The longstanding view that most democratic citizens do not meaningfully engage in politics has in recent years been given new life. Research has both reaffirmed the empirical case for the disinterested and unknowledgeable citizen (Achen and Bartels, Reference Achen and Bartels2016) and provided new theoretical rationales for how democracy functions with such citizens. For example, the literature on “stealth democracy” argues that many voters in the USA abhor partisan politics to such an extent that they would trade a contested, partisan policy-making process for one that empowers unelected experts and mostly relieves voters of the need to pay attention to politics (e.g., Hibbing and Theiss-Morse, Reference Hibbing and Theiss-Morse2002; VanderMolen, Reference VanderMolen2017).
These trends, however, contrast sharply with recent empirical research about citizens in multiparty democracies, which paints a very different picture of citizens' knowledge of, and engagement with, partisan politics. This includes work showing that voters perceive parties' left–right positions accurately (Adams et al., Reference Adams, Ezrow and Somer-Topcu2014; Somer-Topcu et al., Reference Somer-Topcu, Tavits and Baumann2020), know the sizes of the parties (Lee et al., Reference Lee, Haime and Stevenson2019), respond sensibly to parties' nonideological rhetoric (Jung and Tavits, Reference Jung and Tavits2021), and hold parties in governments accountable for their campaign promises (Matthieß, Reference Matthieß2020).
In this article, we seek to add to this growing body of evidence about the partisan knowledge of citizens in multiparty democracies by exploring the nature and sources of their beliefs about one of its central dimensions: the extent to which parties cooperate with one another on a day-to-day basis. This is an important focus of study not only because of its obvious relevance to the question of whether citizens are interested in and knowledgeable about partisan politics, but also because the idea that citizens have sensible beliefs about patterns of partisan cooperation and conflict plays a central role in many contemporary explanations of electoral behavior in multiparty democracies. This includes theories of coalition-directed voting (Bargsted and Kedar, Reference Bargsted and Kedar2009), how voters form expectations about likely coalitions (Hobolt and Karp, Reference Hobolt and Karp2010; Lachance, Reference Lachance2023), how voters perceive party policy positions (Adams et al., Reference Adams, Weschle and Wlezien2021; Lee et al., Reference Lee, Santoso and Stevenson2021; Hjermitslev, Reference Hjermitslev2023), rational retrospective voting (Duch and Stevenson, Reference Duch and Stevenson2008; Hobolt et al., Reference Hobolt, Tilley and Banducci2013), and voters' electoral responses to partisan cooperation (Harbridge and Malhotra, Reference Harbridge and Malhotra2011).
Given the importance of patterns of partisan cooperation to theories of electoral behavior in multi-party democracies, it is correspondingly important that scholars understand the sources and limitations of voters' beliefs about these patterns. Do voters have accurate beliefs about which parties cooperate with one another and which do not? And are these beliefs influenced by contemporary differences in the level of day-to-day cooperation and conflict, or are these events either missed by most voters, or discounted as relatively short-term tactical moves unreflective of more fundamental patterns of cooperation driven by factors like parties' ideological compatibility or a history of cooperation in government?
In this article, we report a first effort to answer these questions using original data that for the first-time measures voters' perceptions of the extent to which party dyads in four countries cooperate with one another. Further, we utilize the Quantified Political Relationships (QPR) data (Weschle, Reference Weschle2018) on real everyday interactions between parties to examine the extent to which voters' perceptions of partisan conflict and cooperation follow the changing real-time portrait of interactions reported in the news.
We find that voters' perceptions of cooperation and conflict among political parties do reflect their actual pattern of interactions, as reported by the media. Further, this relationship holds controlling for policy differences between parties, as well as (current or past) joint cabinet membership. Finally, we find that voters who monitor the political news carefully give a great deal more weight to contemporary events in their perceptions than do less engaged voters.
