Graduate School of Psychology
R
ESEARCHM
ASTER’
SP
SYCHOLOGYT
HESISFreedom of Choice:
The Effect of Self-Selection Learning on Children’s Engagement
Maien S. M. Sachisthal Supervisor: Maartje Raijmakers Second Assessor: Agneta Fischer
Abstract
Recent studies have shown that self-selection learning can have distinct benefits, such as enhancing category learning (Castro et al., 2008). To investigate the effect of self-selection on children’s level of engagement, which is thought to underlie self-selection learning benefits, we measured heart rate changes (cognitive engagement) and facial emotional reactions (emotional engagement) of 8-12 year olds when receiving observational feedback in a categorization task. In the self-selection condition, stimuli could be constructed, whereas stimuli were randomly
presented in the reception condition. Although the task appeared to be very easy for all
participants, self-selection did induce some enhanced category learning. Our findings concerning the effect of self-selection on engagement in the first two blocks, that did not show accuracy differences, were mixed, with cognitive engagement being higher in the reception condition and emotional engagement being higher in the self-selection condition. Implications of these findings are discussed in relation to possible long-term benefits of active learning through enhancing engagement.
Freedom of Choice:
The Effect of Self-Selection Learning on Children’s Engagement
The idea that individuals learn better when they are able to control the flow of their learning experience is widely advocated, especially in relation to children (e.g., Montessori, 1912/1964; Bruner, Jolly, & Sylva, 1976). Learning instances in which the learner can control their learning experience are called active learning and have been shown to lead to enhanced outcomes as compared to more traditional forms of learning, that is, passive learning (Bonwell & Eison, 1991; Freeman et al., 2014; see Markant, Ruggeri, Gureckis & Xu, 2016 for a review). Self-directed learning, or self-selection learning, is a form of active learning, in which learners have voluntary control over decisions such as the timing and order of the presentation of learning materials (see Gureckis & Markant, 2012 for a review). As compared to reception learning, in which the learner merely observes stimuli they are presented with, self-selection learning has been found to have distinct benefits (e.g., Sobel & Kushnir, 2006). In a category-learning paradigm, for instance, participants who were allowed to actively select samples to learn from performed with a higher accuracy in determining category boundaries than participants who were presented with randomly generated samples (Castro et al., 2008). While selection learning is supposedly
beneficial for children, few recent studies have extended the investigation of selection learning to younger populations (e.g., Sim, Tanner, Alpert & Xu, 2015). Partridge, McGovern, Yung and Kidd (2015), for instance, showed that even at the age of 3, self-selection learning has benefits, as children’s information retention was better in the selection condition than in the reception condition. Another recent study on the effect of self-selection on recognition memory in children aged 6- to 8-years-old found that recognition memory was enhanced in the self-selection (active condition) as compared to the reception condition in a simple memory game (Ruggeri, Markant, Gureckis & Xu, 2016). Moreover, self-selection has been shown to improve categorization learning in children at the age of 7 (Sim et al., 2015).
selection learning, namely informational choice and self-pacing and level of engagement. Supporting the role of informational choice and self-pacing, Sim and colleagues (2015) observed that children’s information gathering was systematically driven by uncertainty and prior feedback, indicating that benefits of self-selection learning are due to a cognitive effect. Earlier findings show that adults strategically select information by sampling closer to the true category boundary to maximally reduce uncertainty (e.g., Markant & Gureckis, 2010). Additional support that cognitive processes underlie learning benefits of self-selection learning comes from findings indicating that these benefits are influenced by task complexity (e.g., Enkvist, Newell, Justin & Olsson, 2006). In a recent study, Markant and Gureckis (2014) found that learners were performing better in the self-selection condition only if stimuli had to be classified into rule-based categories and not if the categories were based on an information integration structure, with the optimal classification rule being a linear combination of values along two dimensions. As self-selection learning seems to have different effects on learning outcomes depending on the task at hand, Markant and Gureckis (2014) argue that engagement cannot be one of the main factors underlying enhanced learning in self-selection learning conditions. They did, however, not assess task engagement, leaving the possibility that task-engagement may be influenced by self-selection.
Highlighting the importance of engagement, Fredricks, Blumenfeld and Paris (2004) posited that engagement is a megaconstruct and once it is established, it evolves, thereby
contributing to improvements in other outcomes of interest. Tasks contributing to a heightened level of engagement could, for instance, lead to a heightened level of interest in learners, which is an important determinant of science learning and science career orientation (Nugent et al., 2015). If self-selection learning does have a positive effect on levels of task engagement, this could positively influence not only immediate learning but also long-term outcomes, such as scientific interest, making studying self-selection learning especially relevant in relation to children and their formal and informal education.
Engagement
Not much attention has been paid to measuring a possible increase in engagement of learners in controlled, experimental self-selection learning research. Studies using objective, quantitative measures of engagement are especially lacking, as previous studies on level of engagement in self-selection learning mainly relied on self-reports or reports from educators to estimate learners’ levels of engagement (e.g., Henderson & Yeow, 2012; Milman, Carlson-Bancroft, & Boogart, 2012). Partridge and colleagues (2015), for instance, used reaction times as an objective measurement of engagement in a task in which children had to learn novel object-word pairings, but did not find a difference between conditions (self-selection vs. reception). They did, however, find that even when only one stimulus was presented, the selection group performed better, indicating that engagement might have played an important role nonetheless. Their findings thus point to the importance of using more fine-grained methods to measure engagement. Reaction times might be seen as an indicator of behavioral engagement, which is just one form of engagement, as suggested by the multidimensional model of engagement by Fredricks and colleagues (2004). In addition to behavioral engagement, the model includes emotional and cognitive engagement. Although their model was aimed at assessing classroom-based engagement, it can be meaningfully applied to task engagement. Emotional engagement in the context of task engagement can be conceptualized as the affective states learners experience during learning, which may be influenced by the task they are doing. Cognitive engagement, on the other hand, is associated with learners’ willingness to exert effort, such as closely monitoring task feedback to ensure successful learning.
