Abstract
Relying on data from a natural experiment in Spain, I produce an unbiased estimate of the extent to which strategic voting occurs in multi-member districts. I show that voters have fully adapted to the different incentives provided by distinctive electoral systems in Spain since the first election and also that they behave strategically only when the opportunity to do so is present. That is, contamination effects do not seem to exist when voting strategically.
Introduction
District-level strategic voting – a vote for a party (candidate) other than the most preferred one in order to avoid wasting the vote on hopeless candidacies in a district – is a crucial but slippery phenomenon in mass elections. First, it affects the party system. As is well known, strategic voting imposes an upper bound, M +1, on the effective number of competitors that will appear in equilibrium, where M is the number of seats to be filled in a district (Cox, 1997). However, measuring strategic voting is not a simple endeavour, given that it entails comparing the actual (or stated) vote with the likely and unobservable vote in a counterfactual situation where incentives for behaving strategically are absent. That is, the so-called fundamental problem of causal inference is particularly acute here (Holland, 1986).
The three existing approaches to measuring strategic voting, the aggregated inference methodology, the direct individual methodology and the indirect individual methodology (Alvarez and Nagler, 2000; Blais et al., 2005), differ in how they obtain estimates of the true rank-ordering of parties or candidates for each voter (i.e. the sincere vote). And the crucial consequence is that, with few exceptions (for instance, Blais et al., 2005), there is a great deal of variation in the estimates of strategic behaviour reported in the literature. For instance, in the 1987 British general election, an important case study for scholars interested in the problem of strategic voting in multiparty elections (Galbraith and Rae, 1989), strategic voting ranges from a low of 6.3 percent reported by Evans and Heath (1993) to a high of 17 percent reported by Niemi et al. (1992). 1 Although much less is known about strategic voting in multi-member districts (Cox and Shugart, 1996; Gschwend, 2007; Lago, 2008), it could be argued that things are even more difficult in Spain. District magnitude is correlated with urbanization or third-party membership (Monroe and Rose, 2002) and then a problem of severe multicollinearity arises to disentangle the true impact of district magnitude on voting behaviour.
In this article, I use unique data from a natural experiment – ‘a situation in which an intervention of “nature” approximates the property of a controlled experiment’ (Robinson et al., 2009: 342) – in the region of Andalusia, Spain, to measure the extent of strategic voting in multi-member districts. Elections in Andalusia have two particularly interesting features. First, in national and regional elections, the electoral system is exactly the same: seats are elected from the same eight multi-member districts using the D’Hondt method of proportional representation with closed party lists for each of the provincial-level electoral districts and with a formal threshold of 3 percent of valid votes at the district level. However, there is a crucial difference between both electoral systems. In regional elections, district magnitude is greater than 5 – the cutoff point to observe strategic voting in multi-member districts – in the eight districts, and the third national party, the Communist Party (the PCE) and its heir, the United Left (the IU), is a viable competitor everywhere. On the contrary, in national elections there are two districts with a magnitude of 5 and the PCE/IU has never gained a seat there in the 10 elections held until now; whereas in the 6 remaining districts, the magnitude is greater than 5 and the IU has gained seats there in at least one of the elections. Second, in six of the eight regional elections held in Andalusia since 1982, the regional and national elections were concurrent. In sum, we are holding everything constant while we are changing the value of the key causal (or treatment) variable, which is district magnitude or the institutional incentives for strategic voting.
In dwelling on this natural experiment, I make three contributions. First, the natural experiment allows us to produce unbiased estimates of the extent to which strategic voting occurs in multi-member districts. Second, taking advantage of the fact that Spain is a young democracy, I revisit the developmental argument of strategic voting (Tavits and Annus, 2006) (i.e. the strategic behaviour of voters increases with time as a result of a learning process). Third, a surprising omission in the literature on electoral systems is the consideration of ‘contamination’ or ‘interaction’ effects on the micro-foundations of the district-level logic of strategic voting. Similar to the well-known presidential coat-tails effect, it can be hypothesized that a coat-tails effect from the more important arena, the national, to the less-important one, the regional, exists. That is, voters in regional elections might make up their minds on which party or candidate to vote not on the basis of the institutional incentives they face in the election at hand, but on the basis of the incentives provided by national elections. If voters do not fully adapt to the different incentives provided by electoral systems and use the information about the electoral prospects of competitors in the more important arena in the less important election as well, they could vote strategically in both arenas, although strategic voting is unnecessary in the latter. This argument is tested in this article. In the end, my estimation reveals that about 9 percent of voters cast a strategic vote when presented with the opportunity to do so; additionally, this strategic behaviour apparently does not increase with time and does not contaminate electoral arenas in which voters do not have the opportunity to behave in a strategic fashion.
