Economic voting goes local: evidence from the Italian regions

Abstract

This article investigates the pattern of economic voting at the regional level in Italy. It focuses on the elections held in 18 out of 20 Italian regions from 1995 to 2020. Retrospective voting is examined by using the theory of economic voting, measured at the subnational level. By providing some inferential models and controlling for the impact of phases of recession, this article tests the hypothesis whereby the incumbent regional government is rewarded (or punished) by voters in the event of a good (or poor) state of the regional economy. It mainly considers macroeconomic variables, focusing on the relationship between the unemployment rate (at both national and regional levels) and the electoral performance of the incumbent executive. The empirical analysis shows that, particularly during periods of ‘quiet politics’, economic voting also occurs at the local level and thus the regional unemployment rate affects regional rulers' electoral outcomes.

Introduction

Elections are often seen as a means to influence policymakers' behavior and can be studied as an instrument of democracy, as per Powell (2000). The concept can be traced back to Key (1966) and his research on the ‘responsible electorate’, which highlighted citizens' tendency to cast a retrospective vote. He stated (perhaps excessively enthusiastically) that ‘to be sure, many individual voters act in odd ways indeed; yet in the large the electorate behaves as rationally and responsibly as we should expect, given the clarity of the alternatives presented to it and the character of the information available to it’ (Key, 1966: 7). Key has inspired many scholars interested in studying how voters use elections to reward and punish incumbents.

Starting with the reconstruction of Healy and Malhotra (2013), scholars' interests can be traced to three main themes: (1) whether and under what conditions voters can cast a retrospective vote; (2) how voters express retrospective voting; and (3) the conditions under which retrospective voting can promote electoral accountability.

Retrospective voting, based on the mechanism of anticipated reactions (Friedrich, 1963), can play a crucial role in making electoral accountability effective. Examining whether regional governments are held accountable for their policy performance has become increasingly relevant and debated. While regional governments do not – or rarely – shape the state's economic policy, they can certainly impact their regions' well-being. Voters may hold them accountable for worsening or improving of the state of the regional economy and punish or reward them accordingly at the ballot box in local elections. In this vein, the growth of political and policy decentralization which, in recent decades, has affected both established and new democracies (Hooghe et al., 2010; Queralt, 2012) has contributed to this trend.

Following recent studies in the United States (Atkenson and Partin, 1995), Canada (Gélineau and Bélanger, 2005; Anderson, 2008; Gonzáles-Sirois and Bélanger, 2019), Germany (Thorlakson, 2016), Spain (Riba and Diaz, 2002; Aguilar and Sánchez-Cuenca, 2008; Fraile and Lewis-Beck, 2010) and France (Fauvelle-Aymar and Lewis-Beck, 2011), in this article we will examine the conditions and functioning of ‘sanction-and-reward mechanisms’ at the subnational level in Italy. The analysis focuses on the functioning of economic retrospective voting in Italian regions between the electoral cycles of 1995 and 2020. The empirical analysis is mainly based on the theory of economic voting (Anderson, 2000; Lewis-Beck and Paldam, 2000; Gélineau, 2013), according to which voters reward or punish governments based on overall economic performance. Examining economic voting at the subnational level also means studying how vertical relationships between different levels of government affect voters' choices. In fact, from a theoretical perspective in regionalist or federal political systems, ‘when voters […] evaluate government performance, voting retrospectively, they must assess the degree of responsibility each order of government bears for a given policy outcome’ (Cutler, 2004: 19). This is by no means simple or self-evident, because of the several links between the different levels of government which, although in various forms and intensity, characterize all multilevel political systems.

In this article, we aim to determine if and under what conditions regional voters in Italy use elections to cast a retrospective vote based on factors such as the state of the regional and national economy. The paper is structured as follows: in the next section, we will discuss the literature and the rationale for examining the Italian case. In the third section, we will describe the dependent, independent, and control variables used in our empirical exploration. In the fourth section, we will present the empirical analysis and discuss the main results. Finally, we will provide some general concluding remarks.

