Actually, the goal of a line maker is not to even out the bets on both sides. That is a common misconception.
Agreed...
But to the original question- 16-5 is not that unusual really in terms of spreads. In strictly mathematical terms, the probability of a coin flip hitting 16 or more times out of 21 is 1.33%. However, if Vegas significantly underestimates a team going into the season, it's going to take them quite a while to adjust. This is also partly how KenPom does things too. As a sample function to look at simply, consider that if they outperform in their first game, in game 2 their value of NU might be 90% preseason projection and 10% game 1. The next game might be 80% preseason projection and 20% YTD performance. At that rate it will take 10 games to fully incorporate the YTD performance. Add in that most of those early games are against crappy opponents so have somewhat limited value, and you then almost have to reset that metric once B1G play starts cause you have relatively few usable data points before then.
To put it simply, the outcomes relative to the spread are not independently distributed, and as such direction of the errors over the course of a season will tend to be correlated. So the distribution of spread outcomes over the course will have higher kurtosis (larger tails) than might be implied by a normal distribution (which pure coin flips would approximate to).
For reference, the best team against the spread in the NFL this year was the Pats, and they were 13-3. That is not unusual.
That's probably more of a technical answer than you might have bargained for, but hopefully it's helpful.
