Is class voting emergent in South Korea? While we know that age and region have highly significant effects on voting in Korea, this still leaves open the question of whether there is evidence of class voting and the extent of its substantive significance. This question is important for two reasons. First, inequality has risen in Korea, raising the question of whether class voting might have become salient despite the prevalent notion that class does not matter in Korean politics. Second, there is a debate about reverse class voting: circumstances in which low-income individuals in fact vote for right parties more than high-income voters. The question of class voting has intrigued many scholars of Korean politics in the midst of rising socioeconomic inequality, and a number of studies found the absence of class voting or the existence of reverse class voting. While some studies found evidence of class voting, their findings were restricted to subjective class identity or particular segments of the population.
In this research note, we seek to understand whether class voting is emergent in Korean elections and what substantive effect class has on voting choices. We critically examine the findings of extant studies and offer a new empirical investigation on class–vote ties over six elections between 2004 and 2014.
First, we recognize that the conceptualization (and consequently the operationalization) of social class in existing studies remains highly tentative, ranging from income levels to occupational classifications, home ownership, and self-reported class identity. Social classes can be conceptualized in various ways, as demonstrated by class voting literature in advanced democracies.Footnote 1 Of two prominent class schemas, Erickson and Goldthorpe (Reference Erikson and Goldthorpe1992) use four dimensions (employment status, manual and non-manual occupations, agricultural and non-agricultural occupations, and service and labor) to propose 11 categories. Erik Olin Wright (Reference Wright1997) takes a more materialist approach to social classes, highlighting material interests generated by relationships to authority and skills, and suggests 12 classes. Yet, in their empirical analysis the class categories shrink to five or seven due to data limitations. We concur that voters’ partisan choice emerges from their economic interests, which cluster as a result of shared circumstances in incomes and employment conditions (Chan and Goldthorpe Reference Chan and Goldthorpe2007). But we face similar issues if we use complex class categories in our analysis, because the currently available survey data do not include such information in a consistent way over multiple elections.
As an alternative, we use income as a simple and straightforward representation of objective class position (Hout, Brooks, and Manza Reference Hout, Brooks and Manza1993). Income is the most direct indicator that shows the scope of disposable material resources that lock people in a certain social status. This is an especially valid approach in the Korean context because the labor market is highly fractured and even the same occupation offers variegated material conditions. For instance, manual workers are divided by the firm size, by their employment in the primary firm or subsidiaries, and by regular or irregular employment status, and their incomes greatly vary by their employment categories. Moreover, the lower strata of the self-employed, whom Wonik Kim (Reference Kim2010) categorizes as petty bourgeoisie, earn as little as irregular workers in small firms and see themselves as low-income class (Yoonkyung Lee Reference Lee2014).
Thus we operationalize class in two ways in our statistical analysis. First, we use income as the best available measure of objective class in our survey data. Second, we use self-reported class position as an indicator of subjective class identity. We posit that one's objective class position, defined by income, influences her/his subjective class position, which in turn affects her/his vote choice in elections.
Using the Korean General Social Survey data (KGSS 2004, KGSS 2010, KGSS 2011, KGSS 2012, KGSS 2013, and KGSS 2014), our own empirical investigation shows that Koreans do engage in class voting, although the magnitude of the class effect is smaller than that of region and age.Footnote 2 We find that previous studies have failed to adequately control for age effect and examine the interaction effect between age and class. Class measured by income levels as well as by subjective identification is found to have a significant impact on vote choice when region and age are controlled for. While regional and generational cleavages still dominate electoral politics in Korea, class has become another factor in vote choice, although its effect is rather weak. In particular, we demonstrate that the poor electorate does engage in class voting. This finding overturns those of existing studies that claim poor voters are responsible for reverse class voting.
ELECTORAL CLEAVAGES AND CLASS VOTING IN KOREA: DEBATES AND CONTRASTING FINDINGS
Class voting is premised on the notion that class affects citizens’ vote choice and expects that social classes would support a political party that best represents their economic interests. In its simplest form, working-class voters are expected to cast their ballot to leftist parties that advocate redistributive policies financed by high levels of taxation on the wealthy, whereas upper-class voters are expected to endorse conservative parties that promise strict protection of individual property rights via low levels of taxation.
