You are on page 1of 11

Time spent playing video games is unlikely to impact well-being

Matti Vuorre1*, Niklas Johannes1*, Kristoffer Magnusson1,2, Andrew K. Przybylski1*


1
Oxford Internet Institute, University of Oxford, 2Centre for Psychiatry Research, De-
partment of Clinical Neuroscience, Karolinska Institutet, & Stockholm Health Care Ser-
vices, Region Stockholm

Video games are a massively popular form of entertainment, socialising, cooperation, and competition.
Games’ ubiquity fuels fears that they cause poor mental health, and major health bodies and national
governments have made far-reaching policy decisions to address games’ potential risks, despite lacking
adequate supporting data. The concern-evidence mismatch underscores that we know too little about
games’ impacts on well-being. We addressed this disconnect by linking six weeks of 38,935 players’
objective game-behaviour data, provided by seven global game publishers, with three waves of their
self-reported well-being that we collected. We found little to no evidence for a causal connection be-
tween gameplay and well-being. However, results suggested that motivations play a role in players’
well-being. For good or ill, the average effects of time spent playing video games on players’ well-being
are likely very small, and further industry data are required to determine potential risks and supportive
factors to health.
Keywords: Video games; well-being; play behaviour; human motivation;

Preprint under review.

Billions of people play video games 1 and their popular- and relevant evidence to inform policymakers on whether to
ity has led public stakeholders to express concerns, but also regulate video games 8,9. However, current evidence doesn’t
hopes, about their effects on players. On the one hand, many meet these criteria and tells us little about the causal links
have sounded alarms about video games’ addictive qualities between video games and mental health 10,11.
and potential harm to players’ well-being, which has led to Why is current evidence inadequate? On the theoretical
far-reaching and widely contested health policy decisions 2,3. level, there are vastly different approaches to investigating
On the other hand, video games may help players relax and the effects of games. Researchers studying the psychology
recover 4–6; they even may serve as a treatment to some psy- of play highlight that games can help players try out differ-
chological disorders 7. Consequently, games have the poten- ent social roles, experience power in a safe environment, or
tial to affect well-being on a global scale, for good or ill. It experience a state of flow 12–16. Other approaches relying on
is therefore critical that researchers provide robust, credible, social learning have focused on how seeing violent behav-
iour rewarded can lead to real-world aggression 17—alt-
hough more recent studies found little to no link between
*MV, NJ, & AKP declare equal contribution to this work. game play and aggression 18–20. Earlier research traditions,
This research was supported by the Huo Family Foundation. The such as work focused on television, invoke the displacement
raw data and annotated analysis code supporting this work are hypothesis to explain the effects of media, taking for granted
available at https://osf.io/fb38n/. A public preprint of this work is the idea that engaging with media displaces face-to-face in-
available at https://psyarxiv.com/8cxyh. The authors declare no
conflicts of interest. The funder had no role in study design, data
teraction, leading to lower well-being 21,22. Yet again, recent
collection and analysis, decision to publish, or preparation of the studies find little to no evidence for the hypothesis nor that
manuscript. The industry partners reviewed the study design and newer forms of interactive media have larger effects over
assisted with play behaviour data collection and recruitment, but time 23–25. In contrast, research based on theories of human
had no role in collection of participant survey data, data analysis,
decision to publish, or preparing the manuscript. motivation indicates the reasons why people use technology
(i.e., quality of behaviour) is more relevant to their well-be-
Correspondence concerning this article should be addressed to ing than the amount of time they spend engaged (i.e., quan-
Andrew K. Przybylski, 1 St Giles, Oxford OX1 3JS, United King-
tity of behaviour). The extent to which play is enjoyable and
dom. E-mail: andy.przybylski@oii.ox.ac.uk.
intrinsically motivated rather than extrinsically motivated is
Preprint under review 2

