Over the last number of months as an Associate Editor at Emotion, as well as a reader of the literature, I have seen an increasing number of scholars refer to “discrete emotion models” and pit them against “dimensional emotion models” in order to set up a primary hypothesis and its alternative. This is a false dichotomy. My challenge as an editor is that I have not been able to find a singular “quick read” piece of writing in the literature articulating why a discrete versus dimensional contrast doesn’t make sense (although there are longer papers on this that do make this point such as this, this, this, and this – if you know of others that are near and dear to you, send them my way). I did a quick – entirely nonscientific – review of papers with the keyword “emotion” and “discrete or dimension or dimensional” published in 2017 and 2018 and there is some evidence that this is a problem in the published literature as well. My first take was “I should write a manuscript about this”, and I may, ultimately. My goal here, however, is not to call people out for this analytical problem per se but to provide an explanation as to why pitting discrete emotion models against dimensional emotion models is logically problematic, so that folks can fix it if it is in their writing (or thinking) or not make it in the first place. I don’t see any way around that goal if I were to undertake writing a manuscript which would require appropriate scholarly referencing. So, writing a blog post seemed like a good option.
There are lots of theories and models of emotion that focus on discrete emotions – how they come to be, their outputs, their functions, and so on. They vary widely in their hypotheses about all of those features of emotion. But there is no dimensional theory of emotion. Dimensions – typically valence and arousal – are characteristic of affect [i], not emotion. According to a number of the major theories in our field, affect is thought either to be the causal foundation of emotion or one of the most important components of emotion. Saying that affect is a causal foundation or critical component of emotion is very different, however, from saying that affect is emotion – and so, theories of emotions which recognize differences between affect and emotion are not reductionist about emotions. According to these theories, whatever else emotion is, it is something “more” than just affect. It is therefore a mistake to talk about theories of emotions as if they are dimensional. As a result, there is no way to pit “discrete emotion models” against “dimensional emotion models” because a) there is no dimensional emotion model, only dimensional affect models and b) emotion theories that hypothesize that multiple dimensions of affect underlie emotions in some way or another frequently hold that these emotions can be separated into distinct types and kinds, and therefore are, metaphysically speaking, discrete entities.
It goes without saying that there are many different theoretical and methodological approaches to the study of emotion. Over the last decade or so, two types of theories have dominated discussion and study (despite both being around for much longer) – what we call the Classic View of Emotion (CVE) [ii] and Theories of Constructed Emotion (TCEs). I use this terminology with very specific intentions, not just as shorthand labeling to differentiate my writing from that of other scholars.
The CVE says that emotions are precipitated by events and then produce a stereotypical set of outputs in the face, voice, autonomic/peripheral nervous system, behavior, and, or brain. There is a predictable mapping between the types of events that cause emotions and specific emotions. Further, each emotion is thought to have some sort of specific pattern of physiological and behavioral outputs. Because of this, to understand the emotion state of another individual, an observer (who could be the self) can “read” the outputs and, because the relationship between the outputs at the emotion causing them is mapped, infer the identity of the emotion causes the relevant outputs.
The interesting unit of analysis from the CVE can be at any step of the process – the event that causes the emotions, the process by which the effects of the event are translated into the emotion, the phenomenological experience of the emotion, the process by which the emotion generates to the outputs, the type of outputs generated by the emotions. CVEs stipulate that some special set of, or number of, emotions cannot be reduced to more fundamental or basic parts. This belief is reflected in the moniker of one type of CVE – “basic” emotion theory.
“Basic” emotion theories typically stipulate that a small set of emotions (e.g., 5 or 6) are unique kinds. This means that they are held by the theory to be fundamental or basic [iii], irreducible, and have a modular neurobiological architecture [iv]. These theories sometimes stipulate that more “complex” emotions (e.g., guilt) may be built via combinations of basic emotions, but even those more emotions must follow the event–>emotion–>output schema according to this theory (a view articulated here, for example). Basic emotion theories argue that relationships between events, emotions, and outputs is biologically hardwired and evolutionarily conserved; these assumptions have guided investigation into the neurobiological basis of emotion (e.g., in nonhuman animals, and in humans) and evolutionary emergence of emotion (for example).
Basic emotion theory is arguably the dominant theory that embodies a CVE, but it is not the only theory, hence my use of “CVE” and not “Basic Emotion Theory” (or something similar) to name this perspective. For example, most appraisal theories embrace a CVE insofar as they evaluate the specific pattern of cognition (called appraisals) that people make following events that lead to emotion – that is the process by which the event translates to emotions.
CVEs do say there are discrete emotions, whether they stipulate that a small set of emotions is biologically basic and hardwired or focus on the process by which an event precipitates an emotion or even the numbers of discrete emotions that exist. The unit of analysis is typically specific, discrete emotions and these emotions are thought to be the source of the lion’s share of variance in outputs. These are the theories, I believe, that scholars think they are referring to when they discuss “discrete emotion models” because of their emphasis on specific emotions (like happiness, sadness, fear, disgust, elation, and so on). Further, CVEs do not typically discuss or evaluate the importance of valence and arousal – the dimensions of affect.
