AFDELING PSYCHOLOGIE FACULTEIT DER MAATSCHAPPIJ- EN GEDRAGSWETENSCHAPPEN UNIVERSITEIT VAN AMSTERDAM
The Informative Value of Feedback and the
Role of Sustained Attention and Ambiguity in
Reception and Year 6 Children
by Sandra Kamphuis
Student Number: 10096388
Supervisor: mw. prof. dr. M. E. J. Raijmakers Date: 20-08-2015
The Informative Value of Feedback and the Role of Ambiguity and Sustained Attention in Reception and Year 6 Children
Table of Contents Abstract………p. 3 Introduction ……….p. 4 Method………...p. 8 Sample ………...p. 8 Experimental tasks ………p. 8 Discrimination learning ………p. 8 Sustained attention……….p. 10 Reaction time………..p. 11 Procedure………...p. 11 Results ………p. 12 Conclusion/Discussion………..………...…...p.20 Appendix………...p. 24 A. Task Instructions……….p. 24
B. Visionary Overview Tasks………...p. 26
Abstract
Hypothesis testing is a process that has been proven to be very difficult for young children at times. The current research has attempted to provide further insight into the underlying mechanisms that contribute to this difficulty, by assessing the influence of feedback quality (e.g. is feedback ambiguous or non-ambiguous), the informative value of the given feedback and the role that sustained attention possibly plays in the hardships some children experience in hypothesis testing. Results indicate that the quality of the given feedback is not any more relevant for younger children than for older children and that both younger and older children seem to ascribe similar values to given feedback. How young children perform on hypothesis testing tasks varies enormously, but the variation that is found cannot be explained by sustained attention abilities.
The Informative Value of Feedback and the Role of Ambiguity and Sustained Attention in Reception and Year 6 Children
Introduction
As technology becomes intertwined with our daily lives it is becoming more common for schools to try to utilize technology in order to support their students in the learning process. More recently, devices such as tablets and laptops are being used more frequently in combination with classroom teaching. These changes mean that there is a growing need for investigation on how these electronic devices possibly influence childhood education.
When children play educational games or homework assignments on a tablet or computer, feedback is mostly given in an emoticon related manner, such as smiley’s, ticks or crosses or simply markings that are red (incorrect) or green (correct). However, research has shown that emoticon related feedback is at times very difficult for children to process correctly (Schmittmann, van der Maas & Raijmakers, 2012). It has shown that there is a large percentage of children that simply do not seem to learn from the given feedback, especially amongst younger children (e.g. 4-5 years of age).
Feedback learning is a very complex process in young children. What the exact mechanisms are behind this process is yet unclear. Tasks that are used to test feedback processing are tasks such as the Dimensional Change Card Sort, DCCS (Zelazo, 2006) and discrimination learning tasks (Kendler, 1979). Discrimination learning tasks require the participant to infer underlying rules by learning from given feedback. In order to do this successfully the participant must be able to use hypothesis testing. In a successful hypothesis-testing situation there are two situations: 1. Win-Stay, where a person’s hypothesis turns out to be correct, which leads to the continuation of the rule. 2. Lose-Shift, where a hypothesis turns out to be false, which results into a switch to a new hypothesis (Schmittmann et al. 2012). Several researches have shown that there are great developmental differences in feedback learning and the ability to correctly apply these elements of hypothesis testing (e.g., Eppinger, Mock & Kray, 2009; Schmittmann et al., 2012).
In the learning task from Schmittmann, children are required to simply choose one out of two presented pictures in order to deduct an underlying rule. The underlying rule in this case would be something similar to ‘All the yellow pictures are correct’. The children are for example, to choose from a green chicken and a yellow rabbit. A child can make up the rule in his or her mind to consider that all the chickens are to be selected. When the child would tap the green chicken on the screen to select it, a sad smiley would appear, indicating the mistake
the child had made. Using hypothesis testing the child should now deduct that his or her rule is incorrect, and switch to a new rule until the correct rule is obtained (as described earlier, the Lose-Shift strategy). Results from Schmittmann et al. (2012) suggested that in this task 4 and 5 years old children could often not change dimension during hypothesis testing, which becomes clear in a process called perseveration. This means that the child is unable to let go of his or her first sorting rule, regardless of the given feedback. The task that was used in the research of Schmittmann and colleagues (2012) proved to be very hard for young children (e.g. 4-5 years old) to complete successfully, whilst the task has shown to be significantly easier for children around the age of 8. However, why this was the case is yet to be specified. The current research has attempted to acquire more insight in this occurrence by administering a discrimination-learning task (a variation on the task used in Schmittmann et al. 2012) in reception (aged 4-5) and year 6 children (aged 10-11).
