User Assessment in Serious Games and Technology-Enhanced LearningView this Special Issue
Research Article | Open Access
Lotfi Derbali, Claude Frasson, "Assessment of Learners’ Motivation during Interactions with Serious Games: A Study of Some Motivational Strategies in Food-Force", Advances in Human-Computer Interaction, vol. 2012, Article ID 624538, 15 pages, 2012. https://doi.org/10.1155/2012/624538
Assessment of Learners’ Motivation during Interactions with Serious Games: A Study of Some Motivational Strategies in Food-Force
This study investigated motivational strategies and the assessment of learners’ motivation during serious gameplay. Identifying and intelligently assessing the effects that these strategies may have on learners are particularly relevant for educational computer-based systems. We proposed, therefore, the use of physiological sensors, namely, heart rate, skin conductance, and electroencephalogram (EEG), as well as a theoretical model of motivation (Keller’s ARCS model) to evaluate six motivational strategies selected from a serious game called Food-Force. Results from nonparametric tests and logistic regressions supported the hypothesis that physiological patterns and their evolution are suitable tools to directly and reliably assess the effects of selected strategies on learners’ motivation. They showed that specific EEG “attention ratio” was a significant predictor of learners’ motivation and could relevantly evaluate motivational strategies, especially those associated with the Attention and Confidence categories of the ARCS model of motivation. Serious games and intelligent systems can greatly benefit from using these results to enhance and adapt their interventions.
It is widely acknowledged that learners’ psychological and cognitive states have an important role in intelligent systems and serious games (SGs). For instance, engagement and motivation or disaffection and boredom obviously affect learners’ wills and skills in acquiring new knowledge . SGs cannot, therefore, ignore these states and should take them into account during learning process. One important learners’ state is motivation which plays a crucial role in both the learners’ performance and the use of intelligent systems over time . Motivation is generally defined as that which explains the direction and magnitude of behaviour, or in other words, it explains what goals people choose to pursue and how they pursue them . It is considered as a natural part of any learning process. Several researches have showed that motivated learners are more likely to be more engaged, to undertake challenging activities, and to exhibit enhanced performance and outcomes [4, 5]. Therefore, it is of particular relevance to study motivation and its role in improving learners’ performance during gameplay.
Learners’ interactions with Intelligent Tutoring Systems (ITSs) and especially SGs have always been considered to be intrinsically motivating. One possible explanation is the fact that ITSs generally use pictures, sounds, videos, and so forth, that are considered, crudely, as motivational factors. Intrinsic motivation is possibly gained through challenge, curiosity, control, sensory stimuli, interaction, and fantasy when using SGs [6, 7]. However, many researchers have argued that learners’ negative emotions or amotivational states such as boredom or disengagement have been known to appear following a certain period of interaction with computer systems. These states can be overwhelming to learners and cause motivational problems and decrease learning benefits [2, 8, 9]. Once learners’ psychological and cognitive states are identified, intelligent systems are in a much better position to act upon them. In this perspective, several studies have described intelligent systems that can provide adapted emotional or cognitive strategies for coping with, or at least reducing, the negative learners’ states [8–12]. Computer systems can also use motivational strategies which are the actions (or tactics) taken in order to scaffold learners’ motivation toward tasks and goals of learning process and to make learning easier, faster, more enjoyable, more self-directed, and more effective. However, it is surprising to find so little mention of the motivational strategies and relatively little is known about coping with motivational problems, which motivational strategies should be used, and to what extent they are employed. Within the researchers who have tackled this issue, some have found that SGs seemed to show a promising potential from a motivational standpoint. It has been consistently shown that SGs have inherent motivational properties and different strategies, allowing them to be used for improving educational applications [7, 13–15]. Game designers, for example, employ a range of Artificial Intelligence (AI) techniques (e.g., controlling the behaviour of the nonplayer characters, providing performance feedback) to promote long-term user engagement and motivation .
Moreover, evaluating systems interventions is obviously related to differences in learners’ performance (successful completion of tasks) or judgement (self-report questionnaires). However, by using only performance or judgement in evaluating motivational strategies, intelligent systems risk obtaining delayed or imperfect evaluations or interrupting learning process by repeatedly using self-report questionnaires. This may offer misleading information regarding the impact of motivational strategies on learners’ motivation. Therefore, it is of particular relevance to investigate new ways of evaluating motivational strategies. One promising way is the use of physiological sensors. This is notably explained by the significant results of recent studies involving physiological sensors to assess motivational learners’ states as well as emotional and cognitive systems strategies [10, 17–19].
The present paper examines the implication of different physiological sensors to evaluate some motivational strategies employed in SGs and to highlight the corresponding learners’ patterns. To this end, we use an existing SG called Food-Force presented by the United Nations World Food Programme (WFP) and intended to learn players about the fight against world hunger. The ultimate objective of this intervention study is to assess learners’ motivation when motivational strategies haves been used by SGs. We ask the two following research questions: Can we empirically find physiological patterns to evaluate the effects of motivational strategies on learners’ motivation during interactions with Food-Force? If so, Can these patterns feed AI models to predict the level of learners’ motivation to the motivational strategies? Hence, two hypotheses are postulated: (1) it is possible to model learners’ physiological reactions and trends towards motivational strategies in an SG environment and (2) we can discriminate between effective and ineffective motivational strategies using physiological manifestations as well as self-report questionnaires. We designed an experiment using an existing SG called Food-Force and combined both the theoretical ARCS model of motivation and empirical physiological sensors (heart rate (HR), skin conductance (SC), and electroencephalogram (EEG)) to assess the effects of motivational strategies on learners.
The organization of this paper is as follows. In the next section, we present previous work related to our research. In the third section, we explain our empirical approach in assessing learners’ motivation. In the fourth section, we describe the theoretical ARCS strategies to support motivation and the studied strategies in Food-Force. In the fifth section, we detail our experimental methodology. In the sixth section, we present the obtained results and discuss them. Finally, we give a conclusion in the last section, as well as present future work.
