![]() |
|
Ladder Discussion Everything related to altitudeladder.com and the ladder servers goes here. |
![]() |
|
Thread Tools | Display Modes |
#1
|
|||
|
|||
![]()
I would like to see which side (left or right) each player was playing on be recorded. Ultimately the data would then be presented on the website. Despite most maps being mirrors, I do believe there have to be some trends as too which side has higher win % and different players playing better on different sides.
While the data doesn't have much potential to change ladder play, leagues would definitely benefit from this data being spaded. |
#2
|
|||
|
|||
![]() |
#3
|
|||
|
|||
![]()
Looking at u lost city
|
#4
|
|||
|
|||
![]() Quote:
|
#5
|
|||
|
|||
![]()
I agree this is interesting data, but we need "difference from expected win percentage given player rankings" -- not just raw win percentages. Depending on how the balancer works (and for smaller sample sizes, chance) it's very plausible that one side was expected to win 53%+ of the time (barring any left/right considerations) for a given set of games on a given map. Before claiming vindication on a particular asymmetric map, you might want to look into the perfectly symmetrical maps showing even more skewed left/right win percentages
![]() Last edited by lamster; 11-22-2011 at 06:01 AM. |
#6
|
|||
|
|||
![]()
Statistics my friend.. Given sufficiently large sample size you can just take the average win percentage.
|
#7
|
|||
|
|||
![]()
That is that the allocation of whether a team is given left and right is completely random. I mean if the team with the higher win percentage alway is assigned the right side, this will not hold.
|
#8
|
|||
|
|||
![]()
Most of these sample sizes are too small to actually make any definitive conclusions.
To put it in perspective, take Ball_Grotto for example. 410 games, left wins 43.4% of the time. That's 27 games short of a 50% balance and this is the most extreme example. (is this map even asymmetrical?) With Ball_lostcity we're looking at 13 games. Once you factor in Lamster's explanation and some other misc. factors, some amount of deviation should be expected. When we get into the 1000+ games played / map, than we can start changing maps. Also tbd_cave has been the historically not symmetrical map and it has the most games played and is only 1.26% away from 50% I forgot most of my statistics learnin but I think that if you actually did the calculations, even with this sample size, the side of the map you are on affecting the outcome of the game is well below 1%, probably somewhere closer to <.1% |
#9
|
|||
|
|||
![]()
Slightly offtopic, but it's clear to me why ball_lostcity favors the left side: the wall of specchat that clogs the bottom makes it impossible for the right team to mount any sort of offense along the bottom corridor.
|
#10
|
|||
|
|||
![]()
aYeah you have to do some power calculations, but 400 games can be considered a big smple size. T-values will not budge too much with an aditional 600 games played. But this is all easily done in stata or spss.
I suspect if there is such big differences in a symmetric map it is because ladder does not assign teams to either right or left side perfectly random. edit: or there exists some human failure, that makes it easier to play on the left side, because we are western and used to read from left to right lol Last edited by beefheart; 11-22-2011 at 06:49 AM. |
#11
|
|||
|
|||
![]()
make request, already exists, **** yea.
I would agree with Smush that sample sizes are too small to jump to any conclusions. But I'm happy this information is being calculated nonetheless. With near 10k games in only 3 months, and possibly season2's data made available, it is not unreasonable to think we will at some point record enough to gain insight. lam you bring up a good point. I really have no idea how exp win % is calculated. I wouldn't mind someone explaining how that is calculated. I suppose the least biased set of data would be season1 when teams were supposed to be dead even (exp win% 50/50). Unfortunately I don't think any info was collected back then. For anyone that cares, I calculated the overall Left Win % for all 9522 games and you get 48.855% (4652/9522), which is only 109 games off 50%. When you split it by ball and tbd things look a bit more interesting: ball 48.245% 2969/6154 108 games off 50% tbd 49.9703% 1683/3368 1 game off 50% |
#12
|
|||
|
|||
![]() Quote:
Quote:
|
#13
|
|||
|
|||
![]()
these maps seem pretty darn fair from a statistical perspective.
if you require (conventionally) 95% confidence to reject the null hypothesis, then we cannot reject the null hypothesis for the vast majority of the maps. for example, for ball_lostcity, the left side has won 53.69% of the time in 352 matches. but the standard error for 352 matches is roughly 2.665%, meaning that a 95% confidence interval (1.96 standard errors) gives us a range of 50% +/- 5.22%, or a range of 44.78% to 55.22%. because 53.69% is within that range, there's really no reason to think that if you played it infinite more times it'd approach 50.00000%. another way to think about this is to calculate how many standard errors away from 50.00% each map is. ball_lostcity is 3.69% above 50%, so it's 3.69/2.665=1.385 standard errors above 50%. if it's within two standard errors, then we can't say it's unfair or imbalanced. so i'm forced to conclude that there's no reason to think ball_lostcity is imbalanced. there is, however, one map that (according to this analysis) may not be balanced. the left side has won ball_grotto only 43.24% of the time in 414 games, which is 6.76% less than 50% and thus 6.76/2.457=2.75 standard errors less than expected. in other words, we're 95% confident that ball_grotto is unfair and imbalanced. but there's a catch! there are 31 maps that are played with any frequency. and if we're imposing a 95% confidence interval, then we'd expect that, purely by chance, one in twenty maps would be outside that confidence interval, even if all of the maps were truly fair and balanced. and ball_grotto isn't really that far away from 50% (2.75 standard errors only). so according to my meta-analysis, i conclude that all of the maps are essentially fair and balanced -- or, more precisely, there is no reason to think from these statistics that they are unfair or imbalanced. |
![]() |
|
|