Multilevel political context and economic voting: the Italian case

Anyone who considers electoral accountability a virtue of democratic regimes agrees that the central mechanism in inducing rulers to act responsively is the voters' ability to oust them. According to a large stream of literature (see, for instance: Powell, 2000; Pasquino and Pelizzo, 2022), retrospective voting can be an effective accountability tool if at least two conditions are met. First, voters must know who is responsible for policy decisions. Second, voters must be able to vote for or against the incumbent. While the second condition is necessarily satisfied in a democratic regime, the first condition is characterized by a wide variability that depends essentially on the institutional infrastructure of the political system.

With cognitive and emotional voter biases (Achen and Bartels, 2004; Healy and Lenz, 2014), the uncertainty about who is responsible for the decisions distorts the incentives posed by retrospective voting. This limits the ability of voters to reduce policy-makers' moral hazard. Although some scholars question this assumption (Géalinau, 2013), the relevance of clarity of responsibility (Powell and Whitten, 1993; Powell, 2000) is emphasized by these considerations. Its essence consists in the perceived level of ‘unified control of policymaking by the incumbent government’ (Powell and Whitten, 1993: 398). Factors such as coalition governments, party fragmentation, and the absence of clear government alternatives hinder the clarity of responsibility (Fisher and Hobolt, 2010).

For similar reasons, multilevel political systems – federal or regional – provide a rather complex scenario concerning the effect of retrospective economic voting. Federalism and regionalism are considered potentially harmful to the clarity of responsibility since they increase the number of actors involved in the decision-making process, thereby making it difficult to attribute responsibilities to specific actors.

From a theoretical standpoint, regional economic voting can also be a function of the type of decentralization of the political system. In this regard, Figure 1 places three possible configurations of the political system along a continuum, with centralized systems (e.g. France) at one end and decentralized systems with shared power (e.g. Germany and Italy) at the other end. Decentralized systems with separate institutions sharing power (e.g. USA) fall in the middle. In this respect, centralized systems are characterized by a unified government, while decentralized systems with separate institutions sharing power have clarity between the policy areas assigned to the central government and those belonging to subnational governments. Finally, decentralized political systems with shared power are characterized by the fact that the central state establishes the general principles and the regions the detailed norms on many issues.

Figure 1. Clarity of responsibility and political decentralization.

Source: Authors' own compilation.

In centralized political systems, regional voters should decide based on national factors, following the logic of the second-order elections model (Reif and Schmitt, 1980; Fauvelle-Aymar and Lewis-Beck, 2011). In decentralized political systems – whether with separate or shared power – regional retrospective voting could be cast by voters. In both cases, voters can perceive that regional governments play an important role in the country's politics, making it plausible that they could affect the region's economic conditions. If voters perceive the role of regional governments through cognitive shortcuts (Healy and Malhotra, 2013), it is implausible to expect a significant difference in electoral behaviors recorded in decentralized systems with separate power and in those characterized by shared power.

Although from a theoretical point of view, the dispersion of power, being called to limit the action of the governmental majority (Dahl, 1956), reduces the clarity of responsibility, the empirical scientific literature on this topic has provided evidence that is far from consistent. Especially regarding the relationship between regional macroeconomic variables and the electoral performance of incumbents at the subnational level.

Starting with the United States, scholars have come to quite different conclusions. Some studies have confirmed the hypothesis that the state's economy's condition impacts the incumbent governments' electoral performance (Hansen, 1999; Niemi et al., 2015). By contrast, although less recent, other studies have highlighted the opposite trend (Pletzman, 1987; Chubb, 1988).

