Statistical models beyond linear regression - applied statistics for political scientists

Political scientists are in high demand as analysts due to their ability to both understand societal questions and answer them using empirical observations. This course puts you in that category of job seekers. You acquire a set of transferable skills that are valuable, regardless of whether you contemplate a career as an academic or a policy analyst for a governmental body, NGO or in a consulting firm. The course also provides a toolkit for students looking use quantitative methods for their master’s thesis.

This is an applied methods course. I explain the statistical theory behind models, but the emphasis is on understanding when different models are useful, how to employ them and interpret the results. The course helps students 1) identify appropriate statistical models that describe different data types and 2) interpret these models in a meaningful way.

Many of the phenomena political scientists take interest in are best classified into categories or events. This includes outcome variables like voters’ choice of party, number of social media posts in a time span, or time between violent events. Furthermore, these events are often not independent from each other. Instead, they are nested in groups. Survey respondents come from several countries, media posts are published by specific social media users, and violent events take place as part of specific conflicts.

The course introduces students to a number of models that are specifically designed to describe the underlying phenomenon that generates such data, while possibly leveraging their nested structure. Our focus will be on observational data. The purpose is to help students gear the statistical analysis towards a realistic description of the data while answering questions of political interest.

Topics include

• introduction to R as a statistical software
• binary outcomes (binomial models)
• categorical outcomes (multinomial and ordered regression)
• count outcomes (poisson, negative binomial and hurdle models)
• event history data (survival models)
• hierarchical (nested) data (hierarchical/multilevel models)
• missing data (imputations)