https://orcid.org/0000-0002-3032-4992
![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
ABSTRACT
Do voters update their perceptions of party positions? How certain do voters feel about their perceptions? We argue that perceptual updating and perceptual certainty depend on how divided or unified political parties are in their policy positions. We argue that voters do not accurately update their perceptions and that they become less certain about their perceptions when divided parties shift positions. For unified parties, we expect voters to accurately update party positions and become more certain about their perceptions. We test our arguments in the context of the European integration issue in the UK. Using original newspaper data on British parties’ messages on European integration and five waves of the British Election Study (2014–2016), we find evidence in support of our arguments. The paper has important implications for research on party position-taking and its consequences on voters.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 Another concern is that the UK media system has traditionally been different from other European media systems. Hallin and Mancini (Citation2004) note that the UK’s Liberal Model is characterized by the dominance of market mechanisms, while the other European models have more active state involvement. Yet, as they also note, “by the beginning of the twenty-first century, the differences [between different media models] have eroded,” and “the Liberal Model has clearly become increasingly dominant across Europe” (251).
2 The heuristics-based framework also informs comparative work on voters’ knowledge of party positions (e.g., Adams, Ezrow, and Wlezien Citation2016). It shows that coalition membership is an important cue about party positions. Coalition membership is party-provided information, even though it is not in the form of messages, but in the form of party behaviour.
3 We argue that message clarity declines as internal division increases. Nonetheless, we acknowledge that clarity may also depend on how ambiguous the party is, independent of division (Rovny Citation2012; Citation2013; Somer-Topcu Citation2015; Lehrer and Lin Citation2018). We do not argue that clarity solely depends on party dissent/unity. As Rovny (Citation2012) explains, “Position blurring, understood here as the deliberate misrepresentation of party positions on some dimensions, is conceptually distinct from intra-party dissent” (272). One way to distinguish the two concepts might be to think about the dimension of competition and determine whether the party has incentives to blur on that dimension strategically. We do believe, however, that the implications of strategic ambiguity and intra-party dissent go together since strategically ambiguous parties are also likely to decrease perceptual accuracy and certainty. Nevertheless, the relationship between dissent and ambiguity is an empirical question, which we leave for future studies.
4 We clarify that we are interested in both updating and certainty because they are concepts that go together. The level of certainty about a party’s position depends on whether and how voters perceive the party’s position. Although our results for certainty are substantively stronger (see below) and certainty is a more novel concept, because certainty follows updating, we study both outcomes.
10 One final practical reason for our focus on European integration in the UK is the unique availability of survey data on voters’ perceptions and certainty about party positions.
11 Newspaper data is only one type of media data. Nevertheless, as Peterson (Citation2019) finds, “newspapers remain an important contributor to political awareness in a changing media landscape, even for those with limited political interest” (1).
12 Three research assistants coded each article in the CCD dataset. For each coded statement, they also indicated how confident they are. The final data include only the responses on which either two or more students agreed (while being at least “mostly confident”) or where at least one assistant was “fully confident” in their coding.
13 To be clear, we fully acknowledge that our estimates of party positions from newspaper coverage are not perfect. There are issues of media bias that cannot be eliminated even though the CCD project took great care to minimize such issues by, for example, using both left-wing and right-wing broadsheets and recording proportions, as opposed to just absolute numbers, of issue mentions. Therefore, we use the word “estimates” when we talk about measures from the CCD data to be clear that we are estimating parties’ positions by taking stock of the strengths of the CCD project, in spite of its limitations. However, as reporters have much agency about what to report and how, it would be an interesting topic of future research to investigate how the CCD data, and media portrayals generally, differ from other existing measures of party positions.
14 We included immigration coverage into our measure because in the UK around the time of this study, immigration was highly associated with EU politics. See, e.g., “Immigration: Brexit an opportunity to fix ‘broken’ system” (https://www.bbc.com/news/uk-politics-37177937). Moreover, according to the 2014 and 2017 Chapel Hill Expert Surveys, the correlations between experts' mean assessments of positions on the EU issue and the immigration issue for Labour, the Conservatives, UKIP, and the Lib Dems are 0.972 (in 2017) and 0.999 (in 2014). The correlations are 1 (in both 2017 and 2014) if we only examine the united UKIP and Lib Dems. In terms of expert-level (not averaged) correlations, the values are 0.914 for all parties and 0.962 for UKIP and the Lib Dems in 2014. In 2017, the correlations are 0.842 for all parties and 0.961 for UKIP and the Lib Dems. These high correlations are additional evidence that the EU and immigration issues were intertwined.
15 Although the CCD follows the measurement approach of CMP, we do not argue that CMP’s approach is not without flaws. There are alternative measures of party positions, such as the Chapel Hill Expert Survey, that complement the weaknesses of CMP’s emphasis-based approach.
