Who benefits from the retrospective vote? Bringing in new parties

Abstract Democracies across the world have experienced the rise of new political parties. The dominant view of the model of (retrospective) economic voting implicitly assumes that the main beneficiary of electoral change is the established opposition. However, the rise of new political parties affects how we think of retrospective (economic) voting. This article presents a more nuanced picture of electoral change and considers when electoral turnover benefits established opposition parties, and when new political parties. The theoretical model introduces different macro-economic and macro-political motives for electoral turnover. Using a novel dataset on electoral change, covering 59 democracies worldwide, it is found that high levels of corruption discredit the entire political establishment, and promote the rise of new parties. The effect of economic hardship is more nuanced. Low economic growth mainly benefits the established opposition because voters look for an established alternative within the political system. Rising unemployment, by contrast, promotes the rise of new parties.

The recent 'meteoric rise' of political newcomers (Carreras 2017), who have been entering the political scene with two-digit vote shares or even becoming the largest governing party in the first elections in which they compete, is no longer unique only to countries in the political periphery.The rise of new political parties now also affects how we think of retrospective (economic) voting.The model of retrospective voting expects voters to defect from the incumbent government in response to low economic performance (Dassonneville and Lewis-Beck 2014;Fiorina 1981;Lewis-Beck 1986).Periods of economic recession lead to a swing in voting between incumbents and (established) opposition parties, though the question of whether and when voters lose faith both in the incumbent and in the established opposition remains unanswered.Hence, new political parties have attracted interest within the literature on party systems (Mainwaring et al. 2017;Powell and Tucker 2014;Sikk 2012).
This article taps into this question, and connects the literature on party system change and on retrospective voting, asking who benefits from the retrospective vote, when the incumbent fails?It scrutinises the electoral response to economic cycles as well as to corruption for two different types of non-governing parties: established opposition parties and new parties.It uses one of the most comprehensive samples of retrospective voting covering 59 democracies worldwide, over 25 years .This allows us to examine the differential effect of corruption and of economic hardship on the success of new and established opposition parties.
The empirical analysis builds on two new measures, which differentiate vote shifts into: • firstly, vote shifts between the governing parties and opposition parties ('opposition gains and losses'; cumulative, in vote percentages); • secondly, vote shifts benefitting new political parties (cumulative, in vote percentages), covering the rise of parties with no previous electoral legacy, as well as previously electorally insignificant parties.
The two indicators analyse electoral change from a retrospective voting perspective, looking at changes between incumbents and opposition parties (not counting vote shifts within the governing coalition or between established opposition parties).
Our results point to two genuinely different retrospective voting effects, depending on whether voters react to governance issues (corruption) or to the economic dimension (economic crises versus unemployment).While factors associated with competition over economic programmes (economic growth) affect the electoral success of established opposition parties, factors leading to generalised mistrust and 'antipartyism' (corruption and rising unemployment) offer opportunities for political newcomers to gain voters.
The remainder of this article proceeds as follows.We first present our theoretical argument, distinguishing between the vote gains of established opposition and new parties.We then discuss our measures and report the empirical models.The final section concludes.

Theoretical argument
Retrospective performance voting has been identified as a key factor of electoral change: voters reward or punish incumbents for the results of their government action.In this section, we innovate by asking not only in what circumstances government parties lose, but also who benefits from retrospective voting.
The classical literature on retrospective voting has focussed on economic performance as the simplest indicator of government performance (Fiorina 1981; for an overview: Lewis-Beck and Stegmaier 2000; Nadeau et al. 2013).The classical measure thereof, growth, is often complemented by measures of unemployment or inflation (e.g.Tavits 2008).More recently, scholars have increasingly scrutinised the electoral reaction to other issues on which voters might share common preferences, and which might affect popular satisfaction with governments.Such 'valence issues' (Enelow and Hinich 1982: 117) include war casualties, the government's handling of natural disasters (Healy and Malhotra 2013: 294-95), good governance (Bowler and Karp 2004), and in particular corruption (cf.Crisp et al. 2014;Peters and Welch 1980).Most of the literature finds that good handling of difficult situations and good governance benefit incumbent parties.However, results on the electoral effects of corruption for the incumbent are more mixed (e.g.Anduiza et al. 2013;Fernández-Vázquez et al. 2016; for a review: De Vries and Solaz 2017).Aiming to explain the heterogeneity of the effect, scholars have pointed to economic performance (Klašnja and Tucker 2013;Zechmeister and Zizumbo-Colunga 2013) and to institutional characteristics as potential contextual factors. 1  Building on this literature, we focus on the economy and aspects of good governance (represented here by corruption), not only due to their prominence in the literature, but in particular because both issues have the potential to become salient across a large set of democracies, at least when the economy performs badly or under high levels of corruption (Bågenholm and Charron 2014).

