Big data are those labeled, for strange reasons, with the capitalized "Big". Nevertheless, they are still "data", altough with some specific characteristics: large volume, high frequency and, most notably, unpredictability - data come in the many different forms, they are raw, messy, unstructured, not ready for processing, and so on. Still, these data convey a lot of information to social scientists and good statistical techniques are required in order to extract meaningful results from them. In this workshop we will focus on a specific type of Big data, namely digital texts, both from social media as well as other sources (such as legislative speeches or electoral programs). The aim is to provide an introductory guide to this exciting new area of research, while also offering guidelines on how to effectively use statistical methods on texts for social scientific research by discussing the advantages, but also the limits, of each approach. The attention will be devoted to five main areas: 1) scaling methods that allow to estimate the location of actors in some policy space; 2) supervised classification methods, including machine learning algorithms, that allow to organize texts into a set of pre-defined categories; 3) unsupervised classification that allow to discover new ways of organizing texts into a set of unknown categories; 4) semi-supervised classification methods; 5) network analysis.
An elementary knowledge of R, plus a curiosity towards applied statistics, are good prerequisites for the lab sessions.
Lab sessions are a crucial part of the course: they are offered for "hands-on" experiences to learn the techniques and the statistical methods discussed during classes. All the datasets, replication files of the lab sessions and reference texts will be made available at a dedicated URL before the beginning of the course. Enrolled students should bring their own laptop with R, RStudio and the relevant packages previously installed and functioning (instructions will be circulated beforehand).
Materiale di riferimento
Benoit, Kenneth. 2020. Text as Data: An Overview. In: Luigi Curini and Robert Franzese, Sage Handbook of Research Methods in Political Science and International Relations, London: Sage, 461-497
Grimmer, Justin, and Stewart, Brandon M. 2013. Text as Data: The Promise and Pitfalls of Automatic Content Analysis Methods for Political Texts. Political Analysis, 21(3): 267-297.
Laver, Michael, Kenneth Benoit, John Garry. 2003. Extracting Policy Positions from political texts using words as data. American Political Science Review, 97(02), 311-331
Proksch, Sven-Oliber, and Slapin, Jonathan B. 2008. A Scaling Model for Estimating Time-Series Party Positions from Texts. American Journal of Political Science, 52(3): 705-722.
Curini, Luigi, Hino, Airo, and Atsushi Osaki. 2020. Intensity of government-opposition divide as measured through legislative speeches and what we can learn from it. Analyses of Japanese parliamentary debates, 1953-2013, Government and Opposition, Volume 55, pp. 184-201
Robert, Margaret E., Brandon M. Stewart, Dustin Tingley, Christopher Luca, Jetson Leder-Luis, Shana Kushner Gadarian, Bethany Albertson, David G. Rand. 2014. Structural Topic Models for Open-Ended Survey Response, American Journal of Political Science, 58(4), 1064-1082
Curini, Luigi, and Robert Fahey. 2020. Sentiment Analysis. In: Luigi Curini and Robert Franzese, Sage Handbook of Research Methods in Political Science and International Relations, London: Sage, 534-551
Barberá, Pablo and C. Steinert-Threlkeld Zachary. How to Use Social Media Data for Political Science Research. In: Luigi Curini and Robert Franzese, Sage Handbook of Research Methods in Political Science and International Relations, London: Sage, 404-423
Further readings can be suggested during the course. Please check regularly the home-page of the course on Ariel.