Positive feedback increases feedback-based learning and motivation more
than negative feedback in adolescents.
Elbrig Jansma (11642467)
Department of developmental psychology, University of Amsterdam Bachelor thesis Psychobiology
Prof. Dr. Hilde Huizenga & Anne-Wil Kramer June 19, 2020
Abstract
Feedback has been shown to improve performance in learning. Since, improved performance means that students do better at school, feedback-based learning is important in education. In adolescents, it has not been established if either positive or negative feedback is more
beneficial for learning. Therefore, the effect of feedback valence on feedback-based learning in adolescents was studied. Moreover, motivation in education is important for improving performance as well. Communication of expectations and communication from teachers, which include feedback, are aspects that increase motivation towards learning. Again, in adolescents it has not been established which feedback valence is the most beneficial for motivation in learning, so the effect of feedback valence on motivation in feedback-based learning in adolescents was studied as well. This was tested by means of a reinforcement learning task in which adolescent VMBO students learned through positive and negative feedback. To measure motivation, an effort discounting task was used. In addition, a Raven’s Progressive Matrices test was conducted to explore the interaction of cognitive ability and to see if the same feedback valence should be used in different educational levels. The results indicated that positive feedback significantly increased performance and motivation in
feedback-based learning relative to negative feedback. No interaction of cognitive ability was found. In conclusion, this shows that in the education of adolescents more use should be made of positive feedback to improve feedback-based learning and motivation, regardless of educational level.
Key words: Feedback valence, feedback-based learning, motivation, adolescents, education.
Introduction
It has been proven that feedback, both positive and negative, is an important aspect of
enhancing performance, making it one of the key foundations for successful learning (van den Bos, Cohen, Kahnt & Crone, 2012; Raaijmakers, Baars, Schaap, Paas, & van Gog, 2017; van Duijvenvoorde, Zanolie, Rombouts, Raijmakers & Crone, 2008). In education, enhancement of performance would mean that students do better at school, which makes feedback-based learning important in education. In feedback-based learning, people learn by establishing which actions provide the most reward through trial and effort. Therefore, reward signals and trial and error learning are the most important aspects of feedback-based learning (Sutton & Barto, 1998). In educational environments, students learn through reward signals as well, because they receive positive and negative feedback. In the context of enhancement of performance in education, it is therefore relevant to observe if positive and negative feedback affect feedback-based learning differently.
Van den Bos et al. (2012) used feedback-based learning to look at differences in sensitivity to feedback between negative and positive feedback by determining the learning rate. The learning rate measures the impact of feedback on future behaviour and thus, indirectly measures the effect feedback valence could have on learning. They did not find a significant difference between the learning rates of positive or negative feedback in adolescents, which is supported by research of Christakou et al. (2013). However, Hauser, Iannaccone, Walitza, Brandeis and Brem (2015) showed that in adolescents the learning rate for negative feedback was significantly increased and that they learned faster from negative feedback than adults. These studies only focused on the learning rate in feedback-based learning but van
Duijvenvoorde et al. (2008) looked at performance in feedback-based learning, which measures learning as well. In their study, it was found that children, early adolescents and adults performed faster and more accurately on a feedback-based learning task after positive feedback relative to negative feedback. Thus, the findings on the effect of feedback valence on feedback-based learning in adolescence remain inconsistent and are mostly based on
learning rate instead of performance. Therefore, the following research question will be studied: What is the effect of feedback valence on the performance in feedback-based learning in adolescents?
This research will also look at the effect of feedback valence on motivation in feedback-based learning, since motivation increases the will to learn in students and therefore is considered important in the context of performance enhancement in learning (Christophel, 1990). Motivation is a complex concept that consist of multiple motivational variables such as self-efficacy and goal-orientation, which subsequently influence effort (Brookhart, Walsh & Zientarski, 2006). When motivation to learn is absent, it is unlikely that this effort will be exerted (Harlen & Deakin Crick, 2003). Instructions and both communication from teachers and communication of expectations are aspects that stimulate motivation toward learning (Christophel, 1990). This communication includes feedback, which can be positive or
negative, therefore it is interesting to observe whether feedback valence affects motivation in feedback-based learning differently.
