作者：Jim Cummings

2011年游戏会议的热门话题无疑是用户分析，或是运用参数量化和模式化玩家在在线游戏中的行为和决策。很多谈论话题都是围绕采用什么指标，何时采用这些指标，这些指标应在塑造新设计中扮演什么角色。

例如，在题为“How Analytics Make Players Happy”的演讲中，ngmoco谈及他们用于提高游戏玩法体验的核心衡量指标（游戏邦注：例如日均用户、用户日均收益及持续时间）。他们的观点是不同玩家存在不同游戏需求和愿望——有些希望在前进过程中受到指引，有些追求非常规玩法，而有些则希望获得地位。ngmoco策略在采用分析法的公司中颇具代表性：

* 根据玩家的不同需求和积分情况将玩家分成不同类型（“新型”、“休闲”和“强大”玩家，这些玩家有些是“付费”玩家，有些不是）。

* 辨别特定危险行为（例如：ARPU降低或回访用户减少）

* 通过针对性的设计调整消除这些现象（例如，提高只能通过硬通货币购买的游戏内容的福利；延迟新内容在随后关卡的解锁）

ngmoco将此过程形容成针对“不快玩家”的“路线调整”。但有人也将此形容成通过营收相关的行为指标衡量用户粘性。

arpu from roxology.com

Damion Schubert在其关于创造吸引“休闲”和“硬核”玩家的讲话中也谈到玩家细分和分析方法。关于玩家细分和归类，Schubert表达的理念也大同小异，称休闲&硬核属于错误的二分法，相反，他觉得设计师应基于“投入程度”区分玩家。此外，我们可以根据用户的投入程度将其进行分类（感兴趣–> 休闲–>忠诚–>投入–>硬核），通过把握他们所处的位置，你能够在特定时刻更好迎合他们的需求。

就此来看，此设计“投入程度”观点听起来和ngmoco的看法类似；虽然ngmoco及其他设计师似乎都很满足于玩家的金字塔分布归类方式，但Schubert表示，设计机制应促使玩家朝硬核一端发展——也就是说，希望实现收益最大化的设计师应通过参数、归类和定向设计功能引导更多用户转变成高投入的鲸鱼型玩家。

讨论话题随后转变成“分析方法应在多大程度上影响游戏设计？”讨论中，来自大小工作室的设计师纷纷发表他们对基于分析的设计的看法及自身作品的现行方案。其中呈现的对立局势非常明显，很多设计师发现，其工作室的分析部门常常都无法科学把握，继而合理运用参数报告——这显然会令开发者质疑他们的设计调整建议。有不少设计师发现他们的团队通过分析法调整客服支持而非优化游戏设计——工作室追踪玩家在游戏中的行为，若他们发现玩家反复进行有失合理性的操作，他们会发邮件告诉玩家游戏的优先选择方案。

此方式令许多设计师颇感叹息，他们称这是变相运用分析方法限制游戏风格的多样性——即便其真实意图是提高用户体验。我们也许会觉得这点在下述情况中表现更突出：设计师有意将玩家“幸福感”或“投入程度”同上述基本营收指标等同起来。

所以若行为参数主要用于呈现强化粘性的设计，那么它们是否本身就会减少和归类各种用户行为？是否存在其他选择方案？

关于这点，Scherlis指出如下两方面内容所存在的差异：1）通过营收指标辨别和调整用户行为；2）运用所谓的“预测分析法”。后者（游戏邦注：其通过分析同伴压力及创意行为传播等各种变量给游戏玩家的社交圈建模）能够用于创造迎合玩家未来倾向的设计，而且不像关键衡量指标那样明确同营收绑定。换而言之，通过采取各种行为措施，分析法能够让设计师顺利预测各类玩家的决策和偏好，而非简单促使玩家日益汇集至漏洞底部，创造更多ARPU。

此外，Scott Rigby在分析法及运用基于数据的设计方面有自己的看法。根据多年学术和行业研究，Rigby发现，玩家都存在相同的普遍需求，那些能够满足这些需求的设计才是最具粘性的设计。关于这点，他表示，这属于刺激性参数，而非行为方面的数据（游戏邦注：而这是分析法所关注的内容）。

就上述内容来看，游戏分析法所涉及的参数内容似乎非常混杂，从而带来许多新问题，主要涉及收集用户数据的目的及意义，以及这在呈现新设计方面所扮演的角色。虽然我们目前主要侧重营收相关的行为，以及促使玩家行为朝更优先的玩法风格靠拢，但新变量会让设计师创造既能保证用户行为多样性、投入性，且又有益于创收的功能。此外，分析法的性质及其同设计的关系依然处在发展中。（本文为游戏邦/gamerboom.com编译，拒绝任何不保留版权的转载，如需转载请联系：游戏邦）

GDC Online – Hunting Whales: Will Analytics Kill Off Diversity of Gameplay?

by Jim Cummings

A hot topic at this year’s conference has definitely been player analytics, or the use of metrics to quantify and model player behaviors and decisions in online games (social games and MMOs in particular).  Across post-mortem sessions, developer roundtables, and chats in the hallway, there’s been plenty of discussion of what indicators to use, when to use them, and the role they should play in shaping new designs.