To further investigate the role of engagement in young learners’ self-selection learning, we will measure 8- to 12-year-olds’ emotional facial reactions as well as their heart rate changes to feedback in a rule-based categorization task, in which this age group has been shown to
successfully learn category boundaries across 80 trials (Rabi & Minda, 2014). In the current study, we will primarily focus on cognitive engagement; additionally, we will look at learners’ facial
emotional expressions as an indicator of emotional engagement and their self-reported levels of task engagement. Engagement has been described as enjoyment, concentration and a sense of challenge (Rose-Krasnor, 2009), highlighting the importance of positive emotions during
emotional task engagement. To measure emotional engagement, we will use a method frequently used in marketing research (i.e., FaceReader), in which engagement is operationalized as facial emotional expressions of joy and surprise while watching an advertisement (Texeira, Wedel & Pieters, 2012).
To measure cognitive engagement, heart rate change is used. Previously, heart rate has been employed as an informative index of feedback processing (e.g., Jennings & Van der Molen, 2002), showing that for adults, children and adolescents (8 – 12 years) heart rate changes reflect feedback monitoring, especially in learning conditions where feedback actually informs learning (e.g., Crone, Jennings & Van der Molen, 2004; Mies, Van der Veen, Tulen, Hengeveld & Van der Molen, 2011; van Duijvenvoorde et al., 2013). Negative feedback leads to heart rate decelerating, whereas positive feedback elicits an acceleratory recovery (Jennings & Van der Molen, 2002). Moreover, heart rate deceleration after negative feedback has been shown to be more
pronounced in good performers, indicating that heart rate change may be specifically sensitive when individuals aim to use feedback to adjust their behavior for future performance (e.g., Jennings & Van der Molen, 2000; Crone et al., 2004). If individuals are cognitively engaged in a task, their heart rate is thus likely to be more sensitive to feedback than if they are not cognitively engaged.
In the current study, following Markant & Gureckis (2014), learning will be solely observational, so that we will not have an indication whether participants received positive or negative feedback. Still, when (unexpected) feedback is monitored more closely, that is, if individuals are cognitively engaged, heart rate changes are expected to be more sensitive to feedback, thus varying to a greater extend. We will therefore look not only at the overall change of heart rate after receiving feedback, but also focus on heart rate variance as an indication of
cognitive engagement, with greater variance implying closer monitoring of positive and negative feedback. Lastly, we will measure self-reported engagement before and after the task as to have an indication to how engaging the children found the task (e.g., Maier, Waldstein & Synowski, 2003).
To summarize, the current study aims to extend the research on the effects of self-selection learning on children by taking a closer look at the role of engagement in such a learning task. More specifically, children’s facial emotional reactions during a task as well as their heart rate changes are recorded as to establish whether the learning environment of self-selection (vs. reception) has an effect on children’s emotional as well as cognitive engagement during a categorization task. With earlier studies showing that self-selection can have distinct benefits in rule-based categorization tasks (Markant & Gureckis, 2014), we expect that children in the self-selection condition perform better, that is, reach higher levels of accuracy in the rule-based categorization task, than children in the reception condition. This finding would replicate earlier studies showing that children can benefit from self-selection in easy categorization tasks (Sim et al., 2015). We further expect that children’s levels of both cognitive as well as emotional
engagement are higher in the self-selection condition. Children are thus expected to show more happiness and surprise emotional facial displays in the self-selection condition and to show a more variable heart rate variance than those children in the reception condition; indicating that they would be monitoring feedback more closely.
Methods Participants
83 children aged between 8- and 12-years-old (39 female, 44 male; MAge=9.25, SDAge=1.15) were
recruited at Science Center NEMO in Amsterdam. One child did not start the task, because he/she was too scared to play the game. Children were recruited together with their parent and were briefly told the general aim of the study as well as the measures that would be taken (i.e., the electrocardiogram measures and the recording of their faces during the task). Parents gave their
informed consent for their child to participate in the study and for the use of the data for analysis, at the same time children began the task.
Accuracy, Self-reported Engagement and Cognitive Engagement. For 73 participants (34 female, 39 male; MAge=9.27, SDAge=1.18), all 10 ECG measures for the first two blocks were
recorded and were usable. As many children started to drop out after the second block, we decided to only analyze the data of children who had enough data points for the first two blocks. With our study primarily focusing on cognitive engagement, we tested the hypotheses concerning accuracy and self-reported engagement on the sample of children who had all 10 ECG measures of the first two blocks, that is, that there were no technical difficulties with the ECG measures of these children, so that no measure had to be removed due to artifacts, for instance.
Emotional Engagement. 61 participants (29 female, 32 male; MAge=9.29, SDAge=1.16)
had at least 5 usable measures of emotional engagement in each of the first two blocks.
