Edited by:
Reviewed by:
*Correspondence:
This article was submitted to Educational Psychology, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
We hypothesized that embedding educational learning in a game would improve learning outcomes, with increased engagement and recruitment of cognitive resources evidenced by increased activation of working memory network (WMN) and deactivation of default mode network (DMN) regions. In an fMRI study, we compared activity during periods of learning in three conditions that were increasingly game-like: Study-only (when periods of learning were followed by an exemplar question together with its correct answer), Self-quizzing (when periods of learning were followed by a multiple choice question in return for a fixed number of points) and Game-based (when, following each period of learning, participants competed with a peer to answer the question for escalating, uncertain rewards). DMN hubs deactivated as conditions became more game-like, alongside greater self-reported engagement and, in the Game-based condition, higher learning scores. These changes did not occur with any detectable increase in WMN activity. Additionally, ventral striatal activation was associated with responding to questions and receiving positive question feedback. Results support the significance of DMN deactivation for educational learning, and are aligned with recent evidence suggesting DMN and WMN activity may not always be anti-correlated.
Games offer incentivised conditions that are remarkably effective in engaging players in goal-directed behavior (
Efforts to localize working memory function during learning have converged on a dorsal fronto-parietal network that is activated in demanding tasks that require the processing of material presented (
In light of the above arguments, if gamification increases engagement with a goal-directed educational learning task (without additional self-referential or creative processing), we might expect to observe increased WMN activity and decreased DMN activity with gamification. Such predictions could be tested using a well-theorized learning game environment designed to engage its player. The mechanisms by which games, including learning games, incentivise their players remain to be fully understood, but some insights can be provided by our emerging understanding of the reward system. Midbrain dopamine neurons which project to the ventral striatum (VS) fire in response to cues that predict reward (
To explore how “gamification” of a learning context might influence engagement and learning, we used fMRI to measure brain activity when adults were studying in three conditions in a within-participants design. Our design also allowed the educational learning achieved between conditions within a defined time period (a learning window) to be meaningfully compared. In this way, it provided an indication of the comparative efficiency of the three approaches, which has been identified as an important issue for those interested in evaluating the effectiveness of digital game-based learning (
This study was carried out in accordance with the ethical procedures of the University of Bristol (Graduate School of Education) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki.
Informed consent was obtained from 24 healthy student volunteers (6 males, 18 females) who responded to an advertisement placed in the Education Department of the University of Bristol. The department’s population has a strongly international profile and the study was conducted in both Spanish and English to facilitate recruitment. The first 12 volunteers in each language category who met criteria (i.e., right-handed, no metal, no psychoactive medications) were recruited. The mean age of participants was 34.7 years,
For the sake of ecological validity, learning corpora were constructed that addressed a diverse range of educational topics (including history, biology, mathematics, grammar, electronics, music, horticulture). Four corpora were developed, each consisting of 10 topics that were exclusive to that corpus. For each topic, learning content was generated that consisted of a screen of text and pictures, along with an associated pair of questions (resulting in 10 screens of content and 20 questions for each corpus). Corpora (content and questions) were designed to provide an educational challenge that extended beyond factual recall. To achieve this, each of five levels of educational learning objective, as defined in educational terms by Bloom’s taxonomy (
The four corpora were permutated across the three scanning conditions of each participant, with the remaining corpus being reserved for the Game-based condition when they competed with their partner as their partner was being scanned. In this way, each participant encountered each learning corpus only once. Additionally, within each subgroup (
Three different sets of 40 multiple-choice questions (120 questions) on the learning content were also generated. These were designed to test participants knowledge and understanding of the learning content immediately before (pre-test) and after (post-test) being scanned and following a period of 3–4 weeks (retention test). Each set comprised 1 question on each of the 10 screens of content within each of the four corpora. These were novel questions, in the sense that they were similar in form, but did not replicate, the questions that participants experienced during scanning. The three sets were allocated for use as pre-, post-, and retention tests, with balanced permutation of question set across the three types of test within each subgroup of Spanish and English speakers.
Participants experienced three experimental conditions that represented three learning contexts (Study-only, Self-quizzing, and Game-based). Following study of each topic, the Study-only condition required participants to observe an exemplar question and answer, the Self-quizzing condition required them to select an answer for the question, and the Game-based condition required them to compete with a friend to select an answer, and to game their potential points on a wheel of fortune. These conditions were implemented using an interface resembling that used by Zondle Team Play (
In the Study-only condition, participants were presented with a question accompanied by only one answer option (the correct response). To balance conditions and ensure wakefulness, participants were still required to select this answer, after which their score would increase by 10 points.
