You will never participate alone. Personal networks and political participation in Belgium

ABSTRACT This paper looks at the mobilizing effect of personal networks on the individual propensity to favour some types of political participation over others, in a context of changing participation repertoires. We rely on original egocentric network data gathered via a unique online survey conducted among a quota sample of 2801 Belgian citizens. We show that dominant political behaviour(s) in a network diffuse as byproduct of social proximity and influence: the more someone has been exposed to a certain type of participation in the past, the more this person is likely to be recruited in the same type of participation in the future (engagement), or, if this person was already active, to retain the same participatory behaviour (retention). Moreover, our results point to a cross-over dissuasive effect across types of participation that keeps citizens away from certain participatory behaviours. In particular, exposure to online and instiutionalized participation in their personal network decreases respondents’ likelihood to engage in non-insitutionalized participation. Overall, we stress the added-value of a meso-level approach that embeds citizens in their personal network to understand their participatory choices.


Introduction
Political participation has faced major changes in the last decades (Marien, Hooghe, and Quintelier 2010). Institutionalized types of participation such as associational and party memberships are in decline (Dalton and Wattenberg 2000;Marien and Quintelier 2011;van Haute, Paulis, and Sierens 2017) while non-institutionalized and online types of participation are expanding (Klingemann and Fuchs 1995;Norris 2002;Stolle, Hooghe, and Micheletti 2005;Gibson and Cantijoch 2013;Theocaris and van Deth 2018).
In this context, a central question for participation scholars is to explain how citizens make a choice and favour some types of political participation over others. However, the classic explanatory models developed in the literature heavily focus on the determinants of the intensity of participation in general rather than its nature, or of a specific type of participation. What is still largely missing, is a better understanding of how citizens make a choice within the action repertoire that is offered to them (Tilly 1993). the results of our analyses are presented and discussed. We show how participatory habits and characteristics of citizens' personal network predict their likelihood to engage in politics and their choice between different types of participation.

Political participation and the role of social networks
Political participation is a central field of research in political science. It has generated numerous debates and explanatory models. These models focus mainly on micro-and macro-level factors. Micro-level studies revolve around three main models: (1) the resources model, which emphasizes the role of socio-economic status and inequalities in participation (Brady, Verba, and Schlozman 1995); (2) the socio-psychological model, which stresses the role of attitudes that favour participation; and (3) the rational choice model that highlights individual motivations behind participation (Olson 1965;Whiteley 1995;Bäck, Teorell, and Westholm 2011). Macro-level studies have partly looked at the political opportunity structure of participation (e.g. Norris 2002;Gallego 2008;Hooghe and Marien 2012;Dalton 2017). These studies focus on cultural, structural and institutional explanations shaping the 'structures of opportunities for civic engagement' (Norris 2002, 25).
These studies have contributed to our knowledge of who gets involved, why, and under which context. Yet they are not without shortcomings. Micro-level studies tend to consider individuals out of their social contexts as 'atomized actors floating unanchored' (Knoke 1990(Knoke , 1058. Furthermore, they consider participation as an individual undertaking despite its often very collective nature (Leighley 1995;Lazer 2011). Lastly, 'opportunities to participate are not equally or randomly distributed in the population' (Leighley 2008, 46) and individual-level inequalities in participation may in fact reflect inequalities in participation opportunities, as shown by Rosenstone and Hansen (1993).
If macro-level studies put individuals back into their context, they do not shed light on how the process works: How and when do political opportunities lead to actual participation? Under which circumstances do resources, attitudes, and motivations get turned into participatory behaviour? Answering these questions calls for an investigation of intermediate explanatory factors, and especially the role of mobilization by social groups or agencies. Yet these factors have largely been neglected in the literature.
Recently, the mobilizing effect of social relations has attracted a growing attention as a result of the increasing inter-connectedness of our modern societies. Most notably among the pioneers, Brady, Verba, and Schlozman (1995) adapted their resources model to recognize mobilization as a crucial factor when they state that citizens who do not participate in politics do so either because they can't (they lack the resources), because they don't want to (they lack the mindset and motivation) or because nobody asked (they are isolated 'from recruitment networks through which citizens are mobilized to politics') (1995,271).
