European party systems have experienced crucial changes over the last few decades. New political forces have arisen in many countries channeling political discontent and grievances directed against political parties and/or political elites. The emergence and success of these new political forces have been facilitated by economic performance failures, lack of responsiveness, and perceptions of political corruption (Hawkins and Rovira Kaltwasser 2018; Andreadis et al. 2018). New political forces have emphasized the mismanagement and unresponsiveness of political elites, and on many instances they have also appealed to voters by using a populist vocabulary that emphasizes the contrast between the people and self-serving elites. This portrayal has questioned the contemporary relevance of the divisions that historically structured mass politics in Europe, that is, first and foremost, of the left-right division.

Now, to what extent has the Great Recession altered patterns of ideological spatial competition in Europe? And more specifically, do political and attitudinal changes connected to the Great Recession affect party positions within the left-right ideological space? To examine the effects of these socioeconomic and political transformations we analyze spatial party competition in three European countries -Spain, France, and Italy - that share intense and long-lasting left-right divides and which have experienced different degrees of economic duress during the Great Recession (very high in Spain and Italy and comparatively lower in France). In all these countries new and important political forces have emerged at the national level (Movimento 5 Stelle in Italy, En Marche! and La France Insoumise in France, and Podemos and Ciudadanos in Spain). Some of these forces have explicitly articulated a populist message emphasizing the opposition between the people and a self-serving and corrupt political elite. The fact that for these three countries there are surveys available that allow us to examine key variables in this exploration (such as ideological self-placements, perceptions of ideological party locations, economic evaluations, and populist attitudes) further justifies the interest of our focus in these three cases.

Our article focuses on the analysis of the ideological positions of parties in Italy, Spain, and France. It compares actual ideological positions (as perceived by all voters) to the ideal or optimal ideological party positions predicted by spatial competition models based on proximity and directional voting, always on the assumption that parties will choose those positions that allow them to maximize their vote shares.

We are interested first in identifying the attitudinal factors that condition vote choices and that thus affect party optimal ideological positions in these three countries. In the second place, we are interested in examining party deviations from the optimal ideological locations estimated on the basis of spatial models. In this analysis, we pay special attention to the effects of attitudinal variables directly connected to the Great Recession. There are some previous works on spatial competition and ideological positions in these countries. Adams, Merrill, and Grofman (2005) and Curini (2015) have conducted studies on optimal ideological positions in France and Italy. Queralt (2012) has contributed a spatial analysis of voting in Spain that includes both proximity and directional models, but his work does not identify ideal ideological positions. None of these works address the effects of government evaluations and political attitudes on ideal ideological positions.

Different analyses have shown that both populist attitudes and economic evaluations have affected vote choices in West European countries (Fraile and Lewis-Beck 2014; Hernández and Kriesi 2016; van Hauwert and van Kessel 2017; Lavezzolo and Ramiro 2018; Andreadis et al. 2018). We lack information, however, about the degree to which such attitudes have exerted any influence on party ideological positions. Does the inclusion of such variables in our models affect estimates of ideal positions? Are there cross-national differences in the effects of such variables on ideal ideological positions? And are these differences connected to the duress of the economic crisis, as in the case of the effects of economic evaluations (Hernández and Kriesi 2016)? Do estimates become more or less realistic after the inclusion of such variables? And finally, can we identify some partisan similarities (ideological, attitudinal, or organizational) among the parties displaying higher differences between ideal and real positions?

In order to conduct the analysis we followed the unified model of party competition developed by Adams, Merrill and Grofman (2005, 19-23), which predicts ideal party positions according to Nash (that is, the set of positions from which no vote-maximizing party has incentives to deviate unilaterally (Adams, Merrill and Grofman 2005, 40)). In that situation, no party would improve its electoral share by unilaterally modifying its position. Our models take into consideration both policy (ideological) preferences and non-policy characteristics and attitudes (including here party identification). We first estimate conditional logit models explaining vote choices (using both quadratic proximity and direction as ideological predictors) and then, following Adams, Merrill and Grofman (2005), we estimate the Nash equilibria for party positions using the parameters provided by our analysis. We derive Nash equilibria by using the iterative algorithm developed by Merrill and Adams (2001), as implemented in the nopp R package (Nash Optimal Party Positions) developed by Curini and Iacus (2017). The original package only deals with the proximity model, therefore we have developed a new R-project to implement the directional model. For this analysis we have used, for the Italian data, the 2013 ITANES post-electoral survey with 1208 respondents.1 The French and Spanish data were collected online in December 2015 (Spain) and April 2017 (France) by the company Netquest. Respondents were selected using a quota sampling in order to avoid biases in the distribution of gender, age, and territorial location.2 The final sample was 1208 respondents in Spain and 803 respondents in France.