a consistent predictor of players’ well-being 6,26–30. Taken to- Methods


gether, these different approaches to studying games, and
mixed findings therein, have made it challenging to deliver Participants and procedure
actionable evidence to health policymakers 9. We collaborated with game publishers who recruited
Compounding these theoretical challenges, there are four players with emails to participate in a three-wave panel
notable methodological weaknesses present in the evidence study. Seven publishers participated with the following
base. First, in typical studies, participants complete experi- games: Animal Crossing: New Horizons (Nintendo of Amer-
ments designed to imitate play 19. Such experimental manip- ica; N = 13,646), Apex Legends (Electronic Arts; N = 1,158),
ulations of game play facilitate causal inference but don’t Eve Online (CCP Games; N = 905), Forza Horizon 4 (Mi-
accurately capture play as it occurs naturally 31. Second, nat- crosoft; N = 1,981), Gran Turismo Sport (Sony Interactive
uralistic studies of game play have almost exclusively relied Entertainment; N = 19,258), Outriders (Square Enix; N =
on retrospective self-reports of play, which are biased and 1,530), and The Crew 2 (Ubisoft; N = 457). The emails tar-
inaccurate indicators of actual behaviour 32–35. Using self-re- geted the general English-speaking player bases of these
ports has been necessary because, until recently, video game publishers in Australia, Canada, India, Ireland, New Zea-
companies have declined to share objective game play data land, South Africa, United Kingdom, and the United States.
to study player well-being. Third, even the few studies that Publishers invited active players of the selected game to par-
had access to accurate data were cross-sectional and didn’t ticipate. Active play was defined as having played the re-
adequately address potential causal effects of games on spective game in the past 2 weeks to 2 months; variability in
well-being 26. Fourth, investigations have typically been lim- this interval between publishers was due to differences in
ited to a handful of specific kinds of games, rendering con- how many players regularly played a given game, so that an
clusions regarding general video game play tentative at best adequate sample could be invited.
36
. We didn’t plan to recruit a specific number of partici-
The present study addresses these limitations by a) stud- pants, but instead aimed to collect the largest possible sam-
ying naturalistic play and not artificial experiments; b) col- ple of players from each participating game publisher, given
laborating with game publishers to obtain accurate measures their player base sizes and capabilities of contacting them.
of objective behaviour rather than relying on self-reports; c) The invitation email campaigns were carried out from
collecting three-waves of longitudinal data to go beyond May to September in 2021, with different publishers inviting
cross-sectional correlations; and d) examining a variety of between 30,000 and 1,729,627 players. The emails had short
different games rather than focusing on a single title, genre, invitations to the study and contained a link to our survey
or type of player. Going beyond these limitations we adopted hosted at our institution. Upon entering the survey, partici-
three goals for this study. Our primary objective was to esti- pants first read information on the purposes and procedures
mate the causal effects of video game play on well-being: of the study, were given the option to consent, and then pro-
the extent to which the quantity of play—the average daily ceeded to the survey measures. We counted anyone who
hours played over a two-week period—impacts players’ consented to the study and responded to at least one survey
well-being. Our secondary objective was to estimate the ex- item as a participant, but participants could leave responses
tent to which well-being affects subsequent video game blank or quit or withdraw at any time. Therefore, the sample
play, because reciprocal effects—that well-being affects sizes per analysis vary because of this variable missingness.
play—are likely 23,37. As a third objective, we followed re- The publishers then recontacted all players who partici-
search emphasizing subjective experiences of players 30, and pated at the first wave two and four weeks later with invita-
investigated how the motivational dimensions of play re- tions to participate in the following waves of the study. Be-
lated to shifts in well-being. cause of the technical implementations of the email cam-
To address these objectives, we relied on a sample of paigns and late responding, the intervals between responses
38,935 players of seven popular video game titles, who were were not exactly two weeks, but instead varied slightly (IQR
invited to respond to three surveys about their well-being = [13.7, 15.9 days]). We didn’t implement a response win-
and motivational experiences over a span of six weeks. We dow, but participants could complete the surveys at any time
collaborated with the games’ publishers to get behavioural suitable for them.
data on the participating players’ play. Merging the longitu- Initial response rates ranged from 0.1% to 3.0%, in line
dinal survey responses with players’ behavioural data al- with a previous study that used similar recruitment methods
lowed us to estimate the potential causal impacts of video 26
, but retention rates were greater for the following waves
games on well-being over time.
Preprint under review 3