2. Related Research
Csikszentmihalyi  observed that people enter in a “flow” state when they are fully absorbed in activity during which they lose their sense of time and have feelings of great satisfaction. Games generally catalyze conditions of flow state by their clear goals, balance between challenges and skills, immediate feedback, progress, and control. Furthermore, Ryan and Deci  defined the Self-Determination Theory (SDT) and distinguished between intrinsic motivation (to understand the subject) and extrinsic motivation (for the reward of a certificate or employment). They assumed that the individual is normally inclined to be active, motivated, curious, and eager to succeed. They also recognized that some people mechanically perform their tasks, or even people passive and unmotivated. They reported that environments that facilitate the satisfaction of psychological needs can boost the internal dynamism of people to maximize their motivation and to maximize results in personal development and behavior. Ryan and colleagues  have studied the SDT and stated that the motivational pull of computer games is attributed to the combination of optimal challenge and informational feedback. Bartle , one of the pioneers of the massively multiplayer online games and known for his work on the first MUD (An MUD (originally Multiuser dungeon, with later variants Multiuser Dimension and Multiuser Domain) is a computer program, usually running over the Internet, that allows multiple users to participate in virtual-reality role-playing games.), distinguished several motivational profiles among players: the killer (competitiveness), the per-former (success), the explorer (curiosity), and socializer (cooperation). He reported that some players strive to achieve all the challenges offered by the gameplay while others seek the company of other players, or want to discover the whole virtual world.
Nevertheless, the effectiveness of any study regarding the assessment of learners’ motivational states depends on two important factors: (1) the choice of proper assessment tools and (2) the accuracy of the selected tools. For example, Schunk et al.  used Keller’s ARCS model (see next section for a description of this model) and proposed several rules to infer motivational states from two sources: the interactions of the students with the tutoring system and their motivational traits. Some researchers have analyzed log files and have established correlations between learners’ actions in log files and their motivational states (e.g., ). Other researchers have used physiological sensors to assess learners’ motivation and correlate physiological learners’ responses to some dimensions of motivation such as attention and confidence (e.g., [17, 25]). They have identified that the combination of various physiological sensors may provide perfect measures of learners’ states and consequently enhance systems intervention strategies. They have involved a variety of sensors to assess physiological learners’ states and responses to stimuli in Computer-Based Education (CBE) environment: mouse, electromyogram (EMG), respiration (RESP), HR, SC, and more recently EEG. For instance, Conati  has used biometric sensors (HR, SC, EMG, and RESP) and facial expression analysis to develop a probabilistic model of detecting students’ affective states within an educational game. Arroyo and colleagues  have used four different sensors (camera, mouse, chair, and wrist) in a multimedia adaptive tutoring system to recognize students’ affective states and embed emotional support.
Others studies have shown that learners have also been known to experience a lower sense of relatedness to the educational systems , thus increasing their feeling of isolation and possibly leading to further motivational issues. For example, learners lack the substantial self-monitoring skills that CBE systems require and possibly start “gaming” the system . In addition, CBE systems generally place fewer restrictions on learners and learners must take greater responsibility for their educational experiences. Hannafin and colleagues  recommended that students need more support and must be empowered to acquire the necessary skills to effectively progress in an educational environment. For example, it has been found that when collaborative learning strategy is used, a fewer errors are made than in individual learning situations, resulting in better outcomes performance, increased confidence, and decreased frustration levels of the learners [29, 30]. Dörnyei  has reported that motivational strategies are used not only to maintain students’ motivation but also to generate and increase it. He has defined that more than one hundred motivational strategies can be used by teachers in the classroom. These strategies integrate the creation of the basic motivational conditions, the generation of initial motivation, the maintaining and the protection of motivation, and the encouragement of positive and retrospective self-evaluation. Furthermore, efforts to overcome learners’ motivational problems have mainly been focused on tutor’s strategies or instructional design aspects of the systems. For example, Hurley  developed interventional strategies to increase the learner’s self-efficacy and motivation in an online learning environment. She extracted and then validated rules for interventional strategy selection from expert teachers by using an approach based on Bandura’s Social Cognitive Theory and by observing the resulting learners’ behaviour and progress. Goo and colleagues  showed that tactile feedback, sudden view point change, unique appearance and behaviour, and sound stimuli played an important factor in increasing students’ attention in virtual reality experience. Arroyo and colleagues  evaluated the impact of a set of noninvasive interventions in an attempt to repair students’ disengagement while solving geometry problems in a tutoring system. They claimed that showing students’ performance after each problem reengages students, enhances their learning, and improves their attitude towards learning as well as towards the tutoring software.
3. Assessment of Learners’ Motivation
There are a few studies that have particularly considered the evaluation of motivational strategies. It is of particular significance for this research work that motivational strategies are identified and their impacts on learners are evaluated in a specific CBE environment, precisely serious games. In addition to ARCS self-reported questionnaires, the present study uses three physical sensors (SC, HR, and EEG) to assess motivational strategies while interacting with a serious game called Food-Force. We first need to present the tools used to measure motivation itself.
3.1. ARCS Model of Motivation
In the present study, the ARCS model of motivation  has been chosen to theoretically assess learners’ motivation in SG. Keller used the existing research on psychological motivation to identify four components of motivation: Attention, Relevance, Confidence, and Satisfaction. His model has been used in training and games and has also been validated in numerous studies with all education levels and in many different cultures (e.g., [3, 34, 35]), and therefore, it is of particular interest in our study.(i)Attention: to attract learners’ attention at the beginning and during the process of learning. Diverse activities should be considered to maintain students’ feelings of novelty thus the attention can be sustained.(ii)Relevance: to inform learners of the importance of learning and to explain how to make the learning meaningful and beneficial.(iii)Confidence: to allow learners to know the goal and to believe that the goal can be achieved, if enough effort (physical and/or intellectual) has been made.(iv)Satisfaction: to provide feedback on performance and to allow learners to know how they are able to perform well and apply what is learned in real life situations.