In the next section of this article we present the theoretical foundations of our analysis of ideal positions and our main empirical expectations regarding these cases. We present then our proximity and directional vote-choice analyses for these countries and the Nash equilibria that we infer from them. Our analysis pays special attention to the differences between ideal and real positions and to the effects that crisis-related attitudes have on the estimation of ideal positions in these three countries. We conclude with a summary of our main empirical findings.

Our analysis of ideological party positions follows the path opened by Downs spatial model of party competition, which asserted that political parties will adopt ideological platforms in order to maximize their vote shares (Downs 1957, 96 and the following). Downs (1957, 115) also assumed that voters would choose parties based on ideological proximity. By contrast, the directional model developed by Rabinowitz and Macdonald 1989 hypothesized that individuals would choose parties on the basis of the intensity with which they advocate a particular policy direction. Spatial analyses of voting and party competition have continued to use both proximity and directional models of voting (Blais et al. 2001; Maddens and Hajnal 2001; Cho and Endersby 2003; Lacy and Paolino 2010; Meyer and Müller 2014; Henning, Hinich, and Shikano 2015; Kropko and Banda 2018) without reaching conclusive evidence favoring one of these models over the other (Lewis and King 1999). In fact, some analyses have shifted their focus to the study of the characteristics (individual, partisan and systemic) that favor the prevalence of either of these two models of voting (Maddens and Hajnal 2001; Kropko and Banda 2018). Given the continuing presence of these two models in spatial analyses, our work uses both of them in order to estimate ideal party positions in these three countries: Spain, Italy and France.

Different studies have shown the two-dimensional character of European public opinions. Kitschelt (1994, 20-33) and Kitschelt with McGann (1995, 4-19) revealed the presence of an economic (socialism versus capitalism) and political-cultural (libertarianism versus authoritarianism) dimension structuring public opinions in Western Europe. Hooghe, Marks and Wilson 2002 also showed that, in addition to the classical, economic left-right dimension, a new politics GAL/TAN dimension (green, alternative, libertarian versus traditional, authoritarian, nationalist) was crucial to account for party positions regarding the European Union. More recently, but also before the Great Recession, Kriesi et al. (2008, 4-20) showed that globalization processes led to the formation of a two-dimensional policy space in six European countries (France, the UK, Switzerland, Austria, the Netherlands, and Germany) pitting demarcation and integration in two domains, the economic and the cultural. Finally, Beramendi et al. (2015, 14-23) have hypothesized the existence of a two-dimensional public opinion space pitting statism versus promarket orientations and universalist versus particularist preferences. In turn, these public preferences would correspond to the main public policy alternatives citizens face (state versus market, and consumption versus investment policies). The first dimension stressed by Beramendi et al. (2015) equals classical left-right economic divides, whereas the second one shares important features with the political-cultural dimensions identified by previous comparative analyses.

In spite of the conceptual disputes over the character of these two dimensions (in particular, the political-cultural one), these different contributions have revealed the presence of two critical dimensions of conflict in European countries. If this is the case, what are the grounds for maintaining an analysis focused on a single, classical dimension (left-right) of party competition? One crucial reason for continuing with such an analysis is that even if the space of public preferences is clearly two-dimensional, “party systems tend to organize along a single dimension whose ends correspond to the left-libertarian and right-authoritarian poles”, as the Kurella and Rosset (2017, 16) study of party systems in England, France, Germany, the Netherlands and Switzerland has shown. Shikano 2008 has already revealed some of the possible factors (differences in issue salience and/or non-separability of dimensions) leading to this single-dimensional party competition structure. Whatever the reasons, the fact is that this single left-libertarian/right-authoritarian super-dimension articulates and reflects economic and political-cultural divisions, as in Kitschelt’s 1994 and 1995 classical analyses on European social democracy and radical right, respectively. It is also for this reason that left-right positions continue to be used as indicators of party positions in substantive dimensions by authors that have shown the emergence of new substantive policy issues in European politics (Kitschelt and Rehm 2015: 189). Furthermore, as Vries, Hakhverdian, and Lancee’s 2013 study on the evolution of left-right self-placements in the Netherlands has shown, the increasing importance of cultural divides has not led to the demise of the left-right dimension, but to the change of its substantive content, which has become more tied now to attitudes towards immigration than to redistributive preferences. Of course, this single-dimensionality of political competition may lead to significant mismatches and representation gaps between the preferences of voters and the positions of parties, particularly for groups of cross-pressured voters, such as left-authoritarian and right-libertarian citizens (Kurella and Rosset 2017). The presence of congruence inconsistencies in party systems was already emphasized by Kriesi et al. (2008, 326-27).