In the Game-based condition, each of the four options featured a plausible response to the question, and participants were asked to choose the correct response during the 2.8 s question response window. In this condition, at the end of this period, the second circle would illuminate. This indicated that the participant and their competitor had 2.8 s to press the first button held in their left hand again, should they wish to see their winnings for a correct answer “gamed” on the wheel of fortune, and not press it if they did not wish to game their points. At the end of this gaming decision window, the response of both participant and their competitor to the question were revealed (inside the first circle) and the correct response was marked with a tick. At the same time, a participant and/or competitor who had not decided to game their points but had answered correctly would see their score increased by the points available for the question. After 2.8 s of displaying this information, a spinning wheel of fortune appeared. At the end of a further 2.8 s, the wheel of fortune stopped on either blue or white (with 50% probability). For both the participant and competitor, a correct response and prior commitment to game their points would result either in gaining double the points available for the round if the wheel landed on blue, or in losing their points for the round if it landed on white. The outcome of the spinning wheel of fortune remained on the screen for 2.8 s, before the next trial began. In this condition, points began with one point for the first two questions and increased by two points every two questions such that, over a 20 trial block, 19 points were offered for the last two questions (averaging 10 points per question – as in the other conditions). The order in which screens of learning content appeared was automatically randomized for each participant, with each screen appearing twice with a block, but always with a new question. Participants were made aware that, all else being equal, the average value of gaming was equivalent to the value of not gaming.
The Self-quizzing condition was identical to the Game-based condition except that no competitor was present and 10 points were provided for selecting the correct response. The average value of points for correct responses across the three conditions was, therefore, equivalent.
To ensure equal durations of trials in all conditions, a 2.8 s rest was provided in the Study-only and Self-quizzing conditions instead of a gaming decision window. A spinning wheel of fortune still appeared in the Study-only and Self-quizzing conditions for 2.8 s, but it was made clear to participants that this had no implication for their scores.
An advertisement requested volunteers to apply for “Brain School,” in which successful candidates would have the opportunity to learn some interesting general knowledge while having their brain scanned. Candidates were called to attend a preliminary meeting with researchers in pairs. They completed an initial survey of medical history to ensure they met scanning criteria, and the task and three conditions were then explained to them. Both candidates then experienced a shortened (15 min) version of the procedure together with fMRI simulation, each competing with the other candidate in the pair during the Game-based condition. Each applicant experienced two trials in each condition, using content and questions not employed in the main scanning experiment. This simulation helped ensure volunteers were aware of the procedures they were consenting to, and also helped acclimatize them to the scanning environment. Pairs who passed safety criteria and were able to provide informed consent for scanning were then offered a slot on one of four scanning days.
On arrival at the scanner, each member of a participating pair completed a 40 question pre-test (featuring 10 novel questions for each corpus) which assessed their prior knowledge of the material they would be learning. Inside the scanner, one member of each pair was functionally scanned while experiencing all three conditions in quick succession, with each condition comprising a block of 20 trials. In Study-only trials, participants simply had to study each slide of learning content, and the exemplar question and answer that followed it, acknowledging the latter with a press of a button. In the Self-quizzing condition participants were required to choose an answer for the question from four options, receiving 10 points for a correct response. In the game-based condition, the other member of the pair (as competitor) was also responding to this question from the control room, with both participant and competitor winning points for correct answers that escalated (from 1 to 19) over the block and that could be doubled or lost on a wheel of fortune. Following the three conditions, a structural scan was completed before the participant left the scanner. The pair of participants then swapped their roles, with the other member of the pair experiencing three conditions while being scanned and their partner acting as competitor in the Game-based condition. When scanning was completed for both members of the pair, the participants individually provided a brief self-reported evaluation of conditions in terms of engagement and stress, and were then asked to complete a 40 question post-test. Self-reports of engaged and stressed participants felt were indicated on a 5-point Likert scale, where 1 was labeled “not at all” and five was labeled “extremely”. After a period of 3–4 weeks, participants were recalled for another 40 question retention test.
