We start by showing under which conditions LLMs can be useful in the measurement of rev. preferences (e.g. when voting is heavily skewed, or severe select. bias exist). We maintain, however, that nothing beats subject-matter expertise and manual human hand-coding.
We find the following below.
