Abstract
Sociological approaches to inquiry on emotion in educational settings are growing. Despite a long tradition of research and theory in disciplines such as psychology and sociology, the methods and approaches for naturalistic investigation of emotion are in a developmental phase in educational settings. In this article, recent empirical studies on emotion in educational contexts are canvassed. The discussion focuses on the use of multiple methods within research conducted in high school and university classrooms highlighting recent methodological progress. The methods discussed include facial expression analysis, verbal and nonverbal conduct, and self-report methods. Analyses drawn from different studies, informed by perspectives from microsociology, highlight the strengths and limitations of any one method. The power and limitations of multimethod approaches is discussed.
School and university classrooms provide an ideal natural milieu for investigating emotions during social interactions. Such research is necessary because studies based on psychological frameworks have begun to associate positive emotions with engagement with challenging projects and creative problem solving, and negative emotions have been associated with poor academic performance and school attrition (Pekrun & Linnenbrink-Garcia, 2014). Historically, research in psychology has relied on laboratory settings and experimental designs to investigate emotions and different approaches are now required that adopt “exploratory methodologies, qualitative data, nonexperimental designs, field studies, and idiographic strategies” to complement existing studies (Pekrun & Linnenbrink-Garcia, 2014, p. 664). Three broad questions require further consideration: (a) how can theoretical understandings of emotion in education be progressed?; (b) what approaches should be used for empirical investigation of emotion?; and (c) on which phenomena should emotion research interest be focused? (Pekrun & Linnenbrink-Garcia, 2014).
This article offers a partial response to the issues raised by Pekrun and Linnenbrink-Garcia (2014) by describing multimethod studies of emotion in educational contexts, informed by sociological frameworks of emotion. Practical examples of empirical studies by the author and colleagues (see Ritchie, Hudson, et al., 2014) are used to illustrate how research can be designed and conducted in natural settings. The studies reviewed and the analytical examples provided in this article relate to the second question posed by Pekrun and Linnenbrink-Garcia (2014). Advantages and limitations of specific methods and suggestions for further refinements are considered.
A Sociological Perspective on Emotion
Before presenting the review, it is pertinent to offer the sociological theoretical orientation to emotion that informs the studies reviewed in this article. Emotions are biological and cultural processes involving physiological states of arousal (Turner, 2007). They are coproduced by people engaged in interactions which can lead members of a group to become aware of one another’s emotions resulting in collective effervescence and synchronized conduct (Collins, 2004).
Multimethod Investigations of Emotion Within Sociological Frameworks
Conceptualization of Emotional Climate
Research on collective emotional climates of university classrooms (Bellocchi, Ritchie, Tobin, Sandhu, & Sandhu, 2013) and school classrooms (Tobin, Ritchie, Hudson, Oakley, & Mergard, 2013) has drawn on perspectives of individual | collective dialectics and event-oriented sociology (Tobin & Ritchie, 2012). In these studies, interactions between self and other involve the bidirectional elicitation of emotions resulting in a collective climate. As individuals engage in common social processes and structures, they experience the emotional climate (EC) of the group (Barbalet, 1995; Tobin et al., 2013). As well as involving intersubjectivity through experiences of the same transient emotions, EC relates to general states of emotional energy of longer duration. EC is valenced just as emotions are positively (satisfaction-happiness) and negatively (aversion-fear) valenced (Turner, 2007), and varies in intensity in the same way that emotions do (Bellocchi et al., 2013). Based on these conceptual aspects, EC is a meso-level phenomenon whereas emotions experienced by individuals are microlevel phenomena.
