Party Discourse as a Conditioning Factor

When including parties in economic voting studies, there are three main objects of study: (1) the type of economic discourse and campaign that will most likely be followed by opposition and incumbent parties,Footnote ²² (2) the conditioning role of party positionsFootnote ²³ and (3) third-party competence signals.Footnote ²⁴ These three characteristics have been found to be significant in activating the link between economic factors and the vote. The effect of party discourse on public economic perceptions, however, is still a prominent gap in the literature.

Figure 1 shows what would be a traditional expectation of political conditioning of the economy. In a perfect setting, we would expect citizens’ perceptions to match the cues given by the real economy; however, we know that this is not the case. As Parker-StephanFootnote ²⁵ has found, for instance, partisan attachments alter this linear relationship and make government partisans more optimistic, and opponents more pessimistic. However, this partisan effect only exists in a middle-ranged zone, since in ‘too good’ or ‘too bad’ economic climates, citizens converge in their perceptions.Footnote ²⁶

Fig. 1

Political conditioning of the economy

Following this line of argument, we argue that party discourse can effectively condition citizens’ economic perceptions in this middle zone. However, instead of discourse having the capability to affect these perceptions in a positive and/or negative way (as depicted in Figure 2a), we argue that it can only affect them negatively (as Figure 2b shows). Specifically, we reach this argument by merging the theory of issue ownershipFootnote ²⁷ with that of negativity bias.Footnote ²⁸

Fig. 2

Political conditioning by party discourse

Issue ownership theory is an influential framework in economic voting studies, which allows us to derive observable implications regarding the effect of party discourse. It predicts that voters will evaluate parties or candidates on important issues, and that some issues will naturally advantage or disadvantage one party over the others.Footnote ²⁹ In issue ownership terms, the party should prime an issue that it owns. The theoretical presumption is that when a party owns an issue, it is perceived by the electorate as being better and more credible at handling problems in that issue area.Footnote ³⁰ For instance, in the United States, Democrats are considered to own the issue of health care, and Republicans are believed to own the issue of national security.Footnote ³¹ Similarly, in Western European countries, socialist parties are perceived to own welfare politics, while right-wing parties are believed to own the issues of immigration and economic growth, and green parties are believed to own environmental issues.Footnote ³² When a party owns an issue, and public opinion considers that issue important, the theory predicts that the party’s candidates should benefit electorally. A more refined interpretation of issue ownership has led researchers to consider some specific economic topics as likely to benefit the electoral prospects of certain types of parties. From this perspective, inflation is likely to benefit center-right parties, whereas unemployment is likely to connect with the issue priorities and competence reputation of center-left parties.Footnote ³³ Recent comparative economic voting studies, however, have found mixed evidence for this refined version of issue ownership applied to the study of the economy and the vote.Footnote ³⁴

Owning the economy (or having a strong political reputation of handling it effectively) becomes crucial according to this theoretical framework. The issue owner is thus expected to have more economic credibility and therefore to be able to condition public opinion more clearly. This expectation implies a direct effect of the issue owner’s discourse beyond (and independently of) the real economy. The idea that parties’ economic discourse is a sole function of the macroeconomy would imply that the former is in some sense a mirror image of the latter. If incumbents are to influence public opinion when the economy is performing well, and if challengers have to do so when the economy performs badly, the true driver of the effect is the macroeconomy; thus party discourse has very little leverage to shape public perceptions. If party discourse has any substantive relevance in the process of the political conditioning of economic perceptions, parties with a better reputation of handling economic issues should be able to persuade voters beyond how things are really going.

