Introduction

Video games have attracted a lot of attention in recent decades, from the growing population who enjoy them1, to the media who often revile them2, and amongst researchers who continue to explore both beneficial and detrimental aspects of gaming3,4. Some research suggests that playing video games can be associated with better performance across multiple domains of cognitive and executive functioning such as attention, visual spatial awareness, psychomotor control, working memory, and inhibition and planning, for review see: Bediou et al.5. These outcomes are typically attributed to a transfer of cognitive and perceptual skills acquired during gaming, to other tasks which engage these same skills, thereby resulting in performance benefits6.

Action video games (AVGs) in particular, defined by their fast-paced style of gameplay, appear to facilitate skill transfer more effectively than most other activities, where learning is confined much more locally4,7,8. Evidence from both correlational and experimental design AVG studies has suggested that performance across various cognitive, attentional and perceptual tasks improves after a sufficient amount of time spent playing AVGs9,10. Further, there is some evidence that AVGs yield advantages in real-world settings, with a handful of studies finding performance boosts in areas such as laparoscopic surgical procedures11, piloting of remote military robotics12 and driving13.

Notably, all these examples suggest that skill transfer is taking place, despite significant contextual and situational differences between AVGs and the other tasks being performed. This runs counter to conventional wisdom which holds that practice-based skill development only results in domain-specific learning which does not transfer beyond the scope of the original environment14. For example, learning appropriate strategies and move-sets will make someone a better chess player but will do little to improve their performance on checkers or backgammon. By comparison, evidence appears to indicate AVGs yield a “far-transfer” of skills, the so-called ‘holy grail’ of education and learning practices15. If substantiated, this would indicate that AVGs hold tremendous promise to broadly improve cognitive and perceptual skills and benefit a range of other tasks.

Despite this promise, the existing literature is far from consistent16, with conflicting results sometimes reported even amongst AVG studies with methodologically similar designs e.g. Feng and Spence17,] compared to Roque and Boot18. This has led some researchers to claim that AVGs are ineffective for cognitive training18,19,20, while others vigorously defend their potential5,10. In light of this considerable uncertainty, we propose a different methodological approach to investigating the potential benefits of AVGs that may aid in reaching a consensus.

To date, most research investigating the association between cognitive performance and video gaming has used self-report measures of experience (i.e., asking how long/frequently video games have been played) to discriminate between ‘gamers’ and ‘non-gamers’21,22. This approach seems based on an implicit assumption that experience is synonymous with skill level and thus that experienced players perform better because they have superior skills derived from their time spent playing games. However, there may in fact be significant differences in skill between individuals who self-report similar amounts of experience. Such differences could arise due to factors related to the individual (i.e., engagement, motivation, intention, dexterity, age, and deliberate practice) and/or to the nature of the specific video games being played (i.e. gameplay elements, difficulty, stakes, and rewards)23,24,25.

To illustrate, consider two gamers with a similar initial skill level who each spend one thousand hours playing AVGs. One does so with deliberate focus, practicing to improve their skills in the AVG and aiming to play professionally in e-sport tournaments. The other plays simply for entertainment and without a specific focus or aim to challenge themselves. Though both will have accrued the same amount of gaming experience, via number of hours played, the relative skill level reached for the respective players will differ greatly. Though extreme, this example illustrates the potential for measures of gaming experience to mis-classify “experienced” players as being more skilled than they really are. This, in turn, would likely influence the outcome of any study that is investigating the impact of playing AVGs on other tasks.

In light of this point, we propose that directly measuring AVG performance (we refer to this as AVG proficiency) may better reflect the individual differences in skill gained through playing an AVG over experience alone26,27. This is due to proficiency more directly indexing the cognitive and perceptual processes that are being honed while playing AVGs. To date, examples in the literature where an AVG proficiency measure is used are extremely limited. Kokkinakis et al.28 found a positive Spearman’s rank order correlation between player rankings on a multiplayer online battle arena game (League of Legends) and fluid intelligence. Similarly, Millard et al.29 found a positive Spearman’s rank order correlation between veterinary students’ performance on target shooting style video games and performance on a set of simulated laparoscopic tasks such as making incisions and accurately guiding equipment. Although correlational, these outcomes suggest AVG proficiency could be related to performance on other tasks. However, crucially, because AVG experience was not measured in either study, the potential added benefit of assessing proficiency over experience remains unclear.

