Examining the “Referendum Theory” in Taiwan’s Local Elections

Abstract

This paper analyzes the extent to which the performances of local and national governments can shape local election outcomes. Specifically, we use various waves of survey data from Taiwan’s Elections and Democratization Studies (TEDS) to explore whether a person’s assessments of local and central government performances affect his/her vote for the incumbent party candidate. Our empirical findings partially verify the so-called “referendum theory” and can be summarized as follows: First, voters who hold a positive assessment of the performance of local government are more likely to vote for an incumbent who seeks reelection, but this is not necessarily the case for an incumbent party candidate in an open-seat contest. Second, Taiwan’s local elections cannot be regarded as referenda on the central government because the central government approval rating does not consistently affect vote choices across different types/levels of local elections.

Keywords

Local elections midterm elections referendum theory government performance approval rating incumbency

Introduction

Empirical studies suggest that voters usually punish or reward the performance of the national government in what are viewed as second-order elections between two major elections (Erikson, 1988; Tufte, 1975). Taiwan’s past experience seemed to correspond with such scholarly findings (Hsiao and Yu, 2008; Wu and Lee, 2004; Yu, 2011). As Taiwan’s local elections always take place between the two major national elections (i.e. presidential elections), regardless of whether it is the Democratic Progressive Party (DPP) or Kuomintang (KMT), the ruling party of the central government in Taiwan almost always loses in such “midterm exams.”

From a viewpoint of timing, Taiwan’s local elections can surely be regarded as “midterm exams” of the national government performance. Yet, from a viewpoint of scope, Taiwan’s second-order (i.e. local) elections, unlike US Congressional midterm elections, have no national component at all.¹ Nevertheless, as the recent partisan divide appears to run deeper in Taiwan (Hawang, 2016; Hsiao, 2019; Hsiao and Yu, 2020), it is arguable that local elections may function like national referenda. Indeed, the major opposition party has consistently asked voters to cast their votes of confidence against the ruling national government in all of the past local elections. However, the extent to which the voters have responded accordingly to such request should be an empirical question and needs further investigation.

Do voters actually regard local elections as a test of the ruling party’s performance? In Taiwan’s Election and Democratization Study (TEDS), one related question included in the 2010, 2014, and 2018 municipality election surveys directly addressed such a concern.² Figure 1 shows that in the 2010 survey, less than 50% of the respondents in the surveyed municipalities regarded the mayoral election as a vote of confidence in President Ma Ying-jeou’s administration. Except for those in Taipei, most respondents in Taichung and Kaohsiung thought that it was just a local election. In contrast, as the 2014 local election took place approximately 8 months after the mass social movement (i.e. the sunflower movement in March) against the Ma administration, the majority of the respondents across all the municipalities thought that the mayoral elections should be regarded as a vote of confidence in the ruling KMT government. The same pattern can be found in the 2018 survey data, even though the ruling party has become DPP since 2016. Overall, Figure 1 suggests that voters usually tend to link local elections with the performance of the central government, but they do not do so all the time. As local elections elect local governments/assemblies that pay attention to local affairs, we should not take for granted that those elections are assumed to function as “midterm exams” for the ruling parties. Then, the question becomes: under what conditions or to what extent do voters account for national factors when they vote in local elections?

Figure 1.

Attitudes toward municipality mayor elections, 2010, 2014, and 2018.

Taiwan is a useful case for answering this question.³ First, the past relevant literature about midterm elections is mainly drawn from the US, where midterm elections always include national-level (i.e. Congressional) offices. As mentioned above, Taiwan’s local elections do not have any national components. The case of Taiwan can be regarded as a clean test of the effect of national-level (government) performance on local vote choice. Second, Taiwan is also a country with a consolidated party system, thus voters can use the party label to connect performance across offices. Third, local government is important and has strong autonomy in Taiwan, particularly compared with its democratic counterparts in East Asia such as Japan and South Korea. And last, the different size of various local governments makes it possible to compare the effect of national-level performance on districts with higher and lower levels of electoral nationalization.

In short, the main purpose of this paper is to analyze the extent to which national factors affect the outcome of local elections in Taiwan. Specifically, we explore whether the performance of the central government overrides the importance of local politics in local elections. If so, under what conditions do national factors play significant roles in local elections? If not, what might the key variables that determine local election outcomes be? The rest of this paper is structured as follows: the next section reviews the major literature on vote choice in local elections, emphasizing national and local factors as the two competing explanatory variables as well as laying out conditional hypotheses with respect to incumbency and levels of districts, respectively; following this, data and measurement for our empirical analysis are introduced; findings are then presented; and a final section concludes the paper.