An important factor that might contribute to the difficulty of learning from feedback is the quality of the given feedback. In feedback-learning tasks, feedback takes different forms. First of all a child can be given positive feedback, in other words, a confirmation that a child has made the right decision. Secondly a child can receive negative feedback, which would confirm a child has made the wrong decision. In addition, the interpretation could also take different forms: ambiguous or non-ambiguous. As an example, a child might be able to choose between a blue frog and a red snail. The rule that the child administers could be the following: ‘All frogs are correct’. In the situation where the actual correct sorting rule would be ‘All blue figures are correct’, the child would obtain positive feedback whilst choosing the blue frog, which could be interpreted as a confirmation of the chosen rule, although the rule that the child is using is incorrect. In this case, we call the feedback ambiguous. The ambiguity of feedback could be a possible explanation for the finding that the task that is used in Schmittmann et al. (2012) was so hard for young children. The difficulty that is observed with young children on sorting tasks is not frequently found in children who are older (e.g. >10 years old) which suggests that these children are already able to successfully apply hypothesis testing to a hypothesis testing task (Schmittmann et al. 2012). In this light we expect that young children will perform worse in ambiguous sorting tasks (resulting in perseveration) than in non-ambiguous sorting tasks, whereas older children will perform equally well in ambiguous sorting tasks than in non-ambiguous sorting tasks.
The expected differentiated effect of ambiguous feedback possibly derives from the difference of informative value that is ascribed to positive feedback. Research from Eppinger
do. Their research showed that both adults and children showed similar accuracy in a hypothesis testing task when feedback was valid, whereas children (age 10-12) showed a decreased accuracy when feedback was partially invalid, due to the probabilistic nature of the task that was given. The probabilistic nature of the task was created by giving the participant the correct feedback (positive when the choice was correct and negative if the choice was incorrect) in only 80% of all choices. The nature of this task can be compared to the ambiguity of feedback as described above: feedback is not to be trusted 100% of the time. It could be said that children are more sensitive to feedback than adults. It seems to be that whereas adults perceive the probability of feedback merely as an error (which results in them hardly being affected by random negative feedback), children perceive the negative feedback as a direct reflection on their own work (Eppinger et al. 2009).
As the quality of the given feedback seems to be likely to play a role in the hardships that young children experience whilst attempting a discrimination learning task as used in the research of Schmittmann and colleagues (2012), it is important to obtain a better understanding about the processes that underlie these phenomena. As stated earlier we expect young children to show greater difficulty whilst handling ambiguous feedback than older children will. As young children tend to perseverate in the task that is used in Schmittmann et al. (2012) we expect that in the current research the young children that receive unambiguous feedback will outperform the children that receive ambiguous- or probabilistic feedback. Consequently we expect older children to perform equally well whilst receiving non-ambiguous, ambiguous or probabilistic feedback as well as outperforming the young children at all times.
Based on the research of Eppinger et al. (2009) we furthermore expect that the young children in our sample will assign different informative value to both positive- and negative feedback than the older children. Former research has established that hypothesis testing in a task as used in Schmittmann et al. (2012) consists out of two phases. The first phase consists out of creating a rule and testing this. We call this phase 1: hypothesis testing. The second phase consists out of applying the correct rule that is found during phase 1. We call this phase 2: applying the rule. Considering the idea that young children assign different informative value to given feedback than older children, we expect that the phase of the task will be irrelevant for the value that young children assign to positive feedback, whereas the phase of the task will show to be relevant for the value that older children ascribe to given feedback. This will be measured using reaction times, where we expect that the reaction times will not differ for young children (as is expected based on the results of Eppinger et al. 2009), but will
differ for older children, between both phases. The difference that we expect to find with older children is that the reaction times will decline once the rule has been discovered; as the feedback will become less valuable once the children are convinced they have deducted the correct rule.
As we are only able to research the effects of positive feedback based on a deterministic task, we will add a probabilistic learning task, in which the feedback will be valid for 80% of the time. The probabilistic nature of this task will be the same as explained earlier in the task used in Eppinger et al. (2009). In this task it will be possible to observe the effects of negative feedback on reaction times in both phases of the task. This is not possible during a normal ambiguous task as there will be no negative feedback given during the second phase, applying the rule (as all participants by this point have deducted and are successfully applying the underlying rule). Eppinger et al. (2009) found in their research that children (age 10-12) could perform equally well as adults, unless probabilistic feedback interfered; then their performance dropped substantially. Based on the results of Eppinger et al. (2009), we anticipate to find no decline in reaction times for both younger and older children after receiving negative feedback, as the incongruent feedback will form an interference factor on performance, leaving the children unable to perceive the given feedback as less valuable or irrelevant.
Another factor that might contribute to the findings that younger children experience greater difficulties whilst performing on a discrimination-learning task than older children is the possible difference in sustained attention ability. A successful learner on a task such as used in Schmittmann et al. (2012) needs to be able to focus his or her attention on the relevant stimuli whilst simultaneously ignoring influences of irrelevant stimuli. Tasks such as the Track-It task are thought to measure such abilities (Fisher, Thiessen, Godwin, Kloos & Dickerson, 2013). Research has shown that younger children (e.g. 4 years of age) are less able to focus their attention for an extended period of time than older children are (Eppinger et al. 2009). The current research will control for this factor by administering a task for measuring sustained attention ability (Track It, Fisher et al. 2009), as this might be a contributing or even explanatory factor for the poor performance that is observed in young children on a discrimination learning task.
