Habitual video gaming predicts multitasking performance while the role of cognitive capacity remains inconclusive

Abstract

Video gaming has been shown to improve information processing. Enhanced attention control functions appear to illustrate cognitive substrates thereof. Additionally, attention control functions represent constraints of cognitive capacity—an established predictor of multitasking performance. Thus, cognitive capacity may explain an association between video gaming and enhanced multitasking performance via attention control functions. To investigate this, we assessed the short-term memory and multi-attribute task battery (MATB) task performance of 60 individuals with different levels of video gaming experience; and conducted structural equation modeling to test if video gaming experience predicted multitasking performance; and if this effect was mediated by cognitive capacity. We used two functions of the theory of visual attention computational modeling framework as indicators of cognitive capacity; four MATB performance measures as indictors of multitasking performance; and self-reports on video game play time and expertise to model video gaming experience. Video gaming experience predicted multitasking performance but we found no evidence for cognitive capacity mediating this relationship. Thus, the role of cognitive capacity in the association between video gaming experience and multitasking is inconclusive.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-25260-5.

Video gaming represents a popular leisure activity. Approximately 61% of the American population regularly play video games for at least one hour per week¹. Opinions about the impact of video gaming on the human mind are polarizing ranging from concerns about video games causing addiction, health issues, and antisocial behavior^2,3 to appreciating them for eliciting joy, challenging the intellect, and enabling to socialize^1,4,5. Aside this, video gaming has been related to alterations in cognitive performance, for instance, enhanced (visuospatial) information, attention, and working memory processing^6–8. Early research on video gaming found that these effects were associated with elevated attention functions predominantly related to playing action video games^9–11; and generalize to environments beyond video gaming (far transfer): for instance, video gaming has been shown to improve flight (simulation task) performance and surgical proficiency^12–16. Thus, video gaming poses a promising tool of cognitive enhancement.

Likewise, video gaming effects are notorious for being unreliable—an issue probably related to inconsistent procedures of operationalizing video gaming effects and publication bias^17–20. Video gaming effects are commonly operationalized based on performance differences between action video game players and control individuals (playing either no/little or other video games than action video games). In this context, video game player types are determined based on information on how prevalently video games of certain genres within specific time frames had been played: for instance, individuals may be classified as action video game players if they had played action video games for at least 5 h per week during the past year; as non-video game players if they had played no or hardly any other video games than action video games during the past year; and as non-action video game players if they had regularly played other video games than action video games during the past year¹⁸. However, these reference values are artificial (despite being data-driven) and inconsistently used in research^6,18,20,21. Furthermore, concerns were raised that meta-analytic evidence in support of this approach may be biased¹⁹.

Aside this, for this approach to produce reliable results, action video games would need to be “construct pure”, meaning that all action video games should exhibit the same clearly defined game play characteristics distinguishing them from video games of other video game genres. However, this is not the case: for instance, there are many action video game subgenres emphasizing different (action) video game play characteristics, e.g., first-/third-person shooter vs. action role play adventure games. On top of that, some video game genres share game play characteristics with action video games, e.g., multi-player online battle arena, real-time strategy, and sports/driving video games, which may be an explanation for why playing these video games has been observed to go along with similar alterations in cognitive processing as with playing action video games^22–26. While this line of argumentation highlights action video games as superior in fostering cognitive improvements, it also illustrates that it is not entirely clear yet which action video game play characteristics exactly contribute to cognitive alterations. Given that the video gaming industry is rapidly evolving and thereby producing even more subgenres sharing game play characteristics²¹, established approaches of investigating video gaming effects may not be sufficient anymore for researching video gaming effects, which raises the question whether a paradigm shift is required in future research on video gaming effects.