The current study examines whether the inclusion of video game proficiency is better than experience alone at predicting performance in a driving task. Driving is a real-world task which engages many of the same types of cognitive skills needed for AVGs, such as directed attention, object perception, psychomotor control, and executive functioning30, while placing significant loads on a driver’s cognitive resources. It is also a good example of a literature where there is some evidence that AVGs may benefit driving performance31 but findings remain mixed13,32. Conclusively demonstrating a relationship between AVG proficiency and driving would open the door to a range of future interventions to improve driving safety.

We assessed AVG proficiency using participants’ performance in a commercially available AVG (Quake III Arena). Driving performance in a simulator was measured via vehicle control (maintaining lane and speed) and spare cognitive capacity (Detection Response Task; DRT; see: Howard et al.31). To examine whether AVG proficiency might differentially predict driving performance when drivers are experiencing elevated cognitive load, participants completed the driving task twice – once while performing a concurrent distraction task and once without this distraction task. We investigated whether differences in driving performance were better predicted by differences in participants’ AVG experience or proficiency using hierarchical linear modelling (HLM).

Methods

Participants

A power analysis using G*Power33 indicated that 78 participants were required to detect a small/medium effect using regression. We aimed to recruit 120 undergraduate student participants, and ended up with 126. Of these, ten either withdrew or experienced technical issues which resulted in incomplete data, leaving a final dataset of 116 participants (43 Male, 72 Female, Age: M = 20.8 years, range 17–34, SD = 3.44). Younger adults were selected because they represented an age range at greater risk for driving incidents34,35 and are also relatively more likely to play AVGs. Participants held a current Australian car drivers’ licence and received course credit, plus a bonus of up to AUD$5 in exchange for their participation. All participants provided informed consent, and the study was approved by the UWA Human Research Ethics Office in accordance with relevant guidelines and regulations.

Materials

Driving simulator

A medium fidelity simulation running Oktal SCANeR Studio software (v 1.4; https://www.avsimulation.com/en/scaner/) was presented on three 32-inch monitors (140° field of view), housed within an Obutto gaming cockpit (Fig. 1). Participants were seated ~ 85 cm from the central monitor, which presented the front windscreen view, with the rear vision mirror presented at the top and digital speedometer presented at the bottom. Side windows, including side view mirrors, were presented on the two adjacent monitors. A Logitech G27 steering wheel and pedal set was used to control the vehicle, configured for automatic transmission and Australian left-hand driving conditions.

Fig. 1
figure 1

Left image shows the driving simulator cockpit and displays. Photograph of Setup. Right image shows central monitor with four lanes and DRT stimulus visible. In software screen shot from Oktal SCANeR version 1.4 (https://www.avsimulation.com/en/scaner/).

Full size image

Participants drove in the left-most lane of a four-lane, gently winding road (Fig. 1). There was no traffic in the participant’s lane and the other lanes were lightly populated with vehicles (approx. five per minute). There were no intersections and participants were instructed to drive at 50 km/h and remain in their lane. Vehicle speed and lane position data were recorded continuously at 1000 Hz and down sampled to 50 Hz for analysis.

A distraction task was also intermittently presented to participants while they were driving. It consisted of a series of audio-based addition questions with two randomly generated operands between 0 and 10 (e.g., “9 plus 4 equals?”) presented via a Samsung Galaxy A 8” tablet installed with an Android version 11 operating system. Participants answered each question aloud into a headset microphone and a tone was played to confirm each response before the next question was presented.

