Cooperate But Divided at Heart: Analysis of an Opposition Elite Survey During Autocratization

When opposition elites cooperate to counteract authoritarian incumbents, they often deemphasize their fissures. But behind the scenes of cooperation, will their attitudinal divisions remain? If so, on what dimensions? Though these questions have important implications for the tenacity of cooperation and future political developments, the current literature does not shed much light on them. Our research answers these questions by examining the case of Hong Kong, which has undergone widespread repression. We surveyed district councilors in an autocratization episode in which opposition elites were highly cohesive. Using the Kolmogorov–Smirnov metric, we find that opposition factions share similar economic attitudes. Yet, despite their cooperative behavior, they still hold divided attitudes on the center-periphery and democratization strategies dimensions. These findings contribute to the burgeoning literature on autocratization by highlighting the (in)congruence of opposition elites’ attitudes and behavior.


Introduction
The literature on hybrid regimes has long established that there are cleavages within opposition camps (Gandhi and Reuter, 2013;Greene, 2007;Howard and Roessler, 2006;Magaloni, 2006).For one thing, opposition elites hold different ideologies as they represent different classes, ethnicities, religions, and territories (Cannon, 2004;Gandhi and Ong, 2018;Selçuk and Hekimci, 2020).For another, opposition elites can be divided because of the democratization strategies they favor, like the use of violence, compliance with laws, and election boycotts (Buttorff and Dion, 2016;Ortmann, 2009;Smith, 2013).But despite these cleavages, cooperation is still possible to counter authoritarian incumbents.According to previous research, cooperation behavior among opposition factions is likely when the regime is debilitated (Ong, 2022;Wahman, 2013).Once different factions choose to cooperate, they often deemphasize their fissures and instead prioritize the goal of anti-authoritarian incumbents (Ong, 2022).However, behind the scenes of cooperative behavior, will their attitudinal divisions remain?If so, on what dimensions?
In this article, we distinguish opposition elites' attitudes and behavior and highlight that the two notions need not align with one another.Understanding the (in)congruence of attitude and behavior among opposition elites is an important topic since it has profound implications for the tenacity of elites' cooperative behavior and the political developments of hybrid regimes.For instance, in studying the diverging transition paths between Tunisia and Egypt, Nugent (2020) argues that even though opposition elites cooperated to topple the regime, their polarization heavily affected their willingness to compromise and further cooperation after the regime collapsed.Although this work inferred elites' polarization from interviews and did not measure their self-reported attitude, it directs us to a crucial insight.That is, even though opposition elites cooperate to counteract the regime, their attitudinal divisions (or their absence) do matter because they can shape opposition elites' willingness to compromise and cooperate further.If the attitudes of opposition factions are heavily divided on certain dimensions, their cooperative behavior is likely to be short-lived.In turn, the hybrid regime can co-opt one faction over the others in future episodes.Contrarily, if elites' attitudes are unified on different dimensions, their cooperative behavior is likely to sustain, which can give the authoritarian incumbents a tougher resistance.However, despite the importance of analyzing the (in)congruence of attitude and behavior among opposition elites, the current literature does not shed much light on this topic.
Our research fills this gap by analyzing Hong Kong's opposition elites in an autocratization episode-from the beginning of the anti-Extradition Law Amendment Bill (anti-ELAB) movement to the enactment of the National Security Law (NSL).In this autocratization episode, the hybrid regime no longer used targeted repression as before, but instead adopted widespread repression.In response, opposition elites from different factions were strongly cohesive and cooperative (Lee, 2020;Leung et al., 2022).To test the (in)congruence of attitude and behavior among opposition elites, we conducted a firsthand survey with district councilors.This large-N survey was possible as there was no serious candidate filtering in the District Council (DC) election.The opposition camp benefited from an unprecedentedly high turnout driven by the movement and won a landslide victory.We took advantage of this rare opportunity and asked district councilors about their attitudes on different socio-cultural, economic, and political topics.Using the Kolmogorov-Smirnov (KS) metric, we find that opposition elites share a similar economic attitude.However, contrary to our expectations, they are attitudinally divided on the center-periphery dimension and on how to confront authoritarian incumbents.In short, our analysis shows an incongruence of attitude and behavior among opposition elites: even though widespread repression fostered elites' cooperative behavior, the attitudinal differences in mainlander issues and democratization strategies did not dissipate.
The contribution of our research is threefold.First, this research enriches the literature on autocratization by bringing in the notion of opposition elites' self-reported attitudes.Although the literature on autocratization is still growing, most studies focus on the opposition elites' behavior (Cleary and Öztürk, 2022;Tomini et al., 2022).These studies do provide valuable insights; however, our research is the first that directly taps into the implicit attitude of opposition elites.By doing so, we highlight the (in)congruence of attitude and behavior among opposition elites, which has implications for the tenacity of cooperation and future political developments.Second, our findings on Hong Kong can shed light on other hybrid regimes that similarly experienced widespread repressions, such as Myanmar, Sri Lanka, and Tunisia.Previous literature often suggests widespread repression fosters elite cooperative behavior because of collective victimhood (David et al., 2022;Nugent, 2020).However, behind the scene of unity, we have little knowledge about opposition elites' attitudes during autocratization.Third, our large-N survey analysis provides a firsthand quantification of attitudinal divisions among opposition elites in hybrid regimes.When studying divisions among opposition elites, the current literature does not directly measure elites' attitudinal divisions (or their absence).Part of the problem stems from the fact that scholars can only study a small number of elites using qualitative techniques.But even for quantitative studies that used roll-call votes, party manifestos, or social indicators, they still do not measure elites' attitudinal divisions.So, by quantifying opposition elites' attitudinal divisions using the KS metric, we likewise make a methodological contribution in this field.
Our article proceeds as follows: we first review the literature on opposition elites' division and cooperation in hybrid regimes.Since previous works on opposition elites did not tap into their attitude, we argue that one easily neglects the (in)congruence between their attitude and behavior.In the second section, we elaborate on the case of Hong Kong and present our expectations regarding the attitudinal divisions among opposition elites in the autocratization episode.The third section describes our elite survey and the motivation for using the KS metric.This section is followed by the presentation of our results.We conclude by discussing how these findings advance our understanding of opposition elites' dynamics when hybrid regimes undergo autocratization.