Who gains when the incumbents lose?
In the 'sanctioning model' of retrospective voting, which we rely on, citizens judge the government's competence based on past performance.'On the basis of that assessment [the elector] votes for or against the political incumbent' (Lewis-Beck and Stegmaier 2008: 30).However, this literature rarely investigates the beneficiaries of incumbent losses.
We focus on the functions that established opposition parties and political newcomers can fulfil in offering voters an opportunity to vote retrospectively.In some contexts, retrospective voting might lead to a pendulum swaying continuously between government and opposition, especially when there are two large parties.When bad government performance leads to a loss of trust in political institutions, we argue, new parties will flourish.
The classical literature on party system change highlights the importance of institutional, as well as socio-cultural and structural factors for the entry of new parties (e.g.Harmel and Robertson 1985).A newer generation of studies in the field of party system change looks at new parties as carriers of electoral volatility.They find that economic performance is primarily associated with electoral shifts among established parties, whereas for party system change, other factors, such as the age of democracy, or turnout change are more relevant (Mainwaring et al. 2017;Tavits 2008).
Our research question is more closely aligned with the literature on retrospective voting, and the reaction to economic performance and other, non-economic aspects of government performance.Accordingly, the distinction between incumbents and the opposition, as well as the distinction between electoral gains and losses is more central in our work.Different types of performance, which motivate voters to (retrospectively) sanction the incumbent, do not necessarily have the same beneficiaries.In particular, there are factors that may be addressed in programmatic competition (on which either established or new parties can capitalize).Other factors are more likely to affect trust in political institutions and political parties, and they produce an unspecific (not programmatically rooted) sentiment of 'antipartysm' (Bélanger 2004).These different factors are also reflected in the type of electoral reaction.The incumbent's (economic) policies and the country's economic performance are a 'natural part of the political game' (Bågenholm and Charron 2014: 906-7).Economic downturns might shift voter preferences and evaluations on the economic dimension of party competition.Thus, the performance translates into a debate about the right (economic) programmes, benefiting established opposition parties offering an alternative programme.Bad governance, and in particular corruption, is not about politics-as-usual.Research on the micro-foundations of the political consequences of corruption shows that it primarily erodes the trust in political elites and in institutions more generally (Chong et al. 2015;Winters and Weitz-Shapiro 2013). 2 Likewise, micro-level studies report that unemployment may reduce trust in institutions and contribute to political alienation (e.g.Algan et al. 2017).Armingeon and Guthmann (2014) report similar effects for a larger array of economic repercussions of the global financial crisis of 2008.Thus, under economic or political conditions that erode citizens' trust in political institutions and parties, such as high levels of corruption or rising unemployment, we might expect citizens to spurn participation in elections, or to vote for parties that are not part of the political establishment.
The electoral prospects of various types of political outsiders are subject to a growing body of literature.Carreras (2017) looks at the appointment of heads of government with no conventional institutional trajectory, while most other studies look at the electoral stage and the electoral success of political outsiders (Engler 2016;Lavezzolo and Ramiro 2018), or at parties with no prior government experience (Hobolt and Tilley 2016).Studies of new parties find them to be successful in times of crisis (Hobolt and Tilley 2016;Hug 2001: 118-9;Vidal 2018).Rico and Anduiza (2019) point to discrepancies between the subjective perception of the economy and objective indicators.Studies of anti-establishment parties with centrist positions link their success to rising rates of corruption (Hanley and Sikk 2016), and a crisis of political trust (Engler 2020), rather than to economic recessions.Turning to political extremists, Anderson (1996) and Arzheimer (2009) find that unemployment benefits the radical right.Lubbers et al. (2002) present contrary evidence for the 'sociotropic' effect of unemployment, i.e. that it is linked to the national unemployment rates, while Kurer (2020) reports counter-evidence from the 'pocketbook' model, showing that individuals who lose in economic status do not shift to the radical right.
While most of this work looks at one specific type of party, Algan et al. (2017) systematically compare various types of newcomers or populist parties, and find similar effects for all except the far right.These results are close to our reasoning about the different effects of corruption and economy: we argue that one factor -the economy -allows the opposition to offer an alternative economic programme, whereas the loss of political trust (e.g.linked to corruption and a loss of economic status) instead offers opportunities for newcomers.
Accordingly, hypotheses 1 and 2 stipulate a differentiated retrospective voting effect on established opposition parties and newcomer parties for the economy and for corruption.
H1a: Low or negative economic growth will benefit established opposition parties more strongly than new political parties.
H1b: High unemployment rates will benefit new political parties.H2: High levels of corruption will benefit new political parties.