In the case of these three countries, the empirical analysis of party positions in the GAL/TAN, economic left-right, and general left-right dimensions confirms the single dimensionality of party competition at the supply-side level. The Pearson correlation between general left-right and GAL/TAN party positions is 0.89 in Spain (14 parties), 0.94 in France (13 parties), and 0.90 in Italy (13 parties) (Polk et al. 2017). The Pearson correlations between the general left-right and the economic left-right are 0.99 for Spain, 0.91 for France, and 0.88 for Italy. That is, at the supply-side level, and judging from the evaluations of experts, there is a very strong connection between the positions parties adopt in the left-right dimension and the positions they take in the general cultural GAL/TAN and economic left-right dimensions. Therefore, also in these three countries the position of parties in the left-right dimension allows voters to identify the party programmatic offers in the two main substantive public policy dimensions, that is, the economic state versus market, and economic libertarian versus authoritarian (or GAL/TAN) divisions. Consequently, both the findings of previous comparative works on European party systems and the empirical characteristics of these three party systems justify the adherence to an analysis of spatial competition that uses left-right positions as the best indicator of substantive party policy proposals.3

Historically, the party systems of these three countries were characterized by the importance of left-right divides. We can assume that the substantive content of the left-right dimension will have experienced more important changes in those countries in which GAL/TAN divides have become more prominent in the political arena (particularly in France, in connection to the issue of immigration, to a lesser degree in Italy, at least until 2013, and to an even lesser extent in Spain). Still, as previously indicated, the associations between left-right party positions and GAL/TAN party positions remain extremely high in these three countries.

Although these countries were characterized by diverse political regime trajectories in the past, they also share some important historical traits (such as the key role of left-right divides, the importance of the disputes around the position of the Catholic Church, and the impact of Communist parties in the 20^th century). In addition, these three cases have recently witnessed the emergence of new parties that have become key political players. These countries, however, differ substantially in the extent to which they have been affected by the Great Recession, Spain and Italy having suffered to a greater extent than France during this period. If we compare the pre-crisis peaks and the Great Recession troughs from 2007 to 2014 (Bozio et al. 2015, 409), we observe that per capita GDP dropped by 15.6% in Italy, 9% in Spain, and 4.2% in France. As for the unemployment rate, it rose by 13.7 percent points in Spain, 3.4 points in Italy and 1.6 points in France. That is, in both respects France was the least severely hit country during the Great Recession, whereas Spain suffered the most in terms of unemployment and Italy in terms of per capita wealth (Bozio et al. 2015, 409).