(21.2% to 92.1% between games and waves). At the conclu- Well-being


sion of the study, the publishers sent the behavioural data of
In line with a previous study 26, we queried for partici-
all consented participants to our team for joining with survey
pants’ affective well-being with the scale of positive and
data for analysis. At no point did we collect personally iden-
negative experiences (SPANE) 38. Participants reflected on
tifiable data or risked participants’ privacy. Instead, the sur-
how they had been feeling in the past two weeks, and re-
vey URL captured a hashed ID per player. The game pub-
ported how often they experienced six each of positive (e.g.,
lishers created those hashed IDs for each player account. We
“Pleasant) and negative feelings (e.g. “Unpleasant”) on a
used these hashed, anonymous IDs to connect each player’s
scale from 1 (Very rarely or never) to 7 (Very often or al-
survey responses to their behavioural data.
ways). As recommended, we subtracted the mean of nega-
We were interested in studying active players, defined as
tive items from the mean of positive item to form a global
playing during the study period. Of the 108,880 consented
measure of affective well-being 38.
players, 85% responded to at least one well-being survey
We also measured participants’ general life satisfaction
item; of those, 42% had played during the study, leading to
with the Cantril self-anchoring scale. This one-item measure
a final sample of 38,935 active players. Of these players,
asked participants to “Please imagine a ladder with steps
77% identified as male, 21% as female, 1.8% as third or non-
numbered from 0 at the bottom to 10 at the top. The top of
binary gender, with other responses missing or declined. The
the ladder represents the best possible life for you, and the
participants’ median age was 34 years (interquartile range
bottom of the ladder represents the worst possible life for
(IQR) = [25, 42]) and they had a reported median 23 years
you. On which step of the ladder would you say you person-
(IQR = [16, 30]) of experience playing games.
ally feel you stood over the past two weeks?”. Participants
The study procedures were granted ethical approval by
responded on a scale from 0 to 10 39,40.
our institute’s Central University Research Ethics Commit-
tee (SSH_OII_CIA_21_011). All participants provided in- Motivational experiences
formed consent to participate in the study and to have their We also measured players’ experiences and motivations
gameplay behaviour data provided by the respective game with the player experience and need satisfaction scale
publisher and reported being 18 years or older. (PENS) 30,41. This scale asked participants to reflect on their
Measures past two weeks of playing the game title, and reporting on
their experienced sense of autonomy, competence, related-
Video game behaviour
ness (only when they reported having played with others),
Participating publishers provided us with each partici- intrinsic motivation, and extrinsic motivation on subscales
pant’s game play data from two weeks prior to the first wave with 3-4 items. Participants responded on a scale from 1
up to the third wave. Different publishers had access to data (Strongly disagree) to 7 (Strongly agree). For the analyses
at different levels of detail, but all of them provided the start reported below, we created indicators for intrinsic and ex-
and end times (i.e., durations) of each game session. How- trinsic motivations by taking the means of the respective
ever, due to differences in the games and logging systems, subscales.
each game’s session start and end times were measured In addition, participants were asked to estimate their time
slightly differently because different events could count as spent playing the specific title in the past two weeks in hours
session start and end indicators. A participant could have and minutes.
multiple game sessions over the two-week period for each
Data, code, and materials
wave. We therefore added their session durations to obtain
total game time in the two weeks preceding each survey The data, annotated code required to process and analyse
wave. In doing so we excluded all sessions with durations them, supplementary analyses, and survey materials are
below 0 or above 10 hours, as extreme values can sometimes available at https://osf.io/fb38n/ (henceforth referred to as
happen due to logging errors. In the analyses below, we used Online Supplementary Materials, OSM). That page also in-
the average number of hours per day in the two-week period. cludes details on the number of invitations and responses,
Some games included additional behavioural data, such data processing such as how session durations were cleaned,
as information on in-game achievement and social aspects, exact item wordings, complete surveys, and additional data
but we don’t report on those variables here. not analysed here.
Preprint under review 4

Results weeks prior to the period addressed by the survey questions


(e.g., game play during weeks 0 – 2 affecting well-being at
In the analyses reported here, the sample sizes vary be- weeks 2 – 4). Scatterplots of data representing these lagged
tween analyses/variables due to differential data missing- associations are shown in Figure 2.
ness. We used R for all data analyses 42. Below, numbers in To best estimate this effect over time, we specified a bi-
square brackets indicate 95% credibility intervals. variate random intercepts cross-lagged panel model using
The amount of time played was variable across the average hours played per day and the well-being scores as
games and showed a small decrease on average over the the target variables (RICLPM 44,45). We determined this
three waves, which was likely attributable to differential at- model as most appropriate because it separates stable traits
trition (Figure 1, top panels; bwave = -0.09, [-0.15, -0.04]). and their correlations from within-person differences over
Players’ affective well-being decreased slightly (bwave = - time 46 and attempts to adjust for time-varying impacts on
0.05, [-0.10, -0.01]) and life satisfaction slightly increased game play and well-being by including their covariances.
(bwave = 0.05, [0.03, 0.08]) over the six weeks of the study. We estimated the model separately for affect and life satis-
Players’ intrinsic (bwave = -0.18, [-0.23, -0.13]) and extrinsic faction, and for each game. We constrained the (cross-)
motivations (bwave = -0.01, [-0.03, 0.01]) also slightly de- lagged paths to equal across the two intervals because there
creased over time (Figure 1). is no a priori reason to expect their variance over this time
period. We used the lavaan R package for estimation, and
treated missingness with full information maximum likeli-
hood methods 47.

Figure 2. Scatterplots of well-being scores and hours played


at the previous wave. Points are individual participants’
Figure 1. Distributions of key variables (rows) over time and data, and lines and shades are generalised additive model re-
by game (columns). We excluded hours of play per day over gression lines with 95%CIs. The x-axis is truncated at 3
3 (2.6%) from this figure for clarity. hours per day for this figure for clarity.

Effects between play and well-being over time Our investigation wasn’t about these specific games; in-
stead, we considered them a sample from a broader universe
This study was designed to take the first steps toward
of games. Therefore, we summarised the RICLPM parame-
identifying the real-world causal impacts of game play on
ters across games with random-effects meta-analyses of the
well-being over time. We reasoned that potential effects
unstandardized cross-lagged parameters (see OSM for the
would play out in the time span of the study, potentially cu-
corresponding standardized parameters). We focused on un-
mulate over successive sessions of play, and that a two week
standardized parameters because they facilitate understand-
period would best represent the trade-off between allowing
ing magnitudes on the variables’ natural scales 48. For these
sufficient time for effects to manifest, but not long enough
meta-analyses, we assumed weakly informative prior distri-
for them to dissipate 43. Following this reasoning, we identi-
butions on the average effect (a normal distribution with
fied the following as the most plausible causal pathway: the
cross-lagged association between game play in the two
Preprint under review 5

