The public’s preferred level of involvement in local policy-making

The three aforementioned research objectives were tested across five empirical studies. Table 1 summarizes the main characteristic of our five studies. As shown in this table, we started our research endeavour with an explorative study (Study 1), which aimed to obtain a general estimate of the desired relative weight of citizen and government involvement in policy decisions (Objective 1). This explorative study was supplemented with four additional studies (Studies 2–5), in which we used a variety of different research methods and designs (see Table 1 for an overview) to map what particular combination of citizen and government involvement at the local level people consider to be optimal (Objective 1). Studies 3, 4, and 5 additionally also examined the existence of possible clusters within the citizenry based on their preferences (Objective 2), with Studies 4 and 5 also exploring the defining personality characteristics of these clusters (Objective 3). Sensitivity power analyses using the WebPower package³⁴ and the simr package³⁵ in R showed that our studies were all sufficiently powered to detect the reported effects (see Appendix A for more details). Unless mentioned otherwise, the data of our studies were analyzed using SPSS (version 27). The datasets and data analysis scripts are publicly available at: https://osf.io/zs9cj/.

Participants in our studies were all recruited via Prolific (www.prolific.co), an online research platform that provides a detailed description of the demographics of their participant pool, and which can be used by researchers to a priori select specific groups of participants (for more detailed information on this platform, see^36,37). In all five studies, we recruited a gender-balanced sample of adult participants living in the United Kingdom. No further selection criteria were used. The five samples were independent and did not overlap. The UK is a highly appropriate context for the purpose of our research, because both representative and more direct forms of democracy are already used at different levels of government³⁸, including the local level. The Edinburgh Road Tolls Referendum is an example of a local referendum, held in February 2005 by the City of Edinburgh Council, about whether voters supported the Council’s proposed transport strategy³⁹. An example of a local citizens’ jury is the Leeds Climate Change Citizens’ Jury, which was commissioned in 2019 by the Leeds Climate Commission to ensure citizens’ voices were heard in Leeds’ vision for achieving carbon zero emissions⁴⁰. Finally, an example of a local participatory budgeting project is ‘You Decide!’. This project was carried out in 2009–2010 in Tower Hamlets, a dense urban Borough in the East End of London⁴¹. These examples illustrate that, in the last decade, various forms of direct citizen participation have been used by UK local governments.

Study 1

Our first study (N = 200) was an explorative study in which participants were asked to indicate how decisional power in local policy-making should be balanced between citizens and the government. To this ends, we employed a constant-sum approach⁴², which allowed us to directly compare the relative importance that people ascribe to the direct involvement and decisional power of citizens and the local government (see “Methods” section for more details).

Results of study 1

Figure 1 shows the distribution of participant’s preferred decisional weight. Our analyses revealed that, on average, participants preferred citizens to have a decisional weight of 49.6% and the local government to have a decisional weight of 50.4% (SD = 19.79). A closer inspection of participants’ preferences (see Fig. 1), revealed that 37.0% of the participants (74 out of 200) preferred the government to outweigh citizens. Of the remaining participants, 28.5% (57 out of 200) preferred citizens and the government to both have a weight of exactly 50%, and 34.5% (69 out of 200) preferred citizens to outweigh the government. What is particularly interesting, however, is that only a very small percentage of participants (less than 5%) preferred either the government or citizens to have full decisional control (i.e., a decisional weight of 100% for one of the two actors)—see Fig. 1. These findings hence show that most people indeed seem to prefer a model in which citizens and the government both have a considerable weight, but also that the preferred amount of citizen (vs. government) involvement varies strongly across individuals (with some preferring small and others preferring large levels of citizen involvement).

Study 2

Our second study (N = 270) was an experimental study in which participants were randomly assigned to one of five conditions which varied the decisional weight that citizens have—relative to the government—in a local policy decision-making scenario (citizen vs. government weight: 0% vs. 100%, 25% vs. 75%, 50% vs. 50%, 75% vs. 25%, 100% vs. 0%). Participants were asked to indicate to what extent they found the presented decision-making model acceptable, legitimate, fair, democratic, effective, efficient, appropriate, and justified (1 = not at all, 10 = very much so).