Another issue concerns the lack of a conclusive model of an algorithmic solution as to how video gaming effects are elicited. There is a large body of research implying that enhanced top-down (distractor) suppression and/or target enhancement may account for improved information processing in (action) video game players^27–30. But computational research based on the theory of visual attention (TVA)^31,32 showed that video game players rather display enhanced speed of information processing (i.e., the number of elements that can be encoded per second) than active attention top-down control (e.g., target enhancement) and visuospatial attention deployment^33–36. This inconsistency highlights that research might focus too strongly on attention control functions and thereby limit itself. Video game players may employ different cognitive strategies for task execution relying more or less on attention control functions depending on which task they are supposed to perform. In support of this, it has been shown that action video game players outperformed non-video game players in an attention demanding task despite conserving attention resources as indicated by a smaller activation of an attention brain network^29,37.

A similar strategy can be observed when individuals are supposed to handle extreme task load requirements in multitasking conditions, e.g., by focusing on one task at the cost of other tasks as opposed to by shifting attention between tasks at the cost of peak performance in either one task (i.e., the stability-flexibility dilemma)³⁸. These strategies are characterized by cognitive control states relying on different cognitive functions. Maintaining cognitive control is exhausting – an effect that has been debated to relate to cognitive control functions drawing resources from a limited cognitive capacity³⁹. The exact nature of said cognitive capacity is debated but short-term/working memory and attention and executive functions have been frequently discussed as substrates thereof^40,41. Therefore, we suggest that cognitive capacity illustrates a more flexible predictor of complex performance by encompassing (not relying) on attention control functions (in line with task requirements).

TVA computational modeling illustrates a promising approach to operationalize cognitive capacity in this context: it was shown that multitasking decrements relate to reduced TVA speed of information processing and short-term memory capacity – with alterations in short-term memory capacity interacting with task difficulty and age^42,43. Thus, TVA speed of information processing and potentially short-term memory capacity illustrate substrates of a cognitive capacity. Given that, TVA speed of information processing was observed to be enhanced in both action and non-action video game players^33–35, TVA speed of information processing might illustrate a vital resource of cognitive capacity explaining why video game players display an enhanced performance in complex task conditions, e.g., flight (simulation), surgery, and other multitasking environments^12,13,16,26.

To investigate this, we recruited 60 individuals to undergo cognitive and multitasking assessments. All individuals (except one) had played video games on a regular basis. We hypothesized that (H1) video gaming experience would predict multitasking performance, and that (H2) this effect would be mediated by TVA cognitive capacity. We tested these hypotheses using structural equation modeling (SEM). Video gaming experience was operationalized as a latent variable based on the sum of self-reported average video gaming play times of 7 video game genre categories per week in the past year, the sum of self-reported average video gaming play times of said 7 video game genre categories per week prior to that past year, and the sum of self-reported expertise scores on said 7 video game genre categories acquired through an established questionnaire (which can be acquired from the Supplementary Material provided by¹⁸⁾. The 7 video game genre categories comprised: (1) action first/third person shooter, (2) action role-play (adventure), sports, and driving, (3) real-time strategy/multi-player online battle arena, (4) non-action turn-based role-play and fantasy, (5) turn-based strategy, life simulation, and puzzle, (6) music, and (7) other video games. This questionnaire may also be used to subdivide individuals into action-video game players, non-action video game players, and non-video game players (information on the classification scheme can be inferred from the Supplementary Material provided by¹⁸ as well). But we did not use it for this purpose. We acknowledge that the likelihood for finding video gaming effects may be diminished because of this as our hypotheses were based on research on foremost action video game playing. But we feel that subgrouping individuals into video game player types may not illustrate a reasonable approach anymore as video game player classification is artificial and requirements for applying this approach are not (sufficiently) given.

Multitasking performance, in turn, was operationalized by means of four performance measures derived from the open-source version of the multi-attribute task battery (openMATB)⁴⁴. The openMATB constitutes a low-fidelity flight simulator. Hereby, performance is operationalized as a function of the number of misses in a system monitoring and a communication task; and the deviation from optimal levels in a resource management and a tracking task, respectively. Using OpenMATB task performance to operationalize multitasking performance for our research purpose was particularly promising as (action) video gaming has already been shown to improve OpenMATB proficiency¹².

