What makes a good game? Using reviews to inform design

Abstract

The characteristics that identify a good game are hard to define and reproduce, as demonstrated by the catalogues of both successes and failures from most games companies. We have started to address this by undertaking a grounded theoretical analysis of reviews garnered from games, both good and bad, to distil from these common features that characterize good and bad games. We have identified that a good game is cohesive, varied, has good user interaction and offers some form of social interaction. The most important factor to avoid is a bad pricing. Successfully achieving some of these good factors will also outweigh problems in other areas.

Full text

What makes a good game?
Using reviews to inform design
Matthew Bond
Advanced Interaction Group
School of Computer Science
University of Birmingham, Edgbaston
Birmingham, B15 2TT, UK
+44 (0) 121 414 3729
ug56msb@cs.bham.ac.uk
Russell Beale
Advanced Interaction Group
School of Computer Science
University of Birmingham, Edgbaston
Birmingham, B15 2TT, UK
+44 (0) 121 414 3729
R.Beale@cs.bham.ac.uk
ABSTRACT
The characteristics that identify a good game are hard to define
and reproduce, as demonstrated by the catalogues of both
successes and failures from most games companies. We have
started to address this by undertaking a grounded theoretical
analysis of reviews garnered from games, both good and bad, to
distil from these common features that characterize good and
bad games. We have identified that a good game is cohesive,
varied, has good user interaction and offers some form of social
interaction. The most important factor to avoid is a bad pricing.
Successfully achieving some of these good factors will also
outweigh problems in other areas.
Categories and Subject Descriptors
H.1.2 User/Machine Systems
General Terms
Design, Human Factors.
Keywords
Games, grounded theory, analysis
1. INTRODUCTION
At present, video game development accounts for billions of
dollars of US expenditure, and US video game companies
directly employ 24,000 people. The market for games has also
been growing disproportionately fast [10]. Clearly, not all of
these games have the same degree of success; a game needs
certain properties to make it successful and to make the most of
this growing market. Gamers often rely on review sites to
make purchasing decisions, and so a game that receives a high
average review score will have considerably better sales than a
bad game. Therefore it is desirable to know what makes a game
get a good review rather than a bad or mediocre review.
The aim of this paper is to determine what earns a game a good
review, and, by extension, what features should be prioritised
during the production of a game to make it successful.
This paper tries to determine practical categories which should
be considered during game development and, to an extent,
marketing and distribution. It is different from most of the other
approaches in that it uses Grounded Theory [5] to develop a
hypothesis from the data source, instead of using an existing
hypothesis. We determine the elements that characterize good
and bad games from an analysis of the reviews of a wide range
of games, providing a bottom-up, practical approach to
determining factors that make for a decent gaming experience.
2. RELATED WORK
Csikszentmihaly’s paper on optimal experience
(Csikszentmihaly, 1991) first used the word flow. Flow
describes a phenomenon experienced in many fields, often
described as “being in the zone”. While experiencing flow, a
person will feel absorbed, lose sense of time, and focus entirely
on the task in hand. This state can be brought on by an
intrinsically rewarding task with clear goals, immediate
feedback, balanced difficulty and a sense of control over the
situation.
GameFlow, Sweetser & Wyeth’s [11] adaptation of
Csikszentmihaly’s model of flow [1] for game experience,
groups game-design criteria into each of these categories, and
shows that gaming is an example of flow experience. Their
approach makes for an interesting comparison, since it takes a
top-down approach, but emerges with similar conclusions to
ourselves.
Our own work developed in the opposite direction to
GameFlow, with no prior knowledge of flow theory. Both
GameFlow and our categories share similar ‘leaf node-criteria’;
it is the higher level categories which are significantly different.
Also, some of the specificity of our leaf nodes is not captured
by GameFlow’s criteria.
As such, we see our categories as a more practical abstract level
approach to game design than GameFlow, due to the more
domain specific terminology used. However the same low level
details were revealed by both studies.
Both GameFlow and our own work identified social interaction
to be a key part of gaming experience that was unaccounted for
by traditional flow theory. Whether this is something that
should be revised into flow or whether it is entirely domain
specific is still an outstanding research question. The
GameFlow model never brings into question the price of a
game, something we found to be of critical importance,
although it is debatable whether price contributes to the design
process of game development.
