I used to teach stats, and one of the fun exercises I would use to show that most people vastly underestimate streakiness is the following:

Assign 2 teams to generate a sequence of 100 coin flips.  One team should actually flip a coin and record the results.  The other team should just make up the results.

Teacher leaves the room while this is happening, returns after the teams have written their results down.  Teacher has to guess which list was generated by actual coin flips.  It never fails that the coin flips generate more - and longer - sequences of all H or all T.  The team that just makes up the results will almost never have more than 3 H or 3 T in a row, and often never have more than 2 in a row.  Real sequences of 100 coin flips have many more streaks than most people presume.

A similar exercise that blows most people's minds the first time it's presented to them: how many people do you need to have in a room before the chance that 2 or more of them share a birthday is greater than 50%?  Answer: 23.  Most people assume it's around 180.

And don't get me started on the Monty Hall problem...

Posted by teaparty on 10/10/2011 1:58:00 AM (view original):
Switch to the other curtain...

Always.

Posted by contrarian23 on 10/10/2011 1:53:00 AM (view original):
I used to teach stats, and one of the fun exercises I would use to show that most people vastly underestimate streakiness is the following:

Assign 2 teams to generate a sequence of 100 coin flips.  One team should actually flip a coin and record the results.  The other team should just make up the results.

Teacher leaves the room while this is happening, returns after the teams have written their results down.  Teacher has to guess which list was generated by actual coin flips.  It never fails that the coin flips generate more - and longer - sequences of all H or all T.  The team that just makes up the results will almost never have more than 3 H or 3 T in a row, and often never have more than 2 in a row.  Real sequences of 100 coin flips have many more streaks than most people presume.

A similar exercise that blows most people's minds the first time it's presented to them: how many people do you need to have in a room before the chance that 2 or more of them share a birthday is greater than 50%?  Answer: 23.  Most people assume it's around 180.

And don't get me started on the Monty Hall problem...

Interesting.  Knew about the birthdays.  Didn't know about the coin flip exercise.  You can get some sense of streakiness when you use SIM matchup  I'm always SImming two of my teams against each other or against someone else's team to see who's better, and it's amazing how many streaks of five, six or seven games you get even with relatively evenly matched teams.  It's also interesting how poor a predictor 50 sims is of longer term results.  You have to get up to at least 100 to 200 sims to get an accurate sense of where you'll come out after 500 sims, which is what I like to use as a benchmark.

Sure - the "average" streak will be 2.  But most people then make an error in their logic, and assume that the incidence of longer streaks will be much less frequent than they actually are.  There is, of course, a 1 in 8 chance of getting 3 H in a row, 1 in 16 of 4 in a row, 1 in 32 of 5 in a row, etc.  All of which still averages out to "average streak length = 2."  But you are more likely than not to see a string of five heads together somewhere in a sequence of 100 coin flips.  In the example I gave, the team that just makes up the outcome will literally almost never have a run of 5 H in their results.

If anyone is interested, the analysis of the "birthday problem" is here.  The math gets ugly further down the page, but the first few paragraphs are pretty self-explanatory.

Well, I (and many other owners) use the OL in open leagues for many of the same reasons you describe:
-- Cash in 200K scrubs
-- Take advantage of an unexpectedly good AAA player to upgrade another position
-- Correct a drafting mistake (example: I went light on IP, but ended up in a division with 2 Coors teams and the fatigue is killing me)

What I don't do is dump a "cold" player in the hopes of getting improved performance from someone else.  This has nothing to do with the question of whether you drafted the best possible team; it has to do with whether you believe players perform better or worse than statistical probabilities would dictate.

This last idea is where the rubber meets the road though, because many owners believe there is no advantage at all to be gained from doing this.  Others (and I put you in this category - my apologies if I am misunderstanding your argument) believe that you can and should do this to optimize your team - even after you pay the WW fees.  I have never seen the inner workings of the SIM algorithm, so I freely confess you could be right.  So my question is "how much value do you think you gain?" and therefore "how much would you be willing to pay for the ability to do this AND to limit other owners from being able to do it?"

Posted by contrarian23 on 10/10/2011 8:25:00 AM (view original):
If anyone is interested, the analysis of the "birthday problem" is here.  The math gets ugly further down the page, but the first few paragraphs are pretty self-explanatory.

I was going to guess that people bang more in the spring lol.

Since when have we not seen the innerworking of the sim algorithm????

To broow a phrase,
Luck... protects fools, small children and teams managed by crimsonblue.