The impact of organisational attributes on political participation: results of a multi-level survey from Switzerland

Authors


Abstract

Abstract: The notion of civil society associations as schools of democracy has resulted in models of political participation that place an emphasis on membership in civic associations as a means of developing personal skills that are conducive to political participation. These suppositions seem well established. It is still contested, however, to what extent the particular characteristics of the associations that offer such membership have an impact on civic engagement. Expanding recent research that mainly builds on group characteristics aggregated from the characteristics of the associations’ members, we apply the Swiss part of a unique multi-level data set, the CID-data, which provides information about approx. 1000 activists from about 400 associations. This data allows us to measure group characteristics, such as the function of an association and its connection to the local elite, directly and thus provides us with a special opportunity for a multi-level analysis of activists nested in organisations.

Introduction1

Two of the most central questions of political science are “why do people participate in politics at all?” and “which factors can explain their motivation to vote, protest and articulate their opinion?”; one of the most popular explanatory factors is membership in voluntary associations. These are supposed to serve as “schools of democracy” (Tocqueville, 2002).

A substantial body of literature tries to prove this claim using elaborate data sets, however, it has to date never been fully corroborated. In light of this, we would like to contribute to this literature in two ways: first, we apply a unique data set, namely the Swiss Citizen Involvement and Democracy (CID) survey, which is composed of linked samples of organisations and their members or “activists”. This allows us to employ a hierarchical regression model that combines independently measured organisational characteristics with individual level data. Second, in doing so, we establish that the degree to which associations are embedded in a local context, and particularly the degree to which they are connected to the local elite, is an essential factor regarding their ability to foster political participation. In addition, we also show that organisational influence is linked to the goal of the organisation, although in a less direct way than is commonly expected.

Membership in organisations mobilises, enhances individual skills and influences attitudes

The early socio-economic-model (SES) of political participation (Verba and Nie, 1972) developed from the presumption that the willingness to participate in political life depends mostly and directly on the socio-economic status. This over-simplified explanation was challenged by the “Civic Voluntarism Model” of Verba et al. (1995) and Brady et al. (1995), who alluded to an old claim of de Tocqueville (2002), namely the importance of civil society organisations as a playground in which democrats can emerge, and thus to their function as “schools of democracy”. One of the main endeavours of Verba et al. (1995) was in fact to develop a model about the learning of “civic skills” relevant to political participation. While basic communicational and organisational skills are assumed to be acquired in school, and hence can be operationalised as the amount of education received, further abilities will then be obtained as an adult at work2 or through membership in voluntary organisations (Brady et al., 1995), where individuals engage in activities (“skill-acts”) such as the organisation of meetings or giving speeches.

In the study by Verba et al. (2000), the measure for “skill-acts” actually accounts for a substantial amount of the variance of political participation, relative to either psychological (beliefs) or economic (costs) factors. Consequently, the authors saw this as evidence for de Tocqueville’s claim regarding the importance of civil society organisations for the development of democratic virtues. This is also supported in more recent studies (von Erlach, 2005; Howard and Gilbert, 2008).

In addition to opportunities for training skills, voluntary organisations may also provide a suitable setting for the recruitment and mobilisation of additional members (Leighley, 1996; Teorell, 2003). Further, membership in an organisation most probably also has socialisation effects (Hooghe, 2003; Mutz, 2002), due to, for example, the interaction between members.3

By employing the population data of the Swiss CID survey (see footnote 7), von Erlach (2005) attempted to simultaneously test many of these possible effects of organisational membership. While he exclusively focused on political discussion as a form of political participation, he was able to confirm that exercising particular tasks either at the workplace or in a voluntary organisation, and being exposed to “cross-pressures” as a member, increases the propensity for political deliberation.

Membership in organisations enhances the size and quality of individual social networks

In reaction to the skills argument, and very closely related to the socialisation-hypothesis discussed above, Lake and Huckfeldt (1998) claimed that, in addition to any skills acquired, it is also the quality of a person’s social relations that are influenced by membership in civic organisations. The quality of the networks that are offered by such membership has been of strong interest to scholars engaged in social capital research, inspired by Coleman (1990) and propagated by Putnam (1993, 2000). One prominent approach has been to differentiate between “bonding” and “bridging” associations, thereby distinguishing organisations that are rather homogenous from those that have a heterogeneous membership structure (Coffé and Geys, 2007; Freitag et al., 2009), or by observing multiple memberships of the single members (Freitag et al., 2009; Paxton, 2007). Especially the latter has been found by some studies to make a difference with respect to political participation (Djupe and Gilbert, 2006; Mutz, 2002; Quintelier, 2008; Teorell, 2003).

Yet, even a decade after Lake and Huckfeldt’s (1998) claim, Quintelier (2008) admits that it remains unclear, why exactly multiple memberships foster political participation more than, for example, the time spent in one organisation, suggesting further that the former effect depends also on the type of organisation. This leaves us with the question which differentiation of the organisations (by their goals or some other characteristic) helps to explain political participation. In the remainder of this text, we propose local embedment and two organisation typologies as possibilities for such a differentiation.

How embedded organisations support civic engagement in the local context

One possible solution to the puzzle of the effect of multiple memberships on political participation is suggested by Wollebaek and Strømsnes (2008): challenging the socialisation theory, they suggest that a dense network of voluntary associations that is visible within the community, demonstrates to members and non-members alike the merits of collective action and civic engagement. Therefore, while multiple memberships increase the density of the organisational network, it is also the embedment of the organisations in the local context that strengthens civil society. This corresponds to the neo-institutionalist approach propagated by Stadelmann-Steffen and Freitag (2011), who explore how the political and institutional context shapes individual preferences and behavioural options. The underlying assumption is that those citizens, that are members of organisations which have relations to (local) state institutions (for whatever reason), will have better opportunities to interact with and hence learn about the institutional context they live in. This perspective similarly integrates with earlier sociological studies showing that having a central position in local networks (which implies being connected to local state actors also) is a success factor for voluntary organisations with respect to goal achievement and influence (compare the review article by Galaskiewicz, 1985: 294).

Following these approaches, we would expect that a useful distinction between organisations that are more or less successful in fostering individual political participation lies in their “visibility” and embedment in the (local) public sphere. For Switzerland and West Germany, for example, Dayican et al. (2010) found that connections to the local political elite, provided through such an organisation, affected subjectively perceived “political empowerment”.4 Similar positive effects on various forms of political participation were established by LeRoux (2007) in relation to organisations that receive government funding. Closely related to that argument is that by Han et al. (2011), who demonstrate how organisations with skilled and committed leaders have more political presence and thus supposedly should be better able to support individual political participation.5

Accordingly, we hypothesise that associations function as a bridge between individual citizens and the local elite and consequently, that membership in associations that have strong ties to the local political elite offers better opportunities to become engaged. From such contact, the members of these associations learn that they are indeed taken seriously and that it thus makes a difference whether one expresses one’s preferences (to the local elite) or not. For example, if we take the case of local parties, such as the “Social-Democratic Women of Berne”, or that of special migrant groups’ associations, such as the “Association Venezolana Suiza”, which serves as a meeting point for people of Venezuelan origin, it is plausible that such associations have closer links to local authorities than cultural associations or sports clubs, probably due to their representational function of certain societal interests.

