The Inherent Predictability of Legislative Votes: The Perils of Successful Prediction

Most social science fields find prediction, even in the post-diction sense, to be very difficult. Predictive accuracy is generally low. Thus one would expect great enthusiasm once a field is able to move up to 80-90 percent correct prediction. This level has now been achieved in the legislative roll call analysis area. That would seem to suggest an important theoretical breakthrough has occurred in the area--until it becomes apparent that this same level of predictive success has been achieved by several different and competing theories. How does one go about choosing between theories in such situations? Are statistical criteria irrelevant at this juncture? How is this uniformly high success to be explained?