The dual processing self-regulating model
The Dual Processing Self-Regulating Model from Boekaerts (2011) describes the essential role of emotions in learning. Boekaerts (2011) claimed that emotional states guide the learner's behavior onto one of two possible pathways. She proposed a well-being and a growth pathway as self-regulatory strategies, depending on how the task is assessed. Tasks that do not fit the current mental model trigger negative emotional states, which are detrimental for knowledge increase, leading the learner to take the well-being pathway. Tasks that correspond with the learner’s goals cause positive emotional states and thus open the growth pathway, resulting in knowledge increase. Measuring learner's emotional states can therefore propose a statement about learning success.
Furthermore, it is possible to switch from one pathway to the other. If learners are on the growth pathway and detect indicators for failing, they shift to the well-being pathway (Boekaerts, 2011). Determining this emotional shift in real-time enables immediate support and therefore guides the learner back on the growth pathway (see Arguel et al., 2017; D'Mello & Graesser, 2014). We want to find an appropriate “on-the-fly” measure that can identify negative emotional states during learning with CBLEs, as a step towards the primary goal of guiding and keeping the learner on the growth pathway.
Given that emotions are concomitants of learning, it is necessary to differentiate these academic emotions specifically (Pekrun & Stephens, 2012). Academic emotions, which can be seen in Table 1, are related to achievement, classroom settings, and learning. They are bound to success and failure, but also to the process of learning itself (Goetz & Hall, 2013; Pekrun et al., 2002, 2017). Multiple research approaches address academic emotions (e.g., confusion: D'Mello et al., 2014; boredom: Goetz & Hall, 2013; Pekrun, 2006; Pekrun et al., 2002). The underlying concept of this work is the Three-Dimensional Taxonomy of Academic Achievement Emotions from Pekrun (2006), which classifies academic emotions in three dimensions: their valence (positive or negative), activation (activating or deactivating), and object focus (activity or outcome; see Table 1). Enjoyment, for example, is, according to Pekrun (2006), a positive and activating academic emotion, during an activity (e.g., studying). In comparison, sadness is defined as negative and deactivating academic emotions triggered by pro- or retrospective failure (e.g., upcoming or past exams).
In the psychophysiological literature, the term “arousal” is more common than “activation” (e.g., Berntson et al., 2017; Lang et al., 2009; Levenson et al., 2017; Potter & Bolls, 2012). To have consistent terminology in this article, we refer to the term “activation”.
Negative academic emotions usually trigger task-irrelevant thoughts and decrease the resources required for the task. Therefore, learning performance may decline if a learning goal seems unachievable due to prevalent negative academic emotions. However, negative activating academic emotions can also cause intense motivation to prevent failure, resulting in solving the task and increasing learning performance (Pekrun & Stephens, 2012). The shift from detrimental and conducive emotional states is also supported by Boekaerts’ Dual Processing Self-Regulating Model (2011; see chapter “Theoretical framework”), where learners switch from the well-being pathway to the growth pathway. Depending on the learner's assessment and the apparent solvability of a task, emotional states can change, and even knowledge can increase despite experiencing negative emotions during learning (Boekaerts, 2011).
Furthermore, task difficulty can affect academic emotions due to cognitive incongruity (Pekrun & Stephens, 2012). If the task seems too tricky or non-solvable, negative academic emotions are triggered, resulting in low learning performance (Baker et al., 2010; D'Mello & Graesser, 2014). Otherwise, positive academic emotions arise if a learning task can be solved, leading to high learning performance (Kang et al., 2008; Pekrun & Stephens, 2012).
In the present study, we decided to focus on negative activating academic emotions to reduce complexity. Besides, it is more valuable to properly understand the physiological appearance of negative academic emotions and cope with them to promote learning. We are interested in whether learners show an increase in knowledge despite the task causing negative academic emotions, or say it with Boekaerts’ approach if there is an increase in learning, a shift from the well-being to the growth pathway has happened.
Psychophysiological measurements for academic emotions
Psychophysiological measures (e.g., EDA, electromyography, eye-tracking, or electrical activity of heart and brain) are well-elaborated to index cognitive tasks and emotional states (see Berntson et al., 2017; Dawson et al., 2017; Levenson et al., 2017). Psychophysiological measurements aim to conclude from physiological reactions to psychological processes (e.g., emotions or attention; Pinel & Pauli, 2012). Here, the essential statement is that physiological processes are intertwined with human behavior (Cacioppo et al., 2017). Based on psychophysiological data, conclusions concerning emotional processes can be drawn. Psychological conditions cannot be associated with a separate isolated physiological reaction. The complex reaction pattern must always be considered (Cacioppo & Tassinary, 1990). For example, an electrodermal reaction can indicate an arousing situation or a deep breath. Both situations show the same result—an increase in the electrodermal curve—but they are very different in their respective meaning. Therefore, there is no one-to-one relation between a single physiological response (e.g., an increase in EDA or HR deceleration) and a specific emotion (e.g., frustration). For example, an increase in EDA cannot identify frustration, and frustration does not express solely in changing EDA. Adding HR as a measure for valence can specify the increase in EDA since negative emotions express in HR decrease (see sections “Electrodermal activity” and “Heart rate”). Therefore, the psychophysiological pattern composed of EDA and HR curves must be considered to identify emotional states. The attribution from physiological response patterns to actual psychological meaning requires an accurate experimental design, appropriate data analyses, and interpretation (Cacioppo et al., 2017).