Electoral gains and losses of new and established opposition parties
The empirical section analyses which parties benefit from voter dissatisfaction with government.In contrast to previous analyses, we distinguish between established and new opposition parties, and we consider two types of retrospective voting: the economy and corruption.
Thus, we count and explain the electoral gains and losses of opposition parties, with elections (nested in countries) as our unit of analysis.We calculate opposition vote change, as the cumulated vote change for all parties of the established and new opposition, varying between negative (vote losses) and positive values (vote gains).New parties' success is measured as the total (cumulative) vote share gained by new political parties in an election; vote gains can be concentrated on a single party or split between multiple political parties.To measure opposition vote swings, we have compiled one of the largest datasets to date on vote change.The dataset includes almost all democracies in Europe, South, Central and North America, Asia and Oceania (Polity IV score ≥ 6 at election time) with at least 500,000 inhabitants, and for a period of 25 years (1990-2014) (see Online Appendix A2).A few countries and elections were excluded due to a lack of data or stable party affiliations.We partially build on data from Mainwaring and Zoco (2007), Bochsler (2009), Lane and Ersson (2007), Ruth (2014) and Bochsler and Hänni (2019).For all countries, we have used our own coding criteria (see below) in order to match the parties over election periods, and we have significantly extended earlier data collections in order to cover the full time period.We rely on information from the EJPR Yearbook, Electoral Studies' Notes on Elections, and country-specific information to update previous sources. 3 As long as political parties are stable the calculation of vote change is straightforward, and we can compare the (aggregated) vote difference between two elections.Many of the countries covered in our project are characterised by frequent party splits and mergers.For splits and mergers, we follow Sikk's (2005) advice: where two parties merge, we measure whether the new party's election results improve or worsen overall as compared to their joint vote share in the previous election.If a party splits, we compare the election result of the old party with the joint vote share of the successor parties (see Online Appendix A).Party splits and mergers are not considered new parties and are counted as part of the established opposition (unless they were in government).
Indeed, one major challenge is the identification of new parties.There are several approaches to measuring electoral change through new parties.Most of this work measures party system change as organisational change, assessing the likelihood or the frequency at which new parties enter the electoral arena, or the supply-side (Harmel and Robertson 1985;Hug 2001;Tavits 2006).A newer generation of studies in the field of party system change looks at new parties as motors of electoral change, dissecting electoral volatility benefitting established parties and electoral shifts associated with changes in the set of parties (Mainwaring et al. 2017) or with party entry or exit (Emanuele and Chiaramonte 2019).Our empirical approach differs from this work in two ways.First, we split overall electoral change (volatility) into net gains and losses by a specific type of party.Second, we distinguish between incumbents and non-incumbents (oppositions and new parties).This allows us to run classical models of retrospective (economic) voting, explaining the electoral fortunes of incumbents versus the opposition, though in our case extended to a further type.
This extension to new parties requires an operationalisation of 'new parties' , a term that is employed very differently in previous work: Hanley and Sikk (2016) identify genuinely new parties that 'are not successors to any previous parliamentary parties, have a novel name and structure, and do not have any important figures from past democratic politics among their major members' (Sikk 2005: 399).Others link novelty to government participation: Hobolt and Tilley (2016) study 'new challenger parties' , which are parties that have never been in government since 1970.