The ARCS questionnaire asks students to rate ARCS-related statements in relation to the instructional materials they have just used. Examples of items related to each ARCS component are as follows.(i)“uses questions to pose problems or paradoxes.” (Attention);(ii)“uses language and terminology appropriate to learners and their context.” (Relevance);(iii)“provides feedback on performance promptly.” (Confidence);(iv)“makes statements giving recognition and credit to learners as appropriate.” (Satisfaction).
3.2. Physiological Sensors
Considering the motivation as a state of both cognitive and emotional arousal, we have decided to combine several noninvasive physiological sensors in order to empirically evaluate the motivational strategies in serious games context. Besides the SC and HR sensors which are typically used to study human affective states , we have considered relevant to use the EEG sensor in our proposed approach. Indeed, brainwave patterns have long been known to give valuable insight into the human cognitive process and mental state . More precisely, our EEG analysis relies on differences between slow and fast wave ratios (i.e., “attention ratio” or Theta/low-Beta) which are correlated with responses to motivational stimuli and emotional traits [38, 39]. For instance, low-level attention is characterized by “a deviant pattern of baseline cortical activity, specifically increased slow-wave activity, primarily in the Theta band, and decreased fast-wave activity, primarily in the Beta band, often coupled” . The power of the EEG “attention ratio” can be explained by Putman and colleagues . According to the authors, a negative correlation exists between the attention ratio and learners’ Attention level. A high Theta/low-Beta ratio is usually correlated with excessive Theta and consequently inattentive state. Conversely, a low Theta/low-Beta ratio is normally correlated with excessive Low-Beta brainwave activity reflecting normal state in adults.
4. Motivational Strategies
The key issue in this paper is related to the identification and assessment of motivational strategies in SGs that support and enhance learners’ motivation. We define a motivational strategy as the use of a game element (or factor)  susceptible of providing motivational support for players. Motivational strategies in SG are the key to finding and harnessing players’ motivation to learn and achieve their goals. For example, a virtual companion in SG can offer encouragement to players as well as offering additional aid in their current task. This is can be considered as a motivational strategy related to the Confidence category of the ARCS model only if it increases learners’ belief in competence and consequently their effectiveness. Otherwise, it is simply an SG element and not a motivational strategy. Each of the four categories has also subcategories that are useful in identifying learners’ motivational profiles and in creating motivational tactics (or strategies) that are appropriate for specific situations in SG .
Attention Getting Strategies (Capture Interest, Stimulate Inquiry, and Maintain Attention)
Before any learning can take place, the learner’s attention must be engaged. The challenge with the attention is to find the right balance of consistency, novelty, and understanding how people differ, what tactics to use, and how to adjust the tactics for the learners and how the tutor will be able to keep them focused and interested.
Relevance Producing Strategies (Relate to Goals, Match Interests, and Tie to Experiences)
It is very difficult for students to be motivated to learn if they do not perceive there to be any relevance in the instruction. To stimulate the motivation to learn, it is best to build relevance by connecting instruction to the learners’ backgrounds, interests, and goals.
Confidence Building Strategies (Success Expectations, Success Opportunities, and Personal Responsibility)
When people believe that they have little or no control over what happens to them, they experience anxiety, depression, and other stress-related emotions. In contrast, when they believe that they can predictably influence their environment by exercising their efforts and abilities in pursuit of their goals, then they are more motivated to be successful.
Satisfaction Generating Strategies (Intrinsic Satisfaction, Rewarding Outcomes, and Fair Treatment)
One of the most important elements of satisfaction is intrinsic motivation; that is, if learners believe that they achieved a desirable level of success while studying topics that were personally meaningful, then their intrinsic satisfaction will be high. Another component of satisfaction is based on social comparisons and comparisons to expected outcomes.
The present study invited participants to play the freely downloadable SG called Food-Force . It is an initiative of the World Food Program (WFP) of the United Nations intended to educate players about the problem of world hunger. Food-Force is comprised of multiple arcade-type missions, each intended at raising players’ awareness towards specific problems regarding worldwide food routing and aid. Food-Force also presents players’ objectives in a short Instructional Video before the beginning of each mission. A virtual tutor also accompanies the player throughout each mission by offering various tips and lessons relative to the obstacles and goals at hand. All participants have never played Food-Force before. We have investigated in details six motivational strategies in Food-Force in order to answer our main research question. (Can we empirically find physiological patterns to evaluate motivational strategies during serious game play?) These strategies are related to the four categories of the ARCS model, see Figure 1.
Keller’s ARCS motivation theory tells us that the learner’s motivation is also aroused by the mean of “solving a problem or resolving an open issue…” called inquiry arousal. Mission 2 (nutritious meal preparation) presents to learners a challenging problem that consists of finding the right combination of different food items (rice, beans, vegetable oil, sugar, and iodized salt) to create a nutritious and balanced diet, all at a target cost of 30 US cents per person per meal. It has been investigated in our experiment to study the Problem Solving strategy used by Food-Force.
According to Brophy , situational interest is triggered in response to something in the situation (e.g., unexpected sound) that catches our attention and motivates us to focus on it and explore it further. Keller’s ARCS motivation theory also argues that the learner’s motivation is possibly gained by a perceptual arousal (novel, surprising, or incongruous events). We have decided to investigate the three Alarm Triggers as a strategy supporting motivation in mission 5 (UN. food delivery) of Food-Force. An example of an alarm trigger is shown in Figure 2.
Motivational strategies rely on some game elements that make a lesson content relevant to the learners. Keller has reported that the instructor has to tie instruction into the learner’s experience by providing examples that relate to the learner’s work. In Food-Force, Instructional Video segments that draw on players’ existing knowledge have been used in order to show them the real application of presented mission in the field and connect each mission to the problem of world hunger.
