Visions in Methodology 2015
May 13-16, 2015
Department of Political Science, University of Kentucky
Hosted by Tiffany D. Barnes and Emily Beaulieu
Conference Assistant: Greg Saxton
This years meeting is sponsored by:
National Science Foundation
The University of Kentucky Gaines Center for the Humanities
The University of Kentucky Office of the Vice President for Research
The University of Kentucky Department of Political Science
The University of Kentucky Quantitative Initiative for Policy and Social Research
The University of Kentucky Department of Sociology
Featured Senior Scholars:
Diana Mutz, University of Pennsylvania
Key note address: Friday, May 13th, 2:20pm. William T. Young Library Auditorium
Heather Stoll, University of California, Santa Barbara
Key note address: Friday, May 13th, 3:15pm. William T. Young Library Auditorium
Thursday Professional Development Session:
Friday Professional Development Session:
Presentations accepted for VIM 2015
Political Violence and Issues of Measurement and Methodology
Sponsored by The University of Kentucky Gaines Center for the Humanities
Learning to Protect and Serve in Latin America: Building Relationships between Police and the Communities They Serve
Mary Fran T. Malone (Associate Professor, Political Science, University of New Hampshire)
How can democracies build police forces that fight crime and protect the rights of citizens? Most Latin American countries have grappled with this question, as they have struggled to create new police forces while simultaneously confronting sharp increases in violent crime. To understand how Latin American democracies can develop civilian police forces that do indeed protect and serve their communities, I examine the evolution of police-community relations in Chile, Colombia, Nicaragua, and Panama. In particular, I rely upon public opinion data gathered by the Latinobarometro and the Americasbarometer from 1990 to the present, and trace public attitudes towards police forces over time. Using binomial and ordinal logistic regression, I examine whether public experiences with the police (e.g., frequency of patrols, presence in the community, solicitation of bribes, etc.) determine levels of trust in police over time. To measure trust in police, I create a new measure that taps into how much citizens trust police vis-à-vis other state institutions. I also include a behavioral measure of trust in police, based upon respondents’ willingness to report crimes.
Discussant: Abby Córdova, University of Kentucky
State Cooperation with International Criminal Tribunals: An Investigation of International Warrant Enforcement
Emily Hencken Ritter (Assistant Professor, Political Science, University of California, Merced)
Co-author: Scott Wolford (University of Texas at Austin)
International criminal tribunals (ICTs) cannot apprehend suspects, and states hesitate to put forth costly effort to arrest those indicted for war crimes. Yet many suspects have been arrested or surrendered to ICTs of their own accord. Understanding why some suspects are arrested and others are not can illuminate why states will cooperate with international justice more generally. We present a formal model of a suspect who surrenders or evades arrest and a state that devotes some level of effort to apprehension. We draw on this theory as well as interviews conducted at ICTs in The Hague to present international-, state-, and suspect-level expectations over when and how suspects are likely to surrender or be captured. We use these insights to model the time until capture or surrender in an event history framework, utilizing newly collected data on all individuals indicted by the International Criminal Tribunal for the Former Yugoslavia (ICTY).
Discussant: Anne Meng, University of California Berkeley
Predicting civil wars with higher order interactions
Adeline Lo (PhD Candidate, Political Science, University of California San Diego)
If we wish to predict civil wars, we may require a new approach, placing more emphasis on variable selection for better prediction. The universe of variables to sift through should include any and all information available, but also the higher-order interactions amongst all variables. Higher order interactions become more difficult to capture as the number of explanatory variables grow due to the well-known curse of dimensionality. Given that civil wars are highly complex processes, it is likely that using only marginal information may result in discarding important information embedded in higher orders interactions. Recent advances in big data analysis catered towards uncovering higher order interactions lend themselves to application to political science questions. I suggest that important higher order interactions in existing political science data can be uncovered by the Partition Retention method and illustrate with an application to civil wars data. My findings show that variable sets as large as 4 or 5 variables interact to predict civil wars. Using these identified variables and variable sets to predict boosts correct prediction rates on out of sample testing sets from 86.2% to 97.6%. True positive rates are improved from 5.67% to 90%. The application demonstrates possible gains in correct prediction rates for political science phenomena like civil wars when including a research step for identifying very predictive sets of variables.
Discussant: Sara Mitchell, University of Iowa
Unpacking the Effect of the Changing Sex Ratio on Women’s Political Representation in Post-Conflict States
Melanie M. Hughes (Associate Professor, Sociology, University of Pittsburgh)
Endings of major armed conflict – especially those fought over the political system or central government – are associated with increases in women’s political representation. One explanation for this effect is that during civil wars, men disproportionately die or are imprisoned, leaving openings for women to step into the leadership void. That is, the changing sex ratio could explain why civil wars are associated with increases in women’s legislative presence. Critics of this view suggest instead that changes in the gender ratio are useful empirically only as a proxy for conflict intensity, which in turn is correlated with the magnitude of post-conflict gains in women’s legislative representation. This paper uses advanced quantitative techniques to try to adjudicate between these competing explanations by unpacking the relationships among conflict intensity, the changing sex ratio, and women’s post-conflict legislative success.