How the goals or functions of the organisations might make a difference

While associations can be distinguished by their closeness to politics6, they also differ in terms of their individual goals and the specific sector they are active in.

Such more general typologies of organisations have always been suspected to be relevant with respect to the political activities of their members. Brady et al. (1995) found that civic engagement was supported by all sorts of associations, except in the case of sports clubs; they even discovered a negative association between a strong emphasis on sports in school and later civic engagement. Indeed, some sports associations do not seem to offer any opportunities to develop civic skills, given that their goals are largely predetermined and, in the case of high school sports, any planning is carried out by adults (Kirlin, 2002). Furthermore, sports clubs can also exacerbate gender and racial prejudices, and thus may even impact negatively on the democratic understanding or the level of civic skills among their members (Dyreson, 2001). There is, however, also counter-evidence (Auld, 2007), namely that sports clubs in disadvantaged communities actually build social capital and foster community development, as Skinner et al. (2008) showed on the basis of several case studies from Anglosaxon countries (compare also Groeneveld et al. (2010) for European countries and Harvey et al. (2007) for Canada). However, most of these studies fail to compare sports associations to other voluntary organisations, a point taken up by Seippel (2006) in a comparative study in which he shows that the effect for sports organisations is weaker than for other voluntary associations but stronger when citizens are not only members in a sports organisations but in a couple of other associations (multiple membership).

In accordance with the latter finding, purely empirical association typologies have not led to clear results. Based on the CID-survey, Maloney et al. (2008), in their study of organisations in Aberdeen and Mannheim, tested an empirically derived typology that comprises ten types of associations, ranging from family to sports to religious organisations, and found no evidence for the impact on political participation. This result was largely confirmed by van Deth (2010a)– using data of the European Social Survey – who showed that only religious organisations might cause an adverse effect on satisfaction with democracy and political engagement. Van Deth (2010b) also specifically examined political discussion and political interest but found a similarly negligible effect of organisational characteristics. Finally, linking such empirical typologies to concepts such as “bonding” and “bridging” types of organisations (Garcia Albacete, 2010) has not led to any further insights either.7

Instead, Kriesi and Baglioni (2003) suggested a theoretically more convincing typology that is based on two orthogonal dimensions of the functions these organisations usually accomplish: the first dimension differentiates between “services for constituencies” and “advocacy” (mediation between government and constituencies), the second between low and high membership involvement. These two dimensions thus result in four types of organisations:

  • 1) “Service” associations: these deliver professional services (e.g. medical assistance) and are mainly led by a small group of highly qualified and specialised people, without broad membership involvement.
  • 2) “Activating” associations: these are service organisations, such as sports and cultural clubs, with high membership involvement.
  • 3) Organisations for “representation”: these take the form of parties and interest groups, which, operating mostly as advocacy associations, are led by a small group of activists, and thus have only limited membership involvement.
  • 4) Organisations for “mobilisation”: these are advocacy associations that pursue some broader interest, and, being dependent on the mobilisation of large groups, entail high membership involvement.

We prefer this typology because it contains fewer dimensions, yet a more stringent link to political participation. Obviously, we expect organisations for “representation”, and possibly also for “mobilisation”, to better prepare their members for political participation, while “activating” and “service” organisations are not expected to have a remarkable impact.

Method and Data

We use a multi-level regression approach to establish the impact of the types of organisations and of the degree to which they are embedded at the local level on a composite index of political participation, while accounting for the control variables that are suggested by the literature in this field.

To achieve this, we draw on the Swiss “activist survey” of the CID-data.8 This data consists of linked samples of organisations and their “activists” from eight Swiss municipalities of various sizes. It was intended, to actively identify the participants of the “activist sample” by consulting the members list of the organisations. Many organisations, however, did not agree with this procedure and decided to distribute the questionnaires among their members on their own (Baglioni, 2004, 2007).9 Hence, not all of the questionnaires that were returned were actually filled out by “activists” in the more narrow sense.10

The survey contained 20 questions that could be answered with either “yes” or “no” in relation to participation in a variety of political activities11, these ranging from participation in protests to participation in elections, thereby offering several opportunities to measure political participation using this data. The relevant variables are listed in the second table12.

Based on this list of “political participation” variables, we construct the dependent variable in two different ways: first, we conduct an exploratory factor analysis13 (principal components) with orthogonal rotation.14 This produces three factors of fairly evenly distributed explanatory power that together account for about 40% of the variance in the data (Table 1), as well as one additional factor with a somewhat smaller variance component. Table 2 lists the factor loadings of the variables employed for the four factors. While the first factor loads rather highly on virtually all variables that measure direct contact to politicians and bureaucrats, the second factor loads on variables that measure “expressive” but probably less personal forms of participation, such as demonstrations. The third factor loads on variables measuring less expressive but more anonymous forms of participation, such as participation in boycotts. The fourth factor loads highly on one variable only, namely on the one that measures fundraising activity. Table 3 lists the descriptive statistics for the three first components, also differentiated by issue domain the organisation is active in. Obviously, sets of members from “representing” organisations have a larger mean across all forms of participation than members from other organisations.

Table 1.   Principal-components factor analysis: variance proportions of the first four factors
FactorVarianceProportionCumulative
Factor13.0830.1620.162
Factor22.4800.1310.293
Factor32.0910.1100.403
Factor40.3710.0750.478
Table 2.   Factor loadings (only those greater than 0.3 are listed)1
VariableFactor 1 (contact with elite)Factor 2 (expressive form of participation)Factor 3 (quiet form of participation)Factor 4 (fundraising)
  1. 1 The results remain rather ambiguous for the “Worked for a party” and the “Others” variables as well as for the variable on participation in elections. However, for the latter two, this is not very surprising (compare also the following footnote of this table).

  2. 2 The variable regarding participation in elections and referendums had not been asked as a part of the item battery for political participation, but appeared later in the questionnaire.