Method
Sample
The original sample consisted out of 120 participants. From all participants 2 were excluded from the analysis, as they did not complete all trials in the discrimination-learning task. The total sample that was left for analysis contained 118 participants. The sample consisted out of 64 boys and 54 girls in two age groups; 58 children that were in reception (‘groep 1/2’ in the Netherlands) (35 boys and 23 girls) and 60 children that were in year 6 (‘groep 7’ in the Netherlands) (29 boys and 31 girls). The children in reception had an average age of 4.59 (SD=0.59) and the children in year 6 had an average age of 10.33 (SD=0.66) All participants were randomly selected and are recruited from three regular primary schools and the Science Centre ‘NEMO’, in the Netherlands. The gender distribution did not differ over both age groups (χ²(1)=1.74, p=0.19). All participants were randomly assigned to one of three conditions; the ambiguous condition (N=38, 17 females, 21 males), the non-ambiguous condition (N=39, 19 females, 20 males) and the probabilistic condition (N=41, 18 females and 23 males). The gender distribution was also equal across conditions (χ²(2)=0.21, p=0.90). All participants were required to have normal or corrected-to-normal vision and active informed consent was obtained in all cases from parents or caretakers.
Experimental tasks
Discrimination learning
A discrimination-learning task was administered in which children chose one of two stimuli (right or left) that were presented on one card on a computer screen. The rule was not told but found by trial and error with feedback. The task used in the current study is an adaption of the task used in Schmittmann et al. (2012). The test consisted of two card sets: A) With cards that resulted in ambiguous feedback (See Figure 1) and B) With cards that resulted in non-ambiguous feedback (See Figure 2). In both card sets there were an equal number of different cards that needed to be discriminated. Each condition consisted out of 48 trials.
Cards stated in Figure 1 have values of two dimensions (shape and colour) in common with another card. The cards are ambiguous because the child could get positive feedback on a choice even when the child does not have the right rule in mind (for example, the child chooses to sort based on ‘blue’, so therefore picks the blue frog. A smiley face appears telling the child he or she has made the right choice,whilst the actual sorting rule is ‘frog’, not ‘blue’. This will lead the child to choose wrongfully in the second trial).
Figure 2. Example Cards Non-ambiguous Feedback
Cards stated in Figure 2 have no dimensional value in common with another card. The set of cards described above are non-ambiguous because the child could not, according to our hypothesis, get the chance to get a false rule confirmed with positive feedback, as was described earlier (for example, the child chooses to sort on ‘blue’, so therefore picks the blue frog. A smiley face will in all cases mean that the child made the right choice, as there are no other pictures that are either blue or frog-shaped). In the card set that is shown here the child was required to learn two separate rules, for example ‘All frogs are correct’ and ‘All rabbits are correct’, or to learn separate pictures (‘The blue frog’ and ‘The yellow rabbit’ are correct).
From the card sets used, three conditions followed: the ambiguous condition (with the ambiguous card set), the non-ambiguous condition (with the non-ambiguous card set) and the probabilistic condition, which was also created out of the non-ambiguous card set. The only difference between the non-ambiguous condition and the probabilistic condition was that in the probabilistic condition the given feedback was congruent with the given response 80% of the time, and incongruent with the given response 20% of the time (averaged over all participants over all trials). Congruent feedback in this case meant that the given feedback was positive when the choice was correct and negative if the choice was incorrect. Incongruent feedback on the other hand meant that the given feedback was negative when the choice was correct and positive when the choice was incorrect. This enabled us to observe the effect of negative feedback.
An indeterminacy that had to be taken in consideration is the fact that children might appear to have deducted the correct rule; whilst in reality they are showing preference for a certain dimension of the stimuli. If the underlying rule happened to be ‘All blue pictures are correct’ and the child’s favourite colour was blue, then he or she might have appeared to have deducted the rule successfully, whilst not having done so at all. As former research has shown that the preference of a certain stimuli can cloud the results concerning the sorting ability of children (Schmittmann et al. 2012) the current research started the discrimination learning task with 6 blank trials, in which we determined if a participant showed preference for a certain dimension (e.g. a preference for a certain shape or a preference for a certain colour). In order to counteract negative influences, each child was required to sort the cards during the test-phase of the discrimination-learning task on the dimension that did not show to be preferred. If the child did not show preference during the first blank trials then the sorting dimension was chosen randomly.
Sustained attention
In order to determine if younger children did not simply achieve worse results than older children because their sustained attention ability is not as well developed, we also administered a sustained attention task called Track-It. Track-It is a computerised task which presents each child with a 9-block grid (See Figure 3), in which several (the number can be adjusted) figures are shown. One of these figures has a red circle around it when each trial starts and is therefore the figure that needs to be followed throughout that trial (e.g. target figure, which changes every trial). When testing commenced all figures started moving around the grid along a random trajectory. All children were asked to visually track the target figure and identify the last location where the target was seen at before it disappeared (Fisher, Thiessen, Godwin, Kloos & Dickerson, 2013). Directly afterwards the children were requested to point out what the target figure looked like. The task in the current study consisted of one target figure, which was accompanied by 6 distractor figures (all different shapes). Each trial had a duration of 600 milliseconds and each child completed a total of 7 trials, which was shown to be the most successful setting for our target group (as tested through a pilot study). The child received positive feedback by seeing a happy smiley after each trial, regardless of the reaction, in order to keep motivation high.