National versus local factors in local elections

Over the past few decades, numerous studies have utilized empirical data to test the theory of retrospective voting—that is, rather than choosing among candidates on the basis of specific issues of public policy, voters may rely more on assessments of government performance (Fiorina, 1978; Key, 1966). In other words, if the performance of the incumbent party is “satisfactory,” the voters will retain the incumbent governing party in office with their votes, whereas voters will vote against the incumbent when they evaluate the performance of the incumbent party as “not satisfactory.”

To what extent can the theory of retrospective voting be applied to understand midterm/second-order elections? Researchers have fiercely debated whether it is valid to interpret second-order elections as barometers of national public opinion toward the national government. There are several theories speaking to this divergence of opinion. According to the referendum theory developed on the basis of US American politics, the midterm electorate usually punishes the president’s party when the government performs poorly. Tufte’s (1975) notable effort coupled macroeconomic conditions and opinion poll ratings in explaining vote shares of the president’s party in congressional elections. Specifically, his “negative voting” argument suggested that citizens are more likely to turn out to vote against the president’s party in midterm elections when they hold negative views on the current president or government. Thus, the linkage between the president’s performance and the outcome of midterm congressional (i.e. legislative) elections can be explained by the referendum theory. It is logical that voters might express their opinions about the national government through another election that focuses on national issues such as congressional elections.

In addition to the linkage between the executive and legislative branches, empirical studies also employ aggregate-level data over time and find strong support for the notion that national economic conditions affect gubernatorial candidates’ electoral results (Chubb, 1988; Kenney, 1983; Stein, 1990; Svoboda, 1995). The national effect is potentially tied to the president: gubernatorial candidates who are of the same party as the president will benefit when the national economy is strong and suffer when it is weak, whereas candidates from opposition parties experience the reverse. Beyond studies on US gubernatorial elections, Leon (2014) used Spanish aggregate-level data and also showed that regional politicians’ electoral performances are correlated with that of their national counterparts, particularly from a viewpoint of economic conditions.

By using individual-level survey data, early studies on gubernatorial elections found that vote choice is a function of party identification, the evaluation of national economic conditions, and the president’s performance (Simon, 1989; Simon et al., 1991). Yet, King (2001) posited that a referendum test on the incumbent governor should be included as a key local-level variable to explain vote choice in gubernatorial elections. His empirical analysis found that the performance of the incumbent governor has a greater effect when he or she seeks reelection, and this effect, though it becomes minor, is also present in open-seat contests. Indeed, it is not a straightforward fact that voters pay more attention to the president’s performance than to the performance of the local government when they choose their state’s chief executive. Brown and Jacobson (2008) showed that governors can better insulate themselves from national political conflict than senators. That is, governors are executives who cannot avoid responsibilities for unpopular actions of the state government, and senators, as national politicians, are inevitably tied up with either positive or negative national political waves. Thus, the two competing hypotheses—namely, the national referendum hypothesis and the local referendum hypothesis—need further empirical tests to see under what conditions the former is more valid than the latter and vice versa.

In Taiwan, two levels of government (henceforth “local” and “central”) have different authority over certain aspects of governance. On a small island such as Taiwan, electoral expectations should be shaped by both local and national conditions. Previous studies on Taiwan’s mayoral elections showed that both local and national conditions have significant effects on determining the vote decisions of the electorate (Hsiao and Yu, 2008; Wu and Lee, 2004; Yu, 2011). However, some puzzles remain unsolved—that is, under what conditions and to what extent might local and national circumstances combine to shape electoral expectations? We argue that two contextual variables should be taken into account when studying Taiwan’s local head elections: first is the incumbent factor. For races with incumbents, retrospective performance always plays an important role in determining vote choice, as voters may easily attribute responsibility to the incumbent (Miller and Wattenberg, 1985; Peffley, 1984). In non-incumbent (or open-seat) races, it is relatively difficult for voters to attribute responsibility to particular candidates. Even successor candidates (from the incumbent’s party) may only take partial credit/blame for the past government performances, as they exert no or very limited influence on the status quo policies (Campbell et al., 2010).

Thus, if an incumbent runs for local head reelection, we expect that his/her past performance is always important in determining vote choice. On the other hand, in an open-seat contest, as none of the candidates needs to take full responsibility for the local government’s past performance, the central government’s recent performance can be an accessible and substantial reference point for voters in casting their votes. It does not necessarily suggest that the assessment of the central government performance should be more important than that of the local government in determining vote choice in non-incumbent races. But we expect that the central government performance plays a more considerable role for voters in open-seat contests than in incumbent races. In short, the type of race (i.e. non-incumbent/open-seat race vs. incumbent race) may decide the relative importance between national and local accountabilities.

