A video-game-based method to induce states of high and low flow

Joessel, Freya; Pichon, Swann; Bavelier, Daphne

doi:10.3758/s13428-023-02251-w

A video-game-based method to induce states of high and low flow

Original Manuscript
Open access
Published: 20 October 2023

(2023)
Cite this article

Download PDF

You have full access to this open access article

Behavior Research Methods Aims and scope Submit manuscript

A video-game-based method to induce states of high and low flow

Download PDF

Freya Joessel^1,2,
Swann Pichon^1,2,3 &
Daphne Bavelier^1,2

1153 Accesses
1 Citation
9 Altmetric
1 Mention
Explore all metrics

Abstract

Flow has been defined as a state of full immersion that may emerge when the skills of a person match the challenge of an activity. It is a special case of being on task, as during flow, keeping focused on the task feels effortless. Most experimental investigations of the neural or physiological correlates of flow contrast conditions with different levels of challenge. Yet comparing different levels of challenge that are too distant may trigger states where the participant is off task, such as boredom or frustration. Thus, it remains unclear whether previously observed differences ascribed to flow may rather reflect differences in how much participants were on task—trying their best—across the contrasted conditions. To remedy this, we introduce a method to manipulate flow by contrasting two video game play conditions at personalized levels of difficulty calibrated such that participants similarly tried their best in both conditions. Across three experiments (> 90 participants), higher flow was robustly reported in our high-flow than in our low-flow condition (mean effect size d = 1.31). Cardiac, respiratory, and skin conductance measures confirmed the known difference between a period of rest and the two on-task conditions of high and low flow, but failed to distinguish between these latter two. In light of the conflicting findings regarding the physiological correlates of flow, we discuss the importance of ensuring a low-flow baseline condition that maintains participants on task, and propose that the present method provides a methodological advance toward that goal.

Do Not Disturb: Psychophysiological Correlates of Boredom, Flow and Frustration During VR Gaming

The Flow Observational Grid: an Observation-Based Solution to Assess Flow States

Article 24 January 2021

Experimental Paradigms to Investigate Flow-Experience and Its Psychophysiology: Inspired from Stress Theory and Research

Introduction

Flow is a state of intense attentional focus that has been defined to describe an optimal experience, enjoyable and intrinsically motivating, which can emerge when the challenge of an activity is matched to the skills of the person (Csikszentmihalyi, 1974; Nakamura & Csikszentmihalyi, 2014). Importantly, flow is a special case of being on task while remaining focused on that task feels effortless (Csikszentmihalyi, 1990; Ullén et al., 2010) and contrasts with other on-task behaviors, not associated with flow, which typically require an effortful maintenance of attention. Identifying unobtrusively whether or not someone is in flow state is an important and yet-unresolved issue in experimental works aiming at investigating the psychological and physiological correlates of flow. Indeed, like sleep, one cannot get in and out of flow on demand, and asking someone about their flow level will take them out of that state. Experimental investigations of flow usually contrast conditions of low, medium, and high challenge in the same task, where the medium condition is assumed to be the “flow” condition, as it is where the challenge of the task is aimed at being the closest to the skills of the participant. It is thus crucial to choose a task and difficulty levels such that not only can high flow emerge in one condition, but the participant also remains on task in the other condition(s) that serve(s) as the baseline state. Indeed, as the difficulty of the task becomes either low or high, mental states such as boredom (low difficulty) or frustration (high difficulty) may emerge, thus pushing the participant off the task. It would then be unclear whether differences between the flow condition and such baseline conditions may be ascribed to flow state rather than differences in being on or off task. We review below the various video game-based paradigms that have been used to experimentally manipulate flow, before introducing a methodical approach that not only ensures different flow levels, but also ensures that participants remain on task, truly committed to trying their best in conditions of different challenge levels (replacing studies like those of Bian et al., 2016, that have only one flow condition).

Many real-life activities have been associated with the experience of flow, but few can be experimentally manipulated in a lab. Video game play behaviors offer this possibility and are known to elicit strong flow experiences (Csikszentmihalyi, 2014; Sweetser & Wyeth, 2005). Some have even suggested that making boring tasks playful is a way to induce flow and overcome boredom (Mathwick & Rigdon, 2004). Therefore, a majority of experimental studies in the literature have resorted to some sort of play behavior to induce flow, with a strong historical focus on sports playing. Yet, to extract meaningful neural or physiological markers of flow, sports manipulations are far from optimal given the large muscle activity and motion they entail. As such, studies that investigated the neural and physiological correlates of flow have often used computerized tasks, especially video games, in which they manipulated the level of difficulty across different experimental conditions.