We know that the Great Recession led to the severe punishment of governments in Europe in general and in Western Europe in particular (Hernández and Kriesi 2016, 221). The punishment of governments has extended over time in Western Europe, and it has led to the success of new parties in general and populist parties in particular (Hernández and Kriesi 2016, 221). In fact, many recent empirical contributions have underlined the associations between economic evaluations and conditions, populist attitudes (Spruyt, Keppens, and van Droogenbroeck 2016; Rico, Guinjoan, and Anduiza 2017), and the vote for challenger parties (Bosch and Durán 2017; Lavezollo and Ramiro 2018; Hobolt and Tilley 2016; Andreadis et al. 2018).We do not know however whether the Great Recession has had any substantial effects on spatial ideological competition in Europe. On the one hand, given the complexity of partisan ideological alignments and governmental responsibilities during the crisis we cannot assume that the Great Recession will have led parties to move in similar ideological directions across nations and parties (to the left, to the right, to the extremes, etc.). We know however that economic evaluations and populist attitudes, two variables directly connected to the economic crisis, affect the vote-choices of citizens. And we can assume that vote maximizing parties will take such attitudes into consideration when adopting ideological positions. For instance, according to Schofield (2007), parties with low valence, due in some cases to their governmental responsibilities during the crisis, might choose to adopt extreme ideological positions in order to keep the support of core groups of supporters. By contrast, parties that are popular due to their lack of responsibilities during the crisis, and parties that are in tune with the extension of public populist and anti-elite attitudes might increase their vote shares by moving to centrist positions. It is impossible to summarize the different effects that attitudes related to the crisis may have in different party systems. However, we can assume that, at least in cases such as these, where left-right divisions have played a crucial role in the structuring of party systems, the more intense the economic crisis, the more affected party ideological positions will be by public attitudes directly connected to the crisis. More specifically, we can hypothesize that the inclusion of economic judgments and populist attitudes directed against the political class will make a greater impact on the estimation of ideal party positions in those countries that have been more severely hit by the crisis (Spain and Italy, in this analysis). Furthermore, we can assume that the ideal positions estimated on the basis of models using economic evaluations and populist attitudes as predictors will be closer to real party positions in those countries that have been more severely hit by the crisis, again, in Spain and Italy. That is, we hypothesize that models including economic evaluations and populist attitudes will have stronger implications for left-right positions in countries that have been more severely hit by the recession, and that the inclusion of these variables in our models will make ideal party positions more realistic.

Following Adams, Merrill and Grofman (2005), the first step in our analysis consists in the prediction of vote-choices through a conditional logit model. We have restricted our analysis to the parties for which a sufficient number of respondents was present in our survey - five parties in Spain and France and six in Italy. Specifically, our analysis restricts the sample to the respondents who voted for the selected parties and with data available for all the variables we have considered (631 individuals in Italy, 523 in Spain, and 482 in France). These parties are: Sinistra, Ecologia, Libertà (SEL), Partito Democratico (PD), Movimento 5 Stelle (M5S), Scelta Civica (SC), Lega Nord (LN) and Il Popollo delle Libertà (PDL) in Italy; Unidad Popular-Izquierda Unida (IU), Partido Socialista Obrero Español (PSOE), Podemos (Ps), Ciudadanos (Cs) and Partido Popular (PP) in Spain; and La France Insoumise (FI), Parti Socialiste (PS), En Marche! (EM), Les Républicains (LR), and Front National (FN) in France. In the case of Podemos we have considered as voters for this party all respondents who voted for the alliances in which Podemos participated in Catalonia, Galicia, and Valencia.

We have conducted our conditional logit models using Adams, Merrill and Grofman’s 2005 unified proximity model of voting. We implement the proximity model where the utility is given by the negative of the squared distance between the voter’s and the party’s location in the left-right dimension - scales 0 to 10 in Spain and France and 1 to 10 in Italy, and the directional model of party competition (Rabinowitz and MacDonald 1989), where the utility is given by the product of the difference between the respondent’s position and the status quo (neutral point) and the difference between the mean party position and the status quo, taking in this case the center of the scale as the neutral point.

But we also wanted to control for the effects of other policy and non-policy factors to voter utility. In this respect, a typical voter utility function with both policy and non-policy factors can be expressed as follows (Merrill and Adams 2001; Adams, Merrill and Grofman 2005; and Calvo and Hellwig 2011):

U_i (j) = αV_ij + βp_ij + γ_jZ_i + ϵ_ij (*)

Where U_i (j) represents the utility of voter i for voting party j and ϵ_ij have standard Type 1 extreme value distributions.

The different variables can be grouped into two types:

–Alternative specific variables, which vary with the alternatives: V_ij and p_ij.

p_ij is a vector that describes characteristics of party j related to voter i, each coordinate equals to 1 if i identifies with party j and 0 elsewhere. As Adams, Merrill and Grofman 2005: 37-39 have shown, parties have incentives to present policies distant from the center in the direction of voters leaning towards them for non-policy reasons. Furthermore, party identification acts as a stabilizing element that limits the ideological flexibility of parties and makes the identification of Nash equilibria possible (Kurella and Pappi 2014). Party identification is thus a critical variable for these analyses.4

We have also put to the test the valence model developed by Schofield 2006

U_i (j) = λ_j+ αV_ij + βp_ij + γ_jZ_i + ϵ_ij

which includes a non-policy related valence term λ_j that captures the perceived competence and reputation of party j.5Ideology and party identification coefficients are basically similar to the ones reported here. Again, we obtain unique Nash equilibria in all the cases. For the sake of simplicity, we do not include full statistical results but they are available on demand.6