Cognitive capacity was fit using TVA speed of information processing (C) in Hz and short-term/working memory capacity (K). These were modeled from accuracy performance in a visual short-term memory task using a TVA computational modeling framework⁴⁵. This approach has already been successfully related to video gaming effects – both in action and non-action video game players³⁵. We used both TVA parameters as variables to fit a latent construct as there are indications that both parameters may serve as resources of cognitive capacity^42,43. Furthermore, TVA C and K parameter values are known to correlate⁴⁶. Thus, both parameters were likely to fit a construct pure latent variable. However, video gaming has also been shown to impact exclusively on TVA C but not K parameter values^33–35, while (action) video game players frequently displayed enhanced short-term memory task performance^8,47–49. Therefore, we also tested if TVA C or K parameter values individually may serve as cognitive capacity variable to explain the relationship between video gaming experience and multitasking performance. Additionally, we conducted a series of control analyses testing if military experience/flight experience⁵⁰, age and/or gender^51,52, and/or media multitasking^53,54 affected any of these measures. Based on our literature review^12,33,34, we expected 60 individuals as sufficient to find the anticipated effects. Nevertheless, we estimated effects based on 5000 Bootstrapping samples and conducted post-hoc power analyses to control for reliability.

Results

Hypotheses testing

The data fit well to the anticipated models irrespective of whether both C and K values were used to fit one cognitive capacity construct or either one of the TVA parameters, see Table 1. However, there were only significant associations between video gaming experience and multitasking performance but not between video gaming experience and cognitive capacity nor between cognitive capacity and multitasking performance; and thereby no significant indirect effects indicating that TVA cognitive capacity may explain the association between video gaming experience and multitasking performance, see Table 2.

Table 1.

Model comparisons between models with different mediator variables (Cognitive Capacity as latent construct based on C and K or either C or K, Military Experience (Military Service Duration in years), Age, and Media Multitasking) of the relationship between Video Gaming Experience and Multitasking Performance.

Model	N	N Bootstrapping	χ2	p	CFI	TLI	RMSEA	SRMR	BIC
No Mediator	59	5000	14.21	0.438	0.98	0.99	0.04	0.06	1043.98
Cognitive Capacity (C and K)	59	5000	26.02	0.421	0.98	0.99	0.04	0.07	1381.20
C	59	5000	19.76	0.42	0.98	0.99	0.04	0.06	1217.43
K	59	5000	19.76	0.425	0.98	0.99	0.04	0.06	1216.51
Military Experience	59	5000	19.77	0.425	0.98	0.99	0.04	0.06	1206.81
Age	59	5000	19.75	0.426	0.98	0.99	0.04	0.06	1216.96
Media Multitasking	59	5000	19.76	0.426	0.98	0.99	0.04	0.06	1213.18

Estimates were acquired using Bootstrapping with 5000 samples; CFI: comparative fit index; TLI: Tucker-Lewis index; RMSEA: approximated root mean square error; SRMR: standardized root mean square residual; BIC: Bayesian Information Criterion. C: theory of visual attention speed of information processing; K: theory of visual attention short-term memory capacity.

Table 2.

Structural equation modeling parameter estimates.

Model	Btotal	Bdirect	Ba	Bb	Bindirect
Cognitive Capacity (C and K)	– 0.51 [– 0.90, – 0.14]	– 0.55 [– 1.08, – 0.08]	0.26 [– 0.13, 0.72]	0.15 [– 0.73, 1.16]	0.04 [– 0.19, 0.30]
C	– 0.53 [– 0.92, – 0.15]	– 0.54 [– 0.94, – 0.16]	0.22 [– 0.24, 0.68]	0.05 [– 0.11, 0.21]	0.01 [– 0.05, 0.08]
K	– 0.51 [– 0.89, – 0.14]	– 0.53 [– 0.93, – 0.16]	0.25 [– 0.20, 0.72]	0.08 [– 0.08, 0.24]	0.02 [– 0.05, 0.09]

C: theory of visual attention speed of information processing; K: theory of visual attention short-term memory capacity. total/direct: association between video gaming experience and multitasking performance; a: association between video gaming experience and cognitive capacity as a function of both C and K or either C or K; b: association between cognitive capacity and multitasking performance; indirect: a x b.

