|You were so preoccupied with whether or not you could,|
you didn't stop to think whether you should.
Continuing work from last year's playoffs, I am sharing live win probability graphs on my site for all regular season games (although I'm still working out some kinks in the implementation).
In addition to the standard win probability metrics from last year, there is a new table that accompanies each graph, which I'm calling the "SO" Calculator ("SO" being short for Significant Other).
Most of us have been in the position where our wife/boyfriend/child/etc. does not care about a game we are watching and simply wants to know when it's going to be done. And most of us lie. If there is 4:55 on the clock, well say "about five minutes". And with a straight face too.
But now, with the power linear regression, you can restore honesty to your relationship. Or get called on your bullshit.
Here's how it works: As the games progress, there is a table above the graph updates real time. It looks like this:
Projected end times are shown in Pacific time (my model, my time zone).
This is based on a local regression model that estimates average real time remaining in a game as a function of game time and scoring margin. Close games tend to take longer, both because of the possibility of overtime, as well as the increased number of timeouts, fouls, and free throw shots. Compare the expected remaining time for a game with a one point differential to a 15 point game:
One additional variable the model considers is whether the game is nationally televised. Due to more commercials, nationally televised games (aside from those on NBA TV) tend to take fifteen minutes longer than local games.
Just a note that the model projects averages, which can be misleading. For a close game in the final minutes, the average will be skewed by the probability of overtime. If the game ends in regulation, the expected time left may be only 5 minutes, but for an overtime game, you may have 20 minutes to go. The calculator averages those two possibilities into a potentially misleading 12 minute estimate.