davidpomerenke / protest-impact

Can we use causal inference and natural language processing to quantify the impact of protest movements on public discourse?

How do protests shape discourse?

Causal methods for determining the impact of protest events on newspaper coverage

Master thesis project by David Pomerenke

Abstract

Protests can have an impact on newspaper coverage, not only by prompting reports about the protest events themselves, but also by bringing attention to the issue that they are concerned with. But they may also have negative impacts by distracting from existing constructive discourse on the issue. Quantitative media analyses can uncover these impacts. The problem is that protests and media coverage are in a complex causal relation: They mutually influence each other, and external events may cause both protests and coverage to increase at the same time. To deal with observed and unobserved confounding, I evaluate multiple causal methods: Besides the classical repertoire of regression and instrumental variables, I investigate aggregated synthetic controls and inverse propensity weighting. I show that all methods reduce bias but do not completely remove it, except perhaps the synthetic control method. My analysis of climate protest events in Germany shows that the protests generally tend to increase not only protest-related coverage but also other coverage related to climate change; but there are differences between the various protest groups. This sheds empirical light on a theoretical debate about possible backfiring effects.

➡️ Download the report

➡️ Download the German Protest Registrations dataset

Repository structure

├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
├── models             <- Trained and serialized models, model predictions, or model summaries
├── notebooks          <- Jupyter notebooks.
├── pyproject.toml     <- Requirements for installation with [Poetry](https://python-poetry.org/).
├── report             <- [Quarto](https://quarto.org/) project with the source for the report.
└── src                <- Source code for use in this project.
    ├── data
    ├── features
    ├── models
    └── visualization

License

(c) David Pomerenke 2023. (For now.)