Following the interest of this paper in retrospective voting, we introduce a new operationalisation for the electoral consequences of retrospective voting.We follow Sikk (2005: 394) and consider parties as organisational units, analysing any vote shifts between these units.Similar to previous research which has distinguished between organisational change and vote shifts (Mainwaring et al. 2017;Powell and Tucker 2014), we only measure (aggregated) electoral gains and losses that are due to changes in voter support or (de-)mobilised voters (see Online Appendix A1 for a more detailed comparison of the measures).We define new parties conventionally, as entities with no organisational predecessor, competing for the first time in elections.Some of these new parties might include established politicians who switch parties (Hanley and Sikk 2016).We follow the general approach and count them as new parties.However, we disregard vote gains or losses that are due to organisational change (changes of electoral alliances, mergers or splits).A party is considered to have split if entire branches quit and form a new party (e.g.SVP-BDP split in Switzerland, 2007), or a party breaks into two parts (e.g.SRS-SNS in Serbia, 2012).
A second version of our measure consists of cumulative electoral gains and losses of new parties and previously electorally marginal parties which have previously not been included in the government. 4 Thus, our measure includes (previously irrelevant) party organisations that are reformed, and arise electorally (Powell and Tucker 2014).Instances thereof can be found across the world, where typically a political entrepreneur reinvigorates a marginal party: the Italian Lega Nord emerged in the 1992 elections with 8.2% of the vote as the successor of completely marginal local movements, counting a total of 0.5% of the vote in previous elections.In Bolivia, Evo Morales converted the sleeping left-wing movement, MAS, for his electoral success in 2001 (from 3.7 to 20.9% of the vote).In Lithuania, the Liberal Union rose from 1.9% to 15.3% in the 2000 elections, with the shortlived Prime Minister Rolandas Paksas as a prominent candidate and party switcher.
In the models included in this article, we rely on an index of electoral gains and losses of parties that either did not run or gained less than 5% of the votes in previous elections. 5Of course, this will also include small or regional parties with long political records.However, the cumulation of vote gains and losses avoids the measure being systematically driven by small but established parties. 6 Likewise, we employ two equivalent operationalisations for established opposition parties.Either these are parties who have participated in the previous elections (in the first model) but were not part of the government at the end of the legislative period (excluding interim governments), 7 even if they won only a handful of votes.Or, in the second model, we only include electorally relevant non-governmental parties, with at least 5% of the votes in previous elections.In analogy to our definition of new parties, we consider parties resulting from mergers or splits of opposition parties, and parties merely changing names as established opposition parties.Robustness checks use a seat threshold instead of a vote threshold, in order to assign electorally relevant parties without a relevant parliamentary group (e.g. the Front National in France, or UKIP in the United Kingdom) to the same category as new parties (see Online Appendix A3).
Figure 1 reports gains and losses of new and established opposition parties since 1990.We see little variance over time, but there are much fewer elections with successful new parties towards the end of the period.However, there is great variance in the success of new parties between elections: elections where newcomer parties win landslide victories are the exception, but they occur throughout the period of analysis.Opposition parties, by contrast, not only gain, but also lose votes.On average, they gain more votes than they lose as indicated by the positive median.This means that the incumbency disadvantage (Roberts 2008), which is in favour of the opposition remains, even after we exclude new parties from the opposition.