Second, different levels of local elections should not be regarded as the same. It is arguable that voters in urban areas are simply different from those in rural areas. Blais et al. (2003) demonstrated that rural voters are more likely to form preferences for their local candidates than their urban counterparts. Zaller (2004) argued that voters in large districts by population tend to be rather ignorant about local affairs and are more likely to support incumbent candidates. In contrast, voters in small districts tend to pay more attention to local issues and are more likely to support challengers. Oliver and Ha (2007) found that voters in suburban areas have better knowledge about their candidates and issues. Thus, unlike voters in big cities, suburban voters tend to cast their votes based on candidates’ issue positions rather than just party labels. In short, the information gap forms a rural–urban divide in voting behavior, as rural voters tend to be more informed about local candidates and issues than their urban counterparts.

In developing democracies, scholars also found reasons to differentiate rural voters from urban ones. For example, Bawumia (1998) posited that in Ghana, rural voters were more likely to support the incumbent military leader Jerry Rawlings in the 1992 Ghanaian presidential election because they had benefited for almost a decade from the Structural Adjustment Program introduced by Rawlings’ previous military junta (i.e. Provisional National Defense Council, PNDC). Koter (2013) also showed that in Senegal’s presidential elections between 1978 and 2012, rural voters have tended to support the incumbent, as they were more susceptible to clientelism than urban voters. As to the comparisons of voting patterns between rural and urban voters, the abovementioned findings based on either developed or developing countries pointed to the same trend, that rural and urban voters have different concerns when casting their votes.

Since 2010, Taiwan has had five major municipalities (i.e. Taipei, New Taipei, Taichung, Tainan, and Kaohsiung) that consist of more than 60% of the entire population in Taiwan. Furthermore, in 2014, Taoyuan became the sixth municipality, and now approximately 70% of the Taiwanese population resides in these six municipalities. It is arguable that municipality mayoral elections are different in nature compared with county magistrate or city mayor elections due to the former’s scope and importance. Thus, municipality mayoral elections always draw significant media attention and become major battlegrounds for high-profile, ambitious politicians. For example, both President Chen Shui-bian and President Ma Ying-jeou were Taipei municipality mayors before running for president in 2000 and 2008, respectively. The two major candidates in the 2016 presidential election, DPP’s Tsai Ying-wen and KMT’s Eric Chu, ran against each other in the 2010 New Taipei municipality mayoral election.⁴

Thus, we argue that in Taiwan’s local elections, people who live in municipalities (i.e. urban voters) are more likely to be influenced by national factors than those who live in counties/cities (i.e. rural voters). This is due to the following two reasons: first, in a comparison between urban and rural voters, the former tend to use Internet news platforms or social media to receive political information (Biswas et al., 2014), and they are more likely to receive information regarding central government performance and utilize such assessment to cast their votes even in local elections. In contrast, rural voters tend to have stronger local networks and are more likely to be involved in local affairs. Thus, in local elections, they are more likely to cast their votes based on the assessment of local government performance. Second, municipalities have direct linkages with the central government. For example, the six municipality mayors are allowed to participate in the weekly meeting of the Executive Yuan (i.e. cabinet meeting). They can elaborate their policy demands directly to the central government. Because of a closer tie between the municipality governments and the central government, municipality elections are expected to have a stronger linkage with the performance of the central government. In short, the differences in voter preferences and government structures between municipality and county/city lead to our hypothesis that the national factors play a relatively more important role in municipality elections than in county/city elections.

Based on the abovementioned literature review, we spell out two sets of hypotheses with respect to the referendum theory in local elections and test them by employing Taiwan’s survey data. The first set is called “referendum hypotheses” (i.e. H1 and H2) and the second set is called “conditional hypotheses” (i.e. H3 and H4):

Referendum hypotheses in local elections

Hypothesis 1 (H1: national referendum hypothesis): The performance of the central government affects vote choice. Specifically, the more a voter is satisfied with the central government performance, the more likely he/she will vote for the ruling party candidate in the local election.

Hypothesis 2 (H2: local referendum hypothesis): The performance of the local government affects vote choice. Specifically, the more a voter is satisfied with the local government performance, the more likely he/she will vote for the local incumbent party candidate.

Conditional hypotheses in local elections

Hypothesis 3 (H3: incumbency hypothesis): The performance of the local government is more important than the performance of the central government if the incumbent is seeking reelection.

Hypothesis 4 (H4: levels of district hypothesis): The performance of the central government is more important than the performance of the local government when the election takes place in a municipality; the performance of the local government is more important than the performance of the central government when the election takes place in a county/city.

In the next section, we utilize multiple waves of survey data to see whether Taiwan’s local elections can be regarded as “referenda.” Specifically, we analyze the determinants of vote choice in local head elections (i.e. county magistrate, city mayor, and municipality mayor) and examine the extent to which voters consider local and central (national) government performances when they cast their votes in local elections.

