The two interpretations of local political support derive from theoretical assumptions on the constitution of political support. In the first line of reasoning, local political support among citizens derives from local democracy's own merits. Here, local political support is understood to be relational and evaluative. A central paradigm reads that “A trust B do to X” (Hardin, 1999: 26). Citizens (A) evaluate the regime and its institutions (B) on a specific matter (X). Building on this paradigm, the institutional approach to political support focuses its attention to the performances and procedures of the object of support (cf. Martini and Quaranta, 2019).

We can further distinguish between three object-specific explanations of political support: input, throughput, and output, also referred to as government respectively of the people, with the people, and for the people (Schmidt, 2013). Input performance refers to the representative function of political institutions, namely to represent the views of all citizens. The relevance of input traits to political support is found at the macro level of proportional electoral rules (e.g., Aarts and Thomassen, 2008; Marien, 2011) and at the meso-level of citizens' assemblies (Werner, 2020). Output performance relates to (the effectiveness of) policy outcomes, often understood as economic performance (e.g., van Erkel and van der Meer, 2016). Finally, throughput points to “the practices that go on in the “black box” of governance”(Schmidt, 2013: 5). This is commonly summed up as the procedural quality of government, and includes accountability, transparency, and inclusiveness, or – inversely – corruption (Grimmelikhuijsen, 2012; Grimes, 2017; van der Meer, 2017). Input, output, and throughput may be studied via objective, exogenous indicators or via citizens' subjective perceptions and evaluations thereof (Mishler and Rose, 2001; Aarts and Thomassen, 2008). For the purpose of this study, we rely on the subjective perceptions as the minimal requirement for object-specific evaluations.

When we transpose this argument from the national level to local political support, the overarching hypothesis reads that political support at the local level is embedded in evaluations of local input, output, and throughput performance. In other words, citizens are expected to judge local democracy on its own merits. That would be visible in empirical research when local political support is related to local political evaluations.

The literature on local democracies offers some evidence on this relationship, albeit only for output evaluations. DeHoog et al. (1990) show that the perceived quality of city provisions stimulates satisfaction with local government. This is echoed in works by Fitzgerald and Wolak (2016) and Denters (2014). By contrast, input and throughput legitimacy are less well-studied. Nevertheless, a longstanding argument favoring local democracy states that municipalities offer better opportunities for the voice of citizens (Dahl, 1967). Indeed, research shows a negative effect of population size of cities on participation (Oliver, 2000) and trust in local government (Rahn and Rudolph, 2005; Montalvo, 2010; Hansen, 2015). Similarly, citizens mention the linkage function (encompassing representatives, accountability, responsiveness, and transparency) almost twice as often as a reason to trust local government compared to the state and federal government in the United States (Jennings, 1998). In other words, input and throughput legitimacy are likely to matter, particularly at the local level.

All in all, our first set of hypotheses reads:

The rivaling perspective on local political support poses that it is not an object-specific evaluation, but rather derivative of a more diffuse attitude. Although local governments are embedded in a structure of multilevel government, this multilayered component tends to be overlooked when local political support is studied in a vacuum. Only few studies examined the influence of national politics on local political support (Fitzgerald and Wolak, 2016; Muñoz, 2017). These studies, next to a substantively more comprehensive literature on supranational support, propose two mechanisms in which local political support is primarily derived from more general, nationally-oriented political support.

The trust-as-syndrome mechanism (e.g., Harteveld et al., 2013) – also known as the “congruence” (Muñoz et al., 2011) or “equal assessment” model (Kritzinger, 2003) – states that people have a general attitude toward the political system that informs their attitude toward different polity levels (Anderson, 1998). This general attitude may itself be rooted in social (interpersonal) trust (Zmerli and Newton, 2017) or psychological predispositions (Harteveld et al., 2013; Ares et al., 2017).

The cue-taking mechanism suggests that people use their perceptions of national politics as a heuristic (cue, proxy) when they are asked about less salient levels of government. Anderson (1998: 576), for instance, argues about support for the European Union: “Citizens compensate for a gap in knowledge about the EU by construing a reality about it that fits their understanding of the political world. For most people, this means that they rely on what they know and think about domestic politics.” Many studies on the European Union have argued that citizens employ the national level as such a heuristic (Anderson, 1998; Rohrschneider, 2002; Kritzinger, 2003; Hobolt, 2012; Harteveld et al., 2013; Armingeon and Ceka, 2014; Torcal and Christmann, 2018).