Using a race game, Tozman et al. (2015) contrasted three conditions where difficulty was varied with the goal of inducing different levels of flow: an easy condition where participants drove on a straight road, a medium-difficulty condition where the racetrack was slightly bendy, and a hard condition where participants played on a very bendy road against nine opponents stronger than them. The same principle was used in Harris et al. (2017), who asked participants to complete, in a race game, ten laps on tracks at three fixed levels of difficulty. In a somewhat similar paradigm, Peifer et al. (2015) had participants play Pac-Man through five increasing levels of difficulty, ranging from very easy to very hard, with the third level being of medium difficulty. The levels of difficulty were determined first by asking participants to rate the difficulty of ten levels, five of which were then selected for the final experiment. Unlike other studies, Nacke and Lindley (2010) did not use a high-difficulty condition. Using the game Half-Life 2 as a basis, they implemented a first, easy condition where the level was designed to be very linear and repetitive, with no progression in difficulty, a second flow condition designed according to criteria for a flow experience in terms of pace and difficulty progression, and a third condition designed to be more narratively rich, with high potential exploration but with less emphasis on fitting the pace to the player.

In all these studies, the medium level of activity was framed as the flow condition of interest. However, the difficulty in the “medium” condition was not fixed for all and not titrated to the individual skills of each participant. As such, these paradigms contrasted fixed difficulty levels in rather similar ways as 1990s literature investigating the physiological correlates of effort (Althaus et al., 1998; Backs & Seljos, 1994). Although this allows us to understand how flow, task difficulty, and effort relate to each other, it is generally agreed in the literature that the flow state cannot be reduced simply to different levels of task difficulty or effort. Rather, flow corresponds to a unique psychological state that may emerge when the challenge of the task meets the skill level of the individual, which to be accurately investigated requires a baseline condition that also keeps participants on task (instead of considering baseline low-flow conditions, where participants are likely to become bored or too frustrated and be off task).

More recent studies have attempted to tailor the difficulty of a “medium” flow condition to the skills of participants to ensure it was associated with an optimal level of challenge. We consider below two approaches to doing so. A first approach is to use an adaptive algorithm that adapts task difficulty in real time while participants are playing. For example, in Keller et al. (2011), participants played the game Tetris in one of three individually tailored conditions of difficulty: easy, medium, and hard. In the easy condition, the Tetris pieces would fall at a slow pace, with no possibility for the player to accelerate the fall. In the medium condition, the difficulty level or falling speed of the Tetris pieces was adjusted during the game as a function of the player’s performance. Finally, in the hard condition, the speed was maintained so high that it was difficult to even clear one line, putting the player at risk of quitting. In Núñez Castellar et al. (2019), the level of the game was adapted as a function of the participant’s performance in the “medium” flow condition, while in the other two conditions, the participant just repeatedly played the game’s easiest or hardest level, respectively. In the same vein, Klarkowski et al. (2016) had participants play a session of Left 4 Dead 2 against enemies that provided no resistance, against enemies that dynamically adapted their difficulty to the player’s skills, or against enemies that made the game close to impossible to play. Although not using a game, Ulrich et al. (2016) had participants complete an arithmetic task that involved summing the numbers presented on a screen in three difficulty conditions: a very easy one, a second one where the difficulty was dynamically adapted to the participant’s level, and a third one where the difficulty was clearly above the participant’s skill level.

While this first strategy ensures that challenge in the medium-difficulty flow condition is matched to the skills of the participant, it leaves a number of key differences between this condition and the easy or hard fixed condition. In particular, the task difficulty in the medium condition is adaptive and individualized, while it is not in the other two. It would seem preferable not to confound flow with the use of closed-loop adaptive algorithms, which have been shown to engage brain circuits in a rather unique way (Mishra & Gazzaley, 2014). Another uncommon implementation of this approach can be found in de Manzano et al. (2010) and Ullén et al. (2010), who invited professional pianists to play two pieces of classical music. Participants were asked to bring a piece of their own choice that they knew they could play well and enjoyed playing, while the other piece was selected by the researchers to be extremely difficult, even for professional pianists. This latter way of proceeding suffers, however, from other confounds such as differences across conditions in memory content, emotional value, or familiarity, to cite a few, as well as the possibility that participants may not choose a piece that is actually matched to their skills due to motivational considerations (such as fear of failure). This method thus appears equally problematic.

The second approach is to determine the level of the player before the experiment occurs and set the difficulty of the compared conditions accordingly. This approach typically attempts to contrast an optimal challenge, high-flow condition to low-flow conditions induced through either easy or hard challenges, but does not use adaptive algorithms during the experiment. This is what Harmat et al. (2015) did using Tetris. In a short game session before the flow induction experiment, the participant’s optimal level was determined by having the participant play Tetris for 6 minutes in adaptive mode. The level achieved at the end of the training period was taken as the optimal flow level for that participant. Then the levels for the easy and hard challenge conditions were determined as being three levels above the optimal level (frustration) and three below the optimal level (boredom), respectively. Using the same game, de Sampaio Barros et al. (2018) implemented a slightly different method, whereby the easy Tetris condition was set at the easiest difficulty level of the game, the medium flow level was determined beforehand as the level at which the participant could make four lines with 30 pieces, and the hard level was implemented by doubling the falling speed of the medium level. Finally, Tozman et al. (2017) invited national- and international-level chess players with an Elo ranking to play a game of chess at the lab. Participants had to play either against a computer at their Elo ranking level or against opponents 400 Elo above or below their level, which corresponds to a 10% or 90% chance of winning, respectively.

