We use a panel-data framework to study the effects of print censorship on early-modern England’s cultural production. Doing so requires distilling dispersed qualitative information into quantitative data. Integrating the historical record implicit in a large language model (LLM) with facts from secondary sources, we generate an annual index of print censorship. Applying a machine-learning (ML) algorithm to a major corpus, we construct document-level measures of the innovativeness (quality) and volume (quantity) of cultural production. We use pre-existing topic-model estimates to apportion each document among distinct cultural themes-three affected by censorship and five unaffected. We thereby assemble a yearly theme-level panel for 1525-1700. We use local projections to estimate censorship’s dynamic effects. Paradoxically, censorship raises the level of innovativeness in censorship-affected themes relative to non-affected themes. Censorship has a temporary chilling effect on the quantity of cultural production, with output recovering within a decade. Our findings are robust to the use of an instrumental-variable approach addressing the endogeneity of censorship. Our findings are unchanged when using three alternative LLMs to produce the censorship index. Using LLMs and ML to measure hard-to-quantify phenomena like censorship and cultural production, we provide new insights into the drivers of cultural evolution.
Here is the full paper by Peter Murrell and Peter Grajzl. Via the excellent Kevin Lewis.
The post Data on the effects of censorship in early modern England appeared first on Marginal REVOLUTION.