Performance indicators
The models in this article explain these gains (and losses) for opposition parties in terms of macroeconomic performance and corruption.
Economic performance is measured by economic growth (World Bank 2018).We also include results for changes in unemployment levels, another prominent measure in the literature on economic voting, 8 and have tested several transformations of growth, capturing possible differences between positive growth and economic recessions. 9In line with other macro-level studies of economic voting, we consider performance indicators from earlier parts of the legislative period as less important, and we assume a one-year lag in the effect, lagging both economic indicators by one year for all elections taking place before November.We assess the level of Corruption with an indicator from the International Country Risk Guide (The PRS Group 2013), which evaluates corruption within the political system.Values range from (1) (high risk of corruption) to (6) (no risk of corruption between politics and business).We invert the scales of corruption and of growth to facilitate the interpretation.High values thus indicate high levels of corruption or negative economic growth. 10 While the previous literature has used the Corruption Perception index (CPI) (Mishler and Willerton 2003: 131), we favour our alternative operationalisation, as the coding of the CPI has changed several times and can only be used for comparisons over time as of 2012, when a completely new methodology was introduced (Transparency International 2013).

Control variables
We consider several contextual factors and their direct and indirect influence on vote gains and losses of opposition parties.We include the age of democracy in our analysis, as new parties have for a long time been a phenomenon of younger democracies.Democratic experience is measured based on the number of (uninterrupted) years a country has scored as a democracy in Polity IV (a value ≥6 on the combined Polity score).We count the number of years since the last transition.
We assess the the clarity of government (parties') responsibility because retrospective voting requires voters to be able to clearly assign responsibility for the current political and economic situation.We operationalise clarity of responsibility with the size (number of parties) of the government coalition.Higher values refer to lower responsibility.When all large parties are included in a governing coalition, anti-incumbency voters might not have a viable opposition party to vote for, and instead switch to a new party (Arzheimer and Carter 2006).
We assess the effect of permissive electoral rules which lower the hurdles for new parties and make their success more likely.We operationalise the electoral rules as the country-wide effective threshold of votes needed to gain at least one seat in parliament.The effective threshold is based on a country's district magnitude, as well as its legal threshold (if applicable).We use the log of the effective threshold to account for the diminishing effect of larger districts.
We further include binary variables for regional effects and for the last major economic crisis of 2008.The success of new parties long differed between regions, and the last economic crisis (coded as elections after November 2008) led to electoral turmoil in many established party systems, potentially affecting gains and losses of opposition parties.
Last, but not least we control for the direct effect of presidential elections, which we code as regimes with a popularly elected fixed-term president (for presidential and semi-presidential regimes) (Carreras 2017;Cruz et al. 2016;Elgie 2016), and the number of effective parties at the electoral level (Enep) (Merkel et al. 2018).Both factors may affect opposition gains and losses due to clarity of responsibility and their effect on the number of electoral alternatives.
Based on the previous literature and on our theoretical reasoning, it is possible that these factors moderate the effect of government performance.Accordingly, we have tested whether they moderate the effect of economic performance and corruption, relying on interaction effects (not reported in the results).

Results
When governments are driven out of office, opposition parties benefit.This article analyses who wins when government parties lose, between 1990 and 2014, a period characterised by landslide victories for political outsiders in elections in Italy (2013), Bulgaria (2001 and2009) and Bolivia (2002 and2009).Gains of new parties are clearly no longer a phenomenon of new democracies, but have reached some of the oldest democracies in the world.By distinguishing between different types of opposition parties, we depart from previous literature, which has either focussed on the distinction between government and opposition parties, or between established (within system) and outsider (extra system) parties.
The literature uses different modelling strategies to estimate the effect of retrospective economic and corruption voting, and the gains and losses of parties on the aggregate.We rely here on hierarchical models with yearly fixed effects and electoral gains and losses of new and established opposition parties as the dependent variables.Electoral gains and losses of opposition parties in country j and time-point i are explained by an overall mean (α 0j ), time-dependent country variables and controls (ij), and time-invariant controls (j), unexplained country variation (μ j ) , and unexplained time-variation (ε ij ).

