10 Apr 2011

Bribes, Karma and Recommendations

Patch 4.1 is approaching, although when it’ll actually arrive is currently anyone’s guess. Content and changes are being crammed in on a regular basis, making predicting the arrival date sheer guesswork at the moment.

Call to Arms is one of the latest features to be squeezed into the patch. Essentially the change amounts to a bribe to persuade tanks (and occasionally healers) to queue by themselves for heroic groups by offering them an additional reward when they complete the dungeon.

The rewards on offer to those that respond to the Call are a mix of the following:

  • More gold
  • A chance of a rare gem, flasks or potions (basically more gold)
  • A good chance of a dungeon drop only non-combat pet
  • A rare chance of a dungeon drop only mount

Although the Call to Arms can in theory be triggered for any role, it’ll always go for the one that’s needed most. In the majority of cases this will be the tank, although occasionally it’ll be healers that are most in need. It is highly unlikely that this will trigger for DPS roles.

The goal is that this will reduce the queue times for DPS, providing them with a better LFD experience.

It will not work.

Firstly, the benefit (if there is any) is likely to be short-lived. Once a tank has stocked up on the rare items they want, they no longer have an incentive to queue. There will be faster ways to make gold – just look at the number of blogs offering solid gold-making strategies and advice. If Blizzard want to avoid tanks tailing off again then they’ll need to either increase the size of the bribe or continually swap it around to lure people back. Have a look at research on demand decay and demand stimulation for more information on this phenomenon.

Secondly, the difference will likely be insignificant. Due to the relationship between the tanks and the group (you need one tank per group), it’s really easy to calculate the effect that increasing the number of tanks in the queue will have. If the current rate of tanks joining LFD results in a queue time of x for DPS, twice the number of tanks will result in half the queue time. If we want to get to queue times of x/y minutes for DPS then we’ll need y times as many tanks joining the LFD queue. And that’s assuming the healers are there to meet demand as well.

To put that in numbers, going from a 60 minute queue to 6 minutes for DPS will need ten times as many tanks joining up.

I’m happy to be charitable to Blizzard and assume that they might, if they’re lucky, pull off a tenfold increase in the number of tanks entering LFD on their own. But I promise you that it won’t sustain itself longer than a few weeks. If more DPS start signing up to take advantage of shorter queues then the whole effect counterbalances.

Thirdly, as put so eloquently by The Daily Blink, the change will attract the wrong sort of tank. The tank who’s mostly in his greens, or DPS blues, or PVP gear just to fudge the iLevel hurdle. The kind of tank that’s squishier than a plushie and sends the healer into meltdown. This isn’t about tanks that want to do a good job, it’s about players that aren’t really interested in tanking and are only there to milk the bribe.

The comments on various forums are littered with players saying that this will give them a reason to tank on their alt, or to use their healing offspec more. The mainspec tanks and healers seem to be mysteriously quiet, yet it’s these folks that are the ones you want to encourage the most.

And that’s just the start of it.

The Community Effect

Bribing someone to complete a particular task isn’t something that sits well with everyone. There are tanks out there who don’t like the idea of being bribed, while some others feel that the price isn’t high enough for them to enter LFD. That said, it might result in the end of the current (although rare) situation where players pay a tank in gold to run a heroic instance with them.

There’s also resentment from within the DPS community that they’re being relegated to third class citizens in the game through the creation of a reward that they are highly unlikely to obtain. Add to that the idea that 1) the tank gets the reward but the queue time doesn’t significantly drop and 2) that you get a tank who can’t tell his sword from his shield and is just there for the “phat loots” and you get DPS that wonder what the whole point of Call to Arms was anyway. You can see it happen.

There’s also the feeling that DPS will have paid for the right to be abusive because they’re not getting the special goody bag. By throwing a sweetner to those in demand, Blizzard seem to be silently acknowledging that the LFD experience can be abusive, and that there’s not much they’re either willing or able to do about it. Such tacit approval of players’ willingness to don the asshat will only make community divisions worse. You think things are bad as a tank in Pugs now? Wait till you’re getting paid by Blizzard to be there.