It is important to raise learner’s confidence by offering suitable feedback. According to , negative or positive Informative Feedback tells learners what they are doing. It works much better then Controlling Feedback which simply tells them what to do. For Informative Feedback used in Food-Force, comments like “What was a dangerous drop! Try to be more accurate and watch the wind gauge” (mission 3) or “Won’t arrive immediately, but that might be ok for you” (mission 4) indicate the effects (or benefits) of actions taken by the player.
The learner is open to a brief instructional explanation that will help build the right mental model and/or correct misconceptions. Explanatory Feedback resulted in much better learning than Corrective Feedback , which can be automated in many authoring tools with only a few key strokes. The virtual companion of Food-Fore makes comments, such as “Rice: we need a lot of rice. It provides nutrition and energy” (mission 2), to explain user actions.
Motivational strategies aimed at increasing learner’s satisfaction usually focus on allowing students to display their work, encouraging them to be proud of themselves and celebrate success, and using rewards. Displaying Score strategy is used at the end of each mission in order to show players their current scores and their overall progress.
5. Experimental Methodology
Thirty-three volunteers (11 females) took part in the study in return of a fixed compensation. Participants were recruited from the University of Montreal. The sample’s mean age was 26.7 (SD = 4.1). Following the signature of a written informed consent form, each participant was placed in front of the computer monitor to play the game. Set on a fictitious island called Sheylan riven by drought and war, Food-Force invites participants to complete 6 virtual missions that reflect real-life obstacles faced by WFP in its emergency responses both to the tsunami and other hunger crises around the world. All participants have played only the first five missions of Food-Force. A pretest and posttest were also administered to compare learners’ performance regarding the knowledge presented in the serious game. We have used 10 multiple choice questions about general problem of world hunger. Figure 3 presents a flow diagram of the experiment.
5.2. Data Collection
The motivational measurement instrument called Instructional Materials Motivation Survey (IMMS) was used following each mission to assess learners’ motivational state. It is derived from four categories of ARCS motivation model. Due to time constraints and in order to achieve minimum disruption to learners, we used a short IMMS form which contained 16 out of the 32 items after receiving the advice and approval from Dr. John Keller. IMMS used a 5-point Likert-type scale (where 1 is strongly disagree and 5 is strongly agree). Furthermore, two cameras were also used to simultaneously record learners’ facial expressions and game progress. Physiological data was also recorded in synchrony to both camera feeds throughout the experiment. The Galvanic Skin Resistance (GSR) electrodes and the blood volume pulse (BVP) sensor were attached to the fingers of participant’s nondominant hands, leaving the other hand free for the experimental task. BVP sensor is a blood volume pulse detection sensor housed in a small finger worn package, to measure heart rate (HR). GSR electrodes measure the conductance across the skin (SC). An EEG cap was also conveniently fitted on participant’s head and each cerebral sensor spot slightly filled with a saline nonsticky solution. EEG refers to the recording of the brain’s spontaneous electrical activity over a period of time as recorded from multiple electrodes placed on the scalp. HR, SC, and EEG recordings were managed by the Thought Technology ProComp Infiniti Encoder . This encoder has 8 protected pin sensor inputs with two channels sampled at 2048 samples per second and six channels sampled at 256 samples per second. The first two channels were used to record HR and SC. The last six other channels were used to record EEG at sites Fz, F3, C3, Pz, A1, and A2 according to the international 10–20 system. A ground is located at Fpz. Electrode impedances were maintained below 5 KΩ. Participants were asked to minimize eye blinks and muscle movements during physiological recordings. Furthermore, an additional notch filter is typically used to remove artifact caused by electrical power lines (60 Hz in Canada). According to , “the reference electrode should be placed in a location that is not susceptible to artifact. An extra midline electrode is suitable”. Thus all EEG sites were referenced online to Cz. Electrophysiological data were recorded during the whole of the experiment. A 60s-baseline was also computed before the beginning of the game.
5.3. Data Analysis
The offline processing of the HR, SC, and EEG data was performed using BioGraph Infiniti software. EEG data were rereferenced offline to the mean of the activity at the two mastoid leads (A1 and A2). For each site , the corrected is calculated using the following formula:
Four participants (2 females) were excluded from the EEG analysis because of technical problems at the time of recording. Technical Fz recording problem with some participants leads us to exclude all Fz data from our analysis. Furthermore, manual editing of the recorded signals has been carried out to remove artifact-contaminated data caused by muscle activity and eye blinks or movements. The EEG raw signal is filtered through a band pass filter from 2 to 48 Hz. A necessary normalization technique (min-max ) was applied to HR and SC physiological data using the baseline data. Indeed, normalizing the data keeps the physiological patterns for individual subjects and establishes a common metric for intersubject comparisons. Min-max normalization performs a linear transformation on the original data. It has the advantage of preserving exactly all relationships in the data. We have normalised each HR and SC data using the following modified formula : where and refer, respectively, to maximum and minimum values during interaction period and baseline refers to the average value of physiological data before the beginning of the game. These normalized physiological data reflect signal changes from baseline.
EEG data were segmented into one-second epochs and power spectral densities were calculated for each epoch using Fast Fourier Transformation. Power spectral data were averaged within Theta (4–8 Hz) and Low-Beta (12–20 Hz) bands. For each epoch of every participant the attention ratios (Theta/low-Beta) were calculated as described in Section 3.
5.4. Percent of Time (PoT) Index
We have defined an index representing players’ physiological evolution throughout each mission of the serious game with regards to each signal signification. This index, called Percent of Time (PoT), represents the amount of time, in percent, that player’s signal amplitude is lower (or higher) than a specific threshold. The PoT index is a key metric enabling us to sum up players’ entire signal evolution for a mission. A simple method would be to choose the mean players’ signal amplitude of each physiological sensor as the threshold. The PoT index of HR (or SC) for each player was calculated using values above the HR (or SC) threshold, whereas the PoT index for each EEG sites was calculated when player’s attention ratio was below the threshold since we are looking for positive evolutions. Figure 4 illustrates an EEG attention ratio evolution during 20 seconds. The PoT for the selected 5-second window was 80% (4 values below divided by 5 values) and 70% for the entire 20 seconds (14 values below divided by 20 values).