Earlier studies that investigated the effect of feedback valence on motivation mainly found that positive feedback improved motivation. For instance, Venables and Fairclough (2009) studied the influence of positive and negative feedback, and the perception thereof, on effort investment and motivation in adults. They found that motivation was most declined after negative feedback, but that more effort was invested, presumably to compensate for failure, showing that positive feedback improved motivation whereas negative feedback increased invested effort. These findings are in line with the research of Burgers, Eden, van
Engelenburg and Buningh (2015), who examined the effect of feedback valence on motivation in adults through brain-training games and determined that positive feedback increased motivation. They further determined that negative feedback increased immediate gaming-behaviour and therefore also invested effort. In addition, these findings have been found in children as well, as Mabbe, Soenens, de Muynck and Vansteenkiste (2018) showed that in children the motivation to complete puzzles was increased through positive feedback. These findings of increased motivation following positive feedback are supported by multiple studies (Sansone & Harackiewicz, 2000) Thus, the effect of positive feedback and negative feedback on motivation has been studied in children and adults, but not in adolescents and also not focused on motivation in learning. This leads to the second research question: What is the effect of feedback valence on motivation in feedback-based learning in adolescents?
To study both research questions, a reinforcement learning task with an easy, a medium and a difficult condition will be used, in which the adolescent VMBO students will learn through both positive and negative feedback. Afterwards, to operationalise motivation, an effort discounting task will be done to determine the motivation for the reinforcement learning task. The motivation in the effort discounting task is assessed by a so-called indifference score, which is part of the cognitive effort discounting paradigm (Westbrook, Kester & Braver, 2013). Furthermore, a Raven Progressive Matrices test was added to measure cognitive ability (Langener, Kramer, van den Bos & Huizenga, 2020). In the Netherlands namely, secondary schools have different levels, which are classified on cognitive ability. Hence, it is of social relevance to determine if the same feedback valence should be used in all levels. Therefore, it will be explored if cognitive ability interacts with the effect of feedback valence on feedback-based learning or motivation.
There is no expectation on the effect of feedback valence on reinforcement learning, because of inconsistencies in the findings of previous research. Furthermore, it is
hypothesized that there will be a main effect of positive feedback on motivation in learning, meaning that positive feedback will have more positive effect on motivation in feedback-based learning than negative feedback. Thus, it is expected that the indifference score on the effort discounting task will be significantly higher for the positive feedback condition.
Materials and methods
The research described in this thesis was conducted during the coronavirus outbreak in 2020. Introduction and methods sections were written before closing of secondary schools. Since data had to be gathered at secondary schools, data-collection was impossible. Therefore, the results and discussion sections are based on simulated, that is fake, data. Thus, this thesis only serves educational, and no scientific purposes.
Participants
Adolescent VMBO students between the age of 13 and 15 were chosen to participate in this study. The students were recruited from a secondary school in the Netherlands. The study was approved by the ethics review board and all participants had to sign an informed consent prior to participation. The data consisted of 117 participant of which 43 were women and 74 were male.
Procedure
The data collection was done at a secondary school in the Netherlands where 7 classes of 2nd or 3rd graders have been tested. The participants with a diagnosis of neurological or
psychological problems were excluded. To promote participation the participants were able to win a chocolate bar and received the final reward they got in the effort discounting task. First, it was explained to the participants what the tests would entail and the informed consent had to be signed. Then, we administered the reinforcement learning task, the effort discounting task and lastly the Raven’s Progressive Matrices task. All tasks together took approximately 40-50 minutes. When the participants were done they received their reward and the winner of the chocolate bar was announced.
Experimental design and materials
Participants had to take three tests. First, the reinforcement learning task which consisted of 6 blocks defined by difficulty (easy, medium and difficult) and valence (positive and negative feedback). Second, an effort discounting task was done in which they had to choose which subtasks they would like to redo and for which reward. Last, they had to take a Raven’s Progressive Matrices test to measure their cognitive ability. The reinforcement and effort discounting tasks were made in this research and programmed in Neurotask (Neurotask, 2014). The shortened Raven Progressive Matrices test was made in Qualtrics. All of the tasks were accomplished on a computer.
