Institut for Statskundskab

TY - JOUR

T1 - Collaborating with the Machines

T2 - A hybrid method for classifying policy documents

AU - Loftis, Matthew

AU - Mortensen, Peter Bjerre

PY - 2020

Y1 - 2020

N2 - Burgeoning government activity has enabled an important and growing research program into the content and determinants of policy agendas around the world. However, tightening research budgets and the vast scale of available information force political science to aspire to do more with less. Meeting this challenge requires innovation in managing and preparing data. This paper makes two contributions to the practice of data coding to measure the content of political agendas. First, we propose a method of combining human content coding of political agendas and automated computer classification to classify large data sets. Second, we present software and supporting tools to apply a well-known algorithm for automated text classification, the Naı̈ve Bayes classifier. We demonstrate its usefulness for coding large sets of highly unbalanced multiclass data of the sort used to study the political agenda and demonstrate how our hybrid approach can maximize the returns on research budgets.

AB - Burgeoning government activity has enabled an important and growing research program into the content and determinants of policy agendas around the world. However, tightening research budgets and the vast scale of available information force political science to aspire to do more with less. Meeting this challenge requires innovation in managing and preparing data. This paper makes two contributions to the practice of data coding to measure the content of political agendas. First, we propose a method of combining human content coding of political agendas and automated computer classification to classify large data sets. Second, we present software and supporting tools to apply a well-known algorithm for automated text classification, the Naı̈ve Bayes classifier. We demonstrate its usefulness for coding large sets of highly unbalanced multiclass data of the sort used to study the political agenda and demonstrate how our hybrid approach can maximize the returns on research budgets.

KW - Classifying policy documents

KW - Machine coding

KW - Policy agendas

U2 - 10.1111/psj.12245

DO - 10.1111/psj.12245

M3 - Journal article

VL - 48

SP - 184

EP - 206

JO - Policy Studies Journal

JF - Policy Studies Journal

SN - 0190-292X

IS - 1

ER -