One more change, which I think most people will like. After seeing the model bounce around a fair bit as a result of individual generic ballot polls, we went back and looked at whether it needs to use a longer-term/slower-moving generic ballot average. The short answer is yes.
So we've made the generic ballot average in our model less sensitive. The generic ballot is really quite "bouncy", almost uniquely so among the polling measures that I've studied, and it's generally best to wait for several weeks before concluding that something has changed.
The model will use a slightly more aggressive setting for the generic ballot as we approach Nov. 6 but even then it's best to be skeptical of sudden shifts. You might say the generic is a great measure of the political *climate* but doesn't always forecast the *weather* reliably.
