Introduction
One of the best-established propositions in the literature about representation in contemporary democracies is that “accountability, if it is to be properly exercised, first requires citizens to make attributions of responsibility” (Rudolph, Reference Rudolph2003b: 700). With the intensifying global trend towards decentralization over recent decades,Footnote 1 increasing attention has been devoted to how voters attribute responsibilities between levels of government within states.
Roughly speaking, there have been two main research questions on the attribution of political responsibilities: to what extent multilevel governance undermines clarity of responsibilities and hence affects economic voting (for instance, Anderson, Reference Anderson2006; Cutler, Reference Cutler2004; Hobolt and Tilley, Reference Hobolt and Tilley2014; León, Reference León2010), and more recently, under which circumstances citizens are more likely to learn about responsibility attribution over time (León, Reference León2012). We focus in this paper on how people learn, as the scant existing research is, in our view, subject to several substantive and methodological shortcomings. The lack of a model of learning grounded in social psychology and cognitive science does not allow us to account for differences in learning across policy areas and individuals: not all individuals learn in the same way about the same things.
Multilevel governance improves the quality of public policy and reduces the administrative costs by decreasing the distance between government and citizen as well as enhancing governmental understanding of citizen preferences (see Escobar-Lemmon, Reference Escobar-Lemmon2003; Weingast, Reference Weingast1995). However, if the electoral control of incumbents falters as a result of an inability to assign responsibility in multilevel governance, then incumbents do not really fear losing the next election and then they will not be representative (Anderson, Reference Anderson2006). Given this potential trade-off between efficiency and accountability in multilevel governance in the short term, the long-term outputs of decentralization crucially depend on understanding how people learn. If citizens are able to learn about political responsibilities, then the trade-off progressively disappears as time goes by.
We argue that the accuracy of individuals' attribution of responsibilities depends on the attention they pay to political issues, particularly those individuals who derive benefits from having information about specific policy areas. Apart from group-serving biases and institutional designs, we show that there is a third variable affecting responsibility attribution: the saliency of issues at a specific point in time. That is, differences in the attribution of responsibilities can be found not only across individuals and institutional settings, but also over time.
We focus on Spain as an example of a decentralized democracy. Spain is a quasi-federal parliamentary democracy (Linz, Reference Linz, Goldwin, Kaufman and Schambra1989) in which the regional level of government holds significant competencies in the welfare state, such as education and health care, while the central level maintains other competencies, such as the administration of justice, defence or immigration. The recent process of devolution resulted in a fragmentation of competencies which has diffused the attribution of responsibility by citizens. The economic crisis in Spain also provides an ideal scenario for testing our argument about learning as the saliency of policies has dramatically changed in only a few years. The spectacular increase in the unemployment rate, from 8.3 per cent in 2007 to 25.0 per cent in 2012, accompanied by the significant reduction in the importance of previously crucial issues, such as immigration or housing for instance, allows us to test whether the public interest in issues that arise and individuals' self-interest drive the attribution of responsibility.
We rely on individual data from two surveys in late 2007 and 2012 to disentangle the mechanisms that account for learning about responsibility attribution in multilevel government systems. Our results show that the (unintended) positive consequence of the current economic crisis and the enormous increase in unemployment rates is that citizens are now more able to accurately attribute the responsibility for political decisions in relation to this policy than some years ago. Learning is particularly significant among those most affected by the economic crisis, the working population. This is the bright side of the economic crisis.
The rest of the article is organized as follows. In the next section, the previous literature and our argument about citizens' learning about responsibility attribution are presented. The third section describes the data and methods. The fourth section discusses the results of the empirical analysis. Section five presents our conclusions and suggestions for further research.
Arguments
When explaining the outputs of institutions (that is, how they affect the behaviour of political actors), learning is crucial. Institutions matter when the actions of citizens or elites are driven by the incentives provided by the rules of the game. Of course, a necessary condition for this effect of institutions is that actors are aware of those incentives. Therefore, the expected outputs of institutions are often not observed immediately but in the medium- or the long-term, once actors have good information. For instance, the electoral systems literature indicates that the strategic behaviour of parties and voters produces a long-term equilibrium relationship between the size of party systems at the electoral and legislative levels (Cox, Reference Cox1997). Since actors' instrumental rationality is a standard concern, differences over time within countries in the level of co-ordination or the speed at which the equilibrium is reached are a function of the quality of the information about the electoral prospects of competitors. This is what Tavits and Annus (Reference Tavits and Annus2006) define as the developmental argument of strategic voting (that is, the strategic behaviour of voters increases with time as a result of a learning process).