Overall, these three studies were successful at inducing a higher flow state in the optimal challenge condition than in the easy (often termed boredom) or hard (often termed frustration) conditions. Of note, two of these studies also investigated the physiological correlates of flow using cardiac and respiratory measures. While Harmat et al. (2015) did not find any difference between their conditions in terms of heart rate or its variability, Barros and collaborators (2018) observed increased heart rate with increasing difficulty, but no difference in heart rate variability. Given their low cost and ease of use, whether physiological measures, such as heart rate and heart rate variability, can be relied upon as online measures of the flow state remains an interesting avenue to explore. In Experiments 2 and 3, we investigated how peripheral physiological measures may be used to provide objective markers of flow.

The present work builds upon these studies and the goal of identifying online markers of the flow state. To this end, we propose to manipulate flow by comparing two conditions with different levels of challenge carefully chosen to induce low and high levels of flow while ensuring that participants remain on task, i.e., try their best in both conditions. In this way, the present study furthers recent flow studies that have rather contrasted their optimal challenge condition with other conditions that differ in their incentive to stay on task because their level of difficulty is set to induce either boredom or frustration. In doing so, participants may be more likely to disengage from the task in the easy or hard conditions than in the high-flow condition of interest. In the present study, we propose controlling that participants stay similarly on task in a low-flow baseline condition and high-flow condition of interest.

Hence, the present work introduces a method to reliably generate two key experimental conditions to reveal possible neural or physiological markers of flow, while controlling for matched on-task behavior. As reviewed above, a first methodological choice is to contrast conditions that induce more or less flow while limiting possible confounds such as differences in adaptive procedure or in familiarity across conditions. Our approach will thus closely mirror that of Harmat et al. (2015) in terms of conditions contrasted, and in particular avoid contrasting one condition with a closed-loop adaptive mechanism to one without. Unlike Harmat et al. (2015), our experimental design includes two conditions instead of three: one where the challenge is matched to the skills of the participant and thus the probability of experiencing flow state is high, and a second one of greater challenge where the probability of experiencing flow state is lower. Importantly, these two conditions were designed such that participants remained on task and kept trying their best in each of them. To control that this was indeed the case, we tracked how much participants tried their best in each condition using the Effort/Importance subscale of the Intrinsic Motivation Inventory, which includes items such as “I did my best while I was playing the game” (see SM-02 in the supplementary materials for the full list of items). After establishing this method, we ask whether such different flow states induce different levels of peripheral physiological activity using cardiac, respiratory, and electrodermal recordings.

Experiment 1

Experiment 1 aimed at developing a procedure to induce low and high levels of flow level between two comparable on-task conditions. We devised a procedure to induce either high or low levels of flow by manipulating the challenge/skill balance within a video game, while at the same time ensuring that participants stayed on task, trying their best in each condition. To this end, we used the video game Unreal Tournament 2004 (UT2004), a first-person shooter game that provides the typical high-pace, absorbing, and challenging type of gameplay beneficial for maintaining flow. The game also provides a straightforward measure of performance computed as the kill-to-death ratio (KDR) within a level, and offers the possibility to finely manipulate difficulty by tuning the skill and the number of artificial intelligence (AI) opponents. We present an individualized procedure to tailor the difficulty of the game to the participant’s skills during a training phase using the KDR as a performance criterion. This approach was used to generate two experimental conditions: one “high-flow” condition where difficulty was matched to the participant’s skills, and one “low-flow” condition where difficulty was too high for the participant but still within a reasonable range to avoid having them give up the game.

Our procedure first involved finding the level of the game to induce flow given the participant’s skills, using the same game mode and map for each participant. We operationalized this level as the level where participants would have a KDR between 2 and 0.5, meaning players were at most twice as likely to kill than be killed (KDR = 2) or to be killed than killing (KDR = 0.5). Within this KDR range, players roughly achieve as many kills as deaths, a sign that the skills of their opponents (i.e., the challenge of the task) are balanced to their own skills. Once this level was determined for a participant, it was set as the high-flow level for this participant. The low-flow condition was then created by increasing the game difficulty by three levels (on a scale from 1 to 8 preset difficulty levels in the game). By matching skills to game challenges for each participant, we optimized the probability that flow would emerge in the high-flow condition. In contrast, by increasing the challenge above the participant’s skill level by a systematic value, we made sure that all participants experienced the same contrast in challenge between their high-flow and low-flow conditions. Crucially, this increase in difficulty was set to keep the game within a playable range for a wide variety of participants, ensuring that participants kept trying their best at the task even if somewhat frustrated.