Exploring Emotional Climate: Meso-Level Data and Analyses
Classroom EC has been investigated using digital numbered keypads—or clickers—during classroom instruction (Tobin & Ritchie, 2012). EC intensity (i.e., very positive, positive, negative, very negative) and valence is assigned to a 5-point scale where 5 = very positive EC, 4 = positive EC, 3 = neutral EC, 2 = negative EC, and 1 = very negative EC. Participants input their perceptions of the class EC based on the 5-point scale by using the numbered clickers at 3-minute intervals. A USB device attached to a laptop receives incoming signals from the clickers. Line graphs are later constructed using the mean EC ratings against the 3-minute time intervals. Peaks and troughs in EC curves are used subsequently to identify events from video data for further analysis of microlevel nonverbal conduct. This is achieved by aligning time intervals on the graph with video time codes. An example of one analyzed graph is reproduced in Figure 1 (Bellocchi et al., 2013). The study took place in my preservice teacher education class as student groups presented debates about controversial issues (e.g., climate change, nuclear energy alternatives, and genetically modified foods).
Mean student EC ratings during Debate 1.
In that study, student perceptions of EC were related to various interactional factors. For example, when I addressed the class through long monologues (see Figure 1, educational implications discussion), low positive EC values were recorded. When debaters delivered engaging presentations, or when class discussions resembled friendly conversations (see Figure 1, voting discussions), the EC reached high positive values. In-the-moment emotional experiences of individuals can be identified through detailed microanalysis of video and audio data and related to collective EC. Subsequently, we have studied the collective emotional climate of teacher education classes in combination with the identification of emotions experienced in-the-moment by individual students (Bellocchi, Ritchie, et al., 2014). In the sections that follow, I describe some techniques used in analyzing individual emotional experiences.
Researching In-The-Moment Expressions of Emotion: Microlevel Analyses
Measures such as mean EC ratings conflate moment-by-moment changes in emotions experienced by individuals. Collective measures that aggregate individual data do not account for short-duration feelings produced during interactions (i.e., within the 3-minute intervals), which, if used alone, provide a narrow picture of emotional experiences in classrooms.
Microanalysis of Classroom Events
“In-the-moment” emotional expressions involve fleeting feelings that occur on short time scales of seconds or at most minutes. Studies that combine different methods including facial expression analysis (Ekman & Friesen, 2003; Ekman, Friesen, & Hager, 1978/2002), the study of proxemics and gestures (Harrigan, 2008), prosodic analysis (Juslin & Scherer, 2008), and physiological variables such as pulse rate (Tobin & Ritchie, 2012), are beginning to provide multifaceted understandings about emotion in interaction. For a more recent nonintrusive measurement tool used in emotion research, see the article in this issue on infrared thermography by Clay-Warner and Robinson (2014). The discussion that follows is organized under section headings based on the methods used for microanalysis of emotions. Inevitably, this has led to a fragmentation of the multiple methods used within each of the studies presented. For this reason, I return to discussing the same studies under each subheading.
Facial expression analysis
A growing number of studies are applying facial expression analysis for investigating emotions in education (e.g., Bellocchi et al., 2013; Ritchie, Tobin, et al., 2013). On-line training tools (Ekman, 2003; METT3.0/SETT3.0 COMBO, software available at https://www.paulekman.com) are available for identifying facial expressions of emotion that are informed by Ekman, Friesen, and Hager’s (1978/2002) Facial Action Coding System. These tools enable researchers to develop skills at identifying six emotions including fear, anger, disgust, surprise, happiness, and sadness. Emotions that are more complex can also be studied from the combination of facial expressions for the six basic emotions (Ekman, 2003; Ekman & Friesen, 2003). Researchers can quickly identify emotional events when immersed in the field and subsequently during video analysis with this training. It is important to note that these techniques have their limitations. Studies have shown that associations between facial expressions and emotion labels can vary according to cultural and conceptual contexts (see Gendron, Roberson, van der Vyver, & Barrett, 2014).
To illustrate the application of facial expression analysis in educational contexts, I will provide a brief example from research based on my own university teaching. Figure 2a contains an example of the facial expression for happiness. To capture my facial expressions, I wore a head-mounted minicamera that tracked my head movement to produce a direct image of my face at all times. This eliminated the challenge of interpreting expressions when my head was tilted as would have occurred if a stationary camera was placed on a stand or a hand-held camera were used by another researcher to capture my expressions.
a. Happiness facial expression.