Two main sub-fields have contributed to the growing attention to these 'network' factors and their effect on political participation. First, sociologists have early on looked at the role of personal networks in the recruitment processes in non-electoral politics (e.g. Klandermans and Oegema 1987;Passy 2003;Diani and McAdam 2003). They underline that personal networks allow individuals to socialize, to be recruited, and thereby enhance their disposition to participate, but also to repeat their participation in social movements, protests, or radical activist groups (Schussman and Soule 2005;Saunders et al. 2012). These 'network capitalists' argue that social relationships are important as they give access to resources of various kinds (material, immaterial, instrumental, emotional, etc.). They can thus have consequences on various social and behavioural outcomes, such as political participation. Moreover, it has been shown that the structure formed by interpersonal relationships (network density, size, etc.) influences social and political collective action, as the relationships between individuals make them interdependent in their contributions to a common goal (Gould 1993;Siegel 2009).
Second, political communication and election studies have analysed individuals' decision to engage in elections and campaigns as a by-product of the form and content of their social networks (Huckfeldt and Sprague 1987;Coleman 1988;Knoke 1990;La Due Lake and Huckfeldt 1998;Huckfeldt et al. 2000;Burt 2000;Putnam 2000;McClurg 2003;Zuckerman 2005;Sinclair 2012;Campbell 2013;Ahn, Huckfeldt, and Ryan 2014). Election studies have also looked at social networks through their analysis of campaign techniques, including canvassing, door-to-door, get-out-to-vote initiatives, or the role of opinion leaders and mobilizing agencies (e.g. Rosenstone and Hansen 1993;Green 2000a, 2000b).
However, these studies are limited to electoral participation and how 'networks' of political discussion among peers influence the decision to vote, vote choice, or the probability to take part in election campaign activities (like donations). Few researchers have extended this framework to other types of political participation, or to the analysis of the choice between types of participation.
This paper fills these gaps and looks at the mobilizing effect of personal networks on the individual decision to participate politically and the choice between types of political engagement. More specifically, it analyses how individuals turn into types of politically active citizens and the role that the 'others' play in that process. To do so, it mobilizes the literature on behavioural diffusion in social network analysis (SNA) (Valente 1995;Rogers 2003) applied to participatory behaviours. We argue that individuals' personal networks and the characteristics of these networks play a decisive role. We assume that, in their personal networks, citizens (ego) get stimulus from other relevant agents (alters), and that these may influence the way ego behaves. As stimulus, we focus on the type of participation that alters are engaged in, and on their social and political profile.
From network diffusion theories, we can expect that the exposure to participation stimuli matters. The more agents are exposed to a (political) behaviour in their proximate environment, the more they will be pushed to adopt the same (political) behaviour. This is known as the 'exposure' effect. Depending on how much someone is exposed to politically active agents, 'networks may increase individual chances to become involved, and strengthen activists' attempts to further the appeal of their causes' (Diani 2004, 339). Exposure to peers' political engagement is shown particularly relevant to explain mobilization processes in social movements, sects, clubs or any kind of voluntary associations (Diani and McAdam 2003), but also in politics and political groups (Beck 2002;McClurg 2003;Zuckerman 2005;Sinclair 2012;Morisi and Plescia 2018). Nickerson (2008) finds a contagion effect in his study that shows that voting intentions and voting choices are viral within households, using a field experiment with a 'get out the vote' mailing as the treatment. In another field experiment, Sinclair (2012) supports the same finding, but finds out that it does not diffuse to other households. In the same vein, Fowler (2005) stresses that a modest degree of contagion can cause a chain reaction which, in turn, substantially raises aggregate voter turnout. He estimates that one person's decision to vote can affect up to four other individuals. Similarly, Siegel (2009) simulates the structure of social networks and shows that network ties channel civic participation. Therefore, when alters are themselves already committed to some type(s) of political participation, they can then stimulate ego's probability to engage in the same type of political activities as a result of peer-pressure and mimicking behaviour. Without ignoring the tricky causal relationship between the exposure to participation and the individual behavioural output (McAdam and Paulsen 1993), we follow the theoretical direction put forward by a substantial amount of research. Hence, we hypothesize that the higher ego's exposure to a specific type of political participation in his/her social network, the higher ego's likelihood to engage in the same type political participation (H1).
