Why do people play games? A meta-analysis

TL;DR: It can be posited that games are multi-purpose ISs which nevertheless rely on hedonic factors, even in the pursuit of instrumental outcomes, as well as the ways in which they are used.

About: This article is published in International Journal of Information Management.The article was published on 2017-06-01 and is currently open access. It has received 207 citations till now.

Summary

INTRODUCTION

• During the last decade games have become an established vein of entertainment, consumer culture, and essentially a common part of people's daily lives (Mäyrä & Ermi 2014; Yi 2004 ).
• Developments in information technology have pushed games into a variety of areas of human life, and have diversified in terms of their different uses, motivations, and users.
• Mathematical meta-analysis provides a highly accurate means of calculating the reasons and motivations why people use games across differing theoretical approaches and contexts of study.
• Analyzing such a wide body of literature allows us to see the phenomenon from a wider perspective, and therefore the study is not limited by the narrow scope that singular studies are obliged to adopt.

2. METHODS & PROCEDURE

• The procedure began by gathering empirical studies related to game usage from academic literature repositories.
• Systematic review processes result in straightforward search, with a low ambiguity in the inclusion of studies and the details needed for a replicable review process, as well as a transparent methodology (Boell & Cecez-Kecmanovic 2014; Oates 2015) .
• After the elimination of unsuitable search hits, selected details were coded from the remaining set of valid studies.
• This step included the identification and combination of variables with similar definitions.
• The frequencies of the details are reported in the next section to provide an overview of the study of game usage.

2.1. Sources of data

• Following Webster and Watson (2002) and Ellis (2010) , the analysis procedure commenced with a literature search.
• The search procedure was undertaken in the Scopus database (November 2015) which is the largest abstract and citation database of scholarly literature (Elsevier B.V 2014) .
• Scopus is also the most relevant repository for studies within the disciplines where literature on why people adopt and use different technologies is being published.
• The literature search in Scopus was conducted using a search string which consisted of three main parts: 1) use-related keywords, 2) game-related keywords and 3) subject area specifications.
• The authors decided to exclude natural and medical sciences as these areas led to hundreds of false positives.

2.2. Inclusion criteria

• Eight inclusion criteria were used to assess the mass of research (985 entries) found by the literature search.
• Fourthly, the entries were inspected as to whether their topic or research question concerned the use of games.
• Inevitably, this was the largest omission category with 616 rejected search hits.
• 20 study articles were omitted from their analysis for reasons of insufficiently reported results, ambiguous variable definitions, and unclear study methodology.
• After the inclusion process, a total of 48 studies were eligible for inclusion in the analyses.

2.3. Coding procedure

• The analysis of the selected studies followed a two-stage process, and was conducted in accordance with the guidelines of Webster & Watson (2002) .
• The first step of the literature review framework is an author-centric analysis in which studies are listed in a table and selected details from the papers are entered in columns.
• For this review, the details included 1) the reference, 2) the context of the study (the game type and the actual game if disclosed in the study) 3) data collection and analysis methods, 4) sample size, 5) the variable correlation matrix, RUNNING TITLE: A meta-analysis and 6) the underlying theoretical framework of the study.
• These tables are reported in the results section.

2.3.1. Coding decisions

• The theoretical framework of the study refers to the basis of the research model used.
• It was decided to code the theoretical frameworks with a single value.
• In other words, studies which combine different theories are treated as different entries in the coding.
• Not all games are solely entertainment products and games are increasingly being utilized to motivate the use of information systems which are designed for mainly utilitarian purposes (Hamari & Koivisto 2015) .
• Since these two game types are relatively different, the authors identified and categorized games into separate categories of hedonic and utilitarian games.

2.3.2. Combining variables

• Since their meta-analysis combines variables across the studies, the authors had to pay extra attention when they coded the variable names.
• Inevitably, some studies used different terminology for RUNNING TITLE: A meta-analysis similar variables, as well as using the same names to describe relatively different concepts.
• Regardless of this, the authors combined and separated the variables accordingly, using a reasonable level of abstraction.
• In the end, by carefully combining similar concepts with varying details, the authors were able to extend their results across a larger scale.