There are reasons, however, to suspect that the credibility of the issue owner’s messages will differ depending on whether it is in or out of power. In her study about how voters evaluate economic competence in Britain, ButtFootnote ³⁵ shows that retrospective evaluations of economic performance influence evaluations of the incumbent, but not the challenger. While citizens can rely on an up-to-date performance record to judge the incumbent (through the real macroeconomy, for example), judgements about the discourse and hypothetical performance of the challenger can be more responsive to ideological predispositions and partisan cues.Footnote ³⁶ The experience of living under the current government is certain and readily available, and therefore the discourse of the incumbent should have a much more limited power to persuade voters. By contrast, the evaluation of the prospective performance and current discourse of the challenger can be more subject to partisan sources of information and persuasion. While the incumbent has to deal with ‘real events’, the challenger can be judged ‘only by talk’.Footnote ³⁷ This is consistent with De Boef and Kellsted’s findings that US presidents cannot always use the power of the press to maintain high approval ratings, and that while they are in power they ‘cannot use rhetoric to consistently dupe the public’.Footnote ³⁸

The stronger potential for opposition parties to place economic issues on the agenda and condition public perceptions beyond the macroeconomic facts is also supported by research in political psychology, which documents negativity bias in information reception. Negative information has a significantly larger effect on impression formation.Footnote ³⁹ From a political behavior perspective, ClaggettFootnote ⁴⁰ showed that there is an asymmetric relationship between macroeconomic conditions and the response of the congressional electorate. Contractions reduce the vote share of the incumbent party, but it is not rewarded for economic expansions.Footnote ⁴¹ Lau also found evidence that negative information is more influential than positive information in affecting evaluations of presidential candidates.Footnote ⁴² More recently, media studies have provided evidence that negative views are more likely to attract the media’s attention and to be placed on the public agenda.Footnote ⁴³

Based on these predictions, and if party discourse is able to shape public evaluations of economic performance, one can expect that the owner of the economy will be more successful at shaping public perceptions. Moreover, both the constraints of real facts for incumbent parties and the negativity bias framework expect the challenger’s discourse to be more likely to place economic issues on the agenda and to condition public opinion. Table 1 summarizes these predictions.

Table 1

Theoretical Expectations

Our reasoning implies that important partisan sources of political conditioning of economic evaluations may be at play, and that this should be largely independent from the real economy and individual ideological predispositions. However, our expectations acknowledge the role of the real economy and classic accountability, in the sense that incumbent parties’ real performance should be more important than their discourse. Moreover, we only expect party effects among economically credible parties. It is important to stress that our theory does not intend to resolve the controversy between classic and revisionist perspectives. Our attempt is rather to include a substantively important factor that has been overlooked in the literature, and to deepen our understanding of when and why external partisan cues should shape public opinion.

Spain: From Bounty to Crisis

The expectations of different economic voting models have been largely tested in Anglo-American democracies. The reward-punishment logic is a very intuitive model to apply in two-party majoritarian democracies, in which there are usually single-party governments and one clear challenger. We believe that extending the test to a multiparty proportional system like Spain, which is a less-studied context in the economic voting literature, can increase the comparative power of the propositions sketched out above. Spain is also a relevant context in which to test our theoretical propositions because it went from relatively good macroeconomic conditions to being one of the Western democracies most punished by the global economic meltdown. This is thus a conservative test for party discourse, given the presumably important effects of macroeconomic conditions.Footnote ⁴⁴

The time span analyzed covers four electoral cycles. The two first cycles (March 1996–March 2000 and April 2000–March 2004) had a right-wing single-party government, and the two last cycles (April 2004–March 2008 and April 2008–November 2011) had a left-wing single-party government. This allows us to vary the ideological orientation of incumbent parties. We focus on the two main parties competing in general elections in Spain: the left-wing socialist PSOE and the right-wing conservative PP. Despite the presence of many other parties in the Spanish political spectrum, the PP and PSOE were able to govern two electoral cycles each without the need for a coalition partner. While acknowledging the role of other parties in the Spanish national parliament, we keep the distinction between one main incumbent and one main challenging party that is inherent in the theory described above.

