Chapter Ⅲ METHOD
3.5. Apparatus
3.5.2 emWave
The emWave Desktop system uses an ear sensor to determine HRV, based on human pulse signals, in order to identify human emotion. The emWave system has a patented emotion recognition algorithm for identifying emotional states. With this system, we can look at user-friendly interface to assess how human emotions affect HRV. The emWave system calculates the Coherence Ratio of these three color indexes, presents them as a percentage of a participant's emotional state. The red stick is the low-frequency zone of the power spectral density, which represents a change in sympathetic activity, and the red stick represents a negative emotional state. The blue stick is the medium-frequency zone of the power spectral density, representing changes in parasympathetic nervous activity, or a peaceful emotional state. The green stick is the high-frequency zone of the power spectral density, representing parasympathetic nervous activity changes, or a positive emotional state (Chen & Wang, 2011).
Figure 14.emWave hardware
Figure 15.emWave interface
3.5.3 Neurosky
Brain wave measurement tool manufactured by Neurosky was adopted, which was one type of non-invasive brain wave measurement instrument. It was used to detect neuron electric triggering activity, and it has the earphone appearance. Meanwhile, three sensors were used to contact three locations on the skin: Two ears at the lower sides and the forehead. Forehead is a location that can be placed with sensor easily, and cortex is also the source with high cognitive signal and consciousness.
3.6. Physiology signals analysis 3.6.1 Eye movement variables
To summarize the eye movement patterns on each learning environment, 5 eye-movement measures were used (F. Y. Yang et al., 2013), including total fixation duration (TFD), number of fixations (NF), and average fixation duration (AFD), percentage of viewing time (PVT) and frequency of saccade path (FSP). Meanwhile, to analyze the attention distributions on different medium components (look-zones) on post-test problem, 4 eye movement measures were used (F.
Y. Yang et al., 2013), including percentage of time spent in zone (PTSZ), fixation count (FC), percentage of total fixations (PTF), percentage of time fixated related to total fixation duration (PTFRTFD).
These eye movement measures represent cognitive activities related to reading, comprehension and movement of attention. In brief, the fixation measures such as NF, FC and TFD, indicate the period of time needed to acquire new information (Rayner, 2009). Meanwhile, the average fixation duration, AFD, while reveals the time for information processing, could be influenced by the nature of the task given to the participants (Rayner, 2009). On the other hand, the percentage measures including PVT, PTSZ, PTF and PTFRTFD were employed in the study to show attention distributions in terms of reading time and fixation durations for different target areas of interest. In addition, the times of saccade paths indicating the back-and-forth scanning between different zones were recorded. These times can reveal the processes of integration between different modes of information (Holsanova, Holmberg, & Holmqvist, 2009). Table 10 provides the definitions of these eye movement and other measures.
In order to describe an entire picture of a participant’s visual attention on the stimuli image, we analyzed the Hot Zone images produced by DataViewer programming. In the image, the
mapped color varied from individuals’ total fixation duration at each pixel on the screen. The darker the red color was, the longer the total fixation duration accumulated on a pixel. Therefore, the red or orange spots (i.e., Hot Zones) represented locations where the information had been processed longer and more deeply by participants while green colors represented locations where the information had been minimally processed.
Table 10.Definitions for the eye-movement measures
Eye-movement measure Definition
1. Total fixation duration (TFD) Sum of durations of all fixation points on a slide
2. Number of fixations (NF) Sum of number of all fixation points on a slide
3. Average fixation duration(AFD) Average duration of a fixation point
4. Percentage of viewing time (PVT) Total fixation duration divided by the total time shown
5. Frequency of saccade path (FSP) Times of saccades divided by total time tracked
6. Percentage of time spent in zone (PTSZ) Total time in a look-zone, such as a text or picture zone, divided by total time tracked 7. Fixation count (FC) Number of fixation points in a zone
8. Percentage of total fixation (PTF) Fixation count divided by number of fixations
9. Percentage of time fixated related to total fixation duration (PTFRTFD)
Total fixation duration in a zone divided by total fixation duration of the whole slide
In order to further analyze the scan patterns of participants in both post-test performance groups, two lag sequential analyses Bakeman and Quera (1995) were conducted for each group on the sequences of Look Zones fixated of all participants.