Even in the case of Spain, the evidence is mixed. Using an approach based on individual data and focusing on the analysis on Catalonia, Queralt (2012) explained the absence of regional economic voting by looking at the level of voters' information. On the contrary, Riba and Diaz (2002), in relation to Catalonia's case, show a significant propensity for regional economic voting, especially in cases of negative economic performance. More recently, it has been shown that the level of regional economic voting is a function of the clarity of responsibility of each Comunidad Autónoma (León and Orriols, 2016). Toubeau and Wagner (2018), examining some regions of Canada, Germany, and Spain, instead showed how the diffusion of regional economic voting does not depend so much on the level of clarity of responsibility, as on the congruence between the parties that make up the national government and the parties of the regional government.

The Canadian case was also widely examined. Some analyses have rejected the hypothesis relative to the existence of regional economic voting, arguing that, even at the subnational level, national dynamics eventually prevail (Gélineau and Bélanger, 2005). In contrast, recent studies show a more nuanced picture, in which regional economic voting plays an important role, especially after the explosion of the 2008 financial crisis (Gonzáles-Sirois and Bélanger, 2019).

The same description applies to Germany, where older studies show that economic factors influence elections in the Länder but provide no clear evidence about the influence of national or regional economies (Lohmann et al., 1997). More recent works, on the other hand, show a greater propensity to resort to regional economic voting (Olsen, 2012). This trend is far from present in the case of France, where, according to the few existing studies (Fauvelle-Aymar and Lewis-Beck, 2011), regional elections are used by voters to express their opinion or judgment on the national executive.

In this context, the Italian regional laboratory is particularly promising for at least two good reasons. The first pertains to institutional aspects, while the second has to do with the current state of knowledge in this field of research.

Although the Italian constitution was promulgated in 1948, the implementation of the regional institutions came only in 1970, when the law instituted decentralization. The first regional elections were held in the 15 ordinary regions. Since 1948, however, the statutes of the first four special regions were approved by constitutional law: Sicily, Sardinia, Valle d'Aosta, and Trentino-Alto Adige. In 1963, the statute of Friuli-Venezia Giulia was finally approved.

Between 1999 and 2001, there was a further advance in the process of decentralization of the Italian political system due to a profound change of Title V of the constitution. In 1999, the direct election of the president of the region was introduced (albeit in a formally transitional form) together with the full statutory autonomy of the regions. Yet, in 2001, the relationship between the state and the regions was effectively reversed, leaving the former with exclusive legislative power and the latter with power over everything not explicitly attributed to the state. This has reduced the distance, as well as the power asymmetry, between the regions with special and ordinary statutes.

Regardless of the opportunities offered by the new Title V, in most cases the electoral system of the regions is inspired mainly by the 1995 majoritarian electoral system, which was introduced before the 1999 reform that allowed regions to legislate on electoral issues. The regions with special statutes differ most from the structure envisaged in 1995 and 1999. From this point of view, Trentino Alto-Adige and Valle d'Aosta have peculiar characteristics determined by adopting a proportional electoral system and, above all, by the absence of the direct election of the region's president.¹ For these reasons, these two regions were excluded from our analysis.

Although the decentralization of the Italian political system started several decades ago, the literature on regional economic voting is virtually absent. Some studies have investigated the impact of some regional macroeconomic variables at the national level (Giuliani, 2018, 2022a), while others have focused exclusively on economic voting at the national level, both in relation to the dependent variable and the independent variables (Bellucci, 1991, 2003; Lewis-Beck and Nadeau, 2012). To our knowledge, no study has yet focused on economic voting at regional elections. With this article, we intend to start filling this gap. Based on the theoretical discussion above, we thus propose the following hypothesis:

H1. The higher the regional rate of unemployment, the worse the electoral performance of the regional incumbents.