The distraction task was presented in three continuous 6-min blocks during the drive, interleaved with three 6-min blocks without the distraction task. Block order was counterbalanced between participants. Prior to each block, the driving simulation was paused for 30 s, and participants were given instructions that they would either “need to answer maths questions during the next block” or “not need to answer maths questions during the next block” depending on whether a distraction was to be presented. Thus, the total duration of the drive was 39 min, divided into two equal sections, separated by a short rest break.

Detection response task

A modified DRT was used36, in which participants responded to a series of small (0.49° of visual angle) red dot probes presented on the central monitor. A total of 160 probes were presented across four horizontal eccentricities (40 probes each). ‘Inner’ probes were presented randomly at locations between 6.80° and 10.13° of visual angle to the left or right of the vertical midline, and ‘outer’ probes were presented randomly between 18.14° and 21.17° of visual angle to the left or right of the vertical midline. All probes appeared randomly within 2-4° of visual angle above the horizontal midline. Probes were presented for 2s, or until a response was made, and separated by a random interval of 10–15 s. Participants were instructed to respond to probes as quickly and accurately as possible by pressing a designated response button on the steering wheel.

Action video game

Participants played the ‘Gold’ version of Quake III Arena (Gold Version; id Software. Quake III Arena, 1999; https://www.gog.com/en/game/quake_iii_arena). An additional downloadable mod, OSP Tourney, 2003 was also installed to facilitate match administration. This mod enabled additional logging and scoring features but did not make any alteration to core gameplay, mechanics, or AI (https://www.moddb.com/mods/osp). Quake III Arena was chosen because it is a ‘pure’ first person shooter game, lacking any story, mission, strategy or tactical elements, with a very fast, reaction-based, gameplay style37. Compared to modern equivalents, Quake III Arena, also has low level of realism, minimal gore, and enables modification to allow for standardisation and recordable output, making it suitable for assessing proficiency outcomes.

Participants completed two training and one experimental ‘match’. In a match, participants competed against three AI controlled opponents. All three AI opponents appeared as a glowing blue skeleton. AI behaviour was governed by both difficulty settings and character model used, meaning all three opponents followed the same set of programmed behaviours38. Matches were conducted as a ‘free-for-all’, with AI opponents targeting each other as well as the player.

Questionnaires

Participants completed questionnaires assessing driving experience, age, gender (options: male, female, other), and video gaming experience (Dale et al.21; Green & Bavelier4; available at https://osf.io/pkmdv/). Experience playing seven different types of video games, including first- and third- person shooting style games and action role playing games, was assessed based on the average number of hours spent playing each week for each type of game, using a 6-point scale (Never, 0–1 h, 1–3 h, 3–5 h, 5–10 h, 10 + hrs). Participants responded based on their gameplay experience over the last 12 months, and again for their experience prior to this period. Given the focus of the current study, an additional item, not included by Green and Bavelier10, was included to specifically assess experience playing driving games (“How often do you play racing / driving style video games, i.e., Need for Speed, Gran Turismo, FORZA, or similar”) with four options (‘Never’ to ‘Often’).

Procedure

Participants first completed the simulated driving task, beginning with a 10-min training scenario, with instructions, to familiarise themselves with driving in the simulator. During this training, they practiced driving, responding to DRT stimuli, and performing the distraction task. Participants were instructed that they would receive an initial bonus of up to AUD$5, which was reduced if they drove either too fast or too slow, in order to incentivise adherence to the speed limit. Following this training, participants completed the driving scenario as outlined above.

After driving, participants completed the AVG task. Match 1 was a 5-min training and familiarization game in which participants were given basic coaching by the experimenter on how to move and navigate, fire the weapon, and other basic gameplay requirements. Participants were instructed to shoot as many targets as possible while avoiding taking damage and dying as much as possible. Match 2 was an 8-min practice game with difficulty set to the easiest mode (‘I can Win’). Match 3 was the 8-min testing trial with difficulty set to hard (‘Hurt me plenty’).

Lastly, participants completed the online questionnaires. The entire experiment took approximately two hours to complete. Materials and analysis code for this study are available by emailing the corresponding author.