Sequential analysis has been applied to many published studies focusing on behavioral analysis of learning processes (Hou, 2010b; Jeong, 2003). The analysis method can effectively infer the overall behavioral path patterns during a certain time period of all users. This method was calculated through rigorous conditional probability, expected values matrix, and statistically testing sequential relations between each behavior. Therefore, this study conducted sequential analyses of a group of participants’ massive eye movement transition-frequency data during the period of problem solving task, and could infer the characteristics of a group of students’ eye movement transition paths.
3.6.2 Emotion variables
Based on the distribution of frequency zones, when a learner's emotional state is negative, peaceful or positive, the Coherence score will be calculated as 0, 1 and 2, respectively .The value of Heart Rate Artifacts (HRAs) is zero when the human emotion detected is in normal situations, whereas the value of HRAs is one when the human emotion is detected in abnormal situations.
The emWave system identifies learner emotional states every 5 s. In this study, identifying the percentages spent in positive or negative emotions was applied to assess the effects of two learning environment on learning emotions. In computing the percentages of positive and negative emotions, the Accumulated Coherence Score (ACS) has the key role, whereas the method for computing ACS based on different coherence states and HRAs is as follows (Chen &
Wang, 2011):
WhereCV t is the coherence value at the ith sampling time. Coherence (t) is the coherence
( ) state at the tth sampling time; HRA(t) is the HRA value at the tth sampling time; and m is the times of emotional states that are recognized.Based on the CVs obtained by Eq. (2), the ACSs of positive, peaceful, and negative emotions during learning can be respectively formulated as follows:
1
Therefore, the occupied percentage of positive and negative emotions can be respectively formulated as follows:
Positive Emotion
Negative Emotion
=
ACS of Negative Emotion
ACS ofPositive Emotion+ACS of Peaceful Emotion+ACS of Negative Emotion 100%
(7)
According the Eq. (6) and Eq. (7), the occupied percentage of positive and negative emotions during learning can be respectively obtained.
3.6.3 Brain wave variables
NeuroSky is comprised of a complex combination of artifact rejection and data classification methods. According the NeuroSky proprietary Attention & Meditation eSense algorithms, NeuroSky can report the wearer’s attention state. Attention score is calculated each second. (The range of attention score is 1 to 100; 1=very low attention level and 100=very high attention level.)
3.6.4 Data analysis
PAHSE 1
To measure the motivation score by
physiology signals and compare the motivation,
affective experiences and cognitive load of learners in digital-game based learning and traditional static e-learning environments. The attention in ARCS model was measured by NeuroSky, the affective experiences was measured by emWave and the cognitive load was measured by eye-tracker. The physiology signals variables in PAHSE1 were all shown in Table 11. Learner attention is recognized by the Neurosky system, it was used to detect neuron electrictriggering activity, and it has the earphone appearance. According NeuroSky proprietary Attention & Meditation eSense algorithms, NeuroSky can report the attention score each second.
The range of attention score is 1 to 100 (1=very low attention level and 100=very high attention level). Learner affective experience was recognized by emWave system, which uses human pulse physiological signals to identify Coherence score every 5 seconds. Coherence score have 0, 1 and 2 (0 = negative emotion, 1 = peaceful and 2 = positive emotion).
Cognitive load was recognized by eye tracker. To summarize the cognitive load between Digital Game-Based Learning and traditional static e-learning, 5 eye-movement measures were used, including total fixation duration (TFD), number of fixations (NF), and average fixation duration (AFD), percentage of viewing time (PVT) and frequency of saccade path (FSP).
Table 11.Physiology signals variables in PHASE 1
Physiology signals variables Definition Reference
Eye movement variables:
Total fixation duration (TFD) Number of fixations (NF) Average fixation duration (AFD) Percentage of viewing time (PVT) Frequency of saccade path (FSP)
Cognitive load
Sum of durations of all fixation points. (millisecond, ms) Sum of number of all fixation points. (piece)
Average duration of a fixation point. (millisecond, ms) Total fixation duration divided by the total time. (%) Times of saccades divided by total time tracked. (piece)
When above
Positive emotion occurrence times divided by the total emotion state occurrence times. (%)
The range of attention score is 1 to 100; 1=very low
When AT gets bigger, learning
The attention data, emotion data and eye-movement data were exported to Excel, and SPSS was then applied for further statistical analyses. In phase 1, the descriptive statistics and one-way ANOVA were performed to find differences in attention, emotion and eye movements between the different learning groups (DGBL vs. Static e-learning). In addition, post-test scores were analyzed by ANCOVA, in which instructional strategy (DGBL vs. static e-learning ) was the between-groups factor, and pre-test scores were treated as covariates in order to control for the effects of pre-existing between-group differences on subsequent analysis.