Results of study 2

An analysis of variance (ANOVA) revealed that participants’ overall evaluation of the decision-making models (i.e., the mean of acceptable, legitimate, fair, democratic, effective, efficient, appropriate, and justified) differed significantly across the five experimental conditions, F(4, 265) = 43.67, p < 0.001, partial η² = 0.397. Figure 2 displays participants’ overall evaluation of the different decision-making models. As shown in this figure, participants’ overall evaluation was highest in the condition in which citizens and the government both have a decisional weight of 50%. Post-hoc comparisons with Tukey’s honestly significant difference (HSD) correction showed that the overall evaluation in this condition differed significantly from all the other conditions (all ps < 0.001), with exception of the condition in which citizens have 75% and the government has 25% weight (p = 0.129). These findings thus indicate that there is, on average, a clear preference for models in which citizens have at least 50% weight in the decision, and although in case of unequal weight people seem to prefer citizens rather than the government to have a greater say, support decreases when citizens envision having absolute control in local policy-making processes.

Study 3

This third study (N = 294) aimed to replicate and extend our prior findings using yet another research design. More specifically, we employed a similar setup as in the second study, but this time we used a within-subjects (instead of a between-subjects) design to administer the different decision-making models, which allowed us to test for the potential existence of different preference patterns. In order to be able to pinpoint more precisely where the general optimal level of desired citizen involvement is located, in this third study we included a total of eleven different decision-making models. These models ranged from 0% citizen and 100% government weight up to 100% citizen and 0% government weight, in small steps of 10% (see “Methods” for more details). For each resulting model, participants were asked to indicate the extent to which they found that particular model appropriate, justified, and acceptable (1 = not at all, 10 = very much so).

Results of study 3

A repeated measures ANOVA revealed that the eleven decision-making models significantly impacted participants’ overall evaluation (i.e., the mean of appropriate, justified, and acceptable), F(10, 284) = 71.73, p < 0.001, partial η² = 0.716. Figure 3 visualizes how participants scored the different decision-making models. We found the overall existence of an inverted-U (or, to be more precise, an inverted-V) curve. As shown in Fig. 3, participants’ overall evaluation increased up until the point where citizens and the government both have a decisional weight of exactly 50%. After this point, higher degrees of citizen involvement negatively impacted participants’ overall evaluation. Pairwise comparisons with Sidak adjustment for multiple comparisons showed that the model in which citizens and the government both have a decisional weight of 50% differed significantly from all the other models (all ps < 0.046).

It is important to realize, however, that the existence of such an overall pattern does not preclude the possibility of distinct clusters (i.e., subgroups) of individuals within the population, reacting differently to increasing levels of citizen (vs. government) involvement. To test this possibility, a k-means cluster analysis was conducted to categorize participants into different clusters, based on their responses to the different decision-making models. Because we hypothesized the existence of three subgroups within the citizenry (see “Introduction”), we extracted three distinct clusters, which are mutually exclusive (meaning that participants can only belong to one particular cluster). Figure 4 visualizes the corresponding curve of the three extracted clusters. Cluster 1 consists of a subgroup of participants (n = 94; 32.0%) who preferred the government to outweigh citizens. As shown in Fig. 4, in this first cluster, the decision-making model in which citizens have only 30% weight and the government has 70% weight was rated most positively. Cluster 2 contains a subgroup of participants (n = 139; 47.3%) who preferred the decision-making model in which citizens and the government both have an equal weight. Note that this particular subgroup closely mirrors the overall pattern (compare Cluster 2’s curve of Fig. 4 with the general pattern displayed in Fig. 3). Finally, Cluster 3 contains a subgroup of participants (n = 61; 20.7%) who preferred citizens to outweigh the government. Figure 4 illustrates that in this third cluster, the curve peaked when citizens have 70–80% weight and the government has only 20–30% weight. Interestingly, from Fig. 4 it can be derived that also in the ‘unequal balance’ clusters (i.e., Clusters 1 and 3), the model in which either the government or citizens have an absolute say (i.e., 100% weight) was clearly not considered the optimal model.

