Thanks Yertle - can you run it one more time? I just need a single row of the output e.g.:

Team Vision: 0 Entry Seat: 2 Player Seat: 1 Target Seat: 2
000

stdClass Object
(
    [logid] => 42659421
    [gameid] => 52459
    [turnid] => 165
    [timestamp] => 1299636467
    [action] => transfer
    [s] => 2
    [gp] => 
    [tid] => 0
    [fromid] => 15
    [toid] => 14
    [count] => 3
    [seats] => 
    [cards] => 
    [ad] => 
    [dd] => 
    [bmod] => 
    [al] => 0
    [dl] => 0
    [hidden] => 
)

Soooo, when I first clicked on that link this time I believe it showed the stats correctly, but it didn't have the debug info.  Then when I went back to the link or manually entered it I get the incorrect stats with the debug:

(For the record this is on Firefox beta 4.0, but I'm also seeing it on IE 8 too)

Team Vision: 0 Entry Seat: Player Seat: 1 Target Seat: Fog:
000

stdClass Object
(
    [logid] => 42025776
    [gameid] => 52459
    [turnid] => 1
    [timestamp] => 1299272948
    [action] => join_game
    [s] => 
    [gp] => 2e0545c8fcabba49666d7f1ab9325c74
    [tid] => 0
    [fromid] => 0
    [toid] => 0
    [count] => 0
    [seats] => 
    [cards] => 
    [ad] => 
    [dd] => 
    [bmod] => 
    [al] => 0
    [dl] => 0
    [hidden] => 
)

Ah ha. OK this should be fixed now.

ps - the problem only happened after a turn had been taken in the game so please test while you are in the middle of a turn to be certain it is properly resolved.

Interesting problem!  Appears to be more better from what I see in my games!  Awesome work and thanks tom!

I'm glad this is fixed.  I was curious - what are people's personal best/worst on the luck stats?  Obviously it matters what game type it is (as to how high/low you can get).  I got +46.89 on a Gotham game that ended today...

I believe this +46 is very very lucky but you have to consider how many fights you had to do to get to this figure. Another column with a kind of normalised luck for 100 dice would be very usefull. Tom ?

Toto wrote:

I believe this +46 is very very lucky but you have to consider how many fights you had to do to get to this figure. Another column with a kind of normalised luck for 100 dice would be very usefull. Tom ?

Toto, have you read this thread?

http://www.wargear.net/forum/showthread/1458

There are other threads if you do some digging.

This is an ongoing and complicated topic. Right now the use of z-scores is pretty much at the top of the list of possibilities.  Bear in mind that most standard deviation related calculations are potentially very draining on the system. 

Toto wrote:

I believe this +46 is very very lucky but you have to consider how many fights you had to do to get to this figure. Another column with a kind of normalised luck for 100 dice would be very usefull. Tom ?

I think the number of battles is there - looking at the number of Kills and Deaths - that gives you an idea kinda.

On the normalized luck, I believe that would just be 0 (in the long run, luck evens out).

Cona Chris wrote:

 

On the normalized luck, I believe that would just be 0 (in the long run, luck evens out).

I've heard this before and I take issue with it.   I would say the opposite.  In the long run luck rarely evens out,

Take the normal distribution for 1 million fair coin tosses.   Even though it is the most likely outcome, the probability that you would flip exactly 500,000 heads and 500,000 tails is extremely low.   More to the point, it is extremely likely that one of the two choices will predominate.

Translate that to WG luckspeak, and I'd say it's likely that no one on this site who has played more than 100 games has a net luck score that's less than 1.0

M57 wrote:

Take the normal distribution for 1 million fair coin tosses.   Even though it is the most likely outcome, the probability that you would flip exactly 500,000 heads and 500,000 tails is extremely low.  

True, but the chance you come close is extremely high.

Amidon37 wrote:

M57 wrote:

Take the normal distribution for 1 million fair coin tosses.   Even though it is the most likely outcome, the probability that you would flip exactly 500,000 heads and 500,000 tails is extremely low.  

True, but the chance you come close is extremely high.

Define close.

Let's be more specific.  Let's flip a coin 1 million times and look at a few very possible results.

With a luck score of 100, you would have to roll 500,100 heads.  The probability of rolling 500,100 heads is ~42%.   This would give you a z-score of 0.2, which on my proposed scale where 100 is phenomenally lucky would be a revised luck score of 6.6

A luck score of 1000 seems pretty unlikely, but in fact it will occur approximately 2.2% of the time. for a z-score of 2 and a revised luck score of 66.

What would it take to get a z-score of 3 (a 99 on the revised luck score scale)? About 3000 more heads than tails (501,500 heads), which would be the equivalent of a current 1500 WG luck score.

M57 wrote:

Amidon37 wrote:

M57 wrote:

Take the normal distribution for 1 million fair coin tosses.   Even though it is the most likely outcome, the probability that you would flip exactly 500,000 heads and 500,000 tails is extremely low.  

True, but the chance you come close is extremely high.

Define close.

Let's be more specific.  Let's flip a coin 1 million times and look at a few very possible results.