2.4. Meta-analytic approach

• Reviewing published research can be divided into two overall approaches: 1) traditional qualitative method (also known as the narrative method) in which the conclusions of reviewed studies are practically summarized using words, and 2) meta-analysis which is a mathematical and quantitative approach, and where the effect sizes of the reviewed studies are combined using calculations (Ellis 2010 ).
• The narrative approach has been found to be insufficient when synthesizing findings from contradictory results, especially for a large number of studies (Hunter & Schmidt 2004 ), whereas the meta-analytic approach provides more comprehensive results with estimates for effect size, different metrics for reliability, and information about different kinds of bias.
• Moreover, unlike the narrative approach, meta-analysis does not suffer from increased complexity in interpreting large amounts of studies.
• Instead, meta-analysis addresses the discrete limitations of individual studies and settles conflicting findings (Paré et al. 2015) .

2.4.1. Meta-analysis calculation model

• More specifically, meta-analysis is a mathematical and statistical method for combining the results of previous studies that address a similar research problem (or the data/results which can be used to address a similar research problem) (Glass 1981) .
• In the approach of Hedges et al., raw correlations are z-transformed before combining the effects, and weights of n -3 are used instead of the original sample size (n) for each study.
• A meta-analysis Both meta-analytic calculation approaches include at least two different models, namely to account for fixed and random effects.
• Thus, using a fixed effect generally produces less variance as well as tighter confidence intervals.

2.4.2. Test of heterogeneity

• The authors verified their model approach using tests for heterogeneity.
• The heterogeneity of their data was tested with Q-statistics and I 2 -values for every relationship that was analyzed in meta-analysis .
• The Q-statistic (Cochran 1954) is the classical measure for heterogeneity while the I 2 -value represents the percent of the variance explained by the heterogeneity of the data, and the minimum of 0 % indicates that all variability is instead due to sampling error within trials (Higgins & Thompson 2002) .
• All Q-estimates were statistically significant at p < 0.01 and each RUNNING TITLE:.
• Thus, the random effect model is seen as a proper approach for conducting this particular meta-analysis.

2.4.4. Moderator analysis tests

• The purpose of their moderator analysis is to examine the difference in meta-analysis results between two different types of games.
• Similar to actual meta-analysis, the test also requires some decisions regarding the calculation model to be used.
• First, one must choose between a fixed or random effect model, depending on how the within group estimates are to be calculated.
• Similar to the main meta-analysis, the authors had no reason to believe that even studies within the same game categories would have such identical research conditions, that a fixed effect could be assumed.
• On the other hand, the authors had no reason to assume different variances for these groups, so the same within studies estimate for variance is used for both subgroups.

2.5. Meta-analytic structural equation modeling

• In addition to their correlation based meta-analysis, the authors used structural equation modeling to further investigate the effectiveness of a technology acceptance model in explaining the use RUNNING TITLE:.
• While structural equation models are often used with raw questionnaire data, using a meta-analytically pooled correlation matrix as data is relatively uncommon, and thus the authors were required to address some differences between these methods.
• Researchers have previously used medians, totals, arithmetic and harmonic means (Cheung 2015) , as well as minimums (Schepers & Wetzels 2007) of meta-analysis sample sizes.
• Next, using a correlation matrix as data input in structural equation modeling requires estimates of the standard deviation of the variables.
• While it could have been possible to estimate the standard deviation for each variable (e.g. by weighted averaging deviation drawn from each study), it would have reduced their sample size significantly.

3.1. Descriptive information of the analyzed literature

• The search process resulted in 48 relevant research articles for inclusion in the analysis (listed in Table 1 in ascending alphabetical order of the reference).
• The articles had been published between the years 2004 and 2015.
• The theoretical basis of the reviewed studies was also examined.
• Rather a high amount of studies ( 12) used no specific theoretical framework explicitly, and instead the variables were adopted from various theories, different studies, or were created by the authors.
• When combining this with less frequently used combinations of the model, the authors get a total of 23 research models that were based on TAM.

3.2. Variables

• Since the data contains 758 unique correlation pairs, the scope of analysis must be limited.
• As a statistical method, meta-analysis prefers multiple findings for establishing relationships and therefore it is reasonable to only include the most frequently studied correlations in the analysis.
• Thus, the authors focused on variables that were included in at least three independent studies.
• User perception of how enjoyable, entertaining and fun playing games or specific game is (Davis et al. 1992; Van der Heijden 2004) .
• Perceived Ease of Use (PEOU) 22 "The degree to which an individual believes that using a particular system would be free of physical and mental effort" (Davis 1989) .

Attitude (ATT)

• Therefore, attitude towards using games includes opinion on whether playing games is good idea and how much people like playing games.
• Perceived Usefulness (PU) 18 "The degree to which an individual believes that using a particular system would enhance his or her job performance" (Davis 1989 ).
• Perceived usefulness was defined loosely as any sense of usefulness in playing games.

