Some political processes, both natural (e.g., election campaigns) and artificial (e.g., Deliberative Polls), are broadly informative, exposing people to sizable quantities of information. The natural measure of the individual-level learning (G) is the observed knowledge gain (g): the increase in the proportion of knowledge questions answered correctly. Yet taking g as a left- or righthand-side variable in analyses seeking to explain it or use it to explain other variables often produces disappointing results. Observed post-event knowledge (k) often does better.
Here we confirm this pattern, examine its roots, and suggest one simple escape. We first show mathematically that the correlation between G and g can be negative and, even when positive, less than that between G and k. Numerical simulations then show that these are not mere possibilities. Indeed, g is a generally weak and occasionally perverse indicator of G; k, a much stronger one, should be used instead.