1. What drives perceptions of partisan cooperation?
In this note, we are interested in both the overall sensibility of voters' perceptions of partisan cooperation and conflict, and whether the most important drivers of these perceptions are short-term cooperative or conflictual events or longer-term relationships between parties that are rooted in their histories of co-governance and/or their relative ideological profiles. Thus, the main hypothesis that we seek to test is:
Hypothesis 1: Voters' perceptions of the extent to which any two parties cooperate or conflict will track news reports about real-world instances of cooperation and conflict between the parties.
The theoretical motivation for this is straightforward. When parties publicly cooperate or conflict with one another, these events are reported in the news. Further, under almost any model of how individuals process and use such mediated messages, they will have a discernible impact on aggregate perceptions of partisan cooperation and conflict. For example, this follows directly from on-line processing and memory-based models of perception formation (Zaller, Reference Zaller1992; Taber and Lodge, Reference Taber and Lodge2006).Footnote 1
Of course, contemporary political events are not the only potential drivers of citizens' perceptions of partisan conflict and cooperation. Beginning in school and early political socialization, most citizens of Western democracies learn about two other characteristics of (pairs of) parties that are likely relevant to their inferences about partisan cooperation and conflict: the broad ideological relationships between parties and the typical patterns of cooperation between parties in cabinet. Thus, we will also examine the following hypotheses:
Hypothesis 2: Voters will perceive any two parties to be more cooperative the more the parties agree on policy.
Hypothesis 3: Voters will perceive any two parties to be more cooperative the longer the two parties have served together in cabinet.
Finally, voters clearly differ in the extent to which they are interested in politics and in how closely they follow the political news (their levels of “habitual news reception”). The theoretical narrative that motivates hypothesis 1 is simply that cooperative and conflictual events between parties will drive voters' perceptions of partisan conflict and cooperation because voters see reports of these events in the news. If voters do not see such reports, the path from the events to perceptions is broken. We should therefore expect that evidence for hypothesis 1 will be less prevalent among citizens who follow the political news less faithfully:Footnote 2
Hypothesis 4: The relationship described in hypothesis 1 will be stronger for individuals with higher levels of habitual news reception.
2. Measurement
In this section, we briefly describe how we measure our dependent variable (voters' perceived levels of cooperation and conflict between each party dyad), our measure of real-world cooperation and conflict between parties (hypotheses 1 and 4), and Habitual News Reception (hypothesis 4). Additional detail on the measurement of these variables, as well as the other variables in hypotheses 2 and 3, can be found in Appendix 6.
2.1 Measuring perceptions of partisan cooperation
Our measure of voters' perceptions of partisan cooperation comes from a question we asked in seven original surveys that we conducted in Demark (2019), Germany (2018, 2019), Canada (2017, 2019), and the UK (2017, 2019).Footnote 3 In each survey, respondents were shown a pair of parties and asked the following question:
How often do you think these two parties cooperate with each other in [your country's] federal politics? [answer categories: (0) “Never cooperate” to (10) “Almost always cooperate”]
We asked each respondent about each pair of parties that had seats in the legislature at the time of the survey.Footnote 4 We did not allow respondents to skip the question for any dyad or to say “Don't Know.” Instead, they were encouraged to guess if they were uncertain.Footnote 5
As an illustration, Figure 1 shows the distributions of all party-dyads from our survey in Germany in 2019. Clearly, this figure (which is typical of the others) suggests that German voters have no trouble answering the question in sensible ways. The highest average cooperation score (6.58 on a 0–10 scale) is between the CDU/CSU and the SPD, which had been in a governing coalition at the time of this survey for six years. Likewise, the traditional pairings of the SPD and Greens and CDU and FDP come next at 5.85 and 5.6, respectively. Finally, German voters—quite sensibly—identify the least likely cooperators as the various left parties (Green, Linke, and SPD) with the far-right AfD. These sensible patterns of results extend to the larger set of countries and cases (see Appendix 3).
Figure 1. Distribution of perceived cooperation among German parties in 2019. Dashed lines indicate average cooperation scores.
