Comments on: Is Alex Semin A Streaky Shooter?

By: What We Learned: Only one solution to this Phil Kessel trainwreck « harveyrobb

Tue, 08 Feb 2011 11:07:10 +0000

[...] Capitals: Alex Semin is consistent after all. At least until someone else runs some stats that tell a different [...]

By: Neil - RMNB

Neil - RMNB — Mon, 07 Feb 2011 17:48:18 +0000

@Will I never said he was exactly like one or the other, just perhaps more consistent than we give him credit for.

@Colcurry I did the same for Phil Kessel as a FanShot over at Pension plan Puppets. Turns out he is WAY more consistent that people give him credit for. Also, will have Ovechkin post later today.

I agree, a 9.6% shooting percentage isn’t bad, but it is when comparing to an average of 14%. Plus, keep in mind we are looking for a cold shooter – not a terrible one.

Feel free to run one yourself using a 5 game moving average, calculate the standard error and lmk what you find.

By: Will

Will — Mon, 07 Feb 2011 16:06:51 +0000

This doesn’t make sense. Ignoring Mr Streaky, Semin is streakier than Mr Consistency. Mr Consistency is streakier than Mr Streaky in 2 of the 3 columns. I only have a basic understanding of Markov chains but either this data is a fluke or it is a bad model.

By: colcurry

colcurry — Mon, 07 Feb 2011 15:34:13 +0000

I think you need to run your simulations with other players to provide some context – do players always outperform their consistency/streaky models? Also, you might want to play around with the parameters a bit. A shooting percentage of 9.6 isn’t particularly bad, and the probability of maintaining a given state (90%) seems rather high.

Why not look at, say, 5 game moving averages and calculate the standard error? I guess it all depends on how you define consistency.

By: 28ISGREAT

28ISGREAT — Sun, 06 Feb 2011 22:15:13 +0000

Excellent post!

By: Russian Machine Never Breaks

Russian Machine Never Breaks — Sun, 06 Feb 2011 17:12:59 +0000

@KHtaD Yea. (HMM = hidden Markov model). #geeksrule

By: KHtaD

KHtaD — Sun, 06 Feb 2011 17:04:57 +0000

So this is essentially a 2-state HMM without the learning algorithm? I’ve been looking at doing a full-blown HMM for Semin’s shooting with a learning algorithm to discern what his true states or what the data look like if he is, in fact, modelable with a hidden Markov process.

By: Ian

Ian — Sun, 06 Feb 2011 07:04:59 +0000

Neil, just FYI. When I first read this post, it blew my mind. Great work.

By: bb

bb — Sun, 06 Feb 2011 04:25:50 +0000

I agree with Sakurafire up there.

By: DarkStranger44

DarkStranger44 — Sun, 06 Feb 2011 01:13:23 +0000

Good post!