To measure flow, we used the Flow State Scale (Jackson & Marsh, 1996) that assesses, through separate subscales, the nine components of flow described by Nakamura and Csikszentmihalyi (2014): challenge-skill-balance, unambiguous feedback, clear goals, concentration, sense of control, action-awareness merging, transformation of time, loss of self-consciousness, and autotelic experience (which refers to the phenomenon that the activity is experienced as enjoyable and intrinsically rewarding). Given our aim to manipulate flow while having participants trying their best in both conditions, we used the Effort/Importance subscale from the Intrinsic Motivation Inventory (IMI; Deci et al., 1994) to check that participants similarly tried their best in both conditions. This subscale asks “how hard participants tried to do the task,” with the exact phrasing of the four questions we used for this scale being “I did my best while I was playing the game,” “I put a lot of effort into playing the game,” “I tried very hard on this game,” and “It was important to me to do well at this game.” We refer to this subscale as “Best-Try” rather than “effort” to prevent the possible confusion that may arise from the different definitions of effort by other motivation frameworks (see Experiment 2 [E2]). Second, we used the Perceived Competence subscale from the IMI to check that our experimental manipulation of challenge successfully impacted participants’ perceived competence at the task. We expected perceived competence to be higher in the high-flow condition, where skills and challenges are matched, than in the low-flow condition, where difficulty exceeds skills. Third, we also used the Interest/Enjoyment subscale of the IMI to check that interest remained high across our two conditions, even if possibly lower in the low-flow than the high-flow condition. Finally, we used as a more exploratory measure the Pressure/Tension subscale of the IMI to assess how tense and pressured participants felt while playing. Given that our experiment did not involve performance-contingent incentives (e.g., a given goal to reach in the game or monetary rewards linked to performance), external pressure, or a social component, we expected this scale to remain rather low across both conditions, with possibly greater pressure/tension in the low-flow, more difficult condition.

The primary goal of Experiment 1 is to present a procedure to induce in each participant a high-flow state versus a low-flow state while controlling for best-try, as defined above. This novel contribution is outlined in detail below in the section “Individualized procedure to titrate high-flow and low-flow experiences.” Here our hypotheses were as follows and will remain the same for all three experiments: (i) the Flow State Scale (FSS) score will be higher in the high-flow (matched challenge) condition than in the low-flow (over-challenging) condition; (ii) the best-try score will be similarly high in the high-flow (matched challenge) and the low-flow (over-challenging) condition.

A secondary, pragmatic goal of Experiment 1 was to assess the possibility of inducing such flow variations while playing the game without the associated soundtrack. Indeed, it has been previously suggested that in-game sounds may have a positive influence on the emergence of flow (Nacke et al., 2010). However, with an eye toward using this paradigm to track the neural bases of flow, and given our aim to port it to a magnetic resonance imaging (MRI) scanner (Experiment 3), it was important to check the role of sound stimulation, if any, in our flow manipulation. As such, we also checked that our flow manipulation would be effective even without in-game sound.

Materials and methods

Participants

Twelve healthy individuals—11 men and 1 woman between 18 and 32 years of age (mean age = 22, SD = 2.9 years)—participated in this study. They were covertly recruited from the Bavelier lab pool of participants. Each participant in this pool has filled out a series of questionnaires, including one about their video game play habits, and has been asked to play in the lab at least one 10-minute session of the game Unreal Tournament 2004 (UT2004). For each play session, the number of kills and deaths during the match were recorded.

Participants were recruited based on the following inclusion criteria:

1.
Their familiarity with first- or third-person shooter (FPS) games: Participants had to have played on average 1 to 10 hours of shooter games per week over the past year. This ensured that participants were already familiar with the basic game play mechanics of shooters.
2.
Their score during their in-lab 10-minute play: Participants’ kill-to-death ratio (KDR, number of kills over number of deaths) in Unreal Tournament during our screening procedure had to be between 0.5 and 4. In this sample, all participants completed two sessions. We utilized the KDR obtained from their second session, as it provides a more accurate estimate of their skill level in the game, given that participants typically acclimate to the game during their initial session.
3.
Their age: Participants had to be between 18 and 35 years old.
4.
Finally, all participants had to be right-handed and to have normal or corrected-to-normal vision.

The sample size for this and the following two experiments could not be based on power computations, as the expected effect size was unknown given the aim of establishing the feasibility and robustness of the novel methodology we present.

Participants were given information about the experiment and were asked to provide informed consent in accordance with the Declaration of Helsinki, under a protocol approved by the ethics commission of the Faculty of Psychology and Educational Sciences of the University of Geneva, and were paid for their participation.

Experimental design

All participants were asked to play UT2004 successively in four different conditions, crossing the factors of flow level (high, low) and sound (with sound, no sound). In the high-flow conditions, participants played a free-for-all deathmatch against four other players controlled by the computer (called bots) at a level that was adapted to their skill (i.e., the challenge matched their skill), while in the low-flow condition, they played a free-for-all deathmatch against four bots that were too skilled for them (i.e., the challenge exceeded their skill). Participants played at each level twice, once with game sounds activated and once with them deactivated. The 12 participants were assigned to one of four orders counterbalancing the flow and the sound factors, such that the first two games had sound on and the last two games had no sound or vice versa. Whether low or high flow was then presented first was counterbalanced, keeping the same order for sound and no-sound runs. Condition orders were interleaved across participants, with participant 1 performing the experiment in order 1, participant 2 in order 2, participant 3 in order 3, participant 4 in order 4, and back to 1 for participant 5, and so on until all participants had been recruited.

Since each player’s skill is unique, we developed a novel procedure to determine the level of the game that matches the skills of each participant. This individualized procedure is described in detail before the results section.