The idea is to analyze, in a joint venture, PoT indexes of HR, SC, and EEG signals to determine, or at least estimate, relations between the motivational strategies used in the serious game and the physiological learners’ responses. To that end, various AI models have been constructed using gathered data in order to classify learners in two distinct classes: “Below” and “Above”. Indeed, subjects have been separated into two groups based on their self-reported scores of the ARCS model after each mission of game: those with scores below the overall average (group “Below”) and those with scores above the overall average (group “Above”). For instance, the evaluation of an Attention Getting strategy (e.g., Alarm Trigger or Problem Solving) used by Food-Force will consider the Attention scores to determinate the “Below” and “Above” groups of subjects and compare their physiological reactions . The same procedure has been applied for all other strategies. Consequently, the members of each group are different from one strategy to another. A detailed description of all these possibilities is given in the following section.
6. Experimental Results
Before presenting our results, we considered it necessary to quickly explain the statistical approach used in this section. Indeed, we could not rely on the usual parametric statistical tools such as ANOVA and t-test because (1) our sample population is small ( participants), (2) no justifiable assumptions could be made with regards to the normal distribution of the data, and (3) normality tests run on our data confirmed its nonnormal distribution. Hence, nonparametric Friedman’s ANOVA by ranks (counterpart of the parametric one-way ANOVA) and nonparametric Wilcoxon’s signed ranks test (counterpart of paired sample t-test) have been used. However, value is interpreted in the same manner in both approaches and to that effect, reported significant values were all computed at the 0.05 significance level (95% confidence).
6.1. Performance and Motivation
In order to determine if the IMMS scale is reliable, a Cronbach’s Alpha was run on IMMS data gathered after the first mission of Food-Force. The simplified IMMS yielded reliability (Cronbach’s Alpha coefficient) of 0.88 for the overall motivation measure and Cronbach’s Alpha for Attention, Relevance, Confidence, and Satisfaction was 0.91, 0.71, 0.79, and 0.87, respectively. These reliability coefficients are analogous to those found in  and showed that the motivational measurement instrument used in the present study was highly reliable.
Since we intend to study several motivational strategies in different missions within the Food-Force game, we evaluated the effects of these strategies on learners’ performance as well as their motivation. We have then conducted statistical tests and we have obtained several results regarding knowledge acquisition (pre- and posttests) and learners’ motivation (ARCS scores). The results of Wilcoxon signed ranks test displayed in Table 1 showed a significant difference between the participants’ scores of the pre- and posttests in terms of knowledge acquisition (, ). Number of correct answers after finishing the game is significantly higher than that of correct answers before start playing. The results of Friedman’s ANOVA by ranks between ARCS scores are displayed in Table 2. Significant differences for the general motivational scores as well as each category of the ARCS model were also observed between missions, except for Relevance (motivation overall score: F(1,4) = 10.16, ; Attention: F(1,4) = 19.51, ; Relevance: F(1,4) = 7.38, ; Confidence: F(1,4) = 16.8, ; Satisfaction: F(1,4) = 10.85, ). Nonsignificant results of the Relevance category can be explained by the fact that the Relevance Producing strategy (Instructional Video) presented between missions was roughly the same: video segments explain the goal of each mission or its real application in order to connect each mission to the problem of world hunger. Conversely, the Attention category which showed the strongest difference and rank has used various game strategies throughout the missions. Indeed, Food-Force maintains learners’ attention by using Alarm Trigger when they are confronted with an unexpected situation such as attacks to the convoy by local rebel forces or flat tires of trucks (mission 5). It also includes mental tasks that require concentration and attention: drop food from the air without risking human lives (mission 3) and guide a convoy of trucks safely to a feeding centre while overcoming challenges from clearing land mines to rebuilding bridges and negotiating with local rebel forces (mission 5). Finally, learners’ attention is possibly gained by using Problem Solving strategy such as finding the right combination of different food items (rice, beans, vegetable oil, sugar, and iodized salt) to create a nutritious and balanced diet, all at a target cost of 30 US cents per person per meal (mission 2).
*Significance at the .05 level.|
*Significance at the .001 level.|
*Significance at the .01 level.|
*Significance at the .05 level.|
*Significance at the .01 level.|
*Significance at the .05 level.|
An example of an Alarm Trigger used in mission 5 is shown in Figure 2. As described in Section 4, Alarm Trigger is a motivational strategy (Attention Getting strategy) associated to Attention category of the ARCS model. We have then considered self-reported Attention scores to separate participants into two groups: a “Below” class (4 females and 7 males) representing participants who reported an Attention score below that of the overall mean and an “Above” class (5 females and 13 males) presenting the opposite (a score above the overall mean). Three alarms in mission 5 have been investigated. They are a sound trigger followed by Food-Force logistics officer’s comments used to help players to overcome challenges—from clearing land mines to rebuilding bridges and negotiating with local rebel forces. To detect physiological changes for each player, we considered two 5-second windows computed before and after each alarm and calculated their means (, ). Fifteen Wilcoxon signed ranks tests (3 alarms × 5 physiological sensors) were run between Before Alarm and After Alarm data and significant results were obtained for all data. The 3 Alarms Triggers and learners’ physiological trends are presented in Figure 5. Each dot on the graph represents the difference between the two means for each alarm (). Figure 5 shows almost complete opposite trends for all physiological data between the “Below” and “Above” classes, except for SC. The physiological analysis pointed towards the fact that the “effect” of an Alarm Trigger seems to decrease over time. We can see on Figure 5(a) that the effect of those alarms on SC seems to slowly fade after the second alarm, contrary to the popular belief. Indeed, one may think that intervening with color and sound tends to capture learners attention, but our findings seem to indicate that this is only partially true. There seems to be a certain “adaptation” on the part of the learner with regards to SC at the very least. Nevertheless, any permanent diagnosis regarding learners’ attention level in reaction to an Alarm Trigger based only on SC at this point may be hasty or even wrong for there are numerous other physiological trends to consider first. Indeed, even if no clear trends were found in HR, the cerebral data provided clarity in distinguishing between the two classes.