The reinforcement learning task started with either the negative feedback condition or the positive feedback condition, which was randomly assigned to each participant. Before each feedback condition there was a practice block of two word pairs that were each shown 10 times, thus 20 trials. They had to learn which spelling of the word pair was correct by using the received feedback. In the positive feedback condition a check mark was presented on the screen after the correct spelling was chosen and a dash was presented after the incorrect spelling was chosen (figure 1). For the negative feedback condition a dash was presented after the correct spelling was chosen and a cross was presented after the incorrect spelling was chosen. Both the positive and the negative feedback condition consisted of three blocks. An easy block in which the correct spelling of two non-existent word pairs had to be learned, a medium block in which the correct spelling of three non-existent word pairs had to be learned and a difficult block in which the correct spelling of four non-existent word pairs had to be learned. The words all consisted of 4 - 6 letters, an ei/ij, au/ou or g/ch, varied at least one letter from an existing Dutch word and acted in accordance with the Dutch language rules.
For each participant it was pseudo-randomly determined which spelling was correct, this did not change during the blocks. These blocks were randomized in each feedback condition and consisted of 40 trials, thus for example each word pair in the easy block was shown 20 times and each word pair in the difficult block was shown 10 times. In total the participants had to do 2 x 120 = 240 trials. For all 6 conditions, the performance on the task was measured as the percentage of correct answers. After each block the percentage of correct answers was shown to enhance the motivation. There were short breaks after each block and before the other feedback condition. A percentage of 80% congruence was used in the feedback trials to avoid a ceiling effect in learning, meaning that 4 out of 5 feedback trials were congruent. The order of congruence was fixed in set-size and feedback condition to enable equal task difficulty. In every trial a fixation cross was shown first, then the non-existent word pairs were shown with a picture, lastly after the participant chose an answer, the feedback was shown. The duration of each trial was approximately 3 - 4,5 seconds, thus the task took about 240 x 3- 4,5 = 12 - 18 minutes.
Figure 1. Negative and positive feedback condition. This figure shows the trial of a negative and positive feedback condition in the reinforcement learning task (Kramer et al., 2020a). In the effort discounting task the participants had to repeatedly choose which block of the learning task they wanted to redo and for what reward. The participants had to choose if they wanted to do a low effort task (2-word pair block) and receive a small reward of €1 or if they wanted to do a high effort task (3- or 4-word pair block) and receive a higher reward of €2. The task consisted of 4 conditions and in each condition there were 5 trials, thus the
participants had to make 4 x 5 = 20 choices. The different choice conditions consisted of an easy vs. medium task and an easy vs. difficult task for both the negative valence and the positive valence (table 1). When the participant chose to do a high effort task the reward of the next low effort task that could be chosen was increased, whereas the reward of the next low effort task that could be chosen was decreased after the participant chose to do a low effort task. The final reward for the low effort task at the end of the trials was considered the indifference point, which portrayed the amount of effort the participants were willing to invest into the high effort task compared to the low effort task (figure 2) (Westbrook et al., 2013). A high indifference point indicated that the person was willing to invest much effort into the task, while a low indifference point indicated that the person was willing to invest less effort into the task. Since, there were 4 conditions, each participant had 4 indifference points. At the end of the task the participants had to redo one of the 20 choices, which was randomly picked.
This was mentioned at the beginning of the task, to make sure honest choices were made. There was no time limit on the trials, thus the duration of the trials and task depended on the response time of the participants. The task was estimated to take approximately 5 to 10 minutes.
Feedback Easy vs. medium task Easy vs. difficult task Negative 2-word negative vs. 3-word
negative 2-word negative vs. 4-word negative Positive 2-word positive vs. 3-word
positive
2-word positive vs. 4-word positive
Table 1. The 4 different choice conditions of the effort discounting task.