Figure 3. Example of Starter Grid of the Sustained Attention Task: Track-It
Reaction time
Each participant completed a two choice simple reaction time task in order to determine average reaction time. This task was administered so we were able to control for the two-choice reaction time. In this task children were asked to place their finger in the centre of a touch-screen (for a complete overview on visual task overviews see Appendix B. Visionary
Overview Tasks). At the moment they placed their finger in the centre a fish appeared, which
was either ‘swimming to the left’ or ‘swimming to the right’ (See Figure 4). All children were asked to move their finger as quickly as they could to the side where the fish was supposedly swimming (looking) towards (i.e. left or right). 48 Repetitive trials were used to determine the average reaction time of each participant (for a complete overview on task instructions see Appendix A. Task Instructions).
Figure 4. Fishes as Used in Two-Choice Simple Reaction Time Task
Fish swimming to the left Fish swimming to the right
Procedure
All children were tested at their own primary school or in the NEMO Science Centre, in a private, quiet room. Once the participant was made familiar with the researcher, the touch screen was introduced. The basic information of all participants was noted before testing
commenced. Each participant was explained that the testing was voluntary and that if they desired, testing could be stopped on their request.
Testing begun by completing a sustained attention task; Track It. The experimenter explained the rules to each child (for a complete overview on task instructions see Appendix A. Task Instructions). After receiving confirmation that the child understood the rules of the task, 7 trials were completed. The child received positive feedback from the computer through a smiley face after each trial in order to keep motivation high.
Each participant continued with a two-choice simple reaction time task. The experimenter explained the rules of the task to each child (for a complete overview on task instructions see Appendix A. Task Instructions). All participants were told they had to do this task as fast as they could, with as little mistakes as possible.
After a short break the experimenter checked if the child was ready to continue, after which the discrimination-learning task was administered. Each child was randomly assigned to either the non-ambiguous, ambiguous or probabilistic condition. After completing 6 blank trials to determine if a child showed preference for one of the used figures, the test trials commenced (if a child indeed showed to have a preference for one of the figures/dimensions, the underlying rule was determined on the opposite dimension. For example if a child consistently selected the frogs, he or she had to sort according to the underlying rule ‘All snails are correct’). Each child completed a total of 48 trials, 6 blank trials and 42 test trials. After the child completed the tasks they received positive feedback and they were returned to the classroom/main area of the Science Centre.
Results
Results will be presented in the order of the posed hypotheses. Firstly, we will present results concerning the possible difference in difficulty between an ambiguous
discrimination-learning task and a non-ambiguous discrimination-discrimination-learning task on basis of accuracy data. These results will be discussed per age category. Secondly, we present results regarding the informative value of feedback on basis of the reaction times. Lastly we will consider which role sustained attention played in the results that are obtained.
Ambiguous vs. non-ambiguous tasks
We hypothesized that young children would perform worse on ambiguous sorting tasks than on non-ambiguous sorting tasks, whereas we expected that older children would perform equally successful on ambiguous sorting tasks than in non-ambiguous sorting tasks.
In order to test this, a repeated measures analysis of variance (repeated measures ANOVA) was performed, using accuracy as a dependent variable, age category (category 1: age 4-5, category 2: age 10-11) and condition (ambiguous, non-ambiguous & probabilistic) as a between variable and block as a within variable. The block variable was based on accuracy per block (% correct). The learning data were divided into 4 equal blocks, consisting of 12 trials each (block one consisting of the first 12 trials, block two consisting of the following 12 trials etc.). Mauchly’s test revealed that the assumption of sphericity was violated
(χ²(5)=27.44, p<0.001). As the assumption for sphericity was violated, the within subject
effects were corrected using the Greenhouse-Geisser values.
The results showed that there is a main effect for block (F(2.48)=54.53, p<0.001), meaning that there is in fact a learning curve throughout the trials (multivariate analysis confirm that there is in fact a main effect for block, regardless of the violation of sphericity, F(35,84)=35.84, p<0.001).
Furthermore it was found that there is no significant main effect of condition (non-ambiguous condition versus (non-ambiguous), F(1)=3.23, p=0.076 (See Figure 5). However, it is important to note that the data shows a significant trend towards a performance difference in ambiguous and non-ambiguous tasks.
Figure 5. Mean Percentage Correct Per Block For Ambiguous and Non-Ambiguous Condition
Furthermore, pairwise comparisons showed that there is a main effect for age (F(1)=9.344, p=0.003). On average over all blocks it is shown that the year 6 children outperformed the reception children (Year 6 M=0.907, SD=0.25, Reception M=0.801, SD=0.25). For the non-ambiguous condition year 6 children obtained a correct response average of M=0.924 (e.g., 92.4% correct, SD=0.035, e.g. 3.5%) whereas the reception children obtained a correct response average of M=0.846 (e.g., 84.6% correct, SD=0.034, e.g. 3.4%). For the ambiguous condition year 6 children obtained a correct response average of M=0.890 (e.g., 89.0% correct, SD=0.034, e.g. 3.4%) whereas the reception children obtained a correct response average of M=0.756 (e.g., 75.6% correct, SD=0.036, e.g. 3.6%) (See Figure 6).