Control analyses

There were relationships between video gaming experience and military experience (military service duration in years) (B = 0.72, 95% CI [0.26,1.23]) and media multitasking and multitasking performance (B = 0.17, 95% CI [0.004,0.33]). But these effects likely played a minor role, given that a model with video gaming experience as the only predictor of multitasking performance represented the best model, see Table 1. Hereby, video gaming experience had a positive impact on multitasking performance (as multitasking decrements were reduced) (B = -0.52, 95% CI [– 0.92, – 0.15]). For an outline of the model and all parameter estimates, see Fig. 1. This effect was independent of gender. Also, there was no indication that individuals with flight experience demonstrated a superior performance in any of the performance measures.

Fig. 1.

Model with video gaming experience as a predictor of multitasking performance (best model). Multi-Attribute Task Battery performance measures served as indicators of Multitasking Performance. Communication and System Monitoring performance were operationalized based on the average number of misses, and Tracking and Resource Management performance based on the average root-mean squared error and average absolute difference values, respectively. Thus, Multitasking Performance was better the smaller the respective measures. Expertise and Play Time served as indicators of Video Gaming Experience. All measures were z-standardized. Numerical values illustrate un-standardized and bootstrapped estimates. Values in brackets indicate 95% confidence intervals. Latent variables are displayed as circles, manifest variables as squares.

Post-hoc statistical power analysis

The association between video gaming experience and multitasking performance (model without mediator) was found with a statistical power of 94.54%. In fact, data of 31 individuals would have been enough to observe the effect at 80% statistical power. Given the unexpectedly small effect sizes whose direction was also partially inconsistent with the literature our hypotheses were based on, a much larger sample would have been required to detect a relationship between video gaming experience and cognitive capacity and between cognitive capacity and multitasking at 80% statistical power (see Table 3). Thus, our mediation analysis was not of sufficient statistical power (see Table 3).

Table 3.

Post-hoc power analysis results.

	Btotal		Bdirect		Ba		Bb		Bindirect
Model	Power N = 59	N 80%	Power N = 59	N 80%	Power N = 59	N 80%	Power N = 59	N 80%	Power N = 59	N 80%
Cognitive Capacity (C and K)	92.33	38	92.33	38	24.16	319	12.09	726	4.64	817
C	95.50	20*	95.50	20*	25.37	477	22.11	1457	7.19	1829
K	91.46	34	95.50	20*	18.55	313	20.23	518	3.45	738

N_80%: sample required for 80% statistical power; C: theory of visual attention computational modeling speed of information processing; K: theory of visual attention short-term memory capacity. total/direct: association between video gaming experience and multitasking performance; a: association between video gaming experience and cognitive capacity as a function of both C and K or either C or K; b: association between cognitive capacity and multitasking performance; indirect: a x b; *simulations started at N = 20 with power > 80%.

Discussion

In conclusion, our data supported H1 according to which gaming experience predicts multitasking performance. In contrast, our results did not align with H2 according to which this association was mediated by TVA cognitive capacity: we did not find a significant relationship between video gaming experience and any TVA cognitive capacity variable nor between any TVA cognitive capacity variable and multitasking performance. The effect of video gaming experience on multitasking performance was of sufficient statistical power; and neither military experience, age, gender, or flight experience explained the relationship between video gaming experience and multitasking performance better. In the following, we will discuss if and how our research outcomes may relate to our previous discussion on different approaches of operationalizing video gaming effects. Furthermore, we will address the role of cognitive capacity in video gaming and multitasking research.