The Root Problem

Let’s be honest, tanks and healers don’t want to use LFD because of the amount of abusive behaviour they get dealt with. It’s not so much that the job they perform is hard but that they have responsibility heaped on them – pulls, mob marking, tactics and pacing are usually placed on the tank’s shoulders. Get it wrong and watch the muck-flinging commence as the DPS try to cover up that they can’t watch their aggro, can’t crowd control and can’t follow target assignment.

In fact, with each LFD pool now covering an entire region (US, Europe, etc) it’s fair to say that you’re so unlikely to come into contact with the same group of players again that you might as well be anonymous. It’s as if the tool spawned a version of Gabriel’s Greater Internet Asshat Theory. These people are so unlikely to have an effect on your continued gaming that you might as well treat them however you like.The lack of consequences to their actions mean that players can (and frequently do) behave as they please.

Back in the early days of Warcraft this whole culture of player abuse was much less of a problem. Levelling was a much longer process, meaning that characters were less disposable. If you earned a reputation for being a bad or abusive player word would get around quickly and you’d find it hard if not impossible to get pick-up-groups. Being found out as a ninja was pretty much a kiss of death.

Faction changes, server transfers and renames eroded this. A reputation could be wiped clean with a new identity, but you’d have to pay cash to get it. Only the biggest scams were worth paying that kind of money for.

LFD removed the final barrier to paying for the consequences of your actions. There’s now no consequential obligation to be reasonable to other players through LFD, so why bother? You’re unlikely to meet them again and they’re unlikely to impinge on your game once the dungeon is finished.

The problem isn’t how to persuade tanks and healers to use LFD more in it’s current form, or to shoehorn more DPS into a group. It’s not even that players don’t like to tank or heal. It’s about how to remove the negative aspects of a system that’s so random as to be virtually anonymous.

The Karma Alternative

A frequently suggested enhancement to LFD is to introduce some kind of karma or reputation based system in order to prioritise players that work well in LFD groups. On the face of it this sounds great – just rate your fellow group members at the end of the run and the high scorers get shorter queue times as a result.

Unfortunately there’s a few problems with a straight karma system like this:

  • There’s no “agreed definition” on what makes a good player
  • There’s a perception that the system can be “gamed” through player responses, such as organised or block voting.

One suggestion is to go back to server-bounded queues to enforce reputation, but is that practical when looking at the vast imbalances in population that we’ve seen? It’s likely that LFD and accessible server transfers are the only thing preventing wholesale realm mergers.

Stealing Netflix’s Playbook

So if we can’t agree on what we like in a LFD player and yet still want a better form of group matching than straightforward “time in queue”, what other options are out there? It just so happens that there’s a form of selection and suggestion algorithms tailored to do this very job.

If you’ve bought anything online then you’ve probably come across a box on a website that suggests other things you might be interested in to go along with your past or current purchases. Sometimes these are pretty basic, based on what other people bought to go with the thing you’re purchasing. Sometimes they’re a bit smarter and based on the things that you’ve bought or browsed. Amazon’s great at doing this.

The smartest ones not only look at the things you’ve said you like, but also what other people with similar interests to you also like. And from these heavy-duty recommendation engines, Netflix is the one people point to.

For Amazon the reason for having the engine is obvious – they want to sell you more stuff. For Netflix it’s more about keeping you as a subscriber, so it’s in their interest to feed you with a steady stream of movies that it’s likely that you’ll enjoy.

The great thing is that the ability to game the system is much more limited. Since you’re tracking a network of preferences (both explicit and implied) across a number of relationships the ability to influence one single ‘popularity’ figure is diminished. Besides, the only person you’d be gaming is yourself by reducing the cluster of people you would be grouped with.

How would you go about building a recommendation engine? Ask a simple question at the end of each dungeon group:

“Would you group with this player (in this role) in the future?”