Contacted politician0.7442   
Contacted organisation0.5860   
Contacted civil servant0.7204   
Worked for party0.55310.5240  
Worked for other organisation0.4801  0.3322
Contacted media0.5080  0.3444
Worked for political group 0.4022  
Wore badge 0.6940  
Participated in public demonstration 0.6820 0.3146
Participated in strike 0.5544  
Participated in rally0.44820.6478  
Contacted solicitor 0.3138  
Signed petition 0.30410.5202 
Participated in boycott  0.7031 
Bought certain products  0.7835 
Donated  0.5391 
Helped in fundraising   0.6700
Others0.4496  0.4744
Participated in elections, referendums20.4250 0.4509−0.3800
Table 3.   Descriptive statistics of the dependent variables, grouped by type of organisation
  Composite index (excluding contacts)Composite index (including contacts)First Principal Component “contacting political elite”Second Principal Component “expressive participation”Third Principal Component “anonymous participation”
ActivationMean1.9382.706−0.233−0.263−0.249
MobilisationMean2.5833.489−0.0650.036−0.025
RepresentationMean3.8785.2570.6210.5320.194
ServiceMean2.9543.9950.0560.1750.153
OthersMean2.4433.373−0.084−0.1310.046
TotalMax9.04112.0792.6613.7091.775
 Min00−2.154−1.569−2.081
 Mean2.6153.581000
 Stand. Dev.2.0082.607111

Our second approach to the construction of a dependent variable is to build one single index. In doing so, we weigh each action by its “exclusiveness”, i.e. the share of individuals (activists) who answered the questionnaire and indicated not to have performed the same action. That is, if an individual has participated in protests and signed petitions, the value of his/her action is the sum of the share of individuals who have not taken part in protesting and signing petitions. This approach is more differentiated than indices that just add several dummy variables to account for various types of political participation (McClurg, 2003).

The assumption behind our index is that more “exclusive” forms of political participation require more civic virtue and more moral, as well as probably organisational, support. Since one of our explanatory variables measures the contact an organisation has with the political elite of its municipality, the inclusion of contacts of the individual with the political elite in the dependent variable bears the danger of drawing tautological conclusions. In order to avoid this, we compute two variants of our composite index: in the first, all variables that measure contact between the individual and the elite are excluded, these being the variables “contacted politician/organisation/civil servant/media/solicitor” (see Table 2), most of which load relatively highly on the first dimension of the principal component analysis listed in Table 2. In addition, we also construct an extended second variant of our participation index that includes all variables. However, this is to be used as the dependent variable only for replicating the analysis (see last column of Table 5).

Table 5.   Hierarchical regression models (random intercepts)
 (1)(2)(3)(4)
Composite index (excluding contacts)Composite index (excluding contacts)Composite index (excluding contacts)Composite index (including contact items)
  1. t statistics in parentheses. *< 0.05, **< 0.01, ***< 0.001

  2. If model 2 is replicated based on the sample of models 1 & 3, the results do not change substantially.

Number of memberships (proxy)0.155** (3.00)0.192*** (4.15)0.163** (3.07)0.216*** (3.69)
Early political socialisation0.118 (1.29)0.120 (1.38)0.152 (1.61)0.079 (0.72)
Degree of personal engagement0.373* (2.44)0.338* (2.43)0.436** (2.81)0.478** (2.73)
Amount of time active0.320*** (3.60)0.182* (2.23)0.298** (3.21)0.213* (2.08)
Higher education (dummy)−0.022 (−0.12)0.037 (0.22)0.036 (0.19)0.243 (1.14)
Skills0.160*** (4.16)0.141*** (4.10)0.163*** (4.13)0.250*** (5.79)
Political interest0.688** (3.16)0.782*** (4.04)0.839*** (3.85)1.053*** (4.33)
Male0.364 (1.86)0.161 (0.89)0.398 (1.93)0.224 (0.98)
Age0.005 (0.55)0.010 (1.19)0.004 (0.42)0.016 (1.60)
Employed−0.008 (−0.03)0.184 (0.89)−0.0414 (−0.16)0.475 (1.83)
Perceived influence of own organisation0.036* (2.30)0.042** (2.87)0.045** (2.79)0.070*** (3.81)
German speaking0.0823 (0.45)0.0835 (0.50)0.162 (0.87)0.143 (0.68)
Years being member of organisation−0.0203* (−2.12)−0.0129 (−1.36)−0.0193 (−1.93)−0.0209 (−1.74)
Index for contacts of organisation with local elite0.195*** (3.93)0.180*** (3.59)0.217*** (4.03)0.220*** (3.70)
Number of members of organisation−0.000 (−0.59)−0.000 (−0.50)−0.000 (−1.15)−0.000 (−0.07)
Degree of formalisation of decision making process in organisation−0.634** (−2.75)−0.534* (−2.34)−0.629* (−2.46)−0.527 (−1.95)
Mobilisation (goal)1.463* (2.10)   
Representation (goal)1.351*** (4.18)   
Service (goal)0.233 (0.82)   
Religion (goal)0.166 (0.21)   
Others (goal)1.017* (2.50)   
Mobilisation (function) 0.489 (1.09) 0.582 (1.10)
Representation (function) 0.948* (2.40) 0.957* (2.04)
Service (function) 0.387 (1.23) 0.250 (0.67)
Others (function) 0.404 (1.42) 0.376 (1.11)
Interest (goal)  0.869 (1.90) 
Bridging (goal)  0.332 (0.52) 
Others (goal)  −0.006 (−0.02) 
Constant−3.598*** (−4.78)−3.430*** (−4.76)−3.799*** (−4.83)−4.955*** (−5.59)
var(Organisation)0.230** (−2.60)0.463* (−2.30)0.408* (−2.15)0.504 (−1.61)
var(Residual)1.790*** (5.92)1.918*** (7.51)1.809*** (5.94)3.067*** (12.77)
log-likelihood−486−667−495−725
Chi-Square271.964230.819215.983303.854
p-Value0.0000.0000.0000.000
Number of organisations7711077108
Observations276365276355

It has to be noted that, in general, it is the more “expressive” forms of political participation that are chosen by the least number of people (see last column of Table 2). Hence, our index gives most weight to these types of participation and correspondingly, the second factor of the factor analytic solution mentioned above is highly correlated with our index (r = 0.78). We nonetheless believe, however, that it is worth examining the difference in the results derived from these two dependent variables.

Descriptive statistics of the independent variables used in the analysis are listed in Table 4. The rows contain means and standard deviations for the whole sample as well as for sub-samples of respondents corresponding to Kriesi and Baglioni’s (2003) types of associations.

Table 4. Descriptive statistics of the independent variables, grouped by type of organisation1
 Membership in other organisationsEarly socialisationCommitmentTime invested
MeanSdMeanSdMeanSdMeanSd
Activation1.9811.6052.4370.9463.2420.6663.2651.072
Mobilisation2.4102.1642.5620.9793.1600.7983.0001.162
Representation2.7291.9642.8740.9513.4960.6042.8471.119
Service2.8227.1312.6310.9973.3300.6873.1711.297
Others2.0391.7602.6200.9523.3140.7013.0961.196
Total2.3193.7122.6020.9703.3090.6903.1181.180
 EducationSkillsPolitical InterestMale
MeanMeanSdMeanMean
Activation0.4367.0852.3910.4370.637
Mobilisation0.3667.6723.5090.6250.734
Representation0.4888.0622.8360.8800.567
Service0.5128.0952.6400.7190.455
Others0.4417.4272.7680.5740.498
Total0.4557.6002.7530.6130.555
 AgeEmployedPerceived influence of organisationGerman speaking
MeanSdMeanMeanSdMean
Activation45.15814.0830.7116.9495.0910.479
Mobilisation44.88217.2670.6518.1075.9100.458
Representation46.19414.8460.77014.8236.5940.659
Service46.14414.2790.72410.9446.4950.540
Others45.77915.0630.6368.2615.6860.496
Total45.67214.7820.6959.5206.4540.520
 Length of membershipContacts with local administr.Number of membersDegree of formalisation
MeanSdMeanSdMeanSdMean
  1. 1 Although “early socialisation”, “commitment” and “time invested” are measured on an ordinal scale, we treat these variables as if they were measured on an interval scale, as discussed in the text. For the dichotomous variables, we only list the means.