Kristian Kiili’s work on Game-based learning [3] applied the
idea of flow to educational game design. It determined
gameplay to be the most important part of game design, with
storytelling, graphics, sound and balance to be auxiliary factors.
© The Author 2009.
Published by the British Computer Society
418
HCI 2009 – People and Computers XXIII – Celebrating people and technology

Federoff [3] also outlines gameplay as the most important
factor, followed by ‘game mechanics’. Interestingly, our
findings contradict both of these studies in this respect.
Back in the early 1980’s, Malone [7] determined three criteria
for the creation of an enjoyable game and discussed applying
his findings to the creation of educational games. He found
challenge, fantasy and curiosity to be the most important
features in good game design. His findings are less relevant
today, overlooking the importance of variety and a social
aspect, and the importance and meaning of ‘fantasy’ is
questionable.
In Simulation and Gaming, Myers’ identifies challenge, social
and meditation as the core elements of a game [9]. These
findings, especially the importance of a social aspect, tie in well
with our own findings.
3. METHODOLOGY
Grounded theory is an empirical research strategy which works
from some raw data to form hypotheses about some situation
[2]. In this case, the exploration was to discover what makes a
game ‘good’ or ‘bad’.
Some introductory work was done to find an acceptable data
source. The criteria for the source were that it had to be up to
date and have numerical scores, as well as having some form of
text for coding. It is assumed here that a successful game is one
that will sell a lot of copies: however, sales data for games is
not published for many games, and has no body of text from
which to find a hypothesis, so we need to find a more detailed
source than just sales figures.
By comparing sales of Playstation 2 games as published by VG
Chartz [12] with their respective review scores from Metacritic
[8] there was a Pearson correlation of 0.3289 (i.e. reasonable
strength, but not strong), suggesting that a high review score is
reasonably indicative of strong sales. Metacritic is a review
aggregator, and only provides a numerical score. Gamespot
provide both a numerical score, and a body of text related to
each score. A strong correlation between Gamespot scores and
Metacritic led to the decision to use online game reviews from
Gamespot UK [4] as the data source, since it was the text and
phrases in the reviews that we needed to analyse.
Samples of particularly good and bad reviews from Gamespot
UK were analysed and coded to the near point of saturation.
Note that we are certain these categories are incomplete,
because full saturation was not reached: saturation in Grounded
Theory occurs when the data no longer adds to the current
hypothesis.
The coded criteria for success and failure were then organised
into a hierarchical structure, based on emergent relationships
between those categories. This gave a set of core categories
summarising success and failure. These were then strengthened
by coding a second smaller sample of reviews into the
hierarchy. In total 25 reviews were used for the original data
coding.
Having established these core categories, the importance of the
categories was determined. This was done by comparing the
latest game releases at the time to the core categories, and
recording whether the review mentioned a category.
4. RESULTS
4.1 Categories
For game success, 13 core categories were found, and for game
failure, 12 categories were found. These give a good intuitive
overview of what to aim for in game design.
Good Factors Bad Factors
Gameplay Poor gameplay
Environment Poor environment
Storytelling Poor storytelling
User interaction Poor user interaction
Customisation
Social Lack of social
Variety Lack of variety
Technical Technical issues
Cohesion Lack of cohesion
Maintenance
Price(value for money) Price (worthless)
Franchise Failure of Franchise
Quantity (lots) Quantity (little)
Annoyance
Table 1: Good and bad factors in game design
Table 1 shows the overview of what to aim for, and what to
avoid, in game design. However it offers no idea of the relative
importance of each category.
Each category has a sub tree of criteria, of which the category
heading is a summary. A sample of these criteria for ‘good’ is
given in Table 2. The full versions of these trees are very large;
however the table below gives a good feel for the criteria found
within each category.
4.1.1 Interpretation of good and bad factors
With the exception of quantity and franchise, a “good or bad x”
means that some or all of the criteria of x are covered in a good
or bad way, where x is a category.