Hypotheses

Since AVG proficiency reflects underlying skill better than AVG experience, we expect that proficiency should predict improved driving performance better than experience. Accordingly, we hypothesised that greater AVG proficiency (H1a) and experience (H1b) would predict improved speed control (operationalized as speed consistency), but that proficiency would be a stronger predictor than experience (H1c). We also hypothesized that greater AVG proficiency (H2a) and experience (H2b) would predict improved lane maintenance (operationalized as lane position consistency) but that proficiency would be a stronger predictor than experience (H2c). Finally, we hypothesized that greater AVG proficiency (H3a) and experience (H3b) would predict improved DRT performance (faster probe responses), with proficiency expected to be a stronger predictor than experience (H3c). We also expected that these relationships would be stronger for distracted compared to non-distracted driving conditions due to the greater need for cognitive and perceptual resources to maintain driving performance when distractions were present.

Results

Data analysis

Driving performance was assessed via metrics of vehicle control (speed control, lane maintenance), and spare cognitive capacity (DRT performance). Speed control was measured as the standard deviation of participant vehicle speed (in km/h), with lower values indicating better performance and a more consistent speed control. Lane maintenance was measured as the standard deviation of vehicle lane position (in metres), with lower values indicating better performance and more consistent vehicle position relative to lane centre. DRT hits were quantified as responses made 100−2500 ms after probe onset. DRT response times (RT) were calculated only for hits. Responses made outside of this window were counted as false alarms (M = 2.08, SD = 1.95). Video game proficiency scores were calculated by dividing the number of ‘kills’ in Match 3 by the number of participant character ‘deaths’. This generated a kill/death ratio where higher scores indicated more kills and fewer deaths, and thus better performance39,40,41.

It has been suggested that the video game questionnaire responses should not be treated as continuous variables for regression analysis see:21,40. We therefore applied a categorical approach to quantify experience using responses to four questions looking at experience playing: (a) first and/or third person action shooter games and (b) Role Playing / Adventure action games, both in the past twelve months and prior to the past twelve months. These four questions make up the formal criteria used to assess an AVG player, and the types of games assessed have all been associated with improved cognitive performance10,21.

The four questions each had six possible responses for a total 24 categories. To reduce the number of categories, each participant was assigned a category based on their highest response across the four questions such that someone endorsing responses 0–1 h, 0–1 h, 5–10 h, and 3–5 h to the four questions respectively would be scored 5–10 h. We then condensed this to three categories for the final analysis: <1 h (N = 59), 1–5 h (N = 34), >5 h (N = 23) spent playing AVGs per week.

Each of the driving metrics presented was captured under a set of within subjects’ conditions. Lane maintenance and speed control were each recorded under both distraction absent and distraction present conditions. DRT accuracy and RT were similarly recorded under distraction absent and distraction present conditions, separated further by inner and outer eccentricity conditions. DRT measures are effectively nested (see Fig. 2) such that each of the 116 participants provided a total of four measures each for accuracy and RT. Similarly, each of the 116 participants provided two measures each for lane maintenance and speed control.

Fig. 2
figure 2

Diagrammatic representation of the data structure for DRT accuracy and RT, speed control and lane maintenance measures. .

Full size image

To account for the nested data structure, we used HLM41,42 to analyse our data. A HLM has advantages in parameter estimation methods and algorithms, model assumptions and data requirements, compared to traditional regression models, and is able to account for individual differences due to random variance43. These analyses were conducted using maximum likelihood estimation, in R Studio (v. 2024.04.2; https://posit.co/download/rstudio-desktop/) using the lme4 package44. As we were interested in investigating whether differences in participants’ AVG experience or their proficiency (between-subjects effects) predicted differences in average driving performance across participants, we conducted HLMs for each driving metric, using a four-step model building process.