Since we see emotions as a two-dimensional model, both, valence and activation must be examined to capture emotions comprehensively. Then, merging EDA and HR data reveals a physiological pattern, which can identify emotional states (e.g., Barrett & Russell, 1999; Eteläpelto et al., 2018; Larsen & Diener, 1992; Levenson et al., 2017). Furthermore, only the valence can declare if the emotion is positive or negative, which is crucial for successful learning. We chose EDA and HR since these are easily measurable, non-invasive, sensitive to psychological states, and well-elaborated (see sections “Electrodermal activity” and “Heart rate”). Based on established research about psychophysiological measurements, we used EDA to capture the activation and HR to measure the valence of academic emotions. We do not further address the third dimension “object focus” because it refers to whether the emotional state is seen as activity or outcome (see Table 1), which is not relevant for our purpose.
A standard psychophysiological measurement in many different research areas is EDA (e.g., attention, information processing, and emotion). Its popularity is the simple measurability and the sensitivity to many psychological states and processes (Dawson et al., 2017). EDA changes are associated with emotional activation, emotionally arousing thoughts or events, which induce an increase of electrical conductivity of the skin (Bradley, 2009). The EDA is solely controlled by the sympathetic nervous system (SNS) and, therefore, a direct reflection of activation (details see section “Heart rate”; Dawson et al., 2017; Lang et al., 2009). The interpretation of EDA changes depends on the stimulus material and the surroundings (Dawson et al., 2017). For example, an increase in EDA in an emotional surrounding can be interpreted as increased emotional activation. When somebody gets frightened, the increase in EDA can be traced back to the occurring attentional shift towards the unexpected stimulus (Bradley, 2009). Therefore, the more controlled a laboratory setting is, the more reliable is the interpretation of a change in EDA (Dawson et al., 2017). Moreover, having more than one measure (e.g., HR and self-reports) leads to a more accurate reconstruction of the learner's psychological state (Lang, 2014).
The most used method of recording EDA are skin conductance level and skin conductance response, both measured in microSiemens (μS). The tonic skin conductance level measures the conductivity of the skin in a particular situation and ranges from two to 20 μS. The phasic skin conductance response shows temporary fast changes in the conductivity of the skin caused by discrete events and ranges from one to five μS (Dawson et al., 2017).
Besides the primary function of pumping blood through the body, the heart also reveals information about emotion, attention, activation, and information processing (Berntson et al., 2017; Lang et al., 2009; Potter & Bolls, 2012). HR is, like EDA, easily measurable, non-invasive, and associated with many different psychological states. The HR shows the frequency of a cardiac cycle and is measured in beats per minute (bpm; Berntson et al., 2017). The most promising measurement is an inter-beat interval (IBI). Here, the time between two peaks of the cardiac cycle is tracked. The most prominent peak of the cardiac cycle is the R-spike. The time between two R-spikes is called RR-interval (Potter & Bolls, 2012).
Fluctuations in the HR can tell if a stimulus is pleasant or unpleasant, meaning HR is sensitive for measuring valence (Greenwald et al., 1989). Pictural stimuli (everyday objects or exciting scenes), which were assessed as pleasant (e.g., a beautiful landscape or erotic pictures), lead to HR acceleration, and pictural stimuli, assessed as unpleasant (e.g., dirty laundry or mutilated bodies), cause HR deceleration (Ijsselsteijn et al., 2000; Lang et al., 1993, 1997; Palomba et al., 1997). The valence of the pictural stimuli (pleasant or unpleasant) was evaluated and standardized by the International Affective Picture System, which can be used to explore emotion and attention (Lang et al., 1997).