Why is this relevant? Well, Warcraft is also subscription based, with a goal is to keep you playing your $15 or £9 a month. At endgame, there’s little content to choose from except raiding and heroic instances. And just as with Netflix giving you films you’ll enjoy, it’s in Blizzard’s interest to put you in dungeon groups that you’ll have fun in. If you’ve got a guild that you can regularly do that with then fantastic. If on the other hand you’re reliant on LFD because outside of raid days your guild is like the Mary Celeste, this is going to be a huge bonus.

The challenge for Warcraft is in creating multi-way recommendation in the same way, and that’s going to be a challenge. Netflix solved the problem back in 2009 by offering a million dollar prize for the best solution. If Blizzard are going to start offering incentives, wouldn’t it be better for them to ask on ways to improve the game rather than bribing us to play a broken system?

The Need for Solutions

There’s a simple reason why Blizzard need to solve this problem – subscriber churn rates. In order to grow the number of subscribers playing a game, you need more people joining up than you have leaving.

Cataclysm was deliberately designed to stimulate the number of new subscribers by improving all the old-world content. What it didn’t do to a huge extent was look at the reasons why people leave, namely burnout with endgame content. As the number of new joiners slows it’ll become more important for Blizzard to focus on this in order to avoid a dip in subscriber numbers.

In the longer term, it’s important for MMO developers to examine this problem as it’s a serious wrinkle in the gameplay experience that they’re currently unable to control. By solving this problem, they gain the ability to deliver more rewarding co-operative group challenges that can be spontaneously or dynamically assembled.

In other words, you get the groups you like. Surely that’s worth something?

Like this? Try these other related posts:

Tags: , , , ,

16 Responses to Bribes, Karma and Recommendations

  1. Rezznul says:

    You have made some very valid points. I love that idea of a recommendation engine, that’d be wonderful.

    I understand their wanting to make random queueing as quick as possible for everyone, but this way seems to be so polarizing. Hopefully, they can take the feedback they’re getting and tweak things out to make it appealing for everyone.
    Rezznul recently posted..Raiders of the Lost Noob A Tree Grows in Azeroth — Part 4My Profile

    • Gazimoff says:

      I really hope they start looking beyond pure game mechanics or strategies and towards more online commerce or social networking techniques. There’s just so much that can be borrowed in terms of promoting good co-operative grouping that they’d be doing themselves a disservice by ignoring it.

      Sometimes it’s better to fix things properly than quickly.

  2. Guillin says:

    Very good article, just by simply announcing this Blizzard has unintendedly increased the angst amonst players in an already frustrated playbase. It’s impossible to see this a short term fix, with long term reprecussions and for me it’s an example of Blizzard not actually willing to fix the issue(s) but just putting a band aid over it and hope that people forget about what’s underneath.
    Guillin recently posted..3 Issues Affecting WoW Dungeoning Right NowMy Profile

    • Gazimoff says:

      Yep, they’re pushing people further into guild grouping through the reputation mechanics etc. They don’t want to fix it, they want to make it redundant or irrelevant. Fact is that it remains the only tool to get groups for a large chunk of the playerbase.

  3. Meli says:

    Ultimately, this all comes down to forcing people to behave with civility. I am not sure that can be done on the internet. Until people start treating others with respect, this situation will only continue to deteriorate.

    Reading your thoughts, there is a certain merit to what you suggest, but I would like to look at some of the other things in place. I see a number of people around the web talking about Blizzard encourage ‘guild’ groups. There is no question premades can be much more effective, but that means giving up the random stat bonuses for true LFD groups. For some, particularly groups that are just barely over over the 329ilvl gear check, that can be a deal killer.

    Though my mage has been my main for years, I also have a tank (and a healer too). I’ve geared the tank almost exclusively through heroics and rep rewards and though I continue to run heroics, I will often Q as DPS while I do dailies. In my experience, though many tanks like to play the victims in this charade, a fair number of them are equally guilty of being ass hats, so this isn’t entirely a DPS vs the world issue.