Electoral change of opposition
Results are reported in Tables 1 and 2 below, while Table 3 reports the descriptive statistics.Model 1 reports the total gains and losses of both opposition and new parties jointly to show the overall size of the anti-incumbency change.Models 2-5 then split the vote gains (and losses) into gains and losses of new parties (models 2 and 3), and into gains and losses of established opposition parties (models 4 and 5).
While models 2 and 4 consider all parties participating in the previous elections as 'opposition parties' , models 3 and 5 do so only for parties which were electorally significant, gaining at least 5% of the votes, whereas insignificant parties are considered as 'new' (results remain robust if we lower this threshold to 3% or 1% of the votes).Table 2 is equivalent to Table 1, but reports the effect of rising unemployment instead of negative growth.
If we estimate electoral gains and losses of all established opposition and new parties jointly, we find that non-incumbent parties benefit mainly during times of negative economic growth.Rising unemployment and corruption do not affect the overall vote share of opposition parties (model 1).However, once we disaggregate the opposition vote into new and established opposition parties, it becomes evident that different mechanisms are at play (see Figure 2).Established opposition parties benefit from negative growth (Table 1), 11 while new parties benefit from high corruption and rising unemployment (hypotheses 1a, 1b).More specifically, under negative growth of the economy, voters turn to established opposition parties rather than new parties.For each percentage point of negative growth, they gain  about half a percentage point of votes.Under rising unemployment and high levels of corruption, by contrast, voters support political outsiders.
Our findings on economic growth corroborate findings from cross-country studies, but stand in contrast to the study of individual countries reporting that newcomers benefit from a bad economy (Marcos-Marne et al. 2020; but see discussion on perceptions below).The effect of unemployment is similar to that of growth and corroborates earlier results of comparative research on Central and Eastern Europe (Hanley and Sikk 2016), albeit for a larger set of countries.
We provide for additional analyses, relying on alternative operationalisations of economic crises.First, we reproduce Powell and Tucker's (2014: 125, 138) contested findings about the effect of long-term growth (since transition) 12 for a larger sample of cases (Table 4, specification 2).Our results also point to long-term effects, but this does not alter our main findings about more short-term effects.Further, we test whether extreme economic crises might have similar effects to high levels of unemployment, benefitting political outsiders (Mainwaring et al. 2017).Some of our results point to electoral gains by political newcomers in the aftermath of the global financial crisis of 2008, when interacted with economic growth, or in specifications with non-linear transformations of growth, capturing primarily negative economic growth (Table 4, operationalisations 3 and 4).However, these results are not robust across different specifications of the dependent variable, nor across model specifications, or to the inclusion of further controls.We relegate the question to studies covering more elections held in different periods of economic crises.
With regard to corruption, our results generalise Hanley and Sikk's (2016) findings, which also report a positive effect of rising corruption on new parties' vote gains in a worldwide set of countries (hypothesis 2).We obtain the same results if we use a (time-invariant) variable measuring the level of patronage among a country's political elite 13 instead of corruption (Online Appendix B, Tables B10 and B11).Both corruption and patronage are associated with more important vote gains by new political parties; high levels of patronage also lead to vote losses of established opposition parties.To ensure that our results are not driven by regional effects, we replicate the analyses for a subset of the cases (see next section).
Turning to political legacies, the effect of democratic experience is notable.New parties are more successful (and vice-versa for established opposition parties) in Latin America and Central and Eastern Europe than in Western Europe.Vice-versa, the effect of the age of democracy is mostly driven by collinearity with other variables and regional fixed effects. 14 The remaining control variables have no or very limited effect on gains and losses of opposition parties.We find no or unstable effects for the openness of the electoral system, the presence of presidential elections, the size of the government coalition, or the economic crisis of 2008.
We further tested if context factors moderate the effect of corruption and the economic conditions, but none showed significant effects.Our results are robust if we run models without control variables (see Online Appendix, Tables B1 and B2).standard errors in parentheses: + p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001, election fixed effects included.all measures of growth are inverted.transformed 1: Growth rates above 2% are discounted by 50%.curvilinear 2: logarithmic transformation, where the effect of negative growth is stronger: growth T = (log(growth + 98)-log(100)).100.
Our analysis looks at structural factors and measures that (as far as possible) are based on objective assessments.Needless to say, not only objective indicators of economic performance and evidence of corruption count.Rather, the popular perceptions of corruption and of economic performance are shaped by the public discourse, and politicised by political actors.In particular, the opposition or new parties may aim to politicise corruption to increase its salience when casting a vote (Bågenholm and Charron 2014).Our results speak to some of these arguments: we have reassessed our models with a measure of the perception of corruption in the public sector, by experts and business people (Transparency International 2013).On the one hand, we find that our results are robust regarding corruption and economic growth (Online Appendix B2, Table B12).On the other hand, models measuring corruption perceptions report a lower effect of unemployment (Online Appendix B2, Table B13).Earlier research has shown that unemployment is the most important time-variant factor affecting perceptions of corruption (Melgar et al. 2010).If with growing unemployment rates, new parties more successfully politicise corruption, then unemployment would be partly related to an indirect effect on the success of new parties. 15 However, our research design, covering a large number of elections, structural factors and aggregated perception values does not allow us to take into account political agency or heterogeneous effects among the electorate (cf.Marcos-Marne et al. 2020).We refer to comparative research at the level of parties or individual respondents for an analysis of subjective perceptions of the economy and corruption, and their link to political trust and voting behaviour.
We have controlled for several other factors that are discussed in the literature as affecting the success of new parties (or related types, such as protest parties), but for which we do not have data for the complete sample (results are reported in Online Appendix B1). 16To test for the influence of party-voter linkages we control for the strength of linkages and the extent of party membership (operationalisation and models, see Online Appendix B1, Tables B3 and B4).Stronger party-voter linkages reduce the success of new parties but do not alter the role of growth and corruption.When controlling for party membership, economic growth is no longer significant at conventional levels in some of the models -most likely due to the low number of cases that remain when this variable, rather than the (insignificant) effect of party membership.We further test for effects of the degree of electoral competition and polarisation (Abedi 2002).We include a measure for the issue congruence between political parties and voters and the issue dimensionality in a party system (Online Appendix, Tables B5 and B6).Neither of the two variables affects the main results, nor do they significantly influence the success of new or established opposition parties.Further, we control for the effect of globalisation, a potential mediator of retrospective economic voting (Online Appendix, Tables B14 and B15) (Hellwig and Samuels 2008).

