Replication Data for: Legislators talk less about the future as they age

This folder contains materials necessary to reproduce the results contained in the above article and additional analyses contained in the supplementary information.

The folder contains all the materials necessary to reproduce the results *starting from the data on parliamentary speech*. This involves:

- fine-tuning a large language model
- making predictions for parliamentary corpora
- aggregating these predictions to form country-specific datasets
- estimating various country-specific models

The first two steps are particularly compute intensive. We therefore additionally include the data-sets used to estimate these different models (in [aus|can|gbr|irl]/outputs/), so that researchers who are *only* interested in the quantities presented in the paper can replicate these models.

Note that we include the code we used to fine tuned the model, but the code used to generate predictions relies on the version of this fine-tuned model archived on HuggingFace.