All of the images were collected on a Siemens 3T Magnetom Skyra MRI scanner at the Clinical Research Imaging Centre (CRIC) at the University of Bristol. The functional images were collected using an echoplanar T2∗ gradient-echo EPI sequence to measure BOLD contrast over the entire brain (36 contiguous 3 mm-thick axial slices;
Visual stimuli were presented with a personal computer running in-house software implemented in Visual Basic (Microsoft), back projected onto a translucent screen and viewed through a mirror attached to the head coil. The presentation timing was controlled and triggered by the acquisition of the fMRI images.
The functional MRI data were pre-processed and analyzed offline using SPM8 (Wellcome Department of Cognitive Neurology, Institute of Neurology, London). For each subject, the functional images were first spatially corrected for head movements using a least-squares approach and six-parameter rigid-body spatial transformations (
The images were subsequently analyzed using a random-effects approach. At the first stage, the time series of the functional MR images obtained from each participant were analyzed separately. The effects of the experimental paradigm were estimated on a voxel-by-voxel basis, according to the general linear model extended to allow the analysis of fMRI data as a time series (
For inference at group level, these contrasts were subjected to a second level analysis in which random effects group statistics were generated. Regions of interest (ROIs) were
Behavioral results were analyzed using IBM SPSS Statistics, version 22. The average number of questions correctly answered by participants during the self-quizzing and game-based conditions were similar (see
Descriptive statistics of percentage correct scores achieved by individuals during learning in the Self-quizzing and Game-based conditions, and for decisions to game following correct responses in the Game-based condition (
Self-quizzing | % Questions answered correctly by participants | 51.4 | 13.9 |
Game-based | % Questions answered correctly by participants | 49.8 | 14.7 |
% Questions answered correctly by competitors | 46.0 | 14.1 | |
% Occasions participants chose to game points | 65.9 | 28.2 | |
% Occasions competitors chose to game points | 73.7 | 29.6 |
Means and standard deviations of the pre-test, post-test and retention scores for learning content experienced in each of the three conditions are provided in
Descriptive statistics for scores (out of 10) achieved in the pre-test, post-test, and retention tests (
Pre-test |
Post-test |
Retention test |
||||
---|---|---|---|---|---|---|
Study-only | 3.92 | 1.53 | 5.50 | 1.77 | 3.63 | 1.93 |
Self-quizzing | 3.42 | 1.64 | 5.58 | 2.00 | 4.17 | 1.53 |
Game-based | 3.21 | 1.69 | 6.25 | 1.82 | 4.38 | 1.55 |
Descriptive statistics of measures of immediate and retained learning calculated as the increase in post-test and retention test scores relative to the pre-test score.
Immediate learning |
Retained learning |
|||
---|---|---|---|---|
Study-only | 1.58 | 1.86 | -0.29 | 2.53 |
Self-quizzing | 2.17 | 2.50 | 0.75 | 2.13 |
Game-based | 3.04 | 2.48 | 1.17 | 1.97 |
Looking at the two time points separately using one-way ANOVAs, immediate learning was not significantly different across conditions (
A 3 × 2 × 2 (condition × time × language) ANOVA revealed no additional main effect of first language (
Across participants, the means of self-reported measures of engagement and stress were all in the direction Game-based > Self-quizzing > Study-only (see
Means (with standard deviations in parentheses) of self-reported measures of engagement and stress (on a scale of 1–5) for the three conditions (Study-only, Self-quizzing, Game-based) reported immediately following scanning.