Figure 2. Example choices effort discounting. This figure shows an example of choices in the effort discounting task, which eventually will lead to an indifference point (Kramer, van Duijvenvoorde, Krabbendam & Huizenga, 2020b)
Last, the participants had to do a shortened Raven’s Progressive Matrices test (Langener et al., 2020), which estimated the cognitive ability. The task consisted of a large figure on the screen of which a part had been left out. The participants had to choose from 6 smaller figures which one fitted into the large figure and into the pattern (figure 3). There were 15 items in total, meaning that the results could lead to a maximum cognitive ability score of 15.
Figure 3. Example question Raven. This figure is an example of one of the questions from the Raven Progressive Matrices test (Langener et al., 2020).
Data analysis
For the data analysis SPSS was used.
Research question 1: The independent variable is feedback valence and the dependent variable is the percentage of correct answers. From the collected data, the percentage of correct answers from all 3 blocks of both the negative and the positive feedback condition was assessed. By using a repeated measures ANOVA, the mean percentage of correct answers of both conditions within each subject were compared, which would show whether there was a significant difference in the performance between both feedback conditions.
Research question 2: The independent variable is feedback valence and the dependent
variable is the indifference score. In the collected data were 4 indifference points, 2-3P, 2-4P, 2-3N and 2-4N. By using a repeated measures ANOVA, the means of the indifference points between both feedback conditions within each subject were compared, which would show if there was a significant difference in motivation (indifference scores) between both feedback conditions.
Exploratory analysis: To study the interaction between cognitive ability, feedback valence and feedback-based learning or motivation the same repeated measures ANOVA was done, however cognitive ability was added as the covariate (repeated measures ANCOVA).
Results
The data from all the participants was used, none were taken out. The mean age was 14.77 (SD = 0.77).
Effect of feedback valence on reinforcement learning
Figure 4. Feedback effect on feedback-based learning. This graph shows the mean percentage of correct answers from the reinforcement learning task for the positive and negative feedback conditions.
The mean performance on the reinforcement learning task for the positive and negative feedback condition for all within subjects can be seen in figure 4. The assumption that the mean difference had a normal distribution was met. A significant effect of feedback valence on feedback-based learning was found, thus the performance on the task differed significantly between the positive and negative feedback condition, F(1, 115) = 18.92, p < .001. This is also evident from figure 4, the mean percentage of correct answers from the positive feedback condition (M = 0.77, SD = 0.07) is substantially higher than the mean percentage of correct answers from the negative feedback condition (M = 0.67, SD = 0.07). In short, the statistical result and graph indicate that positive feedback has significantly more positive effect on feedback-based learning than negative feedback.
Effect of feedback valence on motivation
Figure 5. Feedback effect on motivation. This graph shows the mean indifference score from the effort discounting task for both the positive and negative feedback conditions.
The mean indifference score on the effort discounting task for the positive and negative feedback condition for all within subjects can be seen in figure 5. The assumption that the mean difference had a normal distribution was met. A significant effect of feedback valence on motivation was found, thus the indifference score differed significantly between the positive feedback condition and the negative feedback condition, F(1, 115) = 411.76, p < . 001. This is also evident from figure 5, the mean indifference score from the positive feedback condition (M = 1.43, SD = 0.08) is considerably higher than the mean indifference score from the negative feedback condition (M = 1.07, SD = 0.08). The statistical result and the graph indicate that positive feedback has significantly more positive effect on motivation than negative feedback.
Exploratory analysis
The assumptions of the factorial repeated measures ANCOVA were met. No significant interaction between cognitive ability, feedback valence and feedback-based learning was found, F(1, 115) = 2.003, p = .160. Furthermore, no significant interaction between cognitive ability, feedback valence and motivation was found, F(1, 115) = 0.131, p = .718. In short, the results indicate no interaction effect of cognitive ability on the effect of feedback valence on both feedback-based learning and motivation.
Discussion
The aim of this research was to study how feedback, positive or negative, affects feedback-based learning or motivation in feedback-feedback-based learning in adolescents.
