Musashi Hinck (né Jacobs-Harukawa)

Coding, Debugging, Designing and Optimizing for Social Science Research

Detecting Electoral Constraints on Twitter with Unsupervised Text-as-Data

Jul 2019 » Research Text-as-Data Twitter Machine Learning Electoral Systems
Detecting Electoral Constraints on Twitter with Unsupervised Text-as-Data

For those interested, here’s a link to my master’s thesis where I compare the tweets of New Zealand MPs by electoral system. (New Zealand has a mixed electoral system where some MPs are voted in on a single national list, and the rest by SMDP).

Abstract

A wealth of political science literature notes the connection between the constraints created by electoral systems, the incentive for re-election, and legislator priorities. This study measures the effect of these constraints on legislators’ online communication strategies. Using the MMP representational electoral system in New Zealand, I compare a variety of text-as-data models to find systematic differences between List and SMD MPs in a dataset of 15,381 tweets by New Zealand legislators. A number of patterns emerge: List MPs are more likely to engage in rhetoric critical of spending, whereas SMD MPs are more likely to advocate reform. List MPs are also more likely to use Twitter to celebrate or congratulate, whereas SMD MPs are more likely to respond directly to other users.