mean 0 and standard deviation of 0.5) and its standard devi-


ation between games (a Student’s t distribution with 7 de-
grees of freedom, mean of 0 and standard deviation of 0.25).
We then focused on our first research objective: deter-
mining the extent to which game play affects well-being.
The meta-analysis of play time and affect indicated that, on
average, video game play had little to no effect on affect,
with 67% posterior probability of a positive effect (Figure 3,
top left). The 95% most likely effect sizes of a one-hour
daily increase in play on the 13-point SPANCE scale indi-
cated that the effect was not credibly different from zero:
The magnitude and associated uncertainty of this effect sug-
gests that there is little to no practical causal connection be-
tween gameplay in the preceding two weeks on current af- Figure 3. Random effects meta-analytic estimates of the un-
fect. standardized cross-lagged effects of average daily hours
The meta-analysis of play time and life satisfaction indi- played on well-being (top left: on affect; top right: on life
cated similar results: Increasing one’s play by one hour per satisfaction), and well-being on average daily play time (bot-
day would lead to a -0.05 [-0.16, 0.07] point change in the tom left: affect’s effect on play time; bottom right: life sat-
11-point life satisfaction scale (Figure 3, top right). The pos- isfaction’s effect on play time). Shaded areas indicate pa-
terior probability for a negative effect was 83%. rameters’ approximate posterior densities. Percentages be-
To put these numbers in perspective, the average range low the average parameters indicate posterior probabilities
of changes in players’ average weekly hours of play in this of the effects’ directions. The points and intervals indicate
sample was 0.48 hours. Thus, if an average player went from posterior means and 95% credibility intervals (CI; repro-
their minimum amount of play to the greatest, their affect duced in text on the right).
would change by 0.013 [-0.04, 0.08] units on the SPANE
The RICLPM included other parameters of subsidiary
scale, and their life satisfaction by -0.02 [-0.07, 0.04]: On
interest. First, the autocorrelation parameters indicated that
average, the amount of time a person spends playing games
affect and life satisfaction were modest predictors of them-
is unlikely to have a meaningfully large impact on their well-
selves (baffect[t-1] = 0.39 [0.28, 0.52]; blife satisfaction[t-1] = 0.20
being.
[0.11, 0.30]). Second, the covariances of the random inter-
We then turned to our second research objective: inves-
cepts, indicating the extent to which people who tended to
tigating the reciprocal effects of affect (Figure 3, bottom left)
play more were also more likely to report higher well-being
and life satisfaction (Figure 3, bottom right) on play. The
was significantly greater than zero only for Animal Crossing
average meta-analytic estimate across games indicated that
and Outriders, for both well-being measures, replicating our
a one-unit increase on the 13-point SPANE affect scale
previous findings and extending them to the life satisfaction
would lead to a 0.01 [-0.03, 0.04] hour per week change in
outcome26. However, it was notable that the positive corre-
play. Similarly, life satisfaction had a practically zero effect
lation was not replicated across the other game titles. Third,
on game play (0.00 [-0.01, 0.01]). This suggests little to no
the within-person gameplay-well-being covariances were
practical causal connection linking well-being to subsequent
overall not significantly different from zero. See OSM for
gameplay.
details on these parameter estimates.
In sum, analyses testing our first two objectives provided
little evidence in favour of the idea that shifts in time spent
playing games, on average, has negative or positive effects
on players’ well-being, or that well-being directly affects
how much time individuals spend playing games.
Role of motivational experiences in well-being
We then turned to our third research objective; investi-
gating the roles of motivational experiences during play in
players’ subsequent well-being. We conducted additional
RICLP models and used intrinsic and extrinsic motivation
Preprint under review 6

scores (subscale means) in place of play time in the model


with well-being scores. These data are shown in Figure 4.

Figure 5. Random effects meta-analytic estimates of the un-


standardized cross-lagged effects of intrinsic motivation
(top) and extrinsic motivation (bottom row) on well-being.
Second, extrinsic motivation impacted affect and life sat-
isfaction negatively, such that a one-unit increase in extrin-
sic motivations was associated with a -0.9 [-0.21, 0.03] unit
change in affect, and -0.10 [-0.26, 0.05] unit change in life
satisfaction (Figure 5, bottom row). Similar reciprocal ef-
fects of affect and life satisfaction on subsequent intrinsic
and extrinsic motivation were also in evidence (see OSM).
These parameters indicated smaller but positive effects of
well-being on subsequent intrinsic motivation, and negative
effects of well-being on extrinsic motivation. In contrast to
Figure 4. Scatterplots of well-being scores on intrinsic (top) the play time results, these analyses provided consistent ev-
and extrinsic (bottom) motivation in the previous wave. idence that the motivational quality of video game play was
We ran four models: Affect predicted by each motiva- reliably linked to player well-being. Play experienced as a
tion, and life satisfaction predicted by each motivation. As volitional experience was positively predictive of subse-
above, we constrained the lagged and cross-lagged coeffi- quent well-being whereas the opposite was true of engage-
cients across waves. After conducting these models sepa- ment characterised by a sense of compulsion.
rately for each game, we then aggregated the results with
random effects meta-analyses as above. The results of these Discussion
analyses are presented in Figure 5.
Given their ubiquity and prominent place in the global
First, there was an average positive effect of intrinsic
health conversation it is crucial to understand whether video
motivation on both affect and life satisfaction, with posterior
game engagement, on its own, influences psychological
probabilities in excess of 96% in favour of a positive effect
well-being. However, evidence so far has suffered from sev-
(Figure 5, top row). These effects were consistent across
eral limitations, chief among them inaccurate measurement
games in both magnitude and direction and indicated that on
and a lack of explicit and testable causal models. Here, we
average a one-unit increase on the intrinsic motivation scale
aimed to remedy these shortcomings by pairing players’ ob-
would translate to a 0.10 [0.00, 0.19] increase in affect and
jective behavioural data with self-reports of their psycholog-
0.18 [0.07, 0.30] unit increase in life satisfaction.
ical states over six weeks. Our study provides a preliminary,
yet essential link in a new chain of evidence which will help
us understand how time spent playing games influences the
health of players.
Across six weeks, seven games, and 38,935 players, we
report results demonstrating our most pronounced hopes and
fears surrounding video games may be unfounded: Overall,
Preprint under review 7