Activation13.32611.3216.5441.541284.268292.9740.448
Mobilisation16.31614.6597.5821.884437.896563.5570.542
Representation15.73111.5998.8412.8222451.7563717.9880.822
Service11.25311.2937.5332.449763.0391668.6820.366
Others9.3619.7656.4741.981157.254183.5040.289
Total12.27611.4877.1472.270647.1371729.1040.441

The first variable listed in Table 4 is a proxy for the number of associations an individual is engaged in.15 While, on average, people are members of about two (classes of) organisations, this is a little higher in the case of members of associations active in the fields of “Representation” and “Activation”.

To control for some of von Erlach’s (2005) variables, we also tested variables measuring early socialisation at home, the avoidance of conflict and the amount of cross-pressures in the organisation, as well as engagement and integration. In our final model, however, we include only the variables that proved to be statistically significant in one of the models tested, namely “early socialisation” (an ordinal variable measuring different degrees of political discussion at home during adolescence) and “engagement” (an ordinal variable based on a self-assessment of the degree of engagement in the organisation), as well as an ordinal variable that measures the amount of time spent in the organisation during one month, as indicated by the individual him/herself. No big differences can be detected between the mean values of these variables for the different types of organisations listed in Table 4. Since we are interested in these variables only as control variables, we simply treat them as interval scales, as was done in von Erlach’s study (2005).

Another binary control variable measures whether the respondent has a regular (full- or part-time) job or not, which can be interpreted as an SES-variable or as a proxy for the opportunity cost of time.16 On the whole, this seems to be the case for more than two thirds of the respondents in the full sample; it also more or less holds for virtually all sub-groups. Education as a further control variable is covered by a binary measure for standard or higher education (university degree, higher education after standard professional formation). Here, about half of all “activists” have such additional formation, and since education is lowest for individuals in “activating” groups, and highest in the case of those in “other” groups, this indicates that the “other” category might be a rather special assembly of organisations. However, more important for the research question at hand is the additive index that measures the frequency of various skills-acts (charing meetings, writing letters etc.) as indicated by the respondents.17

We also employ a control variable for individual preferences related to political participation, namely whether and how much a respondent is generally interested in politics (a measure that is again well above the mean for participants in “representing” associations and the special group of “other” organisations). In addition, we include measures for age and sex.18

The impact of the organisational level is measured on the basis of several covariates that have been collected through an independent questionnaire that was distributed to the organisations directly. Measurement of the organisations’ characteristics, such as the number of members or the degree of formalisation of the decision-making procedures, was thus carried out exclusively at the organisational level.19 While religious associations are by far the largest organisations, “mobilising” and “service-oriented” organisations have the most formalised structures. It is assumed that the more democratic an association’s structure, the more it fosters the civic engagement of its members, however, this measure is quite a poor proxy since it only captures the number of formalised institutions an organisation comprises (such as an assembly, a board of directors, or a treasurer).

Our first explanatory variable is an additive index that measures whether an organisation has regular or only occasional contact with important actors of the municipality (the local administration, the mayor, members of the local parliament, and local political parties). The organisations could indicate whether they usually have none, occasional or regular contact with each type of actor, and hence the additive index ranges between 4 and 12, with higher numbers indicating more intense connections between an organisation and the local political elite. The expectation is that the more often an organisation makes contact with the local authorities, the more it is embedded in the local context and thus the more its members will engage in political actions. Not surprisingly, the “representing” organisations have the highest mean score. In addition, we employ an individual level variable that closely relates to the organisational level in that it asks how much a respondent believes the organisation to be influential. Here again, the “representing” organisations receive the highest mean score.

Finally, we include indicators for the various operationalisations of the different types of organisations. Our operationalisation of the associational typology (“representation”, “activation”, “service” and “mobilisation”) is relatively straightforward as two questions in the CID-survey directly address this issue: one asked about the primary goal of the organisation and the other about its main activity (or function). We list the 14 respective activity-categories as they were contained in the questionnaire in the Appendix in Table 8, and indicate how we use each category to operationalise the categorical variable by which we measure Kriesi and Baglioni’s (2003) typology. In Table 7, we list all 36 goal-categories, and again allott each goal-category to one category of Kriesi and Baglioni’s, as well as to Garcia Albacete’s (2010), categories. We apply and test these alternative typologies in the following econometric model.

Table 8.   Operationalisations of Kriesi and Baglioni’s (2003) typology of the organisational activity categories (functions)
1representationRepresentation
2mobilisation of membersMobilisation
3self-helpMobilisation
4recreation, sportsActivation
5meetingsActivation
6membership servicesService
7services for othersService
8consultingService
9social integrationActivation
10fund raisingOthers
11recruitment of members and donorsMobilisation
12promotion of voluntary activitiesMobilisation
13promotion of rights/advocacyRepresentation
14othersOthers
Table 7.   Alternative operationalisations of Kriesi and Baglioni’s (2003) and Garcia Albacete’s (2010) typologies of the organisational goal categories
  Mimicking of Typology of Kriesi and BaglioniBridging and Bonding Typology according to Garcia-Albacente
1social serviceServiceBridging
2healthServiceOther
3disabledServiceBonding
4pensionersServiceBonding
5military servicesServiceOther
6religious activitiesReligionBridging
7educationService and RepresentationOther
8povertyService and RepresentationOther
9foreigners, minoritiesService and RepresentationOther
10sportsActivationBridging
11youthActivationBridging
12clubActivationBonding
13parentsService and RepresentationBridging
14culture, musicActivationBridging
15hobbiesActivationBridging
16researchServiceOther
17economic developmentService and RepresentationOther
18environmentMobilisationOther
19animals’ rightsMobilisationBridging
20peaceMobilisationOther
21humanitarianService and RepresentationBridging
22womenRepresentationBonding
23human rightsMobilisationOther
24childrenService and RepresentationOther
25local developmentService and RepresentationOther
26politicalRepresentationBridging
27economicRepresentationInterests
28trade unionRepresentationInterests
29professionalRepresentationInterests
30consumersRepresentationBridging
31familyService and RepresentationOther
32employmentService and RepresentationInterests
33housingService and RepresentationOther
34crimeService and RepresentationOther
35othersOthersOther

Econometric Model

Due to the structure (and the sampling strategy) of the data, we opt to apply a hierarchical random-intercepts regression model, which considers the individual activists as the first and the organisations as the second level.20 Although the organisations would again be nested in municipalities, we drop this third level since the number of municipalities is small and the variance component of this level only minor.