The crucial – but in this paper untestable – difference between trust-as-syndrome and cue-taking, is that the latter assumes a hierarchical ordering of levels of government in public opinion. The national political level is the first-order polity in terms of the attention citizens pay to the political dynamics, and the knowledge they have about it. Other polity levels, such as the supranational and the municipal are secondary (De Blok et al., 2020), as has been long recognized in electoral research on local and regional elections (Heath et al., 1999; Rallings and Thrasher, 2007; Marien et al., 2015). Yet, both mechanisms suggest a strong relationship between national and local political support. Ultimately, both mechanisms lead to the same hypothesis:

Building on the literature on support for the European Union (e.g., Harteveld et al., 2013; for an overview see Muñoz, 2017), we expect that local political support is related to national political support more strongly than in object-specific evaluations of local democracy's input, throughput, and output performance.

Up to this point, we have introduced two rivaling interpretations of local political support, one arguing that it is object-specific and evaluative and the other that it is not. In a multilayered government structure, we should consider the conditions under which these interpretations are more or less likely to be valid.

First, in a context of multilevel government we may expect that the validity of the two interpretations depends on the extent to which people are embedded in their local democracy. Feeling part of the political community is the most fundamental, diffuse mode of political support (Easton, 1975; Norris, 2011). We may observe this local embedding in citizens' length of residence, their feelings of attachment to the municipality, their offline or online involvement in local political activities, and their support for local parties (rather than local chapters of national parties). Local embedding is likely to stimulate local political support (Denters, 2014; Fitzgerald and Wolak, 2016). More importantly, we expect that the locus of local political support is more likely to be local among citizens that are themselves more strongly embedded in their local political community: ceteris paribus, citizens are more likely to base their local political support on local, object-specific evaluations, and need to rely less on other cues or heuristics, when they feel part of the political community.

Second, political sophistication - a mixture of political interest, attentiveness and knowledge - plays a consistent role in public opinion research (Zaller, 1992). People are more likely to use heuristics when they have less knowledge of or interest in that object (Lupia, 1994; Kahneman, 2011). There is ample evidence for this conditional relationship regarding political support in multilevel settings. Within the European Union, higher levels of political knowledge, attention, and education decrease the degree to which national political considerations are used as a proxy for EU attitudes (Desmet et al., 2012; Hobolt, 2012; Armingeon and Ceka, 2014; Muñoz, 2017).

Education is an important discriminator between citizens with low vs. high levels of political sophistication. Most notably, the high educated are not only more likely more likely to have a better understanding of and interest in politics (Bovens and Wille, 2017); they are also more likely to respond to their governments' actual performance (Hakhverdian and Mayne, 2012; Mayne and Hakhverdian, 2017; van der Meer and Hakhverdian, 2017). This suggests that the association between local political support and local political performance is likely to be stronger among the higher than among the lower educated.

Although the theoretical approaches we build on differ in terms of the theorized causality, we restrict our theoretical assumptions and analyses to associational only. While the institutional approach theorizes input, output and throughput evaluations to be causes of political support, this applies primarily to objective indicators. Subjective indicators, that is, evaluations of performance, are potentially more endogenous to political support (e.g., De Blok et al., 2020). Furthermore, the multilayered approach does not focus on causality but the relationships between trust in different objects. Hence, this multilayered approach sees national and local support either as equals in one syndrome of trust, or national support as dominant and spilling over in local support (cue-taking). In this paper, we do not make any claims on causality, and instead investigate the associations between evaluations of local political performance, national political support, and local political support. In this way, we aim to assess the locus of local political support, namely the extent to which it is rooted in attitudes on local performance or national support, without any claim as to whether those roots are causes or associations.

To test these hypotheses, we need (1) data on political support (democratic satisfaction and/or political trust) at the local and national levels, (2) detailed evaluations of input, throughput, and output performance by local governments, and (3) information about the extent to which citizens are embedded in their municipality.

These demands are met by the first Dutch Local Election Study (DLES). The first round of the DLES was collected in the Spring of 2016, shortly after a decentralization process was implemented in the Netherlands and 2 years before nation-wide local elections would take place. DLES 2016 would serve as the baseline measurement for future editions. DLES 2016 is a national sample that uniquely contains a wide range of measures (on citizens' evaluations of their municipality's performance on input, throughput, and output, as well these citizens' political embedding in their municipality) that were collected for the purposes of this study.