2.2 Measuring partisan cooperative and conflictual events
To measure the patterns of real-world episodes of cooperation and conflict between parties as reported in the news, we use the QPR data developed by Weschle (Reference Weschle2018). It uses large-scale machine-coded event data from news reports as inputs to latent factor network models to provide a measure of the cooperative or conflictual tone of the relationships among a large set of political and societal actors.
The event data are taken from the Integrated Crises Early Warning System (ICEWS) project (Boschee et al., Reference Boschee, Lautenschlager, O'Brien, Shellman, Starz and Ward2015). As its source material, it takes reports published in several large media repositories that collate material from hundreds of national and international news sources. These stories are machine-coded by a natural language analysis system to identify and extract an event's source and target as well as the event type, which is classified into one of more than 350 categories. The QPR dataset focuses on domestic interactions only and further processes the events in two ways: First, every actor is hand-coded as being partisan-political, nonpartisan-political, or societal. All politicians from the same party are aggregated into a single partisan actor. Second, the various event types are dichotomized into either cooperative or conflictual.Footnote 6
To see what the data look like in practice, consider a 2018 report on the fallout of a speech by German President Frank-Walter Steinmeier of the SPD, in which he had called for a new immigration law. The article reports that “Greens co-leader Cem Ozdemir… welcomed the intervention of the president.” In our data, this is coded as a cooperative interaction between the Green Party as the source and the SPD as the target.
Next, the large number of such cooperative or conflictual events recoded in the ICEWS data is summarized using a latent factor network model (Minhas et al., Reference Minhas, Hoff and Ward2019). The basic idea is to infer the network that gives rise to the observed patterns of interactions by locating all actors in a low-dimensional social space. The relation between a pair of actors can be computed from their positions in this space (for technical details, see Appendix 6). Larger positive scores indicate a more cooperative relationship, and more negative values a more conflictual one.Footnote 7 Importantly, cooperation scores not only take direct interactions into account, but also third-order relations (e.g., a friend of a friend is a friend). This is important since voters likely consider such higher-order relations when assessing levels of partisan cooperation and conflict. For example, if unions cooperate with one party but conflict with another, this provides voters with information that the two parties likely have conflictual relations.
We use cooperation scores based on all reported events in a country in the 365 days before the first day a given survey was in the field.Footnote 8 For a few of the parties in our surveys, we do not have a cooperation score. This is the case if the ICEWS data do not contain any events involving the party in the relevant period, which can happen with smaller parties. In our preferred specifications, we drop all dyads for which we do not have cooperation data. However, results are robust when imputing scores of zero or the country-specific mean for dyads without a cooperation score (see Appendix 2).
2.3 Measuring habitual news reception
Our scale of habitual news reception relies on the well-validated technique introduced by Price and Zaller (Reference Price and Zaller1993), which asks respondents about the details of three increasingly widely reported news items, chosen in the days just before the survey was fielded. In addition, we added three items intended to tap knowledge that could only have been obtained from mediated sources in the medium term (the inflation rate and the occupants of several elected offices). Using this combination of items may be more accurate for respondents who may have not been monitoring the news in the days before the survey, but usually do so.Footnote 9
3. Research design and methods
To explore hypotheses 1–4, we estimate a series of regression models in which a row of data is a respondent-[party-dyad] and the dependent variable is the respondent's perceived dyadic cooperation score (0–10). Our independent variables are the QPR cooperation score, measures of parties' ideological distance (actual or perceived), and variables capturing their joint cabinet membership (actual or perceived, past, or present).
Since each respondent answered about many dyads and each dyad was asked of many respondents, we use multilevel regression models that include crossed random intercepts for respondent and dyad. Because the QPR score and the manifesto-based RILE measure of party left–right positions are themselves estimated quantities, we propagate their standard errors through our empirical models by calculating bootstrapped standard errors (see Appendix 7).
Finally, as explained in more detail in Appendix 6, we have measured each of the theoretical concepts in our hypotheses in several different (but equally justifiable) ways. For the events data and histories of cabinet participation, we have used various time windows. Likewise, for party policy differences and patterns of joint cabinet participation, we have both perceptual and objective measures. We therefore estimate a series of models, each of which includes one possible measure. For robustness (and because we have no strong theoretical reason to prefer any one measure of each concept), we estimate models for each permutation of these various measures (see Appendix 2). As it turns out, the substantive results are remarkably similar across all specifications. In the results reported below, we therefore choose six representative models that capture the substantive story well.