Two further hypotheses can usefully supplement the main hypothesis. Retrospective voting at the regional level could be influenced by two significant factors: the general condition of the national economy and the congruence between the political color of the regional government and that of the national government. Concerning the first aspect, just as during global recessions a part of the electorate does not tend to attribute responsibility to the national government for the country's economic situation (Lobo and Pannico, 2020), during national recessions, it seems less plausible that voters can perceive the role played by the regions as an independent one. Arguably, during a national economic downturn, the general state of the economy becomes more salient. For this reason, we expect voters to exercise a relative economic vote, based on the benchmarking effect of the national economy with regard to the regional vote (Arel-Bundock et al., 2021; Giuliani, 2023). Concerning vertical congruence, however, it is likely that the performance of regional governments is influenced by political overlapping (or congruence) between the two levels of government. As Toubeau and Wagner (2018) argue, in case of political incongruence between the two levels of government, support for the regional incumbent is relatively more limited, especially when the state of the regional economy is negative.

H2. In times of national economic recession, it is more likely that regional voters will compare local and national states of the economy to make their electoral choice.

H3. Regional voters are more likely to punish incumbent regional rulers when their party (or coalition) is not in power at the national level, especially when economic conditions are bad.

In the following sections, we will empirically test these three hypotheses.

Variables and data

Dependent variable

Our dependent variable is the electoral performance of the ruling parties in Italian regions. According to Powell and Whitten (1993: 393), this variable is operationalized as ‘the gain or loss in percentage of the vote received by the incumbent governing party (or coalitions of parties)’. Figure 2 illustrates the trend in the electoral performance of the regional executive for the 18 selected regions from 1995 to 2020.

Figure 2. Incumbent's electoral performance in 18 Italian regions, 1995–2020.

Source: Authors' own compilation.

The cost of governing, already observed and analyzed for the national executives in European countries (Narud and Valen, 2008), is also apparent at the subnational level. With very few exceptions, such as those observed in Apulia and, more recently, in Campania, Emilia-Romagna, and Veneto, regional executives tend to lose support while in office. On average, incumbent regional governments experienced a decline in their vote share of 8.2 percentage points during the period under investigation.

However, the magnitude of the electoral losses varies from region to region, and in some circumstances, it reflects the overall ‘swing’ of the electoral pendulum from left to right (or the other way around), during specific election cycles (for instance, to the left in 2013–15 and to the right in 2018–2020) (Tronconi, 2015; Massetti, 2018; Tronconi and Valbruzzi, 2020). At the same time, our data show that those regional executives facing elections during times of economic downturn, especially in the period 2008–2015, experienced larger electoral losses (for a similar perspective at the national level, see Bremer et al., 2020).

Independent variables

Following the theoretical framework discussed in section 2, our main independent variable concerns the state of the economy at election time. In general, according to classic economic voting theory investigated especially at the national level, the incumbent government benefits at the ballot box from economic improvement and suffers from economic decline. In our study, the concept of ‘economic improvement’ (or deterioration) has been operationalized by considering the unemployment rate. This choice is mainly due to two reasons: first, unemployment is the main – or one of the most important – concerns of Italian citizens;² second, (un)employment is widely and commonly used in the literature on economic voting, representing in many analyses the main predictor of voting behavior (Dassonneville and Lewis-Beck, 2013; Giuliani, 2022b).

Since we aim to investigate whether the retrospective economic voting considers local or national factors – that is, the economic performance at either the regional or the country level – the unemployment rate is measured at both levels. More specifically, at both regional and national levels, unemployment is calculated as the average of the values of the four quarters preceding the vote, including the one in which the elections are held.

In addition to this economic variable, the analysis also includes a possible moderating or accelerating factor for the logic of economic voting. In fact, as anti-incumbent voting may be stronger during recessionary periods, we have added a variable that taps into the consequences of economic downturn. Operationally, this variable is coded as 1 if the economy is in recession, and 0 otherwise. More specifically, Italian economy is considered to be in recession when there are two successive quarterly declines in gross domestic product (GDP), a measure of the nation's output.

