goodr

Since beta, I've been a strong raider but recently, a friend convinced me to try arena. Now I'm definitely one of those guys who understood the raiding universe very well, but was quickly humbled by pvp. Instead of playing, I went and spent my time doing what I usually do with gaming - I made some useful stuff because I didn't quite like what was available out there.

My most recent project is part rankings site (yawn, another?), part historical archive of ratings and a big part about the really interesting data we want to dig up from the armory.. "if only we could". These tools aren't meant for everyone. Hopefully they encourage exploring, questioning and discovery. Hopefully we learn from it - in the same ways many of us approached wow web stats, swstats, recount, etc etc as a vehicle to make really interesting discoveries.

It's pretty vanilla at the moment, I want to make sure my basic approaches make sense to more than just me. I wanted to streamline a lot of what is usually presented in the rankings, highlighting specific data. I strip away a lot of mundane labels & information and exchange them for things I felt were more relevant and appropriate for reading at a glance. For instance, it's much more apparent when a team as a significant lead over the next team.

I have a ridiculous list of features that are almost done, being fleshed in, or on the horizon. I'd love to hear what you guys think so far and any suggestions you may have. Let me know if you like something and want more of it or don't like something so much and want me to tone it down. What other trend analysis would you like to see next?


One of the first discussions I've had with a few folks is about the display of team composition in team rankings. I'm currently using the 'default' way blizzard orders these classes, but someone suggested another way he felt was more relevant - ordering it by arena related roles:

warrior
rogue
hunter
mage
warlock
shaman
priest
paladin
druid

Which tends to give us MS debuff, ranged dps in the middle and healers at the end. This makes a lot of sense to me, but spreads out the 'drain' capable folks.

It's a good thought, but should it warrant a departure from the 'default'? What are your thoughts?

Thanks again for taking the time to look at this and give me your criticism. If you find it useful for yourself, don't forget to add the teams you're interested in tracking and it should show up by the next day.

playerversus.com - world of warcraft arena stats

 

Halfdane

Here's something I'd like to see:

What is the average gear rating for a rank of x across every server.

Say x = 1500. As a first cut, get the sum of the item rankings of the gear of each team member. The average of the team members is the gear average of the team.

Is it the case that a 1500 on BG9 has almost all of their honor/arena gear? What about BG6? If I want to get higher rankings with my current gear and skill level, which battlegroup should I transfer to?

 

goodr

That's an excellent idea.

One of the ways I used to rate recruits for raiding was via average itemlevel of gear. While that doesn't tell everything, it gives a pretty good sense of the value of the gear. In the arena environment, that's even truer with mostly one route for upgrades.

I agree with the usage of itemlevel as a way to compare yourself and your team to other teams of similar comp and gear level. For example, if I have a druid, warrior team with an average itemlevel of 140, what is the average rating? That's also a good thought on cross battlegroup comparisons as well.

 

Sillia

If you are going to do that, you should also build in some other requirements, like accidentally picking someone who isn't currently wearing pvp gear, or speccing for pvp. Spec is pretty difficult to tell, but you can check their total resilience rating as a filter. If they have under X resilience, you don't count them (200+ is a decent number), no matter what their current spec and gear, since you can tell that the gear they are wearing probably wouldn't provide meaningful data anyway.

 

◊ foolish_fool

Might want to be careful with this, considering that for some, pvp gear might not have that much resilience (T6 paladins come to mind).

edit: Straight up average Itemlevel (maybe excluding trinkets) is probably in itself an interesting question for comparing accross battlegroups. If you could give the average Itemlevel for teams, say, 1050-1150, 1150-1250 etc. up to 2450-2550 (at the top end you wouldnt expect that much difference, but still it is interesting to see) accross battlegroups, that would be cool.

 

goodr

You make a good point and it goes back to why I haven't presented gear or spec related stats. I have a method for more reliably assessing that soon, but until it's implemented, the data we look at for that kind of assessment is easily tainted by the situations you mentioned.

 

Aadar

Using ItemLevel has some faults as those people that have moved from raiding to PVP will have a huge Item Level but very low resilience and by as useful a chocolate teapot.

So perhaps you have 3 pieces of data; "Sum of ItemLevel of Items with Resilience", "Sum of ItemLevel of Items with NO Resilience", and "Sum of ItemLevel (regardless of resilience)".