In this research it was determined that positive feedback had a significantly more positive effect on feedback-based learning than negative feedback. There was no expectation
beforehand, because of inconsistencies in the findings of previous research. However, our findings do not agree with most of these previous findings, which mainly found that there was no difference in the effect of feedback valence or that negative feedback had more positive effect. This could be explained by the fact that most of these previous studies used learning rate as a measure of feedback-based learning, while we used performance. Learning rate is a measure of the impact of feedback on future behaviour and indirectly measures the effect on learning, while performance is more a direct measure of learning. Therefore, this research is more suitable to answer the research question. This is supported by the fact that our findings are in agreement with the finding from van Duijvenvoorde et al. (2008) who used
performance on the task as a measure of feedback-based learning as well, and found that the performance was improved after positive feedback.
Furthermore, it was determined that positive feedback increased motivation in feedback-based learning more than negative feedback. This was expected, thus our findings are in agreement with our hypothesis and previous findings. Moreover, most of these studies used different tasks, such as a cognitive task or a brain-training game, to study motivation whereas this research specifically focused on motivation in feedback-based learning (Venables & Fairclough, 2009; Burgers et al., 2015; Mabbe et al., 2018). As mentioned before, feedback-based learning is important and specifically related to education, while the tasks used in previous research are not. Therefore, our results do not only support the finding that positive feedback increases motivation in adolescents, but further indicate that it specifically applies to motivation in education.
Last, no interaction between cognitive ability, feedback valence and both feedback-based learning and motivation was found, meaning that there is no difference in the effect of
feedback valence on feedback-based learning or motivation between adolescent with a high or a low cognitive ability. To our knowledge, this is the first research that explored the
interaction between cognitive ability, feedback valence and feedback-based learning or motivation. Therefore, we can cautiously assume that cognitive ability does not interact with the effect of feedback valence on feedback-based learning or motivation in learning.
However, more research is needed to support this finding, possibly with a more advanced measure of cognitive ability than a shortened Raven’s Progressive Matrices task.
A limitation of this research might be the ecological validation of money as the reward in the effort discounting task, since in school, students do not receive money to increase their motivation or to base their decisions on. Moreover, money is an external reward that only improves short-term motivation (Urminsky, 2018). While, long-term motivation is better and more important in education. However, this is the first experimental task used to measure motivation in learning, since mostly questionnaires are used. An experimental task can determine a causal relation and is a more objective measure of motivation than a
questionnaire. Therefore, this operationalization provides a better and more realistic view on motivation in learning than the general methods. In addition, the participants consisted of only VMBO students, which is one level of the educational system in the Netherlands. Therefore, if we want to explore the interaction of cognitive ability and see if the same feedback should be used in different educational levels, this group might not be variable enough. Thus, to be
able to conclude with more certainty that there is no interaction of cognitive ability and make the results more generalizable, this research could be replicated with a larger group of
participants with more variety in educational levels. Finally, for future research it might be interesting to examine if in adolescents the communication style of feedback could affect motivation in learning. Research showed that an autonomic-supportive feedback style, compared to a controlling feedback style, improved motivation and enjoyment of a task or game in children and adults (Mabbe et al., 2018; de Muynck et al., 2017). Despite these findings, little research has been conducted in adolescents on the effect of the communication style of feedback on motivation in learning. The findings of such a research would help to further improve the motivation of students in education.
To conclude, in adolescents positive feedback has more positive effect on both feedback-based learning and motivation than negative feedback and is not affected by cognitive ability. Therefore, we suggest that in the education of adolescents more use should be made of positive feedback.
References
Burgers, C., Eden, A., van Engelenburg, M., & Buningh, S. (2015). How feedback boosts motivation and play in a brain-training game. Computers in Human Behavior, 48, 94-103.
Brookhart, S.M., Walsh, J.M., & Zientarski, W.A. (2006). The Dynamics of Motivation and Effort for Classroom Assessments in Middle School Science and Social Studies. Applied Measurement in Education, 19(2), 151–184.
https://doi.org/10.1207/s15324818ame1902_5
Christakou, A., Gershman, S.J., Niv, Y., Simmons, A., Brammer, M., & Rubia, K. (2013). Neural and Psychological Maturation of Decision-making in Adolescence and Young Adulthood. Journal of Cognitive Neuroscience, 25(11), 1807–1823.
https://doi.org/10.1162/jocn_a_00447
Christophel, D. (1990). The relationships among teacher immediacy behaviors, student motivation, and learning. Communication Education, 39(4), 323–340.