Discussant: Mirya Holman, Tulane University
Multiple systems estimation (MSE) at the example of lethal violence in Kosovo 1999
Jule Krüger (Postdoctoral Fellow, Political Science, University of Michigan)
Co-author: Kristian Lum
Conflict and/or violence are often measured by counting the casualties in a specific area and period. An unbiased account of conflict lethality is a prerequisite for testing key hypotheses with empirical data. Unfortunately, the options available to reliably assess the number of victims are limited. Visibility and security issues involved in conflict situations inhibit the collection of random samples or full enumerations required for statistical inference. A promising method to make full use of the ‘convenience samples’ usually available to conflict scholars is multiple systems estimation (MSE). We introduce this technique and apply it to the case of lethal violence in Kosovo (March- June 1999). Kosovo provides a unique opportunity for evaluating this method and its performance because more than ten data collections are available for this conflict. Performing MSE at various ‘snapshots in time,’ we compare the estimates given the available data, our knowledge of the conflict, and the underlying data-generating processes. As the highest criterion, we evaluate all estimates against the ‘ground truth’–a newly available census of war victims provided by the ‘Kosovo Memory Book’ (2014). Based on our temporal comparison of estimates and the true count of victims, we discuss the advantages, disadvantages, and promises of MSE for future empirical research.
Discussant: Jillienne Haglund, University of Kentucky
Sunk Costs and Citizen Support for Military Operations Abroad
Patricia L. Sullivan (Associate Professor, Public Policy, University of North Carolina at Chapel Hill)
We know that support for foreign military operations declines over time and as the human toll of a war climbs, but the rate of this decline varies considerably from one conflict to another and from one point in time to another within the same engagement. In this paper, I explore whether a well-established psychological bias, the “sunk cost trap”, helps explain variation in support for sustaining foreign military operations—among individuals, across conflicts, and over time. I test hypotheses about the conditions under which individuals will favor withdrawing from an ongoing war effort with individual-level public opinion polling data from 60 separate surveys conducted over the course of nine U.S. military interventions between 1960 and 2013. The analysis leverages variation in casualty rates across time, and individual “exposure” to casualty information across space within wars, to disentangle the effects of casualties and war duration, two factors that have been confounded in many studies. The effects of casualties on support for sustaining a war effort, and the conditioning effects of individual characteristics and other attitudes about the war, are estimated using Bayesian hierarchical and multilevel logistic regression models.
Discussant: Carew Boulding, University of Colorado Boulder
Thursday May 14
Patronage and Party Institutionalization in Authoritarian Regimes
Anne Meng (PhD Candidate, Political Science, University of California, Berkeley)
Why do some dictators institutionalize their ruling parties after coming into power, whereas others do not? Ruling parties in autocracies are an important channel through which dictators can solve a multitude of governance problems, such as resolving distributional conflicts or co-optation of elites. However, autocratic parties vary a great deal in their levels of institutionalization, defined as the extent to which party organizations operate independent of particular leaders. I develop a framework for understanding the costs and benefits faced by dictators when institutionalizing a ruling party after they take power. I argue that autocrats can institutionalize their parties by creating formal constraints on their personal authority. Leaders of the CCP in China or the PRI in Mexico, for instance, created term limits and procedures for party officials to move up the hierarchy. Using an original dataset on portfolio allocations and constitutional term limits in African one-party regimes between 1970-2005, I show that leaders who were either very weak or very strong at the start of the regime were the least likely to institutionalize the ruling party. In addition to providing insight on the conditions under which we might expect party institutionalization, this paper also presents novel ways to measure ruling party strength in autocratic settings.
Discussant: Carew Boulding, University of Colorado Boulder
Starving Your Enemies and Your Friends: How Career Incentives Influence Fiscal Transfers in Decentralized Countries
Jane Lawrence Sumner (PhD Candidate, Political Science, Emory University)
In decentralized countries, subnational governments are often heavily reliant upon transfers from the central government. Yet, not all subnational governments receive equal funding, even if they have equal responsibilities. In explaining those patterns, many have found a positive relationship between co-partisanship and funding: elected officials from the central government’s party receive more funding than others do. These findings, though robust, explain a limited number of cases: in many countries, subnational governments are selected through non-elective processes. Analyzing subnational officials’ career incentives lets us situate elected officials on a single dimension with non- elected officials. In this paper, I use both electoral and non-electoral data to develop a measure of that dimension. I then test a hypothesis that the positive co-partisanship finding is actually part of a nonlinear relationship revealed when we extend the spectrum, where the lowest and highest on the scale receive less funding than those in the middle.