It is remarkable that we were able to replicate an association between video gaming (experience) and multitasking performance irrespective of a video game player type classification, while the original effect our hypothesis relied on was specifically related to action video gaming¹². This may be related to video game genres sharing game play characteristics with action video games and hence similarly impacting on cognitive processing^21–25. However, such a conclusion may be premature, given that we were not able to find a relationship between TVA speed of information processing and video gaming – even though this association had been shown in both action and non-action video game players with different levels of gaming experience^33–35. One reason for this may be statistical power. Video gaming effects are commonly larger in cohort than training studies as a result of (extreme) differences in group characteristics⁶. As we did not consider such group differences, our effects may have been diminished (as discussed above). In support of this, enhanced TVA speed of information processing had been found predominantly when video game player groups had been compared but was reduced to a visuospatial field benefit when investigated as a function of action video game play training³⁴. Thus, our research outcome may indicate that video game player subgrouping may be beneficial for investigating small to moderate video gaming effects (but not required for investigating large effects).

The null-effects regarding the associations between video gaming experience and cognitive capacity and cognitive capacity and multitasking performance contradict our claim that cognitive capacity may be a substitute to attention control when explaining the impact of video game play on cognition. We argued that attention control may not be sufficient to explain video gaming effects when attention control functions play a minor role in task execution. In support of this, (action) video game players and action video game trainings have been shown to go along with improved performance in paradigms requiring predominantly attention deployment while such an effect was notably less reliable while performing tasks requiring other cognitive operations^6,7,19–21. Furthermore, TVA research showed that (action) video gaming impacted on other cognitive function than top-down attention control^33–35. Additionally, it was demonstrated that enhanced TVA speed of information processing as observed in (action) video game players could not be artificially elicited using non-invasive brain stimulation although visuospatial attention deployment was altered thereof^35,36.

With this being said, the question remains which of these factors may explain how video gaming experience may impact on multitasking performance. We speculate that the participants in our study likely developed strategies based on try and error to successfully perform the openMATB: for instance, by shifting attention between tasks as opposed to focusing on one task³⁸. Indeed, openMATB performance has been shown to be strongly affected by such strategic behavior—even in collaborative settings^55–57. In contrast, the participants would not have benefitted from such a strategy while performing the short-term memory task³⁵. There, the participants had to memorize as many white shapes presented on an invisible circular display as possible while memory displays had been presented at very short exposure durations (between 16.7 ms and 200.4 ms). Attention shifting would have impeded encoding of all shapes and decreased performance. This dissimilarity in task requirements may also explain why we did not find a significant association between TVA cognitive capacity variables and multitasking performance even though multitasking decrements have been related to reduced TVA C and K parameters in previous literature^42,43.

Conclusion

In conclusion, video gaming experience may predict multitasking performance—even irrespective of gaming preferences, e.g., action video gaming. There was no evidence for cognitive capacity explaining this effect. Future research may test if this null-effect was related to an insufficient sample size or predominantly attention control functions accounting for the effect.

Methods

Sample

One of the 60 acquired data sets got lost due to technical issues. Thus, 59 data sets were used for statistical analyses. We conducted post-hoc power analyses to determine the reliability of our effects. The participants’ sex matched their gender – 45 individuals identified as male and 14 as female. They were between 18 and 29 years old. All individuals (except one) had played video games on a regular basis. All except for two individuals were officer cadets who had served for at least one year in the military. Thirteen individuals had flight experience (that is at least 4 and max. 1600 h of flight practice)—five of them possessed a license. The participants did not receive a monetary compensation but students could acquire course credit. The participants provided written informed consent prior to participation. The experimental design of this study was approved by the local Ethics Board (EK UniBw M 23–53). We confirm that all methods were performed in accordance with relevant guidelines and regulations. For a more detailed description of the sample characteristics, see Table 4.

Table 4.

Sample characteristics.

Variable	N	M	SD	Range	M Bayes	95% CIBayes
Age (Years)	59	22.93	2.58	[18, 29]	23.09	[22.3, 23.88]
Military Service Duration (Years)	57	3	1.5	[1, 7]	3.6	[2.91, 4.3]
Sum of Avg. Video Game Play Time (Hours/Past Year)	55	11.07	7.78	[1, 39]	9.54	[8.38, 10.73]
Sum of Avg. Video Game Play Time (Hours/Before Past Year)	58	18.50	12.42	[1, 50]	14.6	[13.11, 16.1]
Sum of Video Gaming Expertise	56	15.56	7.16	[4, 33]	16.2	[14.58, 17.84]

