The Promise and Perils of Quantitative Research in International Relations1 Petr Ocelík MVPd002 Quantitative Research in International and European Politics 1Yes, this paraphrases the title of the compulsory reading. Plan for today • 1. Introduction round: – What is your background? – What is your dissertation research about? – What is your experience with quantitative research? • 2. Seminar: – What is (and should be) the role of quantitative research in political science? – What challenges need to be addressed when using quantitative methods in political science research? – What is the relationship between statistical inference and causality? Politics and statistics • The history of politics and statistics is closely intertwined. What is (and should be) the role of quantitative research in political science? Aggregation • In 2020, IBM estimated that 2,5 trillion (1018) bytes of digital data are generated globally every day. • The ongoing information revolution is strengthening the demand for methods of processing and analyzing aggregated data. • The ability to analyze such data is not only essential academic, but also a civic, competence. Pattern detection: Simpson’s paradox Navarro & Foxcroft 2018 Navarro & Foxcroft 2018 Transparency, standardization, uncertainty quantification Inference • How do you understand the concept of inference? Inference • Gary King: inference is impossible – we try to use the facts we have to learn about the facts we don’t have. King 2020 Inference • Gary King: inference is impossible – we try to use the facts we have to learn about the facts we don’t have. • Choose: substantive questions of interest • Formalize: quantity of interest (QOI), given question • Collect: data, given QOI, question • Assume: class of models, given data, QOI, question • Estimate from data: best model in class, given the above • Present results: given the above King 2020 Statistical inference population sample data random selection samplestatistics population parameters estimation statistical significance evaluation substantive significance evaluation inferringconclusions measurement What challenges need to be addressed when using quantitative methods in political science research? What kind of challenges we face? • Braumoeller and Sartori (2011) underly (1) specification errors and (2) inference errors. • Specification errors – Statistical models do not connect with the relevant theory – Theory is (A) neglected and/or it is (B) insufficiently developed – Statistical models are imposed to data (phenomena) • Inference errors – Models misrepresent the data (phenomena) – Best models are parsimonious: “Everything should be made as simple as possible, but not simpler.” Reducing specification error: bringing in theory Statistics and theory TEORIE STATISTIKA The golem of statistics The role of theory • What factors decrease the probability of inter-state military conflict? The role of theory • Are democracies less likely to engage in inter-state military conflict? The role of theory • Are democracies less likely to engage in inter-state military conflict? Bruce Russett The role of theory • The theory describes a causal mechanism through which the effect of a given cause (predictor) is transferred to the effect (dependent variable). The role of theory: causal mechanisms • The exercise of power in democracies (predictor) requires the support of a large portion of the public; because military engagement imposes significant costs with public repercussions, political elites in democracies exhibit higher conflict aversion (dependent variable) (Mesquita et al. 1999). The role of theory: causal mechanisms • The theory describes a causal mechanism through which the effect of a given cause (predictor) is translated into the change in the dependent variable. • The exercise of power in democracies (predictor) requires the support of a large portion of the public; because military engagement imposes significant costs with public repercussions, political elites in democracies exhibit higher conflict aversion (dependent variable) (Mesquita et al. 1999). The role of theory Models and hypotheses • A theoretical model is a set of assumed causal mechanisms (e.g., institutional brake) defined to represent a class of phenomena (e.g., MID). • A statistical model is a set of probabilistic relationships between variables (level of democracy, intensity of MIDs, etc.) defined to represent a particular observed phenomenon (selection of MIDs). Models and hypotheses • A statistical model is a set of probabilistic relationships between variables (level of democracy, intensity of MIDs, etc.) defined to represent a particular observed phenomenon (selection of MIDs). • Variables: characteristics of a given phenomenon taking on different values. • Probability: quantifies the degree of uncertainty due to random events (noise) and imprecision (measurement error). • Parameters: quantify the strength and/or direction of relationships between variables Models and hypotheses • A statistical model is a set of probabilistic relationships between variables (level of democracy, intensity of MIDs, etc.) defined to represent a particular observed phenomenon (selection of MIDs). Models and hypotheses • A statistical model is a set of probabilistic relationships between variables (level of democracy, intensity of MIDs, etc.) defined to represent a particular observed phenomenon (selection of MIDs). MID = a – b1*DEM – b2*IO – b3*EIN b1 b3 b2 r r r Models and hypotheses • It is useful to distinguish between theoretical and statistical hypotheses. • Theoretical hypotheses: assume a certain relationship between two or more variables. • E.g.: H1. The democratic nature of the dyad reduces the intensity of MIDs. • Statistical hypotheses: allow generalization (inference) of results from a representative sample to the target population from which the sample was drawn. • H1: b1 ≠ 0 MID = a – b1*DEM – b2*IO – b3*EIN Soukup a spol. 2022 Reducing inference error: choosing adequate model Queenborough 2010 THIS ALSO FUNCTIONS AS A MEME Queenborough 2010 Observational interdependent data • We often work with observational data that is interdependent. • Especially in International Relations, we collect data on set of entities (states) that are interdependent. • Why this could be a problem? Two approaches to inferences • Statistical generalization (inference) from sample to population (design-based) vs. inference from data to model (model-based). Two approaches to inferences • Statistical generalization (inference) from sample to population (design-based) vs. inference from data to model (model-based). MEBn4034 Social Network Analysis in R What is the relationship between statistical inference and causality? Statistics and causality Kellstedt & Whitten 2018 P.S.: What needs to be avoided • Hyper-exactness: the average wage is 37384,425 Kč • Pseudo-skepticism • p-value fetishization • Mechanistic application Soukup a spol. 2022