FaceReader can only calculate facial expressions when the entire face of the participant is visible, so it fails to find the face when parts of the face are covered or cut off.
Procedure
Before the experimental task, the Ag-AgCI electrodes were placed to record the
electrocardiogram (ECG). Children were then seated in front of a computer and were presented with a representation of the task, in which two different crystal balls were displayed as well as the two wizards and were told that different crystal balls belong to either the blue or the green wizard. After having been outlined the task, children were asked how excited they were to do the task to get an indication of their self-reported engagement (pre-measure). Then, the actual task started (see Figure 1 for a schematic overview). The crystal balls were presented to the children in blocks of 10 balls each (with a maximum of 8 blocks) and, depending on the randomly assigned learning condition, children were either able to “design” the crystal balls themselves (self-selection condition) or were presented with a randomly selected crystal ball stimulus (reception condition; adapted from Markant & Gureckis, 2014). In the self-selection condition, children
could modify the crystal ball as follows: They were first presented with a randomly generated crystal ball and in the following trials with the previously created crystal ball and could then, using two sets of plus and minus buttons, change the frequency and orientation of the sine-wave gratings on the crystal ball. Participants could make as many changes as they wanted per trial. After having created their desired crystal ball, they clicked on the “TEST” button to reveal the correct category label. In the reception condition, participants could click on the “TEST” button whenever they were ready, after which the procedure for both conditions was equivalent: The self-selection or random crystal ball was then displayed on a screen with a green wizard (left side) and a blue wizard (right side). Children saw the stimuli for a short, fixed duration and were then presented with the corresponding category label, represented by a green frame around the correct wizard until clicking the “OK” button. After clicking the “TEST” button ending in the first slide, the participants’ facial emotional reactions to the observational feedback were unobtrusively filmed. Children were not asked to make explicit predictions and consequently did not receive corrective feedback, making the learning blocks solely observational. Following every learning block, children were presented with 10 stimuli that are informative for the rule children apply (see Figure 2), and were asked to click the “OK” button below the wizard to which the crystal ball belongs (Rabi & Minda, 2014). After each test block, children were presented with the percentage of correct responses in that test block. When children reached 100% accuracy or when they indicated that they wanted to stop the task, the experimental task was ended. Lastly, children were asked how much they enjoyed doing the task (post-measure self-reported
Figure 1. Schematic representation of the categorization task and the measures used. The self-reported engagement measure was taken before and after the task. The ECG and video recordings were taken during the observational feedback reception in the learning blocks. Accuracy was measured during the test blocks.
Figure 2. Spatial frequency and orientation values for the 10 stimuli that were used in the test blocks. The numbers given are the ones used in the PsychoPy package, which produced the stimuli. The category boundary was set at frequency=.05.
Materials
Categorization Task. The categorization task we used was adapted from Rabi and Minda (2014), who investigated the role of age and executive functioning in children’s rule-based
category learning. In their study, children at the ages of 8 to 9 years did not significantly differ from children aged 10-to-11-years-old in their ability to learn the category boundaries, making the task suitable for our study, as we tested children of ages 8 to 12. Stimuli of sine-wave gratings (Gabor Patches; Figure 3) varying in spatial frequency and orientation were generated using the PsychoPy package (Pierce, 2007) and represented the crystal balls. Values of frequency and orientation ranges were taken from Rabi and Minda (2014) and were sampled around a one-dimensional category boundary, that is, the frequency of the sine-wave gratings. To the bottom of each stimulus, two bars were added, so that the stimulus resembled a crystal ball, as to make the task less abstract.
Figure 3. Example stimuli that belong to one of the two rule-based categories (Rabi & Minda, 2014). Please note that the stimuli we used may be slightly different from the ones used in the study by Rabi and Minda (2014).
Self-report measure engagement. To measure children’s engagement before and after the task, they were asked to answer on a 5-point Likert scale (1=”not at all” to 5=”very much”) the following questions: “How excited are you about doing this task?” and “How much did you like this task?” before and after the task, respectively. The 5-point Likert scale was represented by emoticons that ranged from unpleasant to pleasant facial expressions (see Figure 4).
Figure 4. 5-point Likert scale used for children to indicate their level of self-reported engagement before and after the task.
Data recording and analysis
A majority of children did not finish all 8 learning blocks, which may be due to successful learning of the category boundaries or frustration with the task. To investigate reasons for their early dropouts, we looked at the accuracy data and how it was related to the high attrition rate. We conceptualized dropouts after reaching an accuracy score of .80 or higher as “learned”, dropouts after reaching an accuracy score of less than .80 as “frustrated”.