N: sample size; M: maximum likelihood average estimate; SD: standard deviation; Range: minimum and maximum value; M_Bayes: Bayesian mean estimate; 95% CI_Bayes: lower and upper bound of Bayesian credibility interval; all variables except Military Service Duration and Video Gaming Expertise followed an exponentially modified Gaussian distribution; Military Service Duration and Video Gaming Expertise were skewed normally and normally distributed, respectively. Military Service Duration: there was one civilian individual and one individual did not provide a specified information (hence N = 57); Video Game Play Time (Hours/past Year): four individuals indicated not to have played video games in the past year (hence N = 55); Video Game Play Time (Hours/Before Past Year): one individual indicated not to have played video games in the time span before the past year (hence N = 58); Video Gaming Expertise: two individuals indicated zero expertise and one individual did not provide information on expertise (hence N = 56).

Procedure

Volunteers were recruited using the Online Recruitment System for Economic Experiments (ORSEE) software (Version: ORSEE 3.0.0, https://www.orsee.org/web/license.php)⁵⁸, and assessments were conducted in a university laboratory between October and December 2023. At first, the participants performed a visual short-term memory task³⁵. Subsequently, they performed the openMATB⁴⁴. Afterward, they filled in a questionnaire on their demographics, a media multitasking questionnaire⁵³, and a questionnaire on gaming experience¹⁸. The latter questionnaire constitutes an established tool to also subdivide individuals into action-video game players, non-action video game players, and non-video game players but we did not employ the questionnaire for such purposes (the interested reader may find details on the number of each video game player subtype in the Supplementary Material though). Stimuli had been presented on an EIZO^® color monitor with a 27 inches screen diagonal and a frame rate of 144 Hz at a resolution of 2560 × 1440 pixels. Assessments were conducted in sound-proof cabins. All software was Python based and either run on a Windows or Linux operating system.

Assessment of cognitive capacity

The visual short-term memory task³⁵ required individuals to memorize and report as many white shapes as possible presented on an invisible circle at one of six different exposure durations. It comprised three parts (a training and two experimental blocks). The training consisted of 24 trials and the experimental blocks of 210 trials each. Only performance in response to experimental trials was analyzed. Trials consisted of a sequence of visual events: at first, a gray screen was presented (1002 ms). Subsequently, a white fixation cross appeared in the center of the screen. After 1002 ms, a memory display was presented. Memory displays comprised six white shapes that were depicted on an invisible circle, whereby three shapes were presented in each visual field. There were no shape duplicates in one memory display. Memory displays were presented at one of six exposure durations (16.7, 33.4, 50.1, 83.5, 150.3 or 200.4 ms). Exposure durations were randomly assigned to a memory display, but the number of memory displays presented at either one of the exposure durations was counter-balanced. After the respective exposure duration, memory displays were masked. Mask displays comprised six white squares where random black polygons were depicted on, and covered the locations where white target shapes had been presented to prevent visual afterimages. Individuals were instructed to memorize as many white shapes as possible and to indicate which ones they memorized by pressing keys on a keyboard. Participants were not to guess; and they were allowed to take breaks between trials.

Theory of visual attention (TVA) computational modeling

TVA computational modeling was conducted using the libtva toolbox for MATLAB⁴⁵. The TVA^31,32 proposes that the probability of being able to memorize n elements depends on the rate at which information can be encoded. However, short-term/working memory capacity is limited. Thus, information encoding falls within a race model. The rate at which information is encoded depends on both objective and subjective constraints. Stimulus resolution, for instance, constitutes an objective predictor. Attention weighting, in turn, poses a subjective predictor. Hereby, TVA proposes a hierarchical mechanism with pigeonholing affecting the attentional weight of stimulus categories (for instance, color vs. shape) and filtering the attentional weight of features of said categories (for instance, blue vs. red). The likelihood of memorizing n elements is exponentially increasing the longer the exposure duration⁴⁵. The starting point of this exponential growth is considered the minimum exposure duration (t0) required for conscious information processing; the rate at which the exponential curve increases illustrates the speed of information processing (C); and the asymptotic level constraining the growth poses the short-term/working memory capacity (K)^31,32,45.