This article is therefore a meeting point between Fiorina’s substantive claim that ‘ticket-splitting and other aspects of the relationship between voting for different offices should become a central focus of research, rather than a sideline’ (1992: 130) and Shugart’s methodological proposal to use crucial experiments to accumulate knowledge in electoral studies (2005: 34).
The rest of the article proceeds as follows. The second section presents the theoretical arguments on strategic voting, particularly in multi-member districts. The third section describes the natural experiment, the data and the methods. Then, the results of the empirical analysis are shown and discussed. The last section concludes.
Strategic voting and past studies
Strategic voting is a vote for a party (candidate) that is not the preferred one, motivated by the intention to affect the outcome of the election. Hence, two conditions must be met for a vote to be strategic: the voter must vote for a party (candidate) other than the most preferred one and must have come to that decision because of perceptions of the likely outcome of the election (Blais et al., 2001: 344).
There are two key conditions for the district-level logic of strategic voting: voters are short-term instrumentally rational (i.e. they care only about which party or candidate wins the seats in their district in the present election) and they have rational expectations. When voters have rational expectations, (1) they agree on how preferences are distributed in the electorate, (2) they agree on what share of the vote each party or candidate will likely get and, therefore, (3) they are able to distinguish between the expected last winner of the last allocated seat and the expected first loser (Cox, 1997: chs 4 and 5). Since actors' instrumental rationality is a standard concern, particularly in rational choice theory, cross-national and across district variation in the pervasiveness of strategic voting is a function of the quality of the information about the electoral prospects of competitors. 2
The predominant argument in the literature is that strategic voting should decline as voters' expectations about which party or candidate will win or lose are less clear and coordinated. Electoral volatility, polls and, above all, institutional incentives are crucial for satisfying this informational prerequisite. Electoral systems scholarship agrees that the larger the district magnitude, the smaller the voter percentages that separate winners from losers. Hence, the harder it is to be sure which party or candidate is ‘out of the running’, and in these conditions there are no clear incentives to desert (minor) parties or candidates. As a result, in PR systems with large district magnitudes, strategic voting should not be an empirically relevant phenomenon. Based on the quality of the information, this argument does not provide a very precise idea about when strategic voting ought to occur less, as Cox (1997: 100) recognizes. However, the threshold seems to be above the magnitude of 5, according to empirical regularities based on Japanese, Colombian and Spanish district-level results: strategic voting fades out when district magnitude is greater than five seats, because it gets too difficult to satisfy informational requirements (Cox, 1997: ch. 5; Cox and Shugart, 1996). Therefore, M>5 has been selected as the cutoff point to observe or not strategic voting in the empirical analysis. 3
This process to generate expectations is combined with three approaches to measuring strategic voting (Alvarez and Nalger, 2000; Blais and Nadeau, 1996; Blais et al., 2005). The first approach consists of deducing the presence of strategic voting from aggregated electoral outcomes; in particular, direct district-by-district estimates of strategic voting and the pattern of inter-election vote shifts in different types of constituencies. Based on individual data, there are two other approaches. The indirect approach is to model how perceptions of the race affect vote choice to deduce how many voters would have voted differently if their choice had not been influenced by strategic considerations. The direct approach consists of ascertaining how many voters both voted for a party other than their most preferred and perceived the former to have better chances of winning than the latter.
These three methodologies all suffer from different flaws or biases when obtaining measures of the true rank-ordering of parties or candidates for each voter: the sincere vote.
4
As Alvarez and Nagler (2000: 60) summarize: [T]he implication is that the empirical literature has come up with very different estimates of the extent of strategic voting, even from researchers examining the same election in the same country. We believe that the reason for these divergent results lies in the different methodologies used in these studies to measure or to estimate the extent of strategic voting.
