Behind the Pursuit of ‘Frozen Garlic’: Changing Trends of Get-Out-the-Vote (GOTV) Campaigns in Taiwan

Introduction

How have electorates in Taiwan been persuaded to vote during elections over the last two decades? Are these trends aligned with the literature on electoral mobilization initiatives in political science? To observe the trends of electoral campaigns and their changes in Taiwan, we will go over multiple waves of post-election survey data conducted since the 2001 Special Municipalities and Legislative Yuan election by Taiwan’s Election and Democratization Study (TEDS) (2022)—a long-term, large-scale survey project funded by the Department of Humanities and Social Sciences (DHSS) under the National Science and Technology Council of Taiwan. In this article, I will observe the trends in both central (2001–2020) and local (2002–2018) elections for their respective executives (i.e., President, Mayors) and legislative branches (i.e., Legislative Yuan, or LY hereafter, and City/County Councils). To begin with, I will give an overview of the electoral mobilization literature in political science. I will then briefly describe the data and discuss how I put all these longitudinal data together in the face of some technical challenges. After that, moving on to the data section, I will look at how voters were approached by candidates or heard about these candidates and connect these findings with the literature. For local elections, on top of the means of electoral mobilization, I will also investigate by whom subjects were most influenced when casting their votes. Finally, I will summarize these observations with a short conclusion.

Literature Review

Although this article does not intend to make any strong causal claim regarding electoral mobilization, or more specifically the get-out-the-vote (GOTV) initiatives, but rather to review the use of different GOTV means in Taiwan over the years, it is still helpful to put the findings regarding the tendencies in Taiwan into the context of general findings identified by existing scholarship on GOTV campaigning. Previous research on GOTV initiatives mainly has aspired to answer two general questions: 1. What affects the choices of GOTV means and targets? 2. What affects the effectiveness of these GOTV initiatives?

Studies on the first question have identified three key driving forces that determine where and toward whom to deploy GOTV efforts. Electoral competitiveness, according to the study of Pattie, Johnston, and Fieldhouse (1995), is positively associated with GOTV spending. Also, as observed by the cross-national study of Karp, Banducci, and Bowler (2008), GOTV initiatives are more often used in candidate-based electoral systems than in the proportional representation ones, while active voters are more likely to be targeted by these efforts.

For the effectiveness of GOTV initiatives, present research shows that their efficiency depends on the type of GOTV means, electorates’ perceived turnout, and the partisan context of the constituency. There is a strong consensus that face-to-face canvassing is more effective in boosting the turnout than other more distanced means (Broockman and Kalla 2016; Gerber and Green 2000, 2005; Green, McGrath, and Aronow 2013; Imai 2005; Miller, Bositis, and Baer 1981). The suggested mechanism is that personalized and tangible GOTV efforts could signal candidates’ willingness to spend time and energy on voters (Green and Gerber 2019). This explanation also makes sense of why GOTV campaigns work better on voters with preliminary personal ties to candidates (Fiva and Smith 2017). Less tailored means, such as mailings and calls, have been found relatively less effective (Gerber and Green 2000, 2005; Imai 2005) or only making a difference in specific conditions, like in primary elections (Fortier-Chouinard et al. 2023) or with a precise volume of calls (Kling and Stratmann 2023). Besides the perceived affinity of these initiatives, based on other findings, high perceived turnouts (Arceneaux and Nickerson 2009) and also high partisan strength in the constituency (Huckfeldt and Sprague 1992; Foos et al. 2021) have been found to improve the efficiency of these GOTV inputs.

Context and Caveats

In Taiwan, trends in GOTV approaches adopted by candidates began to be tracked when the TEDS team started their first post-election survey in January 2002 (i.e., TEDS2001), right after the 2001 Special Municipalities and LY elections, with questions emphasizing the LY elections. Since then, they have completed six waves of post-election (face-to-face) surveys for central elections (2001, 2004, 2008, 2012, 2016, and 2020) and seven waves for local elections (2002, 2005, 2006, 2009, 2010, 2014, and 2018; data for 2022 is still unavailable) in Taiwan as of January 2024. The two survey questions of concern in this article would be: 1. “How were you/your acquaintances canvassed?” and 2. “Who influenced your vote most?” The first question was included in surveys for both levels of elections, central and local, and the second one was asked only for the local-level ones.