The level of decentralization in the Netherlands is about the average level EU wide, making it well-suited for an inquiry into the locus of local political support. The Netherlands is commonly classified as a decentralized unitary state: Municipalities have specifically delineated tasks, but are limited in taxation. This is reflected in OECD statistics on local government (which in these statistics also encompasses the provinces and the water authorities). The share of government expenditure that is local is above average (31% of total government expenditure), though considerably lower than in Nordic countries such as Sweden and Denmark. The decentralization process of the mid-2010s raised local expenditures by a few percentage points. By contrast, the share of local tax revenues is slightly below average (9% of total government tax revenues). This discrepancy is primarily because the execution of several costly government programmes – most notably education and welfare – is decentralized to the local level.

Our Dutch case faces two limitations with regard to generalizability, upon which we also reflect in the conclusion. First, in unitary countries with a proportional electoral systems such as the Netherlands, trust and satisfaction in local authorities is typically lower than in federal countries and/or countries with majoritarian systems (Fitzgerald and Wolak, 2016). Therefore, our findings are first and foremost generalizable to unitary countries with proportional electoral systems, and to a lesser extent to either federal countries or majoritarian systems, which have institutionally a stronger local dynamic. Second, although the dominant pattern of trust in local politics exceeding trust in national politics is found in the Netherlands, the average political trust levels are above European average (Muñoz, 2017). Because there are some indications that the gap in support between local and national is larger in countries with the very lowest national support (Muñoz, 2017: 73), the locus of local political support might be less affected by national dynamic in those countries.

The level of detail on these variables came at a cost in the shape of missing values, both on the dependent and independent variables. Respondents apparently found it easier to provide a substantive answer on the items that measure political trust (only 2% missing values on the national and local objects of political trust) than on the SDW items (6% missing values on national SWD and 10% missing values on local SWD). On some of the independent variables, the share of respondents with missing values surges to 10–20 percent of the sample. This is particularly prevalent among variables that seem to require a close, detailed understanding of local politics (see Appendix Table A for the number of cases of each variable). High levels of missing values are potentially problematic, particularly because the missingness does not seem to be at random.

Different strategies exist to deal with missing values. Listwise deletion of all respondents with at least one missing value would lead to a net loss of almost 60% of our sample in the full multivariate models. Missing value imputation has been presented as a superior and more efficient option (King et al., 2001; Lall, 2016). Yet, that only holds under the assumption that data are missing at random (MAR). If they are not missing at random (MNAR), multiple imputation might instead induce further bias whereas listwise deletion does not (Allison, 2014; Pepinsky, 2018). The suggestion that data on the DLES 2016 measures are missing at random is quite unlikely.

We dealt with the problem of missing data by testing the robustness of our models to different strategies of dealing with missing data. On the one extreme, we performed listwise deletion, eliminating all respondents with at least one missing value on at least one of the variables identified above. Full listwise deletion led to a net sample size of 1,100 respondents. On the other extreme, we performed missing value imputation on all missing data in our model. Full imputation led to a net sample size of 2,356 respondents. Additionally, we pursued two moderate strategies that would do more justice to our data and model. Partial listwise deletion entails the deletion of respondents except for those who only have missing values on non-significant components of the sheaf coefficient (see below). This led to a net sample of 1,400 respondents. The final moderate strategy is partial imputation. We first deleted respondents with missing data on more than a quarter of the variables in our model (assuming that the non-random element of missing data would be strongest there), before engaging in missing value imputation on the remainder of the respondents. Partial imputation led to a net sample of 1,995 respondents.

Given the number of missing values, and the relevance of the selection effect to the core hypotheses of this article, we argue that partial imputation is the best strategy and is used in all models presented in Tables 1–7. As a robustness check, we report on the findings of the various strategies at the end of the Result section. Although our findings on political trust do vary in aspect depending on the method chosen to deal with missing cases, these variations do not impact our substantive conclusions. We discuss this in the Result section.

Table 1 offers a test of the first set of hypotheses. It displays the effects of input, throughput, and output evaluations that are summarized in sheaf coefficients. The full model, without the sheaf coefficients that combine the separate parameters, is reproduced in Appendix Table B. Together, the determinants explain ~31% of the individual variance in satisfaction with democracy and ~23% of the variance across municipalities. For political trust, these figures are respectively 27 and 40%.