Materials and apparatus

Flow State Scale (FSS; Jackson & Marsh, 1996)

Flow was measured using a French translation of the Flow State Scale (FSS, Jackson & Marsh, 1996). As no agreed-upon translation existed at the start of our study, we share in the supplementary materials (SM-01) the French translation we developed. Note that participants were presented with the French translation of each item and its original English version in italicized font just underneath. This 36-item questionnaire assesses each of the nine components of flow described by Csikszenmihalyi (Nakamura & Csikszentmihalyi, 2014) using one subscale for each component. Each subscale comprises four items evaluated on a five-point Likert scale. For example, for the component of flow sense of control, participants evaluate the statement “I had a feeling of total control,” or for the component concentration they evaluate the statement “I was completely focused on the task at hand.” The nine components of this scale are known to have good internal consistency, with Cronbach alphas between 0.80 and 0.86 on a sample of 394 participants (Jackson & Marsh, 1996). We report the Cronbach alphas for our studies in the supplementary materials (SM-01). Flow is measured as the average points over all 36 items.

Intrinsic Motivation Inventory (IMI; Deci et al., 1994)

We used a four-item subset of each subscale to keep the questionnaire as short as possible. Our selection of four items was based on the authors’ statement that “the incremental R for every item above 4 for any given [subscale] is quite small,” and the results of a validity study by McAuley et al. (1989) who also used four items per subscale. As in the original IMI, all the items in the IMI were rated on a seven-point Likert scale from “not at all” to “very true.” Our reduced IMI version can be found in full in the supplementary materials along with the Cronbach alphas for each subscale (SM-02). As no agreed-upon translation existed at the start of our study, we share in the supplementary materials (SM-02) the French translation used. Similarly, as for all the scales in these studies, participants were presented with the French translation of each item and its original English version in italicized font just underneath.

Best-try (Effort/Importance subscale)

We based our selection of four items on a previous study that had used the IMI in a video game context (Vos et al., 2011) and the reliability analysis from McAuley et al. (1989). The selected items are “I did my best while I was playing the game,” “I put a lot of effort into playing the game,” “I tried very hard on this game,” and “It was important to me to do well at this game.”

Perceived competence

Perceived competence was measured using a four-item subset of the Perceived Competence subscale. We selected the four items based on a previous study that used the IMI in a video game context (Vos et al., 2011) and the reliability analysis from McAuley et al. (1989). The selected items are “I think I am pretty good at this game,” “I think I did pretty well at this game, compared to other participants,” “I was pretty skilled at this game,” and “I think I was good at this game.”

Interest

Interest was measured using a five-item subset of the Interest/Enjoyment subscale of the IMI. We selected the four items from a previous study that used the IMI in a video game context that showed the best reliability in a previous reliability analysis (McAuley et al., 1989). The items in this subscale were “I enjoyed playing this game very much,” “This game was fun to play,” “I thought this game was quite enjoyable,” and “I would describe this game as very interesting.” We also added the item “I thought this was a boring game” given the importance of avoiding boredom.

Tension

Tension felt by the participant during the game was measured using a four-item subset of the Pressure/Tension subscale of the IMI. We selected the four items from the McAuley et al. (1989) study that had the best reliability. The items included are “I felt pressured while playing the game,” “I felt very tense while playing the game,” “I was very relaxed in playing the game,” and “I was anxious while playing the game.”

Objective difficulty measure

The kill-to-death ratio (KDR) was initially used in our procedure to assess a participant’s skills during the training phase and to titrate the level of challenge of the high-flow and low-flow conditions during the subsequent main experiment. KDR was also monitored during the main experiment as an objective measure of difficulty in both conditions and used as a sanity check to ensure that the two conditions presented different levels of challenge to the participant. We log-transformed the raw KDR given that the difference between a KDR of 0.1 and that of 0.5 is the same as that between a KDR of 2 and a KDR of 10 (both reflect that one game had a difficulty level five times greater than the other).

Individualized procedure to titrate high-flow and low-flow experiences

Our procedure aimed for a balance between skills and challenge, avoiding a situation in which participants either never die (too easy) or die so fast that it is frustrating. Accordingly, we aimed for a KDR between 0.5 and 2 to titrate our high-flow setting. The low-flow level was then set at three levels higher than the high-flow level—for example, if the high-flow level was 4 (in-game name: skilled), then the corresponding low-flow level was 7 (in-game name: inhuman). The maximum level in the game is 8 (in-game name: godlike).

To find the high-flow setting, this procedure built on two different UT2004 game play episodes: a 10-minute screening game play and 5-minute evaluation game play. The screening game consisted of a 1v1 deathmatch against a bot (a player controlled by the computer) with the difficulty level set on experienced (3 out of 8) and played on the map Crash. The evaluation game was a free-for-all deathmatch with four bots, using the Gestalt map, at a difficulty level selected according to their KDR at the outcome of the prescreening. As presented in Fig. 1, if their prescreening KDR was equal to or less than 2, the evaluation game was set on the skilled mode of UT2004. If their prescreening KDR was more than 2, the evaluation game was set on the adept mode of UT2004. All other parameters were similar to those of the other games as listed below. The experimenter then recorded the KDR of the participant and determined their high-flow and low-flow levels according to the procedure shown in Fig. 1.

