To investigate students’ perceptions of play, we applied Q-methodology [41], which is a mixed methods research technique that aims to account for all key subjective viewpoints on a certain matter [42]. The qualitative component of Q-methodology allows the expression of subjective opinions to be considered, and the quantitative component uses statistical analysis in order to group participants with shared opinions. Q-methodology is used to cluster individuals based on shared opinions rather than based on latent variables, which is the case in regular factor analysis (or R-methodology). The Q-methodology technique has been used before in medical education, although for different purposes (eg, for identifying different patterns of self-regulating learning behavior [43-45]). Q-methodology is characterized by two main stages: (1) designing a set of statements and let participants sort that set of statements, and (2) by-person correlation and factor analysis of a sample of the included participants.

We obtained ethical approval for this study from the Netherlands Association for Medical Education (NVMO 2019.1.11).

This study was conducted in May 2019 at the Faculty of Medical Sciences of the University of Groningen, the Netherlands. The 6-year undergraduate medical and the 6-year undergraduate dental curriculum both consist of a 3-year Bachelor and a 3-year Master’s phase. In both curricula, teachers used face-to-face and web-based teaching methods and sometimes apply game-based learning, but not on a regular or structural basis.

We invited all medical and dental students (3000 eligible undergraduate students) to participate in our study by posting an announcement on the web-based learning environment (called Nestor) of the university. Participants were informed about the purpose and procedures of the study. Their participation was anonymous, voluntary, and confidential. Participants had the right to withdraw from the study at any time. All participants provided informed consent. In appreciation for their time and effort, each participant received a 5€ (approximately US $5.66) gift certificate.

Although there is no decisive minimum or maximum number of participants for performing Q-methodology research [43], generally, the number of statements should exceed the number of participants [46], and 40 to 60 participants is considered adequate [47,48]. To achieve a highly diverse sample—which is recommended for Q-sort analysis [46,48]—we purposively selected participants. Therefore, we asked participants to complete a sociodemographic questionnaire (eg, age, gender, and whether or not they considered themselves a gamer) prior to the sorting process that also included a question about the participant’s favorite game. For our purposive sample, we included only participants who had specified a favorite game. We identified the game genre to ensure each game genre was represented evenly in the final sample. In addition, only participants who had performed the sorting task in 12 minutes or more were included. We conducted pilot testing and found that the average sorting time was 25 minutes (range 2-3 minutes) and that reading the statements and swiftly sorting the statements took at least 12 minutes; thus, for less than 12 minutes, the sort was regarded as ill-considered. If there were participants with identical favorite games, only one participant was included. Participants’ preferences for modality (digital or analog) had to be distributed as evenly as possible across game genres; the male-to-female ratio had to be evenly distributed across game genres; and medical and dental student had to be evenly distributed across game genres. If a decision about inclusion or exclusion of a participant could not be made based the preceding criteria, the decision was made by rolling dice.

There is no single correct way to compile a set of statements in Q-methodology [41]. In general, the Q sample size is 40 to 80 statements [41,48], and the number of statements should exceed the number of participants [46]. A set containing too many statements can make the sorting process an exhausting and burdensome task, whereas a set containing too few statements may result in inadequate coverage of the topic of game preferences [41]. By sorting and prioritizing each statement from the statement set, individual participants provided us with a model of their view on their own game preferences. Statements should be carefully selected since their nature limits what can be expressed by a participant [49].

We aimed to develop a set of statements in which each statement was unique and made its own original contribution, and all statements together covered the full range of game preferences. Statements were based on the findings of an earlier focus group study [5] among medical and dental students (n=58) with no experience in game-based learning but widely varying experiences in play and games that had been conducted to obtain perspectives on leisure time and academic education. To make sure that the statement set covered as many aspects of game preferences as possible, we also examined player type studies [17-19,50] that possibly addressed different game preferences. This resulted in an initial set of 136 statements. We grouped the statements into 28 themes, duplicates were removed, and statements were translated into English and reworded to start with the phrase “I like games that….” in order to improve clarity and make sorting more intuitive for participants [41]. The final set (Table 1) consisted of 49 statements and was piloted by 3 medical students. Based on their feedback, we considered the final statement set to meet our abovementioned aims.

Participants performed the sorting procedure using a web app (Q-sorTouch), in which the 49 statements were randomly presented. Participants were asked to drag and drop each statement into 3 piles: agree, neutral, disagree. After sorting all the statements, they had to refine their 3 piles by ranking the statements into a Q-sort grid ranging from −4, extremely disagree, to +4, extremely agree. In Q-methodology, the number of statements that can be assigned to each scale point are fixed and represent a quasi-normal distribution (Table 2) [41]; thus, participants placed the 2 statements with which they disagreed most under −4 and the 2 with which they agreed most under +4.