The literature on attributing responsibility correctly in multilevel states has shown that citizens' perceptions of political responsibility are influenced by group-serving biases—partisanship—and the institutional context, particularly the division of responsibilities. First, information about government responsibility is not randomly distributed among individuals. Individual-level differences are related to voters' economic ideologies and party identification (Rudolph, Reference Rudolph2003a, Reference Rudolph2003b). More recently, Tilley and Hobolt (Reference Tilley and Hobolt2011), Hobolt and colleagues (Reference Hobolt, Tilley and Wittrock2014) and Hobolt and Tilley (Reference Hobolt and Tilley2014) have used the concept of group-serving biases to capture the individual predispositions shaping attributions of responsibilities: individuals tend to give credit to the groups they favour for positive outcomes and to blame rival groups or an exogenous constraint for negative outcomes (Hobolt et al., Reference Hobolt, Tilley and Wittrock2014: 154).
Second, institutional arrangements have a crucial role in shaping attributions of responsibility. As is well known in the literature about electoral accountability (see Anderson, Reference Anderson2007a; Reference Anderson, Dalton and Klingemann2007b), the connection between voters' perceptions of the economy and their votes depends on the extent to which structural features of polities can hinder voters' access to information about representatives' activities. The general argument, based above all on Powell and Whitten (Reference Powell and Whitten1993), is that institutions increase or reduce citizens' capacity to reward or punish incumbents. Once voters have formed their opinion about the state of the economy, the transformation to a vote in favour or against the incumbent depends on (or interacts with) what they believe is the responsibility of the incumbent. Meanwhile, certain institutional measures result in a greater “clarity of responsibility” of governments, expressed in the electoral results of the politicians who introduced these measures. And given the difficulty in deciding whom to reward or punish when this clarity does not exist, voters' ability to rely on their own evaluation of economic results is undermined. Thus, the clearer the responsibility of the government for the results of its own policies, the more the economy affects the electoral support of incumbents. Anderson summarized the evidence in the following way: “These findings suggest that voters' ability to express discontent with economic performance is enhanced when accountability is simple. Voters' economic assessments have stronger effects on government support when it is clear who the target is, when the target is sizable, and when voters have only a limited number of viable alternatives to throw their support to” (Reference Heck and Scott2000: 168). More recently, it has been shown that when there is a dispersion of political authority among multiple levels of government, the clarity of responsibility is diminished (Anderson, Reference Anderson2006; León, Reference León2010, Reference León2012). When there is only one government (national), almost all the responsibility is in its hands. However, if there are also sub-national governments, the assignment of political responsibility is no longer direct; in this instance, voters need to know whether the competence in each matter is shared or exclusive. In two recent academic studies, Duch and Stevenson (Reference Duch and Stevenson2013) and Duch et al. (Reference Duch, Przepiorka and Stevenson2015) show that individuals have general responsibility attribution heuristics that apply to collective decisions; individuals tend to assign responsibility to the decision maker with agenda power and with the largest vote share.
In Spain, a substantial body of literature in the last ten years has investigated the extent to which individuals are able to correctly assign political responsibilities to national and regional incumbents (Herrero et al., Reference Herrero, Goenaga and Ruiz-Huerta2015; Lago and Lago, Reference Lago and Lago2010, Reference Lago and Lago2013; León, Reference León2010, Reference León2012, Reference León2015; León and Ferrín, Reference León and Ferrín2007; López and Rodrigo, Reference López and Rodrigo2012, Reference López and Rodrigo2014). Roughly speaking, the main findings are fourfold: (i) when asked about the level of government with the main responsibility over seven policies (unemployment insurance, health care, education, retirement and disability pensions, transports, and infrastructures), the average number of correct answers is three out of seven attributed responsibilities; (ii) unemployment insurance and pensions (that is, those exclusively in the hands of the national government) are the policies with the highest percentage of correct answers; (iii) there is some evidence of learning over time, particularly in the case of those individuals who are highly informed and interested in politics, and (iv) the relationship between decentralization and clarity of responsibility has a u-shape: responsibility attribution is clearer in regions with high and low levels of decentralization (that is, where one level of government clearly predominates over the other) than in regions with a more intertwined distribution of powers.