• V_ij has different expressions depending on the model we are working with:

Proximity model: V_ij = –(x_i – s_j)²

Directional model: V_ij = (x_i – s_q) (s_j – s_q)

where x_i stands for i’s location, s_j for j’s location (mean of party positions) and s_q for the neutral point –status quo-. For this analysis we take as the objective party location the perceived position of the party according to all respondents. We are aware that there is a continuous debate on the advantages and disadvantages of individually perceived positions versus average positions. We use the average positions perceived by all respondents in order to avoid the risk that citizens may project party positions based on their own party preferences, and therefore, that party locations become endogenous to the voters’ preferences (Macdonald, Rabinowitz and Listhaug 1998), even if this bias has been proved to be modest in some analyses (Merrill and Grofman 1997). In fact, average positions happen to be very similar to the evaluations of left-right party positions by experts. The Pearson correlation coefficients between average perceived positions in these surveys and the evaluations of experts (Polk et al. 2017) equal 0.93 for Italy, 0.92 for France, and 0.99 for Spain.

– Individual specific variables which do not vary with the alternatives, Z_i = (Z_i1, Z_i2, Z_i3)^T when we do not include the evaluation of the government performance and populist public attitude; Z_i = (Z_i1, Z_i2, Z_i3, Z_i4, Z_i5)^T if we do, and in the mentioned special case of Spain in which we consider the regional identification, Z_i = (Z_i1, Z_i2, Z_i3, Z_i4, Z_i5, Z_i6)^T. Table 1 reports the individual specific variables in our analysis. The values of these variables are presented in the Appendix (Table 10).7

As previously indicated, we use of two indicators that we assume to be strongly related to the effects of the Great Recession: evaluations of the government performance and populist public attitudes. Finally, we also use controls for education, gender, age, and in the case of Spain, regional identities as measured by the Linz-Moreno question.8

The choice model maximizes the utility function in (*) and the conditional logit model of vote choice assumes that voter i’s probability of voting for party j is given by

Where n stands for the number of political parties.

This model cannot determine absolute utility. The utility for an individual must be specified with respect to a base value. For these analyses, the PP in Spain, the PD in Italy, and LR in France have been respectively set as reference levels. The normalized utility is given by U_i (j) – U_i (1), where 1 stands for the reference level, and j=2,…n. In fact, we obtain four values for the estimated parameters in the cases of Spain and France, and five in the case of Italy, because the program gives the differences with the base values, those for PP, PD and LR respectively. The descriptive statistics for these data can be found in in Table 2, Table 3 and Table 4 for Italy, Spain and France, respectively.

The parameters α and β in equation (*)U_i (j) = αV_ij + βp_ij + γ_jZ_i + ϵ_ij are respectively the salience of the voter’s proximity preference and of the vector p_ij. γ_j is the vector of estimated parameters for each individual specific variable.

We report in Table 5 the coefficients for ideology in the proximity and directional models for these three cases. Full statistical results are presented in the Appendix (Tables 11, 12 and 13). As we expect, the ideology coefficient plays a crucial explanatory role and is always significant at the 0.001 level. Ideology coefficients (both proximity and directional) are very similar in Spain and Italy, and they are clearly smaller in the French case.

Based on the coefficients estimated by our conditional logit model we have inferred the Nash equilibria of party ideological positions. The different parties are the agents that compete for the voters. We assume that their only expectation is to maximize their utilities, that is, to get as much electoral share as possible. Each agent has to select one among all the possible alternatives or strategies (a point on the left-right dimension) and a Nash equilibrium (NE) is the solution that selects the set of positions from which no vote-maximizing party has incentives to unilaterally deviate. In that situation, no party would improve its electoral share by unilaterally modifying its position. That means that if x = (x₁, …, x_n) is a vector of strategies NE, any deviation of an agent i to x´_i while all others remain without changes in their initial positions, (x₁,…,x_i-1, x_i+1,…,x_n), would lead this agent to lose votes or, at most, to remain equal, but the agent would never improve. To compute these NE we use the iterative algorithm proposed by Merrill and Adams 2001. Assuming that parties maximize vote shares, in each step of the algorithm each party’s position shifts to its vote-maximizing position holding the other parties’ positions constant. This leads to a new vector of party positions and eventually converges to a NE. In fact, citing Curini and Iacus 2017, “there is a unique Nash equilibrium that is independent of the randomly generated starting points used in the algorithm for parties’ initial placements”. Table 6 reports these results.9

In order to assess the effects on party positions of crisis-related perceptions and attitudes, we have also estimated the ideal party positions after conducting models that do not include government performance evaluations or populist attitudes. For the sake of simplicity, we do not report estimates based on models with populist attitudes and without government evaluations nor those based on models without populist attitudes and with government evaluations.