Exposure to a specific type of political participation can not only have an effect on ego's likelihood of engaging in that type of participation, but it can also have a crossover effect on ego's likelihood to engage in other types of political participation. This is a consequence of the normative power of social networks, which has been shown particularly salient for citizens' voting duty (Gerber, Green, and Larimer 2010) or for humanitarian actions (Roblain et al. 2020). Through interpersonal relations, civic norms are imprinted and citizens develop an internalized sense of civic duty (Campbell 2013). A network where institutionalized participation is the norm in terms of political behaviour can push ego to perceive non-institutionalized participation as the 'wrong' participatory behaviour to adopt, and conversely. Furthermore, at the individual level, Dalton (2008) has shown that different sets of norms lead to different types of political participaiton. Hence, we expect that the higher ego's exposure to institutionalized participation in his/ her social network, the lower ego's likelihood to engage in non-institutionalized participation, and vice-versa (H2).
Finally, the overall composition of personal networks matters as well when considering political action (Knoke 1990;Zuckerman 2005). We know that individual socio-demographic characteristics and political attitudes are crucial determinants of political participation and types of participation at the individual level (Marien, Hooghe, and Quintelier 2010). In the same way, the social and political profiles of alters in the network play a significant role too, providing ego's individual attributes a social dimension (Campbell 2013). For instance, it is shown that education acts as a status-sorting mechanism within social networks for some types of participation (Persson 2015), while exposure to people with greater political knowledge conducts to higher levels of political involvement (Campbell 2013). Furthermore, there is evidence that social networks contribute to reinforce social and political inequalities inherent to politics, rather than overcome them. Party organization studies have shown that the homogeneous composition of party members' networks in terms of attitudes and socio-demographics prevents parties from reaching other profiles than the 'usual suspects' (i.e. male, older, educated, interested, etc.) thereby explaining their recruitment shortage in certain social categories (Paulis 2019). Hence, we expect that the higher ego's exposure to alters who have high levels of social and political resources in his/her social network, the higher ego's likelihood to engage in institutionalized participation (H3a). Conversely, it is acknowledged that the youth and women are underepresented in party-based participation while more prone to unconventional politics (Marien, Hooghe, and Quintelier 2010). Hence, we expect that the higher ego's exposure to alters who have low levels of social and political resources in his/her social network, the higher ego's likelihood to engage in non-institutionalized participation (H3b). Marien, Hooghe, and Quintelier (2010) suggested nonetheless that non-institutionalized types of participation strongly reduce or even reverse gender and age inequalities but increase patterns of inequality due to education. From that, we may expect to see a slightly different pattern for gender/age and education, especially when related to noninstitutionalized participation.

Data
The survey method is a common technique to collect and generate data on personal networks (Bernard et al. 1990;Marsden 1990;Crossley 2015;Perry, Pescosolido, and Borgatti 2018) and political participation (Gibson and Cantijoch 2013;Vrablikova 2014). Therefore, in order to get insight into the mobilizing effect of personal networks on the individuals' decision to participate politically, we rely on a unique online survey conducted via Qualtrics, which is both a software and a platform that provides panels of online survey respondents. Our online survey was conducted between June and July 2016 among a quota sample of 2801 Belgian citizens (quotas based on region, age and gender) retrieved from Qualtrics' panels. The sample offers a good representation of the population, with a slight underrepresentation of younger respondents (Table 1). The survey was administered in Dutch and French and conducted in-between electoral cycles to minimize the impact of additional external stimulus on participatory behaviours.

Measurement and description of the dependent variable
Over time, surveys have extended questions related to participatory behaviours, from asking about voting and party-related activities, to non-institutionalized types of political activities like protest, boycott or petitioning (Brady, Verba, and Schlozman 1995;Norris 2002;Vrablikova 2014), and more recently, online types of political engagement (social media like/following, etc.) (Gibson and Cantijoch 2013). Our questionnaire included items relating to these three types of participation (institutionalized, non-institutionalized, and online), which are now extensively covered by most comparative political surveys. We followed the wording of the European Social Survey Programme (ESS8-2016, B15-22) and adapted it in order to measure both the past and future participatory behaviours of our respondents.