time spent playing video games showed limited if any im- tically significant remains an open question for investiga-
pacts, positive or negative, on players’ well-being. This lack tion. That said, these results indicate that games on average
of an effect held across both aspects of well-being measured do not cause great fluctuations in psychological well-being.
here: how players feel and how they evaluate their lives. For the average player in our sample the data indicated the
Similarly, well-being had little to no effect on the time peo- subjective quality of play behaviour may be more important
ple spend playing video games. The best predictors of play- than quantity.
ers’ affective states and life satisfaction seemed to be their Previous studies of player well-being relied exclusively
previous affective states and life satisfaction, respectively, on self-reported behaviour. We studied actual behaviour and
with game play contributing little if at all. found exceedingly small effects. This contrast suggests that
We conclude the effects of playing are negligible be- if a self-report measure is biased, even a small amount of
cause they are very unlikely to be large enough to be subjec- bias might inflate estimates of the association between play
tively noticed. For example, Anvari and Lakens 49 demon- and well-being. In our supplementary analyses (see OSM),
strated that the smallest perceptible difference on the we found that participants on average overestimated their
PANAS scale, a measure similar to SPANE, was 0.20 (2%) play time. This finding underscores the importance of mov-
on a 5-point Likert-scale. We found that an hour per day in- ing beyond self-reports of play—or of at least better under-
crease in play resulted in 0.03 unit increase in well-being: standing error in self-reports 33.
Assuming linearity and equidistant response categories, the This study presents a first step in explicitly stating and
average player would have to play 10 hours more per day empirically evaluating the real-world causal effects of play-
than is typical for them to notice a change in their well-be- ing video games on well-being over time. We collected ob-
ing. Similarly, the meta-analytic model indicated 99.7% servational data, requiring us to state the assumptions under
probability that the effect of a one-hour daily increase in play which causality holds 51: the absence of time-varying con-
time on affect is too small to be subjectively noticeable (life founders, no selection bias, and a correct time lag. In our
satisfaction: 99.6%). Viewed in context, the effects of game main analyses, we were unable to control for other variables
play time we observed are small indeed, and in a range sim- that could potentially create a spurious relation between
ilar to other screen-based activities, such as those estimated game time and well-being 52,53. Figure 6 depicts how an un-
in social media research 50. known time-varying confounder (U) could bias the effect of
We also studied the roles of motivational experiences play on well-being. For example, the amount of leisure time
during game play. Conceptually replicating previous cross- people have might increase both their play now and their
sectional findings 26, we found that intrinsic motivations well-being at the next wave. This would bias a truly negative
have a positive effect on well-being whereas extrinsic moti- effect of play on well-being toward the null. Confounding
vations have a negative effect. The effects of motivations may be even more complex: More leisure time could also
were larger than that of video game play and our analysis mean more boredom, which increases play time, but de-
suggests we can be confident in the direction of these moti- creases well-being, which could bias a truly positive effect
vation effects. In absolute terms, the effect of a one-point of play toward the null. Although we account for some stable
deviation from a player’s typical intrinsic motivation on af- confounding by separating between-person characteristics
fect did not reach the threshold of being subjectively notice- from within-person changes 44, within-person changes in
able (0.10 estimate vs. 0.26 threshold). However, we cannot third variables may still lead to within-person confounds.
be certain a one-point increase (out of a seven-point scale) Without theoretical and empirical identification of these
is considered a large or a small shift—participants’ average confounds, our and future studies will likely produce biased
range on the intrinsic motivation scale was 0.36. Until future estimates of the underlying causal effects.
work determines what constitutes an adequate ‘treatment’,
these conclusions regarding motivations remain tentative.
Taken as a whole, our findings suggest that amount of
play does not, on balance, displace time which would other-
wise be allocated to more rewarding activities and thus un-
dermine well-being. By the same token, our results do align
with a perspective that the motivational quality of experi-
ences during play, on their own, may benefit their well-being
30
. The extent to which such an effect generalises or is prac-
Preprint under review 8