Cognitive Engagement. During the categorization task, the electrocardiogram (ECG) was recorded from three Ag-AgCl electrodes, which were placed at the left and the right side of the thorax and on the upper back of the children. It was recorded with a sampling rate of 600 Hz and R-peaks were detected using Vsrrp98 v8.0, which was developed by the Technical Support Group of the University of Amsterdam Psychology. A low pass filter was applied to remove noise from the signal, which was set to Fc = 55 and Order = 4.0. Additionally, psychologically impossible readings and artifacts, namely values above 1500 and below 462 milliseconds were omitted. The interbeat intervals (IBIs) were visually inspected and sequential IBIs were extracted around the moment of feedback during the learning trials, as done in van Duivenvoorde and colleagues (2013), Crone and colleagues (2004), Groen, Wijers, Mulder, Minderaa, and Althaus (2007) and Somsen and colleagues (2000). The interval of feedback presentation was labeled IBI0, with the two intervals preceding feedback presentation being IBI-2 and IBI-1 and the two intervals following IBI being IBI1 and IBI2. IBI difference scores were referenced to IBI-2 as to create a sensitive index of heart rate change. As van Duivenvoorde et al. (2013), we were
IBI1 and IBI2 with IBI-2. More specifically, not knowing the valence of the feedback participants received, we did not only look at the mean heart rate change per IBI, but also at the variance of the IBIs of interest to have an indication of the cognitive engagement of children. Low mean, but large variance would be an indication of high cognitive engagement, whereas low mean and low variance would be an indication of low cognitive engagement. To test the effect of learning environment (between: self-selection vs. reception) and block (within: block 0 and block 1) on heart-rate responses, we conducted mixed ANOVAs with the mean and variance of sequential IBI difference scores (within: IBI0, IBI1 and IBI2) as dependent variables. When necessary, Greenhouse-Geisser corrections for violations of the sphericity assumption were used. Follow-up tests between blocks and IBIs were Bonferroni corrected.
Emotional Engagement. Upon the onset of the feedback, the facial emotional
expressions of children were recorded. To determine whether children were emotionally engaged in the task, we used FaceReader 6.0 (Noldus, 2014), an automated facial coding (AFC) software package, to analyze the video recordings. FaceReader offers a list of four models to fit (general, children, east-Asian and elderly), so that we analyzed the videos using the model for children between 8 and 12 years. Specific facial action units (AUs), that is, facial muscle movements, were recorded and used in our analysis. Some AUs or the combination of AUs are seen as
characteristic for an emotion (e.g., Duclos et al., 1989; Friesen & Ekman, 1983); AU12,
describing the pulling up of the lip corners, which involves the Zygomaticus major, for instance, is a typical action unit used in happiness displays (Hess & Blairy, 2001). Specifically, we calculated the continuous intensity1 of AU6 and AU12 (happiness) and of AU1, AU2, AU5 and AU26
(surprise). Only trials in which FaceReader analyzed more than 30% of the recorded frames were used for analysis. Common reasons for FaceReader to fail the analysis of frames were that not the complete face of the child was visible in the video or that the child covered parts of their face
1
Per AU, FaceReader specifies if the AU was activated and to which extend it was activated. The activation intensity is expressed by the letters A (least intense) to E (most intense) and is then manually converted into one number by multiplying the number of occurrences of the letters by 1 to 5, respectively and then taking the mean of these numbers (Lewinski, 2013)
with their hands. As the data for both overall emotion scores (happiness and surprise) were skewed to the right, we used the Wilcoxon signed-rank test, a non-parametric test, to analyze the effect of learning environment (between: self-selection vs. reception condition) on emotional engagement per learning block (within: block 0 and block 1). We expected that emotional engagement would be higher in the self-selection condition, that is, that more happiness and surprise was expressed in the self-selection learning environment per learning block.
Results
We plotted playing rates per condition (see Figure 5). To test the effect of condition and block on playing and stopping due to frustration, we analyzed both using a general linear mixed model approach. For playing (vs. not playing), we found that the model including block and the interaction effect of condition and block was the best fit, χ2 (1, N= 82) = 5.88, p<.05, suggesting
that the likelihood of children to keep playing depended not only on the block but that the condition (self-selection vs. reception) affected drop-out rates across blocks. For stopping to play due to frustration, the model with only block was the best fit, χ2 (1, N= 82) = 143.86, p<.001,
suggesting that there was no effect of condition on frustration-induced stopping, that is, that independent of the learning environment, children got frustrated with the task when they did not achieve high accuracy scores in the test blocks.
Figure 5. Left: Proportion of children playing vs. having stopped due to successful learning or due to frustration per block in the self-selection condition. Right: Proportion of children playing vs. having stopped due to successful learning or due to frustration per block in the reception condition.
Accuracy. To determine whether learning environment (self-selection vs. reception) influences categorization accuracy, we conducted a mixed ANOVA, with accuracy scores in the test phases as dependent variable, learning environment as between subject factor, and block (0-1) as within subjects factor. The 2 (between: self-selection vs. reception) by 2 (within: block 0 and block 1) mixed ANOVA revealed a main effect of block F(1, 76)=6.63, p<.05, η2
p=.03; indicating
that learning took place across the two blocks, as children had a higher accuracy score in block 1 (M=.73, SD=.22) than in block 0 (M=.65, SD=.19). The interaction effect of condition*block showed a trend to significance, F(1, 76)=3.79, p<.1, η2
p = .016. The condition effect was not
significant.
Self-reported Engagement. To determine whether self-reported engagement was
affected by the learning environment and the measurement moment, we conducted a 2 (between: self-selection vs. reception condition) by 2 (within: pre and post measurement) mixed ANOVA. The main effect of measurement moment was significant, F(1, 71)=5.57, p<.05, η2
p=.03; with
children reporting a higher level of engagement before the learning task (M=4.04, SD=.39) than after the learning task (M=3.78, SD=.92). Moreover, the main effect of learning environment condition showed a trend to significance, F(1, 71)=2.82, p<.1, η2
p = .02; with children in the
self-selection condition reporting a lower level of engagement (M=3.81, SD=.79) than children in the reception condition (M=4.01, SD=.62). The interaction effect was not significant.