Our main dependent variable measures the respondent's intention to engage in political participation (ego's participation intention). We focus on ego's intention to participate, so that alters' participation in the past can be assumed to precede and predict ego's future intention to participate. We measured this intention with following question: 'Could you please indicate whether it is likely or not that you will engage in the following activities in the future?' (likely/not likely): (1) sign a petition (paper or online); (2) take part in a protest or public meeting; (3) boycott or buy on purpose specific products for political, ethical or environmental reasons; (4) be active for a party and/or a candidate; (5) like or become friend with a candidate or a party on social media (FB or Twitter); (6) give or share political opinions online (on social media or traditional media websites). Note that voting was not included as an item, as voting is compulsory in Belgium. As far as the distribution for each political activity is concerned, Figure 1 emphasizes that petition and boycott were the two most selected items among prospective participants, whereas party activism attracted much less intention.
All in all, 77.5% of our sample reported an intention to participate in (at least one) political activity. This figure indicates a relatively high proportion of prospective participants. We ran a principal component analysis (PCA) 1 bound to three dimensions that match the three types of activities identified in the literature and existing surveys: non-institutionalized activities (petition, protest and boycott), online activities (online following, opinion sharing), and institutionalized activities (party activism). These three dimensions of intention to participate were then used as our three (continuous) dependent variables in the analyses.
Part of these prospective participants were already politically active at the time of the survey (70.5% of the respondents reported to have undertaken at least one activity within the year preceding the survey, see the distribution per types of activity in Appendix 1). Our analyses therefore control for the respondents' past participatory behaviours (see our modelling strategy below). This allows to assess to what extent personal networks are a mobilizing factor for inactive citizens (recruitment), but also a factor that affects the intention to continue being active (retention), and contribute to the debate on novices and repeaters (Saunders et al. 2012).

Measurement and description of the independent variables
In social networks analysis, there are two ways to measure an individual's network. The personal (egocentric) network approach looks at one individual (ego) and its direct personal network, such as family, neighbours, colleagues, etc. (alters), as opposed to the whole (sociocentric) approach that looks at the entire relational structure of bounded groups. In this study, we focus on personal networks even if we acknowledge that a respondent's personal network is only a part of sets of relationships (Hâncean, Molina, and Lubbers 2016). Personal network data can easily be collected via quantitative survey designs (Crossley 2015). The most frequent technique is to ask respondents (1) to raise a list of names of people who satisfy a certain definition of a social relationship (name-generater), (2) to collect information about these people (name-interepreter) and (3) to inform how they are tied to each other (name-interrelater). Personal network data have been collected through surveys since the beginning of the 1960s (e.g. Wellman 1979;Laumann 1973) and were systematized in the 1980s. Burt (1984) was the first to include questions documenting personal networks in the US General Social and Election Surveys. Since then, the technique has been widely developed and accepted, also in online settings (Manfreda, Vehovar, and Hlebec 2014;Crossley 2015;Eagle and Proeschold-Bell 2015;Perry, Pescosolido, and Borgatti 2018).
In order to measure our respondents' personal network, we included a name-generating procedure in our survey. The first step asked our respondents to list the first names of a maximum of ten individuals who they consider important in their social life and regularly interacted with over the last year (name-generator). This treshold was set to map ego's proximate network of 'significant others' (Burt 1984;Bidart and Charbonneau 2011;Crossley 2015;Lin 2001), i.e. individuals with the greatest impact on ego's political attitudes and behaviours. Network studies estimate that ego discusses important matters with less than a dozen people (Degenne and Forsé 1994) and naming between 5 and 10 alters was demonstrated to be enough to capture the network of significant others (Kogovsek and Hlebec 2019;Merluzzi and Burt 2013). Furthermore, any increase in the number of alters to be mentioned by ego increases the cognitive burden of filling in the questionnaire, especially for online surveys more prone to dropout (Ferligoj and  Hlebec 1999; Manfreda, Vehovar, and Hlebec 2014;Marin and Hampton 2007;Marsden 2011).
Our dataset excludes respondents who did not mention any personal network, as well as invalid mentions (typos rather than names) (N = 99). Table 2 reports the distribution of our sample on the number of alters reported by the respondents. On average, respondents report 6.2 alters, i.e. important persons with whom they have interacted in the last year. Only 14.1% of the sample named only one alter, and more than a third mentioned 10 names. While this upper threshold might be seen as a limitation, we argue that it is a good tradeoff between getting an estimate of respondents' proximate network and getting them to provide reliable information about that proximate network. The central explanatory factors in this study are these information about the alters, rather than the structural properties of the personal network (which are only included as control variables).