Opposition Elites' Division and Cooperation in Hybrid Regimes
The divisions among opposition elites have always been a key topic in the hybrid regime literature.Very often, the sources of divisions are presumed to be originated from deep societal divisions, such as class, ethnicity, or religion (Cannon, 2004;Gandhi and Ong, 2018;Selçuk and Hekimci, 2020).Also, opposition elites are suggested to be divided over the usage of institutionalized or un-institutionalized strategies to pursue democratization (Buttorff and Dion, 2016;Ortmann, 2009;Smith, 2013).However, when scholars discuss these divisions, they often do not refer to the implicit attitudes of opposition elites.Rather, they infer divisions (or their absence) from different sources.For quantitative studies, these sources include legislative roll-call votes (Smyth et al., 2019), party manifestos (Wong, 2020), or indices that signify ethnic heterogeneity (Gandhi and Reuter, 2013) and religious fractionalization (Gandhi and Przeworski, 2006).On the other hand, in qualitative process-tracing studies, scholars would infer divisions based on elites' campaign messages (Ong, 2022), interviews (Nugent, 2020;Stepan, 2018), or historical accounts that document societal divisions (Cannon, 2004;Selçuk and Hekimci, 2020).To our knowledge, there is still no research that directly taps into the implicit attitude of opposition elites when analyzing their divisions.
But even though scholars used the above techniques to document opposition elites' divisions, opposition elites still cooperate from time to time.According to previous studies, cooperative behavior is especially probable when regimes exercise widespread repressions that target all opposition factions.For instance, in analyzing Venezuela's autocratization episodes, Jiménez (2021) found that opposition elites were willing to coordinate when Chávez persecuted opposition leaders from different factions.Likewise, there are studies on Turkey showing that an opposition coalition was formed in response to Erdoğan's abuse of executive power that targeted different factions (Cleary and Öztürk, 2022;Somer, 2020).Also, in studying the recent autocratization in Myanmar, David et al. (2022) found that Burmese opposition elites cooperated with Rohingya elites to counteract the military coup d'état that involved widespread repression.Overall, these studies demonstrate that widespread repression fosters cooperative behavior among opposition elites.
The reason for opposition elites' cooperative behavior during widespread repression is related to the notion of collective victimhood.As suggested by Nugent (2020), the kind of repression experienced by opposition elites-targeted repression vs widespread repression-can immensely condition the possibility of cooperative behavior.When the repression is targeted against particular factions, such exclusive experience heightens their factional identities.As the targeted factions perceive themselves as unique in their victimization, it is difficult for them to identify with the broader opposition.Contrarily, widespread repression tends to foster cooperative behavior among opposition elites.It is because the shared experiences of repression across multiple opposition factions strongly "primes feelings of groupness with common victims" (Nugent, 2020: 47).Put differently, the common experience of repressive acts can lead opposition factions to be more aware that they belong to a larger group-the collective victims.This priming process matters because it helps opposition factions to recategorize group boundaries.Specifically, opposition elites will identify more with the broader collective identity and less with their specific factions, as the repressive acts bring their common characteristics to the forefront.In turn, opposition elites deemphasize their fissures and prioritize the goal of counteracting authoritarian incumbents.
However, when opposition elites cooperate under widespread repression, will their attitudes on different dimensions overlap as well?This is the question our study aims to pose.The importance of (in)congruence of behavior and attitudes among opposition elites can hardly be overstated since such (in)congruence can shape their willingness to compromise and cooperate further.Suppose opposition factions cooperate but their attitudes are still heavily divided into certain dimensions, it is easily conceivable that their cooperation will break down easily.It is because such attitudinal divisions may lead them to dispute various issues, and the dictators may mobilize fear and hatred between factions, to shatter their cooperation (Stepan, 2018).But on the other hand, if opposition elites' attitudes overlap on different dimensions under widespread repression, we should expect their cooperative behavior is more tenacious.As such, the tactics of divide and conquer will be less effective, and the cohesive opposition camp can pose a headache to authoritarian incumbents.However, because previous research did not tap into opposition elites' implicit attitudes, the current literature does not empirically test the (in)congruence of attitude and behavior among opposition elites.To fill this gap and explore such (in)congruence, we use the case of Hong Kong, in which the hybrid regime changed its strategy from targeted repression to widespread repression.