In fact, variations in the attention ratio are clearly evident for both classes. We found numerous occasions when two participants from different classes had the same SC and HR trends but have shown very opposite trends in EEG sites, especially C3 area. An example of this situation is illustrated in Figure 6: two participants had the same HR and SC trends but only an opposite trend in C3 helped us identify their respective attention classes. These results seem to show the relevance and importance of adding the EEG in assessing learners’ attention change, even more so when this change cannot be clearly established by the use of HR and SC alone. Thus, the EEG “attention ratio” generally increases for participants who reported a low Attention category score (class “Below”) whereas the same ratio decreases for the learners in the class “Above”.
6.2. Logistic Regression Analysis
Subjects have been separated into two groups according to their ARCS scores after each mission: those with scores below the overall average (group “Below”) and those with scores above the overall average (group “Above”). We have run logistic regressions to predict learners’ group (“Above” or “Below”) for each studied strategy. The dependent variable in logistic regression is usually dichotomous, that is, the “Above” group coded as “1” whereas the “Below” group coded as “0”. Furthermore, logistic regression makes no assumption about the distribution of the independent variables. These variables do not have to be normally distributed, linearly related or of equal variance within each group. Our prediction models used all computed PoT indexes as predictor variables (PoT-SC, PoT-HR, PoT-F3, PoT-C3, and PoT-Pz) and the Enter method for variable selection. Table 3 reports the results of adding five predictors (df = 5) to the regression model. Results indicated that adding predictors to the model has significantly increased our ability to distinguish between “Above” and “Below” groups for all studied motivational strategies, except for Instructional Video (see Table 3: Chi-Square and Sig. values with conventional significance level of 0.05). In addition, Nagelkerke’s values of Table 3 ranged from 46% to 65% and indicated a moderately high relationship between the predictors and the dependent variable. Table 5 showed the classification tables which tell us how many of the cases where the observed values of the dependent variable were 1 or 0, respectively have been correctly predicted. In each classification table, the columns are the two predicted values of the dependent, while the rows are the two observed values of the dependent. Prediction success overall was between 65.5% and 79.3% (see Table 5). The Wald criterion demonstrated that PoT-C3 especially made a significant contribution to prediction (see Table 4). Other variables were not significant predictors. Results of regression models clearly showed that physiological data, especially EEG “attention ratio”, were relevant to evaluate motivational strategies. The most significant differences between groups were shown for Attention Getting Strategies though. One reason may be the limitation of the “attention ratio” (Theta/low-Beta) which seems to be inappropriate to identify EEG patterns other than those correlated with the Attention category. Regarding the physiological analysis, it is preferable to explore alternative EEG frequency ratios based on additional brainwaves such as Alpha (8–12 Hz) and High-Beta (20–32 Hz) in order to highlight other patterns correlated with learner’s motivation. Furthermore, F3 and C3 areas showed more significant differences of PoT trends than Pz area which showed roughly similar trends between groups. This is can be explained by specific functions associated with the middle parietal (Pz) area. These functions incorporate appreciation of form, sensory combination and comprehension (pain, pressure, heat, cold, and touch) which are quite sparse or even absent in all missions. Learners tended to rely mostly on the frontal cortex (F3) because it is known to be strongly implicated in taking quick decisions under pressure. The central region of the brain (C3) seems to be the most solicited when a more “generalized” problem solving approach is used. Not only our results show that physiological data can provide an objective evaluation of motivational strategies for clearly distinguishing between learners’ reactions, but also the relevance and importance of adding the EEG in our empirical study. The obtained results also open the door to the possibility to evaluate other motivational strategies used in different intelligent systems.
*Significance at the .05 level.|
*Significance at the .05 level.|
7. Conclusion and Future Work
In this paper, we have assessed the effects of some motivational strategies in Food-Force on learners’ motivation using the ARCS theoretical model as well as three physiological sensors: HR, SC, and EEG. We have successfully answered our first research question by identifying physiological patterns, especially EEG Theta/low-Beta ratio, to evaluate motivational strategies. We then successfully answered our second research question by using these physiological trends to build prediction models of learners’ motivation. These models were able to moderately distinguish motivating strategies from those with low impacts on learners’ motivation. Our findings showed that SC and HR may reach their limits in some cases for evaluating the impacts of motivational strategies on learners. In fact, no clear trends were found in SC and HR for evaluating some studied strategies. However, C3 Theta/low-Beta ratio has showed different trends between groups for almost all studied strategies. It can give valuable evaluation of motivational strategies.
Statistical and physiological study of our data has given some insights into the assessment of learners’ motivation during playing a serious game. It has shown that physiological parameters are suitable to assess the effects of motivational strategies on learners’ motivation. The obtained results are very encouraging to an ITS because (1) it is possible to assess the effects of tutor’s interventions on learners’ motivation, (2) we can rely on this assessment as a substitute for self-reports that can disrupt a learning session, and (3) it is possible to enrich the Learner Model (which describes learners’ behaviors and evaluates their knowledge) with a motivational component based on our results, thus enabling the Tutor Model (which uses the Learner Model and customizes learning environments by adapting learning strategies in order to respond intelligently to learners’ needs, objectives, and interests) to properly adapt its interventions.
However, one limitation in this work is the assumption that the ARCS categories are independent from each other. Simultaneous strategies in SG can be related to different categories of the ARCS model. One possible extension of the present work would be to consider dependencies between ARCS categories. In addition, we can extend the present work to study more than two classes of motivation. Multinomial logistic regression will be used in this case in counterpart of binary logistic regression. It is also possible to add other variables that can improve the prediction quality of our models. Indeed, some personals characteristics (age, gender, player style, hours spent playing video games, etc.) can be additional predictors for players’ motivation. Furthermore, brain activity can also be better analysed in the future and other EEG analysis methods, such as the event-related potential (ERP) technique, can be used to test whether different events in serious game evoke differential EEG responses. We plan, therefore, to address all these possibilities in a further complementary study.