Robustness of the results
In order to assess the robustness of the reported results, we have complemented the analysis with a range of alternative operationalisations of our main variables (Online Appendix B2), and tests on subsets of the data (Online Appendix B3).
First, we confront our results with alternative operationalisations of our main explanatory variables -different measures of patronage and corruption (see above), and different measures of economic performance (inflation and different transformation of growth (see above)).We do not find any retrospective voting effect for inflation (Online Appendix, Table B9), a variable which showed significant results only for very specific samples of countries (Przeworski 1996;Remmer 1991;Tavits 2005).Results for different operationalisations of growth and patronage have been discussed above.
Second, we provide multiple validations for our dependent variable.First, considering that some opposition parties never enter government, we distinguish political outsiders (including parties in permanent opposition) from entangled opposition parties.Results, displayed in the Online Appendix (Table B22), show that corruption benefits political outsiders, though our distinction between new parties and established opposition parties seems more valid for understanding the two types of economic voting. 17Second, we acknowledge the fact that vote shares may be of limited importance in highly disproportional electoral systems.We therefore also count parties with less than 5% of seats (but more than 5% of votes) as 'new' rather than 'established' opposition parties.Our results are robust to this alternative specification of opposition parties' gains and losses (Online Appendix, Tables B20 and B21).Third, we have controlled for alternative operationalisations of incumbency by identifying the main opposition party (Online Appendix B2, Tables B16 and B17), and estimating whether a party spent at least half of the electoral period in opposition (Online Appendix, Tables B18 and B19).
Finally, we were concerned that the effect of the economy and corruption might be driven by regional effects.Our re-analysis on subsets of cases, excluding Central and Eastern European democracies (Online Appendix B3, Table B23), a region notorious for the formation of new parties (Tavits 2005), alters the effect of growth in model 2 (it becomes significant), but does otherwise not affect our main variables.Restricting the analysis to third-wave democracies does not substantially alter our results (Online Appendix B3, Table B24).

Conclusions
A vast research field deals with the electoral consequences of government performance, also termed 'retrospective voting' .The literature on retrospective voting covers primarily the electoral response to economic performance, and more recently increasingly the electoral response to corruption (Curini 2018).
However, we know little about the beneficiaries of this effect: who wins the elections when the incumbent loses?In this article, we investigate the effect of retrospective voting across 59 democracies worldwide, and over 25 years , in reaction to bad economic performance, as well as in reaction to corruption.However, the consequences of these two scenarios are fundamentally different.In the case of economic downturns, electoral change remains mostly limited to the existing party systems -with the exception of times of severe crisis.A negative growth rate of 10% results on average in six to seven additional percentage points for established opposition parties.The electoral consequences of corruption and high unemployment are more fundamental.Corruption not only affects the incumbent, but it is also related to a more fundamental distrust of the political elites on the part of citizens, and the demand for more substantial renewal.Therefore, voters seek new political entrepreneurs outside the established party system.A shift in corruption from the lowest to the highest level results in gains for new parties of about 10 percentage points.Rising unemployment appears to have a comparable effect to corruption, discrediting the whole political elite to the benefit of new parties.
These findings have important implications for our understanding of party system change.Political crises do not simply lead to swings of the pendulum between established opposition and government parties.To the contrary, in times of high unemployment and corruption, voters lose faith in their political representatives and rush to political outsiders to fix the country.While party system renewal is often associated with young democracies and unconsolidated party systems, our results on retrospective economic and corruption voting equally apply to established democracies.However, our results also call for new comparative research based on individual data, linking the perception of the economy, corruption, political trust and electoral behaviour over a large number of democracies, and for comparative research into the politicisation of retrospective valence issues.
Our results offer only first indications of who the beneficiaries of retrospective voting are, as our definition of new parties includes anti-establishment parties, parties at the political poles, or new centrist parties.Earlier work has shown that some of the variables which we associate with the loss of political trust, have a similar effect on different (partly overlapping) types of political outsiders, newcomers and/or anti-establishment parties (Algan et al. 2017), but not for extremists.While this follows our distinction between retrospective vote issues with programmatic implications, and others related to (the loss of) political trust, systematic cross-country studies, comparing the effect for different types of outsiders would be desirable.
Last but not least, depending on who benefits from the retrospective vote, the consequences for the quality of democracy might differ.Political outsiders might be a threat to the stability of political institutions.However, in democracies that suffer from endemic corruption and political patronage, 18 the renewal of the political elites opens opportunities to break political patronage networks.