Data and measurement

To test the hypotheses regarding the referendum theory in local elections, this paper uses TEDS’ face-to-face survey data collected after the 2005, 2009, 2010, 2014, and 2018 local elections.⁵ It is important to note that TEDS surveys were only conducted in selected counties, cities, or municipalities in each of the local election years. Specifically, the dataset consists of 15 local election surveys, including four county/city-level surveys in 2005, two county-level surveys in 2009, and three municipality-level surveys in 2010, 2014, and 2018, respectively.⁶ Among these 15 local elections covered by TEDS, we drop two cases including the 2010 Kaohsiung and 2018 Taipei municipal mayoral elections. Both of them were three-way battles and it is difficult to identify the incumbent party or challenger party candidates in the elections.^7,8 Thus, we only use 13 election cases for the following data analysis.

To test our conditional hypotheses with respect to the referendum theory, we first divide our cases into two groups by types of races—open-seat contests and incumbent races—and expect to see different impacts of the assessments of government performance on vote choice between the two types (i.e. Hypothesis 3). In addition, because municipality elections are much more important not only to politicians but also voters, we divide the 13 cases into two groups by level of election—namely, county/city level and municipality level. And we expect that the two levels of election have different dynamics in terms of setting different conditions for the referendum theory (i.e. Hypothesis 4). Thus, we shall have a two-by-two analytical framework consisting of 13 cases as follows.

The dependent variable in the following analyses is a vote for the incumbent party candidate, a dummy variable coded by the following rules: If a voter voted for the magistrate/mayor candidate whose party was the incumbent local government party, we code his/her vote choice as 1. If a voter did not go to vote or voted for a challenger candidate, we code his/her vote choice as 0.

As this paper intends to explore why voters vote for the incumbent party candidate from a perspective of the referendum voting model, we need independent variables to test the referendum hypothesis with respect to government performance at both local and national levels (King, 2001; Piereson, 1975; Simon, 1989; Simon et al., 1991)—namely, the variables should test the assessment of local government performance (or local government approval rating) and the assessment of central government performance (or central government approval rating). We utilize the following questionnaires to construct proxies for both assessment measures in different elections. Three different measures for the assessment of local government performance and one measure for the assessment of central government performance are as follows:

Assessment of local government performance⁹

i. Measure of the 2005 surveys:

a. Generally speaking, how satisfied are you with the infrastructure of [interviewee’s household] County in the past four years? Very satisfied, satisfied, somewhat dissatisfied, or very dissatisfied?

b. How about social welfare?

c. How about law and order?

d. How about environment and health?

e. How about transportation?

ii. Measure of the 2009 surveys:

Overall, are you satisfied or dissatisfied with the performance of [Mayor’s name] over the past four years?

Very satisfied, satisfied, somewhat dissatisfied, or very dissatisfied?

iii. Measure of the 2010, 2014, and 2018 surveys:

Overall, are you satisfied or dissatisfied with the performance of [name of municipality]’s mayor over the past four years?

Very satisfied, satisfied, somewhat dissatisfied, or very dissatisfied?

Assessment of central government performance:

How satisfied are you with the overall performance of the central government under the KMT/DPP over the last two years/last three years? Very satisfied, satisfied, somewhat dissatisfied, or very dissatisfied?

For the 2005 survey data, we utilize five questions to construct local government approval ratings (i.e. infrastructure, social welfare, law and order, environment and health, and transportation; very satisfied = 4, satisfied = 3, somewhat dissatisfied = 2, very dissatisfied = 1). These five items are indispensable and conformable to uncover voters’ assessments of local government performance.¹⁰ Hence, we combine these five questions to compute the local government performance score and transform it to a 1–4 scale, which is comparable with the measures of other local government approval ratings used in the 2009, 2010, 2014, and 2018 surveys, as well as the measures of the central government approval rating in all of the surveys (i.e. very satisfied = 4, satisfied = 3, somewhat dissatisfied = 2, very dissatisfied = 1). In addition, we recoded the central government approval ratings by using the identical coding (e.g. 1 if “very dissatisfied” and 4 if “very satisfied”) when the central and local governments were of the same party, but we inverted it when the two executives were of opposing parties (e.g. −1 if “very dissatisfied” and −4 if “very satisfied”).¹¹

We incorporate candidate thermometer ratings and party support as two major control variables. Both of them can be regarded as “classical variables” that have been included in almost every vote choice equation and are invariably associated with the assessment of local/central government performance. To construct a candidate thermometer rating, we use the thermometer score of the local incumbent party candidate (i.e., a 0–10 scale) minus the thermometer score of the challenger party candidate. Thus, the maximum value of the measure becomes +10, and the minimum value becomes −10.

Regarding party support, we coded the respondents into three categories: (1) Support for the local incumbent party, (2) Support for the local challenger party, and (3) Independent voters. In the following regression analyses, independent voters always serve as a base category. Finally, we also include election dummies as control variables to tease out unspecified variation across different elections.

