Our End-of-Year Geek support drive ends in 16 days! It has been your support,
financial and otherwise, that has allowed this site to become what it is today.
Learn More

Sorry for missing last week. We gained a Rising Sun and a Chartstone and lost a Cosmic Encounter and Combat Cammnder. Paths of Glory is in the anchor position. Only one game that entered the charts in 2009 remains and it is in the bottom 10.

No change in top 10, but a few maybe close. Possible upcoming promotions include CoB over 7WD, Scythe over Terra Mystica, and (in maybe a couple months) Terraforming Mars over Twilight Struggle. Although by then some newcomers may be on the scene, KDM and Gaia Project look like contenders.

Why is Gaia Project above KD:M with lower avg. score and fewer votes?

Hmm good question. Does Geek rating weigh against people who’ve only voted for 1 or 2 games? It must do some odd and complicated things,seeing as I hear it has an end of month adjustment.

Last edited Thu Mar 15, 2018 3:03 pm (Total Number of Edits: 1)

I feel like this has come up before, and the answer seemed to have something to do with the lower-ranked game having a greater distribution of scores. Or something. Something fantastically nerdy.

The problem in the equation is people. Eliminate that, everything works.

ShaneXtopher wrote:

I feel like this has come up before, and the answer seemed to have something to do with the lower-ranked game having a greater distribution of scores. Or something. Something fantastically nerdy.

It's a top secret formula. If they told us how it worked, they'd have to kill us. I'm happy remaining alive and ignorant. Also Gaia Project deserves to be above KD:M, because Euros are better than trash.

Why is Gaia Project above KD:M with lower avg. score and fewer votes?

Indeed, this has been noticed in various cases and KDM seems to be particularly affected. Working theories include standard deviation as part of the geek rating formula (higher = worse) and/or a "shill filter" that disqualifies some votes individually. In addition, there is clearly something that happens at the beginning of each month, seemingly an algorithm applied at that time that is different from the ~daily updates.

The admins do not disclose the true answers, so nobody really knows. I guess they keep this secret in part to try to minimize manipulation - people can game the system much more effectively if they know exactly how it works.

I feel like this has come up before, and the answer seemed to have something to do with the lower-ranked game having a greater distribution of scores. Or something. Something fantastically nerdy.

It is called standard deviation. For a game's tracking of it, look at the stats page for a game. It is one reason why a game like Concordia does so well as its standard deviation is 1.18. Counter point, Gloomhaven's is 1.6.

The problem in the equation is people. Eliminate that, everything works.

matthean wrote:

ShaneXtopher wrote:

I feel like this has come up before, and the answer seemed to have something to do with the lower-ranked game having a greater distribution of scores. Or something. Something fantastically nerdy. :laugh:

It is called standard deviation. For a game's tracking of it, look at the stats page for a game. It is one reason why a game like Concordia does so well as its standard deviation is 1.18. Counter point, Gloomhaven's is 1.6.

Concordia is such a bland game and I like it so much. That's why.

I'm not sure why they would use standard deviation/variance/etc. as a parameter in geek rating -- geek rating is (theoretically) a Bayesian average, it doesn't really care about variance.

From reading the other thread, it sounds like it's more of the anti-shill guard -- if a game has a lot of 10 ratings from users who logged in once and rated one game, then geek rating discards those ratings (or, also likely, weights them less than other votes.)

At some point I want to run maximum likelihood estimation just to estimate their parameters for this stuff (and see if I'm wrong about standard deviation being a parameter,) but I'm told that if you peek behind the curtain you'll be violently murdered.

If you've already done it, PM me and I'll murder you be happy to review your results!

I'm not sure why they would use standard deviation/variance/etc. as a parameter in geek rating -- geek rating is (theoretically) a Bayesian average, it doesn't really care about variance.

From reading the other thread, it sounds like it's more of the anti-shill guard -- if a game has a lot of 10 ratings from users who logged in once and rated one game, then geek rating discards those ratings (or, also likely, weights them less than other votes.)

At some point I want to run maximum likelihood estimation just to estimate their parameters for this stuff (and see if I'm wrong about standard deviation being a parameter,) but I'm told that if you peek behind the curtain you'll be violently murdered.

If you've already done it, PM me and I'll murder you be happy to review your results!

Bayes' theorem claims that posterior distribution is proportional to prior distribution times likelihood function, divided by a normalizing constant.

Geek votes are an informative prior. Its expected value depends on variance.

I'm not sure why they would use standard deviation/variance/etc. as a parameter in geek rating -- geek rating is (theoretically) a Bayesian average, it doesn't really care about variance.

From reading the other thread, it sounds like it's more of the anti-shill guard -- if a game has a lot of 10 ratings from users who logged in once and rated one game, then geek rating discards those ratings (or, also likely, weights them less than other votes.)

At some point I want to run maximum likelihood estimation just to estimate their parameters for this stuff (and see if I'm wrong about standard deviation being a parameter,) but I'm told that if you peek behind the curtain you'll be violently murdered.

If you've already done it, PM me and I'll murder you be happy to review your results!

Bayes' theorem claims that posterior distribution is proportional to prior distribution times likelihood function, divided by a normalizing constant.

Geek votes are an informative prior. Its expected value depends on variance.

You're not wrong, but the term "bayesian average" generally means this:

A Bayesian average is a method of estimating the mean of a population using outside information, especially a pre-existing belief,[1] that is factored into the calculation. This is a central feature of Bayesian interpretation. This is relevant when the available data set is small.[2]

Calculating the Bayesian average uses the prior mean m and a constant C. C is assigned a value that is proportional to the typical data set size. The value is larger when the expected variation between data sets (within the larger population) is small. It is smaller when the data sets are expected to vary substantially from one another.

{\displaystyle {\bar {x}}={Cm+\sum _{i=1}^{n}x_{i} \over C+n}} {\displaystyle {\bar {x}}={Cm+\sum _{i=1}^{n}x_{i} \over C+n}} This is equivalent to adding C data points of value m to the data set.

Note there's no STDev term. You pick the value of C (the number of average votes you're going to chuck into the equation) based on how variable you think it'll be, but it won't vary inside the function (e.g., Gaia Project couldn't outrank KDM if the above was the entire formula.)

If they used STDev or whatnot, it'd be a useful prior, but it wouldn't be what's usually called a Bayesian average. It might vary widely based on your discipline, of course!

Cheers!

Last edited Mon Mar 19, 2018 12:30 am (Total Number of Edits: 2)

I'm not sure why they would use standard deviation/variance/etc. as a parameter in geek rating -- geek rating is (theoretically) a Bayesian average, it doesn't really care about variance.

Regardless of how it plays out, I have yet to see a game which one would think to be higher ranked and isn't and have the better STD rating.