Control variables

In addition to the main independent variables related to the state of the economy, the analysis also includes several control variables. The first group of control variables pertains to the institutional or electoral context at the regional level. Specifically, we consider the degree of clarity of responsibility, which relates to the likelihood that voters can correctly assign responsibility to the (regional) executive for its decisions and, above all, the social, economic, and political outcomes of those decisions. We focused on the clarity of responsibility, by adopting the concept of ‘availability of clear alternatives’ as proposed by Anderson (2000) and Bengtsson (2004). This concept is operationalized by three indicators: electoral fragmentation, election closeness, and bipolarism index.

The literature (Powell and Whitten, 1993; Anderson, 2000; Powell, 2000; Hobolt, Tilley and Banducci, 2013) shows that the availability of clear alternatives is greater when voter faces alternative coalitions and a zero-sum form of inter-party competition. In such circumstances, voters can reward or punish incumbent governments for their performance and the state of the economy at the individual or aggregate level.

The direct election of the president of the region, introduced in Italy in 1999, provides voters with a greater chance to identify the main responsible party for policy decisions. On the one hand, this form of government brings the Italian regions closer to presidential regimes (Samuels, 2004). On the other hand, the absence of separation of powers between the executive and legislative – typical of presidential systems – and, vice versa, the merger between the assembly majority and the president, brings them closer to the parliamentary systems centered on the prime minister (Sartori, 1994). The combination of these two characteristics seems to create an ideal institutional environment for the operation of electoral accountability mechanisms.³ However, the regional political regime, and particularly the clarity of responsibility in the regional electoral arena, may be influenced by the pronounced fragmentation of the political supply. According to Bengtsson (2004), the smaller the electoral fragmentation, the higher the probability that voters can correctly identify decision-makers and, therefore, the greater the availability of clear alternatives.

Hence, in this study, the multidimensional concept of availability of clear alternatives has been operationalized by including the level of electoral fragmentation in the regional arena (Anderson, 2000; Nadeau et al., 2002) in our statistical analysis, which is measured using the index developed by Laakso and Taagepera (1979). More specifically, the ‘effective number of electoral parties’ provides a useful and easily understandable measure of electoral fragmentation by considering the relative strength of each electoral party in terms of votes at each regional election. In operational terms, the effective number of electoral parties can be described as ‘1 divided by the sum of the squared decimal shares of the vote for (or seats won by) each electoral party’ (Dunleavy and Boucek, 2003: 292–293). As we mentioned earlier, we argue that the possibility for the voter to correctly assign responsibility is limited or reduced when the electoral arena is highly fragmented.

Our analysis includes a measure of electoral uncertainty useful for grasping a precondition of electoral accountability, namely, the probability of government alternation. Indeed, the electoral geography of Italian regions shows contexts where the electorate is more stable over time⁴ and areas where the electoral support for the main political parties is highly volatile, especially in Southern regions. Our measure of electoral uncertainty aims to capture the degree to which the incumbent government is electorally vulnerable or, in Bartolini's terms (2000: 52), ‘the psychological effect linked to the absence of safety experienced by the governing parties’. In this context, we operationalize electoral uncertainty in terms of closeness, specifically as the electoral margin (in percentage points) separating the two main coalitions in two consecutive elections. We expect that when and where the incumbent regional government faces higher electoral uncertainty, the chances of observing an alternation in power increase. As Bartolini (2000: 52) pointed out, ‘closeness and uncertainty may not result in turnover but still provide their supposed effect on competition’. This effect makes easier for the voter to assign responsibility to the governing parties for their actions while in office. In other words, the lesser the margins between the top-two parties at the time t–1, the higher the availability of clear alternatives at time t, and the lesser the expected electoral performance of the incumbent government.

At the party-system level, a third factor that contributes to a clearer assignment of responsibilities with respect to government decisions pertains to the mechanics of inter-party competition and the existence of alternative coalitions willing (and available) to alternate in power. To empirically understand this phenomenon, we use the ‘bipolarism index’ (Bolgherini and Grimaldi, 2017: 495) measured in the regional arena. The index is calculated by adding up the electoral support (as a percentage of the total vote) of the two largest coalitions. Thus, the higher the value, the more the political system is based on the competition between two coalitions. The theoretical maximum score (100) describes a perfect ‘duopoly’ situation, in which no votes are cast outside the two main electoral blocs. Concerning our study, we can hypothesize that a higher level of bipolarism is associated with greater clarity of responsibility and, therefore, with the possibility of a change of government at the regional level.