We then compare those ideal positions to the actual positions of political parties as perceived by all voters in the sample. Figure 1, Figure 2, and Figure 3 report the actual party positions and the ideal party positions (in Nash equilibria) for these three countries and both voting models with and without our two crisis-related variables.10

Table 7 reports the maximum ideological spread in directional and proximity models for the full models (including government performance and populist attitudes). The data reported shows that the positions estimated by both proximity and directional models have a strong centripetal bias.11 The directional model makes a better job at predicting the spread of ideological positions in Spain and Italy, but both models fail to predict the extreme character of the party positions of the PP and Cs in Spain, the PDL and the LN in Italy, and FI, LR, and the FN in France. To a lesser degree, ideal estimates are also more centrist than real positions in the cases of SEL and PD in Italy, and Podemos and IU in Spain. The PSOE is the only party whose ideal position was more extreme than its real position. In France there is a clear instance of ideological leapfrogging in the ideal positions predicted for the PS, which is now to the left of FI. In general, as Adams, Merrill and Grofman (2005, 104-105) point out, ideological leapfrogging makes more sense for small parties.

The data reported in Table 7 allow us to draw some descriptive comparative inferences as well. In the first place, real ideological spread is highest in Spain, and this despite the fact that this country presents the lowest ideological spread in ideal positions in both directional and proximity models. Average differences between ideal and optimal positions are also highest in this country, intermediate in France, and lowest in Italy. These comparisons suggest that the Spanish party system is the one in which centrifugal logics of representation are more intense.

Figure 4 displays the ideological positions of parties according to experts (Polk et al. 2017) and the average absolute differences between ideal and real positions in proximity and directional models. It shows that deviations between ideal and real positions are higher for radical parties in general and for rightist parties in particular. The association between ideological extremeness and real-ideal deviations reflects the centripetal bias of spatial models. It is more difficult to account for the fact that these deviations are particularly large for rightist parties, irrespective of the party family (Christian-Democracy, Radical Right, Liberalism) these parties belong to. If we take the seven parties with higher real-ideal deviations (PP, Cs, LN, FN, FI, PDL, LR), six of them are located on the right of the political spectrum.12

As previously stated, our data do not allow us to assess the impact that party valences for party activists have on party positions. Our survey data provides us however with general public evaluations of political parties. Analysis of this data allows us to establish that some of the rightist parties displaying strong real-ideal deviations, such as the PP, LN, PDL and LR suffer from poor citizen evaluations (Figure 5)13 and that the Pearson correlation between public evaluations and ideal-real deviations variables is negative (-0.37 for un-weighted data and -0.40 for weighted data). However, neither Cs, nor FN, nor FI, three parties also displaying strong real-ideal deviations, suffer from bad public evaluations. Still, if anything, our data is inconsistent with the idea that parties with high valence enjoy the freedom to move to extreme ideological positions.

We lack data on party valence for militants and also on the ideological positions of party militants, and therefore cannot assess whether parties deviate from optimal positions in order to better channel the preferences of their activists. We have however some information on the degree to which party activists enjoy power in some of these parties. Von dem Berge and Poguntke (2017, 147) and Bolin et al. (2017, 177-180) have provided us with an indicator for Assembly-based Intra-party Democracy (AIPD) which reports data for just seven of the parties included in our analyses (LR and PS in France, PDL, LN, and PD in Italy, and PP, PSOE, and IU in Spain). Thanks to this information we know that it is not the case that parties with strong democratic participation deviate more from optimal positions: in fact, the Pearson correlation between absolute ideal-real deviations and party democracy is although weak, negative (-0.45), parties with more democratic structures tending to be closer to their optimal ideological positions than parties with more hierarchical structures. Furthermore, parties with stronger internal democratic practices tend to have a higher general public valence (Pearson correlation of +0.88). It is possible therefore that parties with lower public valence move to the extremes and deviate thus more from their optimal positions (Schofield 2007: 136), but in these cases the reason for that does not seem to lie in the internal power of party activists. If anything, democratic parties tend to be closer to their optimal ideological positions and tend to enjoy higher public valence.