The inclusion of a random intercept term for the organisations is necessary because it enables us to judge by how much the inclusion of covariates at the organisational level reduces the variance of the organisational level relative to the overall variance. This in turn allows us to identify to what extent we are able to capture the role of the organisational characteristics for political participation.

The specification of our random intercepts model can be notated as follows:

image

with

image

Here, Xij denotes characteristics of the individual, such as age and education, and Zj represents the characteristics of the organisation. Since these are measured directly rather than aggregated from the characteristics of the individual members of an organisation, we refrain from using alternative specifications of the hierarchical model, such as those proposed by Seltzer (2004).

However, we list the results for alternative model specifications in the appendix (Table 9), including the model in which we apply grand-mean centered data as well as a simple OLS-model, with the latter including all the variables at all levels as well as cluster-corrected standard errors. We also include a “fixed-effects” model in which we control for clustering effects by estimating separate intercepts for all but one organisation. As Table 9 demonstrates, the results are not driven by a particular specification of the model.

Table 9. OLS Regression and alternative specification of the hierarchical mixed effects model (random intercepts)
 (1)(2)(3)
OLS (cluster correction)Fixed-effects OLS (cluster correction)Grand mean centred
  1. t statistics in parentheses; *< 0.05, **< 0.01, ***< 0.001

  2. The R2 for model 1 is not adjusted, due to the cluster correction of the standard errors.

  3. If models 1 and 2 are replicated based on the sample of model 3, the results do not change substantially.

Number of memberships (proxy)0.176** (2.72)0.150* (2.05)0.026 (0.97)
Early political socialisation0.094 (0.86)0.194 (1.45)0.065 (1.30)
Degree of personal engagement0.279 (1.77)0.406* (2.22)0.178* (2.22)
Amount of time active0.215* (2.16)0.116 (1.06)0.080 (1.63)
Higher education (dummy)0.037 (0.22)0.0311 (0.16)−0.210* (−2.16)
Skills0.157*** (3.92)0.103* (2.22)0.024 (1.19)
Political interest0.853*** (4.31)0.599* (2.56)0.130 (1.18)
Male0.100 (0.50)0.170 (0.64)0.204 (1.79)
Age0.010 (1.11)0.00449 (0.36)0.003 (0.63)
Part-time employed0.252 (1.04)0.116 (0.40)0.205 (1.57)
Perceived influence of own organisation0.051** (2.82)0.0391 (1.56)0.012 (1.43)
German speaking−0.003 (−0.02)−0.0357 (−0.15)−0.011 (−0.11)
Years being member of organisation−0.011 (−1.06)−0.00110 (−0.07)−0.005 (−0.84)
Index for contacts of organisation with local elite0.164** (3.04) 0.107** (3.24)
Number of members of organisation−0.000 (−0.26) 0.000 (0.35)
Degree of formalisation of decision making process in organisation−0.559* (−2.48) −0.295 (−1.95)
Mobilisation0.322 (0.87) −0.120 (−0.57)
Representation0.950* (2.01) 0.167 (0.55)
Service0.305 (1.08) 0.478 (1.89)
Others0.276 (0.99) −0.003 (−0.02)
Constant−3.331*** (−4.97)−1.609 (−1.89)0.054 (0.76)
var(Organisation)  0.285*** (−4.75)
var(Residual)  0.575*** (−6.03)
(Adj.) R20.4460.495 
F-Test11.658***7.252 
log-likelihood  −492
Chi-squared  73.230
p-value  0.000
Observations365445355

Of course, as Freitag et al. (2009) rightly assert, without repeated measurement for the same respondents, it is not possible to assure the direction of the expected causality, due to the possibility of self-selection processes. Since the organisational characteristics were measured in a separate survey that was sent to the organisations directly, it makes sense to correlate the per-organisation mean of the first factor of the factor analytic solution (which loads highly on most contacting activities as listed in Table 2) with the contact-index at the organisational level. This results in a correlation coefficient of only 0.270, however, which does not strongly support the view that the organisational contact index is simply an aggregation of the contact scores of the individual members.

Results

Table 5 compares the results for alternative specifications of our composite index of political participation and for alternative operationalisations of the typology of associations (as listed in Tables 7 and 8 in the appendix).

Column (1) of Table 5 contains the results for the first variant of our composite index. As expected, the variables for individual skills, political interest, the number of memberships, the degree of personal engagement, the time spent in the organisation, as well as the perception of the importance of the organisation, are positively and significantly correlated with our participation index. The remaining variables corresponding to the individual level, most notably education, early socialisation and employment, are not significant. At the organisational level, it is actually the regularity of contact between the organisations and the local elite that is positively and significantly correlated with individual political engagement.21 Interestingly, however, the degree of formalisation of the decision-making procedures within the organisation is significantly and negatively correlated.

As discussed above and as indicated in column (1) of Table 5, we differentiate between types of associations based on their goals. Consistent with our expectations, this shows that individuals who belong to “representing” associations are significantly more likely to participate in political life than individuals who belong to “activating” associations (i.e. sport clubs), these being the reference category. The same holds, to a similar degree, for “mobilising” associations (and the residual “others” category), while “service” and “religious” associations do not support participatory activities of their members significantly more strongly than “activating” associations. All in all, these results are thus supportive of our hypotheses.

Column (2) of Table 5 replicates this analysis using the alternative operationalisation of the typology of associations that is based on their main function. Here, the results are quite the same for the individual-level variables and for the typology, with “representing” associations being the ones whose members show a significantly higher level of political participation than those of the reference category “activating” associations. Interestingly, the organisations with a representative function have a coefficient that is at least two times as high as that of the other types, including those with a mobilising function. This, however, is not very surprising given the operationalisation of this category (see Table 8): “mobilisation” includes the promotion of voluntary work and recruitment activities, whereas “representation” includes the defense of rights, an activity which is possibly more strongly related to more expressive forms of political participation, such as public demonstrations.

Column (3) of Table 5 lists the results for a model that is identical except for the fact that it uses yet another typology of voluntary associations that is again operationalised on the basis of the associations’ goals, but this time builds on the “bridging”–“bonding” dichotomy. Following Garcia Albacente (2010), we create the goal categories as they are listed in the fourth column of Table 6. Unfortunately, however, relative to the reference category (“bonding” associations), neither “interest” nor “bridging” associations prove to have a significantly different impact on the civic engagement of their members, although all coefficients are positive.

Table 6.   Hierarchical regression models (random intercepts)
 (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)
Composite index (excluding contacts variables)First Principal Component “contacting political elite”Second Principal Component “expressive participation”Third Principal Component “anonymous participation”
  1. t statistics in parentheses; *< 0.05, **< 0.01, ***< 0.001. If models 2 & 3 are replicated based on the sample of models 4–12, the results do not change substantially.

