Day 3

Multiple Hypothesis Testing in Conjoint Analysis

Abstract

Conjoint analysis is widely used for estimating the effects of a large number of treatments on multidimensional decision making such as voting for electoral candidates. However, it is this substantive advantage that leads to a statistically undesirable property, multiple hypothesis testing. With a few exceptions, existing applications of conjoint analysis do not correct for the number of hypotheses to be tested. This paper shows that even when none of the treatments has any effect, the standard analysis pipeline produces at least one statistically significant estimate of an average marginal component effect in more than 80% of experimental trials. Then, we conduct a simulation study to compare three methods for multiple testing correction, the Bonferroni correction, the Benjamini-Hochberg procedure, and the adaptive shrinkage method. All the three methods are more accurate in recovering the truth than the conventional analysis without correction. Moreover, the adaptive shrinkage method outperforms in avoiding false negatives, while reducing false positives similarly to the other methods. Finally, we show how conclusions drawn from empirical analysis may differ with and without correction by reanalyzing applications on public attitudes toward immigration, slum brokers, and partner countries of trade agreements.

Disfavor or Favor? Assessing the Valence of White Americans' Racial Attitudes

Abstract

When citizens’ racial attitudes are associated with their judgments related to race— for example, when people with more negative attitudes toward Blacks are less likely to vote for a Black political candidate—existing studies routinely interpret it as evidence of prejudice against minorities. But theoretically, such associations can represent favoring minorities, disfavoring them, or a combination of both. We provide a conceptual framework to distinguish patterns of favoring and disfavoring against a standard of racial indifference, and test it with a pre-registered conjoint experiment. In our results, one widely-used measure—the Racial Resentment Scale (RRS)—captures favoring of Blacks substantially more than disfavoring. This finding calls for greater care in characterizing white Americans’ racial attitudes and illustrates ways to improve future research designs. We also describe several extensions that integrate the distinction between favoring and disfavoring into the broader study of racial attitudes.

Beyond Factions and Guanxi: Mapping the Information Network of Chinese Cadres

Tailored Information in the Process of Personalization: Evidence from South Korean News Reports

From Political Reform to Governance Reform: how do local governments innovate and why?

Recover Unobserved Heterogeneity in Measurement with Application to Patronage Network in China

Diffusion of State Military Cybercapacity: The Theory of Complementarity In Alliances

Abstract

The last two decades have seen a rapid increase in state military cybercapacity among world powers and in the developing world. Yet how states decide to initiate the development of their military cyberapparatuses is a poorly understood phenomenon. Why have some states decided to assign existing military agencies to deal with cyberthreats whereas others jump directly to creating new military cybersecurity units? Viewing the global spread of military cybercapacity as an example of capacity diffusion, I argue that this choice depends on the type of signal a nation wants to send to its allies. The responses to allied behavior follow the logic of complementarity. If the country’s allies signal toughness by creating new units, then the country may not need to do so and assigns cybersecurity responsibilities to an existing military agency. But if the country’s allies take a softer approach by assigning cybersecurity responsibilities to an existing military agency, then the country has an added incentive to create a new cybersecurity agency in order to signal toughness. To test this argument, I construct a new cross-national time-series data set on state cybersecurity organizations for the 1999-2018 period. The analysis provides robust empirical support for my theoretical argument. My findings have important implications for the study of national security policy.

Why and When Do Sovereign States Accept International Agreements for Labor Rights?

The Electoral Authoritarian Advantage in Access to Credit: Historical Mechanisms in Modern Dictatorship

Abstract

Authoritarian rulers in medieval and early modern Europe who struggled to meet their budgetary demands created representative institutions to broaden their support coalitions and stabilize their regimes. I argue that modern dictators introduce elections for similar reasons. Dictators who face financial difficulties are more likely to introduce multiparty elections to placate their domestic oppositions without relinquishing office. In both historical and modern regimes, representative institutions limited executive authority which increased their access to credit. I test the implications of this argument with data from 88 dictatorships between 1970-2014. I find dictators are more likely to transition to electoral authoritarianism after financial instability as measured by debt crises, rising interest rates on new debt, and IMF loans. I also find that dictators who introduce elections benefit in the long term from lower interest rates on new debt. The results suggest that the logic and substantive basis for authoritarian rulers to create representative institutions are general to both early parliamentary regimes and modern electoral authoritarian dictatorships.