Data and methods: A natural experiment on strategic voting
As Cox (1997: ch. 2) explains, and Cox et al. (2000: 15 f.) or Lago and Martínez (2007) show, elections for two separate legislative chambers in the same country at the same (or nearly the same) time produce natural experiments to see the impact of electoral rules on party and voter strategies. This is exactly what happens in national and regional elections in the region of Andalusia (Spain).
The Spanish Lower House has 350 members elected from 50 multi-member districts and two single districts using the D’Hondt method of proportional representation with closed lists for each district. Each multi-member district is allocated two seats, with the remaining seats allocated by population size. There is a formal threshold of 3 percent of valid votes at the district level. In national elections in the region of Andalusia, there are eight multi-member districts. As can be seen in Table 1 , since the founding election in 1977 the range of district magnitude has been wide. Across all eight districts and all elections between 1975 and 2008, the district magnitude ranges from a minimum of 5 (Almeria and Huelva) to a maximum of 12 (Sevilla, since 1977). Additionally, Almeria and Huelva are the only districts where the third national party, the communist party (the PCE) and its heir, the United Left (IU), have never won seats. On the contrary, the Socialist Party (the PSOE), the main leftist party, has always had much better results than the PCE/IU in all districts and is a viable competitor everywhere. 6 As the evidence provided by Gunther (1989) or Lago (2008) supports, there are strong incentives for district-level strategic voting in the smaller districts (i.e. district magnitude equal or lower than five seats).
Comparing electoral systems in Andalusia
Mean votes PCE/IU (/%): National elections: 8.64; Regional elections; 11.93
In regional elections in Andalusia, the electoral system is the same as in national elections. Seats are elected in the same eight districts with exactly the same geographic area as in national elections and using the D’Hondt formula with closed party lists and a formal threshold of 3 percent at the district level. The crucial difference is that in all districts in regional elections the number of seats to be filled is greater than five. As can be seen again in Table 1, district magnitude across all eight districts and all regional elections ranges from 11 (Almeria and Huelva) to 18 (Sevilla). The main consequence is that the PCE/IU has won seats in all districts in at least four of the eight elections. In short, independently of how expectations are generated, incentives to behave strategically at the district level are clearly absent.
According to Dunning (2008: 283), ‘random or “as if” random of assignment to treatment and control conditions constitutes the defining feature of a natural experiment’. The exogeneity of the assignment process can be derived from an understating of the data generation process (Dunning, 2008: 289; Robinson et al., 2009: 347). Electoral district borders in national elections in Spain were established in 1976, before the first election, and using an administrative division of 1833. The regional electoral system in Andalusia adopted exactly the same district borders. On the other hand, it is hardly arguable that this drawing of district boundaries was correlated with systematic differences in voting behaviour between the national and the regional arena.
Table 2 displays the different incentive structures for strategic voting in Andalusia when both electoral systems are superimposed. In those districts where district magnitude is greater than 5 in national and regional elections (and the PCE/IU is a viable competitor in both electoral levels), voters should behave sincerely (cell B). That is, differences in the PCE/IU’s vote-shares between national and regional elections in each district are a function of idiosyncratic characteristics of each territorial level, but not of district-level strategic voting. However, things are different in those districts whose magnitude is greater than 5 in regional elections but not in national ones (cell A). In regional elections, incentives for voting strategically are again absent, above all when we take into account that the PCE/IU is a viable competitor there. But in national elections, given that district magnitude is equal to 5, (some) PCE/IU supporters should behave strategically. That is, differences in the PCE/IU’s vote-shares in cells A and B are a function of idiosyncratic characteristics of each territorial level plus the impact of district-level strategic voting in national elections.
Incentive structures for strategic voting in Andalusia
On the basis of these incentive structures, the existence of strategic voting in national elections can easily be tested and its extent measured. Comparing the difference between national and regional elections in each district in the PCE/IU’s vote-shares in each cell, if the (relative) difference in A is higher than in B, there is evidence of strategic voting in districts with M = 5 in national elections. But if the difference between national and regional elections in each district in the PCE/IU’s vote-shares in A is equal to that in B, there is no evidence of strategic voting in national elections.