Number of memberships (proxy) 0.193*** (4.06)  0.064* (2.53)  0.030 (1.08)  0.070** (2.60) 
Early political socialisation 0.123 (1.38)  −0.044 (−0.92)  0.065 (1.30)  0.142** (2.84) 
Degree of personal engagement 0.345* (2.42)  0.118 (1.55)  0.178* (2.22)  0.116 (1.45) 
Amount of time active 0.178* (2.11)  0.010 (0.22)  0.071 (1.48)  −0.003 (−0.07) 
Higher education (dummy) 0.0365 (0.21)  0.277** (3.00)  −0.215* (−2.20)  0.200* (2.04) 
Skills 0.140*** (3.95)  0.100*** (5.40)  0.024 (1.21)  −0.010 (−0.50) 
Political interest 0.773*** (3.90)  0.514*** (4.86)  0.135 (1.23)  0.428*** (3.82) 
Male 0.166 (0.89)  0.0810 (0.83)  0.120 (1.13)  −0.077 (−0.74) 
Age 0.009 (1.14)  0.012** (2.94)  0.003 (0.69)  0.003 (0.67) 
Part-time employed 0.178 (0.84)  0.240* (2.15)  0.085 (0.71)  −0.060 (−0.50) 
Perceived influence of own organisation 0.041** (2.73)  0.027*** (3.49)  0.012 (1.36)  −0.013 (−1.51) 
German speaking 0.091 (0.53)  −0.008 (−0.09)  −0.010 (−0.10)  0.160 (1.67) 
Years being member of organisation −0.013 (−1.33)  −0.008 (−1.52)  −0.004 (−0.76)  −0.007 (−1.26) 
Index for contacts of organisation with local elite 0.181*** (3.41)  0.0460* (1.98)  0.109** (3.29)  0.042 (1.71) 
Number of members of organisation −0.000 (−0.48)  0.000 (0.11)  0.000 (0.26)  0.000 (0.22) 
Degree of formalisation of decision making process in organisation −0.527* (−2.18)  −0.0743 (−0.70)  −0.302* (−1.99)  0.109 (0.97) 
Mobilisation 0.504 (1.06)  0.183 (0.89)  0.291 (0.99)  0.098 (0.45) 
Representation 0.944* (2.26)  0.172 (0.94)  0.603* (2.29)  −0.042 (−0.21) 
Service 0.393 (1.18)  −0.0587 (−0.41)  0.130 (0.61)  0.240 (1.56) 
Others 0.417 (1.38)  −0.0451 (−0.34)  0.133 (0.70)  0.231 (1.63) 
Constant2.676*** (22.20)−3.437*** (−4.60)2.999*** (18.96)0.0288 (0.53)−2.738*** (−7.35)0.208** (2.80)0.0225 (0.38)−2.330*** (−5.36)0.0550 (0.67)0.00331 (0.06)−1.497*** (−3.78)0.143* (2.40)
var(Organisation)1.662** (3.28)0.576 (−1.74)1.735** (2.59)0.281*** (−6.66)0.0356** (−3.09)0.295*** (−4.32)0.396*** (−5.80)0.288*** (−4.70)0.494*** (−3.58)0.234*** (−7.24)0.0423** (−3.26)0.120*** (−4.72)
var(Residual)2.421*** (16.49)1.990*** (7.80)2.582*** (10.84)0.733*** (−5.63)0.612*** (−5.40)0.807* (−2.41)0.607*** (−9.08)0.578*** (−5.97)0.592*** (−5.93)0.774*** (−4.69)0.687*** (−4.20)0.760** (−3.11)
log-likelihood−1693−705−751−1110−468−507−1060−493−478−1121−487−478
Chi-Square.211.367..244.092..70.726..86.867.
p-Value.0.000..0.000..0.000..0.000.
Number of organisations158110110157108108157108108157108108
Observations849365365814355355814355355814355355
Variance component of random intercept40%22%40%27%6%26%3933%45%23%6%13%

Finally, column (4) of Table 5 shows the results for the alternative operationalisation of the dependent variable, including also the contacting of political elite. What is most striking is that the coefficient for the regularity of contact between an organisation and the local elite does not become much larger; obviously, the number of political contacts of the organisations is simply not strongly correlated with the number of political contacts of their members. As discussed at the end of the section on the econometric model, this is the only indication we have of our expected causality: since the correlation between the contact score of an organisation and that of individuals is not very pronounced, it is hard to argue that such organisational characteristics are simply an aggregation of the characteristics of the members. Consequently, it is at least not wholly implausible to argue that the type of organisation shapes certain patterns of individual behaviour.

In Table 6, we list the results of variants of the models in columns (2) and (4) of Table 5, using the different factors extracted from the principal component analysis as dependent variables. The model listed in column (2) of Table 6 is thus equivalent to the one in column (2) of Table 5.22 Additionally, we compute two “null-models”, which are intercept-only regressions that include the random intercepts. One of these is calculated for all observations possible and the other for the same observations contained in the complete regression model. Columns (5), (8) and (11) show the same regression models using the first, second and third component of the factor analytic solution listed in Tables 1 and 2 as dependent variables.

What is interesting if one examines the significance and strength of the various coefficients, is that the amount of contact an organisation has with the political elite and the category of “representing” organisations are relevant predictors only of “expressive” individual political participation. For the other two types of participation, the coefficients of these two variables are much smaller (or even negative) and in the case of “anonymous” participation not even significant. Noteworthy is also that the perceived influence of an organisation is negatively (although not significantly) correlated with “anonymous” political participation. Not very surprisingly, for “expressive” political participation the results are very similar to those for our composite index. Furthermore, comparing the coefficients also shows that being well educated is negatively and significantly correlated with expressive political behaviour, while it is positively and significantly correlated with contacting the political elite and “anonymous” participation. This could be an indication that better educated people choose forms of political participation that are more demanding intellectually because they either require more sophisticated argumentation (such as in the case of direct meetings with politicians) or a higher ability to process information (boycotts). In addition, compared to the other types of political participation, expressive behaviour seems to be more dependent on the measured characteristics of an organisation, while in the case of contact with the elite or anonymous participation, individual characteristics turn out to be more relevant. This result nicely corresponds to those of Verhulst and Walgrave (2009), who showed that being a member of a co-staging organisation fosters participation in public protests. Similarly, Stolle et al. (2005) found that boycotters, who are engaged in a very individualised form of protest, are exclusively members of “distant checkbook-organisations” without much direct contact.