Sensitivity power analysis

A helpful reviewer suggested we include a sensitivity power analysis. To do this, we used the simr package45 and estimated the smallest effect size detectable where the 95% CI for power had a lower bound greater than 80. Based on the final sample of 116 participants and alpha = 0.05, we could detect AVG proficiency effect sizes of \(\:\gamma\:\) = 0.198 for DRT RT (95% CI for power: 80.63 to 85.37), \(\:\gamma\:\) = 0.275 for Lane Maintenance (95% CI for power: 82.63 to 87.16), \(\:\gamma\:\) = 0.273 for Speed Control (95% CI for power: 82.32 to 86.88), and \(\:\gamma\:\) = 0.075 for DRT Accuracy (95% CI for power: 81.58 to 86.22).

Lane maintenance

To illustrate this process, we step through how we constructed the HLM regression model for lane maintenance in Table 1. At Step 1, we evaluated whether allowing for random intercepts for each participant was a better fit to the data than having a fixed intercept across all participants. Model comparison suggested that allowing intercepts to vary across participants provided a significantly better fit to the data (Likelihood Ratio Test: \(\:{\chi\:}_{\left(2\right)}^{2}\)= 120.88, p < .001). At Step 2, the fixed and random effects of distraction were added. The fixed effect of distraction contributed significantly (\(\:\gamma\:\) = − 0.215, p < .001), such that a distraction being presented led to an improvement in lane maintenance (reduced lane variance). However, the random effect of distraction (SD = 0.285) did not significantly contribute to the model (Likelihood Ratio Test: \(\:{\chi\:}_{\left(6\right)}^{2}\)= 0.133, p = .936), suggesting that the effect of distraction on lane maintenance was similar across participants. The random effect of distraction was therefore dropped from subsequent analyses. At Step 3, fixed effects of AVG experience were added. As discussed earlier, AVG experience was grouped into three categories. For the analyses, we therefore dummy coded AVG experience by setting <1 h as the baseline/reference category. Having 1–5 h (\(\:\gamma\:\) = − 0.460, p = .021) and >5 h (\(\:\gamma\:\) = − 0.613, p = .008) of AVG experience both improved lane maintenance performance, relative to baseline. At Step 4, the fixed effects of AVG proficiency were added. AVG proficiency contributed significantly (\(\:\gamma\:\) = − 0.287, p = .002), with increased proficiency associated with improved lane maintenance performance. However, with the addition of AVG proficiency, AVG experience was no longer a statistically significant contributor to the final model.

Table 1 Hierarchical linear modeling results predicting lane maintenance.
Full size table

Speed control

The same model building process was conducted for both the measure of speed control, and our other driving metrics, so we present only the results of the final models (i.e., Step 4) for these analyses (see Table 2). In the case of speed control, model comparison found a random effect for participants (SD = 0.774; Likelihood Ratio Test: \(\:{\chi\:}_{\left(2\right)}^{2}\)= 45.55, p < .001) and no random effect for distraction (Likelihood Ratio Test: \(\:{\chi\:}_{\left(6\right)}^{2}\)= 0.944, p = .624), which was dropped from subsequent analyses. There was a fixed effect of distraction such that the presence of a distraction lead to reduced performance (increased speed variance) (\(\:\gamma\:\) = 0.301, p < .001). However, the fixed effect of AVG experience did not predict speed control (\(\:\gamma\:\) = − 0.149, p = .436). The fixed effect of AVG proficiency did contribute significantly, with increased proficiency associated with improved speed control (lower speed variance) (\(\:\gamma\:\) = − 0.227, p = .013).