With a luck score of 100, you would have to roll 500,100 heads.  The probability of rolling 500,100 heads is ~42%.   This would give you a z-score of 0.2, which on my proposed scale where 100 is phenomenally lucky would be a revised luck score of 6.6

A luck score of 1000 seems pretty unlikely, but in fact it will occur approximately 2.2% of the time. for a z-score of 2 and a revised luck score of 66.

What would it take to get a z-score of 3 (a 99 on the revised luck score scale)? About 3000 more heads than tails (501,500 heads), which would be the equivalent of a current 1500 WG luck score.

I would agree that the chance of it being exactly 0.00 is extremely small, but 0.00 is the expected value... which is what I thought Toto was trying to see. 

And yes, luck never is exactly even in the end for everyone - but people are not consistently always lucky or always unlucky in each game.  You'll have lucky games and unlucky ones, and on average they will cancel each other out somewhat (but not exactly) over time. 

If I remember my stats correctly, you would expect your absolute luck (i.e. # of undeserved kills/deaths) to get larger (or smaller) as you play an infinite # of games, but your normalized luck (i.e. luck divided by # of rolls), should approach zero.

Ozyman wrote:

If I remember my stats correctly, you would expect your absolute luck (i.e. # of undeserved kills/deaths) to get larger (or smaller) as you play an infinite # of games, but your normalized luck (i.e. luck divided by # of rolls), should approach zero.

Ozy,

Yes, but we are unconcerned about normalized luck and we are very concerned about our absolute luck (luck over a finite period of rolls).

Cona,

In an earlier post you implicetely expressed an interest in determining how lucky that 47 was by asking if others to post their personal best scores.    My point is that my 20 may be a whole lot luckier than your 46 if I took less rolls to achieve it.Another example: Over time if I rolled 49,500 winners where 49,000 was the expected amount.  Who was luckier in the aggregate, you or me?

We need to develop a stat that gives the current luck stat number perspective.   How do I evaluate a luck stat of 45 that was acquired over the span of 1000 rolls?

M57 wrote:

Ozyman wrote:

If I remember my stats correctly, you would expect your absolute luck (i.e. # of undeserved kills/deaths) to get larger (or smaller) as you play an infinite # of games, but your normalized luck (i.e. luck divided by # of rolls), should approach zero.

Ozy,

Yes, but we are unconcerned about normalized luck and we are very concerned about our absolute luck (luck over a finite period of rolls).

Cona,

In an earlier post you implicetely expressed an interest in determining how lucky that 47 was by asking if others to post their personal best scores.    My point is that my 20 may be a whole lot luckier than your 46 if I took less rolls to achieve it.Another example: Over time if I rolled 49,500 winners where 49,000 was the expected amount.  Who was luckier in the aggregate, you or me?

We need to develop a stat that gives the current luck stat number perspective.   How do I evaluate a luck stat of 45 that was acquired over the span of 1000 rolls?

 I would agree.  Perhaps it's as simple as just dividing the luck stat by number of rolls, with some adjustment for standard deviation?  How that would be done I would have to realy think about though... I'm thinking trinomial distribution (since battles can be lose 2, each lose 1, win 2)? 

I'm sure there is more to it than that because - I would think - if you have a lot of 3 vs. 1 your chances for worse luck are not as great as with 3 vs. 2 just because you can lose more in the later scenario.   Might be more complicated than it's worth. 

Cona Chris wrote:

 I would agree.  Perhaps it's as simple as just dividing the luck stat by number of rolls, with some adjustment for standard deviation?  How that would be done I would have to realy think about though... I'm thinking trinomial distribution (since battles can be lose 2, each lose 1, win 2)? 

I'm sure there is more to it than that because - I would think - if you have a lot of 3 vs. 1 your chances for worse luck are not as great as with 3 vs. 2 just because you can lose more in the later scenario.   Might be more complicated than it's worth. 

 

C, there are a few threads here devoted to this subject.  Luck stat divided by # of rolls is a bust.

Standard deviation is a bear for the system to run as the number of rolls gets high, but something close to a z-score is definitely the goal.

Cona Chris wrote:

 I would agree.  Perhaps it's as simple as just dividing the luck stat by number of rolls, with some adjustment for standard deviation?  How that would be done I would have to realy think about though... I'm thinking trinomial distribution (since battles can be lose 2, each lose 1, win 2)? 

I'm sure there is more to it than that because - I would think - if you have a lot of 3 vs. 1 your chances for worse luck are not as great as with 3 vs. 2 just because you can lose more in the later scenario.   Might be more complicated than it's worth. 

This is pretty close to what we're thinking. For the binomial distribution, the SD is sqrt(p(1-p)n) where p is the probability of success and n is the number of trials (rolls). I haven't looked it up, but I believe trinomial will have a similarly simple SD calculation. So, you'd divide luck stats by SD to get a z-score, and if you were so inclined, you could get a percentile indicating how lucky the rolling was from the z-score. Since our actual distributions are discrete, not continuous, we'd need to do a continuity correction.

As you indicate, if there are a lot of mixed types, even just 3v2 and 3v1, you have to do something more sophisticated. The distribution in that case will be multivariate (success for each type being a distinct variable).

Doesn't a z-score calculate the number of SD's away from the mean?

I'm into it, should be easy to calculate.

Some sort of pooled/unpooled SD should do.