Its a weird situation because analytics do t run the scenario for a tie game. Normally if you are down a goal, you pull the goalie cuz otherwise the opponent can kill time til overtime passing the puck in their own zone if you dont get posession because the other team has no need to score another goal and take risks and can go into defensive shell.
But in a tie game Unless the other team is securing playoff position with 1 point which caos werent, they arent tryibg to go in defensive shell and not just trying to kill clock so you can still capitalize on possible turnovers or counter attacks if you play normally.
It’s not just about beating a defensive shell - it’s about creating a higher event probability. If the status quo means a loss, then increasing the likelihood of that status quo changing is beneficial, even if the increased likelihood favors the other team more than yours.
Think of it this way: Philly is tied, which for them in this playoff scenario equates to a loss. Suppose that each team has a 5% chance of scoring a goal in the last minute of play at even strength. To win, the Flyers need to score AND they need the Caps to not score; the probability of that (assuming for the moment that the events are independent) is 0.05 * (1-0.05) = 0.0475, or 4.75%.
Now suppose that by pulling his goalie, Torts can increase the probability of his team scoring to 15%, at the cost of also increasing his opponent’s probability of scoring to 50%. Now the probability of the Flyers scoring and the Caps not scoring (again assuming independence) becomes 0.15 * (1-0.5) = 0.075, or 7.5%. Torts has thereby improved his team’s chances of winning by increasing the odds of scoring for both teams, even though he increased the Caps’ odds far more than his own.
This is of course a drastic oversimplification. To begin with each team is capable of scoring more than once in the final minute of play. In addition the events are not independent: if the Flyers were to score first they would put their goalie back in, which would then impact the probabilities of both teams scoring from that point forward. So the actual calculation would be far more complex, and that doesn’t even get into the problem of accurately determining the underlying scoring probabilities in the first place, or the fact that we’re talking about calculating probabilities deriving from a single decision while ignoring the impact of numerous other decisions that can be made by both sides. This is intended merely as a thought experiment - a simple way of putting some numbers to a concept that may not be easily quantifiable in practice, but which can be intuitively understood.