Yet, the dominant marker of Korea's electoral politics since the democratic transition in 1987 has been regional cleavages under which voters cast their vote for the political party/candidate that represents their home region (Kap-yoon Lee Reference Lee1998). Also, generational cleavages have been identified to become increasingly salient since the early 2000s. However, the question of class voting in Korea remains under intense debate. On the one side, many scholars suggest that there is a persistent absence of class voting or that poor voters support conservative parties more than rich voters (i.e. reverse class voting). For instance, the low-income group's support for the conservative candidate in presidential elections increased from 51.8 percent for Lee Hoi-chang (as opposed to 46.1 percent for Roh Moo-hyun) in 2002 to 60.5 percent for Park Geun-hye (as opposed to 39.5 percent for Moon Jae-in) in 2012 (Han Reference Han2013). Given the rising socioeconomic inequality in Korea, which is presumed to create a fertile ground for class politics, observers are puzzled by the absence of class voting or the persistence of reverse class voting (Park Reference Park2009; H. Kwon Reference Kwon2010; Son Reference Son2010; Kim and Yeo Reference Kim and Yeo2011; Kap-yoon Lee Reference Lee2011; Han Reference Han2013; Won-taek Kang Reference Kang2013; W. Ko Reference Ko2013).
Various attempts have been made to explain the peculiar phenomenon of reverse class voting. Since the elderly make up the majority of the poor electorate in Korea, conservative voting by older voters has been identified to be accountable for reverse class voting (Han Reference Han2013; Won-taek Kang Reference Kang2013; Ko Reference Ko2013; Cheon and Shin Reference Cheon and Shin2014). Won-taek Kang (Reference Kang2013) finds that reverse class voting disappears once those aged 60 or over are removed from the sample. Han (Reference Han2013) and Won Ko (Reference Ko2013) argue that reverse class voting appears only among the older electorate, or those aged 50 or older.
On the other hand, some scholars argue that class cleavage is emergent in recent elections, albeit the scope of their empirical analyses is limited (Chang Reference Chang2013; Yong-ma Lee Reference Lee2014a and Reference Lee2014b). In an analysis of the 2002 presidential election, H. Kwon (Reference Kwon2010) shows evidence of class voting among middle-income voters compared to low- and high-income voters. However, this study can be interpreted as evidence of class voting among the affluent and reverse class voting among the poor. In a study of two legislative elections of 2000 and 2004, Wonik Kim (Reference Kim2010) also finds the class–vote link by using occupational classifications for social classes. Although Kim's four class categories—professional class, non-manual class, petty bourgeoisie, and working class—present a noteworthy approach, the validity of such operationalization is questionable given the diversity of the material conditions within each occupational class. Yong-ma Lee proposes that class voting appears among the upper class who voted conservatively and the new middle class who supported the liberals in both presidential and local elections (Reference Lee2014a; Reference Lee2014b). However, none of the existing studies finds evidence of class voting among low-income voters.
Another stream of studies advocates the importance of real estate property and geography in understanding the meaning of class politics in Korea. They focus on election results in specific geographies such as the three wealthy districts in Gangnam, Seoul, or traditional industrial towns with a large number of blue-collar workers such as Ulsan and Changwon.Footnote 3 Voters in Gangnam districts who boast the highest land/real estate prices as well as the highest incomes are found to consistently vote for the Saenuri Party (Son Reference Son2010).Footnote 4 Kap-yoon Lee et al. (Reference Lee, Lee and Kim2013) also use assets and home ownership to operationalize class and find significant evidence of the effect of wealth on class consciousness, but only sporadic evidence of class–vote ties in recent elections. A study by Woojin Kang, however, shows that the effect of home ownership was insignificant (Reference Kang2013). In contrast, voters in Ulsan and Changwon, industrial towns with a strong labor movement basis, delivered electoral victories to the progressive Democratic Labor Party (DLP) candidates, reminding us of the importance of labor unions in articulating and mobilizing class interests (Cho Reference Cho2006).Footnote 5 Yet, the focus on home ownership or certain district constituencies in some elections is inherently limited due to the lack of comparable and consistent data nationally or over multiple elections.