The sorting procedure ended when all statements were placed in the fixed distribution and the participants felt that the final sort represented their viewpoint. In the final stage of the data collection, participants provided answers to open-ended questions to elaborate on the rationale behind their sort (eg, why statements were assigned to the extreme ends).

To identify groups of participants with shared, but distinct, viewpoints (ie, who subjectively ranked the 49 statements in a similar way), we conducted by-person factor analysis using dedicated software (PQMethod, version 2.35; developer: J Atkinson), which we later verified with formulas [48] in MATLAB (version R2020a; The MathWorks).

Because each sort was correlated with every other sort, the correlation matrix of the participants’ sorted statement sets (ie, sorts) was used to identify factors (ie, groups of respondents whose Q-sorts were statistically similar) by subjecting the correlation matrix to varimax rotation [41]. Varimax rotation generates a factor solution according to the best mathematical solution (while maintaining an orthogonal basis) [48]. Only factors with eigenvalues >1 and on which at least 2 participants are loaded significantly (P<.01) were accepted [41,48,49], which corresponded to a factor loading >0.37, calculated using 2.58 × (1 / √ (number of items in the Q set) [41,48]. Since our aim was to extract patterns that were unique, participants loading on more than one factor were not used for the construction of a factor. This is in line with the procedures applied in other Q-methodology studies [41,43,44,48].

A range of factor solutions were generated. To describe patterns of the participants’ game preferences, each factor solution was interpreted in conjunction with qualitative data from participants’ responses in the final stage of the sort. To facilitate factor interpretation, ideal Q-sorts were computed for each factor. These so-called factor arrays are weighted averages of sorts loading on that factor [41,49]. A group of 9 independent researchers individually interpreted all factor solutions and were asked to identify the solution with the highest number of viewpoints while providing distinct and clearly interpretable factors.

A total of 102 students volunteered to participate in our study and completed the sorting procedure. On the basis of their statements about their favorite games, we identified 7 game genres: action games (n=7), adventure games (n=6), party games (n=13), simulations or sports games (n=15), strategy games (n=35), puzzle games (n=14), and role-playing games (n=10). Consequently, we excluded 60 participants: 2 participants did not provide their favorite game; 10 participants performed the sort in less than 12 minutes; 36 participants had duplicate favorite games (eg, 11 participants stated the game Settlers of Catan); 9 participants (4 favorite digital games, 2 females and 3 male students) to ensure a more even distribution of these variables; and 3 participants, by the roll of the dice. The sample consisted of 42 participants (dental students: n=13; medical students: n=29) having 41 different favorite games, of whom 31 were female and 11 were male, with a mean age of 23.3 years (SD 4.0; range 18-42). Of the 42 participants, 15 participants considered themselves to be gamers. Nine sorts were confounded, and 3 sorts did not load significantly on any of the factors (factor loading <0.37; Table 3).

Pattern 5 comprised 2 participants with significant factor loadings (female: 2; age: mean 23.5 years, range 2-5 years), and both were both medical students. One student self-identified as a gamer. One student favored action games, the other student favored simulation or sports games. Both students favored a digital modality (n=2).

Having the ability to exploit game mechanics to cheat (26: +3), annoy other players (40: +3), and bluff (22: +3) was important for these 2 participants compared with participants described by the other patterns. Such behavior seemed to be the result of boredom or laziness and not really being interested in the game itself. Notably, these participants were not inclined to invest time to learn new techniques (17: −3) but, paradoxically, wanted to get better in a game (47: +3), did not like to see others learn the game (2: −3), and were inclined to play games differently than intended when the game would take too much time (25: +2; 29: −2).

Figure 2 presents a theoretical framework illustrating different player types in relation to sociability and achievement themes.

We aimed to improve the understanding of game-based learning design, in general, and of game-based learning in health professional education, in particular. We contended that player typology, a concept that has been used to inform game design and game play, may be relevant to game-based learning design. To the best of our knowledge, this study is the first to investigate player types in a student cohort, outside of a game design context, using a methodology deliberately aimed to accommodate the largest variety in game preferences. We found that, in a cohort of medical and dental students with likely similar academic interests and intellectual ability, there was considerable variability in play preferences. We identified 5 distinct and clearly interpretable patterns in game preferences, which can be considered player types: the social achiever, the explorer, the socializer, the competitor, and the troll. Given that only a few game elements are applied in education for health professionals research—predominantly points and rewards [1]—our findings indicate that there is room for improvement within game-based learning design; the current variety of game-based learning designs in education for health professionals seems too limited to be able to tailor game-based learning to students’ game preferences to improve learning through motivation and engagement.