The last control variable refers to the ideological proximity or ‘congruence’ (Massetti and Sandri, 2013) between regional and central governments. The literature has shown that in federal or ‘regionalized’ systems, unlike more centralized political regimes, the congruence between the partisan ideologies of the two levels of government has a significant effect on the policies adopted by the regional executive and, consequently, on the economic performance of the region (Khemani, 2007; Toubeau and Vampa, 2021). In operational terms, ideological congruence in the partisan composition of regional and national governments is treated as a dichotomous variable: a region in which the main ruling party is the same as the controlling central government is coded as 0 (congruence), while a region governed by a different party is coded as 1 (incongruence).

Empirics: model specifications and results

As the dependent variable is continuous, we tested our hypotheses on the existence of economic voting in Italian regions by running ordinary least squares regression models with robust standard errors clustered by regions.⁵ All analyses below include the control variables mentioned in the previous section (electoral fragmentation, election closeness, bipolarism index and vertical partisan congruence). In models 5–9, we also include the possible effect of recessionary periods in Italy. The descriptive statistics for all variables included in our dataset are displayed in Table 1. Additionally, the models include three interactions to assess the role played by the interaction of the economic recession, vertical congruence, and the main macroeconomic variables on the dependent variable.

Table 1. Descriptive statistics

Table 2 includes nine regression models aiming to explain the electoral performance of the incumbent regional government and test the existence of a regional pattern of economic voting. Models 1 and 2 include only our main independent variables, and as can be seen, the relationship between the incumbent's electoral performance and the regional unemployment rate, while remaining statistically significant, weakens with the inclusion of the unemployment rate measured at the national level (equal to −0.713, P < 0.1).⁶ As discussed in section 2, H1 posited the existence of a pattern of economic voting at the regional level in Italy. The results of the empirical analysis across all models show strong evidence for this relationship, albeit at the lowest accepted significance level. In other words, regional voters consider the state of the local economy, among other things, in their voting behavior and, more importantly, in judging the incumbent government's performance.

Table 2. Patterns of economic voting in Italian regions, 1995–2020

Note: Robust standard errors clustered by region in parentheses. ***P < 0.01, **P < 0.05, *P < 0.1.

Therefore, in the absence of other controls for possible confounding factors, both macroeconomic variables – the regional and the national ones – are associated, to a different extent, with the electoral performance of the regional government. More precisely, while model 2 shows a sort of spill-over effect from the national to the local level, the introduction of control variables in models 3 and 4 reveals a somewhat different scenario. By controlling for the effect of ideological incongruence between the incumbent regional government and the central government (model 3), national unemployment rate loses statistical significance, while the regional unemployment rate retains it, suggesting for the existence of absolute (or conventional) economic voting. However, the control variable (vertical incongruence) is not statistically significant and, looking at the overall fit of the model, the differences from model 2 are marginal.

Summing up what we have observed so far, the results of the empirical analysis reveal that voters' electoral choice is somewhat independent of the national economic outlook. This finding can be visualized in Figure 3 (based on model 3), which plots the expected effect of both regional and national unemployment rates on regional government election performance. As evident, the impact of the regional unemployment rate on the dependent variable is confirmed, while the association with national unemployment is less clear.

Figure 3. Predicted effect of regional and national rate of unemployment on the electoral performance of regional government.

Note: Based on model 3 in Table 2.