The second step of the name generating procedure asked respondents to report the type of social relations that links them with these alters. We also asked respondents to report social and political characteristics about these alters, using the same questions asked for ego. We focused on information that respondents were most likely to know and left aside other predictors of participation to minimize false reporting, biases of social projection and subjectivity (Crossley 2015;Aeby 2016). While some of these biases may remain, network scholars have stressed that social influence is not always driven by real and accurate knowledge of what alters do politically, but rather by subjective and normative perceptions (Gerber and Rogers 2009). Hence, we assume that information reported by respondents for their personal network, real or biased, work as cognitive heuristics which orientate respondents' own future behaviours.
In order to test our first two hypotheses (H1 and H2), we measured ego's exposure to a specific type of political participation in their personal network. In the survey, respondents were asked 'Could you indicate if the members of your social network have engaged in the following activities in the last 12 months?' (Yes/No/Don't know): (1) signed a petition (paper or online); (2) taken part in a protest or public meeting; (3) boycotted or bought on purpose specific products for political, ethical or environmental reasons; (4) been active for a party and/or a candidate; (5) liked or became friend with a candidate or a party on social media (FB or Twitter); (6) gave or shared political opinions online (on social media or traditional media websites).
Our first set of independent variables, exposures to alters' participation, measure the proportion of alters who have engaged in each political activity (N of alters engaged in the activity in relation to total N of alters). We have computed a rate of exposure that ranges from 0 (no alter engaged in the political activity) to 100 (all alters engaged in the political activity). When looking at the mean exposure rates presented in Table 3, the rank order of frequency of political activities is identical than the one presented for ego (Figure 1). Petitioning is the political act that our respondents are the most exposed to through their alters (34.1%). It means that, for a network of 10 alters, on average at least three of them are reported to have undertaken this specific activity. It is followed by boycott (22.6%), online opinion sharing (19.8%) and protest (18.7%). Respondents are less exposed to party-based activities like online or offline party activism, as the lower mean exposure rates in Table 3 suggests. This is not surprising since this type of participation is more time-consuming and is therefore less undertook by citizens, translating into lower proportion of both egos and alters active in party-based activities (see the distribution in Appendix 1). Furthermore, party-based participation is overall in decline (Van Biezen, Mair, and Poguntke 2012), while citizens increasingly engage in non-institutionalized types of participation (Norris 2002;Hooghe, Oser, and Marien 2016;Quaranta 2018;Portos, Bosi, and Zamponi 2020). Overall, respondents report lower levels of political participation for their alters than for themselves. This may reveal that social desirability weights heavier on the respondents' self-reporting than on their reporting of their alters' participation, or that some of the respondents are not fully informed of their alters' participatory behaviours.
Like for the dependent variable, we ran a principal component analysis (PCA) 2 on the six political activities with the three fixed factors. It allows us to compute our three independent variables, measuring exposure to three types of political activities: non-institutionalized activities (petition, protest and boycott), online activities (online following, opinion sharing), and institutionalized activities (party activism). In addition, since about a third of the respondents (951 out of 2801) reported none of their alters as engaged in political activities, we computed an additional independent variable, exposure to political apathy, which takes the value of 1 if all the alters are politically inactive, and 0 if not.
The third hypothesis (H3a and H3b) requires to measure the social and political profile of alters. Therefore, our second set of independent variables focuses on the characteristics of the alters, and more specifically their socio-demographic characteristics (age, gender and education) and their reported attitudes towards politics (political interest and party identification). We concentrate on exposure rates to 'high' levels of socio-demographic and attitudinal predispositions that are known to increase the odds of institutionalized participation at the individual level (Van Haute and Gauja 2015; Campbell 2013). Therefore, we dichotomized alters' socio-demographic characteristics to oppose lower and higher levels of resources among alters, to test H3. Table 4 introduces alters' characteristics, measured as the proportion of alters in the respondent's network who are reported to have these predispositions towards insitutionalized politics (N of alters with the characteristics in relation to the total N of alters). As the table shows, the networks of our respondents are quite well distributed across gender, age and education. The mean values suggest that a half of the network is male, older than 45 and holds a high school or university degree. In terms of political attitudes, the average proportion of alters that are perceived as politically interested or as identifying with a party is around 40%. It means that, if someone has named 10 alters, (s)he reports on average 4 of them as politically interested and/or identified with a party. The distribution plots of these variables (Appendix 2) display more variation, meaning that the social and political characteristics of ego's network differs across individuals.