Nevertheless, it is reasonable to assume that activating emotions lead to HR acceleration and deactivating emotions to HR deceleration. However, this relation does not necessarily persist based on the mechanics of the autonomic nervous system, which regulates HR and EDA. The link between activation and valence regarding the HR underlies the dual control of the heart. Its pace is regulated by both autonomic nervous branches, the parasympathetic nervous system (PNS), and the SNS (Berntson et al., 2017; Lang et al., 2009; Levenson et al., 2017). Both systems influence how fast the heart beats, depending on which system is activated. The activation of the PNS leads to HR deceleration, which is associated with attention and cognitive effort (Lang et al., 2009). The activation of the SNS results in HR acceleration, which is related to emotional activation (Lang, 1994). Therefore, HR can be a measure of valence but also activation. Nevertheless, since the PNS is faster and more dominant than the SNS, the activation of the SNS must be potent to overcome the parasympathetic activation (Shaffer & Ginsberg, 2017). A parameter to determine which system is activated is the heart rate variability (HRV), measured by spectral analyses (Berntson et al., 2017; Shaffer & Ginsberg, 2017).
Purpose of the study and research questions
When we consciously experience emotions like love, happiness, anxiety, or distress, we feel our physiological reactions (e.g., faster heartbeat or sweaty hands). However, unconscious emotional states, especially in the context of learning, equally impact our physiological behavior and are thus detectable in psychophysiological curves. Furthermore, psychophysiology allows visualizing emotional processes in real-time (see section “Psychophysiological measurements for academic emotions”).
Various studies have explored emotions in CBLEs and collaborative learning settings in a diverse manner (for a review, see Loderer et al., 2020). However, psychophysiological assessments of academic emotions in educational psychology are underutilized (Pekrun & Stephens, 2012). The present study wants to address this issue and get a unified and clear perspective on academic emotions, CBLEs, and self-reports. Moreover, we captured the valence and activation of academic emotions separately to give a detailed statement about the psychophysiological appearance of academic emotions. It was realized with a simple study design in a laboratory set-up (see Fig. 1) that eliminates potential external influencing factors (e.g., big-fish-little-pond effect; Preckel et al., 2008). The learning setting was designed to evoke negative emotions and guide the learner onto the well-being pathway. This process aims to be made physiologically detectable. Due to the lack of literature, the present work's research question and data analyses were primarily exploratory.
Since psychophysiological reactions unfold over time, they are an adequate measurement for academic emotions, which also occur over time. Self-reports give information about an emotional pre- and post-state of the learner – but they cannot provide details about the progression or reasons for the emergence of emotions. The exploratory research question (RQ) and hypotheses are structured top-down with the broad RQ at the top and the detailed hypotheses at the bottom. The derived RQ targets whether physiological behavior reveals more information about academic emotions and learning:
Can psychophysiological measurements provide deeper insights into learning processes? The explorative character of the RQ allows space for different data analyses and approaches. The term “deeper insights” implies getting information about the ongoing learning process (psychophysiological data) rather than solely having information about the current state of knowledge (self-reports). Moreover, the cause, emergence, and physiological progression of academic emotions provide insights into learning behavior. We formulated detailed hypotheses to follow the top-down approach, referring to negative academic emotions and their physiological indicators. The hypotheses target specific data analyses to find distinct physiological patterns and thus indicators of academic emotions. We state that patterns in EDA and HR indicate negative academic emotions. To meet the requirements of the two-dimensional model of emotions, we formulate a particular hypothesis for each dimension. Valence is captured by HR, and EDA captures activation. Learning requires attention and information processing, which activates the PNS. In the psychophysiological context, this implies that the HR decreases. Moreover, the designed learning environment (see section “Learning environment”) included unpleasant stimuli, leading to HR decrease (see section “Heart rate”). Therefore, we state:
Negative activating academic emotions cause HR deceleration over time (H1). Emotional activating situations cause an increase in EDA (see section “Electrodermal activity”). We want to show that this condition transfers to learning (i.e., academic emotions). The learning materials (see section “Learning environment”) induced negative activating academic emotions. Thus, we state:
Negative activating academic emotions cause increasing EDA over time (H2). To associate learning, HR, and EDA, we formulated the third hypothesis. Task difficulty, analyzed using learning performance, has an impact on academic emotions (see section “Academic emotions”), which can be measured by changes in EDA and HR:
Depending on the learning performance (high vs. low), overall HR and EDA differ (H3).
In conclusion, the Dual-Processing Self-Regulating Model (Boekaerts, 2011) shows that emotions have a crucial impact on learning (see chapter “Theoretical framework”). Since learners cannot always detect detrimental academic emotions, learning success can be affected negatively. We want to show an approach, which makes academic emotions measurable in real-time so that learners can be supported immediately. EDA and HR provide a fruitful measurement for emotions (see section “Psychophysiological measurements for academic emotions”). Based on the Three-Dimensional Taxonomy of Academic Achievement Emotions, we aim to measure both, valence and activation to distinguish between detrimental and beneficial academic emotions (the third dimension "object focus" has no further relevance for our approach). Anger and enjoyment, for example, are both activating but different in their valence. Only if both dimensions are measured, detrimental (e.g., anger) and beneficial (e.g., enjoyment) can be discriminated, and the learner can be supported accurately.