ECPR

Install the app

Install this application on your home screen for quick and easy access when you’re on the go.

Just tap Share then “Add to Home Screen”

ECPR

Install the app

Install this application on your home screen for quick and easy access when you’re on the go.

Just tap Share then “Add to Home Screen”

Using Latent Language Patterns in Parliamentary Debates to Detect Regulatory Capture of Legislative Committees

Europe (Central and Eastern)
Institutions
Parliaments
Security
Decision Making
Policy-Making
Mateusz Kolaszyński
Jagiellonian University
Mateusz Kolaszyński
Jagiellonian University
Dariusz Stolicki
Jagiellonian University

Abstract

One of the principal public choice theories of regulation is the regulatory capture theory, first formulated by Stigler (1971) and Peltzman (1976). Its main claim is that specialized regulators tend to align themselves with regulated industries and favor their interests over the general welfare. One of the underlying mechanisms of regulatory capture is the identification hypothesis: as regulators repeatedly interact with the regulated industry and seek to better understand it, they not only accustom themselves to the industry's way of thinking, but begin to identify with its goals and interests, losing objectivity and a critical perspective. Hard evidence of regulatory capture is difficult to obtain, as true motivations and mental states of legislators are not available to researchers. However, we propose an innovative method for overcoming that difficulty, motivated by successful applications of similar methods in stylometry. We posit that as regulators adopt the industry's perspective, this is reflected in their language patterns, e.g., vocabulary choices. Those differences are usually not directly perceptible to observers, but can be identified using computational methods, for instance with the aid of transformed-based large language models. We analyze whether language patterns of committee members evolve during their committee service in a manner that is not present for their similarly situated colleagues that do not sit on a specific committee, by first embedding their speeches in a high-dimensional vector space, and then applying statistical methods as kernel regression to identify latent patterns. We test the proposed method using the example of Polish Sejm's Standing Committee on Intelligence Oversight.