If the participant played the evaluation game in skilled mode and their KDR was equal to or lower than 0.5, then their high-flow level was set to experienced (3 out of 8, or 3/8) and their low-flow level to masterful (6/8). If their KDR was between 0.5 and 2, then their high-flow level was set to skilled (4/8) and their low-flow level to inhuman (7/8). If the participant played the evaluation game in adept mode and their KDR was equal to or lower than 0.5, then their high-flow level was set to skilled (4/8) and their low-flow level to inhuman (7/8). If their KDR was between 0.5 and 2, then their high-flow level was set to adept (5/8) and their low-flow level to godlike (8/8). No player reached a KDR greater than 2 during their evaluation game.

Procedure

Participants were given information about the experiment and were asked to provide informed consent. They were then asked to go to an isolated soundproof booth and were given a brief overview of the specifics of UT2004 (for instance, what each weapon does or what health packs look like). The participants then played a 5-minute evaluation game (as described in our titration procedure), and the difficulty levels for their high-flow and low-flow games were determined. The participants then played four 10-minute games, crossing the factors flow (high flow vs. low flow) and sound (present vs. absent). The map was different for all four game sessions to avoid confounds from knowing the map better from one game to the other, but followed the same order for all participants (Rankin / Rrajigar / Corrugation / Compressed). A summary of the procedure is given in Fig. 2. The parameters for the game can be found in the supplementary materials (SM-09).

At the end of each 10-minute game session, the participants first completed the FSS and then the subset of questions from the IMI, with the instruction to answer these questionnaires based on their experience during the last game only. After completing these two questionnaires, they were allowed to take a short break before proceeding with the next game. The participants saw the experimenter set the difficulty of game play and were instructed to just enjoy the game and perform as well as possible, even when the difficulty was too high for them. Each time the participants were playing, they were left to play alone in a dedicated room. In both the sound and no-sound conditions, participants were instructed to wear headphones. The questionnaires were filled out on the same computer as the game play was performed using a LimeSurvey interface.

Results

The goal of Experiment 1 was twofold: (i) to evaluate the ability of our proposed procedure to manipulate flow state, expecting higher FSS in the high-flow than the low-flow condition, while keeping best-try matched across conditions, and (ii) to test whether the presence or absence of sound would have an impact on the flow manipulation.

Because we were also interested in quantifying the amount of evidence for the null hypothesis (or in other words, an absence of difference in best-try between the high-flow and low-flow conditions), we opted to use Bayesian statistical analyses. A small guide on how to interpret results from Bayesian ANOVAs and t-tests can be found in the supplementary materials (SM-07). In addition, we report the results of the frequentist repeated-measure ANOVA on the FSS for all experiments in the supplementary materials (SM-08) as means of comparison with the results of the Bayesian analyses.

Using JASP software (JASP Team, 2018), we performed a Bayesian repeated-measure ANOVA with flow (high flow or low flow) and sound (sound or no sound) as within-subject factors on all our dependent variables, namely the flow measure from the FSS and the best-try measure, as well as the measures of perceived competence, interest, and tension. Since there might be differences between participants who started with a high-flow game compared to a low-flow game, we also added an “order” between-subject factor.

The data for the FSS and the Best-Try scale are presented in Fig. 3. Descriptive statistics for all the dependent measures in all four conditions are given in Table 1.

Table 1 Experiment 1. Descriptive statistics (N = 12). For each dependent measure, average and standard deviation in parentheses are reported. The evidence for an effect of flow condition is the Bayes factor of the model containing only the main effect of flow against the null model. FSS, Flow State Scale; Perceived comp., perceived competence. Non-log-transformed KDR values are given for information, as they are more straightforward to interpret than the log-transformed values used in the analyses

Full size table

Flow score (FSS)

The Bayesian repeated-measure ANOVA with flow and sound as within-subject factors and order as a between-subject factor (Table 2) indicated very strong evidence for the model with only the main effect of flow on FSS score (BF₁₀ = 84.6) compared to the null model, and anecdotal evidence against the model with only the main effect of sound (BF₁₀ = 0.34 = 1/2.94) or order (BF₁₀ = 0.61 = 1/1.64) compared to the null model. In addition, adding the interactions did not improve the models.

Table 2 Experiment 1. Model comparisons for the FSS score. Results of the Bayesian repeated-measure ANOVA (JASP) with flow and sound as within-subject factors and order as a between-subject factor (we show here only the 9 models including flow out of the 19). The best-fitting model is the one with flow as a unique factor (in bold). P(M|data) denotes the posterior probability of the listed model, BF₁₀ is the Bayes factor (or evidence) of the model against the null model. We used a uniform prior distribution

Full size table

Best-try (IMI – Effort/Importance)

In line with best-try being matched across flow conditions, there was anecdotal evidence against the model with only the main effect of flow on best-try (BF₁₀ = 0.38 = 1/2.63), anecdotal evidence against the model with only the main effect of sound (BF₁₀ = 0.50 = 1/2), and anecdotal evidence for the model with only the main effect of order (BF₁₀ = 1.41) compared to the null model. In addition, adding the interactions did not improve the models.

Perceived competence (IMI – Perceived Competence)

As expected, there was decisive evidence for the model with only the main effect of flow on perceived competence (BF₁₀ = 3.89×10⁶) compared to the null model, and weak evidence against the models with only the main effect of sound (BF₁₀ = 0.29 = 1/3.45) or order (BF₁₀ = 0.32 = 1/3.13) compared to the null model. In addition, adding the interactions did not improve the models.

