![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
This article argues that despite the importance of national party politics in Britain, constituency-level electoral marginality makes it easier for citizens to select parties based on policy. Using data from the 2005 and 2010 British Election Studies, I find that citizens residing in marginal constituencies are more likely to perceive parties' relative ideological positions correctly, and that electoral marginality promotes knowledge by increasing both the attention that parties pay to a constituency and citizens' engagement in the election. Finally, preliminary findings suggest that citizens do not necessarily perceive differences in the tone of the campaign material.
Acknowledgements
I would like to thank Ethan Scheiner, James Adams, and Matt Buttice for their invaluable input (and never-ending patience) throughout the many iterations of this project.
Notes
1 More recent studies draw our attention to informational differences between coalition and non-coalition systems. These studies point to the ability of voters in coalition systems to determine the coalitions that are more or less likely to form (Armstrong & Duch, Citation2010; Duch et al., Citation2010; Fortunato et al., Citation2011), and how coalition membership alters parties' policy positions (Fortunato & Stevenson, Citation2013). Such abilities demonstrate sophisticated political reasoning, and are consistent with the observation that citizens in coalition systems, such as Denmark or the Netherlands, are substantially more informed than their counterparts in systems that tend to be characterised by single party governments, such as the United States or Great Britain (Stevenson & Vonnahme, Citation2010).
2 While there are no survey questions available that facilitate an analysis of whether marginal respondents receive more campaign materials, 48% of 2010 BES respondents from marginal constituencies report being contacted by a party prior to the 2010 election, compared with only 33% of respondents in safe constituencies.
3 Though the difference in means is modest, it is statistically significant at p<0.05.
4 In 2010, BES respondents were asked, “How closely did you follow the election campaign?” On a scale from 0 “not closely at all” to 3 “very closely”, the average response for respondents in marginal seats was 2.06, compared with 1.87 for respondents in safe seats. The difference between the groups is statistically significant at p<0.01.
5 Both the 2005 and 2010 election surveys rely on a sample of constituencies. The 2005 BES contains 127 constituencies that were sampled, while the 2010 BES (CSES Internet Survey) includes respondents in 482 constituencies.
6 One could also capture political knowledge using an individual's ability to recall the identity of elected officials or candidates (e.g., Jacobson, Citation1992; Pattie & Johnston, Citation2004), but questions of this nature are not included in these surveys.
7 Johnston and Pattie (Citation2011) demonstrate that even though Great Britain may be classified as a three-party system nationally, the majority of constituency races are two-party contests.
8 Respondents who opt not to place one (or both) of the parties are coded as having inaccurate perceptions. Because the inability to place either (or both) of the parties demonstrates a lack of political knowledge, omitting these respondents will bias the sample in favour of the politically sophisticated. However, the results remain unchanged if these respondents are coded as missing and omitted from the study.
9 Marginal constituencies are defined as those where the margin of victory was less than 5%.
10 If respondents who are unable to place the parties are omitted from the analyses, the percentage of respondents who have correct relative placements is considerably higher: 63% in 2005, and 80% in 2010.
11 Pattie et al. (Citation1995) argue that British parties look to previous election results when deciding how to allocate campaign resources, a view that is consistent with Laver's (2005: 264) model of adaptive parties that “look backwards and learn from the past, developing simple rules of thumb that condition future behaviour on the recent history of the system” (see also Somer-Topcu, Citation2009). Constituency-level pre-election polls would theoretically be an alternative way of capturing the expected marginality, however, constituency-level polling is not widely used in British elections. Parties did commission local polls from time to time, but this information is not necessarily made available to the public (Johnston & Pattie, Citation2013).
12 The original question wording for all variables used in the article is given in the appendix. For comparability, all demographic variables are rescaled to fall between 0 and 1. For example, in the 2010 BES, age ranges from 22 to 92 years. When Age is recoded, the minimum value (22 years) is now set equal to 0, while the maximum value (92 year) is set equal to 1. The findings remain unchanged if the original variables are used.
13 The structural equation modelling approach uses the observed variables to estimate the latent constructs, and then estimates the correspondence between the observed variables and their respective latent constructs by determining the amount of variance in the observed variable that is explained by the latent construct.
14 There are limits on candidate/party expenditures per constituency. The spending caps vary by constituency, taking into account both the size of the electorate and whether it is a borough or county constituency.
15 The factor loading for party spending is constrained to 1 to provide a scale for Party Attention.
16 For consistency and comparability, all variables are rescaled to fall between 0 and 1. The factor loadings for interest in the election in the 2005 model and following the campaign in the 2010 model are constrained to 1 to provide the scale for Engagement. For both models, the results are robust to the omission of individual components of engagement, as well as the use of unscaled variables.
17 The covariance between the measurement errors and the disturbance term are assumed to be zero (Bollen, Citation1989; Finkel, Citation1995).
18 AMOS does not use listwise deletion to deal with missing data. Rather, it uses maximum likelihood techniques to obtain parameter estimates in the presence of missing data (Arbuckle, Citation2010). Thus, the number of observations included in the SEM analyses is higher than the logit analyses presented above because all respondents are included regardless of whether there is missing data with respect to their individual-level characteristics.
19 In the AMOS Markov chain Monte Carlo (MCMC) procedure, 2,000 initial samples are generated and discarded. To minimise the amount of information that the priors contribute to the joint posterior density, all parameters are given vague, normally distributed priors with a mean of 0 and variance of 100. Convergence is accepted once the model convergence statistic becomes less than 1.002, and posterior distributions, the trace plots and the autocorrelation plots also confirm that all parameters have converged. The results presented here represent summaries of 30,000 draws from the joint posterior density of the model.
20 The credible interval is the Bayesian equivalent of the confidence interval. When the marginal posterior distribution is approximately normal, the 95% credible interval will be approximately equal to the posterior mean ± 1.96 posterior standard deviations. In contrast to the traditional confidence interval, the Bayesian credible interval is interpreted as a probability statement – i.e., if the Prob(a ≤ θ ≤ b) = 0.95, one can be 95% sure that the true value of θ lies between a and b.
21 Higher numbers indicate that the respondent placed the top two parties further apart.