A further insight can be gained by looking at the variance components of the hierarchical regression models. Since the organisational level is introduced as a random intercept parameter, it is possible to compute the contribution this random intercept makes to the overall variance of the dependent variable. These variance components are listed in the last row of Table 6. From this, it is quite obvious that, when only taking into account the “null-models”, the organisational level captures a rather large part of the variance of the dependent variable, this ranging from 20 to 40 per cent. As expected, the share of the variance explained by the random intercept at the organisational level is much smaller in the cases of “contact with the elite” and “anonymous” political participation. Indeed, organisations seem to play the most important role with respect to “expressive” political participation. In addition, while the inclusion of the organisational characteristics as explanatory variables captures most of the variance of the random intercept for “contacting the political elite” and for “anonymous” participation, this is less so for “expressive” participation. Although “representing” organisations are herein identified as the type of associations with more expressively active members, and even though the number of political contacts seems to have an impact on this type of political engagement, there must be additional effects at the organisational level that have not been identified through the CID-survey.

Conclusion

In this paper, we have uncovered a number of contextual factors that seem to have an influence on political participation, although this influence appears to differ with respect to the actual type of political participation that takes place. Based on a factor analysis of 19 different participative activities, we have identified four such types: contacting the elite, expressive participation, anonymous participation and fundraising activities. While we have found a rather strong dependence of expressive forms of political participation on organisational features such as an associations’ contact with the local political elite, less expressive forms of participation seem to be more dependent on individual traits such as political socialisation and interest. According to our results, expressive forms of political participation, such as demonstrations are pursued by less skilled, less interested and less educated people, who are mobilised on the basis of their membership in a specific organisation. In those cases, associations clearly support and facilitate the political participation of citizens.

These findings are important because they suggest that it is not purely membership in an association that makes a difference with respect to political participation. Indeed, knowing the type and specific characteristics of an organisation helps us to understand the differences in the intensity of specific forms of political participation, and also reveals that members of different types of organisations tend to choose different channels for participating. However, while our results are encouraging, much still remains in the dark.

The data we have used in this analysis is unique in that it is based on a combined survey that includes answers from individual “activists” as well as independently collected answers from the organisations’ representatives. This special feature of the data has helped us to find one, albeit weak, indication for our assumption about the direction of causality. With respect to the typology of associations, our operationalisations have remained somewhat ad-hoc. However, this is a general problem with all research concerning the impact of organisational traits on social capital formation, and without further theoretical development and specifically designed measurement instruments, empirical support for the assumption that organisations matter will remain weak. Because the data we have employed in this study has allowed us to measure the goals as well as the activities of the organisations independently, and because we have found very similar results for both types of operationalisations, we have much confidence in the results that both the embedment and the type of organisation matter.

Footnotes

  • 1

    Apart from people mentioned in the text, we would also like to express our gratitude to Jurga Bucaite, Tom van der Meer, Sonja Zmerli and two anonymous reviewers for comments and suggestions.

  • 2

    Ayala (2000) and Adman (2008) demonstrate that an explicit differentiation should be made between the skills obtained at the workplace and those obtained through membership in a voluntary organisation. However, one could also argue that certain professions, e.g. administrative work, nevertheless serve to enhance abilities like letter writing, which can be of use in associations engaged in certain forms of political activity e.g. Amnesty International.

  • 3

    One has to keep in mind that many of the effects described so far could also be attributed to self-selection processes (Sønderskov, 2011), while self-selection and socialisation could even reinforce each other (Hooghe 2003).

  • 4

    Overly dense linkages between associations and the local administration can also become dangerous or counter-effective however (Lelieveldt and Caiani, 2007).

  • 5

    The importance of context variables, such as the openness of the political system and the general public value orientation of local government managers, is also strongly emphasised by qualitative studies (e.g. Lowndes et al., 2006). Others confirm that the perception of an organisation as “political” or “pivotal” animates individual members to engage more in politics (Baumgartner and Walker, 1988; Lelieveldt et al., 2009).

  • 6

    Erickson and Nosanchuk (1990) found that in apolitical groups, it were only political discussions with fellow members that might inspire political participation. Leighley (1996) even concludes that in apolitical groups, political participation can only be increased non-intentionally, while intentional mobilisation only works in political groups.

  • 7

    Differentiating between “bridging” and “bonding” organisations solely based on the goals of the organisation is problematic, since these concepts crucially also depend on the heterogeneity of membership and on the linkages within the local context (Freitag et al., 2009; Schulz and Baumgartner, forthcoming).

  • 8

    In Switzerland, three surveys were conducted in 1999 as part of the comparative European research project “Citizen Involvement and Democracy” (CID, http://www.mzes.uni-mannheim.de/projekte/cid/): these included a large “population survey” that is representative of the Swiss population, and two surveys of organisations and their “activists” for which a special sampling strategy was chosen: first, eight municipalities (from the French and the German speaking parts of Switzerland and from different classes of municipality size) were chosen. For each of these municipalities, it was the goal to identify all or at least as many voluntary organisations as possible, although it was much more difficult to identify informal groups (Kriesi and Baglioni, 2003). Subsequently, an “organisational survey” was conducted of these organisations which is described in Kriesi and Baglioni (2003). Further, in each municipality, a sample of about 10% of the organisations that had been identified was selected to conduct a survey among their activists. Hence, for the town of Thun, for example, approximately 230 organisations were identified, and the respective data contains valid answers from 15 organisations and 93 of their “activists” (Maloney and Van Deth, 2010). Since it was a great challenge to maximise the response rate given this two-tiered sampling strategy, the data is probably rather far from being representative on the basis of the selected cases. However, the major advantages of the data are that the samples are linked, i.e. it is known to which organisations the activists belong, and that the characteristics of the organisations have been collected independently of the characteristics of their activists. We are indebted to Hanspeter Kriesi and Jan van Deth as well as an anonymous reviewer, who provided us with the data and the relevant background information.

  • 9

    Although the size of the organisations (in terms of members) that returned “activists”-questionnaires varies between 2 and 12000, never more than 15 such questionnaires were returned and hence, these samples are very unlikely to be highly representative of the organisations’ membership. For most organisations (77%), the number of observations lies above 5.

  • 10

    The respondents were asked about their degree of involvement, i.e. whether they are simply “friends” of people who are more actively involved in the organisation, “involved” (without being members), formal “members”, “participating” actively in some form, “donating” money or contribute voluntary “work”. If being a “friend” or just “involved” does not constitute an “activist”, 175 of the 968 subjects in the “activist sample” cannot be viewed as “activists” in the more narrow sense.

  • 11

    The question was: “During the last 12 months, have you attempted to bring about improvements, or to counteract deterioration in society by doing any of the following?”. We have listed the mean (the share of positive answers) of these dichotomous variables, along with information about the sample size, in Table 10 in the appendix.

  • 1In this row, for each category and for the total, the number of respondents with a non-missing value are given for these questions for which the least/the most such responses were counted.

  • 12

    In addition, the survey also contained a question about “illegal activities”. We drop this variable since we are only interested in legal forms of political participation in this paper.

