This thread seems odd to me. 

It began with a request from forum members to define "grind", which many did (with variations on grind = repetitive behavior in games), and has somehow morphed into a request for forum members to "think outside of the box" and "think like a developer".

---

I have a background in human computer interaction and usability and am a proponent of user centered design.  If I am asked to think like a developer I would attempt to determine the goals and need states of the users of the application I am developing.  If (a majority of) users tell me that they dislike a feature of the application I am developing (for example, a repetitive behavior that they do not see as fulfilling their needs), then I would attempt to address that criticism, not dismiss it.  I certainly would not infer that it was a problem of the user - or some bias that they held.

Elimination of *the need for* repetitive behavior to achieve (a) user goal(s) does not restrict one to sandbox type games.  Another way to eliminate the need for repetitive behavior is with a large amount of content, so that there are other options for fulfilling the user goals *of those users that do not like engaging in repetitive behavior*.  This content can be from developers or from users. 

Elimination of the need for repetitive behavior does not mean eliminating the *ability to engage* in repetitive behavior if that is the goal or subgoals of the user.  Some users may find fulfillment in repetitive behavior of various types and it would be wise to support those users and their goals.

---

As a user / gamer, I look for developers that offer me products and content that I feel fulfill my goals / needs.  That is *their* job (they get paid for it), as it is their job to discover what I desire (for example, through market research - which I also have a background in).  I do not have the spare time to design my own perfect game or to search the ends of the interwebs for a game that is not well marketed.

As a user, I feel that grind is objective (a game state requiring repetitive behavior to achieve user goals) and subjective (the point at which I personally feel that engaging in that behavior is not fulfilling / boring).  In massively multiplayer games, I do not enjoy grinds or time-sinks, but I do enjoy novelty - novel combats, the need for novel strategy or tactics, novel interactions with peers, novel exploration, novel story telling or role playing.  Banging on the same keys to see an effect on the same group of polygons for hours on end is not fulfilling to me.

---

In this case, thinking out of the box might mean more than thinking like a developer.  It might mean thinking like fellow gamers / peers and understanding that they may have other perspectives regarding mmorpgs and required repetition in games of that type.  Some enjoy "grinding" and some do not.  Some enjoy "grinding" some of the time but do not want to engage in it all of the time.  Some may have a different perspective on what "grinding" is and whether it is a bad thing or a good thing.  That does not mean that they have to think outside of "their" box.  It certainly doesnt mean that they have to move their thinking so it fits neatly in the box of someone else.

From the OP:

This concept can be easily applied to SOE's record to predict their future. Since SOE screwed up lots of their games to the point where they were unplayable, that gives us a really extreme set of data. The principle of regression toward the mean says that the next game they release won't be messed up anywhere nearly as badly. Just based on statistics, their next game will probably be screwed up somewhat, but not to the point where it's unplayable.

...

SOE has had a really bad record in the past, but The Agency will be virtually flawless; it will be the game they use to redeem themselves and develop a new, better reputation.

This is my first post on these boards.  I'm an avid gamer, and a statistician - so the subject matter of the post caught my eye.

This is a poor understanding of regression to the mean and saying that you have evidence that based on SOE's "bad record in the past" that The Agency will be "virtually flawless" based on regression to the mean  borders on the absurd.  This statement is based in no way on any statistical analysis of performance.

- - -

First, I'll give you an example that someone gave me as an undergrad.  Let's say you went to a parachuting contest.  Everyone jumps and tries to land as close as they can to a target and their distance is measured.  The someone offers a bet - pick one person that you think will get closer on their second jump.  Regression toward the mean leads you to pick the person that landed the furthest away on the first jump.  All other things being equal, and with no other knowledge, that is the safest bet.

But let's imagine the bet was who will be closest to the target on the second jump.  In that case the appropriate bet would be the person who was closest on the first jump.  You presume there are underlying factors (such as skill)  that would be predictive.

If I were going to pick a game I would think would be "virtually flawless", I'd pick a developer whose last game was virtually flawless.

- - -

Secondly, you make note of "extreme scores".  There are many times when you are doing an analysis of data (descriptive or inferential) where you would pay special attention to extreme scores in your data as these are outliers.  In many cases outliers are removed prior to a statistical analysis because they may not reflect the population of interest.  Let's say you are at the parachuting contest and on the first jump one of the contestants lands FAR away - like 100 miles away in another city.  If you looked at the variance in landing distances, this person would be several (lets say 3 or more) standard deviations away from the mean.  In such a case, I might make the assumption that this person was not drawn from the same population as other jumpers (that silly prankster!) and I would exclude them from the analysis.

Outlier analysis is performed and outliers are removed to help ensure that the data being analyzed is drawn from a homogenous population.

If I were going to build a predictive model of the state of games in the future, I would exclude extreme examples.

- - -

Thirdly, by applying the concept of regression toward the mean in this manner you are assuming a normal distribution (which may not be the case for game developers / publishers).  What if the distribution was bimodal (drawn from two populations with different means).  For example, what if the parachuting contest was pro-am with some being professional parachuting contest applicants and the others as amateurs.  Imagine that the distributions of the two do NOT overlap.  Imagine that even out of 100 tries, an amateur jumper would never get closer to the target than a professional.  This would be an extreme bimodal distribution but it would be possible if there are factors (again, skill) that are predictive of performance and the sample was drawn from two different populations.

If I were going to look at the performance of gaming companies, I would not assume performance was unimodal - there may be a group of "good" companies, and a group of "bad" companies.

- - -

Lastly, if I were going to build a predicitive model of future game performance, I would not use regression toward the mean to do so.  Regression toward the mean is a concept, but it is not a statistical model.  Linear regression and analysis of variance are statistical models (in fact analysis of variance can be seen as a special case of regression).  I'd identify factors that I thought may be predictive of future performance (such as prior performance, budgets, assets, etc.) and include them in the model.  These would include endogenous (company specific) and exogenous (outside of company - for example the state of the economy, demand for games in general, demand for certain genres, etc.) factors.  In such a model regression toward the mean is NOT a factor - it's just a concept that describes the most likely direction of movement of extreme scores *when all other factors are equal*.

To put it another way, if I was going invest money in a company I wouldn't just pick the ones the were the worst performers of the prior year.  I'd look at the fundamentals.