Bricks, Molotov Cocktails and Engaged Bystanders: Why Violence Becomes Sustained in Popular Protests

Abstract

Why do violent tactics become sustained in protests? While extant literature has explained why protest violence emerges, research is lacking on how violence becomes a sustained tactic after emergence. This article argues that tactically-moderate protesters, motivated by in-group solidarity, play a collaborative role by providing logistical assistance to violent instigators. We illustrate this mechanism through conducting a list experiment on participants of Hong Kong’s Anti-Extradition Bill Movement. Using a multiple sensitive item design, we distinguish the willingness to facilitate violence from the willingness to exercise it, and estimate the proportions of each group of actors. Statistical analysis further reveals that identification with the militants is associated with stronger willingness to facilitate and exercise violence, although the latter effect is endogenized by their exposure to police violence. Findings enrich the study of protest violence by advancing the measurement of sensitive protest behavior and highlighting a relational mechanism that sustains violent protests.

Women Use More Positive Language than Men in Political Campaigns

Abstract

What explains the type of electoral campaign run by politicians? Prior work shows that parties strategically manipulate the level of emotive language used in their campaigns based on their incumbency status, their policy position, and objective economic conditions. While the literature has demonstrated the important role that party characteristics and real-world conditions have in influencing campaign sentiment use, it has largely ignored the potential influence of non-political, individual level attributes, such as gender. In line with a large literature on politics and gender, we argue that women candidates face strategic incentives to be more positive in their assessment of the world around them. We test this claim with a novel data set that captures the emotive language used in over 164, 783 tweets posted by 2, 662 candidates for elected office in the United Kingdom during the 2017 and 2019 elections. Consistent with prior literature, we find that candidates running from the ruling party use more positive sentiment than those running from the opposition party. Importantly, as we predicted, we also find that women candidates are more positive than their men counterparts – and that this positivity is even higher for incumbent women candidates than nonincumbents. Our analysis has important implications for research on gender and campaign strategies.

Measuring Democracy Discretely

Abstract

There is considerable disagreement regarding the appropriate level of measurement for measuring democracy, with some researchers considering it discrete, while others treat it as a potentially continuous measure. While much is written about the measurement properties of continuous scales, discrete measures are most frequently operationalised in a deterministic manner. This debate is also characterized by disputes over dimensionality, where continuous and ordered measures reflect a common dimension defining democracy, while many categorical approaches concerning the definition of "hybrid" regimes eschew such ordering. To solve this problem, I derive a Bayesian unordered latent class model that simultaneously determines the number of latent categories and the classification of units to the categories through reversible jump Markov Chain Monte Carlo methods. Applying to Polity IV data, I recover a 10 category typology producing an interesting set of hybrid regimes not constrained to a single dimension. The estimates illustrate the inherent uncertainty in these classifications, and demonstrate why discretisation of continuous scales are often contentious.

Show Opponents the Money: Targeted Co-optation in Electoral Authoritarianism

Abstract

Although most research on electoral authoritarianism(EA) demonstrates that autocrats favor supporters and punish opponents after elections, we argue that opponents may still be co-opted when autocrats aim at buying their compliance. We evaluate our propositions by studying allocations of industrial complexes in South Korea during 1971-1988. We find that besides counties overwhelmingly supporting the authoritarian incumbent party in elections, counties that cast fewer votes for the incumbent are still more likely to get industrial complexes. Further analysis using the percentage of Protestants as an instrument variable provides causal evidence on our findings. Our findings highlight various survival strategies implemented by EA, different political incentives between autocracies and democracies in distributive policies, and that improvements in governments’ information do not necessarily make citizens worse off in dictatorships.

Farewell President! Political Favoritism, Economic Inequality, and Political Polarization

Abstract

This paper examines the effect of political favoritism on economic inequality in the short run and political polarization in the long run. We exploit the sudden death of an authoritarian leader – President Chiang Ching-Kuo of Taiwan – in 1988 to generate plausibly exogenous variation in partiality. We find that Chiang’s nationalist regime conducted political favoritism broadly toward political immigrants via cronyism (allocating public sector positions) and also differentially toward specific subgroups of political immigrants via wage discrimination (offering higher payroll to these subgroups within the public sector). Favoritism led to a 6.7 percent immigrant payroll premium, which accounted for more than two-thirds of the immigrant-native payroll gap at the time. This in turn propelled overall income inequality by 4.3 percent. Moreover, political favoritism breeds political polarization in the long run by pulling apart the political views of immigrants and natives. Compared with natives, immigrants who were exposed to favoritism tend to adopt political positions that are aligned with the nationalist party today: they are more likely to support unification with China, and are more inclined to trust the mainland Chinese government and its citizens. Exposed immigrant (native) swing voters are also more (less) likely to vote for the nationalist party today.