  • 13

    By applying factor analysis to a set of dichotomous variables we follow Kim et al., (1977), who explore the applicability of such a dimensional analysis on a set of very similar dichotomous variables. They conclude that classical factor analysis (and the assumption of underlying continuous variables) is feasible if the dichotomous variables are not highly correlated. In our sample, the correlation between the dichotomous variables is always below 0.5. If the underlying factors are assumed not to be normally distributed, classical factor analysis might even be preferred to more adapted methodologies, such as the normal ogive model (Kim et al., 1977).

  • 14

    We also allow for correlated factors (opaque rotation), however, the results did not change much and in all instances fitted the data less well than with the standard orthogonal rotation. Engaging in the theoretical discussion on whether or not the different dimensions of political participation can be seen as independent from each other, is beyond the scope of this article.

  • 15

    While each subject was asked about different degrees of involvement (active member, volunteering, donating money etc.) in other organisations, the answers could only be given for types of organisations (sport, church, environment and more than 20 others). Hence, for each subject and type, the degree of involvement indicated was recoded to a binary variable (credits for the construction of these variables go to Christian Schnaud and Jan van Deth), and then the number of classes for which active involvement had been indicated was summed up.

  • 16

    We also employ household income but since this does not contribute substantially to the explanation we opt to drop this. With respect to children, another possible proxy for the opportunity costs of time, the CID-survey allows only an indicator for whether or not there are children living with the respondent. However, since the impact of children depends on their age, we dismiss this variable.

  • 17

    The questionnaire included one battery of questions about the frequency of skills-acts in the primary and a second battery of questions about skills-acts accomplished in all organisations the subject is a member of. We opt for the second battery, measuring acquired skills more generally, although the results are equivalent. The battery includes questions about the frequency of participation in decisions/meetings, planning or chairing a meeting, preparing or giving a speech and writing an official letter. This frequency is measured ordinally (from 1 = “a few times a week” to 4 = “never or almost never”) and, having re-poled the answer categories, we choose to simply add the values of the different items, which results in an index with a possible value range from 4 (no skills-acts in all four skills-acts categories) to 16 (frequent skills-acts in all four categories).

  • 18

    We test a number of additional control variables that have been used by various other studies. We opt to exclude those that have p-values far from any conventional level of statistical significance in all the models presented in this paper.

  • 19

    We do not use the budget since this is highly correlated with the number of members of an organisation.

  • 20

    There is not much agreement on the effectiveness of hierarchical (multi-level) linear regression techniques applied to data with a relatively small number of observations at the first (individual) level, as well as a relatively large number of groups at the second level, which implies a small average group size. However, we follow Gelman and Hill (2007: 248f and 274f), who argued that the multi-level regression modeling approach is always preferable, particularly if there is large between group (second level) variation (which is actually the case in our data). Hence, even if there is a considerable amount of groups with only two or even just one observation, the multi-level approach is still preferable, although the estimates of the small groups’ intercepts are likely not to be precise. We apply STATA’s xtmixed-packages to estimate the hierarchical models reported below.

  • 21

    The score measuring the contact between an association and other associations in the municipality is not significant and we thus refrain from employing it here, although it was successfully employed by Freitag et al. (2009) in their study on generalised trust. This could indicate a problem of multicollinearity, however, the two scores are not strongly correlated (r = 0.31) and variance inflation factors from an OLS-regression were also not suspicious.

  • 22

    The only difference being that the maximum likelihood estimator is used for the former and the restricted maximum likelihood estimator for the latter, which explains why the figures differ marginally.

  • 23

    Of course, a correlated effect is also contained in any unobserved heterogeneity related to the group level as it might be absorbed by the error term or by a random intercept term.

  • 24

    Including the group-mean of a particular variable (without centring the variable around the group-means) accounts for relaxing the assumption that the between and within group effects are the same (Rabe-Hesket and Skrondal, 2008, 114ff). We thus have nevertheless tested a model that includes the group means of the individual level variables and the substantial results, particularly with respect to important characteristics of the organisations, such as the embedment of the organisation in the local context, do not change. However, for reasons of space, we do not list these results here.

Appendices

Appendix

In evaluation studies of schools and teachers, it is common to apply “grand-mean centring”. Seltzer (2004), for example, proposed the following basic grand-mean centred model:

image

with

image

In this model, Xi denotes variables measured at the individual level (level 1), where we have a unique observation for each subject i; Zj are variables at the organisational level (level 2). Paccagnella (2006) differentiates between “contextual” effects, i.e. when individual behaviour depends on the distribution of the independent level 1 variables within the group, and “correlated” effects, i.e. when individual behaviour depends on the “institutional environment” provided, for example, by the group infrastructure.23 To account for “contextual” effects, Seltzer (2004) also incorporates the (grand-mean centred) group mean (the mean of the individually measured characteristics of those subjects belonging to the same organisation) into the model. In addition to the centering of these variables around their means, a random intercept for the organisation level is introduced into the model to account for unobserved contextual effects. While Seltzer (2004: 271) argues that centering around the grand-mean allows one to account for “differences among classes in their student intake characteristics”, for Paccagnella (2006, 71) grand-mean centering simply changes the interpretation of the intercept and the slope parameters, but does not help to differentiate contextual from correlated effects. He thus proposed the following model (Paccagnella, 2006, 75):

image

Here, contextual effects are captured by the group mean inline image, as well as by the directly measured characteristics of the organisations Zj and the level 2 random intercept μ0j. The problem with this model is however that there should be sufficient level 1 observations for computing group means. Since there are only about 5 observations per organisation in our data, we refrain from incorporating the group mean24, as it would not make much sense to estimate the group mean of an organisation that has several hundred members on the basis of only 5 observations. Thus, in Table 9 we only list the results for the first alternative specification, grand-mean centering, along with two more classical OLS-models, the first simply accounting for corrected standard errors according to a clustering of observations in organisations and the second being a “fixed-effects” model in the sense that the group-means have been subtracted from all variables (including the response and the error term, which is equivalent to a model including dummy variables to identify the organisations; compare Rabe-Hesket and Skrondal, 2008, 111ff). As the coefficients show, the results are not very sensitive to these re-specifications.

Tobias Schulz is a research fellow at the Swiss Federal Research Institute for Forest, Snow and Landscape (WSL). In his research, he is most interested in institutions and political behaviour, mainly concerning civil society and environmental policy at the local, national and international level. Address for correspondence: Swiss Federal Institute for Forest, Snow and Landscape Research, Zürcherstrasse 111, 8903 Birmensdorf. Tel.: +41 (0)44 739 24 77; Email: tobias.schulz@wsl.ch

Stefanie Bailer is assistant professor for global governance at the ETH Zurich. Her research interests encompass decision making at the domestic and international level, and civil society and its development. Her articles have appeared in International Political Science Review, European Union Politics and Political Studies. Address for correspondence: Center for Comparative and International Studies, ETH Zürich, Haldeneggsteig 4, 8092 Zürich. Tel.: + 41 (0)44632 75 95; Email: bailers@ethz.ch

Ancillary