In short, we argue that a vote for the local incumbent party candidate can be regarded as a function of the local government approval rating, central government approval rating, (difference of) the candidate thermometer rating, and the voter’s party support.¹² Because our dependent variable is dichotomous, we use a logistic regression model in the following analyses. Due to the analytical framework we adopt in Table 1, we set up four models for different scenarios—that is, an incumbent race at the county/city level, an open-seat contest at the county/city level, an incumbent race at the municipality level, and an open-seat contest at the municipality level.

Table 1.

Level of Election and Type of Race, 13 Cases.

		Type of Race
		Open Seat Race	Incumbent Race
Level of Election	City/County Level	2005 Yunlin County 2005 Taipei County 2009 Taoyuan County	2005 Kaohsiung County 2005 Taichung City 2009 Yunlin County
Level of Election	Municipality Level	2014 Taipei Municipality 2018 Kaohsiung Municipality	2010 Taichung Municipality 2010 Taipei Municipality 2014 Kaohsiung Municipality 2014 Taichung Municipality 2018 Taichung Municipality

Findings

Figure 2 shows the results of logistic regression analyses for the four different scenarios with respect to election levels and race types.¹³ When the election has an incumbent running for reelection, regardless of election level (i.e. county/city level or municipality level), the coefficient for local government approval rating is always positive and different from zero at the 0.05 significance level (see the upper two columns in Figure 2). The finding is intuitive: when the incumbent runs for reelection, his/her performance in office is a key factor in determining vote choice. However, if the election is an open-seat contest, the coefficient for the local government approval rating is only significant in the municipality mayoral election but not in the county/city election. In short, local government performance seems to be an important factor in explaining why a voter casts his/her vote for the incumbent party candidates, at least in three out of the four abovementioned scenarios.

Figure 2.

Logistical regression of vote choice for incumbent party candidate.

In contrast, the coefficient for central government approval rating is only significant under one scenario: the incumbent runs for reelection as the municipality mayor. Thus, the assessment of central government performance does not seem to be an effective variable to predict vote choice for local incumbent party candidates, particularly not in county/city-level elections. Even though the opposition party always labels second-order elections between two major elections a “midterm exam” for the performance of the ruling party in the central government, the results of such elections are not necessarily evaluative from a viewpoint of voting behavior. In other words, voters do not usually account for the performance of the central government when deciding how to vote in local elections. Why, then, does the central government approval rating become important in municipality mayoral elections? A feasible explanation is that municipality mayors usually play leadership roles in their own parties and have nationwide popularity, so their electoral prospects are more likely to be affected by national ties than their counterparts at the county/city level. Of course, such an explanation needs a further empirical test that is beyond the scope of this research.

Regarding the control variables, the coefficients for candidate thermometer rating are statistically significant with the expected positive signs in all different scenarios. Specifically, a voter who gives a more favorable rating to the incumbent party candidate than to the challenger party candidate is more likely to vote for the incumbent party candidate. The party support variables on vote choice are also significant in all the equations with expected signs—namely, that local incumbent party supporters tend to vote for incumbent party candidates, whereas local challenger party supporters are less likely to do so. Overall, candidate thermometer rating and party support play significant roles in determining vote decisions in local elections.

To fully explain the coefficient estimates of the logit regression analyses, we also interpret our findings in a substantial way by using the coefficient estimates to calculate predicted probably changes with respect to various levels of local and national government approval rating for both open-seat contests and incumbent races at the county/city level and the municipality level, respectively, holding all other variables constant. Figure 3 illustrates the probability of changes in voting for incumbent party candidates in different scenarios with respect to changes in the local government approval rating. As the local government approval rating changes from the lowest to the highest level (i.e. from 1 to 4) and the central and local governments are of opposing parties,¹⁴ for example in Figure 3(a), the probability of voting for the incumbent in the county/city-level election increases by approximately 44 percentage points (i.e. from 0.22 to 0.66). Similarly, the probability of voting for the incumbent in a municipality-level election increases by approximately 26 percentage points (i.e. from 10% to 36%) when the local government approval rating changes from the lowest to the highest level (i.e. from 1 to 4). When the central and local governments are of the same party,¹⁵ Figure 3(b) indicates that the probability of voting for the incumbent in the county/city-level election increases by 43 percentage points (i.e. from 0.34 to 0.77) and the probability of voting for the incumbent in a municipality-level election increases by approximately 35 percentage points (i.e. from 20% to 55%) along with the changes of local government approval rating. These considerable probability changes verify the hypothesis that an incumbent’s past performance predicts his/her reelection prospect very well.

Figure 3.

Changes in predicted probability of voting for the incumbent party candidate by changes in the local government approval rating. (a) The central and local governments were of opposing parties. (b) The central and local governments were of the same party.