Subjective Norms (SN)

• 14 Perceived social pressure from other people on how acceptable the activity (use of games) is (Ajzen & Fishbein 1980) .
• Also often referred as "social norms" or "social influence".

Flow 8

• Flow is a mental state where a person is fully immersed, deeply concentrated and truly enjoys while performing a certain activity (Csikszentmihalyi 1990) .
• Flow is the optimal experience in use of games and may lead to ignoring real world surroundings while playing.

Perceived Playfulness (PP) 6

• Perception on how focused one is, degree of curiosity and how enjoyable or interesting it is when interacting with games (Moon & Kim 2001; Webster & Martocchio 1992).
• Playfulness is conceptually rather similar to enjoyment and flow.
• A meta-analysis Satisfaction (SAT) 6 Satisfaction is the positive feeling that arises from how well the actual experience meets expected experience (Bhattacherjee 2001; Hernon & Whitman 2001) .

3.3.1. Main findings

• The meta-analysis results for the most frequently studied variables are available in Table 4 and in RUNNING TITLE:.
• As can be seen from both Figure 3 and the result table, the correlation between playing intention and gender is the only non-significant estimate occurring in the meta-analysis.
• Moreover, this estimate has wide confidence intervals and thus the results do not show any signs of a connection between these variables.
• The results show an even wider confidence interval for the correlation between satisfaction and intention, although the estimate is significant and clearly positive.
• A meta-analysis Nevertheless, all other confidence intervals are rather narrow and most failsafe N measures indicate low risk of publication bias (Sabherwal et al. 2006) .

3.3.2. Moderating effect of game type

• The authors also tested differences between hedonic and utilitarian type of games by dividing the sample into two subgroups and comparing the correlations.
• For the sake of simplicity and interpretability, the authors limited the moderator analysis for variables with direct correlations with playing intention.
• According to the moderator analysis results , the game types differed most significantly (Q = 55.988***) in the correlation between Perceived Usefulness and Playing Intention, where hedonic games had a significantly lower estimate (0.392***) than utilitarian games (0.724***).
• A meta-analysis Every estimate in the moderator analysis was rather strong and above the lower bound of the medium correlation class.
• Thus, the authors determined that they could reliably interpret the comparison results.

3.4.1. The overall model

• In meta-SEM (or any regression-based analysis), as the model structure is more complex and more variables are included simultaneously in the model, the set of studies that can be included diminishes from the correlation-based analyses since all of the included studies have to feature all of the variables in the model.
• Finally, perceived ease of use similarly has rather high coefficients for the paths between it and perceived usefulness (0.493***) and enjoyment (0.437***).
• At the same time however, the results also lend support for their hypothesis (discussed from the outset of the study) that it does not seem meaningful to place the whole spectrum of games under one category.
• Instead, it seems to produce uninformative averages from significantly varying data.
• Therefore, in order to investigate the theoretical fit, the role of usefulness and enjoyment, and how they may differ between utilitarian and hedonic games, a meta-SEM model fit investigation is needed.

3.4.2. Modeling use intention of utilitarian games

• Utilitarian games (gamification, serious games etc.) are an intriguing combination of both utilitarian and hedonic systems, where the goals of the systems' use are related to productivity, although the means and the design by which the systems promote productivity are hedonic in nature.
• Utilitarian games can hence be characterized as "productivity through fun".
• Pertaining to the relationships between variables, the results show significant and strong path RUNNING TITLE: Moreover, perceived usefulness has extremely large effect on attitude (0.810***), whereas the path coefficients from both enjoyment and perceived ease of use on attitude remain insignificant (0.039 ns and 0.023 ns , respectively).

3.4.3. Modeling use intention of hedonic games

• Having confirmed a fitting model for utilitarian games in the previous sub-section, the question still remains as to what might be a fitting model for hedonic games.
• Therefore, there is reason to believe that the relationship and role of enjoyment and usefulness might also be oppositely true.
• Similar to utilitarian games, the results for hedonic games show a strong relationship between attitude and playing intention (0.479***).