At the time when the image in Figure 2a was captured, I was reminiscing with my preservice teachers about my experiences as a former school science teacher. I described a humorous situation in which an unwitting comment directed at the school Principal led me to having to teach a junior school mathematics class (mathematics was not my teaching area). The upraised cheeks, smile with teeth showing, creases at the corners of the eyes, and pronounced folds between the nose and upper lip show the movement of facial muscles related to experiences of genuine happiness (Ekman, 2003; Ekman, & Friesen, 2003). This expression shows that reflecting on my “unfortunate” situation was reconstructed as a pleasant experience in the context of the interactions.
In contrast, Figure 2b presents a combination of an upturned nose, squinting eyes, and raised upper lip that indicate disgust later in the same conversation. This expression is combined with a partial smile (i.e., lips pulled back and slightly up at the corners) suggesting a blended emotional experience of happily disgusted (Du, Tao, & Martinez, 2014). At this point during the interactions, I was discussing my decision to abandon a career pathway in research science to become a school teacher with a student who had made a similar career change. The happily disgusted expression indicates that although my recollection of research science was not favorable, it was not entirely negative (i.e., the partial smile). In the context in which I produced the expression, it served to establish a social bond with the student involved in the interaction who shared a similar experience. Establishing a point of similarity and revealing emotions through facial expressions, opens the possibility for students to share their own similar or related experiences, which can lead to emotional attunement between teachers and students.
Investigating emotion through gesture
Another method used to identify bodily expressions of emotion includes analysis of body movements, proxemics (interpersonal and environmental space), and gaze (Harrigan, 2008). The focus of this section is on identifying emotions from body movements. Researchers in psychology have used images of actors performing different gestures to study how research participants interpret gestures as emotional displays. For example, Tracy and Robins (2004) report on the “proto-typical pride display” that involves a slight smile, head tilted back, and arms akimbo or neutral. Bellocchi and Ritchie (in press) have also characterized gestures and facial expressions produced in a naturalistic setting as pride and triumph displays based on Turner’s (2007) theory of emotions. One triumph gesture involved the combination of the powerful fist-pump and a grimacing facial expression. The combination of the primary emotion assertiveness-anger (the grimace) with the primary emotion satisfaction-happiness (the fist-pump), when overcoming a challenge, is interpreted to be triumph (Turner, 2007). Variations of the gesture included facial expressions with a smile and head tilted, which indicated pride.
Vocalization of emotions
The analysis of emotional vocalizations has a long-standing research base (Scherer, 2003). Acoustic parameters of the voice that relate to specific emotions have been investigated predominantly with actor portrayals of emotion (Juslin & Scherer, 2008). Five acoustic parameters are considered most reliable when analyzing emotion through vocal modes: fundamental frequency (F0), F0 standard deviation (SD), voice intensity (M), speech rate (syllables per minute), and high-frequency energy (HF500). Freely available software packages, such as Praat (software available at http://www.fon.hum.uva.nl/praat/), can be used to characterize the acoustic parameters from different utterances and to compare different utterances (Juslin & Scherer, 2008). Audio data collected using voice recorders or sound files extracted from video data are analyzed using this software.
It is possible within a single study to compare utterances for identifying specific emotions through prosody in naturalistic settings (e.g., Bellocchi & Ritchie, in press; Ritchie, Tobin, et al., 2013). In one study, the author has identified irritation initially through a student’s emotional self-report (Bellocchi & Ritchie, in press). To support the data from the student’s self-reported emotion, a short passage of text was constructed, based on audio data of actual classroom interactions, which contained the same phrases and words used by the student during the irritated episode. The student read the passage using his normal voice (i.e., not irritated) to establish a control set of utterances for comparison with the emotive one. Acoustic parameters of the neutral reading (left-hand side, Figure 3) were then compared to the in-the-moment irritated vocalizations (right-hand side, Figure 3). In Figure 3, the acoustic spectrograph produced with Praat shows the intensity (bottom light-colored line, measured in decibels [dB] on left axis) and the pitch (top darker line, measured in Hertz [Hz] on right axis) of the student’s voice.