Opposition Divisions in Hong Kong and Its Autocratization Episode
In this section, we describe previous works that study opposition elites in Hong Kong and then lay out the hypotheses concerning their attitudinal divisions when the regime enacted widespread repression.Talking about the opposition camp in Hong Kong, elites are usually classified into three major factions-mainstream democrats, radical democrats, and localists.Before the anti-ELAB movement, the latter two factions were selectively repressed by the regime (Chan and Ng, 2022;Kwong, 2018;Yuen and Chung, 2018).Previous studies documented that there are three dimensions of divisions that hinder opposition elites' cooperation, even though these studies did not tap into elites' implicit attitudes.These three dimensions are economic, center-periphery, and democratization strategies.In Table 1, we briefly summarize the three factions' positions on these three dimensions and indicate the typical parties and politicians that belong to the three factions.
Regarding the economic dimension, previous research suggested that radical democrats were more progressive than mainstream democrats (Ma, 2007: 106-107;Smyth et al., 2019).These two factions have been divided on different economic issues, such as privatization, housing, and labor issues, with mainstream democrats being pro-market while radical democrats being pro-welfare.On the other hand, the localists' economic stance is more ambiguous.Regarding the center-periphery dimension, this cleavage was mainly driven by the huge inflow of mainlanders and resource competition between mainlanders and natives (Chan et al., 2020;Ma, 2015).The societal divisions contributed to the rise of localists that advocate peripheral nationalism or independence (Fong, 2017).Accordingly, localist elites mobilize anti-mainlander issues, while mainstream democrats tend to be pro-mainlander, as the latter sees cultural connections between mainlanders and Hongkongers (Kwong, 2016).For radical democrats, some are rather critical of the anti-mainlander stance of localists.
Apart from the above two cleavages, opposition elites also revealed dissensus on democratization strategies.In that regard, radical democrats and localists are more confrontational than mainstream democrats.The fissure in democratization strategies was fostered by two main aspects of the political opportunity structure.The first is political stagnation: radical democrats and localists often used different un-institutionalized strategies to speed up the democratization process (Kwong, 2018;Ma, 2011).These strategies include street movements and civil unrest (Ma, 2011;Yuen, 2018), which mainstream democrats hardly affirmed.Another aspect is the regime's targeted repression.No matter in the electoral arena or civil society, radical democrats and localists were selectively repressed by the regime, as they often used uninstitutionalized strategies and politicize their constituents (Chan and Ng, 2022;Kwong, 2018;Yuen and Chung, 2018).On the other hand, moderate democrats were less prone to be targeted, since they more often abided by the regimes' institutional rules.Thus, the targeted repression has created a divided structure of contestation, which exacerbated elites' divisions on democratization strategies.
However, after the anti-ELAB movement began in mid-2019, the opposition elites were unprecedentedly cooperative and deemphasized their fissures (Lee, 2020;Leung et al., 2022).This cooperative behavior was fostered by the change from targeted repression to widespread repression.During this autocratization episode, both the police and pro-government mobs were mobilized to attack participants regardless of factions.Throughout the movement, opposition elites from three factions all fell victim to state violence and mob attacks.Such shared experience of repressive acts brought the broader collective identity-Hong Kong people-to the forefront and their factional identities were much less salient.The motto "no division between peaceful and militant protesters" clearly demonstrates the cooperation among opposition elites in the movement.Cooperative behavior was present not only in civil society but also in the electoral arena.Unlike the previous elections, different factions were much more cooperative in nominating candidates in the 2019 DC election so as to increase the chance of winning the progovernment candidates (Shum, 2021).Moreover, in the spring of 2020, the opposition camp conducted a primary election, whose aim was to coordinate candidate nomination in the 2020 legislature election (Chan, 2021).This primary election was unprecedented because candidates from the three factions agreed to form a united front to challenge the government.