Using relative differences instead of absolute differences, the difference in the PCE/IU’s vote-shares between electoral levels is calculated as follows:
where %(IU) R i is the percentage of votes obtained by the PCE/IU in the regional election in district i %(IU) N i and the percentage obtained by the PCE/IU in the national election in district i.
The dependent variable here is the difference between the percentages of votes obtained by the PCE/IU in each pair of national and regional elections. The pairs of elections analysed are 1982/1982, 1986/1986, 1989/1990, 1993/1994, 1996/1996, 2000/2000, 2004/2004 and 2008/2008. The first year corresponds to the national election. As the ratio surpasses 100, the better the electoral results of the PCE/IU in regional elections in a given district compared to national elections, and the other way round. For instance, if the ratio is 110, the percentage of votes obtained by the IU in the regional election was 10 percent higher than in the national election. And the opposite is true when the ratio is 90.
The key independent variable is M = 5, a dummy variable coded 1 for those districts with magnitude 5 in national elections, Almeria and Huelva, and 0 for the remaining districts.
Results
Table 3 gives the OLS estimation results to test the hypotheses. All standard errors are clustered by electoral district to account for the non-independence in the data structure. According to the first model, there is clear evidence of strategic voting in national elections. The variable M = 5 is statistically significant at the 0.05 level and has the expected positive sign: in the districts with magnitude 5 in national elections, the PCE/IU won 11.64 points more (less) in regional (national) elections than in national (regional) elections compared to the districts with magnitude higher than 5 in both national and regional elections. That is, given that in districts with a magnitude greater than 5 there is no strategic voting and that the constant is substantially higher than 100 (i.e. the IU does better in regional elections), around 9 of each 100 voters of the PCE/IU in regional elections behaved strategically in national elections in these two districts. Or, in other words, around 9 percent cast a strategic ballot when presented with the opportunity to do so (Alvarez et al., 2006). The mean vote-share of the PCE/IU in the eight regional elections in these two districts is 8.79; consequently, around 0.75 percent of the total number of voters in national elections behaved in a strategic fashion.
Strategic voting in national elections in Andalusia
Estimation is by ordinary least squares. Standard errors in parentheses.
**p < 0.05; ***p < 0.01.
However, this level of strategic voting could be the tip of the iceberg. If voters are not aware of the different institutional incentives in each arena, they might also strategically desert the PCE/IU in regional elections, although they had no incentives. Hence, the true extent of strategic voting in national elections should be equal to the previous estimate plus the percentage of voters behaving strategically in both arenas. There are good reasons for thinking that the superposition of electoral systems weakens the informational prerequisite for strategic voting. When significantly different rules are superimposed within countries, and one of the offices/tiers/arenas to be elected is more important than the other, it can be hypothesized that voters will use the information generated in the more important contest to vote in both elections. Some of them will look into the mirror of the more important election to decide their vote in every contest within the country. Given that the political knowledge of most adults is very low – ‘study after study documents the breadth and depth of citizens' ignorance’ (Lupia and McCubbins, 1998; ch. 1) – using the same cues in all elections is a way to simplify the decision-making process.
If the most important office/tier is elected under a stronger electoral system than the less important office/tier, then some voters may be faced with incentives to vote strategically in the former but not in the latter. And if voters do not fully adapt to the different incentives provided by electoral systems and use the information about the electoral prospects of competitors in the more important arena also in the less important, they could vote strategically in both arenas, although strategic voting is unnecessary in the latter. Put simply, the more important election within countries is a distorted mirror for voting strategically in the less important one. In sum, the logic predicting strategic voting in elections where the opportunity to do so is not present as a consequence of contamination effects is similar to that underlying the well-documented phenomenon of ‘coat-tail voting’ in presidential regimes. Similar to what happens in presidential coat-tails (Amorim Neto and Cox, 1997; Golder, 2006), this contamination effect should be more important the higher the temporal proximity between elections, that is, in concurrent elections. Consequently, if the argument is right, we should observe more strategic voting in concurrent elections than in non-concurrent elections.