Prosodic analysis of student vocalizations. Neutral utterance (left) is compared to the same word when irritated (right).
There are marked increases in pitch and intensity in the irritated utterance when compared to the neutral utterance. Analyses that are more detailed are also possible. For example, in relation to the irritation analysis, acoustic parameters including values for F0 (378.95 Hz) and the formants (F1: 637.24 Hz and F2: 1813.87 Hz) that represent vowel sounds corresponded with values in the literature for cold anger (a variant of irritation; Scherer, 2003).
In the past, there has been a lack of consistency across some studies of acoustic parameters that correlate with specific emotions, despite strong evidence that voice cues can indicate the state of emotional arousal experienced by an individual (Juslin & Scherer, 2008). Notwithstanding limitations with the use of actor portrayals of emotional vocalizations, I have presented an example through the “irritation” analysis that can yield meaningful and comparable data within one study. A related approach used in a separate study (Ritchie, Tobin, et al., 2013) compared the emotive utterances of a teacher with similar words or phrases spoken in a neutral tone in the same classroom environment to characterize the acoustic parameters of the emotive expression. Replicating the approaches discussed here in other naturalistic contexts could help to identify other vocalizations associated with specific emotions. A limitation of this approach is that the background noise produced in natural settings impacts on the ability to collect and analyze data when compared to laboratory settings. However, the ecological validity of the approaches described here offers scope for complementing actor-portrayal studies.
Self-reports of emotions
A range of self-report techniques has solely been used to explore emotions in education (see Pekrun & Bühner, 2014, for a review) and in combination with other, direct observation techniques such as facial expression analysis (Bellocchi & Ritchie, in press; Ritchie, Hudson, et al., 2014). In one variation of the clicker studies, an emotion label was assigned to each of the numbered buttons on the clickers (Ritchie, Hudson, et al., 2014). More specifically, number 1 = enthusiastic, 2 = happy, 3 = attentive, 5 = neutral, 7 = disappointed, 8 = annoyed, and 9 = bored (emotion labels were not assigned to the other numbers on the keypad). Participants reported their emotions during classes by clicking a number at 5-minute intervals. Graphs were constructed to represent the proportion, as a percentage, that each emotion contributed to the total number of emotions reported for each time interval in a lesson for all participants (Ritchie, Hudson, et al., 2014). Although there were many instances when students reported emotions experienced in relation to the specific class activities, some of them explained during interviews that they had adopted a “default” response where they selected either neutral or attentive (Ritchie, Hudson, et al., 2014). These students admitted that they were bored at times but did not report this, choosing their default options instead. This outcome is supported by previous sociological accounts which have established that individuals engage in emotional management to represent emotions that are considered to be more appropriate in social encounters (e.g., Hochschild, 1979). Another challenge proposed by Ritchie, Hudson, et al. (2014) to the reliability of the clicker method was identified when one student explained that her feelings of disappointment during a lesson were related to experiences from a previous class. This was despite the instruction given to participants that their ratings were only to apply to the lesson being investigated (Ritchie, Hudson, et al., 2014). Although the student’s report of disappointment is valid in terms of how she felt during the lesson of focus, it overshadowed other possible emotions she experienced in direct relation to the lesson being studied. Due to these limitations for identifying individuals’ emotions, Ritchie, Hudson, et al. (2014) concluded the clickers were most useful in studies of EC.
An alternative approach could be to encourage students to report whatever emotions they are experiencing regardless of whether these relate directly to the specific class they are attending. Eliciting such reports may require higher levels of ethical clearance and may not be suitable for most naturalistic research in education.