Another finding of class voting in recent studies is when class is approached as subjective “class identity” instead of objective measures such as income, occupation, or property ownership (Cheon and Shin Reference Cheon and Shin2014; Seo and Han Reference Seo, Han, Kap-Yoon and Hyun-Woo2014; Hee-Kyung Kang Reference Kang2016). Since the incongruence between objective class position and subjective identity is feasible, it makes more sense that voters would engage in class voting when they have developed subjective class identity.Footnote 6
The inconclusive and even contradictory findings briefly summarized here manifest several analytical deficiencies in the existing scholarship of class voting. First, most of these studies are based on survey data from only one or two elections. Such a narrow scope of empirical examination risks the danger of over-generalization.
Second, even those works that use more extensive data to cover several elections, such as Kap-yoon Lee (Reference Lee2011), Cheon and Shin (Reference Cheon and Shin2014), and Yong-ma Lee (Reference Lee2014a; Reference Lee2014b), produce inconsistent results.Footnote 7 Kap-yoon Lee (Reference Lee2011) finds an absence of class voting, while Cheon and Shin (Reference Cheon and Shin2014) find the presence of class voting only in terms of subjective class identity. Yong-ma Lee (Reference Lee2014a; Reference Lee2014b) finds class voting only among the upper class and the new middle class. These conflicting findings motivate this research note to conduct a systematic empirical investigation that properly captures the causal effects of various factors such as class (objective and subjective), age, region, education, and gender, which all can influence voters’ partisan choice. Of particular importance is to clearly isolate the possible interaction effects between class and age or region in order to reach valid and reliable findings.
Is class voting emergent in Korea? Empirical analysis
Our study of the class–vote association in Korea uses the Korean General Social Survey, an annual survey conducted since 2003. The survey data include two legislative elections in 2004 and 2012, two presidential elections in 2007 and 2012, and two rounds of nationwide local elections in 2010 and 2014.Footnote 8 As such, this empirical analysis is expected to be the most comprehensive statistical examination on the correlation between social class and vote choices in Korea.
We use two income-based measures and a subjective measure of class. The income variable in the KGSS data denotes monthly household income. The continuous income variable is not available for every year, so we use the ordinal income variable that ranges between 1 and 22.Footnote 9 Since the ordinal scale has no theoretical meaning or particular relationship with class or voting pattern, we have regrouped it into either three income classes (the poor, the middle-income class, and the rich) or six income classes (upper-upper, lower-upper, upper-middle, lower-middle, upper-lower, and lower-lower). The monthly median income for Korean households in 2013 is slightly over 4 million won (about USD 3,800). Following the conventional definition of the middle class, those who make between 50 percent (2 million won) and 150 percent (6 million won) of the median income are considered the middle-income class.
Income class 1 (denoted simply as Income): 1 to 6 (unit: won per month)
1 less than 1 million
2 1 million–less than 2 million
3 2 million–less than 3.5 million
4 3.5 million–less than 6 million
5 6 million–less than 10 million
6 10 million or over
Income class 2: Poor, middle, and rich (unit: won per month)
Poor less than 2 million
Middle 2 million—less than 6 million
Rich 6 million or over
Subjective class identity (denoted simply as Class): The respondents ranked their class position between 1 and 10, and we have regrouped it into six subjective classes (upper-upper, upper-lower, middle-upper, middle-lower, lower-upper, and lower-lower). Few respondents selected ranks below 3 or above 6. So, ranks 1 and 2 were grouped into class 1 (lower-lower), ranks 7–10 were grouped into class 6 (upper-upper), and ranks 3, 4, 5, and 6 were respectively converted to class 2 (upper-lower), 3 (lower-middle), 4 (upper-middle), and 5 (lower-upper). The resulting distribution is as follows: 1 (9.0%), 2 (13.7%), 3 (18.0%), 4 (29.4%), 5 (16.2%), and 6 (13.7%).
Ideology: Respondents chose their political orientation between 1 (very progressive) and 5 (very conservative).
Education: We regrouped the survey's 8 categories into 7 categories and the distribution is as follows: 0 (no education: 4.5%), 1 (primary or Chinese classical: 10.1%), 2 (middle school: 8%), 3 (high school: 29.9%), 4 (professional school: 12.3%), 5 (4-year college: 30.3%), and 6 (postgraduate: 5%).
