The role of political leadership for economic development has recently gained interest in economics. However, empirically testing the influence of leaders remains challenging since there is not an agreed-upon measure of “good political leadership”. Taking advantage of recent developments in machine learning, we apply Latent Dirichlet Allocation (LDA) (also called topic modeling) to quantify the thematic contents of leaders’ speeches; then shows that these thematic contents are proxies for the leaders’ professed priorities. We use the U.S governors’ State of the State Addresses (SoSA) as a test case, since these speeches are by design meant to lay out the governors’ priorities. We show that the topical contents of U.S governors’ SoSA has a strong and statistically significant correlation with subsequent budgetary actions. These findings illustrate the usefulness of topic modeling in analyzing political speech and suggest that speeches may be useful in identifying the role of leadership in economic development. (JEL O10, R50, C38)
More detail can easily be written here using Markdown and $\rm \LaTeX$ math code.