It is also noteworthy that since the questions were designed in a slightly different way across waves, between branches of government, and over divergent levels of elections, their results are hard to compare for five reasons. First of all, in executive elections (e.g., mayoral and presidential elections), these questions were asked candidate by candidate, while in legislative elections (e.g., LY and city councils), they were asked as a whole. Therefore, there are often more than one set of answers for one wave of elections. Secondly, these survey waves also varied in terms of the approaches used to structure the questions. Some waves required respondents to answer yes or no for each option, but in others respondents would be asked to fill out all the options that applied. As a result, data from different waves come in different formats. Thirdly, answer options were also not the same across waves and between different levels and branches of elections. For example, internet-related GOTV methods were not included as answer choices until the 2010s. Consequently, longitudinal studies of some such specific options are hard to do. Fourthly, local post-election surveys before 2010 were conducted in different jurisdictions^[1] since the schedule of all local elections had yet to be unified completely. The geographical (e.g., Taoyuan County vs. Taipei City) and institutional (county vs. city and special municipality) heterogeneities between jurisdictions sampled in different survey waves became a huge barrier to our making a systematic attribution for those trends observed in the data across the years. Finally, the aforementioned complications were simultaneously aggravated by institutional reforms at different moments at different levels. There were mainly two institutional shocks involved across the period. The first one would be the reform of legislative elections in 2008 as a consequence of the constitutional amendments of 2005. The seat count was reduced from 225 to 113, while seats in geographical constituencies after the reform were elected by a plurality of votes from single-member districts and no longer by single-non-transferable-votes (SNTV). The second shock would be the municipal reforms in 2010 and 2014. Four new special municipalities (New Taipei, Taichung, Tainan, and Taoyuan) were created and one already-existing special municipality (Kaohsiung) was expanded, through the combination of existing localities or the promotion of a single locality. The reforms centralized local political power in what were formerly county areas by rearranging autonomous townships in these areas, where leaders had previously been elected, into appointment-based districts where heads were designated by their corresponding mayor.

In response to the first and second issues, based on the assumption that the variations in question format did not affect the responses, I combined the answers for each question across different candidates within a single election in the following way. For a given subject, when option A appeared in the answer for either one of the candidates, option A was classified as having been selected in that observation. While the third and fourth issues could hardly be addressed effectively on the researcher’s end, in this study I just ignored the possibility of institutional and geographical heterogeneity (i.e., counties, cities, and special municipalities were regarded as the same sort of entity) and pooled all the data by wave, election level, and branch for ease of analysis.

With these constraints noted, in the following two sections, I will focus more on the rankings, rather than the percentages of these options, within a wave and their changes over the years, since the adoption of rankings instead of absolute numeric differences can partially alleviate the afore-discussed measurement issues and make the relative importance of the different options more comparable across a time series.

Central Elections

Beginning with the results for the presidential and the LY elections, Figures 1 and 2 display the grid plots of bar charts that summarize the top five ways voters (or their acquaintances) were canvassed across all the presidential elections since 2004 and the LY elections since 2001, respectively. (In 2001, subjects were only surveyed as to how they were canvassed by LY candidates, not by the presidential candidates; hence, only the LY data was available for 2001,^[2] and the presidential data starts in 2004.^[3])

Figure 1.Top five ways for presidential candidates to approach voters (2004–2020)

Figure 2.Top five ways for Legislative Yuan candidates to approach voters (2001–2020)

Among all the GOTV approaches, recorded phone message (listed as “phone recording”) or phone call was one of the most common GOTV encounters for voters in central elections. In LY elections, recorded phone message or the use of phone remained one of the top three ways to canvass across the years (except for 2008, where use of phone was not an answer choice for respondents): ranked first in 2001, 2016, and 2020, second in 2004, and third in 2012. Its popularity was similar in presidential elections from 2004 to 2012 (all as the top option), but it suddenly plummeted starting in 2016. As suggested by previous academic research, the decreased abundance of recorded phone messages in later presidential but not LY elections could be explained by the method’s diminishing efficiency when not strategized: it is easier for LY candidates than presidential candidates to personalize their GOTV initiatives due to the smaller size of the LY constituencies.

