How do Voters Form Perceptions of Party Positions?

Abstract Political debates are structured by underlying conflict dimensions, such as left-right and economic and cultural ideology, which form the basis for voter choice and party competition. However, we know little about how voters arrive at perceptions of parties' positions on these dimensions. We examine how the emphasis parties place on the different issues that make up a higher-level ideological dimension affects perceptions of their position on that dimension. Using two population-based survey experiments, we present respondents with either short or long statements that communicate the same issue stances. We then test whether the length of statements affects positional perceptions on the higher-level dimension. The empirical results show support for our hypotheses and imply that political parties – and the context in which they compete – can affect their perceived position even if underlying issue stances remain stable.

In competing for electoral support, political parties face a challenging task: While there are often strong incentives to adapt their ideological positioning to increase their appeal among voters (Adams et al. 2004), such shifts always carry the danger of creating a reputation for unreliability ( Fernández Vázquez 2019;Meyer 2013) and damaging the party brand (Lupu 2014).One potential solution to such Catch-22 situations may be the lever that issue salience provides: By modifying which issues they emphasize, parties may be able to move on ideological dimensions without changing their positions on underlying issues.While this idea has been suggested as a theoretical proposition (Van der Brug 2004) and tested using observational evidence (Giebler, Meyer, and Wagner 2021;Meyer and Wagner 2020), we lack robust causal evidence of the impact of party emphasis on perceptions of the parties' ideological positions.
Therefore, in this article, we provide an experimental test to determine whether party emphasis on specific issues can modify where voters place a party on higher-level ideological dimensions.Higher-order dimensions can be a one-dimensional space, such as that assumed in accounts using the left-right or liberal-conservative dimension (Arian and Shamir 1983;Mair 2007) or more specific ideological dimensions such as the economy or culture (Hooghe, Marks, and Wilson 2002;Kriesi et al. 2008).Perceptions of party positions on ideological dimensions are, at least in part, based on perceptions of party positions on the individual issues tied to these dimensions. 1 So, a party's position on legalizing marijuana tells voters something about its position on the broader law and order dimension, and a party's views on estate tax tell voters something about its position on the economic dimension.
Parties should be able to change voter perceptions of their position on a higher-level policy dimension by changing the emphasis on their positions on individual issues tied to this dimension.In Downs ' (1957, 132) words: 'each party takes stands on many issues' and a party's net position 'is a weighted average of the positions of all particular policies it upholds'.If this is true, perceived party policy positions should differ depending on how much parties emphasize their left-or right-leaning positions on underlying policy issues (Meyer and Wagner 2020).This has implications for how parties shift positions on ideological dimensions as (de-)emphasizing an issue stance can change a party's perceived position on the higher-order dimension (Budge 1994).We therefore expect that the emphasis parties place on each issue position affects the weight of this issue position in perceptions of higher-level ideological dimensions.
In this paper, we test this argument using survey experiments.Previous research has focused on observational data (Meyer and Wagner 2020) but experiments are needed to account for three limitations of observational studies.First, in observational studies researchers cannot be certain on which information respondents base their perceptions.An experimental setting allows us to control the information respondents are exposed to.Important confounderssuch as the credibility of policy stancesare ruled out by design.Second, in observational settings, perceptions are necessarily influenced by prior perceptions and partisan bias (Slothuus 2016).In an experimental setting, we can create a situation where respondents evaluate a completely new party, providing a clean test of the proposed mechanism.Third, observational studies cannot rule out that citizens learn about or infer a party's position on issues such as gay marriage based on the party's broader ideological stances on sociocultural issues.An experimental setup can remove the risk of such reverse causality.
The survey experiments we conduct present respondents with short texts describing a (hypothetical) party's position on two socioeconomic (Study I) and two sociocultural issues (Study II).On one issue, the party has a liberal position, while on the other issue the party's position is more conservative.We then ask respondents to place the party on a higher-order economic and a sociocultural dimension, respectively.The treatment in both experiments is the emphasis on the respective issue statement: We expect that respondents pay more consideration to the issue position that parties emphasize more.Our analysis accounts for the possibility that emphasis (measured as statement length) also contains positional signals.Our findings fully support our expectations.
This study contributes to our understanding of how parties form perceptions of the ideological positions of political parties.These questions have been addressed by a growing body of research, which shows that voters infer positions from diverse sources, including party rhetoric (Fernández Vázquez 2014), party policy shifts (Dahlberg 2009), governing coalition arrangements (Adams, Ezrow, and Wlezien 2016;Falcó-Gimeno and Fernández Vázquez 2020;Fortunato and Stevenson 2013), the extent of public cooperation (Adams, Weschle, and Wlezien 2021), leadership changes (Somer-Topcu 2017), issue salience (Meyer and Wagner 2020), and government policy (Seeberg, Slothuus, and Stubager 2017).Moreover, individual voters also differ in their ability to understand the ideological positions of the parties; for instance, based on their level of political sophistication (while contextual factors also moderate these cognitive processes [Carroll and Kubo 2018]).Yet, one important shortcoming of research into the origins of positional perceptions is its almost exclusive reliance on observational data (for exceptions, see Falcó-Gimeno and Muñoz 2017;Fernández Vázquez 2019).
Hence, our paper provides robust causal insights into a universally important question; namely, how voters formulate perceptions and understandings of political parties.Importantly, our findings show that salience and positional theories are not alternatives, but complementary ways of studying party competition, confirming the conclusions of Meyer and Wagner (2019).This paper also contributes to our understanding of perceptions of party policy change.As issue emphasis matters for positional perceptions, parties can shift their overall position without changing their views on any issue.Given that changing views may be a hard sell to activists and may lead to media questions about inconsistency, this is an attractive strategy for parties.Our results also mean that contextual informationfrom the media or rival partiescan shift a party's perceived ideological position even if the party does not change its rhetoric.Our conclusion further explores these implications.