Table 2 Hierarchical linear modeling results for distraction, eccentricity, AVG experience and AVG proficiency predicting four measures of driving performance.
Full size table

DRT accuracy

Results of the final model for DRT accuracy are presented in Table 2. Model comparison found no random effects for participants (Likelihood Ratio Test: \(\:{\chi\:}_{\left(2\right)}^{2}\)= 0, p = .999), while there were random effects for distraction (Likelihood Ratio Test: \(\:{\chi\:}_{\left(7\right)}^{2}\)= 6.80, p = .033) and eccentricity (Likelihood Ratio Test: \(\:{\chi\:}_{\left(7\right)}^{2}\)= 117.77, p < .001). There were also statistically significant fixed effects, with the presence of a distraction (\(\:\gamma\:\) = − 0.117, p < .001) and presentation of probes at outer eccentricities (\(\:\gamma\:\) = − 0.814, p < .001), leading to reduced performance (reduced DRT accuracy). However, AVG experience did not predict DRT accuracy (\(\:\gamma\:\) = -0.030, p = .245). AVG proficiency did contribute significantly, with increased proficiency associated with improved performance (increased DRT accuracy) (\(\:\gamma\:\) = 0.075, p = .005).

DRT RT

Results of the final model for DRT RT are presented in Table 2. Model comparison found random effects for participants (Likelihood Ratio Test: \(\:{\chi\:}_{\left(2\right)}^{2}\)= 94.05, p < .001), distraction (Likelihood Ratio Test: \(\:{\chi\:}_{\left(7\right)}^{2}\)= 6.16, p = .046) and eccentricity (Likelihood Ratio Test: \(\:{\chi\:}_{\left(7\right)}^{2}\)= 14.85, p < .001). There were also statistically significant fixed effects, with the presence of a distraction (\(\:\gamma\:\) = 0.379, p < .001) and presentation of probes at outer eccentricities (\(\:\gamma\:\) = 0.379, p < .001), leading to reduced performance (reduced DRT accuracy). However, AVG experience did not predict DRT accuracy (\(\:\gamma\:\) = − 0.071, p = .676). AVG proficiency did contribute significantly, with increased proficiency associated with improved performance (increased DRT accuracy) (\(\:\gamma\:\) = -269, p < .001).

Confirmatory analysis

The analysis presented above condensed the measures of AVG experience into three categories based on the highest average number of hours played per week (less than 1 h, 1 to 5 h, more than 5 h). In order to ensure that the observed effects were not due to this method of collapsing experience measures alone, we repeated the above analysis using alternative approaches. For comprehensiveness, three alternatives were tried. The first reduced experience to 12 categories (< 1 h, 1to 5 h, >5 h, for each AVG related question). The second produced six categories using only the responses to the “Action First/Third person shooter games played in the last 12 months” question. The third was to generate a continuous variable based on questionnaire responses. In all instances, AVG proficiency was found to consistently predict driving performance, while experience was never a significant predictor in the final model. The consistent finding across three alternative approaches of operationalising AVG experience increased our confidence in the presented findings, as it suggests that the approach used for scoring experience did not significantly change the findings.

We also repeated the analyses, including age, gender, driving experience, and experience with driving-based video games in the model as control variables. None of these predictors contributed significantly to the final model and did not change the conclusions related to our hypothesis.

Discussion

The results of previous studies investigating whether AVGs facilitate skill transfer to other task domains has been inconsistent. Here, we examined whether this inconsistency might be due to the use of AVG experience (i.e., hours spent playing AVGs) as a proxy measure for skill level (i.e., level of proficiency reached when playing an AVG). It was hypothesized that AVG proficiency would be more representative of gaming skill and would therefore better predict performance on another task. We tested this in a simulated driving task, finding that AVG proficiency predicted all measures of driving performance including: speed control (H1a), lane maintenance (H2a), and spare cognitive capacity (measured via DRT; H3a). By comparison, AVG experience either failed to predict driving performance (speed control [H1b] and spare cognitive capacity [H3b]) or only contributed significantly when AVG proficiency was not accounted for (lane maintenance [H2b]). Thus, we demonstrate for the first time that AVG proficiency is a significant predictor of performance in a simulated driving task, while also presenting the novel finding that the inclusion of AVG proficiency in a regression model is better able to predict driving performance than measures of AVG experience alone (H1-3c).