The PP has a stronger reputation for dealing with economic issues. Apart from an intuitive link between economic priorities and economic issues in Western Europe,Footnote ⁴⁵ the link between economic competence and right-wing parties is particularly likely in Spain given the country’s bad macroeconomic performance during its democratic stabilization (during the 1980s and early 1990s) under socialist rule. Furthermore, the PP is – according to survey data such as the European Election Studies (EES) – the party most able to handle the issue of the economy during the time span analyzed.Footnote ⁴⁶ Previous research on economic voting in the Spanish context shows that the link between macroeconomic conditions and the PSOE’s electoral results is non-existent.Footnote ⁴⁷ This finding confirms the lack of a connection between economic priorities, a reputation of economic competence and the PSOE’s electoral prospects. Rather than economic considerations, an electoral coalition of primarily blue-collar workers, rural voters and voters dependent on state-subsidized income explains the historical disconnection between the PSOE’s electoral victories and economic fluctuations.Footnote ⁴⁸

Figure 3 shows the evolution of the unemployment rate (solid line) and the rate of change in the consumer price index (dashed line),Footnote ⁴⁹ which are the indicators used in the analyses below.Footnote ⁵⁰ Unfortunately, data on GDP is only collected every three months, which considerably decreases the number of timepoints available in the analysis. Thus we only report the results concerning unemployment and inflation, for which we have monthly data.Footnote ⁵¹ The graph shows three vertical lines dividing each of the four governments in our study. There was a clear improvement in unemployment rates when the right-wing PP was in power (1996–2004), but there was a gradual and steady increase in inflation, which can be seen in the mostly positive change over time. In spite of the general feeling that the PP is economically competent – reflected by the famous slogan ‘España va bien’ (‘Spain is going well’) – the PSOE could have used the increase in prices exacerbated by the introduction of the euro in Spain (around the middle of the 1996–2004 period) and the maximal inflation rate value observed during the entire study period to call into question the PP’s handling of the economy. Even if, on average, the PSOE’s incumbency clearly experienced worse macroeconomic performance (especially due to the global economic meltdown in the 2008–11 period), there was still some macroeconomic variation that the challenger’s discourse could have used to its benefit between 1996 and 2004. Similarly, not everything was dramatically good or bad when the PSOE was in office. Even if unemployment and inflation rates both increased to maximal levels at the end of the time span analyzed, unemployment rates went down during the PSOE’s first term (2004–08). Thus the macroeconomic variation seems to be sufficient during each incumbency period to test the effect of party discourse not as a mirror image of macroeconomic variation, but as an independent effect beyond the economic reality.

Fig. 3

Evolution of macroeconomic indicators over time

Party Questioning of Government

In order to measure the issue attention of the two major Spanish parties, we rely on the questions asked by party members in the plenary sessions of the Spanish Parliament. Most ‘plenos’ reserve time for members of parliament (MPs) to question representatives of the government.

Political parties’ attention to issues has received increasing attention in the literature. Most studies have focused on the salience of issues to political parties ahead of elections, by using party manifestos or expert surveys.Footnote ⁵² While these studies have advanced our knowledge of parties as strategic actors in the issue domain, they have shed little light on actual electoral dynamics and how they affect issue politics (and vice versa) – that is, the degree to which parties highlight more or less different issues for short spans of time. Thus plenary questions constitute an ideal source for measuring attention during the electoral cycle.

Party debates in Congress represent a unique source of data that has proved to be useful for many analyses of legislative behavior. The assumption is that parties’ attention to different issues in parliament is a good proxy with which to measure the issue hierarchy in the parties’ general communication strategy. In the same way that manifestos are a source of data to measure issue emphasis (even if most citizens do not read manifestos), research using parliamentary questions and speeches assumes that parties’ communication strategies will reach public opinion through the media and the public debate. Exchanges between legislators in plenary sessions have been used to analyze their ideological positioning,Footnote ⁵³ or to understand the dynamics of who speaks and who does not.Footnote ⁵⁴ In the Danish and Belgian cases, debates in the floor of parliament have allowed the study of who sets the agenda.Footnote ⁵⁵ These two studies, for instance, find that MPs focus mostly on their owned issues and on the issues raised by the media. As can be seen from this short sample, speeches from the parliamentary floor are a relevant source of the dynamics of parliament, but also – and more relevant to this study – of the issue priorities of MPs and their parties.