    I bring this up because in the last 3 days I’ve seen the extremes of the LFD.

    Extreme Suckage: Group makeup, melee heavy, exceptionally geared, heaped abuse on the tank. I was Heals for the run. all of the DPS was over 12k. The tank was pulling a solid 8.5k dps. We lost at least one DPS on every single trash pull in Grim Batol. Though we had some CC, the DPS couldn’t be bothered to avoid it, and though targets were marked and communicated, the though was apparently that they had plate, they didn’t NEED a tank. All of the blame and abuse was directed at the tank for not holding aggro. I don’t care who the tank is, they aren’t holding aoe threat on a secondary target off a dps warrior targeting the off target add. It was the worst run I’ve had in months.

    Extreme Awesome: Group Makeup was a nice mix of range and melee, with plenty of CC. I was tank for this run. No DPS over 8k ( I was top DPS as tank at 8.5k for the run ). Mostly newly 330 geared toons. Vortex Pinnacle, 1 death in the entire run, and that was the DPS warrior that missed the grounding field on the last boss. Even with that low DPS, we cleared the insance in 50 minutes, because everyone played smartly despite their gear. Chatty and playful, it was a fantastic group, and easily one of the most fun runs I’ve tanked in the months.

    Extreme Meh: Group Makeup was extreme nuke (hunter, mage, enh sham), druid and me as DK tank. BRC. It was a zerg. We blitzed it in 40 minutes. We assigned target marks and not another word was said from start to finish. We got our shards and points and moved on. It was in a word; boring.

    Finding a middle ground is going to be tough with a Random Q and short of adding a questionnaire at the end of every run and for every vote kick, I just don’t see how you can get a feel for rating toons reputations, even without the potential for gaming the system. While I am not sure that bribery is the best answer, I know that it will work for keeping my tank in the Q rotation, despite not really needing anything other than points and enchanting mats from the content.
    Meli recently posted..Undivided State of the OnionMy Profile

  4. Pingback: Roundup: Better LFD Solutions Than Call To Bandaid | MMO Melting Pot

  5. Eric says:

    A recommendation engine is an interesting solution to the problem, but I don’t think it would work unless it functioned as a downvote/upvote system. The problem is that, mechanically, both bad and good players would “like” good players and “dislike” bad players, so there would only be one preference network. Netflixs’ and Amazons’ recommendation engines function because there are multiple preference networks on each. On Netflix, for example, someone who likes buddy cop movies probably also likes low brow action flicks and gross out comedies, someone who likes foreign films probably also likes indie movies and critics’ picks, and the key thing is that the two preference networks don’t have much overlap. Because there are multiple preference networks with little overlap, Netflix can narrow down the movies it matches each user with. In a LFD system, though, more or less everyone would “like” the good players and “dislike” the bad players, so we’d all have the same preferences. And for the purposes of putting a group together, it really doesn’t matter if everyone likes PurpleHealer and no one likes Arthasdk. For a Netflix-like system to be effective, there would have to be some people who like PurpleHealer and others who like Arthasdk.

    And if the system functioned as a simple upvote/downvote system, that would get us back to the fact that it could be manipulated. Also, more importantly, the LFD system’s main purpose is to allow people who would otherwise have trouble getting groups to get groups, so social consequences are contrary to the system’s design intentions.

    • Gazimoff says:

      I’ve been working a little more on the design of this one, trying to flesh it out further.

      You’re right in that if you rely on a one-way preference network, you’re scuppered. If you’re purely relying on Joe Likes fred OR Fred Likes Joe, I can agree. A form of 2-way relationship based reccomendation engine is needed based on mutual endorsement.

      Also, you assume that we all agree on what makes a good player. While it’s true that there are some universal constants, ask around and you’ll get different priorities from different people. It’s those differing priorities that create the better preference networks. Some people like a slow and careful tank, others prefer an expressway zerg. Some people prefer one style over the other. Some people prefer a lot of chatter, others complete silence. It’s not as black and white as “good” or “bad”.