Hypothesis
We propose that the cooperative behavior among opposition elites in the autocratization episode should affect their self-reported attitude on different dimensions.First, concerning the economic dimension, we expect the economic attitude of opposition elites to be similar in the autocratization episode concerned.It is because, during the widespread repression, what dominated the opposition elites' political agenda was political issues related to civil rights, the rule of law, and police violence.None of the demands of the anti-ELAB movement concerns economic issues. 1Also, opposition elites from different factions hardly disputed economic issues during the autocratization episode.When economic issues were on the agenda, they were mostly related to foreign capital divestment and consumer boycott of pro-government businesses, which opposition factions hardly oppose.For these reasons, we expect the economic attitude of opposition elites is not divided in the autocratization episode (H1).
Second, regarding the center-periphery dimension, we expect that opposition elites share similar attitudes as well.As described above, all opposition factions fell victim to police repression and mob violence in the autocratization episode concerned.These shared experiences of repression brought their collective identity-Hong Kong peopleto the forefront, and opposition elites deemphasized their factional identities and fissures regarding mainlander issues.Even mainstream democrats, which were "patriotic" and respected the sovereignty of China (Kwong, 2018), used the subnational identity during the movement and election campaigns (Chang, 2020).As this group boundaries recategorization process is likely to take place, we expect opposition factions are no longer attitudinally divided on the center-periphery dimension (H2).
Finally, we hypothesize that opposition elites share similar attitude on the democratization strategy dimension.The reason behind this is related to widespread repression: widespread repressive acts fostered unprecedented cooperative behavior across different factions (Lee, 2020;Leung et al., 2022).Throughout the autocratization episode, elites from three opposition factions were highly cohesive in using both institutionalized and un-institutionalized strategies to challenge authoritarian incumbents.Crucially, even moderate democrats were consistently present in various street movements to support militant protestors.Such participation experience should increase their tolerance for violent tactics and law-breaking activities.Hence, we expect that the attitude toward how to challenge authoritarian incumbents is similar between the three opposition factions (H3).

Data and Method
To test the above hypotheses, we surveyed district councilors who won the 2019 election.District councilors were considered a mobilization force for elites at the legislature elections, as DC seats provided an important resource advantage and local support network (Wong, 2014).Before the 2019 election, DC seats were mostly occupied by pro-government elites and mainstream democrats.This political landscape fundamentally changed after the 2019 DC election, which took place amid the height of the anti-ELAB movement.As there was no serious candidate filtering, voters leveraged this opportunity to voice their discontent (Shum, 2021).In consequence, not only did the opposition camp obtain more than 80% of the seats, but radical democrats and localists also obtained a significant proportion of seats.This gave us a valuable opportunity to directly ask them about their attitudes on different sociocultural, economic, and political topics.
Regarding the data collection process, we sent invitations to all elected district councilors from May 23, 2020, onward through both official means (i.e.emails and Facebook pages) and personal contacts.In terms of our sample, 238 of the 452 elected district councilors completed our questionnaire, which amounts to a response rate of 53%.If we just consider opposition district councilors as the denominator, the response rate even goes beyond 60%.This response rate is exceptionally high even when we compare it to other political elite surveys in liberal democracies-for instance, the response rate in a Canadian elite survey is lower than 30% (For documentation of elite surveys' response rate, see Walgrave and Joly, 2018).
In our survey, we asked district councilors to self-identify which faction they belonged to, namely mainstream democrats, radical democrats, and localists. 2We used a subjective measure instead of objective indicators (e.g.campaign programs or party labels) to operationalize faction for two reasons.First, a substantial amount of district councilors were independents or affiliated with newly emerged organizations during the movement.Using objective indicators to code factions would lead to serious missingness.Second, for those politicians who belong to a traditional party, their campaign programs may not fully align with their party.This is especially the case in local elections and is reflected in our dataset. 3Overall, 202 of them identified with one of the three opposition factions: 46.53% were mainstream democrats (N = 94), 24.75% were radical democrats (N = 50), and 28.71% were localists (N = 58).As our article focuses on the attitudinal divisions among opposition elites, those who identified with the establishment camp (N = 3) and those who did not identify with these three factions (N = 33) are excluded from the analysis.
The context of our data collection deserves some remarks: our data collection period was from May 23 to June 30, 2020, which was several months after district councilors took office.At that time, the movement was very much contained, as the government imposed restrictive measures on civil liberties due to the pandemic.Also, because NSL was enacted on June 30, 2020, we decided to stop our survey collection on that date.Though district councilors might self-censor in our survey, we consider the sensitivity bias not serious for three reasons.The first piece of evidence is "item missingness."If a specific item has a high missing rate, this can signal that respondents do not want to reveal their attitude toward that item.Assuringly, none of the questions we utilized has a missing rate higher than 5% (see Table A1 in the Supplementary Information).Second, it is possible that district councilors self-censor themselves after the NPC approved the Standing Committee (SC) to enact NSL in Hong Kong on 28 May 2020.To deal with this issue, we split the sample into two-one before this critical event and the other after this event.We test whether the two subsamples differ in terms of the three key outcomes of our study, but we cannot find any significant differences (See Table A2 for the Wilcoxon rank-sum test in the Supplementary Information).Third, after our survey ended and NSL was enacted, the news reported that opposition elites no longer used the motto "Liberate Hong Kong, revolution of our times."However, opposition elites still used the wordings "nation," "liberation," and "revolt" in their slogans (Chang, 2020).For these three reasons, we consider the sensitivity bias was not serious in our field period.