However, when accounting for learning about responsibility attribution in multilevel structures, there is no model explaining why some individuals, but not others, are able to learn who is responsible for what, or why individuals learn about some policies but not all of them. The (flawed) assumption is that learning is automatic and passive. This is particularly relevant when there is no common pattern of learning across policy areas depending on the level of government with the primary responsibility or the ideology of the incumbent, as in our case. The information hypothesis—which states that those actors who have greater access to information, who are more educated and who are more sophisticated, should make more accurate responsibility attributions—can hardly explain differences over time and across policy areas for the same individual. We need a theory to explain when people learn and which people learn.
According to the model of human decision making by Lupia and McCubbins (Reference Lupia and McCubbins1998: ch. 2), human learning is not automatic, but active. The prerequisite for learning anything is paying attention; as learning requires effort and effort is a scarce resource for everyone, people choose what and when to learn. Additionally, the purpose of paying attention is to make reasoned choices (that is, people are goal oriented). Therefore, people will pay attention only to those stimuli that are easy to process and strongly associated with the greatest avoidance of pain or the greatest production of pleasure.
There are two substantive implications of this model. First, if attention is a necessary condition for learning, then learning about political responsibilities is only possible in those issues or policy areas whose saliency increases over time. All else being equal, as the saliency of policies varies over time, the attention that people pay to policies should also change. Our assumption is that the saliency of an issue is determined by the changing economic, political or social circumstances and therefore, it is exogenous to individuals and the mass media. Second, there are no reasons to expect that all individuals will pay the same amount of attention to a particular policy area. Similar to what Duch and colleagues (Reference Duch, Palmer and Anderson2000) argue when explaining perceptions of national economic conditions, individuals who derive greater benefits from having information about specific policy areas tend to have a better understanding about responsibility attribution between levels of government. In our paper, people out of work can be expected to pay a greater amount of attention to the decisions about unemployment. Not surprisingly, self-interested attitude—defined as one that is instrumental to the individual's attainment of valued goals, particularly those which bear directly on the material well-being of individuals' private lives (Sears et al., Reference Sears, Lau, Tyler and Allen1980: 671)—is a classic determinant of policy attitudes. Thus, following again Lupia and McCubbins (Reference Lupia and McCubbins1998: ch. 2), it can be hypothesized that there is a positive interaction between self-interest indicators (that is, the expected individual benefit from paying attention to an issue) and the saliency of the issue (that is, the cost needed to process a stimulus into a useful inference).
However, as explained by Lau and Heldman (Reference Lau and Heldman2009: 524), figuring out what one's self-interest is on many political issues can be daunting even for the most politically sophisticated individuals. Therefore, self-interest effects might be limited to the most attentive members of the public: when an issue is salient, it is easier for individuals to make the link between their personal circumstances and the actions of politicians. The “heterogeneous attribution” theory by Gomez and Wilson (Reference Gomez and Wilson2001) and the empirical evidence provided by Funk (Reference Funk2000) also go in this direction.
Our theoretical expectations specify that individuals' learning about responsibility attribution varies depending on the amount of attention they give to the political issue, and their social status or occupation. The hypotheses are as follows:
• When the saliency of an issue increases over time, the accuracy of individuals' attribution of responsibilities does so accordingly. We label this the saliency hypothesis.
• As self-interested individuals seek out information that reflects their economic circumstances, economic self-interest will increase the accuracy of individuals' attribution of responsibilities. We label this the self-interest hypothesis.
• When the saliency of an issue increases, learning should be particularly important for those individuals who derive greater benefits from having information. We label this the interactive hypothesis.
Data and Methods
Our analysis is based on individual data from Spain from two points in time: late 2007, the last year before the economic crisis, and late 2012. Two substantive and methodological reasons explain this decision. First, Spain is the country in the European Union (EU) that has suffered the worst consequences of the current economic crisis in terms of employment levels. According to Eurostat,Footnote 2 Spain was the country showing the highest unemployment rate in 2012 in the EU-27 (25.0%) and, more importantly for our purposes, the highest increase in the unemployment rate between 2007 (8.3%) and 2012. This difference of 16.7 percentage points more than triples the average increase in unemployment rate (4.8 points) in the EU-27.