Assessment of multitasking performance

The openMATB⁴⁴ consists of four subtasks that must be performed simultaneously: in the system monitoring task, individuals are required to press the buttons F1, F2, F3, or F4, in case arrow indicators of control barometers deviate too strongly from their respective mid-line; and F5 or F6, in case either one of two buttons changes color, respectively. In the tracking task, individuals must steer a circle as close to the center of a hair cross as possible using a joystick. In the communication task, individuals must respond to radio calls by adjusting radio frequencies. In the resource management task, the filling levels of two tanks need to be maintained at optimal filling levels by en- and disabling pumps. The participants performed four blocks consisting of five trials. One trial lasted 90 s. In the first block, participants used the first four trials to familiarize themselves with each task separately and the last trial to familiarize themselves with the multitasking condition by operating on all tasks simultaneously. During this performance, the participants were allowed to ask questions in case task instructions were unclear. Afterward, they executed multitasking in each block. These trials served as experimentally relevant trials and the participants were supposed to perform the task in silence without consulting the instructor. Two of each of the four blocks were characterized by different scenarios (that is the procedure describing when an event, for instance, a system failure, happens). The order of block scenarios was counter-balanced. System monitoring and communication task performance were operationalized based on the number of misses; tracking performance based on the root mean squared error (RMSE) of the cursor relative to the mid-point; and resource management task performance based on the absolute mean deviation from optimal fuel levels of both tanks. For a visualization of openMATB performance measures in experimental blocks, see Figure S1.

Data analysis procedures

Preprocessing

TVA estimates and human cognitive capabilities do not necessarily match. For instance, the algorithm could fit the data such that C < 1 (although memorizing e.g., half a shape at 1 Hz is unreasonable)³². We substituted these values based on the following procedure: at first, we determined the distribution according to which TVA parameter values with a reasonable value range were distributed (either as a Gaussian, skewed normal, or exponentially modified Gaussian distribution); then, we estimated distribution parameter estimates thereof using Bayesian procedures⁵⁰. Following this, we generated samples of 1000 data points following the corresponding distribution (and parameter estimates). We took one random value from that distribution that was smaller than the minimum value of the original distribution but larger than the minimum threshold value mentioned above (C > 1); and performed this procedure 59 times to generate a sample of minimum values matching the original sample size. Subsequently, the mean value of this distribution was determined. This procedure, in turn, was performed 1000 times to generate a distribution of 1000 mean minimum values. Unreasonable values were then substituted by percentile values of these simulated distributions. For instance, there were two C parameter values < 1. Excluding these values, C parameter values were exponentially modified Gaussian distributed with M_Bayes ~ 33.99, SD_Bayes ~ 4.82, and λ_Bayes ~ 13.84 (and 11.22 constituting the smallest C value). A generated sample could then comprise values ranging from ~ 8.79 to ~ 10.08—with ~ 9.53 and ~ 10.08 illustrating the 50% and 10% percetile values. These were used to substitute unreasonable C values < 1. Kendall’s rank correlation tests proved that the ranks of the original and substituted values were identical (τ = 1, z = 11.19, p < .001). Thus, our preprocessing approach did not confound our analyses. Figure S2 and Table S1 confirm that the final TVA estimates fell within established value ranges. For the interested reader, we also provide information on the t0 parameter estimates in the Supplementary Material.

Missing data

One participant did not provide information on video gaming expertise. We approximated that value based on the Bayesian mean estimate.

Bayesian procedures

Bayesian estimations were based on distributions generated by means of 4 Markov Chains of at least 5000 iterations. Gaussian, skewed normal, or exponentially modified Gaussian distributions were used as link functions – depending on the distribution of the observed data⁵⁰. Parameter distributions were considered normally distributed. Priors were based on a priori knowledge and data simulations approximating the observed distributions. Details on priors (e.g., parameter distribution assumptions) can be found on OSF (https://osf.io/c9ats/overview?view_only=de6945bf1e6a409c89d0401c6d1035bd).