Considered to be the most effective by the GOTV scholarship, personalized GOTV means are consistently adopted by central-level candidates. The use of campaign workers as a GOTV tool remained steadily prevalent among all the central elections across the years. In LY elections, whenever campaign workers was kept as an option (2001, 2012, 2016, 2020), it always came in either first (2012) or second (2001, 2016, and 2020) place. In presidential elections, it has always been among the top five means for voters to be canvassed since a question about it was included: fifth in 2004, fourth in 2012, third in 2008, and first in 2016 and 2020. A similar tendency can be observed in face-to-face canvassing by candidates themselves for both types of central elections. It remained at least among the top four in both LY and presidential elections whenever it was included in the question as an answer choice. Since 2016, the option has been removed from the presidential question, but it still ranked high (third in 2016 and 2020) in the one concerning the LY. Apart from that, village or neighborhood chiefs, though not as salient as other aforementioned means of campaigning, also deserves our attention. Appearing in almost all waves (except for the 2004 LY survey), it often reached the top five GOTV means (third in the 2016 and 2020 presidential, fourth in the 2008 LY, and fifth in the 2001, 2012, and 2016 LY surveys), indicating that it was still an indispensable part of campaigning for central elections, likely due to the value of the tacit local knowledge on the part of village or neighborhood chiefs, especially for remote constituencies.

On the other hand, potentially due to the fragmentation of the social network structure, led by deepening urbanization (Inglehart and Welzel 2005), the electoral salience of acquaintances, like blood relations (relatives or family, RoF hereafter) and other-than-blood relations (schoolmates or friends, SoF hereafter), gradually decreased as time went by. For presidential elections, “family or relatives” and “schoolmates or friends” took second and third place, respectively, in 2004 and fourth and fifth place in 2008. But after 2012, in which wave both were temporarily excluded as options, i.e., in 2016 and 2020, only less than 0.6% of the respondents were approached in one of these ways, not even close to being among the notable ranks. Their trends in the LY elections were similar. In 2001, RoF and SoF were in fourth and sixth place, respectively. After the decline of the “friend” option and the temporary exclusion of the RoF option in 2004, in 2008 RoF came back a bit as two separate options that got to second and third place. But starting in 2012, both options faded out gradually as key GOTV means, until SoF rose slightly, to fifth, in 2020.

Finally, a role for “candidate support club,” particularly in the presidential campaigns, has begun to emerge. First asked about in 2012 (coming in at sixth place), it went up to the second most common campaigning approach (after campaign workers) encountered by voters in the 2016 and 2020 presidential campaigns. The rise of candidate support clubs could be regarded as a case of the emergence of alternative GOTV means developed to serve the relentless demand for more personalized GOTV activities. This phenomenon well-echoes what GOTV scholars have found in their research works.

Local Elections

After having observed the trends for GOTV encounters in central elections, how about those for local levels? Figure 3 and Figure 4 showcase the five most common means to canvass in mayoral and council elections. From these figures, we can see some similarities between the local and central elections. First of all, campaign workers and recorded phone messages (listed as phone recording) were almost always among the top three most common encounters in both city/county council and mayoral elections, mirroring how popular they were in the central elections (except for the use of phone calls in presidential campaigns). In council elections, campaign workers were the most prevalent canvassing method from 2002 to 2014 and still the runner-up in 2018, while recorded phone messages wandered between second (2006), third (2002, 2005, 2010, 2014), and fourth (2009) place before it reached the top in 2018. The popularity of campaign workers coincides with the insight of the GOTV literature that emphasizes electorates’ preference for some personal touch.