Figure 6. Mean Percentage Correct Per Condition For Reception and Year 6 Children
An independent sample t-test was used in order to determine if there was a statistical difference between the ambiguous condition and the non-ambiguous condition for both age groups. The analysis showed that whereas the assumption for equal variances had not been violated for the year 6 group (F(37)=3.127, p=0.085), this was the case for the reception group (F(36)=6.777, p=0.013). Therefore the results for the younger group were obtained using equal variances not assumed.
It became clear that there is no statistical difference between the non-ambiguous and the ambiguous condition in percentage correct, both for year 6 (T(37)=0.882, p=0.192, one-tailed) and reception children (T(26,57)=1.521, p=0.07) (See Table 1). It is however notable that the difference between the ambiguous and non-ambiguous condition does seem to approach a significant trend for the reception children.
Table 1. Accuracy Mean and Standard Deviation (SD) per Condition for Year 6 & Reception Children
Age Group Condition Mean (Accuracy) SD
Year 6 Ambiguous 0.889 0.143
Non-Ambiguous 0.924 0.098
Reception Ambiguous 0.756 0.220
Non-Ambiguous 0.846 0.127
In conclusion, it has become clear that both reception and year 6 children experience a positive learning curve in both conditions. As expected, it has been shown that year 6 children outperform reception children overall. However, even though the difference between the performance on ambiguous vs. non-ambiguous tasks approaches a significant trend for reception children, results have shown that there is no relevant difference to be found on accuracy between both conditions for both age groups.
Informative value of feedback
Positive feedback
As stated before, we expected the phase of the task to be irrelevant for the value that young children assign to both positive- and negative feedback, whereas we expected the phase of the task to be relevant for the value that older children ascribe to given feedback. This was
measured using reaction time, where we expected that the reaction times of younger children would not vary between the phase of deducting the underlying rule (phase 1) and applying the underlying rule (phase 2). In contrary, we did expect the reaction times of phase 1 to be longer than in phase 2 for the older children.
In order to assure that the reaction times used were a correct resemblance of phase 1 (deducting the rule) and phase 2 (applying the rule) we have used the reaction times on the first two trials that followed positive feedback and the last two trials that followed positive
reaction times for each participant. It is important to note that non-learners were excluded from the analysis, as they have never reached phase 2, applying the rule. Therefore the
current analysis was done over 101 participants, 39 in the non-ambiguous condition, 33 in the ambiguous condition and 29 in the probabilistic condition.
The analysis of covariance (ANCOVA) was performed using the average post correct difference score between phase 1 and phase 2 as a dependent variable, the simple average reaction time as a covariate and age (category 1: age 4-5, category 2: age 10-11) as a between factor. Condition was not taken as a factor in the analyses. The results showed that there is a significant overall difference in post correct reaction time between phase 1 and phase 2 (F(1)=18.42, p<0.001), implying that the reaction times were significantly faster once the rule had been discovered in comparison to when this was not the case (See Figure 7).
Furthermore we found a significant main effect for age on post correct reaction time differences whilst not controlling for the average reaction time (e.g. not adding the simple average reaction time as a covariate) (F(1)=14.51, p<0.001), suggesting that post correct reaction time differences were significantly slower for the older children than for the younger children (Year 6 M=0.334, SD=0.794, Reception M=1.331, SD=1.835). However, when the results were viewed whilst including average reaction time in the model, it showed that the main effect for age was no longer apparent (F(1)=3.484, p=0.194). That is, reception children did not differ on post correct reaction times (M=1.235, SD=0.211) compared to year 6
children (M=0.657, SD=0.388) once controlled for average reaction time. From this we can conclude that the difference that was found between younger and older children in average reaction time between both phases of the task can be explained by the fact that older children simply are faster on average than younger children are (Year 6 M=0460, SD=0.278,
Reception M=0.640, SD=0.939).
Figure 7. Average Reaction Time Difference For the Ambiguous, Non-Ambiguous and Probabilistic Condition Per Age Category
*Note. Difference reaction times are in this graph not corrected for average reaction time.
Negative feedback
We did not expect to find a decline in reaction times in the probabilistic task for both younger and older children after receiving negative feedback, as we expected the incongruent feedback to form an interference factor on performance, leaving the children unable to perceive the given feedback as less valuable or irrelevant (Eppinger et al. 2009).
In order to assure that the reaction times used were a correct resemblance of phase 1 (deducting the rule) and phase 2 (applying the rule) we have used the reaction times on the first two trials that followed negative feedback and the last two trials that followed negative feedback. These reaction times were used to create a mean difference score on negative reaction times for each participant. It is important to note that non-learners were excluded from the current analysis, as they have never reached phase 2, applying the rule. Furthermore it is important to note that the current analysis only included the probabilistic condition, as for the other conditions it is impossible to receive negative feedback in the second phase of the task (as the rule has been successfully deducted). Therefore the current analysis was done over 29 participants, 15 year 6 children and 14 reception children.