Hypotheses testing

We hypothesized that (H1) video gaming experience would predict multitasking performance, and that (H2) this effect would be mediated by TVA cognitive capacity. To investigate this, we used structural equation modeling (SEM): video gaming experience was modeled as a latent variable with the sum over the self-reported average video gaming play times (per video game genre category on a scale encompassing: never, 0 + to 1, 1 + to 3, 3 + to 5, 5 + to 10, and 10+) per week in the past year, the sum over the self-reported average video gaming play times (per video game genre category on a scale encompassing: never, 0 + to 1, 1 + to 3, 3 + to 5, 5 + to 10, and 10+) per week prior to the past year, and the sum over the self-reported expertise scores on the video game genre categories (using a scale ranging from 1 to 7) as indicator variables¹⁸. For the first two sum scores the highest possible value of each response category was taken. Multitasking performance was modeled as a latent variable with openMATB task measures averaged across blocks as indicator variables (whereby larger values indicate worse performance). Cognitive capacity was modeled as a latent variable either with TVA C and K parameter estimates as indicator variables or with either C or K parameter values as independent variables, respectively. All variables were z-standardized. The model fit quality was determined based on the -test result, the comparative fit index (CFI), the Tucker-Lewis index (TLI), the approximated root mean square error (RMSEA), and the standardized root mean square residual (SRMR), whereby p > .05, CFI > 0.95, TLI > 0.95, RMSEA < 0.05, and SRMR < 0.08 indicate acceptable quality, respectively. All model fit measures were bootstrapped based on 5000 samples. The significance of estimates was determined based on 95% confidence intervals (CI). Additionally, we determined the statistical power of the observed effects and determined at which sample size 80% statistical power was acquired using SEM simulations based on 5000 samples.

Control analyses

Furthermore, we conducted a series of analyses to control for the effects of potential covariates. There are indications that either one of the variables (mentioned above) relates to military experience⁵⁰, age and/or gender^51,52, and/or media multitasking^53,54. For this, we included military service duration in years, age in years, and the media multitasking index (MMI) as mediating variables. Moreover, we fitted the best model subdivided according to gender. The influence of the covariates was determined by comparing Bayesian information criterion (BIC) values of models with covariates to models without covariates (whereby smaller values indicate better model fits). Bootstrapping was also applied for these analyses. Similar analyses but with flight experience as covariate were not applicable due to an insufficient amount of data points. Instead, we controlled if the performance of individuals with flight experience was extremely good/bad (3 inter-quartile range / the first/third quartile). Besides, we controlled if the data fit better to models with either individual TVA parameter values or a TVA compound score (average over z-standardized TVA parameter values with t0 being inverted) (see Table S2); and we controlled if individuals demonstrated practice effects using Bayesian procedures (see Table S3 in Supplementary Material for further details).

Software and data

All statistical analyses were conducted using RStudio⁵⁹. Data were (pre-)processed using the dplyr and tidyr packages^60,61. Structural equation modeling was carried out using the lavaan, simsem, semTools and tidySEM packages^62–65. Bayesian procedures were conducted using the brms package⁶⁶. Visualizations were made using the semPlot and semptools packages^67,68. The original manuscript was generated using the papaja package⁶⁹.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Y.H.: Yannik Hilla; S.-M. S.: Sophie-Marie Stasch; W.M.: Wolfgang Mack; Y.H. was in charge of the project conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, visualization of the results, and validation; and Y.H. wrote the original draft of the manuscript. S.-M.S. provided the necessary materials for running the openMATB and for preprocessing openMATB raw data. Additionally, S.-M.S. reviewed and edited the manuscript. W.M. provided the resources for the project, supported the project administration, and reviewed and edited the manuscript.

Data availability

Data, materials and code supporting the findings of this work will be available in a repository on OSF (https://osf.io/c9ats/overview?view_only=de6945bf1e6a409c89d0401c6d1035bd).

Declarations

Competing interests

The authors declare no competing interests.