Each player type has distinct characteristics. Social achievers like to play collaboratively to achieve something meaningful. They like competition and difficult games, preferably in teams. In contrast, explorers are drawn to the game’s story and immersive elements of play rather than winning and team play. Competitors on the other hand, thrive by winning and competition and would rather not depend on others. Socializers (much like social achievers) play for sociability, and interaction with other players is important to them. Yet, whereas competition is important to social achievers, socializers would rather play to find a sense of sociability and togetherness. They look for easy wins just to have a good time. Trolls like games in which they can annoy other players, bluff, and cheat.

Two themes (Figure 2) are salient in the player types that we identified, and likewise in scientific and grey literature on play and player typologies [17,19,35,39,51-56]: sociability and achievement. Competitors and social achievers like the achievement aspect of play, however, explorers and socializers instead preferred playing for the storyline or to enjoy playing together, respectively. Social achievers and socializers are driven by sociability, collaboration, and interaction; explorers and competitors, however, seem less prone to these traits or only need others to prove their supremacy. The troll is more ambivalent about sociability and achievement aspects of games than other player types and is, therefore, situated at the intersection of achievement and sociability.

Interestingly, the main themes identified in our theoretical framework bear similarity to the very first and often-cited player typology [18], which was based on a sample with homogeneity in terms of the preferred game, instead of homogeneity in terms of academic interest and which lacked any empirical basis. The fact that our study (which followed a more rigorous scientific approach) resulted in a similar typology may suggest that the existence of player types in a cohort is stable. Our scientific approach led to the identification of the explorer and the socializer, player types that have also previously been described [18]. We identified the social achiever, a player type that seems comparable with that of “achiever [18].” We additionally identified 2 other player type—the competitor and the troll—whereas in [18] only one other player type, namely the “killer [18],” which, upon close inspection, includes troll-like aspects (eg, annoying other players) as a social component (acting on other players). In our study, these characteristics appeared in other player types. We identified the troll and the competitor as separate player types. The achiever [18], with its social component was therefore interpreted as social achiever. The reason for these differences between both typologies may be that we also included games involving teamplay as a play genre in our framework, since we aimed to avoid selection bias from using only one or a few specific game genres to identify player types. The earlier typology [18] did not include teamplay, probably because it was based on a multiuser dungeon game that included role playing, player versus player, and chat functions but rarely team effort. We also found differences in relation to the dimensions on which the player types varied. Whereas the dimensions world versus player and action versus interaction have been previously described [18] our empirical evidence supported achievement and sociability as player type dimensions. As a result, competitors and socializers were opposites in our framework (instead of “killers” and “explorers” [18]).

The troll as a player type has not been identified in previous studies [20]. Remarkably, however, the troll phenomenon is well known in the field of problematic gaming and internet use [57,58]. Trolling is defined as deliberately trying to create distress or conflict via provocation, for instance, for the purpose of deception or disruption [58]. More than one-third of American millennials said they engaged in the act of trolling [59] and an immensely popular digital game, called Among Us, is based on the concept of trolling (ie, sabotaging and causing chaos [60]). This suggests that the game-related behavior of trolling is not rare or marginal. Although the relevance of this player type to game-based learning design is unclear, this player type might also be pertinent outside the field of education for health professionals.

The player types in this study represent a broad spectrum of views on games and play. One of the strengths of this study is that the comprehensive set of statements was derived from prior research among medical and dental students [5] and supplemented with statements taken from existing player type studies. Furthermore, a solid scientific method was used to account for all key subjective viewpoints on game preferences, and we included of a variety of participants (independent from game context) to prevent selection bias on game genre. In addition, we discussed multiple factor solutions, sought advice from expert authors [41], and verified Q-methodology software results. In doing so, we added a new perspective to literature on player types and game-based learning by identifying 5 patterns that were distinct, characteristic, and could be considered player types.