16 We find that the correlation between the CCD position estimates and existing expert survey data on party positions is 0.803 and statistically significant, suggesting the validity of the CCD measures. This is further discussed in OA2. We would have liked to test correlations with data that reflect parties’ direct messaging, such as manifestos and leadership speeches, but that is very difficult because the CCD data cover inter-election periods. Yet, party manifestos are written for elections only, and it is not obvious what would be the most representative speech to analyze in inter-election periods.
17 One may notice that the divided parties are large and that the united parties are small. This has both theoretical and empirical implications. Theoretically, since party size is most likely to be causally prior to party division, it makes sense to argue about the effect of party division (instead of party size) on voters' perceptions. Empirically, because we examine divided and united parties separately, we are essentially controlling for the influence of party size on voters' perceptions.
18 We note that individual fixed effects models are inappropriate for our data. First, we have a very short panel. Some respondents answered all five waves, but there is attrition and new respondents were included in the middle. Furthermore, the main predictors measure change, so we have data on a respondent for at most four time periods. Second, when we analyse Certainty, the limitations of statistical packages do not allow us to run fixed effects models with an ordinal outcome.
19 The BES does not consistently ask about newspaper consumption, political knowledge, or political interest. Therefore, we use education as a proxy.
20 We do not have controls for media reporting because the CCD project measures (1) coverage for all parties using the left-wing Guardian and the right-wing Telegraph and (2) positions on European integration using shares of the issue discussed out of all issues discussed by the party. Hence, we do not think there is variation in the type or quality of media reporting across parties that would confound the relationship between position shifts and voter perceptions.
21 We do not have control variables that account for party behaviour because we cannot think of observable party-behaviour factors that are causally prior to position shifts and correlated with voters' perceptions. Moreover, such controls are likely to overfit our models because our models include only two parties (of similar size) and already have wave fixed effects, respondent clustered standard errors, and individual-level controls.
22 The number of observations in the second model is smaller than that in the first because the BES asked the certainty questions to a random subset of respondents who were asked the perceptual position questions.
23 We would like to clarify that it is not counterintuitive to find updating in the opposite direction. First, the range of the outcome variable is larger than that of the main predictor. Updating ranges from -10 to 10, while Policy change ranges from -0.10 to 0.08. Therefore, the finding that, for example, voters move their perceptions to the right by 0.796 when the party moves to the left by 0.082 is not a disproportionately large movement on the voter's part. Second, substantively, because Labour and Conservative often delivered both pro-European and anti-European messages, it is not unreasonable that voters perceive the party to have moved toward the opposite direction in which the party moved toward on average.
24 One might ask whether it is difficult for divided parties to make large policy shifts. We find in our data that divided parties indeed make smaller shifts than united parties do. For example, the average absolute policy shift for the Conservatives is 0.057, that for Labour is 0.069, Lib Dem 0.109, and UKIP 0.315. However, we do not think policy shifts by divided parties are conceptually invalid as those parties still move.
25 To clarify, our main predictor is a measure of change, not a measure of ideological position on the left-right scale. When we talk about parties moving from one standard deviation below the mean (-0.252) to above the mean (0.299), we are saying that the party moves to the left by 0.252 and then moves to the right by 0.299. The party ends up moving by just 0.047 to the right. These are not unrealistically big changes in party position in the -1 to 1 scale.
26 As with Policy change, Absolute policy change measures change in position. Hence, when we posit that Absolute policy change moves from 0.091 to 0.417, the party is moving by 0.091 and then by 0.417. These are reasonable shifts, considering that the range of absolute movement by UKIP and the Lib Dems is from 0.027 to 0.56 (see Table OA2.1).
27 We acknowledge that our data have limited time and party variation—two parties and four waves in each model. We have tried to address this by including wave fixed effects, respondent(-party) clustered standard errors, and respondent-level controls, which serve to account for correlations between responses in each wave and about each party. Nonetheless, we acknowledge that we need longer panel data and data on multiple parties (most likely through multi-country data) to have further confidence in the generalizability of our findings.
28 One might ask whether there is a curvilinear relationship between Absolute policy change and Certainty. We discuss the analyses and results in OA6. Moreover, it might be helpful to know party-level correlations between policy shifts and voter perceptions across waves. The analyses and results are explained in OA7.
29 An exception is Ezrow, Homola, and Tavits (Citation2014), who argue that centrist party positions increase voter uncertainty, which in turn negatively affects party performance.
Additional information
Notes on contributors
Jae-Hee Jung
Jae-Hee Jung is a postdoctoral fellow at the Blavatnik School of Government and Nuffield College at the University of Oxford.
Zeynep Somer-Topcu
Zeynep Somer-Topcu is an associate professor in the Department of Government at the University of Texas at Austin.