We find that both local democratic satisfaction and local political trust are significantly related to local evaluations of political performance. This supports H1. In line with hypotheses H1a–H1c, input, throughput, and output evaluations all significantly relate to political support. Of these three types of evaluations, output evaluations have the largest standardized effect on democratic satisfaction (b = 0.15) and political trust (b = 0.36). Interestingly, input evaluations are least important to democratic satisfaction, while throughput evaluations are least important to political trust. This underlines an intuitive difference between the two middle-range indicators of political support. Democratic satisfaction is directed somewhat more toward the process of decision-making. By contrast, evaluations of responsiveness seem to have a relatively stronger effect on trust in political institutions.

Table 1 also tests the extent to which local political support is embedded in its national equivalent. We find strong coefficients both for satisfaction with democracy (b = 0.30) and for political trust (b = 0.55). This supports H2.

Table 2 combines all three types of local political evaluations (input, throughput, and output) in standardized sheaf coefficients. This enables us to determine whether local political support is embedded more strongly in national political support than in local political evaluations. The results are mixed. In line with H3, we find that local democratic satisfaction is more strongly related to national democratic satisfaction (b = 0.31) than to local evaluations (b = 0.22). This would suggest that the locus of local democratic satisfaction is primarily an aspect or derivation of a more general (national) attitude toward democracy. However, we do not find support for H3 on local political trust, where the standardized effect of national political trust (b = 0.55) and local evaluations (b = 0.56) are statistically indistinguishable.

Tables 3–5 report the models that test whether local embeddedness moderates the effects of local evaluations and national political support, namely with local attachment, local residence, and local political activism.

Hypothesis 4a posited that local embeddedness enhances the effect of local political evaluations. This hypothesis finds little support. All conditional effects are non-significant, except for one: Table 5 shows that local evaluations have a significantly stronger effect on democratic satisfaction among citizens who are politically active in their municipality (b = 0.27) than among those who are not (b = 0.20). While this finding is in line with hypothesis 4a, this supporting evidence is found in only one single model.

Hypothesis 4b posited that local embeddedness mitigates the effect of national political support. We find different results for the two modes of political support. The relationship between national and local democratic satisfaction is slightly but significantly mitigated by length of residence, but not by other measures of local embeddedness. The longer the time span that people lived in their municipality, the weaker the correlation between national and local political support. Yet, the relationship is substantially so small, that even among the elderly the marginal statistical effect of national political support remains stronger than that of local political evaluations.

The relationship between trust in national and trust in local political institutions is significantly mitigated by subjective local attachment and by local political activism. These effects, too, are in line with the hypothesis. However, conditional effects are not very common.

Overall, our findings provide some weak indication that that local embeddedness can indeed enhance the sensitivity for political performance at the local level, and concurrently mitigate their reliance on their national political support. Yet, these effects are scarce and modest.

Finally, Tables 6, 7 show the moderation effects of political sophistication. The tables provide unequivocal support for H5a. Among those with higher levels of political interest and/or a higher level of education, local democratic satisfaction, and local political trust are more strongly embedded in local political evaluations. This suggests that political sophistication makes people's local political support more sensitive to the performance of local democracy.

Hypothesis H5b finds less consistent support. Among residents with high levels of interest in local politics, levels of trust in local political institutions tends to be significantly less derivative of their trust in national politics. Yet, other moderating effects of political sophistication on the relationship between national and local political support are not significant. All in all, evidence that the locus of local political support depends on local embeddedness and political sophistication remains tentative.

We performed two sets of robustness checks. First, we analyzed whether our findings hold when we model the significant interaction effects from the previous tables simultaneously. The conditional effects on SWD are highly robust: All significant effects in Tables 3–7 remain significant when estimated in a single model.

Second, we checked the robustness of our main findings under different strategies of dealing with missing data. Table C of the Appendix reflects Table 2 in the main text. We find that the models on SWD are highly robust, regardless whether we employ full listwise deletion, full imputation, or a partial strategy. However, the models on political trust are less robust. Under a strategy of full listwise deletion political trust is more strongly tied to local performance evaluation than to national political support, whereas the two were statistically tied under the strategy of partial imputation we employed in our main analyses. Yet, our substantive conclusion on hypothesis 3 remains the same (supported for SWD, rejected for political trust)⁶. The interactive models are also highly robust for SWD, but less robust for political trust.