PAHSE 2
Before analyzing the problem solving strategy in each group, several LookZones were defined as follows. In post-test problem 1 (Figure 17), LookZone “Title” referred to the area of problem statement. For options, LookZone “correct” referred to the area of option 2 and 3, LookZone “incorrect” included option 1 and 5. For key factor, LookZone 4 and F are the key here to solve this problem. In post-test problem 2 (Figure 18), LookZone “Title” referred to the area of problem statement. For options, LookZone “correct” referred to the area of option 3, LookZone “incorrect” included option 1, 2, 4 and 5. For key factor, LookZone N, W, F and K are the key here to solve this problem.
4
F
Correct Incorrect
Incorrect Text
Correct
Figure 17.The definition scheme for LookZones in problem1
LookZone “Title” referred to the area of problem statement. For options, LookZone “correct”
referred to the area of option 2 and 3, LookZone “incorrect” included option 1 and 5. For key factor, LookZone “4” and “F” are the key here to solve this problem.
Correct Text
Incorrect
Figure 18.The definition scheme for LookZones in problem 2
LookZone “Title” referred to the area of problem statement. For options, LookZone “correct”
referred to the area of option 3, LookZone “incorrect” included option 1, 2, 4 and 5. For key factor, LookZone N, W, F and K are the key here to solve this problem.
Table 12.Physiology signals variables in PHASE 2
Eye-movement measure Definition
Percentage of time spent in zone (PTSZ) Total time in a look-zone, such as a text or picture zone, divided by total time tracked. (%)
Fixation count (FC) Number of fixation points in a zone. (piece)
Percentage of total fixation (PTF) Fixation count divided by number of fixations. (%)
Percentage of time fixated related to total fixation duration (PTFRTFD)
Total fixation duration in a zone divided by total fixation duration. (%)
Hot Zone image In the image, the mapped color varied from individuals’ total fixation duration at each pixel on the screen. The darker the red color was, the longer the total fixation duration accumulated on a pixel.
Sequential analysis Sequential analysis can effectively infer the overall behavioral path patterns during learners’ thinking.
Meanwhile, to analyze the attention distributions on different medium components (look-zones) on post-test problem, 4 eye movement measures were used, including percentage of time spent in zone (PTSZ), fixation count (FC), percentage of total fixations (PTF), percentage of time fixated related to total fixation duration (PTFRTFD). In order to describe an entire picture of learners’ visual attention on the post-test questions, we analyzed the Hot Zone images produced by DataViewer programming. In the image, the mapped color varied from individuals’
total fixation duration at each pixel on the screen. The darker the red color was, the longer the total fixation duration accumulated on a pixel. Therefore, the red or orange spots (i.e., Hot Zones) represented locations where the information had been processed longer and more deeply by participants while green colors represented locations where the information had been minimally processed. In addition, sequential analysis can effectively infer the overall behavioral path
patterns during learners’ thinking. We can observe the problem solving logic of leaners.
In phase 2, the one-way ANOVA were performed to find differences in eye movements between the different learning style group (Active vs. Reflective), learning environment group (DGBL vs. Static e-learning), different learners’ major group (Non-science vs. Science) and different post-test performance group (High vs. Low). In order to further analyze the scan patterns of participants in both post-test performance groups, two lag sequential analyses were conducted for each group on the sequences of LookZones fixated of all participants.