To test this hypothesis, in model 2 I have included the variable Non-Concurrent Elections (coded 1 for non-concurrent elections and 0 for concurrent elections), while in model 3 I have added an interaction between Non-Concurrent Elections and M = 5 to the previous specification. In model 2, Non-Concurrent Elections has the expected negative sign, but the variable is not statistically discernible from zero. And in model 3 the interaction is not statistically significant either. In sum, there is no iceberg below the tip.
Finally, in models 4 and 5 the argument that strategic voting increases with time as a result of a learning process is tested. If the argument of Tavits and Annus is right (2006), we should observe higher strategic voting after the first pair of national and regional elections. I have created a dummy variable in model 4 identifying the first pair of national and regional elections, while in model 5 its interaction with M = 5 is added. Both the dummy variable and the interaction do not have the expected negative sign; that is, the occurrence of strategic voting is higher in the first pair of elections. But they are not statistically significant. 7 However, the absence of contamination and learning effects is far from being a conclusive finding. The variables could be insignificant because the number of observations is low and/or insufficient variance. Clearly, this question needs further research.
The results of an additional test analysing whether strategic voting arises in regional elections in Andalusia as a consequence of contamination effects are given in Table 4
.
8
The dependent variable now is the difference between the percentages of votes obtained by the PCE/IU at the regional and district level in each regional election:
Strategic voting in regional elections in Andalusia
Estimation is by ordinary least squares. Standard errors in parentheses.
*p > 0.1; **p < 0.05; ***p < 0.01.
where %(IU) R i is the percentage of votes obtained by the PCE/IU in a regional election i and %(IU) D i the percentage obtained by the PCE/IU in the same regional election in district i.
As the ratio surpasses 100, the better the electoral results of the PCE/IU in a given district compared to the whole region in a given regional election. For instance, if the ratio is 110, the percentage of votes obtained by the IU in a given district was 10 percent higher than in the whole region. And the opposite is true when the ratio is 90.
My argument is that if voters behave strategically in regional elections as a consequence of contamination effects, when the opportunity to do so is not present, we should observe significant differences between concurrent and non-concurrent regional and national elections. The Duvergerian gravity coming from the national arena should be non-existent or at least weaker in non-concurrent regional elections than in concurrent regional elections. If some voters in regional elections are strategically deserting competitors who are viable in the regional election but not in the national race, then this behaviour should be particularly relevant in concurrent elections. The results displayed in Table 4 are evidence that voters behave as if they knew the different incentives provided by each electoral system, although they had too small a stake in elections.
According to the first model, in those districts with magnitude 5 in national elections, Almeria and Huelva, the PCE/IU gets worse results than in the remaining districts. However, models 2 and 3 show that its electoral support does not significantly increase in non-concurrent elections. In other words, when the Duvergerian gravity is weaker, no statistically significant differences in the vote-shares of the PCE/IU are observed. In sum, voters fully adapt to the different incentives provided by each electoral system. This is not surprising. If voters are sophisticated enough to recognize and have strategic responses to ‘wasted vote’ situations (Duch and Palmer, 2002), they should vote sincerely when the strategic voting opportunity does not exist.
Conclusions
Research on strategic voting agrees on two points. First, strategic voting is crucial for reducing contests with unviable parties or candidates in contests in which at most M + 1 competitors are seriously in the running for seats (Cox, 1997). Second, estimates of the aggregate amount of strategic voting rarely converge, even when researchers examine the same election. To what extent strategic voting is as important empirically as in theoretical models is still an open debate. This paper has dealt with this second point. Using a natural experiment in Andalusia (Spain), where everything is held constant with the exception that in some districts there are very strong incentives for behaving strategically in national elections, but not in regional elections, I have estimated the level of strategic voting in multi-member districts. I found that only a small fraction of the total voters, around 0.75 per cent, behaved strategically in national elections in Andalusia. However, by focusing the analysis exclusively on voters in districts in which they had the opportunity to behave in a strategic fashion, I found that around 9 per cent cast a strategic ballot when presented with the opportunity to do so. Additionally, voters know the different incentives for strategic voting in each territorial level and there are no contamination effects between electoral arenas. Voters did not strategically desert the PCE/IU in regional elections.
Finally, the empirical results also show that the strategic behaviour of voters did not increase with time as a result of a learning process. Apparently, the developmental argument of strategic voting does not hold in Andalusia.