Measurements
To measure district councilors' attitudes on the three dimensions, our survey questions were based on the measurements from the World Value Survey (WVS), Hong Kong Electoral Study (HKES), and Hong Kong Panel Study of Social Dynamics (HKPSSD).We used 14 items that are associated with the three dimensions of our concern: economic, center-periphery, and democratization strategy (See Table A3 for exact wordings of the questions; Tables A4 and A5 for summary statistics in the Supplementary Information).
The five items on the economic dimension are those used in WVS: (i) "Income should be made more equal" vs "We need larger income differences as incentives for individual effort"; (ii) "Government ownership" vs "Private ownership"; (iii) "People can get rich without harming others' interests" vs "People can only get rich at the expense of others"; (iv) "Government should take more responsibility to ensure that everyone is provided for" vs "People should take more responsibility to provide for themselves"; (v) "Competition is harmful.It brings out the worst in people" vs "Competition is good.It stimulates people to work hard and develop new ideas."All five items were measured on an 11-point scale (0-10).
On the center-periphery dimension, we borrowed five items from the HKES: (i) "Immigrants are generally good for Hong Kong economy."(ii) "Hong Kong's culture is generally harmed by immigrants."(iii) "Primary schools should teach the Chinese language in Cantonese rather than in Putonghua/Mandarin."(iv) "New immigrants in Hong Kong are more likely to be the recipients of social welfare benefits than native Hong Kong people."(v) "Hong Kong has too many immigrants coming from Mainland China." Respondents were asked to answer on a 5-point Likert-type scale that ranges from "very disagree" to "very agree." 4  Finally, we measured the democratization strategy dimension with four items, one from HKES, two from HKPSSD, and one original question.These items measure the attitude related to institutionalized strategies and un-institutionalized strategies.The strategies include the use of violent tactics, abiding by the law, the politicization of constituents, and boycotting rigged elections.To measure elites' assent to violent tactics, we asked them whether striving for political rights through violent acts can be a proper means (0 = Never a proper means; 10 = Always a proper means).Concerning compliance with laws, we asked them whether people should obey the law without exception (0 = Obey the law without exception; 1 = Follow conscience on occasions).To measure the politicization of constituents, elites were asked whether they mainly focus on local issues or political issues in DC work after taking office (1 = mainly local issues; 7 = mainly political issues).This item is created based on the fact that some opposition elites in hybrid regimes use political issues to mobilize their constituents, while others shy away from talking about "politics" due to anti-regime concerns (Bedford, 2021;Ortmann, 2009).Finally, for the item concerning election boycotts, we did not directly ask about their attitude toward this issue for ethical reasons.Instead, we used a proxy that asked how they perceived the level of democracy in Hong Kong (0 = Not democratic; 10 = Very democratic) because previous studies suggest there is an association between election boycotts and the perceived democraticness of the regime (Hiskey and Bowler, 2005;Nikolayenko, 2015).Since the construct of the democratization strategy dimension is novel, we also check the construct validity and reliability (See section B1 of the Supplementary Information).
We calculated a composite score for each dimension by normalizing each item to a 0-1 scale, and then taking the average of all items under a particular dimension.This resulted in three composite scores with a range of 0-1 that represents the positioning of opposition elites on these three dimensions.Specifically, along the economic dimension, 0 means progressive and 1 denotes libertarian; along the center-periphery dimension, 0 signifies pro-mainlander and 1 means anti-mainlander; along the democratization strategies dimension, 0 stands for moderate, while 1 represents confrontational.Respondents answering "Don't know" in any of the items are counted as missing in the corresponding dimension.We end up with 187 respondents on the economic dimension, 196 respondents on the center-periphery dimension, and 202 respondents on the democratization strategy dimension.

Measuring Attitudinal Divisions: The Kolmogorov-Smirnov Metric
To identify whether attitudinal division on a particular dimension exists, some scholars use multinomial logistic regression or linear discriminant analysis to test if there is a significant difference in distance between factions.We consider these conventional approaches inadequate because they are essentially comparing the distance between each faction's centroid.To better capture the concept of division, there should be a clear separation between factions regarding the distribution of a composite score.In other words, attitudinal divisions should be measured as a many-to-many relationship between members of the factions, rather than a one-to-one relationship between the averages of the factions.Hence, we propose to measure the distance between the distribution of the composite scores.
In the current literature, dozens of metrics can be used to measure the distance between two distributions (Deza and Deza, 2016).We choose the KS metric, which was previously employed in the elite-mass congruence literature (Andreadis and Stavrakakis, 2017;Golder and Stramski, 2010).We choose the KS metric for several reasons.First, it is a relatively simple and intuitive metric with a range of 0 to 1, where 0 means complete overlap and 1 means complete separation.Second, it can be easily visualized by plotting the empirical cumulative distribution function (CDF).Third, it is a distribution-free and scale-free metric, making it appropriate for use on highly skewed variables.
The KS metric is best known for being the test statistics of the KS test, a nonparametric goodness-of-fit test (Dodge, 2008).The KS metric between two distributions is: , where sup x is the supremum function, and the CDFs of the two empirical distributions.In other words, the KS metric is the maximum difference between their CDFs.There is no objective standard on how large a distance must be to be considered a clear division.Given a range of 0 to 1, we suggest that a KS metric of <0.2 is a small difference or no division, 0.2-0.4 is a moderate division, and a metric of >0.4 is a large division.We use the p-value of the KS test to determine whether the distance between the distributions is statistically significant.