https://doi.org/10.1080/03634529009378813
De Muynck, G., Vansteenkiste, M., Delrue, J., Aelterman, N., Haerens, L., & Soenens, B. (2017). The Effects of Feedback Valence and Style on Need Satisfaction, Self-Talk, and Perseverance Among Tennis Players: An Experimental Study. Journal of Sport & Exercise Psychology, 39(1), 67–80. https://doi.org/10.1123/jsep.2015-0326
Harlen, W. & Deakin Crick, R. (2003). Testing and Motivation for Learning. Assessment in Education: Principles, Policy & Practice, 10(2), 169–207.
https://doi.org/10.1080/0969594032000121270
Hauser, T.U., Iannaccone, R., Walitza, S., Brandeis, D., & Brem, S. (2015). Cognitive flexibility in adolescence: Neural and behavioral mechanisms of reward prediction error processing in adaptive decision making during development. NeuroImage, 104, 347–354.
https://doi.org/10.1016/j.neuroimage.2014.09.018
Kramer, A., Schaaf, J. V., Cousijn, J., Snellings, P. J., Larssen, H., Kuhns, L., & Huizenga, H. M. (2020a). Information or motivation: an fMRI investigation into the effects of negative versus positive feedback. In preparation
Kramer, A., van Duijvenvoorde, A. C. K., Krabbendam, L., & Huizenga, H. M. (2020b) Individual differences in adolescents' willingness to invest cognitive effort: relation to need for cognition, motivation and cognitive ability. Under review
Langener, A., Kramer, A., van den Bos, W. & Huizenga, H.M. (2020). A Shortened Version of Ravens’ Standard Progressive Matrices for Children and Adolescents. Submitted
Mabbe, E., Soenens, B., de Muynck, G.J., & Vansteenkiste, M. (2018). The impact of feedback valence and communication style on intrinsic motivation in middle childhood: Experimental evidence and generalization across individual differences. Journal of Experimental Child Psychology, 170, 134–160.
https://doi.org/10.1016/j.jecp.2018.01.008
Raaijmakers, S.F., Baars, M., Schaap, L., Paas, F., & van Gog, T. (2017). Effects of performance feedback valence on perceptions of invested mental effort. Learning and Instruction, 51, 36-46. https://doi.org/10.1016/j.learninstruc.2016.12.002
Sansone, C. & Harackiewicz, J. M. (2000). Intrinsic and extrinsic motivation: The search for optimal motivation and performance. Elsevier.
Sutton, R.S. & Barto, A.G. (1998). Reinforcement Learning An Introduction. Cambridge, Mass: MIT Press.
Urminsky, O. (2018, May 16). Short-term rewards don’t sap long-term motivation. Retrieved on June 2, 2020 from
https://review.chicagobooth.edu/behavioral-science/2018/article/short-term-rewards-don-t-sap-long-term-motivation
Van den Bos, W., Cohen, M.X., Kahnt, T., & Crone, E.A. (2012). Striatum-medial prefrontal cortex connectivity predicts developmental changes in reinforcement learning. Cerebral Cortex, 22(6), 1247-1255. https://doi.org/10.1093/cercor/bhr198
Van Duijvenvoorde, A.C.K., Zanolie, K., Rombouts, S.A.R.B., Raijmakers, M.E.J., & Crone, E.A. (2008). Evaluating the negative or valuing the positive? Neural mechanisms supporting feedback-based learning across development. Journal of Neuroscience, 28(38), 9495–9503. https://doi.org/10.1523/JNEUROSCI.1485-08.2008
Venables, L. & Fairclough, S.H. (2009). The influence of performance feedback on goal-setting and mental effort regulation. Motivation and Emotion, 33(1), 63-74.
https://doi.org/10.1007/s11031-008-9116-y
Westbrook, A., Kester, D., & Braver, T.S. (2013). What Is the Subjective Cost of Cognitive Effort? Load, Trait, and Aging Effects Revealed by Economic Preference. PLoS ONE, 8(7). https://doi.org/10.1371/journal.pone.0068210