Finally, the last step of our name-generating procedure gave the respondents the opportunity to report in a matrix box whether their alters knew each other (name-interrelater). From that information, an index of network density was computed by dividing the number of actual connections within the network by the number of potential connections. It ranges from 0 (for a network where none of the alters know each other) to 1 (all alters know each other).
The mean value of the density index is 0.5 (SD = .34), suggesting that on average one in two alters in our respondents' personal network know each other. It is not surprising given the nature of the personal networks that are measured (significant others). The distribution shown in Table 5 underlines that density varies from one personal network to another. In this study, density is used as control variable in the analyses. For larger networks the number of ties is usually preferred to density, as density is highly affected by the network size (Crossley 2015). However, in our case the difference is limited given the limited size of ego networks.

Modelling strategy
We used a network analysis software to compute our network measures, exposure to alters' participation and socio-demographic characteristics (independent vairables) and network size and density (control variables). Then, we estimated their effect on ego's intention to participate through a multilinear regression model run in Stata. Our model predicts the likelihood of a respondent's likelihood of undertaking the three participation types. The model first introduces the independent variables that relates to exposure to alters' participation types (H1 & H2) and then those related to alters' social and political profile (H3a & b). Finally, we control for ego's past participatory behaviours, network size and density, as well as other individual predictors of participation (age, gender, education, political interest, political satisfaction and party identification, all measured for ego (see the Appendix 5 for the operationalization and the summary statistics). In addition, we ran the same analysis using a split sample to test our results across two groups of participants: respondents who are currently inactive but report an intention to participate in the future (recruitment of novices), and respondents who were already politically active and intend to continue being active in the future (retention of repeaters). This allows us to differentiate the recruitment vs. retention power of personal networks. Table 6 presents the outcomes of the models. A first interesting result is the relatively high R square when only including the personal network variables, especially for non-institutinalized (17.5%) and online (15.1%) participation. The predictive power of the personal network characteristics is lower for institutionalized participation (4.5%). It means that the (reported) participatory habits and social and political characteristics of alters is a good predictor of ego's likelihood to engage in political activities, especially for non-institutionalized and online activities. Galesic (2008) had emphasized the role of personal networks on vote choices. Our analysis suggests that this finding extends to other types of political participation as well.

Multivariate analyses
The results support our first expectation across the three participation types: the higher the exposure to one type of political participation, the higher ego's intention to engage in that same type of political participation (H1 supported). It is true for all types of participation (non-institutionalized, online, and institutionalized). In other words, personal networks contribute to a dynamics where citizens consider opting for political behaviours that match their networks' behaviours. The larger the proportion of alters engaged in one type of political activity, the higher ego's intention to behave similarly and engage in the same type of political activity in the future. This suggests that dominant political behaviour(s) in a network diffuse within the network as by product of social influence. Networks act as triggers of mobilization. Previous studies had emphasized this trigger effect of networks (Diani 2004;McClurg 2003;Sinclair 2012). What our study adds is evidence of mimicking behaviours leading ego to favour some types of participation over others, thereby contributing to a better understanding of how individuals operate a choice among an action repertoire. When controlling for ego's current level of participation, network size and density, and other individual predictors, the effect of exposure to alters' participation remains stable and significant. It means that, controlling for ego's current level of political engagement in all three types of political activities, being exposed to alters who are actively engaged in a specific type of participation is related to a higher intention for ego to engage in the same type of activity in the future. The effect is even reinforced for ego's intention to engage in institutionalized participation (higher regression coefficient in the full model with controls). We find a similar effect for non-participation: exposure to a fully inactive network significantly decreases ego's intention to participate politically in all three types of activities, although it remains statistically significant only for non-institutionalized participation once the control variables are included in the model. This is probably due to the correlation between political apathy and socio-demographic characteristics.