Valence and arousal are the dimensions of affect, which, according to TCEs forms the basis of discrete emotions. TCEs stipulate that emotions are not hardwired modules. Instead, TCEs say that each emotion is a causal by-product of a complex interaction of more basic parts, and that these parts can vary dramatically depending on age, behavioral context, cultural context, conceptual knowledge, and any number of other important physiological and cognitive factors [v]. Nevertheless, TCEs still recognize that emotions are discrete things. TCEs differ to some degree in the ingredients that they think are critical for the emergence of discrete emotions – hence Theories of Constructed Emotion and not Theory of Constructed Emotion. But all TCEs hold that affect is a critical ingredient.
Affect is a state that is characterized by some degree of valence (hedonics, pleasantness to unpleasantness) and arousal (physiological activation) and can, but need not be, felt consciously. People can report on their affect (“I feel good” or “I feel bad”) or, people can report on their emotions and affective information can be extracted from those reports (e.g., one characteristic of happiness is a good or pleasant feeling). That is, affect can be organized according to dimensions (valence and arousal). That is the case because discrete emotions are organized in a systematic way with regards to valence and arousal. Happiness? Positive valence, just slightly activated/aroused. Depression? Negative valence, very deactivated. Fear, anger, disgust? Very negative, very activated/aroused, and so on.
TCEs say that affect is an essential component of emotion, but they do not say that affect is emotion – and so they do not reduce emotion to affect. According to TCEs, a discrete emotion has affective components, but the affective states that are components of emotions are not alone sufficient to determine the identity of the emotion of which they are part. For example, there are a slew of negative, high arousal emotions which share a negative, high arousal affective state but are not the same emotion – think anxiety, disgust, anger, and fear. The same affective state can thus be part of a number of discrete emotions depending on the context, what the person knows about emotion, past experiences, the language the person speaks, social norms, and so on. As a result, it does not follow to say that emotions, according to TCEs, are (or can be described by) dimensions. Scholars guided by TCEs, just like those guided by CVEs, study discrete emotions – but they may be equally likely to study the various other ingredients that go into cooking up emotions, including affect.
Why does the distinction between discrete emotions and the dimensions of affect matter?
It is important to be clear about whether we are testing hypotheses about emotions (which nearly everyone agrees are discrete entities) or testing hypotheses about affect (which everyone agrees can be analyzed as falling into at least two continuous but bounded dimensions), or testing hypotheses about some combination of the two–for example, when we ask whether (discrete) emotions or (degrees of the dimensions of) affect capture more of the variance in a given situation. Keeping this distinction in mind can help us develop strong, testable hypotheses about the nature of emotion.
For example, the goal of my lab is to understand some of the biological mechanisms that generate the emotions – and we do that work largely in nonhuman animals. While there is huge debate about the nature of nonhuman animal emotions, I have argued that affect is species nomothetic – at least in mammals who share a similar central-to-peripheral nervous system structure. If this is the case, and if a given behavioral or physiological phenomenon is driven by affect and not discrete emotions, then we can hypothesize that it has a homolog in nonhuman animals.
Another issue in the nonhuman animal literature that appears regularly is that scientists will make strong claims about animals’ abilities to perceive emotion stimuli when they are actually testing an affect hypothesis. For example, in a recent study, when goats were shown pictures of human faces generating behaviors typically associated with anger and those typically associated with happiness, goats spent more time investigating the “happy faces”. The take home message, amplified by the media, was that a) goats understand human emotions or “goats can read human emotions” and b) “goats prefer happy people”. That may very well be true, but given the experiment as conducted, there’s actually no way to determine whether goats prefer happy people or simply pleasant, neutral arousal people compared to negative, high arousal people. It’s entirely possible that when a happy face was compared to a serene face, goats might opt for serenity. If this was the case, one starts to build an argument that their choice has nothing to do with valence at all (let alone emotion), but rather arousal – lower arousal faces might be favored.
What CVE and TCE theorists, as well as scientists studying discrete emotions and the dimensions of affect, all have in common is the goal of understanding the mechanisms that generate and subserve emotions. My hope is that remaining clear about the difference between a hypothesis that is about discrete emotions versus a hypothesis that is about the dimensions of affect will speed those discoveries.
[i] Thus, there is a dimensional theory of affect. Valence and arousal aren’t the only dimensions that have been proposed to organize affect, but they are the dimensions that consistently appear in analyses of self-reports of emotion experience and judgements of emotion stimuli. Jim Russell’s 1980 paper is a classic on this . Here’s a resource of a broader discussion of affect. And here’s a recent paper which proposed additional dimensions of affect.
[iv] There are lots of ways that a phenomenon can be modular, for example if the same stimulus or event produces a single emotion, then that Stimulus-Response link is modular. When framed as a neuroscience argument, modularity has typically been interpreted as their being discrete neural and biological circuity for each emotions – such that emotions have, as Lisa Barrett calls them in here book, “fingerprints” in the brain (as well as voice, face, etc.).
[v] For a historical review of Theories of Constructed Emotion. For an edited volume on Theories of Constructed Emotion (note that many of the chapters appear to be accessible on the authors’ websites).