The addition of control variables concerning the concept of availability of clear alternatives (electoral fragmentation, election closeness, bipolarism index) indirectly provides further support to H1. In fact, the results show that the relevance of (regional) economic voting increases when variables concerning the clarity of political alternatives are included in the regression models. When the conditions underlying the availability of clear alternatives are present, the likelihood of a government change at the regional level is greater. Looking at the results of models 4–9, it is necessary to make some clarifications about the behavior of the control variables related to the availability of clear alternatives. In fact, if the relationship between our dependent variable and election closeness has the (positive) expected sign and is statistically significant, the other two variables are either not significant (electoral fragmentation) or go in the opposite direction to that expected (bipolarism index). More specifically, when the electoral margin between the two main coalitions shrinks, the performance of incumbent rulers tends to worsen. In contrast, as the value of the bipolarism index increases, the electoral performance of regional executives does not worsen (as expected) but improves.

Among the many factors that may explain this somewhat unexpected result, we can point to two. First, our measure of bipolarism may contain different combinations of electoral outcomes, such as that observed in Veneto in 2020, when the center-right garnered 77% of the vote and the center-left 15%. Despite the high level of bipolarism (92%), the structure of the competition resembled unipolarism more than a system based on two main alternative coalitions. Second, in the period under investigation, especially since 2010, the electoral breakthrough of the Five-star Movement, placed in an intermediate position between the center-right and center-left, had two effects. On the one hand, it weakened the bipolar structure of competition. On the other hand, it contributed to making electoral races and the very survival of regional governments more uncertain.

The second hypothesis to be tested (H2) relates to the impact of national recessionary periods on the electoral performance of regional executives. Model 5 adds the variable concerning economic recession to the equation and as can be seen, its effect is quite strong, consistent under different model specifications (see also models 6–9) and statistically significant. Holding all other variables constant, on average a recessionary period in the Italian economy brings about a loss of 5 percentage points for the incumbent regional ruler. In this respect, our findings bring some initial evidence if favor of H2. However, what is more important in the present study is that when recessionary periods are included in the regression, the impact of regional unemployment retains its significance, while the effect of national unemployment is put into question.

To further explore this dynamic, models 7–9 interacted economic recession with both national and regional unemployment. The results show that when controlled for economic recession, the direction as well as the strength of the relationship between the electoral performance of regional governments and unemployment at the regional level remain substantially unchanged. This means that economic crises do not significantly alter the association between retrospective regional voters and the local rate of unemployment but contribute to making the assessment of the incumbent regional executive even more negative.

Instead, the interaction between recessionary periods and national unemployment produces unexpected, but theoretically relevant, outcomes. In particular, it reveals that the negative effect of national unemployment on regional governments' electoral fate applies especially during recessionary periods. In other words, economic downturns help to ‘nationalize’ regional elections, getting closer to the logic of the second-order elections model. However, despite the effect of economic crises on the electoral fates of the regional executive, local unemployment retains its predictive power. Again, this provides further support to H1, confirming the existence (and persistence) of an independent and direct pattern of economic voting at the regional level.

However, to better grasp such ‘nationalizing’ effect of economic crises, in model 9 we jointly included both interactions (i.e. economic recession with both regional and national unemployment rates). This helps us in detecting the evidence of some benchmarking effect at the regional level, i.e. ‘the possibility that citizens compare the local [regional] performance to the corresponding national average, or to some past situation before the crisis period’ (Giuliani, 2022a: 81). Overall, the estimated coefficients of the interaction terms confirm our previous findings, especially the existence of a pattern of economic voting at the regional level, but at the same time, reveal the presence a moderating effect exerted by the state of the economy at the national level. In other words, when general economic conditions turn bad, their effects spill over to the local level, affecting (but not neutralizing) the pattern of economic voting in each region.

To disentangle this complex relationship existing in a multilevel setting, Figure 4 displays the predicted effects of regional and national unemployment rates on regional executives' electoral performance in both recessionary and ‘normal’ periods. This helps to better illustrate the effect of local economy on the fate of regional rulers and, in particular, to ‘relativize’ the observed pattern of economic voting and test whether a benchmarking effect exists.

Figure 4. Predicted effect of regional/national rate of unemployment on the electoral performance of regional government, by economic phase.

