Three Papers on Collective Action

Abstract

This dissertation develops a method to construct a collective action event dataset and then uses the dataset to understand the landscape and causes of collective action in China.

Chapter 1 develops Collective Action from Social Media (CASM), a system that identifies offline collective action events from social media posts using convolutional neural networks on image data and recurrent neural networks with long short-term memory on text data. I implement CASM on Chinese social media data and identify 136,300 collective action events from 2010 to 2017 (CASM-China). I evaluate the performance of CASM-China through cross-validation, out-of-sample validation, and comparisons with other publicly available datasets. CASM-China identifies 10 to 100 times more events than other datasets. This chapter is co-authored with Jennifer Pan.

Chapter 2 uses CASM-China data to analyze the spatial distribution, issue salience, forms, targets, and police presence of collective action events in China.

This chapter also describes how these characteristics change over time. The major finding is that the leadership transition marked a critical change in collective action events in China. After the middle of 2013, the number of collective action events decreased.

Collective action events moved from southeast coastal provinces to northwest inland provinces.

Traditional protest issues such as land acquisition and unpaid wages decreased, while homeowners' collective action increased.

Violent and disruptive action decreased, while peaceful petitions increased.

The government was increasingly unlikely to be the target of protests but was more likely mediate between protestors and non-state institutions. Overall, the transition of leadership fundamentally changed the political opportunity structure of collective action in China.

Chapter 3 compares grievances that become collective action events, and those that do not, using a subset of CASM-China data. I found that the probabilities that grievances will become offline collective action events are decreasing over time. Grievances with different issues share the decreasing pattern in their probabilities of becoming collective actions, but they also exhibit considerable differences on an absolute scale. The decreasing pattern still holds after controlling for users' social media usage and their online political engagement levels.