The good and the bad factors, although apparently ‘opposites’,
have different importance data associated with them. Fulfilling
the criteria for a good category may outweigh (or be
outweighed) by fulfilling the criteria for the bad ‘opposite’
category. See the section on importance for more details.
Quantity and franchise are both unusual categories, in that the
‘good-bad’ divide is not as clear. Quantity refers to the amount
of a category that a game has, the more of a good thing, the
better it is, likewise the more of a bad thing, the worse it is.
Franchise being good is defined by using the franchise well,
staying true to it and improving. The bad side is betraying the
franchise, adding bad elements and changing things which
underpinned the original’s success.
The ‘opposites’ of customisation and maintenance do exist, and
were encountered during the importance ranking stage, but
were not coded during the original coding stage. They were the
inability to customise, and difficulty to maintain.
419
M. Bond et al.
HCI 2009 – People and Computers XXIII – Celebrating people and technology

CategorySampleCriteria
GoodGameplayEngaging,fair,balanced,
progressive,fun,innovative,easy
toplay,hardtomaster,objective
based,freedom,compelling,
dynamic,variouspossiblesolutions
GoodEnvironmentImpressiveness,eyecatching,good
lighting,lifelikeeffects,good
soundtrack,goodsoundeffects,
goodmusic
GoodStorytellingMature,progressive,tense,
engrossing,embeddedingameplay
GoodUserInteractionFastfeedback,customisable,
invisiblecontrols,realistic,
functional
CustomisablePowerfuleasypersonalisation,
charactermodification
GoodSocialInteractionMultiplayerco‐op,multiplayer
competition,communication,
sharing
VarietyNonlinearity,choice,differences,
dynamiccombat,variedAI,
emergenttactics,varieddelivery
media
GoodTechnical
Implementation
Welldesignedcamera,unobtrusive
adverts,smoothframerate,
uniformity,freedomtobehaveas
expected
CohesiveSeamlessintegration,storyrelated
togameplay,cohesivestory,
consistentstyle
EasilyMaintainedLowhardwarerequirements,easy
tomaintain,independentof
externalsoftware
Price(valueformoney)Valueformoney,cost,addoncost,
hardwarecost
PriorFranchiseFranchise,establishedgenre
QuantityLotsofgood,notmuchbad
Table 2: Sample tree content for the 'Good game' tree
Annoyance never developed an ‘opposite’ despite being
important to avoid. It may be that ‘annoyance’ is a cumulative
effect of other bad factors, especially high quantities of bad
criteria.
4.1.2 Relative Importance
This set of categories assumes all categories are equal, but
during coding certain categories were clearly more important
than others, occurring more frequently, for example. To
investigate this, for 33 different, equally varied reviews, it was
recorded whether a category was mentioned during a review,
and a first table of importance was developed shown in Table 3.
Quantity is not included here because it is quantity in a way
that is being counted to rank importance. Franchise is also not
included because despite being mentioned in nearly every
review, it was usually only in passing, and with no proper
knowledge of the franchise we could not establish whether it
was a positive or negative effect.
Gameplay and environment came out on top, being mentioned
most in the reviews, for both good and bad games. These results
did not quite reflect our expectations, as they put more
importance on these two items in relation to other categories
than we had expected. We therefore looked for a modified
measure.
4.1.3 Improved measure
We suspected the complexity of the categories was causing this
effect, with some categories encompassing far more criteria
than others, making them far more likely to be mentioned than
those with relatively few criteria. In a rough attempt to
overcome this, the count was divided by the number of criteria
for each category.
This produced Table 4, which was much closer to our
expectations from the coding stage. We do question whether
this is a valid approach. It addresses the fact that, for some
categories, there are many synonyms and alternative ways to
express the same higher level concept, and so some reviews
will use more words to explain the same thing. However, it
may also be that a review addresses subtleties and hence
multiple mentions of words in one category are commenting on
slightly different things, and so deserve being counted twice.
The point of grounded theory is not to be quantitative, however,
so we need to treat these measures as indicative, rather than
precise.