Discussant: Julia Gray, University of Pennsylvania
Can Ethnic Party Bans Scale Back Conflict?: Applying Measurement Modeling to Reassess Ethnic Conflict Data
Kelly A. Gleason (PhD Candidate, Political Science, University of Wisconsin-Milwaukee)
Political Violence Feature
Ethnicized politics has long been thought to create social cleavages leading to conflict. Recent empirical inquiries into the relationship between ethnic parties, bans, and conflict have returned mixed results. Much of the discrepancy in findings can be attributed to various strategies for treating qualitative variables from the MAR dataset as quantitative information. My proposed research aims to deal with this measurement problem by utilizing alternating least squares optimal scaling (ALSOS) regression, a scaling exercise more common to physical science, to transform the ordinal dependent variable to its most appropriate linear form. As an added attempt to reduce endogeneity, I employ causal inference techniques to stratify a sample of most similar cases between ethnic groups with parties, without parties, and those who have been banned from party operation. This scaling solution will both improve this particular model of ethnic conflict and demonstrate the value of ALSOS regression in myriad social science applications.
Discussant: Meg Shannon, University of Colorado Boulder
Multi-Level Emotions: How to measure individual, in-group, and out-group anger?
Ngoc Phan (Assistant Professor, Political Science, University of Southern Mississippi)
Emotional explanations are used throughout political psychology, but scholars disagree on how best to elicit or measure emotions. The methodology used to measure emotions is heavily grounded in self-reported and single measures that fail to integrate the multilevel nature of emotions. In this research note, I summarize how anger is currently elicited and measured, argue that current methodologies are incomplete, and propose a novel way to measure group-level emotions. I build upon current emotional measurements and provide a way to measure different targets and anger at different levels.
Discussant: Erin Cassese, West Virginia University
Surviving Phases: Introducing Multi-state Survival Models
Shawna K. Metzger (Lecturer, University Scholars Program, National University of Singapore)
Co-author: Benjamin T. Jones (University of Mississippi)
Many political processes consist of a series of theoretically meaningful transitions across discrete phases. While regime-switching models allow us to empirically assess hypotheses about transitions between phases in some contexts, there have been relatively few attempts to extend such models to the study of durations. Yet, political scientists are often theoretically interested in studying not just transitions between phases, but also the duration that subjects spend within phases. We introduce the multi-state survival model to political scientists, which is capable of modeling precisely this type of situation. We highlight three attractive features of the multi-state model: its regime-switching properties, its ability to model multiple forms of causal complexity that unfold over time, and the inherent flexibility that this class of models provides researchers. We provide several illustrative examples from different subfields to illustrate the model’s features.
Discussant: Diana Z O’Brien, Indiana University
Learning about Spatial and Temporal Proximity to Events using Regression Trees
Ines Levin (Assistant Professor, Political Science, University of Georgia)
Learning about the impact of localized events taking place over time is a multi-faceted problem, as it requires taking into account the influence of multiple dimensions, including: geographical location, timing, and attributes of events. In this paper, I argue that traditional regression approaches — which assume the existence of a linear, or otherwise known, relationship between predictors and outcomes — are inappropriate for learning about the impact of spatial and temporal proximity to events. Instead, I propose using Classification and Regression Trees (CART), an approach that allows addressing the problem in a non-parametric and efficient manner. I illustrate the usefulness of the proposed procedure by studying the impact of mass shootings on opinions about gun control and by replicating recent studies of the impact of immigration rights protests on political attitudes and behavior of Latino immigrants.
Discussant: Jane Lawrence Sumner, Emory University
Friday May 15
A General Purpose Method for Matching on Text
Margaret (Molly) E. Roberts (Assistant Professor, Political Science, University of California, San Diego)
Co-authors: Richard Nielsen and Brandon Stewart
Matching has become a widespread technique for preprocessing observational datasets so to locate control cases for treated observations and reduce model dependence. However, as social scientists begin to have access to more data for each observation, the number of variables to match on has dramatically increased, making traditional matching methods difficult or impossible to implement. This problem is particularly acute when social scientists hope to find treated and control units with similar accompanying text data, for example when the analyst wants to find control and treated units that have similar writings. We propose a new framework and new tools for matching in a high-dimensional context, with a particular focus on text. We discuss the pros and cons of these different approaches and their relationship to traditional matching methods. We design new metrics for balance checking in high-dimensional datasets and apply these techniques to three applications.