Emotion diaries
The most recent approach used in my research on classroom emotions has involved the use of emotion diaries (Bellocchi & Ritchie, in press). In the earlier example of the irritation analysis, the student had first reported his emotions through a diary. In the emotion diaries, participants are required to indicate which emotions, from a list of 10 (four positive emotions: happiness, pride, enthusiasm, wonder; and six negative emotions: frustration, anger, disgust, anxiety, sadness, embarrassment), they experienced during a lesson (see Ritchie, Hudson, et al., 2014). An A4 page listing the emotions in one column of a table is provided to students. In an adjacent column, students annotate why they experienced each reported emotion during a lesson. Each diary typically yielded 3–5 reports of emotional experiences per student per lesson (Bellocchi & Ritchie, in press). Combining the use of diaries with direct observations in that study yielded examples where feelings of pride and triumph were related to increases in students’ social status, understanding of science concepts, and receiving positive sanctions from the teacher or a peer for achieving success in classroom tasks (Bellocchi & Ritchie, in press).
One challenge with using the emotion diary method is that students can forget certain emotional experiences by the end of a lesson when diaries are issued. An alternative approach would be to ask students to report their emotions at two junctures such as the middle of the lesson and then again at the end. Keeping researcher field notes along with video recordings of interactions also reduces data loss when compared to using a single measure such as diaries. The diary can also be used for conducting stimulated-recall interviews, making it a versatile research tool for multimethod research designs.
Future Directions
The main strength of the growing corpus of educational research reviewed in this article is that none of these recent naturalistic studies relied on single methods for investigating emotions. When data from multiple methods converged, they produced effective ways of exploring emotional experiences during interactions. Contradictions in the different data sources can provide even more interesting issues to explore. For example, contradictions between self-reports and video/audio data could provide one way of identifying the emotional work that is done by individuals to disguise their emotions during interactions. Conflicting data could lead to different understandings about the emotions produced during interactions and what role they play in establishing, maintaining, or disrupting social bonds.
Multimethod research designs may challenge existing research traditions and require shifts in thinking to overcome traditional boundaries between, for example quantitative/qualitative research, nomothetic/idiographic approaches, or exploratory/experimental designs (Pekrun & Linnenbrink-Garcia, 2014). Whereas self-reports can access the perceptions of larger groups of participants within a study, allowing for statistically generalizable findings, naturalistic observations offer in-depth understandings of emotions produced during interactions. However, observational methods rely on participants externalizing their emotions through verbal and nonverbal conduct. In natural settings, it will not always be possible to match every participant’s self-reported experiences with direct observations by the researcher. For example, Bellocchi and Ritchie (in press) reported the perspectives of a whole class (29 students) through interviews and emotion diaries, but then presented microanalyses that involved only those individuals (four students) who displayed their emotions through gestures during classroom interactions. The focus on four students could be taken as a limitation of that study due to the inability to generalize the displays to the remainder of the class. Researchers with concerns of this kind can overcome this, as Bellocchi and Ritchie (in press) did, by interviewing all students in the class to establish how they feel during similar classroom events as those experienced by the four students who displayed their emotions. In this way, although the emotional displays are not statistically generalizable, a broader understanding of how other students feel in the same circumstances is accessible to researchers. Through the practical examples of research designs described in this article, others seeking to investigate emotions in naturalistic educational settings may be better prepared to deal with the challenges faced by researchers in this field.
Future studies could focus on the contribution that different combinations of methods make to our understandings of emotion. It would be of particular interest to explore the way in which participant self-reports and researcher observations (i.e., facial expression analysis, prosodic analysis) compare so that constructs such as emotional management could be explored in educational settings.
Footnotes
Author note:
The Australian Research Council Discovery and Linkage Grants, contract grant numbers: DP120100369, LP110200368, supported this research. Any opinions, findings, and conclusions or recommendations expressed in this article are those of the authors and do not necessarily reflect the views of the Australian Research Council.