thousands of distinct video games. In this study we were


only able to study how playing a sample of seven games in-
fluenced well-being. This limitation constrains the generali-
sability of our findings 36. To truly understand why people
play and what the impacts are, we need to study a much
broader variety of games, genres, and players. Moreover, we
analysed total game time, which is the broadest possible
measure of play behaviour. Although it is necessary to begin
at such a broad level to build an evidence base of video game
Figure 6. Illustration of a causal model where play and well- effects 9,55, future work must account for the situations, mo-
being have reciprocal effects over time. The RICLPM can tivations, and contexts in which people play 56. Although we
deal with C (time-invariant unknown covariates that affect used an appropriate and established statistical model to
play and WB [well-being]), but not an unknown time-vary- study our assumed effects, it required us to assume that play
ing variable (U) that confounds the effect of play on well- time follows a normal distribution, which is unlikely to be
being. t1 = first time point. true. Future statistical work in this area should also examine
Closely related to the question of confounding via a third different distributional assumptions 55. Finally, measuring
variable is self-selection bias. We observed low response objective behaviour with industry provided telemetry comes
rates and notable attrition over waves. It is probable that self- with its own measurement error and differences between
selection introduced bias into our estimates. The simplest publishers. Independent researchers must continue working
example of self-selection is that older and more experienced with industry partners to get a better understanding of be-
gamers could be more passionate about helping researchers. havioural data and its limitations.
It is also plausible that older and more experienced gamers Billions of people play video games, and policymakers,
have figured out a way to fit games into their lives, so that healthcare professionals, and those who make games ur-
games won’t have a negative impact on their well-being. Un- gently need to know if this form of recreation influences the
der this self-selection, a truly negative effect of play on well- well-being of players. This study provides evidence on the
being would be biased toward the null. Similarly, dropout causal impacts of play on well-being using objectively
between the waves could be caused by unobserved variables, logged gameplay behaviour. Our results provide evidence
which would violate the missing at random (MAR) assump- that the impact of time spent playing video games on well-
tion. For example, missingness could be related to lower being is too small to be subjectively noticeable and not cred-
well-being and more play during the wave for which data is ibly different from zero. Therefore, it is essential to cast a
missing. This type of missing data mechanism could mask a wider empirical and theoretical net and maximise the value
negative effect of play on well-being. To get better esti- of future studies by focusing on the qualities of diverse play
mates, we need better measures, which in turn requires sam- experiences, in-game events, and players for whom effects
pling strategies that rely less on self-selection. may vary. Until then, limiting or promoting play based on
Last, we chose a two-week time lag, extending our pre- time alone appears to bear neither benefit nor harm.
vious work 26. Such a large time frame ensured capturing
enough play for sufficient variance. However, any actual ef- Author Contributions
fect’s temporal properties may differ from our assumptions,
and be either too short-lived to be detected with our design, A.K.P.: Conceptualization, Funding acquisition, Investi-
or accumulate over longer periods of time 54. The exact tem- gation, Methodology, Project administration, Resources, Su-
poral properties of games’ impacts must be investigated in pervision, Visualization, Writing - original draft, and Writ-
future work. Taken together, the causal effects we present ing - review & editing. K.M.: Data curation, Formal analy-
depend on our stated assumptions, which future work will sis, Methodology, Software, Validation, Visualization, and
address in detail. Writing - review & editing. M.V.: Conceptualization, Data
Likewise, although we studied the play behaviour and curation, Formal analysis, Investigation, Methodology, Pro-
well-being of many thousands of people across diverse ject administration, Software, Validation, Visualization,
games, our study barely scratches the surface of video game Writing - original draft, and Writing - review & editing. N.J.:
play more broadly. Online games platforms like the Xbox Conceptualization, Data curation, Investigation, Methodol-
Network, PlayStation Network, Nintendo Switch Online, ogy, Project administration, Software, Validation, Writing -
and Steam connect hundreds of millions of players to tens of original draft, and Writing - review & editing.
Preprint under review 9