Finally, we conducted a multivariate analysis of variance (MANOVA), followed by post-hoc comparisons with Tukey HSD correction, to test if and how the three clusters (which were extracted based on participants’ responses to the different decision-making models) differ from each other with respect to our included demographics. The results of these analyses, which are summarized in Table 2, show that Cluster 1 consisted of significantly more male participants than Cluster 3 (p = 0.036). Although Cluster 3 also appears to be the youngest and least educated, the three clusters did not differ significantly from each other in terms of participants’ age and education level (see Table 2).

Study 4

A consistent data pattern emerged in our previous studies, but in these studies we always used a rather broad (i.e., general) description of citizen participation, without a specific context. An important aim of this fourth study (N = 409) was to test if our previously obtained results also hold when participants are presented with more concrete (i.e., specific) cases. To this end, participants in the present study were asked to evaluate eleven decision-making models (the same ones as we used in Study 3) in the context of a specific local decision (see “Methods” for details on the included cases). The decision-making models were presented in a similar way as in our third study (i.e., through a within-subjects manipulation), but in the present study their presentation order was randomized. Moreover, the present study also aimed to explore if we could identify personal trait characteristics associated with the emerging subgroups. For that reason, we additionally also measured individual differences in terms of three ideological attitudes (RWA, LWA, and SDO), two types of cynicism (social and political), the six HEXACO personality dimensions, and ideological left–right self-placement.

Results of study 4

A repeated measures ANOVA revealed a significant effect of the decision-making models on participants’ overall evaluation (i.e., the mean of appropriate, justified, and acceptable), F(10, 396) = 128.17, p < 0.001, partial η² = 0.764. Figure 5 displays the mean overall evaluation of the eleven decision-making models, separately for each of the four included cases. However, since the type of case did not interact significantly with the decision-making models, F(30, 1194) = 1.18, p = 0.231, partial η² = 0.029, we decided to collapse the data across the four different local decisions in all subsequent analyses. Figure 5 illustrates that, for each of the four cases, participants’ overall evaluation of the decision-making models steadily increased up until the point where citizens and the local government both have a decisional weight of 50%. Once this overall optimum was reached, the curves started to drop. Similar to Study 3, pairwise comparisons with Sidak adjustment showed that the model in which citizens and the government both have a decisional weight of 50% again differed significantly from all the other models (all ps ≤ 0.001).

Preference patterns were once more investigated using a k-means cluster analysis, in which we again extracted three distinct and mutually exclusive clusters, based on how participants responded to the different decision-making models. Figure 6 visualizes the curve of the three extracted clusters. From this figure, it can be concluded that we again found a first subgroup of participants who preferred the government to have a greater input than citizens (Cluster 1; n = 96; 23.5%), a second subgroup who preferred citizens and the government to both have an equal input (Cluster 2; n = 176; 43.0%), and a third subgroup who preferred citizens to have a greater input than the government (Cluster 3; n = 137, 33.5%). Interestingly, the pattern of the three clusters extracted in the present study was virtually identical to the pattern of the three clusters extracted in Study 3 (we invite the reader to visually compare Fig. 4 with Fig. 6). And, as in Study 3, in none of the three clusters participants preferred a model in which either citizens or the government has complete say.

We then conducted a MANOVA, followed by post-hoc comparisons with Tukey HSD correction, to test if and how the three clusters (which were extracted based on participants’ responses to the different decision-making models) differ from one another in terms of our included demographics and individual trait measures. The results of these analyses are summarized in Table 3. As shown in this table, we found that participants in Cluster 3 turned out to be significantly lower educated than participants in Clusters 1 and 2 (both ps < 0.001). Moreover, Cluster 1 again consisted of more male participants and also appears to be the oldest, but the cluster differences in this regard were not statistically significant. In light of the individual trait measures, Table 3 shows that participants in Clusters 1 and 2 scored significantly lower on LWA than participants in Cluster 3 (both ps < 0.025). Additionally, participants in Cluster 1 also scored significantly lower on political cynicism (p = 0.034) and Emotionality (p = 0.014) than those in Cluster 3. However, for all the other individual trait measures under scrutiny, no significant differences between the three clusters were found (see Table 3).