For example, in a talk titled “How Analytics Make Players Happy” reps from ngmoco discussed their use of key performance indicators (e.g., daily average users, average revenue per user, sessions duration) for improving the gameplay experiences for their players. Their belief is that different players have different in-game wants and desires – some want guided progression through the game, others want non-routine gameplay, while others are driven to achieve status. ngmoco’s strategy – one which sounds characteristic of many companies using analytics – is to:

* cluster players into a conceptual typology (“New”, “Casual”, and “Power” gamers, any of which may or may not also be “Paid” players) based upon their differing wants and progress-points within a game

* identify type-specific red flag behaviors (e.g., decreases in ARPU or drops in % of returning users)

* resolve these behaviors through targeted, type-specific design changes (e.g., increase the benefits of exclusive content that can only be purchased through hard money; delay the unlocking of new content to later levels)

ngmoco describes this process as “course-correcting” of “unhappy players”. But one might also describe this as the use of revenue-related behavioral indicators as proxies for measuring player engagement.

A similar use of player segmentation and analytics was described by Damion Schubert (Principle Lead Systems Designer, Bioware Austin) in his talk on creating games that appeal to both “casual” and “hardcore” players.  Schubert presented a very similar conception in of player segmentation and herding, suggesting that the casual vs. hardcore dichotomy is actually a false one and that, instead, designers should segment players along a “spectrum of investment.”  Specifically, players can be clustered along a linear progression of commitment to the game (Interested –> Casual –> Committed –> Devoted –> Hardcore), and by understanding their location in this progression, you can best cater to their needs at a given pain point.

To this extent, the “spectrum of commitment” perspective on design sounded much like that described by ngmoco; however, while ngmoco and other developers seem content enough to accept a funnel demographic of player segmentation (many new players, some casuals, and then relatively few “whales” from which to bring in revenue), Schubert suggests that design mechanics might be constructed with the explicit intent of pressuring the player toward the hardcore end of the spectrum – that is, designers seeking to optimize revenue should use metrics, segmentation and targeted design features to usher as many players as possible toward high-investment, whale-like gameplay.   (Note – for more of the gritty details of such an outlook on design, stay tuned for a one-on-one interview with Schubert to be posted by our very own Jared Lorince).

Of course, the question then becomes “To what extent should analytics influence game design?”  This issue was raised at a round table discussion Tuesday afternoon moderated by industry vet Dan Scherlis (now with Sonamine).   In this session designers from studios great and small weighed in with their concerns on analytic-driven design and current practices in their own games.  A certain degree of antagonism was clear, as many designers noted that the analytics departments of their studios often lack the scientific rigor or statistical understanding to make proper sense of metric reports – which, reasonably might cause a developer to second-guess a suggestion for how their design should be altered.  Perhaps the most vehement and commonly shared concern, however, came when one designer noted that their group has employed analytics to revamp not design but customer service: the studio would track player behaviors in-game and if they noted a less than rationally optimal behavior being repeated over time (say,  running 15 minutes across zones rather than using a teleport, or relying on an auto-attack setting rather than branching out through a skill tree for special attacks), they would send emails to players notifying them of these assumedly preferred alternatives.

This approach was bemoaned by a number of designers in the session, arguing that such tactics are another example of how the use of analytics – even when genuinely meant to improve the player experience – serves to limit variability in gameplay styles.  And, we might reason, this concern should be ever more present in instances where designers are intentionally and explicitly equating player “happiness” or “investment” to bottom-line revenue indicators, as discussed above.

So, if behavior metrics are chiefly used to inform designs for enhancing engagement (for the sake of player enjoyment and/or company revenue), do they inherently seeks to reduce and cluster the variance of behaviors?  Is there an alternative?

To this end, while moderating the round table discussion, Scherlis drew a distinction between 1) identifying and course-correcting player behavior through the use of revenue indicators and 2) the use of what his company and others refer to as “predictive analytics.”  The latter – which models the social graph of a game’s players through analyses of variables such as peer pressure or diffusion of innovative behaviors – can be used to make designs that cater to future dispositions of players (whether to the majority average or on the individual-level) and are not as explicitly bound to revenue as are the typical key performance indicators.  In other words, by using different behavioral measures, analytics can allow designers to successfully anticipate a variety of player decisions and preferences, rather than simply using them for optimizing a game’s ability to shuffle players further down a funnel or spectrum of increased ARPU.

Alternatively, another perspective on analytics and the use of data-driven design comes from Scott Rigby (social psychologist and head of Immersyve).  Based on years of academic and industry research, Rigby notes that all players share common, universal needs, and that the most engaging designs (both psychologically and, in turn, economically) are those which best cater to the satisfaction of these needs.  To this extent, he suggests that it is motivational metrics, not behavioral ones, which analytics should be focusing on.   More to come on this in my next post, which will include both a review of the basic ideas underlying Rigby’s “motivation  3.0” model and my discussion with him on its role in game design and analytics.

From what I’ve pieced together here,  it sounds like the scope and particular metrics used in game analytics are reaching a level of sophistication that is bringing up many new questions about the purposes of collecting player data, its meaning, and its role in informing new and/or revised designs.  Though there seems to currently be heavy emphasis on revenue-relevant behaviors and course-correcting towards a preferred gameplay style, new variables may come to allow designers to produce revenue-benefiting features that still permit for variability in player behavior and investment.  The nature of analytics and its relationship to design is still developing.

And on that note, for a more hands-on discussion of the perspectives shaping new analytics, stay tuned for Travis Ross’ upcoming coverage featuring interviews with Nick Lim (CEO and founder of Sonamine) and Dan Scherlis, as well as Dmitri Williams (game researcher and co-founder of Ninja Metrics).（Source：motivateplay）