Objective difficulty (Log₁₀(KDR))

As expected, there was decisive evidence for the model with only the main effect of flow on Log₁₀(KDR) (BF₁₀ = 1.95×10¹⁴) compared to the null model. As for perceived competence, there was weak evidence against the models with only the main effect of sound (BF₁₀ = 0.29 = 1/3.45) or order (BF₁₀ = 0.19 = 1/5.26) compared to the null model. In addition, adding the interactions did not improve the models.

Interest (IMI – Interest/Enjoyment)

There was anecdotal evidence for the model with only the main effect of flow on interest (BF₁₀ = 2.5), and weak evidence against the models with only the main effect of sound (BF₁₀ = 0.29 = 1/3.45) or order (BF₁₀ = 0.43 = 1/2.33) compared to the null model. In addition, adding the interactions did not improve the models.

Tension (IMI – Pressure/Tension)

There was anecdotal evidence against the models with only the main effect of flow on tension (BF₁₀ = 0.65 = 1/1.54), sound (BF₁₀ = 0.46 = 1/2.17), or order (BF₁₀ = 0.62 = 1/1.61) compared to the null model. In addition, adding the interactions did not improve the models.

Discussion: Experiment 1

Experiment 1 confirmed that our adaptation of difficulty to individual skills in UT2004 was effective at changing the state of flow as measured with the FSS. The effect size of the difference in flow between the high- and low-flow conditions was 1.36 (Cohen’s d; 95% CI [0.55, 2.14]), reflecting a strong effect with a mean difference of 0.41 (SD = 0.33).

Experiment 1 also confirmed that this manipulation in flow level can be achieved in the absence of the game sounds. Participants played UT2004 at the same level of difficulty with or without the game sounds, always keeping the headset on. Our flow manipulation (difference in FSS between the high- and low-flow conditions) was numerically smaller without sound (0.35) than with sound (0.48). This finding is also reflected in two subjects reporting greater FSS scores in the low-flow than in the high-flow condition without sound, but none when the sound was on. However, our results yielded inconclusive evidence for or against a main effect of sound or an interaction of sound with flow. Evidence for a difference between high-flow and low-flow conditions when the participants played without sound was observed with an effect size of d = .87 (95% CI [0.19, 1.53]). These results pave the way to using the proposed paradigm in settings where sound may not be available, as shown in the next two experiments.

We also note that our results are unlike those of Nacke et al. (2010), who reported lower flow in a no-sound condition than in a sound-on condition, using the flow dimension of the Game Experience Questionnaire (GEQ; IJsselsteijn et al., 2007; Poels et al., 2007). One attribute to monitor in future studies to better understand the impact of sound on flow may be the degree of video game expertise in the population sampled. Indeed, although measuring immersion rather than flow, Zhang & Fu (2015) reported that presence or absence of in-game sounds did not change immersion in habitual video game players (about 33 hours/week), in contrast to participants playing fewer hours per week (7 hours/week), for whom in-game sounds increased immersion. In line with that proposal, our sample comprised participants playing video games on average 11 hours/week (range [4;31]), with all participants playing more than 2 hours/week of FPS, whereas Nacke et al. (2010) sampled from the general student population without criteria on video game play hours, possibly recruiting participants with less video game expertise. This effect might also be mediated by game type. For example, Wiemeyer, (2013) reported no difference in flow (measured by the flow construct of the GEQ) in exergames in a sample of habitual and non-habitual video game players. Exergames differ, however, in terms of theme, situations to overcome, challenges, and attentional load compared to the first-person shooter games used in Nacke et al. (2010); Zhang & Fu (2015). Whether video game usage habits and game type may explain different sensitivity of flow to in-game sound remains to be fully addressed.

Finally, while a clear difference in flow was observed between high- and low-flow conditions, our Bayesian analyses indicated that the degree to which participants tried their best was relatively comparable across those conditions, with anecdotal evidence for the null (BF₁₀ = 1/2.63). Our novel paradigm achieved this by individually tailoring the game play level to reach a balanced KDR for each participant and systematically going three levels higher than this personalized high-flow level to generate a more difficult, but not overwhelmingly so, low-flow condition. Only two other flow studies have used a similar procedure when using video games to induce flow (de Sampaio Barros et al., 2018; Harmat et al., 2015); importantly, the present study is the first to assess whether best-try behavior is maintained when inducing high- and low-flow conditions with such a procedure.

As predicted, presenting a higher play challenge than the participant’s skill level resulted in lower objective ease of play (KDR) and perceived competence of participants in the low-flow versus high-flow condition.

Finally, we expected similarly high and matched measures of interest as well as similarly low and matched measures of tension. We did observe similarly high above-average interest and similarly low below-average tension scores across the two flow conditions. This pattern of results suggests that our manipulation led to two conditions of relatively high interest without applying unnecessary tension on the participants in either condition.