We examine finally whether public attitudes strongly connected to the Great Recession (evaluations of governmental performance and populist attitudes) have some likely effects on the optimal ideological positions of parties. Table 8 reports the Pearson correlations between real and ideal positions estimated with and without crisis-related variables so that we can assess whether the inclusion of these variables affect the realism of our predictions. We report the coefficients weighted and non-weighted by vote results in the general elections that were closer to the survey date. The Pearson correlation coefficients reveal that estimated positions are slightly more realistic for both proximity and directional models when we include crisis-related variables, the directional model showing the highest degree of association between ideal and real positions. This result holds for analyses with weighted and non-weighted data, with results based on directional models being more realistic than those based on proximity models.

We can also examine the national effects of crisis-related variables by comparing the differences they make to ideal positions in these three cases. Table 9 reports the average absolute differences between ideal and real positions in models estimated with and without crisis-related variables. The larger these differences, the stronger the difference these variables make in the estimation of ideal positions. Table 9 shows that differences are far larger in Spain and Italy than in France, both in proximity and directional models. These results are consistent with our expectation that these two crisis-related variables will exert more influence on ideal party locations in the two countries that suffered most in the Great Recession.

Now, these differences are not uniform across parties, but they are particularly large in some specific cases. Figures 2 and 3 show that in Spain and Italy these differences are particularly large in the cases of PDL, LN, and PP. In all these cases, the inclusion of crisis-related variables pushes these parties towards the right and makes their ideal positions closer to their real positions, which suggests that the inclusion of these variables makes our models more realistic. Interestingly, all these parties had governmental responsibilities at some point during the Great Recession (from 2008 to 2011 in the case of the LN and PDL, and from 2011 to 2015 –the survey date- in the case of the PP). In the case of Italy, however, at the time the survey was conducted the government was presided over by the independent technocrat Mario Monti, who enjoyed the support of the PDL and the PD, among other parties. These differences in the trajectories of governmental responsibility notwithstanding, our findings suggest that, in all these cases, due to the distribution of orientations in the electorate, populist attitudes and governmental evaluations tended to push rightist parties to the right.

We started this analysis with the goal of assessing the degree to which crisis-related attitudes (evaluations of governmental performance and populist attitudes directed against the political class) affected party ideological positions after the Great Recession. With that purpose in mind we followed Adams, Merrill and Grofman’s 2005 unified model of party competition in order to predict ideal party positions with and without crisis-related variables, an analysis for which we used both proximity and directional models. Our goal did not lay in the theoretical and methodological revision of standard models of party competition, but on the use of such models to analyze and identify some likely patterns of ideological change in the face of the Great Recession.

The ideal party positions obtained were strongly correlated with the real positions of political parties. However, as it is typically the case in spatial models, ideal positions had a strong centripetal bias, even if this bias was less pronounced in directional models. More interestingly, given the goals of our analysis, we found appreciable differences between the ideal positions inferred using crisis-related variables and those derived from models that did not include such variables. Ideal party positions based on models using governmental performance evaluations and populist attitudes were slightly closer to the real positions of political parties (see Table 8), which means that the inclusion of these variables made our results more realistic. And also, as hypothesized at the beginning of our analysis, differences in ideal party positions were clearly larger in the two countries that had suffered the most during the Great Recession, that is, in Spain and Italy (Table 9). In these two respects, our expectations were not disconfirmed by our empirical findings. Furthermore, such differences affected in particular parties located at the right of the political spectrum (Figure 4) and suffering from negative public evaluations (Figure 5).

More rigorous tests of our expectations regarding the effects of the Great Recession on partisan ideological positions and spatial competition would require including a greater number of European countries in the analysis, and also examining variations across time in the patterns we previously identified. Unfortunately, lack of available data makes it impossible to develop such kind of analysis at this moment. As of now, however, our small-N comparative analysis works as a plausibility probe for our hypotheses on the effects of the Great Recession on spatial competition and thus paves the way for more systematic analyses on these issues.