Cognitive Engagement. To determine whether learning environment (self-selection vs. reception) influenced cognitive engagement, we looked at the inter-beats-intervals (IBIs) of interest, that is, the IBIs following the moment of feedback: IBI0, IBI1 and IBI2. More specifically, we first looked at the mean difference scores of IBIs and expected that overall
cognitive engagement would be higher in the self-selection condition, that is, that the mean of difference scores of IBIs would be higher in that condition.
First, the mean of the difference scores of IBIs is shown in Figure 5, left. We conducted a 3 (within: IBI0, IBI1 and IBI2) by 2 (between: self-selection vs. reception) by 2 (within: block 0 and block 1) mixed ANOVA. The three-way-interaction of IBI, condition and block was not significant. The two-way interaction of condition and IBI, however, was significant, F(1.7, 124.1)=3.16, p<.05, η2
p = .01, as was the main effect of IBI, F(1.7, 124.1)=4.0, p<.05, η2p = .01.
In order to describe the two-way interaction, we then tested the effect of condition per IBI of interest. After Bonferroni correction, none of the simple effects were significant.
Next, to test whether the variance of the difference scores of IBIs of interest were affected by learning environment and learning block, we conducted a 3 (within: IBI0, IBI1 and IBI2) by 2 (between: self-selection vs. reception) by 2 (within: block 0 and block 1) mixed ANOVA on the variance of the difference scores of IBI scores. The 3-way-interaction of IBI, condition and block was significant, F(1.96, 139.1)=4.76, p<.01, η2
p = .003. Figure 7 shows the effect of condition on
the variance of heart-rate change per IBI for block 0 (left) and block 1 (right). Figure 6 shows the effect of condition on the variance of heart-rate change per IBI across the two blocks (right Figure).
In order to describe the 3-way interaction, we next performed a 2 (between: self-selection vs. reception) by 2 (within: block 0 and block 1) repeated measures ANOVA per IBI, to further investigate the difference in cognitive engagement between learning environments. Testing the effect of condition and block on the variance of the difference score IBI0, we found no significant effects, only the main effect of learning environment showed a trend F(1, 71)=2.94, p<.1, η2
p = .04. Here, contrary to our expectation, the variance of IBI0 was larger in the
reception (M=81.87, SD=62.84) than in the self-selection condition (M=113.74, SD=100.43). Testing the effect of condition and block on the variance of the difference score of IBI1, we found a significant effect of the 2-way interaction between block and learning environment
condition, F(1, 71)=5.19, p<.05, η2
p = .01. Neither of the main effects was significant. In order to
describe this 2-way interaction, we tested the effect of learning environment on the variance of IBI1 per block and found no effects.
Testing the effect of condition and block on the variance of the difference score IBI2, we found no significant effects.
Figure 6. Left: Mean of heart-rate responses per learning environment across blocks 0 and 1, using IBI-2 as baseline. IBI0 represents the onset of feedback presentation. Right: Variance of heart-rate responses per learning environment across blocks 0 and 1, using IBI-2 as baseline. IBI0 represents the onset of feedback presentation.
Figure 7. Left: Variance of heart-rate responses per learning environment in block 0, using IBI-2 as baseline. IBI0 represents the onset of feedback presentation. Right: Variance of heart-rate
responses per learning environment in block 1, using IBI-2 as baseline. IBI0 represents the onset of feedback presentation.
Emotional Engagement. To investigate the effect of learning environment on emotional engagement, we looked at the overall AU scores of happiness (AU6 and AU12) and surprise (AU1, AU2, AU5 and AU20) separately. Per emotion, we ran a Wilcoxon signed-rank test on the effect of learning environment (between: self-selection vs. reception condition) per learning block (within: block 0 and block 1).
The Wilcoxon signed-rank test on the happiness scores in block 0 showed a marginal effect of learning environment (W=580, p=.05), with happiness expressions being expressed more frequently in the self-selection (median=1.93; IQR=.68-3.00) than in the reception condition (median=.88, IQR=.27-2.10). In block 1, learning environment showed a significant effect (W=601, p<.05), with happiness expressed more frequently in the self-selection (median=1.46; IQR=.52-2.68) than in the reception condition (median=.74, IQR=.10-1.90).
The Wilcoxon signed-rank test on the surprise scores showed no significant effect of learning environment, neither on block 0 nor on block 1.
Discussion
While the possible role of engagement in driving the positive effects of self-selection learning has been advocated, studies using objective measures of engagement in self-selection tasks are lacking. The current study aimed to close this gap in the literature by employing two different measures of engagement for children aged 8 to 12 years during a categorization task. More specifically, we measured heart rate change as an indicator of cognitive engagement and facial emotional expressions as an indicator of emotional engagement; both measures were captured when children received observational feedback. Additionally, we measured the accuracy scores of children after each learning block and their self-reported engagement both before and after the task. We expected to find higher accuracy scores in the self-selection condition as
compared to the reception condition. Moreover, if engagement were indeed a driving factor in the advantage of self-selection in a learning task, we would expect that the measures of
engagement would be higher in the self-reception task, that is, that children’s heart rate change would be more sensitive to feedback in the self-selection task, that they would express higher levels of happiness and surprise and that their self-reported engagement was higher after the task for children in the self-selection task.
