I want to start with one of my favorite “Simpsons” clips ever:
After writing the “Does Size Matter?” post, it occurred to me that the rolling average charts we’ve been posting could be considered as being sorted using an arbitrary X-axis. What I mean is that we calculate the rolling average over time, but of course there is nothing that says that time is the factor that affects the change in average (in a later post, we’ll try to correlate changes in predictions to other events, but we haven’t hit a critical mass to do that yet). I, like Chief Wiggum, just put the predictions in an order and try to draw a conclusion.
On the one hand, time is a crucial aspect for predicting the next week. On Sunday night, the previous game is still fresh and the crowd responds to whatever it saw. Then, as news about injuries and such emerges, the crowd responds again. Then on the eve of the game, the crowd has the most information it can have. To that last point, you could assume that the predictions submitted the day before the game would be the most informed. The counter to that, of course, is that the betting lines would move along with any new information so the more informed predictions may not reflect any additional value.
On the other hand, these types of events – the end of a game, the injury reports, the game previews – may also allow for over-indexing. An exciting end of a game in which one team miraculously pulls victory from the jaws of defeat may cause a lot of predicted wins for the victor and a lot of predicted losses for the vanquished. But as the week goes on, perhaps the circumstances of how the two teams ended up there may bubble to the top, and the crowd may resurface the circumstances that required the miraculous comeback in the first place.
All of this leads to the broader question of when is a crowd big enough for us to feel confident in its predictions. According to some basic statistics and survey articles, a 95% confidence level requires about 400 respondents (we’re not trying to separate demographics at this point). And while we’re a far cry from that number at the moment (hey, tell your friends!), once we do get there, it’ll be interesting to explore the circumstances around groups of predictions. Like political polls after a given event, we would likely see swings in one direction or another, but unlike with politics, the common threads that swing the crowd in one direction or another
We’ll explore this more as the season goes on and try to identify when there are shifts in the crowd and whether they mean anything one way or the other. We’ll also try to correlate them to line movements and see if there is a way to identify just when the maximum value is reached.