References 14. Markey, P. M., Ferguson, C. J. & Hopkins, L. I. Video


Game Play. American Journal of Play 13, 87–106
1. Newzoo. Newzoo Global Games Market Report 2021 | (2020).
Free Version. Newzoo https://newzoo.com/in- 15. Przybylski, A. K., Weinstein, N., Murayama, K.,
sights/trend-reports/newzoo-global-games-market-re- Lynch, M. F. & Ryan, R. M. The Ideal Self at Play: The
port-2021-free-version/ (2021). Appeal of Video Games That Let You Be All You Can
2. Aarseth, E. et al. Scholars’ open debate paper on the Be. Psychological Science 23, 69–76 (2012).
World Health Organization ICD-11 Gaming Disorder 16. Swinkels, L. M. J., Veling, H. & van Schie, H. T. Play-
proposal. Journal of Behavioral Addictions 6, 267–270 ing videogames is associated with reduced awareness
(2017). of bodily sensations. Computers in Human Behavior
3. Kardefelt-Winther, D. A critical account of DSM-5 cri- 125, 106953 (2021).
teria for internet gaming disorder. Addiction Research 17. Bushman, B. J. & Anderson, C. A. Violent Video
& Theory 23, 93–98 (2015). Games and Hostile Expectations: A Test of the General
4. Halbrook, Y. J., O’Donnell, A. T. & Msetfi, R. M. Aggression Model. Pers Soc Psychol Bull 28, 1679–
When and how video games can be good: A review of 1686 (2002).
the positive effects of video games on well-being. Per- 18. Drummond, A., Sauer, J. D. & Ferguson, C. J. Do lon-
spect Psychol Sci 14, 1096–1104 (2019). gitudinal studies support long-term relationships be-
5. Reinecke, L. & Hofmann, W. Slacking off or winding tween aggressive game play and youth aggressive be-
down? An experience sampling study on the drivers and haviour? A meta-analytic examination. Royal Society
consequences of media use for recovery versus procras- Open Science 7, 200373 (2020).
tination. Human Communication Research 42, 441– 19. Elson, M. & Ferguson, C. J. Twenty-five years of re-
461 (2016). search on violence in digital games and aggression:
6. Vorderer, P., Klimmt, C. & Ritterfeld, U. Enjoyment: Empirical evidence, perspectives, and a debate gone
At the Heart of Media Entertainment. Communication astray. European Psychologist 19, 33–46 (2014).
Theory 14, 388–408 (2004). 20. Ferguson, C. J., Copenhaver, A. & Markey, P. Reexam-
7. Kollins, S. H. et al. A novel digital intervention for ac- ining the Findings of the American Psychological As-
tively reducing severity of paediatric ADHD (STARS- sociation’s 2015 Task Force on Violent Media: A Meta-
ADHD): a randomised controlled trial. The Lancet Dig- Analysis. Perspect Psychol Sci 15, 1423–1443 (2020).
ital Health 2, e168–e178 (2020). 21. Masur, P. K. Digital communication effects on loneli-
8. Elson, M. et al. Do policy statements on media effects ness and life satisfaction. in Oxford Research Encyclo-
faithfully represent the science? Advances in Methods pedia of Communication (Oxford University Press,
and Practices in Psychological Science 2, 12–25 2021). doi:10.1093/acre-
(2019). fore/9780190228613.013.1129.
9. IJzerman, H. et al. Use caution when applying behav- 22. Neuman, S. B. The displacement effect: Assessing the
ioural science to policy. Nat Hum Behav 4, 1092–1094 relation between television viewing and reading perfor-
(2020). mance. Reading research quarterly 414–440 (1988).
10. Elson, M. & Przybylski, A. K. The science of technol- 23. Dienlin, T., Masur, P. K. & Trepte, S. Reinforcement
ogy and human behavior: Standards, old and new. Jour- or displacement? The reciprocity of FTF, IM, and SNS
nal of Media Psychology 29, 1–7 (2017). communication and their effects on loneliness and life
11. Rooij, A. J. van et al. A weak scientific basis for gam- satisfaction. J Comput Mediat Commun 22, 71–87
ing disorder: Let us err on the side of caution. Journal (2017).
of Behavioral Addictions 7, 1–9 (2018). 24. Przybylski, A. K. & Weinstein, N. A large-scale test of
12. Cowan, K. A Panorama of Play - A Literature Review. the Goldilocks Hypothesis : Quantifying the relations
https://digitalfuturescommission.org.uk/wp-con- between digital-screen use and the mental well-being of
tent/uploads/2020/10/A-Panorama-of-Play-A-Litera- adolescents. Psychological Science 28, 204–215
ture-Review.pdf (2020). (2017).
13. Granic, I., Lobel, A. & Engels, R. C. M. E. The benefits 25. Vuorre, M., Orben, A. & Przybylski, A. K. There is no
of playing video games. American Psychologist 69, 66– evidence that associations between adolescents’ digital
78 (2014). technology engagement and mental health problems
Preprint under review 10