Research Design, Data and Methods
We test the hypothesis using data from two survey experiments that manipulate issue emphasis.Whereas the experiments were fielded in Germany and Austria for reasons of convenience, we expect potential context effects to be minor given that we use completely fictional texts.Nevertheless, respondents with less exposure to programmatic party competition (Kitschelt 2000) could, potentially, be less used to inferring dimensional positions from individual issues (Carroll and Kubo 2018).
These two experiments share a basic setup: We ask respondents to place a party on an issue dimension (for example, the economic dimension) based on (hypothetical) statements the party expresses on two issues.For each issue, we formulate a long and a short version to express differences in issue emphasis.The long version contains about sixty words and is based directly on party statements that we found in election manifestos and party websites.This statement does not contain party labels, and the language was simplified by removing technical terms.The short version is the first sentence of the long version and contains about ten words.This first sentence clearly states the party's position, so the short version is a succinct summary of the position expressed in the long version.
We choose two issues for each dimension.The party expresses a progressive position on one issue and a conservative position on the other.We show respondents the issue statements on both issues in a random order.Respondents thus saw either: (1) a long progressive and a short conservative statement; (2) two short issue statements; (3) two long issue statements; or (4) a long conservative and short progressive issue statement. 2Our key statistical test is based on a comparison of groups 1 and 4 with group 2. Specifically, our hypothesis is confirmed if respondents in groups 1 and 4 give greater weight to the issue position expressed in the longer issue statement when placing parties on the higher-level dimension. 3 To test whether differences across groups are due to different perceptions of the issue positions expressed in the long and the short statements, we also ask respondents to place the party on both issues.Doing so also allows us to distinguish whether changes in text length have a direct effect on the perceived position (for example, economic policy) or whether the effect is mediated by the perceived position of the party on the two underlying policy issues (for example, tax cuts).
Study I was conducted as an online survey in Germany in November 2020.The quota sample from an online access panel (N = 3,150) is representative of German citizens in terms of age, gender, and education.
In this study, we asked respondents to assess a party's economic policy position.We first describe what we mean by left and right positions on the economy. 4We then present statements of a hypothetical party.Box 1 shows the vignettes used in this study.One issue statement (pensions and care for the elderly) expresses support for more services, thus advocating a left economic policy position.The other statement expresses support for tax cuts, a traditional right-wing position on the economy.