Chapter Ⅳ DATA ANALYSIS
For verifying the hypotheses in this study, the results were showed in chapter 4. In section 4.1, physiology signals for learning motivation, affective experiences and cognitive load were represented. In section 4.2, this paper found that DGBL group have better academic achievement but have no significant differences. In section 4.3, the result showed that high working memory capacity learning style group knew better where to look the key factors. In section 4.4, the result showed that DGBL group knew better where to look the key factors. In section 4.5, that result showed that Science major group couldn’t know better where to look the key factors, but Non-science majors’ learners need more clues to solve problem. In section 4.6, we found that successful problem solvers are able to recognize and concentrate on relevant cues. In section 4.7, the result showed that Successful problem solvers inspected the factors in a different pattern from unsuccessful problem solvers. In section 4.8, we found that physiology signals during learning couldn’t explain learners’ learning style. In section 4.9, Pearson correlation coefficient analysis was used and found that earning style (active vs. reflective) could be measure by eye movement variables when learners solving problem. In section 4.10, the summary of hypotheses verified was showed in table. Finally, discussion was summarized in section 4.11.
4.1 DGBL group have more cognitive load than traditional static e-learning group
4.1.1 Physiology signals representation for learning motivation, affective experiences and cognitive load
To compare how Digital Game-Based Learning affect learning motivation, affective experiences and cognitive load, the attention in ARCS model was measured by attention score
(AT), the affective experiences was measured by occupied percentage of positive (PE) and the cognitive load was measured by 5 eye-movement measures including total fixation duration (TFD), number of fixations (NF), and average fixation duration (AFD), percentage of viewing time (PVT) and frequency of saccade path (FSP). A one-way ANOVA was used to examine the differences between traditional static e-learning and DGBL. The results are shown in Table 13.
Table 13.ANOVA Analysis of physiology signals between Static and DGBL Learner
Static (n=16) DGBL (n=16) ANOVA
AT = attention score, PE = occupied percentage of positive, TFD = total fixation duration, NF = number of fixations, AFD = average fixation duration, PVT = percentage of viewing time, FSP = frequency of saccade path.
*** p < .001
A one-way ANOVA showed that there were significant differences in total fixation duration (TFD), number of fixations (NF), and average fixation duration (AFD) and percentage of viewing time (PVT) across the two learning environment. (FTFD (1,30) = 45.297 , p<0.001 ; FNF
(1,30) = 105.966 , p<0.001 ; FAFD (1,30) = 61.247 , p<0.001 ; FPVT(1,30) = 58.333 , p<0.001 .) For attention score (AT), occupied percentage of positive (PE), frequency of saccade path (FSP) and sum of saccade paths (SSP), we did not find significant differences.
4.1.2 Finding
Hypotheses 1 is rejected
According attention (AT) score, DGBL group had better attention score than traditional static e-learning group, but we did not find significant differences. Huang et al. (2010) observed ARCS scores by questionnaire and found that learners started out with a successful motivational processing that consisted of a high attention level. In addition, Tüzün, Yılmaz-Soylu, Karakuş, İnal, and Kızılkaya (2009) found that students demonstrated statistically significant higher intrinsic motivations and statistically significant lower extrinsic motivations learning in the game-based environment. This suggests that many studies proved that learners got better attention and learning motivation in DGBL environment by questionnaire, but we couldn’t prove this argument by the proof of affective computing technique.
Hypothesis 2 is rejected
According occupied percentage of positive (PE) score, DGBL group didn’t have better affective experiences than traditional static e-learning group. Clark et al. (2011) found that majority of the eighth and ninth grade students in Taiwan and United States liked their affective experiences using SURGE game. SURGE game is a serious educational game and not a commercial game, so maybe college student couldn’t feel funny like eighth and ninth grade students.
Hypothesis 3 is supported
The cognitive load was measured by 5 eye-movement measures. There were found the significant differences in fixation duration (TFD), number of fixations (NF), and average fixation duration (AFD), percentage of viewing time (PVT). In TFD and NF data, the result showed that DGBL group less than Static e-learning group. But in AFD and PVT data, DGBL group more than Static e-learning group. Because a parts of learner completed all stages less than 10 min, based on the different learning time for each learner, this study used AFD and PVT to measure
the cognitive load. Therefore, this suggests that DGBL group have more cognitive load than traditional static e-learning group and have significant difference.