Conflict of Interests
The authors declare that they have no conflict of interests in the research.
The authors would like to thank all the reviewers for their insightful and helpful comments. They acknowledge the support of the Tunisian government and the National Science and Engineering Research Council (NSERC) of Canada for this work. They also address their thanks to Pierre Chalfoun for his participation in the experiment setup and his useful comments.
- A. Bandura, Social Foundations of Thought and Action: A Social Cognitive Theory, Prentice-Hall, Englewood Cliffs, NJ, USA, 1986.
- A. de Vicente and H. Pain, “Informing the detection of the students' motivational state: an empirical study,” in Proceedings of the 6th International Conference on Intelligent Tutoring Systems, pp. 933–943, Springer, Biarritz, France, 2002.
- J. M. Keller, Motivational Design for Learning and Performance: The ARCS Model Approach, Springer, New York, NY, USA, 2010.
- T. S. Chan and T. C. Ahern, “Targeting motivation—adapting flow theory to instructional design,” Journal of Educational Computing Research, vol. 21, no. 2, pp. 151–163, 1999.
- D. H. Schunk, P. R. Pintrich, and J. L. Meece, Motivation in Education: Theory, Research and Applications, Pearson/Merrill, Upper Saddle River, NJ, USA, 3rd edition, 2008.
- M. Prensky, Digital Game-Based Learning, McGraw Hill, New York, NY, USA, 2001.
- R. Garris, R. Ahlers, and J. E. Driskell, “Games, motivation, and learning: a research and practice model,” Simulation and Gaming, vol. 33, no. 4, pp. 441–467, 2002.
- I. Arroyo, K. Ferguson, J. Johns et al., “Repairing disengagement with non invasive interventions,” in Proceedings of the 13th International Conference on Artificial Intelligence in Education, pp. 195–202, IOS Press, Los Angeles, Calif, USA, 2007.
- R. S. J. D. Baker, A. T. Corbett, K. R. Koedinger et al., “Adapting to when students game an intelligent tutoring system,” in Intelligent Tutoring Systems, pp. 392–401, Springer, Jhongli, Taiwan, 2006.
- S. Chaffar, L. Derbali, and C. Frasson, “Inducing positive emotional state in intelligent tutoring systems,” in Proceedings of the 14th Conference on Artificial Intelligence in Education, IOS Press, 2009.
- G. Rebolledo-Mendez, B. du Boulay, and R. Luckin, “Motivating the learner: an empirical evaluation,” in Proceedings of the 8th Conference on Intelligent Tutoring Systems, pp. 545–554, Springer, Berlin, Germany, 2006.
- T. Hurley, “Intervention strategies to increase self-efficacy and self-regulation in adaptive on-line learning,” in Adaptive Hypermedia and Adaptive Web-Based Systems, pp. 440–444, Springer, Berlin, Germany, 2006.
- T. Barnes, E. Powell, A. Chaffin, and H. Lipford, “Game2Learn: improving the motivation of CS1 students,” in Proceedings of the 3rd International Conference on Game Development in Computer Science Education (GDCSE'08), pp. 1–5, ACM, Miami, Fla, USA, March 2008.
- D. S. McNamara, G. T. Jackson, and A. C. Graesser, “Intelligent tutoring and games (iTaG),” in Proceedings of the Workshop on Intelligent Educational Games at the 14th International Conference on Artificial Intelligence in Education, pp. 1–10, IOS Press, Brighton, UK, 2009.
- W. H. Huang, W. Y. Huang, and J. Tschopp, “Sustaining iterative game playing processes in DGBL: the relationship between motivational processing and outcome processing,” Computers and Education, vol. 55, no. 2, pp. 789–797, 2010.
- W. L. Johnson, H. H. Vilhjálmsson, and S. Marsella, “Serious games for language learning: how much game, how much AI?” in Proceedings of the 12th International Conference on Artificial Intelligence in Education, pp. 306–313, IOS Press, Amsterdam, The Netherlands, 2005.
- L. Derbali and C. Frasson, “Prediction of players' motivational states using electrophysiological measures during serious game play,” in Proceedings of the 10th IEEE International Conference on Advanced Learning Technologies (ICALT'10), pp. 498–502, IEEE Computer Society, Sousse, Tunisia, July 2010.
- G. Rebolledo-Méndez, S. de Freitas, J. R. Rojano-Caceres, and A. R. Garcia-Gaona, “An empirical examination of the relation between attention and motivation in computer-based education: a modeling approach,” in Proceedings of the 23rd International Florida Artificial Intelligence Research Society Conference (FLAIRS-23), pp. 74–79, Daytona Beach, Fla, USA, May 2010.
- C. Conati, “Probabilistic assessment of user's emotions in educational games,” Applied Artificial Intelligence, vol. 16, no. 7-8, pp. 555–575, 2002.
- M. Csikszentmihalyi, Flow: The Psychology of Optimal Experience, Harper Perennial, 1990.
- R. M. Ryan and E. L. Deci, “Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being,” American Psychologist, vol. 55, no. 1, pp. 68–78, 2000.
- R. M. Ryan, C. S. Rigby, and A. Przybylski, “The motivational pull of video games: a self-determination theory approach,” Motivation and Emotion, vol. 30, no. 4, pp. 347–360, 2006.
- R. Bartle, “Hearts, clubs, diamonds, spades: players who suit MUDs,” The Journal of Virtual Environments, vol. 1, 1996.
- M. Cocea and S. Weibelzahl, “Log file analysis for disengagement detection in e-Learning environments,” User Modelling and User-Adapted Interaction, vol. 19, no. 4, pp. 341–385, 2009.