In contrast, the local government approval rating is not as important in an open-seat contest as it is in an incumbent race. Figure 3(a) and (b) suggest that in an open-seat contest, regardless of whether the central and local governments are of opposing parties or the same party, the probability of voting for the incumbent party candidate increases by about 10 and 20 percentage points at the county/city-level and municipality-level elections, respectively, as the local government approval rating changes from the lowest to the highest level. In other words, an incumbent party candidate in an open-seat contest does not necessarily have a substantial advantage or disadvantage based on past local government performance.

Figure 4 shows the changes in the predicted probability of voting for the incumbent party candidate by changes in the central government approval rating. Specifically, Figure 4(a) suggests that when the central and local governments are of opposing parties and voters are “very satisfied” (i.e. rating = −4) with the central government performance, in the county/city-level incumbent race, the probability that voters will vote for the incumbent is approximately 0.40. On the other hand, if voters are “very dissatisfied” (i.e. rating = −1) with the central government performance, the probability that they will vote for the incumbent is approximately 0.51. A similar pattern can be found that the probability of voting for the incumbent in a municipality mayoral election increases by approximately 11 percentage points (i.e. from 0.15 to 0.26) with respect to the change in central government approval rating from “very satisfied” to “very dissatisfied.” Thus, the probability of voting for the incumbent who is of the opposing party of the central government increases marginally as the central government approval rating changes from the highest to the lowest extreme. As to the open-seat scenarios (i.e. open-seat contests at the city/county level and open-seat contests at the municipality level), the predicted probabilities of voting for the incumbent party candidate almost have no change along with the decrease of the central government approval rating. In other words, when the central and local governments are of opposing parties, the performance of the central government has no substantial impact on voting for the incumbent in an open-seat contest, regardless of the election level.

Figure 4.

Changes in the predicted probability of voting for the incumbent party candidate by changes in the national government approval rating. (a) The central and local governments were of opposing parties. (b) The central and local governments were of the same party.

Given that the central and local governments are of the same party, Figure 4(b) shows very similar patterns to those shown in Figure 4(a). Specifically, the largest magnitude of the probability changes in voting for the incumbent party candidate occurs in the scenario in which an incumbent municipality mayor runs for reelection. However, such a magnitude of the probability changes is not considerable—only approximately a 14 percentage point increase (i.e. from 0.32 to 0.46) as the central government approval rating changes between the two extreme levels. In all other scenarios, the probability of voting for local incumbent party candidates does not change by more than 10 percentage points as the central government approval rating changes from the lowest to the highest level. In a comparison with the impacts of local government approval rating shown in Figure 3, Figure 4 suggests that the central government approval rating has relatively marginal impacts, if any, on vote choice in almost all types/levels of local head elections.

Concluding remarks

This paper utilizes Taiwan’s survey data to analyze the extent to which the assessment of local and central government performances affect voting behavior. Specifically, we argue that in different types of elections (i.e. incumbent race vs. open-seat contest) and levels of elections (i.e. city/county level vs. municipality level), the associations between assessment of government performance and vote choice may vary. As to the four hypotheses we lay out in this paper (i.e. hypotheses on national referendum, local referendum, incumbency, and levels of district), our empirical evidence only partially verifies Hypotheses 1–3 but rejects Hypothesis 4. Specifically, our empirical findings are threefold.

First, when there is an incumbent seeking reelection as a local head, regardless of the election level (city/county or municipality level), voters take the person’s job performance into account before casting their votes. Our findings support the conditional referendum hypothesis that voters with a positive assessment of local government performance are more likely to vote for the incumbent (i.e. H3: incumbency hypothesis). However, when the election is an open-seat contest, local government approval rating only affects vote choice in municipality mayoral elections but not in city mayoral/county magistrate elections. As the assessment of local government performance affects vote choice in three out of the four scenarios, the local referendum hypothesis (H2) is partially verified.

Second, although the referendum theory predicts that voters may evaluate the performance of the national government and use their votes to punish or reward the national ruling party even in second-order elections, such behavior fails to explain the situation in Taiwan. Our findings suggest that the performance of central government matters only in the case of an incumbent race of a municipality mayoral election. In all other scenarios, the central government approval rating is not a significant predictor of how voters will cast their votes. As our empirical evidence does not show strong support for the national referendum hypothesis (H1), Taiwan’s local elections can hardly be considered as referenda on the national (or central) government.

Third, as the impacts of different government performances on vote choice vary between county/city and municipality elections without any observed pattern, our analyses fail to verify the levels of district hypothesis (H4). In fact, as shown in Figure 3, the performance of local government is always more important than that of central government when we compare the magnitudes of impacts in a substantial way. Thus, the rural–urban divide does not necessarily differentiate the importance of central or local government performance in Taiwan’s local elections.

