Agreed. I am not saying I have a quick fix (such as only look at plays occurring in last 4:00 of close games), just pointing out a potential flaw in their simulation design. Other designs may turn out to be equally flawed, but that won't make their conclusion correct.

One relatively crude way to resolve the question would be to reduce the situation to two possible scenarios. You're up by four points with 4:00 left on your own 25, and you can either run three plays and punt (scenario A) or stay aggressive and try to run the clock out or score again (scenario B). In scenario A the opponent gets the ball back at its 30 with 2:00 left. What are the odds that they will drive the length of the field and score a TD before time expires? Scenario B is much more complex because there are so many different possibilities: you might manage to pick up one first down and make your opponent use up his TOs before you have to punt, or you might turn the ball over, or you might pick up three first downs and run the clock out, or you might score a quick TD against a defense that is selling out against the run. I am not smart enough to figure out how to crunch numbers given such complexity.