Age: We use a cubic function of age to capture its nonlinear effect accurately.Footnote 10 When we examine the interaction effect between age and income class, however, we use for the sake of simplicity dummy variables Young (below 50) and Old (60 or over), with the fifties as the reference group. We use interaction terms such as Young*Poor, Young*Rich, Old*Poor, and Old*Rich.
Region: We included three regions, Chungcheong, Honam, and Youngnam, as dummy variables to isolate the effect of regional voting. Reference category is all the other regions.
Conservative voting, or voting for Saenuri (denoted simply as Vote_Sae): The major conservative party during the period of our investigation was Grand National Party (GNP) until 2011 or Saenuri Party from 2012 to 2014.Footnote 11 Voting for Saenuri is coded 1 for those who voted for the GNP or Saenuri candidate, while those who voted for other party candidates or declined to identify whom they voted for are coded as 0. Those who abstained and those who are disenfranchised are excluded from the analyses.
Table 1 presents the percentages of conservative voting (Vote_Sae) and the average ages for six income classes. It shows that the percentages of conservative voting from the lowest income groups 1 (below 1 million won) and 2 (from 1 million to below 2 million won) are 57.7 percent and 52.8 percent, respectively, much higher than the population average of 49.5 percent. This gives an impression of reverse class voting. However, these groups also have much higher average ages (65.6 and 52.7) than other income groups (between 42.0 and 44.7), indicating a concentration of poverty among the elderly. This corroborates the fact that 49.3 percent of Korea's senior citizens are poor, which is the highest poverty rate for the over-65 age group among the OECD nations and approximately four times the OECD's average elderly poverty of 12.6 percent (OECD 2016). Considering such a highly negative correlation between age and income (r=−.41 in Appendix Table A2), we need to separate the effects of age and income on vote choice.
Conservative voting and average age by income class
Now we turn to logistic regression analyses of social class and partisan voting, controlling for various factors such as regional and generational cleavages. As summarized in Table 2, the significance of “Income” (six income classes) effect varies depending on the specification. The coefficient for “Income” is positive but not significant at 10 percent level if we use a linear function of age (not presented), but it becomes significant if a quadratic or cubic function of age is used (model 1).Footnote 12 Because of the very strong age effect on vote choice and the high correlation between age and income, the estimated effect of income is affected by how well age effect is captured. Previous studies have failed to find a significant effect of income on voting or the presence of class voting because of inadequate control for age effect. The marginal effect of “Income” for a typical person, or increased probability of conservative voting as a typical person's income increases by one ladder (from the lower-middle to the upper-middle income class, holding other variables constant at their means), is 1.6 percentage points, according to model 1.Footnote 13
Pooled analysis of determinants of conservative voting: Logit results
Note: Results for election dummies and constants are not presented. Standard errors are in parentheses. Significance levels are denoted as follows: + p < .10, * p < .05, ** p < .01, *** p < .001 (the same for the rest of the Tables).
In models 3–4, we use the dummies for three income classes, using the middle-income class as the reference category. The coefficient for “Poor” consistently appears with a negative sign and with statistical significance in both models, while that for “Rich” is not significant. The marginal effect of “Poor” for a typical person on conservative voting is −4.3 percentage points, according to model 3. This finding is very different from those of previous studies, which have found no evidence of class voting or even evidence of reverse class voting by the poor. When we use the dummies for six income classes instead of an income class variable that ranges from one to six (model 1) or the dummies for three income classes (model 2), we confirm that class voting effect is concentrated on the lowest income class rather than the middle- or upper-income classes. Table 3 shows that changes in the predicted probability of conservative voting due to changes in income class, holding other variables constant at their means, are significant for any upward changes from the lowest income class. An upward change from the lowest to the second lowest income class is associated with 6.1 percentage points higher probability of conservative voting, and that from the lowest to the highest income class with 12.9 percentage points. However, most of the other upward changes are not associated with significant changes in the probability of conservative voting. This indicates that the lowest income group supports the conservative party much less than other groups when age and other effects are adequately controlled for.
Changes in the probability of conservative voting from changes of income
Note: The above statistics are based on a logit regression slightly modified from model 1 in Table 2 (using the dummies for 6 income classes), holding other variables constant at their means.
