange their best practices and experiences in urban sustainable transition. The Dutch policymakers could establish similar programs to facilitate the flow of knowledge among municipalities through workshops, technical assistance, awards, and pilot projects.

Nevertheless, this study also has several limitations. First, it relies on consumption data as a proxy measure for individual environmental behaviors. While consumption data reflect lifestyles and deep habits, they relate to the environmental behaviors within the home and exclude outdoor behaviors such as transportation choices. Second, one of the assumptions underpinning the model is that frequent communication leads to the transmission of information about pro-environmental behaviors. However, it does not account for whether peers actually discuss environmental issues. It is also possible that some people socially interact frequently but rarely discuss environmental issues. Similarly, effective knowledge transfer is not necessarily mutual as people’s knowledge and experience with pro-environmental behaviors can differ significantly. In the robustness checks, this study attempts to control for such variation by using only work-related intercity travel to identify relational peers. Nevertheless, when data allow, future studies could measure the probability of each respondent discussing environmental behaviors, and weigh their likelihood of transmitting pro-environmental messages accordingly.

Third, intercity travel data is used to establish relational proximity and to indicate the intensity of information flows. However, information can also flow over long distances through various other channels such as electronic communication technologies. Using additional relational data such as intercity electronic communication data to imply intercity information flow can complement this limitation. Finally, this study is subject to the Modifiable Areal Unit Problem, which is a methodological limitation common to spatial analyses concerning aggregated data. It refers to that the scale of spatial data aggregation and the way in which the study area is divided into smaller areas can affect the results. This issue can affect the measurement of peer effects by changing the size of spatial units used to aggregate data. Moreover, respondents in this study are aggregated at the municipal level due to data availability, but using a smaller spatial scale can improve the precision of capturing peer effects. Nevertheless, the sample of 53,590 respondents from 352 municipalities provides a reasonable spatial scale for peer identification considering the relatively small geographical size of the Netherlands.