2
Balance tests (Appendix C) suggest that the randomization worked as intended: There are no systematic differences across groups with regards to respondents' age, gender, education level, or their feeling of being close to a political party.
3 Our approach differs from scaling methods for differential item functioning (DIF); e.g.Aldrich and McKelvey 1977;Hare et al. 2015) in two important ways.While these approaches aim to account for different interpretations of issue scales across respondents, we are primarily interested in voters' responses to party-induced changes in issue attention.In addition, DIF studies derive latent policy positions (of parties and candidates) based on a single issue dimension, whereas we argue that these perceptions are shaped by party stances on underlying issue positions.
In each vignette, two aspects are randomized: The length of the issue statement and the order of the two issues.For each issue, the first sentence (equivalent to the short version) summarizes the party's position; in the long version, the remainder of the text justifies the position and outlines (more) concrete policy proposals (for example, minimum pensions and tax reforms).
Respondents are first asked to place the party on a 0-10 economic left-right scale.This variable is the dependent variable in the empirical analysis.Next, we asked respondents to locate the party on the two individual issues: The scale for taxes ranges from 0 ('more taxes and charges') to 10 ('fewer taxes and charges'); for pensions and care, it ranges from 0 ('more benefits for the elderly') to 10 ('fewer benefits for the elderly').The order of the two issue-specific questions was random.
Fig. 1 shows the distribution of responses for both issues. 5To test whether text length (that is, issue emphasis) affects perceived positions, respondents need to identify one issue as economically left and the other as economically right.Indeed, the mean perceived position on pensions is 2.76, and the position on (cutting) taxes is 6.63.
Box 1. Vignettes in study 1 [First/Second], we need to raise pensions and expand health and long-term care provision.In recent years, higher rents and sharply increasing prices have made life noticeably more expensive for senior citizens, which makes it more difficult for them to participate in public life and thus leads to social impoverishment.We, therefore, call for the introduction of a minimum pension of 1,200 euros per month and a sustainable guarantee of health and nursing care for senior citizens.
[First/Second], the income tax in Germany must be noticeably reduced.Germany is a high-tax country.The tax burden continues to rise and people have less and less money in their pockets.We, therefore, call for a tax system that rewards personal effort.The state must handle taxpayers' money responsibly and can only demand as much from citizens as is necessary to fulfil its core tasks.
Notes: Texts in bold show the long version of the vignette.Short/long versions and the order of the two issues were randomly assigned.Study 2 deals with voter perceptions of party positions on sociocultural issues.It was accepted as a proposed module of the online panel conducted by the Austrian National Election Study (Aichholzer et al. 2020).The quota sample from an online access panel (n = 3,024) is representative of the Austrian population concerning age, gender, education, household size, and region.
We ask respondents to evaluate a party's position on the sociocultural dimension based on two policy issues: LGBT rights (same-sex marriage) and openness towards refugees.The vignettes are displayed in Box 2. The party's statement on LGBT rights is progressive, while the statement on refugees is conservative.As in Study 1, we let the length of both texts vary randomly across the respondents. 6 We ask respondents to assess the party's position expressed in the vignette based on an 11-point scale from 'Openness and tolerance' (0) to 'Traditional values/law and order' (10). 7ext, we ask respondents to assess the same party's policy position on both individual issues (LGBT rights and openness to refugees) to test the assumption that the statements are indeed progressive (for LGBT rights) and conservative (openness to refugees).These questions were placed in random order to avoid question order effects.The responses are shown in Fig. 2. The mean perceived policy position on LGBT rights is progressive (mean: 3.53), while that for the openness to refugees is conservative (mean: 7.34).

Results
Figures 3 and 4 show the mean perceived positions for the two higher-level dimensions depending on the emphasis on the two underlying policy issues.In both experiments, we find empirical evidence for our hypothesis: Respondents tend to place more weight on the issue position that is emphasized more.In Study 1 (Fig. 3), the mean perceived policy position ranges from 3.95 to 4.77, and the perception is either to the left or right depending on whether the emphasis is placed on the party's left-leaning or right-leaning issue position.The difference in the mean perceived position of 0.8 scale points is also substantially meaningful for perceptions measured on an 11-point scale (for comparison, the standard deviation of the outcome variable is 2.32).
We find similar effects for the perceived position on sociocultural issues (Fig. 4).The higher the emphasis on the progressive (conservative) issue position, the more the mean perceived policy position moves to the progressive (conservative) pole.The difference in the mean perceived policy position (1.1 scale points) is also substantially meaningful (SD: 2.33).