The analysis of covariance (ANCOVA) was performed using the average post error difference score between phase 1 and phase 2 as a dependent variable, age category (category 1: 4-5, category 2: 10-11) as a between variable and the simple average reaction time as a covariate. The results showed that there is no significant overall difference in post error reaction time between phase 1 and phase 2 (F(1)=0.595, p=0.447), implying that the post error reaction times were not significantly faster overall once the rule had been discovered in comparison to when this was not the case (difference reaction time: M=0.434, SD=3.344). This implies that whilst controlling for average simple reaction times there was no main effect for age on post error difference scores (F(1)=0.007, p=0.936). Indicating that both reception and year 6 children showed an equal post error average difference score (Year 6 M=0.244, SD=0.859, Reception M=0.638, SD=4.817).
Against expectations it was shown that year 6 and reception children show equal average difference reaction times following positive feedback, after controlling for average simple reaction times. The results suggest that although older children are on average faster than younger children; the informative value they ascribe to feedback does not differ per age category. In contrast, for the informative value of negative feedback in the probabilistic learning task the phase of the task is irrelevant; both reception and year 6 children responded
equally fast in the first phase as in the second phase, indicating feedback was not seen as less valuable or irrelevant in the second phase.
Extra findings
A review of the results showed that the performance for both age categories on the
probabilistic task was not as expected, both on reaction times and accuracy (see Figure 8 & 9). Therefore exploratory analyses have been conducted in an attempt to further understand these findings.
In regards of the reaction times it has shown that the average post-correct reaction time of year 6 children was very low on the probabilistic condition (M=0.032, SD=0.882) compared to the reception children (M=1.000, SD=2.146) (See Figure 8). Using a one-sample t-test it became clear that where year 6 children did not perform any faster at all, reception children did actually achieve faster post-correct reaction times (T(19)=2.917, p=0.009) (T(20)=0.165, p=0.871).
Figure 8. Average Difference Reaction Times After Errors per Age Category
*Note. Difference reaction times are in this graph not corrected for average reaction time.
Whilst looking at accuracy, it is interesting to note that whereas year 6 children outperformed reception children on both the ambiguous as the non-ambiguous condition, this did not appear to be the case for the probabilistic condition (See Figure 9). In order to further review these results a repeated measures analysis of variance (repeated measures ANOVA)
was performed over the probabilistic condition, using accuracy as a dependent variable, age category (category 1: age 4-5, category 2: age 10-11) as a between variable and block as a within variable. Results showed an interaction effect between block and age category (F(1)=4.132, p=0.049), showing that whereas the reception children did appear to acquire a state of learning, year 6 children did not appear to do so. A one-sided one sample T-test, with the accuracy on block 4 as a dependent variable, shows that this is indeed the case. Where reception children have reached a higher than chance (>0.5) level accuracy during the final stages of the test (T(19)=2.729, p=0.007), year 6 children did not (T(20)=1.372, p=0.093).
Figure 9. Accuracy per Block for the Probabilistic Condition per Age Category
Sustained attention
As stated earlier there is a possibility that sustained attention ability might influence the performance on the discrimination-learning task. We have seen earlier that year 6 children outperform reception children overall (F(1)=9.344, p=0.003). The current analysis focuses on determining whether the difference in performance variance is (partially) explained by sustained attention ability.
Analysis of the data revealed that whilst there is a substantial spread in how well reception children perform on the sustained attention task there is hardly any spread in the sustained attention performance of year 6 children (See Figure 9); nearly all older children reached a score of nearly 100% correct whereas the younger children seemed to average
around 50% correct (See Table 2). This indicates that the variation that is observed in performance on the discrimination-learning task might be partially explained by sustained attention ability, but only for the reception children. Therefore the current analysis will focus on this age category.
Table 2. Mean Sustained Attention Scores and Standard Deviations (SD) per Age Category
Mean SD
Year 6 0.970 0.077
Reception 0.530 0.385
Figure 9. Spread in Sustained Attention Scores for Reception and Year 6 Children
Using a simple regression analysis it was tested whether sustained attention (partially) explained the variance in performance on the discrimination-learning task. The results showed that sustained attention does not significantly explain variance in the accuracy made on the discrimination learning task for reception children (F(1)=0.151, p=0.700).
Conclusion/Discussion
Hypothesis testing is a process that has been proven to be very difficult for young children at times. The current research has attempted to provide further insight into the underlying mechanisms that contribute to this difficulty. Firstly, the possible influence of the quality of the given feedback was reviewed. It was suggested that ambiguous feedback (e.g. where the
feedback appears to confirm a rule, while in fact this is not the case) would be harder to interpret for younger children than for older children. Furthermore it was expected that older children would not experience any more difficulties as they have already mastered hypothesis testing sufficiently enough to not be led astray by ambiguous feedback. Results showed that even though older children outperformed younger children overall on a hypothesis-testing task, there was no significant difference between non-ambiguous and ambiguous performance, indicating that the quality of the given feedback is not any more relevant for younger children than for older children. The difference between ambiguous and non-ambiguous performance did however approach significance, indicating it might contribute in a minor way to difficulties on a task such as the discrimination-learning task.
Secondly it was attempted to show whether young children ascribed a different informative value to feedback than younger children, as older children have a better understanding of when feedback is more, or less, relevant. Against expectations we found this not to be the case. The observed difference in reactions between older and younger children was simply explained by average reaction speed, indicating that both young and older children ascribe similar values to feedback. However, it was found that both young and older children are able to overview that feedback is less relevant once a rule has been discovered.
