In my past job I built econometric models for a bank. One of my bosses was not a true statistician (and neither am I) but had a great theory about statistical models, which went something like the following:
No matter what direction or magnitude the models tell us that a variable should have, we can and always will come up with an reasonable explanation as to why that is so.
The theory was based on his experience of seeing multiple-variable models where in some cases, two different models would show the same variable having the opposite impact. Sometimes this was justified due to correlation across variables, but often, it was just a product of statistical variance and small sample sizes. For example, with stocks, we have the following table:
The above table can be created for just about any set of asset prices and indicators you can think of. What makes financial speculation so great is that at some point in time, speculators who gambled based on either column would have been right. The same theory applies in NFL handicapping as well. Each week bettors will place their bets on one column or the other, and around 50% will be right.
Contrary to what a lot of people think there is plenty of disagreement among professional bettors. Granted, much of the time, lines are just off and everyone is hammering the same side save for buyback/covering on huge line moves. But for every couple of obvious games there is a game where there is legitimate two-way action from winning players on both sides. Since everyone who wins is using some kind of statistical model to handicap games, most "normal" games don't end up with two-way action. Instead, in my experience the level of disagreement is highest on games where certain aspects of the contest make handicapping more subjective. Examples include:
- Major injuries such as a QB going down
- One team playing far above or below past expectations in recent contests
- One or both teams using unusual strategies
- Extreme weather conditions
Compared to a statistical model, where most will come up with the same answer, in these types of lower sample-size situations there is much more potential for disagreement. In these games handicapping becomes more of an subjective art form, particularly in football where there isn't the huge sample size of games that we have in something like baseball or basketball. Just as in stocks, we can create the following table for the NFL:
There are really two points to be made here. The first is that it is pretty much a complete waste of time to try and explain the outcome of any one game, or even any series of game outcomes. There is always an explanation, and since around half of bettors will be right in any one game, there will always be a chance to hear it from someone.
The second point is that, in my opinion, the tendency even among sharper bettors is to read too much into these lower-sample size situations and deviate too far from what statistical models will suggest. I think there is a tendency for bettors to want to "use the information" when they hear of things like team disunity, injuries (not saying injuries are not important), and the like, because it makes them "feel smarter" since they have a well-defined reason for their bet. This is why you hear professional touts always spout off about things like motivation, trends, and recent injuries. It is much easier to sell someone when you have a reason.
In my experience, assuming you have come to an handicapping approach that is tested and wins long-term, the absolute best bets are those where there is no obvious angle other than that your model/approach likes the play (although there may be some underlying statistical reason that is hard to explain). Usually these are the sorts of games where if you weren't using math, you would always pick the other team. While there are always certain situations where it is just completely obvious that the model has to be wrong, most of the time the impact is not as large as you would think.