Variety
Cohesion
Social
Customisation
Technical
Storytelling
Environment
User interaction
Maintenance
Gameplay
Price
Annoyance
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Negative
16 16 12 4 14 7 8 2 10 12 11 7 24 18 15 14 1 1 25 18 5 5 4
Table 3: Categories discovered through grounded theory analysis
420
What makes a good game? Using reviews to inform design
HCI 2009 – People and Computers XXIII – Celebrating people and technology

Variety
Cohesion
Social
Customisation
Technical
Storytelling
Environment
User interaction
Maintenance
Gameplay
Price
Annoyance
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Positive
Negative
Negative
2 1 2 1 2 1 1 0 1 1 1 0 1 0 2 1 0 0 1 1 1 2 1
Table 4: Normalised categories
4.1.4 Interpretation
Table 5 summarises the our findings into 3 categories, where
categories that received a ‘2’ are of most importance, a ‘1’ is of
high significance and a ‘0’ is relatively unimportant.
Most
importance
Moderate
importance
Relatively
Unimportant
Feature Variety,
Cohesion,
Social, User
Interaction
Customisability,
technical soundness,
good storytelling,
good environment,
good gameplay,
value for money
Easy
maintenance
Avoid Bad pricing Lack of variety, lack
of cohesion, lack of
social, lack of
customisation,
technical issues, bad
user interaction, bad
gameplay,
annoyance
Lack of
customisation,
poor
storytelling,
poor
environment,
difficult
maintenance
Table 5: Simplified summary of key elements in game
design
If a game features an element of most importance, it will
overwhelm the effect of moderate and unimportant categories.
An example of this in practice is ‘F.E.A.R. 2: Project Origin’,
which has storytelling and environment problems, but these are
overwhelmed by the excellent variety and cohesion
demonstrated, and so receives a decent overall score.
The four most important factors are variety, cohesion, a good
social aspect and good user interaction. The most important
thing to avoid is an unreasonable asking price.
One of the more interesting findings is that a bad environment
and bad storytelling do not have a very significant impact on
game success.
5. ANALYSIS
The intention of these results is to provide thought-provoking
heuristics for game development. They are in no way facts, and
should not be read in that way – they are summaries of human
opinion, with the inherent variability that that implies.
These results could be useful to game designers, to ensure they
are including the most important criteria in their game design
and by implications producing a product that scores highly in
reviews; to reviewers, to check they are covering the key
criteria well enough in their reviews; and finally from a buyers
perspective, both in assessing the expressiveness of a review,
and then in assessing whether a game itself is worth buying.
These heuristics are very general. A wide range of games, with
different genres, on different platforms, at different prices and
from different eras was coded. The core categories seemed to
apply across all of these ranges, though we feel that the
importance values for each category probably vary.
One of the most unexpected findings was that gameplay was
not featured as one of the most important categories to fulfil.
Other work has found gameplay to be the most important factor
[3, 6]. We will focus on the discussion of gameplay, however
these points also apply to the other categories.
There are three interpretations of the discrepancy over
gameplay importance. Gameplay may not be as important as
previously thought; we have developed a different set of
criteria; or there may be a mistake in our findings. It is likely
that all three interpretations make a contribution.
Our criteria for gameplay are certainly different to those found
in common literature. Early in the Grounded Theory coding
process, cohesion and variety were criteria for each of the other
categories. They were mentioned so frequently, in relation to so
many things, it was quickly apparent they formed categories of
their own. We removed them from all the other categories,
including gameplay. This lowers the chance gameplay will be
mentioned in a review because that element may well be
captured in a variety or cohesion criterion. A natural effect of
Grounded Theory is that the results ‘make sense’. This is good,
in that it does make sense, and is therefore easy to accept as
right. However, this may also make it harder to dispute where it
is wrong. It can also be seen as ‘stating the obvious’ to those
with prior knowledge in the field.
The biggest fault with our application of Grounded Theory is
that the categories did not reach a high enough level of
saturation. Normally, when a core category becomes apparent,
coding for other categories stops, and the core category is then
saturated, coming back to saturate other categories later.
The aim of our work was to identify the broad categories, and
not focus on one topic. Having identified 13 core categories,
there was simply not enough time to saturate each of them. This
means that the categories may be incomplete, and more
importantly, that there may be categories missing.