Not surprisingly, this trend in the unemployment rate is clearly reflected in the different saliency of public issues in the pre-crisis and the current scenarios. Figure 1 shows the main problems in Spain according to respondents in monthly barometers by the Centro de Investigaciones Sociológicas (CIS) from 2007 to 2012. As can be seen, the share of respondents who thought that unemployment was the main problem of the country has doubled in five years, from 40 per cent to 80 per cent. The changes in the saliency of the issue of unemployment clearly respond to the evolution of the unemployment rate (see Figure 2). Leaving aside the catch-all category of “economic problems,” the saliency of all the remaining issues has clearly dropped from when they were relevant before the economic crisis (immigration and housing) or do not show significant changes when they were not (health care or education).Footnote 3
Figure 1 Main problems for respondents in monthly barometers in Spain, 2007–2012
Figure 2 Saliency of unemployment in monthly barometers and unemployment rate in Spain, 2008–2012
Second, the face-to-face survey interviews conducted by the Centro de Investigaciones Sociológicas, before the economic crisis in 2007, and five years later in 2012, are particularly well-suited to examining the learning of responsibility attribution. The first study (CIS 2734) was conducted in October and November 2007 in five regions. A representative sample of 1,496 people were interviewed in Andalusia, 1,490 in Castile and Leon, 1,500 in Catalonia, 2400 in Galicia and 1,491 in the Basque Country. The second study (CIS 2956) was conducted between September 13 and October 9, 2012, in the same regions: 1,430 people were interviewed in Andalusia, 965 in Castile and Leon, 1,190 in Catalonia, 585 in Galicia and 425 in the Basque Country. The samples are also representative. More interestingly, the questionnaires of both surveys include a similar question tapping respondents' attribution of responsibilities. In the 2007 survey, the question was: “Which is the most responsible level of government (central government, regional government or local government) for the administration of the following services?” In 2012, the question was slightly different: “Which is the most responsible level of government (central government, regional government or local government) if things go well or badly in the following policy areas?” The individuals from the five regions have been pooled, as our arguments about learning should travel across regions. Accordingly, the samples in both years have been weighted to bring regional sample proportions in line with national population proportions based on census data.
The dependent variable is the accuracy of individuals' attribution of responsibility on unemployment. If respondents correctly identify the most responsible level of government in the area (national level), the value of the dependent variable is 1; for both the incorrect attribution of responsibility and don't know/no answer the value of the dependent variable is 0.Footnote 4 Given that the dependent variable is dichotomous, we can run binomial regression models.
Roughly speaking, self-interest can be measured with objective or exogenous indicators (that is, objective measures that distinguish between those who stand to benefit and those who stand to lose for a given policy) and subjective or endogenous indicators (that is, judgments about whether people would personally be harmed or benefited by a public policy given their own account of what constitutes a cost or a benefit to them). Unfortunately, in the two questionnaires that we are using in the empirical analysis, there were no questions asking respondents to indicate whether the policy would have or has had either positive or negative consequences for themselves personally. Therefore, we have to use a crude proxy for self-interest based on the findings of previous research. In our study, self-interest on unemployment concerns whether or not the respondent was part of the working population,Footnote 5 as Sears and colleagues (Reference Sears, Lau, Tyler and Allen1980: 673) and Lau and Heldman (Reference Lau and Heldman2009: 517) do.
There are two key independent variables. First, in order to capture learning about the attribution of responsibility over time, we have created a dummy variable, 2012 year, that equals 1 for respondents in the 2012 survey and 0 for respondents in the 2007 survey. Given the enormous increase in the unemployment rate in Spain, we expect a very significant increase in the accuracy of responsibility attribution for this policy. Second, in order to test the self-interest hypothesis, we have included dummy variables that equal 1 for those individuals who belong to the group with a particular self-interest in unemployment, and 0 otherwise.
Relying on existing research (for instance, Hobolt and Tilley, Reference Hobolt and Tilley2014), we have included five individual-level control variables that capture perceptual biases and informational differences across individuals. Roughly speaking, it should not only be individuals with greater access to information but also, given the negative economic situation, opposition partisans who make more accurate attributions of responsibilities. The variables are the following:
• Male is a dummy variable that equals 1 for men and 0 for women. According to the previous literature, the variable should enter positively in the model as men tend to show higher levels of political knowledge than women (see Fraile, Reference Fraile2014).
• We expect a positive relationship between age and political knowledge, although a decline in this relationship has been demonstrated in advanced ages in previous works (Lau and Redlawsk, Reference Lau and Redlawsk2009). For this reason, Age (in years) and Age (squared) (to capture a potential non-linear relationship) are included in the models. The expected signs are positive and negative, respectively,
• Education is a categorical variable that equals 1 for those individuals with no education or primary education, 2 for those respondents with secondary education and 3 for those individuals with university education. As education is an important determinant of political knowledge, the variable should enter positively in the model.