Models 5 and 6 examine the interaction effect between age and income. The coefficients for the interaction term “Old*Poor” are significantly negative. This indicates that the old-poor are significantly less likely to support the conservatives than the old-middle. While the coefficients for “Old*Rich” are also significantly negative, the positive coefficients for “Rich” and the negative coefficients for “Old*Rich” cancel out for the old-rich. Table 4 presents the predicted probabilities of conservative voting by income and age, based on model 5. Among the young (below age 50) and those in their fifties, the differences between the predicted probabilities of conservative voting by three income classes are not statistically significant, at 5 percent. Among the old (age 60 or higher), however, the 10.6 percentage points difference between the poor and the middle-income group is significant so that their 95 percent confidence intervals do not overlap. Overall, income effect (class voting based on income) is generally stronger for the poor, particularly for the old-poor, than for the rich. It is notable, however, that the income effect is relatively small compared to the large effect of age. Table 4 shows that within each of the three income groups, the differences in the predicted probabilities of conservative voting by age group are very large and significant.
Predicted probabilities of conservative voting by income and age
Note: Predicted probabilities of conservative voting by income and age, setting other variables at their means. 95 percent confidence intervals in brackets.
Subjective class and ideology are highly significant for conservative voting (models 2, 4, and 6 in Table 2), with marginal effects of 2.4 and 15.7 percentage points, respectively (model 2). The coefficient for “Income” or “Poor” declines and often becomes statistically insignificant when subjective class and ideology are included. This suggests that the income class effect partly runs through its effect on subjective class and ideology. The highly significant and strong effect of income on subjective class identity is confirmed from the OLS regression of subjective class, controlling for various factors (Appendix Table A3). This is consistent with previous studies that show income as by far the most important determinant of subjective class in Korea (Byeong-jo Kim Reference Kim2000; Kwang-min Seo Reference Seo2009; Chang Reference Chang2013; Song, Lee and Choi Reference Song, Lee and Choi2013).
Table 5 shows that income class is a significant determinant of ideology (models 2 and 3). Higher income is associated with more conservative political orientation, and the coefficient for income is larger and more significant in model 3 than in model 2, because the age effect is more fully captured in model 3 than in model 2. The rich are significantly more conservative than the middle-income class, while the poor are less conservative on average than the middle-income class (model 4). However, income effect on ideology is weaker than the age effect. The standardized coefficient for income is 0.03, while that for age is 0.18 in model 2 (not reported in Table 5). Also, income effect on ideology is much weaker than its effect on subjective class identity (standardized coefficient = 0.31, from Table A3). Hence, income effect on voting seems to run substantially through subjective class and only weakly through ideology.
Correlates of ideology (1 = very progressive to 5 = very conservative): OLS regressions
Data From: Data for model 1 is from KGSS (2011). Data for other models are from KGSS (2004–2014).
Note: Results for constants are not presented. Robust standard errors are in parentheses.
Considering the strong effect of ideology on voting, we further attempt to identify what issues are most salient in forming Koreans’ ideological orientation. We utilize the 2011 KGSS data for this purpose because the survey asked several relevant questions such as the respondents’ attitude toward North Korea, income inequality, and government responsibility for social welfare. Model 1 in Table 5 shows that attitude toward North Korea is by far the best indicator of ideology. Those who have negative attitudes toward North Korea (“We should be wary of or hostile toward North Korea”) are significantly more conservative than those with positive views (“We should support or cooperate with North Korea”), with an average of 0.33 points higher value in ideology on a 1 to 5 scale (1 representing very progressive and 5 representing very conservative). Opinions on income inequality and government responsibility for social welfare are insignificant or only marginally significant as an indicator of ideology in Korea. Those who agree that the income gap between the rich and the poor is too large and that government should take responsibility for the provision of social welfare tend to locate themselves on the more progressive side of the ideological spectrum, but the effects are only marginally significant at 10 percent level. These findings suggest that policy positions toward North Korea rather than socioeconomic issues largely determine voters’ ideological orientation as well as party politics in Korea.
In Table 6, we show the logit results by election, which includes the 2004 and 2012 legislative elections, the 2007 and 2012 presidential elections, and the 2010 and 2014 local elections. In the upper panel, the income effect is significant in the 2004, 2007, and 2010 elections but becomes insignificant in other elections with positive signs maintained except for the 2012 presidential election. In the middle panel, “Poor” has a negative sign in five elections except for the 2012 presidential election. Thus, the poor were less likely to support the conservative candidates. In the lower panel where both income and subjective class as well as ideology are included, subjective class is significant in four out of six elections. Ideology exerts strong and highly significant effect in every election. While income loses significance in most elections, it is presumably due to the indirect effect of subjective class identity.