We did not find a difference in accuracy scores between the learning conditions across the first two blocks. Our analyses on the first two blocks of the task showed that children did perform better in block 1 than in block 0, indicating that learning took place, but the pace of learning was the same for both conditions. Our null-findings concerning differences of learning across learning environments is likely due to the fact that we could only analyze the data of participants across the first two blocks. In a study by Sim and colleagues (2015), for instance, 7-year-old children’s accuracy scores only diverged depending on the learning condition in the last test block (fourth block), with self-selection learners performing better than reception learners, pointing to the importance of time in self-selection learning paradigms. The same seems to be true for adult learners, as shown by Markant and Gureckis (2014), with the advantage of self-selection over reception in a rule-based categorization task only emerged by the third block. While we did not analyze the effect of learning environment and learning block on children’s categorization accuracy for all 8 blocks, we did model the playing data versus dropping out due to learning or due to being frustrated with the task. Frustration leading to stopping with the task did not differ dependent on the learning environment, but only on the learning blocks. Playing did, however, depend on an interaction between learning environment and learning block, indicating that learning did differ based on the learning environment. By block 3, a majority of children in the self-selection condition had dropped out, whereas a majority of children in the reception task had only dropped out by block 4, indicating that the task was less challenging for children in the
self-selection condition. With time, children in the self-selection condition thus seemed to have had a slight learning benefit.
With the learning environment condition having no effect on learning in the two blocks we could perform our analyses on, the current study cannot clearly determine whether engagement (also) underlies the beneficial effects of self-selection on learning outcomes. Engagement, however, is not only an important determinant for short term learning outcomes, as it has been shown to positively affect long-term outcomes such as interest (Nugent et al., 2015), so that differences of the effect of learning environment condition on engagement are still of relevance.
Firstly, we found that self-reported engagement was, contrary to our expectations,
marginally higher in the reception than the self-selection group, indicating that overall, children in the reception group enjoyed the task more than children in the self-selection group. Moreover, self-reported engagement was lower after than before doing the task for children in both
conditions, which might indicate that the task was not engaging enough or that it was frustrating for children.
Secondly, concerning cognitive engagement, our results on the effect of learning environment on heart rate change indicated that feedback monitoring was, contrary to our expectations, higher in the reception condition. Taking together our findings of the overall heart rate change to observational feedback, which was higher in the reception condition, as well as the variance in heart rate change to the observational feedback in the learning blocks, which was also higher in the reception condition, pointed towards higher cognitive engagement in the reception condition. Children in the reception condition seemingly had to monitor feedback more closely across the first two blocks without gaining a learning benefit from their higher cognitive
engagement. Our results regarding cognitive engagement thus imply that children in the self-selection condition did not have to monitor feedback as closely, while still performing slightly better in the long run – supporting previous research showing a cognitive benefit of self-selection learning (e.g., Markant & Gureckis, 2014; Sim et al., 2015).
Thirdly, concerning emotional engagement, we found, as expected, that children expressed more happiness displays in the self-selection condition than the reception condition. The
difference between learning environments in the expression of happiness was more pronounced in block 1 than in block 0, indicating that the effect of self-selection on task enjoyment might increase with time and that emotional engagement may be positively related to learning
outcomes. No effects of learning environment were found on the expression of surprise. Prior studies have shown, however, that there is a strong dissociation between self-reported surprise and facial displays of surprise, with only 4-25% of adult participants facially expressing surprise (eyebrow raise only) when surprise was induced while self-reported and behavioral measures did indicate that the induction of surprise was successful (Reisenzein, Bördgen, Holtbernd & Matz, 2006). Facial emotional measures of surprise may thus not be a sensible measure to diagnose the presence of surprise reactions, casting doubt on the usefulness of using surprise expressions as an indication of emotional engagement in learning contexts.
Strengths, Limitations and Directions for Future Research
The current study adds to the growing literature of self-selection learning in children by employing possible ways to investigate how self-selection learning may influence engagement in learners. Two engagement measures from different fields of research were used in a first attempt to establish whether engagement underlies the advantage of self-selection learning over reception learning in a categorization task in children aged 8 to 12. While taking into account design
constraints imposed by the nature of self-selection learning paradigms, that is, the need to provide learners with observational feedback only, as to reduce the impact of feedback on learners’ sampling behavior (Markant & Gureckis, 2014), we managed to set up a study
measuring both cognitive as well as emotional engagement. The categorization task we employed was based on a task previously used in a sample encompassing a broader age range (4-years-old to adults; Rabi & Minda, 2014), and used a similar number of trials as the current study, so that the high drop-out rates we experienced were not expected. It should be kept in mind, however,
that our study was set in a science center context, whereas children in the original study were recruited through schools. The highly engaging nature of a science center context and the fact that children were accompanied by their parent, may have negatively affected their willingness to finish all 8 blocks of the task and might have influenced their expectation of the task. We
therefore are planning to run the current study again, this time in a school context, in which children are likely to be less distracted and more motivated to finish the whole task.