Results
To begin with, we visualize the three composite scores of opposition elites' attitudes at the aggregate level, as shown in the left panel of Figure 1.While the mean of the economic dimension is close to the mid-point at 0.57, the mean of the center-periphery dimension is 0.72, and the democratization strategy dimension is 0.68.In other words, the latter two dimensions are substantially higher than the mid-point.That means the opposition elites' attitudes are, on average, anti-mainlander and lean toward confrontational strategies in the autocratization episode concerned.Next, we calculated whether the three attitudinal dimensions correlate since they may or may not be orthogonal with one another.We find that the economic dimension has no significant correlation with either the center-periphery dimension or the democratization strategy dimension (the correlation coefficients are 0.01 and −0.04, respectively).And in line with previous literature (Kwong, 2018), the center-periphery dimension and the democratization strategy dimension are correlated with a coefficient of 0.27 (p < 0.001) (see right panel of Figure 1).Nevertheless, this correlation is at most moderate, meaning the position of an opposition elite along the center-periphery dimension can hardly determine his or her position along the democratization strategy dimension.Importantly, the right panel of Figure 1 already hints that the three factions cluster themselves along these two dimensions.In that regard, mainstream democrats are the least anti-mainlander and most moderate, while localists are the most anti-mainlander and the most confrontational.The radical democrats are somewhere in between the two factions.The clustering along the center-periphery dimension and the democratization strategy dimension runs counter to H2 and H3, as we expect opposition elites to hold similar attitudes on these two dimensions.
To formally test whether attitudinal divisions exist among opposition elites in the autocratization episode, we turn to Table 2.It reports the KS metrics of the pairwise comparisons of the three factions along the three dimensions.To facilitate interpretation, we also visualize the respective empirical CDFs in Figure 2. First, looking at the economic dimension, we cannot find a significant attitudinal division among opposition factions, which is in line with H1.The distance between mainstream democrats and radical democrats is 0.11, and that between radical democrats and localists is 0.09, and they are both statistically insignificant.The largest distance is the one between mainstream democrats and localists with a KS metric of 0.17, which is still rather small in magnitude and not statistically significant.In substantive terms, that means the economic attitude of opposition elites is similar to the autocratization episode concerned.
As a robustness check of this null finding, we drop the item "government ownership vs private ownership of business and industry."It is because, in communist regimes, government ownership may not signify a progressive economic attitude, while pro-private ownership can be backed by skepticism toward the authoritarian incumbent rather than a libertarian economic stance (Tavits and Letki, 2009).Nevertheless, dropping this item does not affect the substantive interpretation of our results (See Table C1 in the Supplementary Information).Taken together, we find that opposition elites share similar economic attitude in the autocratization episode when political issues (e.g.civil rights and police violence) dominated the political agenda.This null finding is certainly not in line with previous studies that discuss opposition elites' economic cleavage (Ma, 2007:   106-107; Smyth et al., 2019;Wong, 2020).But more importantly, this finding has a broader lesson about the congruence between opposition elites' behavior and attitude on the economic dimension.That is, when opposition elites cooperate to counter authoritarian incumbents and did not dispute economic issues during widespread repression, their economic attitude are similar to one another.Next, moving on to the center-periphery dimension, we find that opposition elites do not share similar attitudes.We see that mainstream democrats are relatively close to the mid-point, whereas the other two factions, especially localists, lean more toward the antimainlander pole.The division between mainstream democrats and localists is apparent with a KS metric of 0.41 (p < 0.001), which is rather large in magnitude.This implies that localists are more skeptical toward mainlanders than mainstream democrats.On the other hand, the distance between mainstream democrats and radical democrats is 0.23, and that between radical democrats and localists is 0.22, respectively, where both metrics fall short of statistical significance.
The similarity in anti-mainlander attitude between radical democrats and localists is in line with our expectations.That is, the shared experience of repression did bring the collective identity (i.e.Hong Kong people) to the forefront, which in turn realigned radical democrats' attitude toward localists on the center-periphery dimension.However, the attitudinal division between mainstream democrats and localists on this dimension rejects H2.We consider there are two reasons to explain this finding.For one, even though the widespread repression did bring the collective identity to the forefront, the identity recategorization process was not strong enough to significantly alter mainstream democrats' attitude on the center-periphery dimension.For another, the items we measure mostly concern the dislike of the out-group (i.e.dislike of mainlanders).Thus, if widespread repression merely changed mainstream democrats' attitude in terms of in-group love (i.e.affinity with Hong Kong people) but not out-group hate, our metric may not be able to fully capture it.Regardless, our findings cannot provide full support for H2, as localists and mainstream democrats are still attitudinally divided on the center-periphery dimension.This suggests there is incongruence between opposition elites' behavior and attitude: even though opposition elites cooperated during widespread repression, their attitudinal division on the center-periphery dimension persists.
We finally turn to the democratization strategy dimension, and our finding rejects H3.Here, mainstream democrats are much more moderate than the other two factions, with a KS metric of 0.45 relative to radical democrats (p < 0.001) and 0.54 relative to localists (p < 0.001).Radical democrats and localists are similarly confrontational, as the KS metric between the two factions is 0.22 and is statistically insignificant.That means radical democrats and localists are still more confrontational than mainstream democrats.This suggests opposition elites are nonetheless attitudinally divided on how to confront the regime, even though they deemphasize their fissures regarding how to counter the authoritarian incumbents.
This finding once again speaks to the idea that there is an incongruence between behavior and attitude among opposition elites.That is, even though widespread repressive acts fostered unprecedented cooperative behavior among opposition elites, they are still attitudinally divided on how to confront the regime.One reason for such attitudinal division can be related to the fact that a significant portion of moderate democrats in our sample did not participate in street movements.And even though they participated, they were not on the frontline and played a supporting role.That is why the widespread repressive acts did not bring about a more confrontational attitude among them.Nevertheless, our finding is noteworthy because elites from the three opposition factions were highly cooperative in using both institutionalized and uninstitutionalized strategies to topple the authoritarian incumbents.But behind the scene of cooperative behavior, our finding suggests opposition factions are still attitudinally divided on the democratization strategy dimension.
Section C1 of the Supplementary Information provides several robustness checks to strengthen our analysis.First, one may argue the results above can be confounded by the demographic differences between the three factions.For instance, moderate democrats tend to be older.Due to age differences, they are more likely to hold pro-mainlander views and are more moderate in terms of democratization strategies than other factions.Second, one may challenge that our findings are driven by the incumbent effect since incumbents are disproportionately moderate democrats.Because of the incumbency effect, moderate democrats are more likely to hold moderate views along the centerperiphery and democratization strategy dimensions.To address these two alternative explanations, we repeat the above analysis by excluding respondents aged above 35 and excluding incumbents.These additional analyses are still in line with our main results (See Tables C2 and C3 and Figures C1 and C2 in the Supplementary Information).Therefore, the incongruence between behavior and attitude on the two dimensions cannot be explained by age and incumbent effects.