Box 2. Vignettes in study 2
[First/Second], same-sex couples need more rights.Whether gay, lesbian, bisexual, transgender or transsexual: This should not be an issue in today's world.Therefore, an extension of legal protection against discrimination is needed to ensure the equal treatment of all people in all areas of life.Because it must not matter who you love and kiss when you are looking for an apartment or going for a drink in a bar.
[First/Second], the number of refugees must be significantly reduced.We need an upper limit that makes it clear: We can help refugees up to a certain point, but not beyond that.Those who come must also follow the rules, learn German, and accept our values.Anyone who thinks they can abuse our prosperity and security to build an unfree, unjust and unequal parallel society must leave.
Notes: Texts in bold show the long version of the vignette.Short/long texts and the order of the two issues were randomly assigned.

6
In this study, we also aimed to test the effect of salience effects for the economy as in Study 1. Yet, we failed to choose issues and vignettes that respondents perceived as economically left and right, respectively.With moderate mean perceived positions on both issues, changes in issue salience had no effect on the (mean) perceived position on economic policies.Results are shown in Appendix G.
Next, we aim to distinguish whether the treatment had a direct effect on the perceived policy position or whether the text length only changed the perceived party position on the two underlying policy issues (for example, tax cuts).If our argument is correct, then we should observe a strong direct effect of text length.The treatment should have no indirect effect because perceptions of the party's position on specific issues such as tax cuts should not vary across treatments.8Results using perceived issue positions on each scale as the outcome variable suggest that these vary across treatments, so this alternative pathway is potentially plausible (for more information, see Tables D.1 and D.2 in Appendix D). 9 On economic issues (Study I), we find that longer statements are perceived as more extreme.In Study II, we also observe differences in the perceived issue positions (see Table D.4 in Appendix D), but these differences are smaller than in Study I and are not systematically related to text length (that is, longer statements are not necessarily perceived to be more extreme).
To test whether ideological differences emerge due to perceived issue positions rather than changes in issue emphasis, we turn to causal mediation models (Hicks and Tingley 2011;Imai, Keele, and Tingley 2010a;Imai, Keele, and Yamamoto 2010b).The results are summarized in Figs 5 and 6.10Both figures show the effect of each treatment (that is, text length) vis-à-vis the reference category 'short texts' using the perceived position of the two issues as potential mediating variables.The direct effect is the pure effect of text length on the perceived policy position.The indirect effect is a treatment effect that is due to differences in the perceived position on the underlying policy issues.
On economic issues (Fig. 5), talking more about the (left-leaning) stance on pensions pushed the perceived issue position further to the left.The mediation model suggests that this effect is largely due to the direct effect of text length (left panel) and not mediated by the perceived position of the statement on pensions or taxes.We do not find a similar effect for longer statements on the (right-leaning) position on taxes.While the effect is positive (relative to the baseline  Appendix Tables D.1 and D.2 also examine the variance of perceptions, testing whether longer vignettes decrease uncertainty about party positions.This could be due to information effects, also concerning other issue dimensions implicitly addressed in the longer statements.If this alternative explanation holds, the variance of perceived positions should be lower for longer statements.However, this is only the case for one of the four issues (taxes).We also find no evidence that longer statements (e.g. on taxes) help respondents to place parties more accurately on higher-level dimensions (e.g. the economy).These findings are not in line with the expectation that longer texts decrease uncertainty about party positions.See also Appendix F for more discussion of this alternative explanation.D.4).The direct effect of emphasis change is identical in both mediation analyses.
category with two short vignettes), the magnitude of the effect is small and it clearly misses conventional levels of statistical significance.
Fig. 6 shows the results of the causal mediation model for the perceived position on sociocultural issues.Increasing the length of the right-leaning position on immigration moves the perceived position on sociocultural issues more to the right.In turn, highlighting the (left-leaning) position on LGBT rights shifts the perceived position more to the left.Importantly, these effects can mostly be attributed to the direct effect of text length (left panel).The estimates for the indirect effects mediated by the perceived position on refugees and LGBT rights are close to zero.