Figure 3.Top five ways for mayoral candidates to approach voters (2002–2018)

Figure 4.Top five ways for councilor candidates to approach voters (2002–2018)

Secondly, like the central-level candidates, face-to-face canvassing persisted as a key GOTV activity for councilor candidates but was not as popular for the executive heads (mayors). This incongruence can be attributed to the limited supply issue: local executive candidates usually have a larger constituency than the local legislative ones, and therefore it is harder for them to reach out to their entire electorate in person. Face-to-face canvassing ranked either second (2002, 2005, 2009, 2010, and 2014) or third (2006 and 2018) in council elections. For mayoral elections, it only hit the top five in 2005 (second), 2009 (third), and 2010 and 2018 (fifth), confirming its shrinking scope as a means of outreach in executive elections at both central and local levels.

Thirdly, acquaintances and village and neighborhood chiefs are all notable as people who helped canvass across these local elections. In mayoral elections, either or both SoF (third in 2002, 2006, 2010, 2014, and 2018 and fifth in 2005) or/and RoF (fifth in 2002 and 2006 and fourth in 2010, 2014, and 2018) ranked among the top 5 in all elections but one (2009). As for village or neighborhood chiefs, although they were relatively less pronounced than other popular options, they still played a role in city/county-level campaigns, as denoted by their occasionally showing up among the top five canvassing encounters in local elections (2002 in mayoral contests and 2002, 2005, 2014, and 2018 in council contests).

We now turn to a question regarding who influenced the survey subjects’ vote the most in local elections (see Figures 5 and 6). The patterns are very stable over time in both mayoral and council elections. Throughout the years, in both types of local elections, the most popular key vote influencer was always the subjects themselves. Around 40–50 percent of them decided whom to vote for on their own. The second most common key vote influencer over the years was family. About 20 to 30 percent of the respondents found their family members decisive for their vote choice. Candidates and friends ranked third and fourth, respectively, most of the time, except in the 2009 councilor elections when they swapped positions. However, usually only less than 10 percent of respondents considered either one of those to be influential in terms of the voting decision.

Figure 5.Top five types of people who influenced mayoral voters most (2002–2018)

Figure 6.Top five types of people who influenced councilor voters most (2002–2018)

Conclusion

To conclude, this article provides a preliminary review of canvassing tendencies in Taiwan over the last 20 or so years at both central and local levels using the TEDS data. The trends were generally consistent with the scholarship on electoral campaigning that stresses the electoral efficiency of personalized GOTV engagements. Over the two decades of democratic elections at different levels, campaign workers and recorded phone messages have had the largest scope of outreach among all the GOTV strategies. While the use of campaign workers was more common across all types and levels of elections, phone calls were adopted only when the size of the constituency was small enough for campaigners to tailor the contents of the calls to their target audiences, regardless of the levels and the types of elections (which could be why presidential candidates abandoned the method after 2012). Face-to-face canvassing was used as well, but this type of encounter becomes less common as the size of constituencies goes up, due to the limited-supply issue. The data also revealed that it was common for acquaintances, such as friends, schoolmates, family, and relatives, to approach their social circles and canvass votes for their preferred candidates, although it has become less common in recent central elections. As a more effective alternative, the rise of candidate support clubs in presidential campaigns lately has provided a different personalized GOTV engagement experience. At the same time, village and neighborhood chiefs have been indispensably involved in canvassing at both the central and local levels, even though they were not as prominent as other means. In terms of key vote influencers at local levels, while almost half of voters would make their own choice, still around 20 percent of them were swayed by their family members. Finally, due to the fact that these surveys have been conducted in different ways for the same question, numerical comparisons between waves of data should be taken with caveats. In the future, more rigorous studies will be required to figure out the origins and the ramifications of these GOTV initiatives.

Corresponding Author

Sanho Chung, sanhochung@arizona.edu
School of Government and Public Policy, University of Arizona