Because the GameSpot resource that we used has many authors,
this lessens the problem of a individual source being used for
this study. However this study could easily be extended to
include other review sites, and this may potentially add some
421
M. Bond et al.
HCI 2009 – People and Computers XXIII – Celebrating people and technology

new criteria, strengthen the categories found, and better balance
the importance findings.
Although the current importance ranking works as a rough
estimate of importance, it was massively oversimplified, and
does not use a large enough data set. The current scheme only
marks whether a review does or does not include a factor. This
does not seem to account for games in which a certain category
has many or all criteria fulfilled, instead providing a rough
estimate..
Importance also varies from criteria to criteria. If the
calculation was done on a criteria level, and propagated up to
the higher level categories, the findings would be more far
more useful in game design, as for the most important criteria,
it could be asked “Have we included/prevented x criteria”,
whilst also providing a high level overview.
At present the calculation does not take into account review
scores. With the correct algorithm to calculate roughly how
much each category contributed to the final review score, the
general importance of each category could be better generated.
One possible implication of an improved calculation of the
relative importance criteria is that new reviews could be coded
based on whether the categories are fulfilled, and a review
score generated automatically for the review. It would be very
interesting to see whether this is possible or reliable, and even
more interesting to compare to the subjective scoring system of
the GameFlow scoring system.
At present the data set for importance was 33 different reviews,
however this should be far higher to improve the reliability of
the importance algorithm.
It is worth noting that improving the importance rating would
be futile without coding all the categories to saturation first.
6. CONCLUSIONS
The application of grounded theory to game reviews seems a
worthy one, and this research provided some similar findings to
prior work. It also identified some new criteria, and provided an
intuitive set of high level categories for analysing games. It
suggests that the most important elements of good game design
are cohesion, variety, good user interaction and some form of
good social interaction. The most important factor to avoid is
bad pricing.
These findings provide a cursory set of heuristics for use while
developing, reviewing, or buying a game. These findings
should be of particular interest to new development teams that
do not have the backing of a franchise, especially in the early
stages of game design.
This paper is primarily intended to inspire further work in the
field. Suggested work includes fully saturating the criteria for
each category, and using some form of learning algorithm to
determine the significance of each criteria in relation to the
review score given.
If importance data could be reversed and applied to new
reviews, it may provide a fairly objective method for generating
review scores.
If this were successful, another possible challenge would be for
existing games developers to try and generate a successful
game that disobeys the importance rankings found, in an
attempt to find innovative new techniques for making enjoyable
games.
7. REFERENCES
[1] Csikszentmihaly, M. Flow: The Psychology of
Optimal Experience. Harper Collins, New York,
1991.
[2] Dick, B. Grounded Theory: A Thumbnail Sketch.
http://www.scu.edu.au/schools/gcm/ar/arp/grounded.
html, March 1, 2009
[3] Federoff, M.A. Heuristics and usability guidelines for
the creation and evaluation of fun in video games. .
Bloomington, 2005.
[4] Gamespot. Video Game Reviews.
http://uk.gamespot.com/reviews.html, February 27,
2009
[5] Glaser, B.G. and Strauss, A.L. The discovery of
grounded theory. Aldine, Chicago, 1967.
[6] Kiili, K. Digital game-based learning: towards an
experiential gaming model. 8 (1).
[7] Malone, T.W., What makes things fun to learn?
Heuristics for designing instructional computer
games. . in, (Palo Alto, California, 1980), Association
for Computing Machinery.
[8] Metacritic. Game Reviews from Metacritic.
http://www.metacritic.com/games/, March 1, 2009
[9] Myers, D. A Q-Study of game player aesthetics.
Simulation and Gaming. 375-396.
[10] Siwek, S.E. Video Games in the 21st Century:
Economic Contributions of the US Entertainment
Software Industry. . Entertainment Software
Association, Washington DC, 2007.
[11] Sweetser, P. and Wyeth, P. GameFlow: A Model for
Evaluating Player Enjoyment in Games. 3 (3).
[12] VGChartz. VG Chartz.com. (VG Chartz).
http://uk.gamespot.com/reviews.html, January 15,
2009
422
What makes a good game? Using reviews to inform design
HCI 2009 – People and Computers XXIII – Celebrating people and technology