Experiment 1 presented a procedure to systematically induce both low- and high-flow conditions, all the while ensuring participants stayed on task, as measured by participants equally trying their best (IMI Effort/Importance scale). In Experiment 2, we recruited a larger sample to achieve more conclusive levels of evidence in our main comparisons. We also wanted to investigate whether our so-induced states of low and high flow could be related to different states of sympathetic and parasympathetic activation.

Experiment 2

Experiment 2 is a replication and extension of Experiment 1. First, to extend our understanding of the interplay between flow, motivation, and effort, we wished to address more directly how difficulty level affects the motivational state of the player, varying from flow to frustration or to boredom. Indeed, while participants showed greater performance (both subjective as reported by the IMI–Perceived Competence subscale, and objective as shown by the KDR), it is possible that the participants did not subjectively feel that one condition required more effort than the other. Thus, to disentangle the effect of our game difficulty manipulation on effort from that on best-try, we added to our battery of questionnaires the Rating Scale of Mental Effort (RSME; Zijlstra, 1993), where participants reported “how much effort it took them to do the task they just completed.” To date, only one study has directly investigated the link between flow and how much effort it took participants to play a game at easy, medium, and hard challenge levels (Harris et al., 2017). Using the RSME, the effort produced to complete the medium-challenge, high-flow condition (not tailored to the skills of the participant) was found to be higher than that for the easy condition and lower than that for the hard condition. Thus, RSME-based effort mirrored the difficulty level of the game, and not flow level. Similarly to Harris et al. (2017), we expected effort to change with task difficulty and to be higher in the low-flow than the high-flow condition.

We were also interested in how our measure of flow interacted with a measure of frustration. Indeed, while frustration may lead to the participant giving up on the task, it may also be an integral part of the flow experience, as flow can be experienced in those moments where one overcomes a difficult challenge, which necessarily elicits frustration (Csikszentmihalyi, 1990). In addition, in the video game design literature, it has been suggested that occasionally putting the player in purposefully designed frustrating situations leads to greater immersion in the game (Lyons, 2015; Qin et al., 2010), and observations from purposefully hard games tend to show that while the games can be extremely frustrating, high satisfaction can be derived from overcoming the challenges within these games (Birk et al., 2015). Two frustration questions were thus added as an exploratory measure on how to best control for motivational and effort constructs across our flow conditions.

Additionally, we collected cardiac (heart rate and high-frequency heart rate variability [HF-HRV]), respiratory (respiratory rate and depth), and electrodermal (skin conductance level [SCL]) measures in an effort to assess the physiological correlates of high- versus low-flow state. The most widely accepted model of the physiology of flow to date posits that flow is characterized by “moderate levels of sympathetic activation and moderate parasympathetic activation—i.e., a sympathetic-parasympathetic co-activation” (Peifer & Tan, 2021; Ullén et al., 2010). In contrast, in a low-flow condition such as ours during which challenge is higher than in the high-flow condition, increased sympathetic activation and decreased parasympathetic activation is expected. In terms of physiological measures, lower sympathetic activation in high flow than in low flow is expected to translate into decreased SCL (Boucsein, 2012) and decreased respiratory rate as well as increased respiratory depth and HF-HRV because of parasympathetic activation (Lumma et al., 2015; Wientjes, 1992), and decreased heart rate due to the sympatho-vagal balance favoring parasympathetic activation (Porges, 1995).

The literature remains mixed, however, on observing such a pattern. This co-occurrence of sympathetic and parasympathetic activation in the high-flow condition was reported by Harmat et al. (2015), who found lower respiratory depth in their lower-flow, more demanding condition as compared to high-flow condition. We note, however, that a variety of effects have been reported in this burgeoning literature—for example, decreased sympathetic with similar parasympathetic activation in the high-flow as compared to the low-flow, difficult condition (de Sampaio Barros et al., 2018), increased parasympathetic activation in the high-flow condition compared to an easy and a hard condition (Tozman et al., 2015), increased sympathetic activation in the high-flow condition compared to easy and hard low-flow conditions (Ulrich et al., 2016, 2022), or even increased sympathetic activation and decreased parasympathetic activation in the high-flow condition compared to the low-flow difficult condition (Harris et al., 2017). In the present study, we expected measures reflecting parasympathetic activation (HF-HRV and respiratory depth) to be higher in the high-flow than in the low-flow more difficult condition, and both to be lower than during the rest condition. We predicted higher SCL, heart rate, and respiratory rate in the high- and low-flow conditions compared to rest. Given the mixed results observed in the literature, we did not have any strong hypothesis regarding SCL. In addition, because heart rate and respiratory rate are both influenced by the sympathetic and parasympathetic nervous system, differential prediction between flow conditions remains uncertain. These measures, however, will give us information on the sympatho-vagal balance when participants try their best equally between a high- and a low-flow condition.

Finally, Experiment 1 instructed participants to “just enjoy the game and perform as well as possible,” which may have pressured participants to report similarly high scores on the Best-Try scale in both conditions. To address this potential confound, the participants’ instructions before the experimental games (i.e., the games after the evaluation game has been completed) were modified to make no mention of enjoyment or performance, by stating, “We will ask you to simply play a match of Unreal Tournament against four other players controlled by the computer (also called bots). After each match, we will ask you to fill out a few questionnaires.”