Beyond the direct effect of exposure by type of participation, the results plotted in Figure 2 stress further cross-over effects across types of participation, as expected by H2. Exposure to institutionalized and online participation decreases ego's intention to take part in non-institutionalized participation, and this holds when controlling for individual factors. However, we found no evidence that the relationship works in the other direction: exposure to non-institutionalized participation does not affect intentions to engage in institutionalized or online participation. Moreover, the exposure to institutionalized and online participation are both positively related to each other, albeit not in a significant way after introducing control variables. This is probably because the cumulative cross-over effect is not only present in ego's personal network, but also in ego's current participatory behaviours, the control variables capturing the main effect. The close connection and cumulative effect between institutionalized and online participation may be due to the fact that one of the item measuring online participation is highly party oriented. Overall, our findings confirm a normative 'dissuasive' power of personal networks (Roblain et al. 2020) that keeps ego away from certain participatory behaviours, but only in one direction: exposure to institutionalized and online participation decreases ego's likelihood to be active in non-institutionalized politics, but not vice-versa. This means that H2 is only partially supported. These findings complement Dalton's research (2008) on citizenship norms and participation and studies on the cumulative aspect of participation (Marien, Hooghe, and Quintelier 2010), by showing the role of meso-level, network factors in these dynamics.
Our second set of independent variables, i.e. the exposure to alters' social and political characteristics, have overall a weak effect on ego's intention to participate. Furthermore, the effects do not go in the expected direction, with exposure to resources being linked to a higher likelihood to engage in institutionalized participation (H3a) and a lower likelihood to engage in non-institutionalized participation (H3b). Rather, social and political resources of alters play in different ways. The social composition of ego's network mostly affects ego's intention to engage in non-institutionalized participation: being exposed to a network that is more female and better educated slightly increases ego's likelihood of engaging in non-institutionalized participation, even when control variables are included. The positive relationship between gender and non-institutionalized participation has been shown at the individual level (Marien et al. 2010;Memoli 2016); our study confirms it at the group level, further substantiating the idea that non-institutionalized politics is favourable to female involvement and reduces gender inequalities inherent to political participation. Similarly, our study confirms the positive relationship between education and non-institutionalized participation (Marien, Hooghe, and Quintelier 2010) at the group level. Furthermore, we find that the exposure to older alters slightly decreases ego's online participation, even when controlling for individual factors. Our study confirms the digital divide when it comes to age and online politics.
Finally, the political composition of ego's network also matters for ego's intention to engage in political activities. Being exposed to politically interested alters increases ego's likelihood to engage in non-institutionalized and online activities (but it is not significant once the controls are included); being exposed to party identifiers increases ego's likelihood to engage in institutionalized politics (even with controls). Exposure to alters' party identification negatively impacts non-insitutionalized participation and is positively related to online participation, but the effects do not remain significant when the controls are included. Overall, alters' partisanship matters more than their level of politicla interest, and it reinforces institutionalized participation but deters from non-institutionalized participation, thereby operating in a similar way as exposure to institutionalized politics.
Regarding our control variables, interestingly we can see from the full model in Appendix 3 that people already engaged in institutionalized politics are less likely to continue in the future. This result stresses a major retention issue faced by most political parties, as already emphasized at the individual level by Pettitt (2020). Figure 3 was plotted on the basis of the outcomes of the split sample model (see Appendix 4). It gives support to the three following findings.

Robustness check
The positive effect of exposure to alters' participation is statistically significant for the recruitment and the retention of ego in all three types of participation. It means that alters' participation acts as a trigger for being newly recruited in the same political activities, but also favours retention of ego's engagement in these political activities. Interestingly, this robustness check systematically shows a much stronger effect of exposure to alters' participation on ego's recruitment (blue dots) than on their retention (red dots). This is especially the case for insitutionalized participation. Exposure to alters' institutionalized participation plays less in explaining ego's continuing intention to participate in institutionalized politics. It means that even with a personal network engaged in institutionalized politics, a respondent who is already engaged in this type of participation is more difficult to retain. It highlights retention difficulties for parties and can bring new insights into the debate on membership decline (Pettitt 2020).