By demonstrating that proficiency is a better predictor of task performance than experience, we show that proficiency provides a better index of the cognitive and perceptual skills being transferred, compared to measures of experience. While AVG experience only predicted participants ability to maintain their lane position, AVG proficiency predicted performance on each of the aspects of safe driving measured. Furthermore, including proficiency in the model predicting lane maintenance reduced the contribution of experience to non-significance. The implication here is that individual differences in AVG skills are attributable to more than can be captured by a measure of experience alone.

Further, playing AVGs and driving represent two activities that are completed in very different environments, requiring different actions that are reliant on very different sets of incoming information to perform effectively. As such, the potential relationship observed here provides the first steps towards meeting the criteria of a far-transfer effect15. What has not yet been established is whether AVG skill is causally linked to driving skill, or if a set of cognitive and perceptual skills common to both driving and AVGs may account for our findings, particularly given previous demonstrations of the cognitive benefits associated with AVGs4,5,17. This represents an important direction for future investigation, given that a clearer link between the skills used in each activity has now been established.

Implications

The majority of previous research has relied on indirect measures of experience to assess the effects of video games. The current findings highlight the value of directly measuring gaming proficiency. Measuring proficiency eliminates many of the unknown quantities that accompany subjective, self-report experience measures, including those which affect the degree to which skills improve while playing (i.e., motivation to improve) and may serve to address some of the inconsistency that exists in this field of research. Additionally, proficiency measures can be obtained via a simple, reproducible single player AVG, which eliminates some of the complexity around recruiting from existing player bases or developing custom AVG software. This represents a promising alternative to existing experience-based methods and may open new avenues of video gaming research, supplementing existing correlational and experimental designs.

The current findings also have important implications for driver safety. First, we have identified an additional means of potentially assessing and predicting driving safety and performance outcomes. Second and most importantly, should future studies confirm that the relationship between AVG proficiency and driving is causal in nature, and that improving AVG proficiency through practice leads to improved driving performance, then this could open opportunities to design inexpensive, engaging and readily-available AVG based training packages to improve driver safety outcomes. This is particularly meaningful given that it is engagement in any such training that is key to maximising outcomes. AVGs are known for being highly engaging46, and strongly activate reward and attentional systems that are key drivers of learning47. Such training is therefore more likely to be appealing, with motivation to participate and persist with training, likely to be higher than for most lab-based tasks.

Limitations and future research

Establishing a causal relationship between AVG proficiency and driving performance will be the next challenge for this research. Demonstrating that increases in AVG proficiency, through training, leads to a corresponding improvement in driving performance will confirm that a transfer of skill is occurring, and open-up opportunities for improvements in driver safety. Such an approach would involve pre- and post- training assessments of AVG proficiency, in addition to similar assessments of driving performance. Assuming AVG proficiency increases post training, then corresponding improvements in driving performance would confirm a causal relationship.

Further research is also needed to investigate whether AVG proficiency predicts driving performance in a real-world assessment. While driving simulators have demonstrated ecological validity in assessing metrics of driver skill and safety such as speed and lane position48 simulation is not a perfect alternative to real-world driving conditions. For example, the information presented in a simulation is limited and cannot perfectly replicate all the sensory information obtained while driving or recreate the consequences of a driving error.

Finally, the current investigation used a sample of younger adults, as this population represents a group typically at greater risk of causing or being involved in a road crash35. However, this also limits the conclusions that can be made about the broader population, particularly those aged over 35 years. It is our expectation that a similar relationship between the skills used for driving and those used for playing an AVG might still be observed in an older sample. However, future research should seek to confirm this using an older sample of participants.

Conclusion

The current study takes the first steps towards demonstrating the importance of evaluating AVG proficiency as a more comprehensive method of indexing the skills gained through playing AVGs compared to traditional measures of experience. This, in turn, may aid in better understanding the process of skill transfer between AVGs and other tasks such as driving. For the first time, AVG proficiency was shown to be a better predictor of performance in simulated driving than measures of AVG experience. This finding opens the door to further studies which examine the relationship between AVGs and performance on both cognitive and real-world, with measures of proficiency offering an alternative to existing methods.