Engagement |
Stress |
|||
---|---|---|---|---|
Study-only | 3.83 | 2.31 | 2.54 | 1.98 |
Self-quizzing | 7.33 | 1.43 | 5.54 | 1.86 |
Game-based | 8.7 | 0.93 | 6.88 | 2.40 |
In each condition, positive associations were sought between self-rated engagement and the learning achieved, and negative associations between self-rated stress and the learning achieved (see
Correlation statistics (Pearson’s
Correlation with subsequent measures of learning |
|||||
---|---|---|---|---|---|
Immediate |
Retained |
||||
Pearson’s |
Significance (2-tailed) | Pearson’s |
Significance (2-tailed) | ||
Study-only | Engagement | -0.227 | 0.285 | 0.111 | 0.606 |
Stress | -0.029 | 0.895 | 0.183 | 0.404 | |
Self-quizzing | Engagement | 0.214 | 0.315 | -0.264 | 0.213 |
Stress | 0.052 | 0.812 | -0.082 | 0.710 | |
Games-based | Engagement | 0.533 | 0.007** | 0.171 | 0.424 |
Stress | -0.149 | 0.497 | 0.036 | 0.872 |
No significant increases in activity in WMN ROIs during the learning window were identified when comparing the Self-quizzing condition with the Study-only condition, or the Game-based condition with either the Self-quizzing or the Study-only condition. A whole brain analysis revealed no unhypothesised activations at
Default mode network ROIs significantly deactivated in the Game-based condition compared with both the Study-only and the Self-quizzing conditions (see
A whole-brain analysis at
Regions deactivating for the learning window in the Game-based compared with Study-only condition [
Region | L/R | BA | Peak coordinates (MNI) | Peak |
Size in voxels | |
---|---|---|---|---|---|---|
Cuneus | 1 | R | 18 | 12 -94 16 | 6.21 | 1955 |
Precuneus | 2 | L | 7 | -6 -73 52 | 6.17 | |
Cuneus | 3 | R | 19 | 18 -94 22 | 6.05 | |
Inferior frontal gyrus | 4 | R | 9 | 48 11 34 | 6.09 | 386 |
Middle frontal gyrus | 5 | R | 6 | 36 -1 52 | 5.49 | |
Middle frontal gyrus | 6 | R | 9 | 45 26 31 | 5.17 | |
Inferior frontal gyrus | 7 | R | 47 | 51 20 -2 | 5.37 | 88 |
Inferior frontal gyrus | 8 | R | 47 | 33 20 -11 | 4.99 | |
Medial frontal gyrus | 9 | R | 8 | 6 32 46 | 5.36 | 285 |
Medial frontal gyrus | 10 | R | 6 | 6 14 52 | 5.36 | |
Medial frontal gyrus | 11 | R | 6 | 6 38 40 | 5.09 | |
Inferior frontal gyrus | 12 | L | 47 | -33 17 -11 | 5.32 | 67 |
Temporopolar cortex | 13 | L | 38 | -48 17 -11 | 5.23 | |
Lingual gyrus | 14 | L | 19 | -27 -73 -8 | 4.97 | 19 |
Thalamus | 15 | R | 9 -16 10 | 4.90 | 19 | |
Supramarginal gyrus | 16 | R | 40 | 57 -49 25 | 4.89 | 21 |
Superior temporal gyrus | 17 | R | 13 | 57 -49 16 | 4.59 | |
Middle frontal gyrus | 18 | L | 6 | -27 -7 52 | 4.87 | 18 |
Red nucleus | 19 | L | 0 -22 -5 | 4.72 | 10 |
An analysis at
Regions deactivating for the learning window in the Game-based compared with Self-quizzing condition [
Region | L/R | BA | Peak coordinates (MNI) | Peak |
Size in voxels |
---|---|---|---|---|---|
Inferior temporal gyrus | R | 19 | 45 -73 1 | 7.03 | 2790 |
Superior parietal lobule | R | 7 | 24 -73 46 | 6.34 | |
Precuneus | R | 7 | 18 -70 53 | 6.30 | |
Middle occipital gyrus | L | 19 | -45 -76 4 | 6.78 | 330 |
Middle occipital gyrus | L | 19 | -27 -88 16 | 5.03 | |
Superior frontal gyrus | R | 6 | 21 14 49 | 5.62 | 585 |
Middle frontal gyrus | R | 9 | 48 14 34 | 5.51 | |
Middle frontal gyrus | R | 8 | 39 23 40 | 5.50 | |
Precentral gyrus | L | 9 | -33 23 37 | 4.76 | 11 |
Whole-brain analyses failed [at
Trials in the present study were phase-locked to the scanner repetition time and not jittered. This allowed a pace of delivery and more trials, but prevented estimation of the BOLD timecourse on a time scale of fractions of seconds. Instead, a grosser estimate of change in BOLD response was obtained from contrasts for each of the five sub-periods of the learning window, to show how the level of deactivation of DMN ROIs changed between these 5.6 s periods. These contrasts were calculated for the Game-based condition (which produced greatest DMN deactivation) compared with the Study-only condition, and compared also with the Self-quizzing condition. These are shown in
Activity in the VS during the response window in the Self-quizzing and Game-based conditions was calculated by comparing ROI activity in the conditions with the Study-only condition, where a question and answer was presented with no response required.