While this analysis uses Taiwan as a case study to show that the performance of the local government, particularly for an incumbent, is much more important than that of the central government in determining vote choice, our study does not conclude that “national ties” play no role in local elections. But at least, our empirical evidence suggests that the impact of central government performance on voting behavior in local elections, if any, is meager.¹⁶ Thus, Taiwan’s local elections should not be considered as referenda on the performance of the central government.

It is important to consider the Taiwan case in comparative perspective. For example, comparing Taiwan with its East Asia democratic neighbors such as Japan and South Korea, Taiwan’s local elections look similar to South Korea’s from an angle of electoral campaign—that is, the opposition parties almost always frame local elections as a “midterm exam” of the performance of national/central government and urge the voters to punish the national/central ruling parties in local elections. Yet, past empirical analyses on South Korea’s local elections showed mixed results suggesting that the importance of national factors in local elections varies over time and across voter groups, depending on the salience of national issues in local elections (Choi, 2013; Song, 2009). In Japan, national factors normally play minor roles in local elections. In fact, Japan’s party competitions in some regions (such as Osaka) possess very different dynamics from that at the national level. Of course, there are still some occasions when local elections reflect the public mood of national politics (Jain, 2019). But in general, Japan’s local elections tend to have weak linkages with national politics (Hijino, 2013). In short, the Taiwan case presented here shows the old phrase, “all politics is local” again. While our finding somehow corresponds to the evidence drawn from Japan and South Korea, we hope to incorporate more data from other countries to test the robustness of the finding for future research.

Supplemental Material

JAAS_national_factor_appendix_finalsub_20200923 – Supplemental material for Examining the “Referendum Theory” in Taiwan’s Local Elections

Supplemental material, JAAS_national_factor_appendix_finalsub_20200923 for Examining the “Referendum Theory” in Taiwan’s Local Elections by Eric Chen-hua Yu and Kah-yew Lim in Journal of Asian and African Studies

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Eric Chen-hua Yu appreciates the funding provided by Taiwan’s Ministry of Science and Technology (MOST 109-2410-H-004 -050 –SSS) for this research. Data analyzed in this article were from Taiwan’s Election and Democratization Study, 2005 (TEDS2005M: NSC 94-2420-H004-008-SSS), 2009 (TEDS2009M: NSC 98-2420-H-004-011), 2010 (TEDS2010C: NSC 99-2420-H-031-002), 2014 (TEDS2014: MOST 101-2420-H004-034-MY4), and 2018 (TEDS2018: MOST 105-2420-H-004-015-SS4). The coordinator of multi-year project TEDS is Professor Chi Huang (National Chengchi University). More information is on TEDS website (). The authors appreciate the assistance in providing data by the institute and individual aforementioned. The authors are solely responsible for views expressed herein.

ORCID iD

Eric Chen-hua Yu

Supplemental material

Supplemental material for this article is available online.

Notes

Author biographies

Eric Chen-hua Yu is an Associate Research Fellow of the Election Study Center (ESC) and jointly appointed as an Associate Professor of political science at National Chengchi University in Taipei, Taiwan. He is also affiliated with the University’s Taiwan Institute for Governance and Communication Research (TIGCR). His research interests and recent publications have focused on electoral politics, public opinion, and public policy analysis. Yu received an MS in Public Policy Analysis from the University of Rochester and a PhD in political science from Columbia University.

Kah-yew Lim received his PhD from the Department of Political Science at National Chengchi University, Taiwan. His research interests include big data in social media, public opinion, election studies and political behavior. He can be reached at kylim.ivan@gmail.com.

References

Bawumia

(1998) Understanding the rural-urban voting patterns in the 1992 Ghanaian presidential election. A closer look at the distributional impact of Ghana’s Structural Adjustment Programme. Journal of Modern African Studies 36(1): 47–80.

Biswas

Ingle

Roy

(2014) Influence of social media on voting behavior. Journal of Power, Politics & Governance 2(2): 127–155.

Blais

Gidengil

Dobrzynska

, et al. (2003) Does the local candidate matter? Candidate effects in the Canadian election of 2000. Canadian Journal of Political Science 36(3): 657–664.

Brown

Jacobson

(2008) Party, performance, and strategic politicians: The dynamics of elections for senator and governor in 2006. State Politics and Policy Quarterly 8(4): 384–409.

Campbell

Dettrey

Yin

(2010) Does the presidential record matter less in open-seat elections? The Journal of Politics 72(4): 1083–1095.

Carey

Shugart

(1995) Incentives to cultivate a personal vote: A rank ordering of electoral formulas. Electoral Studies 14(4): 417–439.

Choi

(2013) Political issues, generation gap, and voting behavior in South Korea: The 2010 Seoul mayoral election. East Asia 30(4): 237–254.

Chubb

(1988) Institutions, the economy, and the dynamics of state elections. American Political Science Review 82(1): 133–154.

