Assuming these models are based on historical game data:
1. Games like the pisarcik fumble at meadowlands are probably in and skew stats
2. Hail Mary passes skew stats
Bottom line is this you are cowboys so you opt to be up either 3 4. or. 5. With 2:30 to play against a team with a good field goal kicker in good indoor conditions. That has to be statistically distinct from outdoor games etc. it would be a mistake to believe the stats are universal
Think of it this way: stats mght tell you that the person who is leading in a 10k race at the second to final lap wins 45% of the time and the person in 2nd wins 47% of the time. These stats are based on historical races with their rules strategy and possible slow races with kicks being important. If you are in a fast race maybe the lead 2 guys both die off and 3 spot is better. Everything is relative
Put another way. Does it pass the gut test? Rojo would you rather the cowboys a be down 3 or 4 with 2:30 to play against Green Bay this weekend? In the snow and against the wind? Or in the calm dry sun? If the other team had 3 timeouts left or none. Etc.
Kick the extra point let efren Herrera earn his paycheck
By the way I took statistics, probability, and quantitative analysis in college and I do not think I have ever understood anything as well as the ping pong balls example of Malmo. That is so simple it is brilliant in explanation without the proof theorems.