Notes
1.Such as coalition governments where voters cannot easily assign responsibility (Tavits 2007), or party systems with limited choice (Schleiter and Voznaya 2014).2. The effect differs considerably between partisans of the incumbent and of the opposition (Anderson and Tverdova 2003).3. The dataset is available from www.bochsler.eu/data.4. Similar: van de Wardt et al. (2021).5. Effects remain stable for models with a 1% or 3% threshold.6.The measure correlates at r = 0.88 with the index only for new parties; the respective measures for established oppositions (with/without threshold) correlate at 0.91.For a few elections, the net electoral gains/losses for new parties, including previously irrelevant parties, result in negative values.One example thereof is the Albanian 2009 elections, where very small parties disappeared from the elections due to a change in the election rules; another pattern emerged in 1999 in Switzerland, where small far-right parties lost most of their votes to a large party (the Swiss People's Party).7. We code government participation 90 days prior to parliamentary elections in order not to consider very short technical governments as incumbents.8.The recent literature shows that changes in the rate of unemployment, compared to the previous year (Hernandez and Kriesi 2016;Stanig 2013) are more important than annual unemployment rates.Dynamic measures also fit best with the individual perception of the economy (Nadeau et al. 2013).Studies of economic voting at the individual level tend to ask about economic improvements or comparisons over time (Healy and Malhotra 2013).We also run our models with inflation as a measure of economic performance (Online Appendix B2).Earlier findings report a significant effect of inflation on retrospective voting for a limited set of countries in Latin American democracies (Remmer 1991), whereas Przeworski (1996) finds an opposite effect for Poland, and Tavits (2005) a decreasing effect for older democracies.9.In particular, we have tested whether the (negative) effect of the economic crisis on incumbent votes is more prevalent than the (positive) effect

Figure 1 .
Figure 1.De-composing electoral change: opposition gains, and newcomer party gains.By period.notes: Graph based on 0% threshold for new parties.Boxplots contain 50% of the observations within the grey box, with the white line representing the median.Whiskers represent the 'minimum' and 'maximum' of the data, excluding outliers (indicated by the dots).

Table 1 .
explaining opposition gains and losses in elections.

Table 2 .
explaining opposition gains and losses in elections.

Table 3 .
Descriptive statistics of the main variables used in the main analysis.
Figure2.explaining gains and losses in elections, main regression results.notes: coefficient plot of main regression results from tables 1 and 2, models 2 and 4. coefficients with 90% confidence intervals.

Table 4 .
the effect of economic growth and crises -different operationalisations.
Daniel Bochsler is an Associate Professor of Political Science and Nationalism Studies at Central European University (CEU), and Professor of Political Science at the University of Belgrade.His research deals with political institutions and regimes, elections and political parties, especially in new democracies in Central and Eastern Europe.[BochslerD@ceu.edu]Miriam Hänni is a senior researcher at the Federal University for Vocational Education and Training.Her research includes studies on party competition, responsiveness and vocational education and training.She has recently published in Education Sciences, European Journal of Political Research and Legislative Studies Quarterly.[Miriam.Haenni@ehb.swiss]