4.2 DGBL group have better academic achievement but have no significant differences
Force Concept Inventory (FCI) test score were used as the pre-test: 32.08 (SD = 12.64) and 33.13 (SD = 11.51) for the static e-learning group and DGBL group. In addition to the pre-test, a post-test was also conducted, with mean scores of 43.75 (SD = 25.00) and 50.00 (SD = 31.62) for the static e-learning group and DGBL group, respectively. The pre-test and post-test score between two learning environment (static e-learning vs. DGBL) was showed in Figure 19.Post-test score was analyzed ANCOVA, in which learning environment (static e-learning vs. DGBL) was the between-groups factor, and pre-test scores were treated as covariates in order to control for the effects of pre-existing between-group differences on subsequent analysis.
Before ANCOVA analysis was conducted, the Kolmogorov–Smirnova method was used to test the normality of academic achievement data in this study; a non-significant result (p < .001) indicated that the data were non-normally distributed. Additionally, a test for homogeneity of variance was conducted, with the result revealing no significant effect, F (1, 30) = .091, p = .765;
that is, the data meet the requirement for homogeneity of variance. There were no significant effect be found for learning environment, F (1, 30) = .349, p = .559, and is presented in Table 14.
Pre-test Post-test
Figure 19.The pre-test and post-test measurement between two learning environment
Vogel, Greenwood-Ericksen, Cannon-Bowers, and Bowers (2006) reported that interactive games were more effective than traditional classroom instruction on learners’ academic learning gains and cognitive skill development. As the results in this study, we found that no differences in student learning can be found between learning environments that involve games and those without game elements (Leonard A. Annetta, Minogue, Holmes, & Cheng, 2009; Papastergiou, 2009). Thus, DGBL group have better academic achievement but have no significant differences.
Hypothesis 2 is rejected.
Table 14.Analysis of covariance for academic achievement
SS df MS F p
Covariance Pre-test 174.58 1 174.58 .209 .651
Between-groups Learning environment 291.49 1 291.49 .349 .559
Within-group Error 24200.42 29 834.49
4.3 High working memory capacity learning style group knew better where to look the key factors.
4.3.1 Active vs. reflective
Attention distributions on each LookZone
To analyze the attention distributions on different look-zones between learning style (active vs. reflective), 4 eye movement measures were used, including percentage of time spent in zone (PTSZ), fixation count (FC), percentage of total fixations (PTF), percentage of time fixated related to total fixation duration (PTFRTFD).
In problem 1, one-way ANOVA showed that there were significant differences in LookZone
“Text” (FPTSZ (1,30) = 9.956 , p<.01 ; FPTF (1,30) = 5.148 , p<.05 ; FPTFRTFD (1,30) = 12.124 ,
p<.01), LookZone “4” (F
FC (1,30) = 7.163 , p<.05 ; FPTF (1,30) = 5.987 , p<.05), LookZone “F”(FPTSZ (1,30) = 5.119 , p<.05 ; FFC (1,30) = 6.409 , p<.05 ; FPTF (1,30) = 8.453 , p<.01 ; FPTFRTFD
(1,30) = 4.816 , p<.05) and LookZone “correct” (FPTSZ (1,30) = 6.762 , p<.05 ; FPTF (1,30) = 8.453 , p<.01 ; FPTFRTFD (1,30) = 4.816 , p<.05). The one-way ANOVA results are shown in Table 15. The significant differences LookZone between active and reflective are shown in Figure 20, the red area represent the LookZone was found significant differences between active and reflective.
In problem 2, one-way ANOVA showed that there were significant differences in LookZone
“K” (FPTSZ (1,30) = 40.060 , p<.001 ; FFC (1,30) = 105.043 , p<.001 ; FPTF (1,30) = 92.396 ,
p<.001; F
PTFRTFD (1,30) = 40.427 , p<.001). The one-way ANOVA results are shown in Table 16.The significant differences LookZone between active and reflective are shown in Figure 21, the
4
F
Correct Incorrect
Incorrect Text
Figure 20.Significant differences LookZone between active and reflective group (problem 1).
The dotted area represent the LookZone was found significant differences and reflective see more than active.
Table 15.Visual attention distributions between active and reflective group (problem 1)
LookZone Measures Active (n=16) Reflective (n=16) ANOVA
LookZone: Text = problem description, 4 and F = key factor, correct = correct option 2 and 3, incorrect = incorrect
LookZone: Text = problem description, 4 and F = key factor, correct = correct option 2 and 3, incorrect = incorrect