4. Results
Table 1 shows the results of our models. The most important message from this table is simply that every estimate, in every model, conforms to our expectations. The positive and significant coefficients on cooperative and conflictual events in models A–F all confirm hypothesis 1—suggesting that citizens, on average, do pay attention to the vagaries of contemporary events as reported by the media when thinking about how much parties cooperate or conflict. The effect is substantively meaningful. A one standard deviation increase in reported cooperation is associated with a 0.092–0.116-point increase in the perceived level of cooperation. This is comparable to the effect of a 3.5–4-point decrease in the RILE left–right distance between two parties, or about one-sixth of the effect of the respondent believing that two parties are in a coalition together.
Table 1. Drivers of perceptions of partisan conflict and cooperation
Shaded cells are statistically significant at p < 0.05. Standard errors are bootstrapped to account for uncertainty in manifesto scores and our measure of events.
At the same time, both of our measures of the ideological spread of the dyad are negative and statistically different from zero, as expected in hypothesis 2. When parties disagree more on policy, voters see them as less cooperative. Finally, current joint cabinet membership and a history co-governance are strongly associated with perceptions of greater partisan cooperation. This confirms hypothesis 3.
In models G–L, we present results relevant to hypothesis 4. In these specifications, we interact the cooperation score with our measure of habitual news reception. To better depict how the relationship between media-reported cooperation and perceived level of cooperation differs across different levels of habitual news reception, Figure 2 plots predicted values for the interaction terms (model G), while holding constant the other variables.
Figure 2. Marginal effects of cooperation score on perceived levels of cooperation, by level of habitual news reception.
For respondents with low, medium, and high levels of habitual news reception, a one standard deviation increase in the cooperation score is associated, respectively, with a 0.02, 0.09, and 0.15-point increase in perceptions of partisan cooperation. These results are clearly consistent with our expectation that the effect of conflictual and cooperative partisan events reported in the news will be larger for respondents with higher levels of habitual news reception.
5. Conclusions
Do voters have accurate beliefs about which parties regularly cooperate with one another and which do not? Our original data on voters' perceptions of partisan cooperation and conflict make it clear that, on average, voters perceive levels of cooperation between parties that are consistent with what most political scientists familiar with the history of these parties would expect. In particular, we find that differences in perceived cooperation across party-dyads correspond closely to differences in the policy/ideological compatibility of different parties, their current levels of cabinet cooperation, and their histories of co-governance. Beyond the long-term sensibility of perceptions and their drivers, we have also shown that perceptions reflect the recent record of cooperative and conflictual events that make up the day-to-day interactions between parties as reported in the media. Indeed, these effects are apparent even after we account for the longer-term factors mentioned above. Likewise, our finding that the effects of the ongoing media narrative about party interactions are dramatically larger for individuals who pay close attention to the political news supports the idea that there is a direct connection between media portrayals of partisan conflict/cooperation and voter perceptions.
Of course, we would be remiss if we did not also point out the limitations of the current study. Specifically, no cross-sectional study can fully validate the kind of dynamic story intimated here. While we certainly think the evidence we present is consistent with the idea that voters update their beliefs about party conflict and cooperation in responses to real-world changes in those relationships, in this design we cannot definitively rule out the possibility that some other variable that we do not control for both causes the media to portray a given dyad as more or less cooperative and also influences voter perceptions of this level of cooperation. To rule out such confounders definitively would likely require that we track changes in perceptions over time, ideally with a panel of voters. While we leave that effort to future work, we do think the preponderance of the evidence thus far suggest that citizens not only have a firm grasp of the cooperative or conflictual nature of partisan relationships in their systems, but that they also update these perceptions as the political landscape changes.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2023.20 and replication materials at https://doi.org/10.7910/DVN/THDW2T.
Open access