All our findings should be interpreted with caution given the small effect sizes of all significant results. Moreover, it should be kept in mind that we experienced technical problems, which precluded us to analyze the data of the same sample across dependent variables to
preclude a further loss of statistical power. Concerning our measure of cognitive engagement, the use of heart rate variance as an indication of feedback monitoring is not typically done, and may not be as fine grained as using the mean heart rate change depending on the valence of received feedback (e.g., Crone et al., 2004; Van Duivenvoorde et al., 2013). As earlier discussed, however, providing learners with directional feedback may stand in the way of learning in the self-selection condition, as learners may experience a conflict in sampling informative items or sampling items that will lead to positive feedback (Markant & Gureckis, 2014). A possibility would be to design the task in such a way that children have to indicate to which category each item belongs (as now only done in the test blocks) and that they receive observational feedback after each trial, so that the feedback would still not be explicitly positive or negative, but that trials can be analyzed based on the expectedness of feedback. An additional advantage of this design would be that test blocks would not be any longer necessary to measure accuracy, so that the task can be shortened, if wanted.
The stimuli we used in the test blocks differed from those used in Rabi and Minda (2014). They did not employ distinct learning and test blocks, as learners received explicit feedback after each trial, making test blocks as we used them redundant. In an attempt to create test stimuli that are informative concerning the categorization rule children used to categorize items, we likely
created stimuli that were too far away from the category boundary, making the test blocks too easy. By creating stimuli that are sampled closer to the category boundaries, children may find the task more difficult and may be interested in the task for a longer time.
Moreover, the task used deviated from the self-selection task used in adult participants (Markant & Gureckis, 2014) in two ways: Firstly, and most importantly, we used less stimuli than they did, as their 8 blocks were made up of 16 learning trials and 32 test trials each, whereas we only used 10 trials each per learning and test phase per block. This deviation once more
highlights that time likely plays an important role in the learning benefit of self-selection learning. Secondly, children in the self-selection condition were only presented with random stimuli once per learning block, whereas they were presented with random stimuli trial in their study.
The divergence of our findings concerning the effect of learning environment on cognitive and emotional engagement highlights the need for future research to investigate the multifaceted nature of task engagement. Rose-Krasnor (2009) described engagement as enjoyment,
concentration and a sense of challenge, giving an indication of the broadness of the construct of engagement. Moreover, previous research has found similar dissociations of different
engagement measures in gaming, for instance (Martey et al., 2014). Future studies are needed to establish the relationships between different engagement measures and their relation to outcomes of interest. In the current study, learning was the outcome of interest and our results point to cognitive engagement being an indicator of cognitive load, so that closer monitoring of feedback may not always be related to better learning outcomes. Emotional engagement was higher in the self-selection condition, and might thus be positively related to learning outcomes.
Conclusion
Taken together, our results suggest that self-selection did induce some enhanced category
learning, even considering the apparent easiness of the task, replicating previous findings showing that self-selection can benefit children’s learning (e.g., Sim et al., 2015). Concerning the effect of
self-selection on engagement, our findings regarding the first two blocks were mixed, with cognitive as well as self-reported engagement being higher in the reception condition and emotional engagement being higher in the self-selection condition. Taken together, these results suggest that cognitive engagement as we measured it gives an indication of the cognitive benefit of self-selection as compared to reception learning and that self-selection learning can increase emotional engagement. These results have general implications for informing educational practice, as active learning could, in addition to enhancing immediate learning outcomes, affect long-term outcomes such as interest development by enhancing engagement.
References
Ashby, F. G., & Ell, S. W. (2001). The neurobiology of human category learning. Trends in cognitive sciences, 5(5), 204-210.
Boekaerts, M. (1997). Self-regulated learning: A new concept embraced by researchers, policy makers, educators, teachers, and students. Learning and Instruction, 7(2), 161–186.
Bonwell, C. C., & Eison, J. A. (1991). Active learning: Creating excitement in the classroom. 1991 ASHE-ERIC Higher Education Reports. Washington, DC: Education Resources Informa- tion Center, U.S. department of Education.
Bruner, J. (1961). The act of discovery. Harvard Educational Review, 31(1), 21–32.
Bruner, J., Jolly, A., & Sylva, K. (1976). Play: Its role in development and evolution. New York, NY: Basic Books.
Castro, R., Kalish, C., Nowak, R., Qian, R., Rogers, T., & Zhu, J. (2008). Human active learning. Advances in Neural Information Processing Systems (NIPS), 21, 241–249.
Cootes, T. F., & Taylor, C. J. (2004). Statistical models of appearance for computer vision. Imaging Science and Biomedical Engineering. University of Manchester, Manchester, UK. Crone, E. A., Jennings, J. R., Van der Molen, M. W. (2004). Developmental change in feedback
processing as reflected by phasic heart rate changes. Developmental Psychology, 40(6), 1228– 1238, http://dx.doi.org/ 10.1037/0012-1649.40.6.1228.
Enkvist, T., Newell, B., Juslin, P., & Olsson, H. (2006). On the role of causal intervention in multiple-cue judgment: Positive and negative effects on learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 163–179.
Fredricks, J. A., Blumenfeld, P. C., & Paris, A. H. (2004). School engagement: Potential of the concept, state of the evidence. Review of educational research, 74(1), 59-109.
Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H., & Wenderoth, M. P. (2014). Active learning increases student performance in science,
engineering, and mathematics. Proceedings of the National Academy of Sciences of the United States of America, 111(23), 8410–8415.
Groen, Y., Wijers, A. A., Mulder, L. J., Minderaa, R. B., & Althaus, M. (2007). Physiological correlates of learning by performance feedback in children: a study of EEG event-related potentials and evoked heart rate. Biological psychology, 76(3), 174-187.
Gureckis, T. M., & Markant, D. B. (2012). Self-directed learning a cognitive and computational perspective. Perspectives on Psychological Science, 7(5), 464-481.