have increased. Clinical Psychological Science 9, 823– 38. Diener, E. et al. New well-being measures: Short scales
835 (2021). to assess flourishing and positive and negative feelings.
26. Johannes, N., Vuorre, M. & Przybylski, A. K. Video Social Indicators Research 97, 143–156 (2010).
game play is positively correlated with well-being. 39. Cantril, H. Pattern of human concerns. (Rutgers Uni-
Royal Society Open Science 8, 202049 (2021). versity Press, 1965).
27. Pine, R., Fleming, T., McCallum, S. & Sutcliffe, K. The 40. Kapteyn, A., Lee, J., Tassot, C., Vonkova, H. & Za-
Effects of Casual Videogames on Anxiety, Depression, marro, G. Dimensions of subjective well-being. Soc In-
Stress, and Low Mood: A Systematic Review. Games dic Res 123, 625–660 (2015).
for Health Journal 9, 255–264 (2020). 41. Johnson, D., Gardner, M. J. & Perry, R. Validation of
28. Przybylski, A. K. & Weinstein, N. Investigating the two game experience scales: The Player Experience of
Motivational and Psychosocial Dynamics of Dysregu- Need Satisfaction (PENS) and Game Experience Ques-
lated Gaming: Evidence From a Preregistered Cohort tionnaire (GEQ). International Journal of Human-
Study. Clinical Psychological Science 7, 1257–1265 Computer Studies 118, 38–46 (2018).
(2019). 42. R Core Team. R: A Language and Environment for Sta-
29. Ryan, R. M. & Deci, E. L. Self-Determination Theory tistical Computing. (R Foundation for Statistical Com-
and the Facilitation of Intrinsic Motivation, Social De- puting, 2021).
velopment, and Well-Being. American Psychologist 43. Sauer, J. D. & Drummond, A. Boundary conditions for
55, 68–78 (2000). the practical importance of small effects in long runs: A
30. Ryan, R. M., Rigby, C. S. & Przybylski, A. K. The mo- comment on Funder and Ozer (2019). Advances in
tivational pull of video games: A self-determination Methods and Practices in Psychological Science 3,
theory approach. Motivation and Emotion 30, 347–363 502–504 (2020).
(2006). 44. Hamaker, E. L., Kuiper, R. M. & Grasman, R. P. P. P.
31. Sherry, J. L. Violent Video Games and Aggression: A critique of the cross-lagged panel model. Psycholog-
Why Can’t We Find Effects? in Mass media effects re- ical Methods 20, 102–116 (2015).
search: Advances through meta-analysis 245–262 45. Mulder, J. D. & Hamaker, E. L. Three Extensions of the
(Lawrence Erlbaum Associates Publishers, 2007). Random Intercept Cross-Lagged Panel Model. Struc-
32. Andrews, S., Ellis, D. A., Shaw, H. & Piwek, L. Beyond tural Equation Modeling: A Multidisciplinary Journal
Self-Report: Tools to Compare Estimated and Real- 0, 1–11 (2020).
World Smartphone Use. PLOS ONE 10, e0139004 46. Hamaker, E. L. Why researchers should think ‘within-
(2015). person’: A paradigmatic rationale. in Handbook of re-
33. Kahn, A. S., Ratan, R. & Williams, D. Why We Distort search methods for studying daily life (eds. Mehl, M. R.
in Self-Report: Predictors of Self-Report Errors in & Conner, T. S.) 43–61 (Guilford Press, 2012).
Video Game Play*. Journal of Computer-Mediated 47. Rosseel, Y. lavaan: An R Package for Structural Equa-
Communication 19, 1010–1023 (2014). tion Modeling. Journal of Statistical Software 48, 1–36
34. Parry, D. A. et al. A systematic review and meta-anal- (2012).
ysis of discrepancies between logged and self-reported 48. Baguley, T. Standardized or simple effect size: What
digital media use. Nature Human Behaviour 1–13 should be reported? British Journal of Psychology 100,
(2021) doi:10.1038/s41562-021-01117-5. 603–617 (2009).
35. Thompson, J. J., McColeman, C. M., Stepanova, E. R. 49. Anvari, F. & Lakens, D. Using anchor-based methods
& Blair, M. R. Using Video Game Telemetry Data to to determine the smallest effect size of interest. Journal
Research Motor Chunking, Action Latencies, and of Experimental Social Psychology 96, 104159 (2021).
Complex Cognitive-Motor Skill Learning. Topics in 50. Orben, A. & Przybylski, A. K. The association between
Cognitive Science 9, 467–484 (2017). adolescent well-being and digital technology use. Na-
36. Yarkoni, T. The Generalizability Crisis. ture Human Behaviour 3, 173–182 (2019).
https://osf.io/jqw35 (2019) doi:10.31234/osf.io/jqw35. 51. Hernán, M. A. The C-word: Scientific euphemisms do
37. Houghton, S. et al. Reciprocal relationships between not improve causal inference from observational data.
trajectories of depressive symptoms and screen media Am J Public Health 108, 616–619 (2018).
sse during adolescence. J Youth Adolescence 47, 2453– 52. Rohrer, J. M. Thinking clearly about correlations and
2467 (2018). causation: Graphical causal models for observational
Preprint under review 11

data. Advances in Methods and Practices in Psycholog-


ical Science 1, 27–42 (2018).
53. VanderWeele, T. J., Jackson, J. W. & Li, S. Causal in-
ference and longitudinal data: a case study of religion
and mental health. Soc Psychiatry Psychiatr Epidemiol
51, 1457–1466 (2016).
54. Dormann, C. & Griffin, M. A. Optimal time lags in
panel studies. Psychological Methods 20, 489–505
(2015).
55. Johannes, N., Dienlin, T., Bakhshi, H. & Przybylski, A.
K. No effect of different types of media on well-being.
(2021) doi:10.31234/osf.io/zgb5y.
56. Meier, A. & Reinecke, L. Computer-mediated commu-
nication, social media, and mental health: A conceptual
and empirical meta-review. Communication Research
1–28 (2020) doi:10.1177/0093650220958224.

You might also like