      And just like with movies, some players will reach universal acclaim and some will have niche appeal. Meanwhile, Arthasdk will be the barely-rented straight to DVD.

      The key thing about having an upvote/downvote or yes/no is that in a reccomendation engine you’re only harming yourself by downvoting. If you downvote someone, you’re reducing the pool of potential good matches in the future. If you upvote someone you’re reliant on the opinion being mutual in order to group with them again. The reason being, unlike Netflix where you’d only watch a movie once, you’d be looking to group with good people more often.

      And finally, I’m not sure that it’s fair to say that LFD is supposed to be port of last resort for people who fail to get groups, or that improving it goes against the original design intentions. I think we all want groups that are quick to assemble, are enjoyable to be a part of and have a high chance of success. The current system fails at all three requirements and the band-aid only fixes the first one (and badly at that). Any revision to LFD needs to look at it as a tool for the masses, and not just the theoretical social cast-offs from gaming. To take this narrow-minded approach is shooting yourself in the foot when it comes to system design.

      • Meli says:

        I have been thinking more and more about this today and the more I think about it, the less I dislike it, however, I would alter it slightly.

        Build on Real ID. If you ‘friend’ a player in a LFD group, those players get priority in LDF assembly, and I build a network by running LFD’s. I can ignore bad players, and though I may still get them in groups, when ‘friends’ are q’d at the same time they get preferential treatment and the LFD tool assembles groups based upon friend priorities. This way, if DPS is in Q and a friend tank q’s, the system puts them together, assuming other ‘friends’ haven’t been prioritized into the group already.

        This encourages DPS to establish credentials with tanks and tanks to manage relationships with dps and healers they like running with. In the end, the bads get the same experience they do now but those that work at the networks get faster q’s with better groups all without limiting the talent pool to a single realm.
        Meli recently posted..Undivided State of the OnionMy Profile

  6. Coldbear says:

    LFD = 4chan

    To hell with that.

    I’d rather have a reputation on my server and play with people who have an incentive to play nice, because they know I’m on their server and I’m good and I’ll be there next week, too. And next year.

    If you want that community feel back on your old release-server, then skip LFD and head straight to the old whispers and friends list and trade chat.

  7. SmakenDahed says:

    I like the Karma idea and the ‘would you group with this person in this role?’ Question at the end. Player feedback can be spotty though.

    One thing Blizzard could do is track performance of the group overall. Did the group clear the heroic without wipes? Did the group take a long time? Did they wipe a lot? Tack that onto the player feedback.

    They have a lot of this information tracked anyway – even down to the number of /facepalms you’ve done.
    SmakenDahed recently posted..CuteMy Profile

  8. Phelps says:

    I think that there is one thing that people are missing about a netflix style rating system. The best system would not be to group people with the same ratings together. The best system would be to group together people who rated others the same way.

    Think about it. The guy who thinks he is gaming the system by rating everyone up and not “lowering his potential LFG pool” gets lumped in with other people gaming the system (and the hunter named Beeberfeever4lyfe who loves everything.) The troll who thinks he’s funny rating everyone down ends up… in a group with four other trolls. (Yay!)

    More importantly, the guys who like no-words zergfest roflstomps end up together, because they rate each other up and rate down those who dare to talk. And the people who like competent people who talk when needed end up together. And the people who like to chatter about American Idol or Glee all the way through the instance end up together. And eventually, all three groups end up happier.

  9. lexicorro says:

    Recommendation engine is a neat idea and maybe they could combine it with just gold for call to arms.

  10. Top Rosters says:

    I actually agree that I will not revolutionise the system by any means and as you say to reduce the queue from 60-6 mins would need 10 times the number of tanks. Having said that I think it is a step in the right direction. I posted my own opinion on Call to Arms over at my wow blog.
    Top Rosters recently posted..Holy Paladin 101My Profile