Henderson, S., & Yeow, J. (2012). iPad in Education: A case study of iPad adoption and use in a primary school. In Proceedings of the 45th Hawaii International Conference on System science (pp. 78–87).
Hess, U., Blairy, S., & Kleck, R. E. (2000). The influence of expression intensity, gender, and ethnicity on judgments of dominance and affiliation. Journal of Nonverbal Behavior, 24, 265– 283. doi: 10.1023/A:1006623213355
Jennings, J. R., & Van der Molen, M. W. (2002). Cardiac timing and the central regulation of action. Psychological Research 66 (4), 337–349, http://dx.doi.org/ 10.1007/s00426-002-0106-5.
Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problem-based,
experiential, and inquiry-based teaching. Educational Psychologist, 41, 75–86.
Kuhn, D., Black, J., Keselman, A., & Kaplan, D. (2000). The develop- ment of cognitive skills to support inquiry learning. Cognition and Instruction, 18, 495–523.
Lewinski, P. (2013). Automatic Facial Coding Research Manual (with NoldusTM FaceReaderTM).
Maier, K. J., Waldstein, S. R., & Synowski, S. J. (2003). Relation of cognitive appraisal to cardiovascular reactivity, affect, and task engagement. Annals of Behavioral Medicine, 26(1), 32-41.
for guided methods of instruction. American Psychologist, 59, 14–19. doi:10.1037/0003-066X.59.1.14
Markant, D., & Gureckis, T. M. (2010). Category learning through active sampling. In S. Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual Conference of the Cognitive Science Society (pp. 248–253). Austin, TX: Cognitive Science Society.
Markant, D. B., & Gureckis, T. M. (2014). Is it better to select or to receive? Learning via active and passive hypothesis testing. Journal of Experimental Psychology, 143(1), 94.
Markant, D. B., Ruggeri, A., Gureckis, T. M., & Xu, F. (2016). Enhanced memory as a common effect of active learning. Mind, Brain, and Education, 10(3), 142-152.
Mies, G. W., Van der Veen, F. M., Tulen, J., Hengeveld, M., & Van der Molen, M. W. (2011). Cardiac and electrophysiological repsonses to valid and invalid feed- back in a time-estimation task. Journal of Psychophysiology, 25(3), 131–142.
Milman, N. B., Carlson-Bancroft, A., & Boogart, A. V. (2012). iPads in a preK-4th Independent School. In Proceedings of the International Society for Technology in Education Conference, San Diego, California.
Montessori, M. (1964). The Montessori method. New York, NY: Schocken. (Original work published 1912)
Noldus. (2014). FaceReader: Tool for automatic analysis of facial expression: Version 6.0 [Software]. Wageningen, The Netherlands: Noldus Information Technology B. V. Nugent, G., Barker, B., Welch, G., Grandgenett, N., Wu, C., & Nelson, C. (2015). A model of
factors contributing to STEM learning and career orientation. International Journal of Science Education, 37(7), 1067-1088.
Partridge, E., McGovern, M. G., Yung, A., & Kidd, C. (2015). Young Children’s Self-Directed Information Gathering on Touchscreens. In D. C. Noelle, R. Dale, A. S. Warlaumont, J.
Yoshimi, T. Matlock, C. D. Jennings, & P. P. Maglio (Eds.), Proceedings of the 37th Annual Conference of the Cognitive Science Society. (pp. 1835-1840). Austin, TX: Cognitive Science Society.
Pierce, J. (2007). Psychophysics software in python. Journal of Neuroscience Methods, 162, 8–13. Rabi, R., & Minda, J. P. (2014). Rule-based category learning in children: the role of age and
executive functioning. PloS one, 9(1), e85316.
Reisenzein, R., Bördgen, S., Holtbernd, T., & Matz, D. (2006). Evidence for strong dissociation between emotion and facial displays: The case of surprise. Journal of personality and social psychology, 91(2), 295.
Rose-Krasnor, L. (2009). Future directions in youth involvement research. Social Development, 18, 497-509.
Ruggeri, A., Markant, D. B., Gureckis, T. M., & Xu, F. (2016). Active control of study leads to improved recognition memory in children. In Papafragou, A., Grodner, D., Mirman, D., and Trueswell, J. (Eds.) Proceedings of the 38th Annual Conference of the Cognitive Science Society. Austin, TX: Cognitive Science Society
Sim, Z. L., Tanner, M., Alpert, N. Y., & Xu, F. (2015). Children learn better when they select their own data. In D. C. Noelle, R. Dale, A. S. Warlaumont, J. Yoshimi, T. Matlock, C. D. Jennings, & P. P. Maglio (Eds.), Proceedings of the 37th Annual Conference of the Cognitive Science Society. Austin, TX: Cognitive Science Society.
Sobel, D. M., & Kushnir, T. (2006). The importance of decision making in causal learning from interventions. Memory & Cognition, 34, 411–419. doi:10.3758/BF03193418
Teixeira, T., Wedel, M., & Pieters, R. (2012). Emotion-induced engagement in internet video advertisements. Journal of Marketing Research, 49(2), 144-159.
Van Duijvenvoorde, A. C., Jansen, B. R., Griffioen, E. S., Van der Molen, M. W., & Huizenga, H. M. (2013). Decomposing developmental differences in probabilistic feedback learning: A combined performance and heart-rate analysis. Biological psychology, 93(1), 175-183.