• Political awareness is a dummy variable that equals 1 if the respondent knows the name of the corresponding regional prime minister, 0 otherwise. The expected sign is positive.
• Voting behaviour is a variable that distinguishes between government supporters (the reference category), opposition supporters and abstainers according to respondents' voting behaviour in the last national election.Footnote 6 Due to the lack of a variable measuring party identification in the questionnaires, voting behaviour is a proxy for partisanship. As shown by Tilley and Hobolt (Reference Tilley and Hobolt2011), when government partisans have a negative view of the economic situation, they tend to think that the government is not responsible. Considering the adverse economic situation, opposition supporters should make more accurate responsibility attributions than government supporters, while the expectation is not clear in the case of abstainers.
• Finally, we also include a variable that controls for the region in which respondents live to capture institutional and economic differencesFootnote 7 (see León, Reference León2010, Reference León2012).
The descriptive statistics of the independent variables are shown in Table 1.
Table 1 Descriptive statistics-independent variables
Source: Studies 2734 and 2956 of the Centro de Investigaciones Sociológicas (CIS).
The purpose of this paper is to analyze to what extent responsibility attribution changed between 2007 and 2012. Unfortunately, panel data observing the same set of individuals over several years are not available. Therefore, we employ two cross-section surveys in which the same questions about responsibility attribution are asked to different samples of individuals from the same population. A pooled cross-sectional analysis (Firebaugh, Reference Firebaugh1997) allows us to track trends in the accuracy of responsibility attribution between levels of government and to establish causal inferences. We have pooled the 2007 and 2012 cross-sectional surveys. The dummy variable (2012 year), which identifies the respondents from each survey, captures whether there are differences in responsibility attribution over time controlling for the same variables in the two years. The interaction of the 2012 year with the independent variable of interest shows whether the impact of the latter is significantly different in the two moments in time.Footnote 8
Individuals might perceive that the role of the central government has strengthened as a consequence of the economic crisis, which would make our argument spurious. To counter this and to support our argument, two policies in the hands of the national government are considered in the empirical analysis; unemployment and immigration. If the crucial mechanism is the role of the national government, we should observe that the accuracy in responsibility attribution increases in both policies; but if the mechanism is simply the saliency of issues, it should only be in the case of unemployment that individuals have more information in 2012 than in 2007 about who is in charge. What we found is the latter and this clearly supports our argument.
Results
We start our discussion of the results by looking at the distribution of our dependent variable. The accuracy of responsibility attribution in unemployment (and on other three policies in which the interest has decreased)Footnote 9 is displayed in Table 2. The only area in which the share of respondents correctly identifying the most responsible level of government has increased over time is unemployment (national level): accurate attributions are 14 points higher in 2012 than in 2007. On the contrary, in the other three policy areas, especially in health care and immigration, the attribution of responsibility is less accurate in 2012 than in 2007. The understanding of how responsibilities are allocated between levels of government is worse in these three areas in 2012 than it was five years earlier. In other words, the level of government with the primary responsibility (the same in both years) or the ideology of the incumbent (different at least at the national level in the two moments), does not explain differences over time among policy areas. These results strongly support the saliency hypothesis.
Table 2 Accuracy of responsibility attribution: descriptive statistics*
*In each cell, the share of respondents correctly attributing the responsibility and, in parentheses, the number of respondents.
Source: Studies 2734 and 2956 of the Centro de Investigaciones Sociológicas (CIS).