Lastly it was reviewed whether sustained attention played a role in observed differences on hypothesis-testing performance. It was found that close to none of the performance variance was explained by sustained attention.
In short; the difficulty that young children experience in discrimination-learning tasks is not explained by the quality of given feedback, nor is it understood through a difference in informative value that is ascribed. How young children perform on such tasks varies enormously, but the variation that is found cannot be explained by sustained attention abilities.
Ambiguity of feedback
Against expectations we did not find the ambiguity of feedback to play a role in the hardships young children experience in hypothesis-testing tasks; young children did not perform any worse than older children did whilst receiving ambiguous feedback. Regardless, it is important to note that there was a significant trend in performance difference between the ambiguous and non-ambiguous condition for the younger children. Furthermore, the data that was used for the analysis was left-censored; the testing was stopped before some children
than expected. As currently we have found a significant trend for the younger children, it can be assumed that there is a small effect of ambiguity of feedback on hypothesis-testing performance. Similarly it was found that the overall difference between ambiguous and non-ambiguous performance approached a significant trend, indicating that there might be some influence of ambiguity on performance.
Regardless, this effect, if existing, would be small. From this we can conclude that there must be different factors contributing to the difficulty young children experience in these types of tasks. Future research could possibly invest more time in reviewing the influences of other factors, such as inhibition, working memory or cognitive flexibility.
Reaction times
It is important to note that the touch screen computer device that was worked with in the current research did not always perform smoothly. At times (e.g. certain trials) the device would react slowly. However, we have taken several precautions to assure that the effect of the possible malfunctioning of the device was kept to a minimum. Firstly, each participant completed both 48 trials on the reaction time task and 48 trials on the discrimination-learning task. The large amount of trials has contributed to minimizing any undesirable effects. Secondly we have used the average reaction time as a control measure in the analysis. It is safe to say that if the device resulted in a participant being, for example, half a second slower, this participant would have had the same in both tasks. Using a difference score will have consequently served the purpose of minimizing negative influences.
Informative value
Against expectations we found that not just older children are able to process the relevancy of positive feedback correctly, younger children are able to do so as well. However, for both age groups it proves to be very difficult when this feedback becomes probabilistic. It could be that children simply assign more value to negative feedback than adults; they might see it as a direct reflection on their work and, unlike adults, not being able to see the probabilistic nature of the feedback.
Probabilistic task
The probabilistic task was, in contrary to expectations, perceived to be equally difficult for younger and older children and in some cases the young children even seem to outperform the older children. It was shown that the average post-correct reaction time of year 6 children
was surprisingly low on the probabilistic condition. Where reception children did achieve faster post-correct reaction times, year 6 children did not perform any faster at all. This infers that year 6 children have not made any improvements between phase 1, deducting the rule, and phase 2, applying the rule, on the probabilistic task. This is strange considering the research of Eppinger et al. (2009) where adults seemed unaffected by the incongruent feedback, whereas young children seemed to struggle.
A possible explanation for this finding is that year 6 children are simply not yet similarly far developed as the adults in the research of Eppinger et al. (2009), which causes them to be unable to be as successful on the task as adults are. However, it is noteworthy that the reception children appeared to outperform the year 6 children on the task when we look at informative value that is ascribed to given feedback (e.g. where the reception children
improved on reaction times after deducting the rule, year 6 children did not). It is important to be aware of the fact that the older children frequently reported attempting extremely complex rule deduction (i. e. ‘it must be the left picture twice and after that it has to be the red picture, but only if it appears on the right side’). Consistently with our data it seems to be that the older children are looking for ‘the perfect rule’, explaining the complete feedback pattern, whereas young children seem to ‘give up’ earlier, simply trying the picture that is most frequently correct. Future research should be directed to explaining this better. On that note it would be interesting to further investigate whether for example adolescents, e.g. 16 or even 18 years of age, display a similar pattern of trying to deduct ‘the perfect rule’ as the older children in our sample did. If this is the case it might be that being able to properly deal with probabilistic feedback is directly related to developmental processes, not fully matured until a later age.
The role of sustained attention
It is very interesting to note that sustained attention does not play a role in the variance that was found in performance for the young children. Even though sustained attention ability is an obvious developmental difference observed in younger children, it does not seem to contribute to performance on a task such as done in the current research.
Limitations and strengths
First of all the current research has shown that children as young as the age of 4-5 have an equally good insight in the value of feedback as children as old as the age of 12 seem to have.
contributing factor to difficulties that children experience in feedback learning, it is not a main factor. Interestingly it has been proven as well that sustained attention forms no explanation for variance that is found in performance; indicating that this does not play a major influencing role in discrimination learning.
The cross-sectional nature of the study may however form a limitation for the current research. A solution to this problem might be attempting a similar research on a longitudal basis. This will enable us to review the developmental processes observed in discrimination learning in a better and more complete manner. However, it must be noted that task-learning effects must be then taken into consideration.
Furthermore we have only used one measure of executive functioning in order to explain the results found; sustained attention. In order to obtain a complete image of discrimination learning performance it is suggested for future research to include more aspects of executive functioning, such as working memory and inhibition skills.