Logit Results by Election (2004 Legislative through 2014 Local Election)
Note: Results for control variables (age, age squared, age cubed, education, female, Chungcheong, Honam and Youngnam) and constant are not presented.
Overall, these results by election show the presence of class voting gauged by both objective income class (especially among the poor) and subjective class. However, we did not find evidence of an increasing trend in class voting. The 2012 presidential election in particular presents an exception as the coefficient sign for income variables flips, while subjective class effect is consistent and significant. Some previous studies claim to have found reverse class voting by the poor using data from this election; however, the evidence is too weak to support such a claim (Han Reference Han2013; Ko Reference Ko2013). Of the three panels, the coefficient for income variable (Income, Poor, or Rich) was marginally significant only in the lower panel, which is far from conclusive evidence for reverse class voting even for this particular election. However, it may suggest that the conservative presidential candidate, Park Geun-hye, was successful in courting votes from the poor by representing herself as a champion of economic democracy and social welfare and promising to give every elderly person a basic pension of 200,000 won (about 200 USD) per month. We should not read the results of this election as a sign of declining class voting but as a piece of evidence that class voting can also be substantially affected by the supply side, i.e. the party and candidate strategies (Evans and Tilley Reference Evans and Tilly2012).
Concluding thoughts
For some scholars and observers, Korea might be the last place to look for evidence of class voting. Explanations can be exhaustive because Korean politics has been highly repressive of “class” both historically and institutionally. Still, this study finds that class voting is emergent in Korea when class is measured by income and subjective class identity. We also find that income is the most crucial factor that forms the basis of voters’ subjective class identity. This implies that, under rising economic inequality, material conditions gauged by a simple income level galvanize individuals’ class standing and facilitate the formation of class identity.
These findings mark a departure from the conclusions of existing studies on class voting in Korea, which have maintained that subjective class identity, not income, is the crucial determinant of the class–vote association (Chang Reference Chang2013; Cheon and Shin Reference Cheon and Shin2014; Kap-yoon Lee et al. Reference Lee, Lee and Kim2013). While we realize that the misconception of reverse class voting is created by the concentration of senior citizens in the low-income group, we find evidence for class voting by the poor rather than the mere absence of reverse class voting by the poor (Won-taek Kang Reference Kang2013). We demonstrate that the elderly poor are less likely to vote for the Saenuri Party compared to more affluent senior voters.
However, we should acknowledge that class voting is rather weak, compared to regional and generational voting patterns. We are unable to discern any trend of class voting (its rise or decline) during the ten-year period (2004–2014) of our study, and comparable data for earlier period is missing. Also, we recognize that ideological orientation, which has a strong effect on vote choice, is largely defined by attitudes toward North Korea rather than positions on redistributive issues. For a fuller explanation on how class operates in Korean voters’ partisan choice, we need to add the supply side of class voting—how political parties articulate class issues in electoral completion. Indeed, the 2012 presidential election, in which the conservative candidate obtained substantial support from the poor (although the coefficient was not statistically significant), seems to suggest the importance of investigating the supply side of class politics. However, our study was unable to rigorously examine this issue due to the limits of available data.
Although the electoral mobilization of the left was traditionally stunted in Korea, the political language to capture the realities of rising economic disparities and class differences has expanded in public discourse in recent decades. An important change in this regard was the Democratic Labor Party's entrance into the National Assembly in 2004. With the formal presence of the labor party and the growing pressure from civil society articulating economic disparity, established parties adjusted their policy positions. The Democratic Party adopted “universal social welfare” in its party manifesto in 2010. In reaction, even the conservative Saenuri Party came to advocate the expansion of social welfare programs. In other words, Korean political parties began to engage in programmatic debates of economic redistribution during electoral campaigns. While further studies are required to gain a full understanding of the specific processes through which class, class identity, and class voting are forged, material hardship seems to galvanize the old language of class and class politics even in places like Korea that has long been dominated by conservative political forces.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/jea.2019.10.