To explore the robustness of the initial findings and to test the saliency, self-interest and the interactive hypotheses, we include the two key independent variables and the controls in our logit regression models. We run two specifications: an additive model with the dummy variable identifying the observations corresponding to 2007 and 2012, the self-interest variable and the control variables (model A); and an interactive model in which an interaction term—between 2012 year and self-interest—is added to the previous specification (model B).Footnote 10
The results when explaining the attribution of responsibility in unemployment are displayed in Table 3. First, the additive model (A) confirms that learning depends on the saliency of issue. As expected, we found evidence in favour of learning about responsibility attribution for unemployment: 2012 year is positive and statistically significant at the 0.01 level. This means that the accuracy of individuals' attribution of responsibility is higher in 2012 than in 2007.Footnote 11 The saliency hypothesis is clearly supported. The coefficient of self-interest in the additive model is not statistically significant. This statistically insignificant impact of self-interest is clearly in line with previous research. For instance, Sears and colleagues “found self-interest to have little effect on voters' policy preferences, while symbolic attitudes had major effects … But only four of our thirteen self-interest indicators had even a statistically significant effect” (Reference Sears, Lau, Tyler and Allen1980: 773). Similarly, in the more recent paper by Lau and Heldman (Reference Lau and Heldman2009: 515), they find that “on average across the four issue domains, the various self-interest indicators explained only 1 per cent of the variance over that already explained by symbolic beliefs, while the symbolic beliefs collectively explained 10 times more of the explainable variance in policy attitudes above that already accounted for by self-interest.”
Table 3 Accuracy of responsibility and attribution on unemployment: regression models
Note: Robust standard errors in parentheses.
* p ≤ 0.05. ** p ≤ 0.01.
Source: Studies 2734 and 2956 of the Centro de Investigaciones Sociológicas (CIS).
On the other hand, the variables political awareness and voting for the opposition have the expected signs and are significant, increasing the accuracy of individuals' attribution of responsibility. The squared transformation of the variable age shows that the older citizens tend to make incorrect attributions of responsibility. Apart from the regional dummies, all of them are statistically significant in our models. However, gender, age, and education are not statistically significant.
Second, when explaining the attribution of responsibility for unemployment, the individuals who were in the working population have learned more about which level of government makes the decisions in this area than the rest of the population. The interaction is statistically significant at the 0.01 level.Footnote 12 In other words, as expected in our interactive hypothesis, learning is higher for those who derive greater individual benefits from having information when the issue is salient.
Based on the results of the interactive model in Table 3, Figure 3 shows the marginal effect of the saliency of unemployment on the accuracy of responsibility attribution in this policy area. It is not only those individuals who derive greater benefits from being informed about unemployment (the working population) but also the rest of the population who have a better understanding of responsibility attribution in 2012 than in 2007. However, learning is more substantial for the former than for the latter.
Figure 3 Marginal effect of the saliency of the unemployment issue on the accuracy of responsibility attribution
Conclusions
Learning is crucial for understanding the long-term impact of institutions. Given the global trend towards decentralization, it is essential to clarify whether learning about “who is responsible for what” can affect the well-known trade-off between efficiency and accountability in multilevel states. The ability of voters to correctly assign responsibility is conditio sine qua non for holding government accountable for action and outcomes and then to make them representative.
As there are no reasons to expect that all individuals learn the same about the same things, we have argued in this paper that learning is driven by the attention that people pay to issues and their (economic) self-interests. This is what the models in social psychology and cognitive science state. The current economic crisis in Spain provides an ideal scenario for testing our argument about learning with observational data. The spectacular increase in the unemployment rate, from 8.3 per cent in 2007 to 25.0 per cent in 2012, has affected the saliency of issues between 2007 and 2012. Additionally, it is possible to identify a group with a self-interest: the working population tend to have a better understanding of how decisions are made in that area.
Using a pooled cross-sectional analysis with individual data, we find support for our argument. We have shown that individuals have a better understanding of responsibility in 2012 in unemployment than some years ago. This is clearly the consequence of the very different saliency of a policy area in two moments in time. Additionally, learning is particularly important for the working population. This means that self-interest constitutes a source of heterogeneity in the accuracy of responsibility attribution. The unintended consequence of the deep economic crisis is Spain has been an improvement in the ability of citizens to correctly assign responsibility to incumbents for managing the economy. In other words, the interaction between citizens and institutional arrangements are affected by exogenous shocks such as an economic crisis.
Of course, more investigation should be done to find out the consequences of economic crises on the attribution of political responsibilities. There are three avenues worth exploring for understanding the role of exogenous shocks. First, using panel data instead of repeated surveys is crucial for clarifying who learns and who does not. Second, (lab or survey) experiments are a fruitful avenue to explore how individuals learn to assign responsibilities. Third, it can be hypothesized that individuals' learning about responsibility attribution varies depending on the longevity of the democracy in the state where they live and the decentralization process together with the degree to which the power is shared between national and subnational incumbents.
APPENDIX
Table 4 Difference (in points) in the accuracy of responsibility attribution by Autonomous Communities (2012–2007)
Table 5 Accuracy of responsibility and attribution on three additional policies: Regression models
