Most of your 2 Unit plays have larger differentials than your 3 Unit plays. What's the catch?
At this point of the season I feel like I need to take into account who teams have played to generate their statistics. There's no easy way to adjust for strength of schedule until after this week, so it's less quantifiable. For example, Fresno State has played three good/very good teams (USC, Utah, Nebraska) whereas San Diego State has played one (Oregon State - I can see an argument for NC, but they aren't on the level of any of the three Fresno has played), so the assumption is the data Fresno State has generated is going to be skewed by their quality of competition. Most of these teams also haven't played enough home/road games to produce viable statistics, so I'm relying on their overall numbers more than I usually would. After this week, the strength of schedule can be quantifiably controlled for, and home/road stat validity will improve which should make the outputs better.
Also, once spreads get past a certain point (18-20+ points), unless betting on the dog, I use more of a ratio than straight differentials - basically, a 20 point differential has more value when the line is 5 (4x) than when it's 20 (1x).