Ian G. Anson, University of Maryland, Baltimore County
Text-as-Data in the Study of Political Attitudes
While most research in the fields of political behavior and public opinion have historically relied upon polling as a primary source of data, a recent explosion of interest in text-as-data has revolutionized the field. While some scholars are studying the effects of text-based media communications on political attitudes and behaviors, others are investigating the ways in which everyday Americans express their politics through writing on a variety of platforms. Both top-down and bottom-up theories of public opinion have been reinvigorated by a focus on the large-scale analysis of writing.
In this presentation, I explain how several of my own projects have yielded new insights in the field of economic voting as a result of text-as-data approaches. In particular, I show how elite cues about the economy can be traced from politicians’ tweets and press releases to the perceptions of rank-and-file partisans. These effects are circumscribed by powerful agenda-setting effects, however, stemming from the availability of certain kinds of economic information in media. These results shed light on the motivated reasoning phenomenon—and its potential limits.
Ian Anson is Assistant Professor of Political Science at UMBC. His research interests include voting behavior, public opinion, political psychology, and the scholarship of teaching and learning. Recently, Ian received the 2019 UMBC College of Arts and Sciences Early-Career Faculty Excellence Award. He received his Ph.D. and M.S. from Indiana University in Bloomington, Indiana.
Duncan Buell, University of South Carolina
Developing the Next Generation of Writing Analytics Researchers
With the oceans of text on the web and in corpora (JSTOR has more than 12 million articles, books, and primary sources), making sense of text is increasingly important and increasingly difficult to do without computational assistance. The task of analysis is complicated by the fact that free text is inherently less structured than numerical scientific data. Analyzing text data requires expertise in knowing what questions to ask of corpora and computational expertise in such things as document representation, indexing, thesaurus construction, part-of-speech tagging and sentence parsing, hidden Markov models, topic modelling, the search for specific linguistic features of interest, and the comparison of different genres of text. Tools used can include prepackaged software, but knowing how to program (in Python or R) is necessary in order to get the full benefit of software like NLTK (the Natural Language ToolKit), CLAWS (from Lancaster University), gensim, WordNet, and the Stanford parsing package. As with most data mining on messy data (and language is inherently messy), obtaining meaningful answers is an iterative process of asking questions, then computing, and then refining both question and computation until the results actually make sense. We will describe, among other things, a recent upper level course designed to introduce students to the questions to be asked and the software and algorithms that can aid in obtaining answers.
Duncan Buell is a Professor in the Department of Computer Science and Engineering at the University of South Carolina. He is also a consulting faculty member in the Linguistics program. Recent projects include the book Data Structures Using Java and work within the Digital Humanities.
Bill Marcellino, RAND Corporation
Building Writing Analytic Systems
Software tools for writing analytics generally lack coherence. Tools tend to be specific to a particular purpose, e.g. lexical-level frequency or complexity description, topic modeling, or stylistic analysis. These tools represent a variety of disciplinary traditions (e.g. corpus linguistics, computer sciences, and digital humanities), a variety of software models (e.g. stand-alone software, R and Python libraries, and java-based programs), and with widely varying levels of programming knowledge required. While writing analytics is growing as a field, this fragmentation is a barrier to continued adoption and expansion.
Just as writing is a design art, building a holistic and robust analytic writing system should reflect a design effort. RAND-Lex is a specific example of a deliberate design effort to build a holistic writing analytic environment, representing a defined construct model of written communication embedded in a web-based system, offering built-in descriptive and inferential statistical analysis and machine learning tools. By addressing the entailment between multiple levels of writing features (lexical, lexicogrammatical, and topical), RAND-Lex also explicitly addresses genre in writing analytics.
In my talk I will cover some of the design choices that went into RAND-Lex, the development effort, cost justifications, launch efforts, and present uses of the platform.
Bill Marcellino is a Senior Behavioral Scientist at the RAND Corporation, a non-profit public policy research institution. Bill is a Professor of Text Analytics within the Frederick S. Pardee RAND Graduate School, and teaches qualitative research methods within John Hopkin’s Global Security Studies program. He also serves on the editorial review board of the Journal of Writing Analytics.
Jessica Nastal, Prairie State College
Writing Analytics and Pedagogy in the Two-Year College
Two-year colleges are a critical point of access for more than 7 million students nationwide. Enrolled students represent a diverse range of communities historically underrepresented in higher education, for example, in terms of race/ethnicity, disability, and socioeconomic status. And, two-year college students have historically been placed into developmental coursework at disproportionately higher rates than their peers at four-year institutions—a fact that is driving placement and developmental education reform.
To re-envision how to better support the range of students enrolled in First-Year Writing courses, I have collaborated with research scientists to explore integrating writing analytics tools with pedagogy. In this presentation, I will provide practical examples from our work and will discuss how two-year colleges might benefit from designing curricula and associated programs of research with attention to writing analytics—and how writing analytics might benefit from collaborating with two-year colleges.
Jessica Nastal is Associate Professor and department chair of English at Prairie State College. She serves on the Illinois Community College Board Placement Standards Workgroup, as Developmental Editor for The Journal of Writing Analytics, and as an editorial board member for Composition Studies.