This study had some limitations: (1) In the interpretation of our patterns, we cannot (and do not) claim to be exhaustive with respect to all viewpoints on game preferences in the entire population. While Q-methodology is a method that aims to capture variety and heterogeneity, our participant group was relatively homogenous (medical and dental students). Therefore, we cannot claim that replication of our study in a different educational context would yield the same outcomes. However, by adding statements from prior (nonmedical) studies on player types in the statement set, and by using stratification to provide profuse and varied participants’ opinions, we feel that the quantitative aspect of the Q-methodology (ie, analyzing participants’ rankings using multivariate data reduction techniques) helped us detect meaningful patterns and connections in game preferences. This, in turn, may provide future researchers with a starting point to investigate the generalizability of our results. (2) In a recent study [5], we showed that game elements are possibly context dependent (ie, aspects that motivate play may not necessarily play a motivating role in game-based learning). For instance, although competition was liked and named trivial in play in leisure time, students considered it stressful and unwanted in play focused on learning. Since we did not ask participants to keep a specific learning environment in mind when they answered the question about their game preferences, their answers may not reflect their game-based learning preferences. (3) We aimed to reduce selection bias by selecting participants independent of game context, however, we do not know whether they had a specific game or context in mind when they performed the sorting procedure. (4) We chose to adopt the 5-factor solution after rigorous discussions and with the help of 3 independent researchers. Although this allowed us to detect a new player type (the troll), few students had significant factor loadings on this player type. Nevertheless, this player type adhered to the widely accepted rules for including a factor in Q-methodology and helped explain the largest variety in play preferences [58].

By using Q-methodology, we aimed to explain as much variety in existing game preferences as possible; thus, our player types are extreme ends of a spectrum on game preferences. The factor arrays that construct these player types are the combined average of all sort loadings on that player type. Therefore, there is very little chance that a participant’s sort will load 100% on a specific player type and fully match its definition [41]. Indeed, all sorts demonstrated characteristics of all player types, and no sort loaded 100% on one player type. Yet, most sorts loaded clearly on one player type.

Systematic reviews indicate that, often, game-based learning strategies are selected based on researchers’ personal opinions rather than theory or a conceptual framework [1,2,61]. Additionally, there is a tendency in game-based learning strategies to use scoring and reward, especially in gamification [1,61]. Our taxonomy provides a novel theoretical framework that may help to tailor game-based learning strategies to student preferences. Future research is needed to investigate whether such tailoring would result in increased effectiveness of applying game-based learning in education.

Based on our findings, all player types except explorers might need the presence or participation of other players to be optimally motivated to continue playing. To develop game-based learning strategies that optimally engage and motivate the majority of students, multiplayer options appear to be critical. However, this feature is currently overlooked in game-based learning strategies in current practice [1,11,61].

Our theoretical framework and corresponding factor arrays indicate that preferences for multiplayer modalities can be diverse and are not limited to sociability [62], social media [63], a chat function [34], and message boards [64]. Competitors, for instance, need other players or computerized opponents to triumph over and show their supremacy, social achievers need other players to work with, trolls need other players to annoy, and socializers need other players to have a good time together. By including each player type in a game-based learning-strategy, the complex and dynamic interaction between player types can turn game-based learning into a meaningful strategy for every student. For example, although trolls might only make a small contribution to the overall player population, their actions can have major impact on social play and interaction [65-68], much more than, for example, the actions of social achievers. The inclusion of trolls in game-based learning design can unite socially oriented players by giving them a common foe. Future research should explore how each player type can contribute to multiplayer game-based learning strategies to enhance collaborative learning.

Future research can focus on investigating whether the range of opinions on play vary significantly across students as a function of the academic level or discipline they are enrolled in, for instance, a medical or a nonmedical group, or medical specialization. Such findings would provide an understanding for future student-specific game-based learning designs. Game preferences might be dependent on context [5] or the players' current needs [31]. For instance, in the playground game called Tag, one player is it and chases the other players in an attempt to tag them by touching them. Then the tagged player becomes it and starts chasing the others to tag someone else. This means that, when being it, a player must adopt the competitor player type (ie, competing and winning from the others), while the others (who are getting chased) can adopt the social achiever or even troll player type to act as a group against the one that is it. Likewise, other digital games (eg, Among Us) perhaps also use changing player types, where one is sometimes a troll and, at other times, needs to take on the role of the social achiever [60]. This raises the question whether player types are in search of a specific game design or does the game design elicit different types of behavioral responses (ie, player types). This might also suggest that game designers should adhere to the entire diversity of player types to ensure inclusion of all participants of the game-based learning strategy.

As a first step in this direction, we aimed to investigate the prevalence of player types among medical and dental students. This may not only provide more evidence for the existence of the currently identified typology in education for health professional students, it may also shed light on the true diversity of player types within medical and dental education. Furthermore, it may improve our understanding of whether the current educational strategy focusing on the achievement-oriented player type is effective and can be justified or whether it might be better to tailor game-based learning strategies to individual player types.

We identified 5 clear and distinct patterns of game preferences. These patterns represent player types that differ in terms of the player type dimensions achievement and sociability. Our taxonomy and accompanying factor arrays can be used to tailor game-based learning design to students’ game preferences to optimize game-based learning effectiveness.