 

goodr

What do you think about the ordering of classes in the at-a-glance listing? I want to make it easy to pick up what the group comp is without much processing involved.

 

Hutch

Here's an anecdote on the ordering/filtering:

I visit your site, click the arena link, and then the Hunter filter button.

In the resulting list, the top two teams (ExpertS and washed up has beens) do not display any Hunters.

Further investigating, ExpertS has one hunter on their roster, whose record is 0-0.
Same story for washed up has beens, who also have a warrior and warlock with 0-0 records.

I don't know if it makes any sense to exclude such characters from your data, or if it's even possible. But it is misleading and/or inconsistent to display teams with the <class> filter, whose members of <class> have insignificant or zero contribution to the team rating.

 

fip

I think that the data would still be useful even with these outliers (as I shall refer to them) in there. Honestly in most cases where resilience is needed, even resil blues are superior to pve epics.

Also, most likely this data won't even be useful past ~1800 rating. I think the bigger tell will be in the 1600-1700 range. At 1500 you could still have teams that were just created, but 1600 implies that you have won a few. Much above 1800 rating and I would be shocked to find anyone wearing less than full arena gear. I know that on bg9 at 1500 rating in both 3v3 and 5v5 you will face people in full season 2. Just to throw it out there.

 

goodr

It's been a point of debate for a while. First, it's definitely possible. I've done it and undone it.

Here's an example of a 2v2 team of 3 friends:
- Druid plays 100% of games
- Warrior plays 51% of games
- Warlock plays 49% of games

On the one hand, I want to make the results precise in team composition (avoiding characters bumming a ride, buying points, etc). I also potentially exclude perfectly valid results of alternates on a team. In the example above, I'm inevitably excluding one or the other.

On the other, I say, if a team wants to water down their value, roster, etc by bringing on characters bumming a ride, buying points, etc - that's their problem. I'd actually be happy to give them a reason to keep them honest. Additionally I don't accidentally exclude someone who's really on a team, but plays 49% of the games instead of 51%.

I feel like someone would be more frustrated to not find the team they were expecting because of a few percent in difference as opposed to finding extra teams. One important point to note is that I'm not interested in explicitly ranking teams - especially across battlegroups. I recognize there may be a difference from one battlegroup to the next, one team comp to the next and will leave Blizzard to deal with those issues.

Which one would you prefer?


In the team composition driven stats I'm working on, I take those factors into consideration. Using the same example, I weight the compositions for the sake of stats accordingly. In the case of, say, counting the number of teams in a certain rating range, the Druid+Warrior combo counts as 51% of a team. The Druid+Warlock combo counts as 49% of a team. Until the armory provides us with game-by-game stats, that's probably the best I can do.

Does that approach make sense?

 

Hutch

That approach would result in the 0-0 characters, in any combination with their teammates, counting as 0% of a team. But would that be reflected in the class filtering? So, if this weighting could be applied to the same example, would the Hunter filter only display teams whose hunter member(s) have played at least one game?

 

goodr

It's definitely possible to set a minimum threshold of 1. But then you have to think "You only need to play 1 game to be counted? That's silly. Let's increase it." That's where it gets less obvious. So let's say we set it at 10% or 20% of games. Personally, I still don't feel that warrants counting someone as 'part of the team'. So the filter is ineffective to me.

After that, it's a huge gray area, which is why we're having this discussion.

From a usability standpoint, I can see visitors easily passing over teams without that class in the class-comp listing and making the decision to include or exclude the teams themselves. If I eliminate them from showing up at all, I take that option away from visitors, regardless of their standing on that issue.

To more specifically answer your question, it depends on the context. For non specific browsing through regions, battlegroups, etc, I think being inclusive is a better approach. For situations where I am asserting particular stats (e.g. ratings distribution of teams with XYZ composition), the selection of teams is much stricter and its effect on stats will likely be weighted appropriately.

At least that's my thought right now.



Btw, are the icons in the rankings lists too small for anyone? I felt the were big enough to serve the purpose of quickly scanning and identifying presence of a class or group of classes. I believe most of us are familiar enough with the icons to pick them out even at half that size. In the case of making an individual selection, the icons used are twice the size (e.g. the class filter).