Statistical Analysis

**stiltr01** · Joined: 06 Jun 2007 Posts: 8

Now that we have a substantial number of users (10,000+) it should be easy to get good statistics on the Guitar Hero community as a whole. Even something as simple as community mean/standard deviation would be motivating for some, and give everyone a better sense of where they stand than just simple ranking. The data's all there, I just can't think of a simple way to export it all into a spreadsheet. Anyone else think it's a good idea?

**KurtCobain** · Posted: Sun Jan 06, 2008 7:06 pm Post subject:

**Cliff** · Joined: 06 May 2006 Posts: 3002 Location: Springfield, IL

**MPChedda** · Joined: 06 Nov 2007 Posts: 1299 Location: Evansville, IN

This actually is a pretty decent idea. It would be interesting to see how our community does cumulatively. It would give some of the lower ranked expert players something to look towards to say that they were above the average on SH.
_________________

**ricecake** · Posted: Mon Jan 07, 2008 6:01 am Post subject:

Along this same lines, I think it would be interesting to see the score distribution plotted on a graph, to see if the scores really follow a Gaussian (Normal) distribution or if they're somehow skewed due to the number of awesome players here.
_________________

DLC Quickplay Profile

**SpoonMan** · Joined: 06 Dec 2006 Posts: 3631

^ i think it would vary by song actually, there would be a normal curve for harder songs i'd imagine, but really easy FCs like HSB, yes we can, and closer would ramp up and flatten out i think due to the high number of close to upperbound scores on those songs.
_________________

My stupid GH achievements

My stupid RB achievements
Participate in the Master FC Breakdown Project! GH1 - GH2 - GH80s - GH3 - GH3 DLC - GHA - GHWT - GHM - GHSH - WTDLC - GHVH - GH5 - BH - GH5 DLC
All Guitar-Only Games
WINNING.

**thecaptainof** · Joined: 04 May 2007 Posts: 7571 Location: ¯\(°_o)/¯

I've got an awesome piece of statistical analysis software (Cliff: I'd guess you're familiar with SPSS?) that might be able to very quickly do exactly what you're talking about... and from tomorrow morning I've got very little to do until Thursday, so I'll give it a try.

EDIT:

A quick experiment using only a handful of values (in this case, all the scores entered for Memories Of The Grove medium, since there's only ten of them) gave this:

**tma** · Joined: 03 May 2007 Posts: 1414 Location: Australia

**Cliff** · Joined: 06 May 2006 Posts: 3002 Location: Springfield, IL

**Barfo** · Joined: 10 Oct 2006 Posts: 2596

Ive done decent amount of effort in the past on this sort of thing. The most detailed analysis i did was one time where i took the entire database at that time of GH80s scores (about 2-4k per song at the time, i think?) and plotted them all up in terms of the actual score distribution. I even started to create some measurements and categorizations of all the songs based on the properties of those distributions. Its decidedly non-gaussian distribution overall, except for some of the most tricky songs. Based on the disribution shape you can figure out some rough heuristics about the difficulty of the song per se, for example songs like The Warrior or WGTB look very similar to each other and much different from say PWM and 17 (which also may have some shape similarities). And within distributions with the same basic shape type you can definitely see differences which can hopefulyl be explained by (and perhaps be quanititative measures of) factors such as how easy the song is t FC, how easy/obvious the best non-squeeze SP path is, and also how easy/much points influence the squeezes are. I never really did a complete anaysis of even the GH80s data to try and rank songs based on all of these individual assements, and i meant to at some point revisit this whole idea in reference to GH3, but its not somethign ive gotten around to yet considering how much work it would be to manually go through and pull out the parameters of the distribution curves, and also i cant see a really pressing use for such data, other than just basic scientific interest.
_________________

Watching her, these things she said / "Time," she cried "failed to wait, this time"
***
Hush now / Let it go now / I know it's time to go / Time to let this fall / From my hands

**thegibbonator** · Posted: Tue Jan 08, 2008 10:37 pm Post subject:

**Monk** · Posted: Tue Jan 08, 2008 10:57 pm Post subject:

maybe instead of a distribution curve the scores them selves could be divided up into like a circular or bar graph type thing... like example....

out of the entire community on x song 10% of the ppl scored over 400K and x% scored under 100K

i guess it could still work as a distribution curve but the graph would probably work better since you would have a max of 10 bars/slices on any given single player song. and some songs would only have 1-2 (unless like in mississippi wqueen you diveded that bar up by 25K or 50K increments... Does that make any sense to you guys?

**stiltr01** · Joined: 06 Jun 2007 Posts: 8

Those heuristics are very interesting. I've suspected that there are a few songs that are "too" difficult, in the sense that their cutoffs are so high that a normal person can't be expected to 5* it. In that regard, I think I'd be interesting to see a sort of error function, that is, what % of the community scored worse than a given score.

**Barfo** · Joined: 10 Oct 2006 Posts: 2596

**Boon** · Joined: 25 Oct 2006 Posts: 269 Location: Paris France

Another interesting point using advanced statistics is inferring player profile i.e. what is your expected score for a song based on your other scores.

Each song as different characteristics : hard rythm, difficult chord, long solo, etc. ...
And everyone of us have different capabilities (that could nicely represented using a Kiviat diagram)
Maybe it is possible to go further than mean and standard deviation ?

But at least, those basic information for every song would be very interesting :
- number of player for every ranking (and a bar graph ?)
- number of score for every ranking (maybe interesting)
- mean score (for all player, for all score)
- Standard deviation
- median score
- total number of score
- total number of player having a score for this song
_________________

Goro is a lucky guy... He can play coop all by himself !

My blog (in French)