We also find additional support for the cross-over effects. The plots show that exposure to institutionalized participation decreases above all the recruitment in non-institutionalized politics, and to a lesser extent retention in non-institutionalized an online participation. Exposure to (non)-institutionalized participation decreases mostly the retention in online participation. And finally, exposure to online participation increases recruitment in institutionalized politics and vice-versa, although these relationships do not hold for retention. It means that being exposed to alters who share political opinions and support parties and candidates online increases ego's likelihood to become active in a party offline. This is in line with other studies showing a spillover effect between online and offline spheres (Conroy, Feezell, and Guerrero 2012;Gil de Zúñiga, Jung, and Valenzuela 2012;Vissers and Stolle 2014;Lane et al. 2017). Overall, these findings point to the importance of distinguishing between first-time participants and repeaters, as already pointed by Saunders et al. (2012) for protest participation.
Finally, the magnitude of the effect of exposure to alters' characteristics is not more substantial when the sample is splitted. It nonetheless supports the two observations made earlier: exposure to male alters decreases ego's likelihood of retention in non-institutionalized politics, whereas exposure to higher education categories increases the recruitment in non-institutionalized politics. Additionally, we find that exposure to alters who feel close to a party positively affects only the propensity to be recruited in institutionalized politics, but not retention.

Conclusion
This paper looked at the mobilizing effect of personal networks on the individual propensity to favour some types of political participation over others. Theoretically, the paper fills a gap in the literature that tends to focus on explaining the intensity rather than the nature of participatory behaviours. It also offers a much-needed focus on the role of personal networks to complement micro-and macro-level approaches. Empirically, the paper relies on original personal network data gathered via a unique online survey conducted among a quota sample of 2801 Belgian citizens.
Our analyses point to four main findings. First, our results confirm the interest of adopting a personal network approach to political participation. The participatory habits and, to a lesser extent, the social and political characteristics of alters are good predictors of ego's likelihood ito engage in political activities, and so across the three types of participation considered in the study. It validates the interest of a meso-level perspective that focuses on personal networks.
Second, our results confirm that the higher the exposure to one type of political participation, the higher ego's intention to engage in that same type of political participation. It is true for all types of participation, even when distinguishing between newly recruited participants and those who are retained. In other words, citizens consider opting for political behaviour(s) that match their networks' behaviours, pointing to a process of diffusion as byproduct of social proximity and influence. Networks act as triggers of mobilization but also matter for the retention of participants.
Third, our findings also point to cross-over effects across types of participation. We emphasize a normative 'dissuasive' power of personal networks that keeps citizens away from certain participatory behaviours. The most robust cross-effect goes in the following direction: citizens who are surrounded by alters engaged in the two other types of participation are pushed away from non-institutionalized participation. Additionnally, we show that exposure to institutionalized participation negatively affects the recruitment in non-institutionalized activities, whereas exposure to online participation affects retention the most.
Finally, we show that the social composition of ego's network affects ego's intention to engage in non-institutionalized participation. Being integrated in a network with more women and alters with higher levels of education increases the likelihood of ego's participation in non-institutionalized activities. We show that alters' gender impacts retention, while education affects recruitment in non-insitutionalized politics. Moreover, older personal networks are related to lower likelihood of online participation, stressing a digital divide. We thus substantiate at the group level prior findings found at the individual level. The political composition of the network affects ego's intention to engage in institutionalized participation: exposure to party identifiers favours institutionalized participation, and above all the recruitment of new participants in this type of participation.
Interestingly, our study points toward major retention issues for political parties. It sheds light on the potential reasons behind the decline in institutionalized participation: social reproduction leads to homogeneous networks that make renewal difficult. At the same time, our findings also point to potential solutions to these difficulties. For example, online strategies could be an avenue to revitilize institutionalized politics, as exposure to alters' online participation positively affects citizens' intention to be active in a party.
More largely, we show that the way people participate in politics is not determined in a social vacuum but depends strongly on what proximate, significant others are and do. In doing so, the paper extends the existing knowledge in two directions. First, it provides a better understanding of how citizens make a choice within the action repertoire that is offered to them, of how they favour some types of political participation over others. Second, it contributes to explaining why, in a given macro context and facing an equal level of resources, some citizens (continue to) engage in politics while others won't. It highlights the importance of personal networks as triggers of participation, i.e. the factors that turn potential participators into politically active citizens. It also points to boosters of retention, i.e. factors that encourage already active citizens to renew their political engagement.

Disclosure statement
No potential conflict of interest was reported by the author(s).