Identification of Joint Probabilities of Potential Outcomes using Instrumental Variable Method with a Proxy Variable

Abstract

This paper deals with the evaluation problem for joint probabilities of potential outcomes from data obtained in experimental studies with non-compliance (Kawakami, 2021). To evaluate joint probabilities of potential outcomes, the assumption of monotonicity (Tian and Pearl, 2000; Pearl, 2009), the independence between potential outcomes (Robins and Richardson, 2011), gain equality (Li and Pearl, 2019) and the specific functional relationships between cause and effect (Pearl, 2009) have been utilized. In contrast, this paper provides novel identification conditions for joint probabilities of potential outcomes using an instrumental variable and a proxy variable of potential outcomes. Different from existing identification conditions, the proposed identification conditions do not require specific constraints on the joint probabilities of potential outcomes. The results of this paper can be leveraged for conducting detailed analysis of causal mechanisms that generate statistical dependencies, that is, cases where either one of the variables causes the other, or a third variable causes both.

Bayesian Analysis of List Experiments: Using Informative Priors to Improve the Study of Sensitive Questions

Abstract

List experiments have become a popular tool for political scientists studying sensitive questions. Recent innovations in statistical methodology have made list experiments more powerful by allowing researchers to include estimates from list experiments in multivariate regression models. But a frequent problem in applied settings is that these models are computationally unstable and that estimates are often less precise than desired in small to medium samples. To address these problems, we propose Bayesian formulations of three canonical models for the analysis of list experiments. One great benefit of the Bayesian approach is the ability to include additional information via prior distributions. Specifying informed priors amounts to a principled combination of information which can improve the stability and efficiency of model estimates. It also provides a straightforward framework for the analysis of more advanced list experimental designs such as the double list experiment or designs that include additional information in the form of direct questions or auxiliary information on the sensitive item. Thus, Bayesian analysis using informative priors generalizes and unifies a whole range of different modeling approaches for list experiments. We demonstrate the utility of our Bayesian approach to list experiments using simulations and applied political science examples.

Post-materialism across Generations in China: Economic Prosperity and Economic Inequality

Abstract

With China’s economy growing remarkably over the past four decades, an emerging literature has concluded that China has moved toward post-materialism as Inglehart proposes. Inglehart argues that against the background of prolonged socioeconomic development, younger generations would take survival for granted and then shift their value orientation from materialism emphasizing survival to post-materialism emphasizing individual autonomy, self-expression, and quality of life. Inglehart contends that birth cohort effects or socialization serve as the primary source of value shift, while other scholars contest this argument by emphasizing the importance of period effects, or short-term fluctuations. This paper examines how values have changed across generations in the context of China’s significant economic and social transformation. This paper not only directly tests birth cohorts effects against period effects but also introduces economic inequality to explaining the Chinese story. As Chinese significant economic achievements are accompanied by widening economic gaps, it is necessary to explore whether the shift toward post-materialism has been weakened by rising inequality. Descriptive analyses of five waves of the World Values Survey data confirm the occurrence of the shift toward post-materialism across generations in China. However, this shift has been weakened in recent years. In addition, the results of multilevel modeling suggest that period effects, rather than birth cohort effects, are more influential in shaping value formation. Contrary to conventional wisdoms, economic inequality, rather than economic prosperity, contributes to this rise of post-materialism. This finding highlights the need for more thorough empirical investigation and theoretical building of the influence of widening economic gaps. This paper not only provides a more comprehensive examination of the Chinese case, but also offers a new theoretical perspective of understanding value changes.

A Matter of Trust: How Americans and Australians Diverged in Trust and Media Use during a Global Pandemic

Abstract

Recent surveys show media trust is low in Australia and the United States. Less clear is how much media trust matters for citizens’ uses of different types of media and under what conditions (Strömbäck et al., 2020). This article focuses on trust in journalists and other information sources – medical experts and friends and family on social media – during a time of crisis: the COVID-19 pandemic. Using survey data, we measure citizens’ trust in COVID-19 information sources and the implications of media use and partisanship during the health crisis. We find stark differences in media use and trust in journalists between the US and Australia. The US has partisan patterns of media consumption serving different public spheres. Most Republican voters do not trust professional journalists and were less likely to trust medical experts. While, Democrats are more trusting of professional journalists, health experts and turn to mainstream media for COVID-19 information. Republicans, who tended to rely on conservative media for news, were the least concerned about catching coronavirus and most likely to believe restrictions on movement and gatherings had gone too far. In contrast, most Australians, regardless of partisanship, express higher levels of trust in professional journalists and medical experts for COVID-19 information. Differences in media use, trust and partisanship between the two democracies reflect patently different attitudes to government lockdowns. We conclude low trust in professional journalists and medical experts matters; without it, government agency to manage the health crisis is undermined.