Discussant: Caroline Tolbert, University of Iowa
Dynamic Economic Voting: What Makes the Economy Matter?
Ellen M. Key (Assistant Professor, Government and Justice Studies, Appalachian State University)
Theories of economic voting assume governmental evaluations accurately reflect changing economic conditions. Citizens, however, view the economy and government through a partisan filter, making the degree and direction of the response to stimuli conditional on party politics. A strong attachment to the incumbent’s party leads to more positive political evaluations and a more rosy economic outlook. Out-partisans discount positive economic and political news. As partisan cleavages increase, the connection between economic perceptions and political evaluations decreases. This poses a problem for retrospective voting and democratic accountability which require voters to accurately assess economic performance and assign reward or blame accordingly.
Discussant: Amber Boydstun, University of California, Davis
Measurement Error in Discontinuous Online Survey Panels: Panel Conditioning and Data Quality
Lonna Atkeson (Professor, Political Science, University of New Mexico)
Academics are increasingly relying on online panels as sources of public opinion data to answer research questions. Online panelists represent a new type of respondent, a professional respondent, who takes many surveys in response to requests and rewards. . We consider two separate, but equally problematic measurement concerns that may arise with the use of professional respondents: 1) whether or not repeatedly asking respondents their opinions changes them; and 2) that the high frequency of surveying the same individuals for extrinsic rewards may reduce the overall quality of respondent data. We find evidence of panel conditioning in the form of decreased survey duration times, increased political sophistication and non-differentiation of responses. Furthermore, we provide evidence that online survey panel respondents are less likely to optimize their responses than respondents in traditional modes that use probability based recruitment methods. We discuss the implications of our research on the quality of data provided by online panels.
Discussant: Margaret (Molly) Roberts, University of California San Diego
4:15 PM (William T Young Library Auditorium)
The Consequences of Terror Threat for Public Preference over Female Leadership
Mirya R. Holman (Assistant Professor, Political Science, Tulane University)
Co-authors: Jennifer L. Merolla (Claremont Graduate University), Elizabeth J. Zechmeister (Vanderbilt)
Political Violence Feature
One reason why women have a difficult time attaining higher levels of office is that voters generally perceive men as having traits and issue competencies that are most relevant to executive office. While many scholars have documented the presence of gender trait and belief based stereotypes, we are interested in how contexts of violence affect support for female leaders across country contexts. We assess this issue with novel data from experiments conducted across eight countries in Europe, North America, and Latin America. Participants randomly received a control group or a treatment in which they read an article about the threat of international terrorism or positive conditions in the country. Using abstract support questions and evaluations of female politicians in each country, our analysis focuses on the application of gender stereotypes to female leaders in contexts of threat and non-threat across political and county contexts.
Discussant: Yanna Krupnikov, Stony Brook University
Jule Krüger, University of Michigan
Jillienne Haglund, University of Kentucky
Adeline Lo, University of California San Diego
Sara Mitchell, University of Iowa
Melanie M. Hughes, University of Pittsburgh
Mirya Holman, Tulane University
Patricia L. Sullivan, University of North Carolina at Chapel Hill
Carew Boulding, University of Colorado Boulder
Emily Hencken Ritter, University of California, Merced
Anne Meng, University of California Berkeley
Mary Fran T. Malone, University of New Hampshire
Abby Córdova, University of Kentucky
Kelly A. Gleason, University of Wisconsin-Milwaukee
Meg Shannon, University of Colorado Boulder
Yanna Krupnikov, Stony Brook University
Jane Lawrence Sumner, Emory University
Lonna Atkeson, University of New Mexico
Margaret (Molly) Roberts, University of California San Diego
Caroline Tolbert, University of Iowa
Ngoc Phan, University of Southern Mississippi
Erin Cassese, West Virginia University
Shawna K. Metzger, National University of Singapore
Diana Z. O’Brien, Indiana University
Julia Gray, University of Pennsylvania
Ellen M. Key, Appalachian State University
Amber Boydstun, University of California, Davis
Carolina Tchintian, Rice University
Bethany Nanamaker, Emory University
Alicia Uribe, University of Illinois
Shuai Jin, University of Iowa
The workshop at University of Kentucky is part of a broader effort to support women in the field of political methodology. In addition to providing a forum to share scholarly work, VIM connects women in a field where they are under-represented. VIM began as an implementation of recommendations from a National Academy of Sciences report, the APSA Workshop on the Advancement of Women in Academic Political Science, and the 2006 Political Methodology Long Range Strategic Planning Committee Report. VIM provides opportunities for networking, mentoring, and scholarly progress for women in the political methodology community. Visit the VIM website for more details: http://visionsinmethodology.org/.