Appendix
A. Task Instructions
Instructions Sustained Attention Task
We are now going to play a game in which you have to do two things. Do you see all the figures on the screen? (point out to all different figures) When I press a button all these figures are going to move across the screen at the same time. You will have two jobs. Do you see a figure that has a red circle around it? (let child point out the figure, if he/she names the
figure, ask if they can point out which one that is to verify). I would like you to follow that
figure with your eyes while all the figures are moving. All of a sudden all the figures will disappear. I would like it if you could point out, with your finger on the screen, in which of the little squares (point out the gridlines) you have seen that figure (point out target figure) right before all shapes disappear. Afterwards I also wonder if you will be able to tell me what the figure you had to follow looked like. Let’s give it a try! Are you ready? (start first trial). So, where was the figure that you had to follow with you eyes, right before it disappeared? (wait for the child to point out the location of choice) Well done! Can you also tell me what the figure that you had to follow looked like? You can again point it out on the screen. (wait
until the child has chosen one of the available figures). Well done! We will now do that
another 6 times, but you have to follow a different figure. Do you know what that figure is? (check if the child realises what the target figure is, if not, point out the red circle around the
target figure and ask again until it is certain the child understands). (continue until task is completed).
Instructions Simple Reaction Task
We are now going to play a game with fishes. When I tell you that you can begin, you can put your finger on the target that you see here in the middle (point out target) (for a complete overview on visual task overviews see Appendix B. Visionary Overview Tasks). When you leave your finger on the target, one fish will appear, right here (point out where the fish will
appear). The fish will either be swimming to the right (point out to the right square that is visible on the screen) or swimming to the left (point out to the left square that is visible on the screen). If the fish would have his face facing this way (point out either left or right),
which way would it be swimming towards? (verify if the child has understood, if not, explain
again). In this task it is important that you are as fast as you can, but also that you make as
little mistakes as possible. Do you think you can do that? Okay then, let’s give it a try, put your finger on the target. (check if the child does the task correctly). Well done! Now we are going to that many more times, remember, be as fast as you can with as little mistakes as you can! (continue until task is completed).
Instructions Discrimination Learning Task
We are now going to play a game with pictures. When I say that you can begin you can put your finger on the target in the middle of the screen, just as we did with the fish game. When you have done that, two pictures will appear, here (point out to the right side on the screen) and here (point out to the left side on the screen). You can choose one of both pictures by clicking on one with your finger. In the beginning you can just choose the pictures, maybe the ones that you find pretty or silly, that is up to you. After a little while, the computer is going to tell you whether you have made the right choice, then you’ll see this (point out the happy
smiley face) or if you have made the wrong choice, then you’ll see this (point out the sad smiley face). I would like you to try to make as many right choices as you can. (For the
probabilistic task the following instruction is added: You do have to remember though;
the computer can be a little bit silly!) Which of the smiley faces will you see when you have made a right choice? (verify whether the child understands). Good! Then we will begin. Remember that you can start by just choosing the pictures, it will be a little while before the computer tells you whether you have made the right, or wrong choice. Okay, let’s give it a try,
there are the two pictures. You can now pick one. (after 6 blank trials the computer will start
giving feedback). See, the computer is now telling you whether you have made the right
choice. Try to get as many happy smileys as you can! (continue until task is completed).
B. Visionary Overview Tasks
Start Screen Discrimination Learning Task & 2-Choice Simple Reaction Time Task + Target.
Choice Screen Discrimination Learning Task
Feedback Overview; Correct (Happy Smiley, Left) and Incorrect (Sad Smiley, Right).
References
Eppinger, B., Mock, B., & Kray, J. (2009). Developmental differences in learning and error processing: Evidence from ERP’s. Psychophysiology, 46, 1043 – 1053. Fisher, AV., Thiessen, E.D., Godwin, K.E., Kloos, H., & Dickerson, J.P. (2013).
Assessing selective sustained attention in 3- to 5-year-old children: Evidence from a new paradigm. Journal of Experimental Child Psychology, 114, 275 – 294.
Kendler, T.S. (1979). The development of discrimination learning: A levels-of-functioning explanation. Advances in Child Development and Behavior, 13, 83 – 117.
Kirkham, N.Z., Cruess, L., & Diamond, A. (2003). Helping children apply their knowledge to their behavior on a dimensionswitching task. Developmental Science, 6, 449–476. Schmittmann, V.D., van der Maas, H.L.J. & Raijmakers, M.E.J. (2012). Distinct
discrimination learning strategies and their relation with spatial memory and
attentional control in 4- to 14-year-olds. Journal of Experimental Child Psychology, 111, 644 – 662.
Van Bers, B.M.C.W., Visser, I., & Raijmakers, M.E.J. (2014). Preschoolers can form abstract rule representations regardless of cognitive flexibility. Journal of Experimental Child Psychology, 24, 50 – 66.
Zelazo, P.D., Carlson, S.M., & Kesek, A. (2008). The development of executive function in childhood. In: Nelson, C., Luciana, M. (Eds.), Handbook of Developmental Cognitive Neuroscience, 2nd ed. MIT Press, Cambridge, MA.