Statistically significant activation associated with responding in the Self-quizzing condition, compared with the Study-only condition, was found in the left and right ventral striatal ROIs [
For positive, compared with negative feedback, there was increased activity in left and right ventral striatal ROIs in the Game-based condition during feedback [
For the Game-based condition, positive associations across participants were sought between deactivation of DMN ROI’s (relative to the quizzing condition) and the immediate learning achieved (relative to the quizzing condition), as calculated by pre-test/post-test difference. Deactivation of the left and right PCC was found to be correlated with learning (Spearman’s
This experiment focused on the changes in neural activity when participants studied in three environments that were, by becoming progressively more game-like, intended to increase goal-orientation and engagement with a learning task. The to-be-learned material included a range of different types of knowledge and concepts but, on all trials, success required participants to attend carefully to it by reading and understanding. Self-reported engagement improved with gamification, and the Game-based condition produced higher learning scores than the other (less game-like) conditions. All four ROIs corresponding to the nodes of the DMN in bilateral PCC [7 mm radius spheres at (-7, -51, 26) and (4, -51, 25)] and bilateral aMPFC [7 mm radius spheres at (-7, 50, 14) and (5, 50, 14)] deactivated in the Game-based condition relative to both the Study-only and Self-quizzing conditions. Previous reported activation of DMN with off-task behavior (
Given behavioral reports of increasing unrelated thoughts and decreasing attentional performance over time (
Whole-brain analysis revealed a range of activities that, although not considered core, have also been associated the DMN. These included dorsal medial prefrontal cortex, implicated in one of two DMN subsystems identified by Andrews-Hanna, also left middle frontal gyrus (
The lack of increased activation of WMN with gamification was contrary to the proposed anticorrelation of DMN and WMN networks observed in other studies (
Gamification of the learning environment was intended to improve goal-orientated motivation by stimulating the reward system. Alongside greater self-reported engagement as the context became more gamified and deactivation of putative hubs of the DMN, we observed that answering questions, and receiving positive feedback, in the two more gamified conditions activated the VS. Greater learning was achieved in the most gamified (Game-based) condition, and individual learning differences in this condition were correlated with deactivation of the posterior DMN hubs. Given the associated role of the DMN with internal thoughts and feelings unrelated to the task at hand, it seems likely that incentivisation may have increased goal-orientation, and so possibly reduced occurrences of mind wandering and improved learning. In the Game-based condition and Self-quizzing conditions, compared with the Study-only condition, bilateral activation of the VS was observed when participants were responding during a test of their knowledge. This may reflect dopaminergic activity in response to anticipated outcome, even though, in this epoch-related study, this activity is being captured with a temporal resolution that is very limited (In this study, we attempted to provide participants with a well-paced learning and gaming experience similar to that which might be provided for educational purposes. This prevented the inclusion of jittering which would enable reconstruction of temporal changes in event-related BOLD response beyond the resolution of the repetition time). If rehearsal of learning occurred simultaneously with such a dopaminergic response, this might have contributed to the greater learning achieved in the more gamified conditions. Activation of this region in response to cues indicating monetary incentives for remembering has been found to have roughly linear relationship with the likelihood of subsequent recall (
Sorting and comparison of trials based on correctness of question response revealed increased ventral striatal activation bilaterally for positive feedback in both the Game-based condition and in the left VS for the Self-quizzing condition. Understanding of the brain’s reward circuitry has been established chiefly through its robust response to physical pleasures, and recent studies have shown that the same networks are activated in response to social rewards such as praise (
We have chiefly discussed improvements in learning observed in the Game-based condition in terms of reward system response, but it is important to point out that these effects might, at least in part, be explained in other ways. Alongside improvements in self-reported engagement, self-reported stress was also greater in the Self-quizzing condition compared with the Study-only condition, and increased further in the Game-based condition. All three defining characteristics of the Game-based condition (competition, escalation, and uncertainty) might conceivably have contributed to this stress. The stress would have been experienced in the context and close in time to the “to be remembered” material, and may have triggered hormones (corticosteroids, noradrenaline, corticotropin releasing hormone) suitable for enhancing memory (
We have demonstrated links between deactivation of the DMN and educational learning. We believe our results support the proposed usefulness of the concepts and techniques of cognitive neuroscience in education, and particularly in regard to the design of technology-enhanced learning (
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The authors are grateful to staff at the Clinical Research and Imaging Centre (CRICBristol) for their expertise and assistance with MRI scanning and to all participating students.