Note: Based on model 8 in Table 2.

Our findings suggest two different patterns of economic voting taking place in different economic circumstances. In ‘normal’ economic times, regional voters consider the state of the local economy in relation to the national state of the economy. In times of economic recession, retrospective voting at the regional level might imply a comparison with the present condition of the country (Aytaç, 2018; Park, 2019), with the general economic condition of the country acting as a benchmark. In contrast, in ‘extraordinary’ times, such as economic recession or global crisis, the generalized state of the economy spills over to the local level, reinforcing the negative assessment of retrospective regional voters. To use a formula: if it rains, it rains on everybody.

Our final hypothesis (H3) considers situations in which regional and national governments share the same ideological orientation. The assumption is that regional voters are more likely to punish incumbent regional rulers when their parties are not in power at the national level, especially when economic conditions become critical. In models 3 and 4, we have already seen that the inclusion of the variable concerning vertical (in)congruence does not change the overall fit of the model, and its effect turns out to be statistically irrelevant. However, to test this hypothesis more precisely, in model 6 we added an interaction between the variable concerning the ideological incongruence of governments (national and regional) and the state of the national economy in terms of unemployment. Again, the results show a zero effect of vertical incongruence on the performance of regional governments, and its impact appears to be entirely unrelated to the conditions of the national economy. Thus, this finding leads us to reject our last hypothesis (H3) and confirms the existence of a ‘localized’ pattern of economic voting in Italian regions.

Discussion and concluding remarks

Retrospective voting closely linked to electoral accountability and is a fundamental aspect of democratic regimes, as it enables voters to elect or remove their representatives. Traditionally, scholars have mostly focused their analyses on the national level, often neglecting the subnational level. However, in recent years the growing importance of regional and provincial politics has drawn attention to this area of research. This article aims to contribute to this field by examining the Italian case which has thus far been largely overlooked.

Although our analysis is just a starting point on a topic that deserves further investigation, its main findings are, to some extent, consistent with the expectations of economic voting theory. However, there are some critical caveats. Firstly, we found that regional macroeconomic factors have a significant impact on the voting behavior of regional voters, particularly during periods of ‘ordinary’ or quiet economic conditions at the national level.

The relevance of regional economic factors may be due to Italian voters' increasing awareness of the opportunities available to regions in terms of policy choices and the growing media visibility of regional presidents. In contrast, during periods of economic recession, the severity of the social condition reinforces the impact of subnational factors and a spill-over effect occurs, whereby the national economic conditions affect the regional level. While these insights require further investigation and larger datasets, they shed light on a topic that has unfortunately been overlooked in previous research.

More specifically, our analysis reveals that when asked to choose their own regional executive, Italian electors retrospectively reward or punish the incumbent regional government based on regional macroeconomic variables, especially during ordinary times. In other words, when politics is mainly affected by local factors, voters at the regional level take into consideration and react to these factors in evaluating the performance of the regional executives. Instead, when national or international factors plague the economic system, regional voters react to these circumstances by taking into consideration the condition of the national economy for their electoral choices. In these situations, regional elections turn out to be, at least partially, more like second-order elections that assess the national executive's popularity rather than the regional rulers' performance.

Furthermore, concerning this research's findings, it is essential to stress the relevance of the bipolarity of the regional party system and election closeness, which are both associated with the electoral performance of the incumbent parties. This seems to suggest that economic voting requires, at both the national and regional level, institutional or political preconditions without which the promises of electoral accountability cannot be fully met.

Moreover, our analysis has greatly downplayed the relevance of ideological or partisan attitudes for the regional economic voting dynamics. Regional voters do not seem to be more inclined to apply a reward-punishment model of voting behavior when their party is not in power at the national level. Quite the contrary, regional competitions in Italy are slowly turning into first-order elections in which the incumbent rulers are judged for what they have done or not done while in office. Only exogenous shocks, such as economic crises, can alter this trend.