Central bankers' preferences were measured using what they said in meetings rather than how they voted. Automated text analysis and scaling methods were used to create a new measure of central bank preferences leading up to the financial crisis.