4. DISCUSSION

• The present study contributed to and advanced the theoretical and empirical understanding of the nature of games, and also their role in the vast space of information systems and the related academic domain.
• More specifically, this study has synthesized previous research literature meta-analytically and provided estimates for the correlations between the most common variables featured in studies that investigate why people use games.
• Their results show that games are also dualpurpose ISs in the sense that their use is driven by both hedonic and utilitarian reasons.
• In particular, as the zero-order correlation between the variables was noticeably stronger than the path coefficient in their structural model, the results suggest that enjoyment and usefulness dominate the effects on attitude over the perceived ease of use.
• A meta-analysis but at the same time rigorously, researchers seeking to make their dent in this vein of literature should boldly venture towards more courageous research initiatives and towards including a more varied set of independent variables.

4.1. Limitations

• As far as the authors know, their study is the first of its kind to conduct comparative (meta-)analyses on the use and acceptance of hedonic and utilitarian games.
• The specific game types remain rather abstract.the authors.
• In order to develop a more detailed and fine grained division of game types, the authors would have needed more specific information about the games specifically examined in the literature.
• Undoubtedly not all of the studies measured similarly named variables identically, and so conceptually, merging variables across a variety of studies could be seen as rather questionable.

Figures

Table 1: Included studies in the analysis.

Figure 1: Analysis procedure workflow.

Figure 4: Meta-analysis for difference between hedonic and utilitarian games. INT playing intention, ATT attitude toward playing games, ENJ enjoyment, PU perceived usefulness, PEOU perceived ease of use, SN subjective norms. S small, M medium, L large. H hedonic games, U utilitarian games.

Table 4: Meta-analysis of correlations between most frequently examined variables.

Figure 2: Inclusion criteria for research articles. Vertical arrows represent omitted research articles at each step while horizontal arrows show the remaining articles.

Figure 6: Results for structural equation modeling for hedonic and utilitarian games. *** p<0.001,

Figure 3: Meta-analysis of correlations between most frequently examined variables and playing intention. INT playing intention, ATT attitude toward playing games, ENJ enjoyment, PU perceived usefulness, SAT satisfaction, PEOU perceived ease of use, SN subjective norms, CM critical mass, GEN gender. S small, M medium, L large.

Table 6: Model fit indices.

Figure 8: Modeling results for playing intention of hedonic games. Model fit = AGFI: 0.980, TLI: 0.980, CFI: 0.996, SRMR: 0.014, RMSEA: 0.056.

Table 2: Frequency of theoretical frameworks in the analyzed literature.

Table 3: Most frequently investigated variables in the analyzed literature.

Figure 5: Results for structural equation modeling with total sample. *** p<0.001, ** p<0.01, * p<0.05, ns p>0.05. R2 total variance explained.

Table 5: Meta-analysis for difference in effects between hedonic and utilitarian games.

Figure 7: Modeling results for playing intention of utilitarian games. Model fit = AGFI: 0.916, TLI: 0.962, CFI: 0.992, SRMR: 0.022, RMSEA: 0.116.

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "Why do people play games? a meta-analysis" ?

To address this gap, the authors conducted a meta-analysis of the quantitative body of literature that has examined the reasons for using games ( 48 studies ). The present study contributes to and advances their theoretical and empirical understanding of multi-purpose ISs and the ways in which they are used. 

Q2. What are the main factors that are important in the use of games?

Since most contemporary games include social activities such as multiplayer game modes, chatting, trading and/or tools for sharing the gaming experience, social factors are arguably an important consideration in the use of games. 

Q3. What is the effect of perceived usefulness on attitude?

perceived usefulness has extremely large effect on attitude (0.810***), whereas the path coefficients from both enjoyment and perceived ease of use on attitude remain insignificant (0.039ns and 0.023ns, respectively). 

Q4. What are the other hedonic variables to explain the use of games?

In addition to enjoyment, the literature has suggested two other hedonic variables to explain the use of games; flow and perceived playfulness. 

Q5. What is the relationship between flow and playfulness?

Flow is the extent of deep concentration, the amount of immersion and the optimal hedonic experience (Csikszentmihalyi 1990) in the use of games, whereas perceived playfulness is related to interestingness and curiosity (Moon & Kim 2001). 

Q6. What were the three classes for interpreting effect sizes?

Correlation effect sizes were interpreted using Cohen’s (1988) small, medium and large thresholds, and therefore the three classes for interpreting effect sizes were:• Small (S) for values between 0.10 - 0.30• Medium (M) for values between 0.30 - 0.50• Large (L) for values between 0.50 - 1.00 

Q7. What is the relationship between enjoyment and usefulness?

In particular, as the zero-order correlation between the variables was noticeably stronger than the path coefficient in their structural model, the results suggest that enjoyment and usefulness dominate the effects on attitude over the perceived ease of use.