Table 2 shows some of these interactions in the Spanish Parliament. Specifically, it shows questions from a member of the governing PP party and from the opposition PSOE in a pleno from November 2003. Similarly, it shows questions from the governing PSOE and the opposition PP in December 2004. While all four questions refer to the issue of the economy, the tone with which these questions are asked makes it clear which party is in opposition. The harsher, more negative, questions – or framing – always come from the opposing party, while the subtle – more nuanced – questions come from the government side. The tone of these questions supports the underlying theme of this study. Because the opposing party will always be negative regarding the economy – as it has no advantage to do otherwise – when it draws attention to this issue it should affect citizens’ view of the topic.

Table 2

Examples of Questions to the Government

Note: Questions translated by the authors.

We obtained the plenary debates from the publicly available data of the Spanish Parliament. From these, we selected all the plenary sessions that allowed MPs to question the government. Table 3 shows the total number of plenary sessions at which questions were asked; the number of questions asked, on average, in each session by the two main parties; and the total number of questions asked per legislature, by party.

Table 3

Average Number of Questions Qsked in Plenos and Total per Legislature

It is clear that the PP and PSOE dominate the debates in the plenary sessions. It is interesting to note that regardless of their role, as government or opposition parties, they behave quite similarly. On average, both parties have a similar number of interactions per plenary, which is almost four times more than the rest of the parties combined. It is interesting to note that in the last legislature (2008–11) there is a big drop in both the number of sessions with questions and in the number of questions (especially from the PSOE). However, we see no need for concern over this. We acknowledge that as the crisis worsened, the PSOE implemented its agenda control mechanisms to minimize its exposure; it diminished the number of sessions with questioning and limited the interventions of its own members. Party questions in parliament therefore provide a detailed account of the issue priorities of political parties.

To analyze the questions and obtain issue attention measures, we use a similar quantitative text analysis approach to that of Sagarzazu (Reference Sagarzazu2011), which applies cluster analysis to textual data. This approach is the ‘Bag of Words’ family of text analysis mechanisms that considers words as independent observations.Footnote ⁵⁶ The underlying assumption behind using clustering analysis to obtain issue measures is that words that belong to the same issue have similar patterns of behavior, as opposed to words that belong to different issue areas, which can be identified via clustering techniques.Footnote ⁵⁷ Specifically, we estimate the distances between the correlations of different words and cluster the words based on these distances.

In order to be able to apply quantitative text analysis techniques to a raw text corpus, a series of preparatory steps is required. First, all questions need to be converted into plain text files, all HTML code tags need to be removed and all words need to be changed to lowercase. Secondly, stop words, punctuations, numbers, currencies and party labels are eliminated from the documents. Thirdly, a stemmer was used to convert words to their stemmed root.Footnote ⁵⁸ Fourthly, with the dataset reduced to stem counts, we computed matrices that indicate how many times each stem occurs in each press release. Finally, following common practice in the literature, we removed very frequent words occurring in more than 95 per cent of the questions and very rare words occurring in less than 5 per cent of the texts.Footnote ⁵⁹