Erikson

(1988) The puzzle of midterm loss. Journal of Politics 50(4): 1011–1029.

10.

Fiorina

(1978) Economic retrospective voting in American national election: A micro analysis. American Journal of Political Science 22(2): 426–443.

11.

Hawang

S-D

(2016) Executive-legislative relations under divided government. In: Chu

Y-H

Diamond

Templeman

(eds) Taiwan’s Democracy Challenged: The Chen Shui-bian Years. Boulder, CO: Lynne Rienner Publishers.

12.

Herron

Nishikawa

(2001) Contamination effects and the number of parties in mixed superposition electoral systems. Electoral Studies 20(1): 63–86.

13.

Hijino

(2013) Delinking national and local party systems: New parties in Japanese local elections. Journal of East Asian Studies 13(1): 107–135.

14.

Hsiao

Y-C

(2019) The people’s partisan polarization and its influence on the attitudes toward democracy in Taiwan. Taiwanese Political Science Review 23(2): 41–85. (in Chinese)

15.

Hsiao

Y-C

C-H

(2008) Administrative performance and voter choices: A comparative study of the 2006 Taipei and Kaohsiung mayoral elections. Taiwan Democracy Quarterly 5(2): 1–25. (in Chinese)

16.

Hsiao

Y-C

(2020) Polarization perception and support for democracy: The case of Taiwan. Journal of Asian and African Studies 55(8): 1143–1162.

17.

Jain

(2019) Japan’s local elections will impact national politics. East Asian Forum, 4 April. Available at: https://www.eastasiaforum.org/2019/04/04/japans-pivotal-local-elections-will-impact-national-politics/ (accessed 16 July 2020).

18.

Kenney

(1983) The effects of state economic conditions on the vote for governor. Social Science Quarterly 64(1): 154–161.

19.

Key

(1966) The Responsible Electorate. New York: Vintage.

20.

King

(2001) Incumbent popularity and vote choice in gubernatorial elections. Journal of Politics 63(2): 585–597.

21.

Koter

(2013) Urban and rural voting patterns in Senegal: The spatial aspects of incumbency 1978–2012. Journal of Modern African Studies 51(4): 653–679.

22.

Leon

(2014) How does decentralization affect electoral competition of state-wide parties? Evidence from Spain. Party Politics 20(3): 391–402.

23.

Miller

Wattenberg

(1985) Throwing the rascals out: Policy and performance evaluations of presidential candidates, 1952–1980. American Political Science Review 79(2): 359–372.

24.

Oliver

(2007) Vote choice in suburban elections. American Political Science Review 101(3): 393–408.

25.

Peffley

(1984) The voter as juror: Attributing responsibility for economic conditions. Political Behavior 6(3): 275–294.

26.

Piereson

(1975) Presidential popularity and midterm voting at different electoral level. American Journal of Political Science 19(4): 936–957.

27.

Rich

(2015) Duverger’s Law in mixed legislative systems: The impact of national electoral rules on district competition. European Journal of Political Research 54(1): 182–196.

28.

Rich

(2018) Coattails and mixed electoral systems: Evidence from Taiwan’s 2016 election. Journal of East Asian Studies 18(1): 47–66.

29.

Simon

(1989) Presidents governors, and electoral accountability. Journal of Politics 51(2): 286–304.

30.

Simon

Ostrom

Marra

(1991) The president, referendum voting, and subnational elections in the United States. American Political Science Review 85(4): 1178–1192.

31.

Song

B-K

(2009) Does the president’s popularity matter in Korea’s local elections? Pacific Affairs 82(2): 189–209.

32.

Stein

(1990) Economic voting for governor and U.S. senator: The electoral consequences of federalism. Journal of Politics 52(1): 29–35.

33.

Svoboda

(1995) Retrospective voting in gubernatorial elections: 1982 and 1986. Political Research Quarterly 48(1): 135–150.

34.

Tufte

(1975) Determinants of the outcomes of midterm congressional elections. American Political Science Review 69(3): 812–826.

35.

Wang

(2015) Duverger’s law as scale effect: Evidence from Taiwan 2014 village chief elections. Journal of Electoral Studies 22(1): 109–141. (in Chinese)

36.

C-L

Lee

S-H

(2004) Government performance and voting behavior: Lessons of the 2002 Taipei and Kaohsiung mayoral elections. Theory and Policy 17(4): 1–24. (in Chinese)

37.

(2011) Examining the ‘midterm loss’ in Taiwan: An analysis of the 2009 county magistrate election. Japanese Journal of Electoral Studies 27(2): 77–92.

38.

Zaller

(2004) Floating voters in U.S. presidential elections. In: Saris

Sniderman

(eds) Studies in Public Opinion: Attitudes, Nonattitudes, Measurement Error, and Change. Princeton, NJ: Princeton University Press.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.18 MB