Conclusion
When voters place parties on broader ideological scales, they make use of information on the more specific issues that underly those scales.In this paper, we show that respondents use issuespecific information more if the parties emphasize a particular issue more.In our experiment we varied party emphasis on two issues.The issue that was emphasized more was seen as providing more information on the party's general position on a higher-level dimension.Importantly, while the length of the issue statement did influence perceptions of positions on the specific issues themselves, controlling for these effects did little to change our findings.
This result has important implications for understanding party competition.It provides robust causal evidence that parties can move ideologically without changing any issue positions.This may seem paradoxical but results directly from the fact that higher-level dimensional perceptions are the outcome of an averaging process across lower-level issue stances.This insight provides a clearer understanding of the positional strategies of political parties.It also highlights the importance of issue salience in the arsenal of strategic tools available to parties because it is much easier to dial up or down the emphasis on an issue than to change positions on it.
In addition, our findings show that the political context can shift parties even if they do not change their rhetorical strategies.If other parties or the media can successfully affect the salience of certain issues (Green-Pedersen 2019; Meijers and van der Veer 2019), they move rival parties to the left or right simply by altering the issues being discussed.Hence, the nature and structure of a party system may have an impact on how voters position parties.Of course, salience may change simply due to the changing nature of the public agenda; for example, in the cases of the financial and economic crisis, the refugee 'crisis', or the Coronavirus pandemic.Issue change may also occur more gradually when new topics enter the political arena.While these additional expectations directly emerged from our experiments, our test focused on how parties themselves shift perceptions.Future work should turn to a direct test on how political context factors, specifically aspects of the party system and media discourse, moderate the processes we uncover (Carroll and Kubo 2018).
Our findings also hold implications for understanding how voters perceive the ideologies of political parties.In our tightly controlled experimental setting, our respondents proved adept at translating concrete issue statements into aggregate dimensional positions.This increases confidence in models of politics that depart from specific issue stances and their implications for party support.While we would not claim that voters are highly sophisticated calculators of dimensional positions, our experiments show that respondents are capable of, and quickly employ, the kinds of reasoning assumed by some spatial models.
In addition, future research should try to assess two competing visions of party ideology.The first sees parties' broad ideological positions as paramount, with perceptions of issue stances flowing from these perceived positions.The second sees specific issue stances as foundational, with broader dimensional perceptions constructed from these perceived stances.This second account is plausible, as our experiment has shown.However, we do not know how often respondents think in these ways or which direction of influence is stronger.

Figure 1 .
Figure 1.Perceived policy positions on pensions (left) and tax reform (right).Notes: Respondents' perceptions of the two underlying issues pensions (left panel) and tax reform (right panel) on 11-point scales (N = 3,150).

Figure 2 .
Figure 2. Perceived policy positions on LGBT rights (left) and refugees (right).Notes: Respondents' perceptions of the two underlying issues LGBT right (left panel) and refugees (right panel) on 11-point scales (N = 3,024).

Figure 3 .
Figure 3. Emphasis effects on the perceived economic policy position.Note: x-axis shows the mean perceived position on an 11-point scale (0: 'economically left'; 10: 'economically right').Bars denote 95 per cent confidence intervals.

Figure 4 .
Figure 4. Emphasis effects on the perceived socio-cultural policy position.Note: x-axis shows the mean perceived position on an 11-point scale (0: 'openness and tolerance'; 10: 'traditional values/law and order').Bars denote 95 per cent confidence intervals. 9

Figure 5 .
Figure 5. Direct and indirect effects of text length on the perceived party position on economic issues.Notes: Full results of the mediation analyses are shown in Appendix D (TableD.3).The direct effect of emphasis change is identical in both mediation analyses.

Figure 6 .
Figure 6.Direct and indirect effects of text length on the perceived party position on socio-cultural issues.Notes: Full results of the mediation analyses are shown in Appendix D (TableD.4).The direct effect of emphasis change is identical in both mediation analyses.