Having generated the initial word count matrix, we begin an iterative process in which we cluster words based on the distances between them. In each step we multiply the word matrix by its transposed matrix to calculate a word-to-word matrix in which each cell indicates the number of times two words are mentioned together. In order to correct for overdispersion – caused by too-frequent and too-infrequent joint mentions – we normalize this matrix by means of the squared root of each cell. We then estimate the distances between the word correlations and use this as the measure of closeness between different words needed for the clustering analysis.Footnote ⁶⁰ This measure allows the clustering algorithm to establish groups based on the proximity of the elements.Footnote ⁶¹ The final stage of the process involves estimating the different word clusters. Specifically, we use a hierarchical clustering algorithmFootnote ⁶² that infers relationships from the data themselves.Footnote ⁶³ Clustering techniques are mathematical tools that allow the classification of data according to latent characteristics.Footnote ⁶⁴ The result of applying the clustering algorithm to the distances between the words is a tree-like structure in which words are organized by how close or distant they are between them.Footnote ⁶⁵ In each iteration, the tree was automatically divided into clusters; we tried with varying sizes from five to twenty clusters. Based on this division, we assessed the organization of the words and clusters, and in each iteration we removed non-political words that were not members of any particular cluster but that instead (because they are mentioned in different issues irrespectively) are close to words from many clusters, and as such tend to remain high in the tree structure – and alone.Footnote ⁶⁶ By doing this analysis, we were able to obtain a clear structure of words and clusters based on the co-occurrence of words. This structure was remarkably consistent throughout the different iterations. After four iterations we were able to identify seven politically relevant clusters:Footnote ⁶⁷ economy, education, health, infrastructure, international relations, gender and state matters. The main words in each cluster are shown in Table 4.

Table 4

Top Ten Key Terms per Issue Area

In addition to checking the face validity of the words and clusters, we looked for external validity. Given that we are looking at speeches in the floor of parliament, it can be assumed that the questions will resemble the structure of the parliamentary committees (which is similar to the structure of cabinet portfolios). As such, we checked the seven issue areas with the seventeen permanent legislative commissions. Given that we have fewer issues than committees, we will expect to have several committees grouped into one of our issue areas. Table 5 shows this grouping; the resemblance of the committee structure and the issues identified is quite similar. Most of the committees fall into a single issue area, with the exception of the Taxes and Public Administrations Committee, which deals with economic aspects as well as state matters. The only committee we could not match to only one issue area was the Budget/Appropriations Committee, given that budget discussions tend to be involved in every policy area.Footnote ⁶⁸ As can be seen both from an internal standpoint (face validity) and an external validity check, our issue clusters provide a reasonable classification of parliamentary questions into relevant policy issues.

Table 5

Key Terms in Economic Area

As we are particularly interested in the attention that parties give to the economy, we will focus solely on this issue. In order to illustrate how issue attention varies over the course of the four legislative terms, Figure 4 plots the attention that the PSOE and the PP paid to the economy as a proportion of the total time the party spoke. This figure shows two interesting dynamics that occur in the four legislatures. First, a similar attention dynamic can be found in the VI ^th and VIII ^th legislatures; in these two sections (the first and third) we can see no major differences between the parties. These legislatures are similar not only in the attention that parties give to the economy but also in the fact that they were the result of a change in control of parliament and end in a re-election of the governing party. The second dynamic can be seen in the second and fourth sections, in which the opposition party paid more attention to the economy (PSOE in the second – VII ^th legislature; PP in the last – IX ^th legislature). While the dominance of the PSOE is clearer in the VII ^th legislature than the PP’s dominance in the IX ^th legislature, the overall pattern is there throughout. Coincidentally, the issue attention of each party overrides that of the rival party when both the PP and PSOE are in opposition, and in the period prior to their access to power. This descriptive evidence is consistent with our intuition that parties have more incentives to stress economic issues and to have more potential persuasion power when in opposition than when there is a macroeconomic reality that speaks for their performance as incumbents.

Fig. 4

Issue attention over time

Spanish Monthly Survey

For each month or timepoint with available party discourse data, we have matched monthly survey data conducted by the Centro de Investigaciones Sociologicas (CIS) in Spain.Footnote ⁶⁹ The CIS conducts monthly barometers together with many other types of surveys in Spain.Footnote ⁷⁰ Each of the monthly barometers contains a random sample of the Spanish population with valid answers to our dependent variable (n=2,376–4,874). The pooled dataset that we obtain after matching all the monthly surveys for which we have data on party discourse contains 320,793 individuals across 127 timepoints.