- G. Rebolledo-Méndez, S. de Freitas, J. R. Rojano-Caceres, and A. R. Garcia-Gaona, “An empirical examination of the relation between attention and motivation in computer-based education: a modeling approach,” in Proceedings of the 23rd International Florida Artificial Intelligence Research Society Conference (FLAIRS-23), pp. 74–79, Daytona Beach, Fla, USA, May 2010.
- I. Arroyo, D. G. Cooper, W. Burleson, B. P. Woolf, K. Muldner, and R. Christopherson, “Emotion sensors go to school,” in Proceedings of the 14th Conference on Artificial Intelligence in Education, pp. 17–24, IOS Press, 2009.
- A. P. Rovoi and R. Lucking, “Sense of community in a higher education television-based distance education program,” Educational Technology Research and Development, vol. 51, no. 2, pp. 5–16, 2003.
- M. J. Hannafin, J. R. Hill, and S. M. Land, “Student-centered learning and interactive multimedia: status, issues, and implications,” Contemporary Educational Psychology, vol. 68, pp. 94–97, 1997.
- T. VanDeGrift, “Coupling pair programming and writing: learning about students' perceptions and processes,” in Proceedings of the 35th SIGCSE Technical Symposium on Computer Science Education, pp. 2–6, ACM, Norfolk, Va, USA, March 2004.
- C. A. Collazos, L. A. Guerrero, J. A. Pino, and S. F. Ochoa, “Evaluating collaborative learning processes,” in Proceedings of the 8th International Workshop on Groupware: Design, Implementation and Use, pp. 203–221, Springer, 2002.
- Z. Dörnyei, Motivational Strategies in the Language Classroom, Cambridge University Press, Cambridge, UK, 2001.
- J. Goo, K. Park, M. Lee et al., “Effects of guided and unguided style learning on user attention in a virtual environment,” in Technologies For e-Learning and Digital Entertainment, pp. 1208–1222, Springer, Berlin, Germany, 2006.
- J. M. Keller, “Development and use of the ARCS model of instructional design,” Journal of Instructional Development, vol. 10, no. 3, pp. 2–10, 1987.
- J. V. Dempsey and R. B. Johnson, “The development of an ARCS gaming scale,” Instructional Psychology, vol. 25, pp. 215–221, 1998.
- G. A. Gunter, R. F. Kenny, and E. H. Vick, “A case for a formal design paradigm for serious games,” International Digital Media and Arts Association, vol. 3, pp. 93–105, 2006.
- T. Lin, A. Imamiya, W. Hu, and M. Omata, “Display characteristics affect users’ emotional arousal in 3D games,” in Universal Access in Ambient Intelligence Environments, C. Stephanidis and M. Pieper, Eds., pp. 337–351, Springer, Berlin, Germany, 2007.
- G. F. Wilson and F. Fisher, “Cognitive task classification based upon topographic EEG data,” Biological Psychology, vol. 40, no. 1-2, pp. 239–250, 1995.
- D. J. L. G. Schutter and J. van Hon, Electrophysiological Ratio Markers for the Balance Between Reward and Punishment, Elsevier, Amsterdam, The Netherlands, 2005.
- P. Putman, J. van Peer, I. Maimari, and S. van der Werff, “EEG theta/beta ratio in relation to fear-modulated response-inhibition, attentional control, and affective traits,” Biological Psychology, vol. 83, no. 2, pp. 73–78, 2010.
- M. M. Lansbergen, M. Arns, M. van Dongen-Boomsma Martine, D. Spronk, and J. K. Buitelaar, “The increase in theta/beta ratio on resting-state EEG in boys with attention-deficit/hyperactivity disorder is mediated by slow alpha peak frequency,” Progress in Neuro-Psychopharmacology and Biological Psychiatry, vol. 35, no. 1, pp. 47–52, 2011.
- L. Derbali, P. Chalfoun, and C. Frasson A, “Theoretical and empirical approach in assessing motivational factors: from serious games To an ITS,” in Proceedings of the Florida Artificial Intelligence Research Society Conference, pp. 513–518, Palm Beach, Fla, USA, 2011.
- United Nation's Food-Force Game, World Food Programme, 2005, http://www.wfp.org/how-to-help/individuals/food-force.
- J. Brophy, “Other ways to support students intrinsic motivation,” in Motivating Students to Learn, pp. 220–248, Lawrence Erlbaum Associates, Mahwah, NJ, USA, 2004.
- A. Hiam, “Motivating with informative feedback,” in Motivating and Rewarding Employees: New and Better Ways to Inspire Your People, Adams Media Corporation, 1999.
- R. Moreno and R. E. Mayer, “Role of guidance, reflection, and interactivity in an agent-based multimedia game,” Journal of Educational Psychology, vol. 97, no. 1, pp. 117–128, 2005.
- Pro-Comp Infinity Encoder, Thought Technology Ltd, http://www.thoughttechnology.com/proinf.htm.
- D. Gregory, “Guidelines for digital EEG,” American Journal of Electroneurodiagnostic Technology, vol. 39, no. 4, pp. 278–288, 1999.
- J. Han and M. Kamber, Data Mining: Concepts and Techniques, The Morgan Kaufmann Series in Data Management Systems, Morgan Kaufmann, 2006.
- T. Lin, A. Maejima, and S. Morishima, “An empirical study of bringing audience into the movie,” in Proceedings of the 9th International Symposium on Smart Graphics, pp. 70–81, Springer, Berlin, 2008.
- L. Derbali and C. Frasson, “Physiological evaluation of attention getting strategies during serious game play,” in Proceedings of the 15th International Conference on Artificial Intelligence in Education, Springer, Auckland, New Zealand, 2011.
- S. H. Song and J. M. Keller, Effectiveness of Motivationally Adaptive Computer-Assisted Instruction on the Dynamic Aspects of Motivation, Springer, Heidelberg, Germany, 2001.
Copyright © 2012 Lotfi Derbali and Claude Frasson. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.