Study 5

Our prior two studies revealed the same pattern of results for abstract (Study 3) and more concrete cases (Study 4). However, a limitation of the within-subjects designs that were used in these studies is that they required participants to judge the different decision-making models sequentially, that is, one at a time. Based on the evaluability framework^43,44, it can be expected that people will be more sensitive to different degrees of citizen and government involvement when they evaluate the different decision-making models comparatively. Therefore, in this fifth and final study (N = 297) we used a pairwise comparison methodology to administer the different decision-making models. The same eleven models as in the prior two studies were included, which resulted in a total of 55 pairwise comparisons. For each of these comparisons, participants were forced to choose which of the two contrasted models they considered most appropriate (see “Methods” for more details). Two weeks later, during the second part of the study, participants (N = 240) completed the same individual trait measures as in Study 4.

Results of study 5

We first constructed a scale which numerically describes participants’ perceived appropriateness of the different decision-making models. This scale was estimated with a Bradley-Terry probability model using the Prefmod package⁴⁵ in R (version 4.1.1). The location of each decision-making model on this scale was estimated by means of a worth value. These worth values quantify participants’ perceived appropriateness of a given decision-making model, relative to the other models. Figure 7 visualizes the estimated worth values of the different decision-making models. A visual inspection of this figure shows that the decision-making model that was perceived as most appropriate was again the one in which citizens and the government both have an equal weight.

Preference patterns were subsequently investigated using the klaR package⁴⁶ in R, which allows for the clustering of categorical data. Based on participants’ responses to the pairwise comparisons, three distinct and mutually exclusive clusters were extracted, which are visualized in Fig. 8. This analysis revealed that the general pattern that we found again reflects the mere mean tendency of three distinct subgroups of citizens which each react differently to increasing levels of citizen involvement. As shown in Fig. 8, we again found a first subgroup who preferred the government to outweigh citizens (Cluster 1; n = 101; 34.0%), a second subgroup who preferred citizens and the government to both have an equal weight (Cluster 2; n = 116; 39.1%), and a third subgroup who preferred citizens to outweigh the government (Cluster 3; n = 80, 26.9%). Interestingly, although the estimated worth values displayed in Fig. 8 suggest that in none of the three clusters participants preferred a model in which either citizens or the government have a complete say, a closer inspection of participants’ responses nonetheless revealed that a small percentage of all people (about 10%) do seem to prefer a model in which either the government or citizens have full decisional control (i.e., 100% weight).

Finally, we again tested if and how the three clusters (which were extracted based on participants’ responses to the pairwise comparisons) differ from each other in terms of the included demographics and individual trait measures (which, as mentioned above, were measured during the second data collection phase). The results of these analyses, which are summarized in Table 4, clarify that participants in Cluster 1 were significantly older than those in Cluster 3 (p < 0.001). Similar to Study 3, Cluster 1 also consisted of significantly more male participants than Cluster 3 (p = 0.009). Although the three clusters did not differ significantly in terms of participants’ education level, the data again suggest that participants in Cluster 3 are the least educated. With regard to the individual trait measures, Table 4 shows that participants in Cluster 1 scored significantly lower on LWA than participants in Cluster 2 (p = 0.016) and Cluster 3 (p < 0.001). Similar to Study 4, we again found that participants in Cluster 1 also scored significantly lower on political cynicism than those in Cluster 3 (p < 0.001). Furthermore, in the presents study participants in Cluster 1 also scored significantly lower on Agreeableness than those in Cluster 2 (p = 0.045) and significantly higher on Conscientiousness than those in Cluster 3 (p = 0.006). For all the other traits, no significant differences between the three clusters were found (see Table 4).