The dependent variable that is available in each of the surveys used in the models below is the assessment of the current national state of the economy.Footnote ⁷¹ The economic voting literature tends to rely on retrospective evaluations of the economy, which are unfortunately available in only fifty-eight surveys during the time period analyzed. The wording of the item is: ‘do you think that the current economic situation is better, the same, or worse than a year ago?’ Since we have more timepoints available for the item capturing current national conditions, we use this in the models reported below. In any case, the correlation between the current and the retrospective evaluations is moderately high (r=0.42, p=0.000). When running a factor and a principal components analysis, both items clearly load on the same dimension, with factor loadings of 0.54 and 0.71, respectively. When replicating the fully specified models shown below with retrospective evaluations as the dependent variable, our substantive results remain unchanged.Footnote ⁷² Figure 5 plots the aggregate responses to the subjective national economy question.

Fig. 5

Evolution of subjective economic evaluations over time Note: The solid line represents the percentage of respondents that answers that the economy is good or very good; the dashed line represents neither good nor bad; the dotted line indicates bad or very bad.

We use two indicators to control for ideological predispositions. First, we use the standard left-right scale from 1 to 10, which is well known for summarizing a number of ideological predispositions and for being a good vote choice predictor in Western Europe.Footnote ⁷³ Unfortunately, there is no indicator available for the whole time span that we are analyzing on party identification, as there is in American politics, for instance. Alternatively, we use vote recall in the last national election.Footnote ⁷⁴ Finally, we also include a set of control variables available in the CIS surveys including age, gender, educational level and professional status (being unemployed).Footnote ⁷⁵

Our modeling strategy conceptualizes our dataset as a two-level structure in which different individuals (Level 1) are nested in different months or timepoints (Level 2). In order to obtain accurate standard errors and non-inflated levels of significance for our time-level variables (namely party discourse and macroeconomic indicators), we predict individual economic assessments using hierarchical ordinal logit models.Footnote ⁷⁶ We choose this modeling strategy because our outcome and some of the important determinants of economic perceptions identified in the literature (like partisan predispositions and all of our control variables) are measured at the individual level, while our hypothesized party effects and the macroeconomy are time-level indicators. This also allows us to use the full statistical power of our dataset. Our time or Level 2 equation, however, has a time-series structure, since monthly averages are regressed over time. To make sure that our main results are not biased by temporal autocorrelation, we conducted several tests and included the lagged monthly average of our dependent variable as a predictor. Our substantive results remained unchanged (see robustness checks and the Appendix). All the variables in our models have been centered around their mean in order to obtain meaningful intercept variances. It is important to note that the validity of our strategy relies on the assumption that our time-random effects are not correlated with any of the covariates included in the model. Our statistical model can be formalized as follows:

(1) $$Logit(Y_{{it}} )=\alpha {\plus}\beta _{1} X_{{it}} {\plus}\beta _{2} Z_{{t{\minus}1}} {\plus}\beta _{3} E_{{t{\minus}1}} {\plus}\Theta V_{i} {\plus}u_{t} {\plus}{\varepsilon}_{{it}}, $$

where

Y _it is the level of economic evaluation for a given individual i at time t

α is the cut-off point of economic evaluations when all other covariates are 0

X _it is individual-level ideological predispositions (left-right and vote choice at time t)

Z _t−1 is lagged party discourse

E _t−1 is lagged macroeconomic indicators (unemployment and inflation)

b _k are the fixed parameters for party discourse, macroeconomic condition and ideological predispositions

u _t is the random intercept at the time level

ε _it is the prediction error of each observation

Θ is the vector of fixed parameters for the individual-level control variables (V _i ).