Conclusion
Previous research on hybrid regimes used different empirical sources to identify divisions within the opposition camp.Yet, these approaches did not measure the self-reported attitudes of opposition elites.As such, even when opposition elites from different factions cooperate, we do not know whether their attitudinal divisions remain, and, if so, on which dimensions.In this article, we fill this gap by using the case of Hong Kong, whose widespread oppression fostered unprecedented cooperation among opposition elites.In the large-N survey that directly tapped into the attitude of district councilors, we find that, in line with our expectation, opposition elites from different factions hold similar attitudes on the economic dimension.However, contrary to what we expect, they are still attitudinally divided on the center-periphery dimension and how to confront the authoritarian incumbents.In short, there is an incongruence of attitude and behavior among opposition elites: despite their cooperative behavior in the autocratization episode, their attitudes were still divided into mainlander issues and democratization strategies.
Our research contributes to the literature in three ways.First, it joins the growing literature on autocratization that studies opposition elites, whose dominant focus is elites' behavior (Cleary and Öztürk, 2022;Tomini et al., 2022).Our research broadens this literature by distinguishing opposition elites' behavior and attitude, showing the two notions need not align.Such incongruence of attitude and behavior is important because it signifies opposition elites' willingness to compromise and cooperate further in subsequent autocratization episodes.Back to our case, we noted that whether to participate in the 2021 legislature election once again became the bone of contention among opposition factions.Even though the candidate filtering was much stronger than before, some mainstream democrats still participated or showed intention to participate in this election, while for radical democrats and localists, electoral participation was out of the question.These disputes can be explained by the fact that mainstream democrats and the other two factions are still attitudinally divided on how to confront authoritarian incumbents.On the other hand, opposition elites hardly disputed economic issues after the NSL was enacted, which can be related to the fact that different factions hold similar economic attitude.Hence, identifying the (in)congruence of attitude and behavior among opposition elites can help us understand the subsequent political development of hybrid regimes.
Second, our findings on Hong Kong should also shed light on other hybrid regimes that experienced widespread repression.According to previous works on autocratization, widespread repression fosters elite cooperative behavior because different factions all fall victim to the regimes' repressive acts (David et al., 2022;Nugent, 2020).However, behind the scene of cooperation, we have little knowledge about whether opposition elites' attitudes are divided, even though the tenacity of cooperation and future political developments depend upon them.Our research is novel as it directly studies opposition elites' implicit attitudes and finds the (in)congruence of attitude and behavior.Still, our research is a case study, and the three dimensions of divisions in Hong Kong-economic, centerperiphery, and democratization strategies-may not be fully generalizable to other cases.Therefore, our research should open avenues to study other autocratization cases that have different dimensions of elites' fissures.
The third contribution is methodological, as we provide the firsthand quantification of opposition elites' attitudinal division.The current literature on opposition elites in hybrid regimes has long ignored their implicit attitude for two main reasons.For one, because hybrid regimes usually do not offer a level playing field, scholars mostly use qualitative techniques to study a small number of elites.For another, even for studies that used quantitative techniques, their empirical sources (e.g.roll-call votes, party manifestos, or social indicators) do not directly tap into opposition elites' attitude.By leveraging the rare opportunity in Hong Kong, we were able to conduct a large-N survey that studies opposition elites' implicit attitudes.Through the KS metric that measures opposition elites' attitudinal divisions, we demonstrate their (in)congruence of behavior and attitude on various dimensions.
Nevertheless, this research is not without limitations.First, even though the response rate of our elite survey is rather high compared to other elite surveys, not all opposition elites participated in our survey.Even though we ensure anonymity to minimize respondents' political risk, some opposition elites might still fear answering our survey.If their attitudes on the three dimensions differ significantly from those who participated, our findings concerning the (in)congruence between attitude and behavior may not be generalizable to those missing units.Second, our elite survey data is cross-sectional and does not trace the same opposition elites over time.Thus, we are uncertain how divided opposition factions had been before the anti-ELAB movement broke out, and how their divisions had changed after the emergency decree was imposed.In the later episode, when NSL was enacted, autocratization was further deepened: participants and organizers of the opposition primary election were arrested, key civil organizations were disbanded, and freedom of media was further restricted.By then, we were unable to recontact opposition elites because some of them were arrested or in exile.Still, we believe our research can provide novel evidence concerning the incongruence of attitude and behavior among opposition elites.Third, regarding the scope condition of our research, the cooperative behavior of our study occurred amid autocratization.Yet, from the rich literature on democratization, we know that opposition elites also cooperate when hybrid regimes liberalize (Ong, 2022;Wahman, 2013).Hence, future research can investigate the (in)congruence of behavior and attitude among opposition elites during democratization.C3.Distance Between Factions on the Economic Dimension, the Center-Periphery Dimension, and the Democratization Strategy Dimension (Excluding Incumbents).Figure C2.Empirical Cumulative Distribution Functions of the Three Dimensions by Faction (Excluding Incumbents).

