We’ve included a new app in the Labs section of the site that simulates the end-of-game scoring process using the standard Poisson scoring assumption. This is the same model that’s been used for the last 40 years, but got a richer treatment in Beaudoin and Swartz’s 2010 article in The American Statistician. (One can make the argument that these times are still too conservative, but for now, the fewer assumptions we make, the better.)
This assumes that you know the scoring rates of both teams at even strength, the trailing team with an extra attacker, and the leading team shooting on an empty net; the defaults are reasonable baseline values and can be adjusted with the sliders.
We run 100,000 simulations of goal-scoring times with these rates and these pull times with respect to 3rd-period deficits. The optimal time to pull with 1 or 2 goal deficits is calculated as the point at which the gain in the probability of tying the game is greatest and displayed as red and green lines respectively.