Sweet alabama hot cakes eagle eye! Romer has no "theory". He simply did a probability analysis using observed data. This is not a "theory" - this is simple mathematics. The math is the same whether you're up 6, down 6, or tied. Using Romer's data, if BB thought 83% or better chance of getting the first, then it was a mathematically justified decision.
I made a slight calculation error, and my reading of his tables is inexact. Expectation on your own 25 is a little better than .5, and of your opponent on your 25 a little worse than -3 pts. But using those rough numbers, breakeven is then:
(success).5 = (1-success)3
Solving for (success) yields 6/7 or 85%.
Again, the actual numbers are rough and are mosty illustrative. I suspect Belichick has not only a precise measured function but possibly also adjustments for game conditions. 85% strikes Bayesian-me as a high success rate for 4 and 1, but what do I know