Notes
1.The five demands are (1) withdrawal of the extradition bill, (2) retracting the classification of protests as riots, (3) amnesty for arrested protesters, (4) establishing an independent commission to inquire police brutality, and (5) universal suffrage for the Chief Executive and legislature.2. We include radical democrats and localists as two separate categories for two reasons.First, the notion of "radical democrats" is still commonly used in the recent literature of Hong Kong studies.For instance, Law (2019) classified radical democrats and localists separately, arguing they have separate action repertoire and framings of the democracy movement.Second, around one-fourth of the respondents classified themselves as radical democrats.Given the significant proportion of respondents who self-identify as radical democrats, we do not lump them with localists in our main analysis.3.According to conventional wisdom, the councilors of Democratic Party will be objectively coded as mainstream democrats.However, our dataset shows that 16% of them self-identified as radical democrats and 8% self-identified as localists.4. One should note that, in Hong Kong's daily language usage, immigrants often signify immigrants from Mainland China.

Figure 1 .
Figure 1.Descriptive Statistics.Note: Left panel: Violin plot of the three dimensions.Right panel: Scatterplot of the democratization dimension against the center-periphery dimension by faction (Each point represents a respondent, and jittering is applied to improve readability.The dashed line is a fitted linear regression line).

Figure 2 .
Figure 2. Empirical Cumulative Distribution Functions of the Three Dimensions by Factions.

Figure C1 .
Figure C1.Empirical Cumulative Distribution Functions of the Three Dimensions by Faction (Excluding Respondents Aged Above 35).TableC3.Distance Between Factions on the Economic Dimension, the Center-Periphery Dimension, and the Democratization Strategy Dimension (Excluding Incumbents).FigureC2.Empirical Cumulative Distribution Functions of the Three Dimensions by Faction (Excluding Incumbents).

Table 1 .
Three Factions' Positions on the Three Dimensions, and Their Conventional Parties and Politicians.

Table 2 .
Distance Between Factions on the Economic Dimension, the Center-Periphery Dimension, and the Democratization Strategy Dimension.