Holes in the Case for Mixed Emotions

The Competing Hypotheses

The debate is one about the structure of core affect, which Russell and Barrett (1999) defined as the most elementary consciously accessible feelings that people can experience. According to Russell and Barrett’s circumplex model, core affect can be described in terms of two psychologically irreducible dimensions. One of these dimensions is the bipolar valence dimension, which ranges from unpleasant states (e.g., sadness) through the neutral point (i.e., no valence) to pleasant states (e.g., happiness). If valence is irreducible (see Barrett & Bliss-Moreau, 2009), happiness and sadness should be mutually exclusive (Russell & Carroll, 1999). In contrast to this bipolarity hypothesis, the evaluative space model (Cacioppo & Berntson, 1994) contends that the positive and negative affective substrates underlying the bipolar valence dimension are separable, which gives rise to the bivariate hypothesis, which holds that happiness and sadness can co-occur.

Note that the bipolarity hypothesis does not state that all opposite-valence emotions are mutually exclusive. Watson and Tellegen (1985) highlighted the roles of arousal in emotional experience in a model that is similar to the circumplex model in many respects. They proposed that affective states vary in terms of two largely uncoupled dimensions later termed positive activation and negative activation (Watson, Wiese, Vaidya, & Tellegen, 1999). High-arousal positive states (e.g., excitement) fall near high-arousal negative states (e.g., tension) in the two-dimensional space and can therefore co-occur. Indeed, Watson and Stanton (2017) found that participants in their experience sampling studies reported higher than average levels of both positive and negative activation on some 3.3% of occasions.

The co-occurrence of polar opposite emotions like happiness and sadness remains more contentious. There has been abundant evidence for the bivariate hypothesis (Larsen & McGraw, 2014), but the debate has persisted largely because of questions about the validity of the measures that have been employed to measure core affect. Valid measures of a construct must demonstrate sensitivity to the construct of interest. Consider our initial evidence for mixed emotions (Larsen, McGraw, & Cacioppo, 2001). We simply handed moviegoers a survey as they walked into a theater to watch Life Is Beautiful or as they walked out of the theater. The two critical questions were: Do you feel happy? and Do you feel sad? These questions are presumably sensitive to happiness and sadness, so the finding that people were more likely to say yes to both questions after the film than before provided evidence that people can simultaneously feel both happy and sad. Of course, sensitivity is necessary but not sufficient: measures must also demonstrate specificity. Imagine a zoologist who is searching for a white raven but has trouble distinguishing white from light shades of gray. This zoologist’s eyes will be sensitive to the presence of a white raven, but will lack specificity. In the presence of a gray raven, the zoologist might see a white raven. Similarly, as detailed in what follows, there are threats to the specificity of all measures of core affect that have been used in the study of mixed emotions to date.

Evaluations or Emotions?

Russell (2003) distinguished between core affect and perceptions of affective quality. Whereas core affect refers to the experience of emotion, perceptions of affective quality refer to evaluations of stimuli as pleasant or unpleasant. As Russell (2017) noted, “Feeling bad is one thing, judging something to be bad another” (p. 111). By extension, people might perceive stimuli as having ambivalent affective qualities, but this does not necessitate mixed emotions because evocative stimuli do not always influence core affect. The problem is that our participants may not split hairs as finely as we do and it can be difficult to develop measures that are uniquely sensitive (i.e., specific) to core affect. We might ask questions about core affect (e.g., “Do you feel happy?”) but they might answer by reporting on their perceptions of affective quality (e.g., by telling us whether a film scene is pleasant). ¹

Testing Russell’s concern may require eliciting “mystery moods” (Leander, Moore, & Chartrand, 2009) which occur when people know they are feeling good or bad, but have no idea why. If people are not aware of the evocative stimulus, their perceptions of the stimulus’ affective quality cannot possibly influence their ratings of core affect. Subliminally priming people with positive words (e.g., music) or negative words (e.g., cancer) can elicit mystery moods (Chartrand, van Baaren, & Bargh, 2006). One way to address Russell’s (2003) concern will be to investigate whether subliminally priming people with positive and negative words leads them to experience mystery mixed moods.

The Vacillation Hypothesis

Even if people tell us about their core affect (as opposed to their perceptions of affective quality) when we ask them to, they might not tell us how they are feeling right now, at this very moment, because emotions can change rapidly (Ekman, 1992). Kahneman (1992) suggested that just as we can only perceive one interpretation of the Necker cube at any moment, perhaps we can only alternate between positive and negative emotions. If so, when people report mixed emotions, they might simply be reporting summaries of how they had felt over the course of the last few moments.

We have addressed this vacillation hypothesis by handing participants a computer mouse and having them watch scenes from Life Is Beautiful in the context of what they believed was a study of foreign language comprehension (Larsen & Green, 2013; Larsen & McGraw, 2011). Their task was to press the left button whenever they felt happy and the right button whenever they felt sad. The main finding is that people spend more time pressing both buttons at the same time during bittersweet clips than nonbittersweet clips, which we take as evidence against the vacillation hypothesis. In contrast, Barrett and Bliss-Moreau (2009) suggested that even button press data are susceptible to the vacillation hypothesis. The psychological moment may last for as little as 100 ms (Condon & Barrett, 2013). If core affect can change 10 times a second, we cannot expect people to press buttons fast enough to keep up with those changes.

In future research, we can test just how analogous the experience of mixed emotions is to the perception of the Necker cube. We will first ask participants to spend some time watching a Necker cube. During a subsequent bittersweet film, they will be instructed to press one button whenever they feel happy, a second whenever they feel sad, a third whenever they are experiencing happiness and sadness simultaneously, and a fourth whenever they are vacillating between happiness and sadness in the same way that they vacillated between the two interpretations of the Necker cube. The vacillation hypothesis predicts that people will spend more time reporting that they are vacillating during a bittersweet clip than during a control clip. More important, it predicts that they will spend equally negligible amounts of time reporting simultaneously mixed emotions during both the bittersweet and control clips. In contrast, the bivariate hypothesis predicts that people will spend more time reporting simultaneously mixed emotions during bittersweet clips. To address concerns about demand characteristics, we will tell one group of participants in a follow-up study that we do not expect them to report simultaneously mixed emotions (see Larsen & McGraw, 2011, Studies 2 & 5). The finding that these participants report more simultaneously mixed emotion during the bittersweet clip would provide particularly compelling evidence for the bivariate hypothesis.

Beyond Direct Measures of Emotional Experience

The debate over mixed emotions is a debate about the experience of emotion and self-reports are generally considered the gold standard for assessing emotional experience (i.e., conscious feelings). Even so, all self-report measures have limitations and the strengths of some of the measures that have yielded evidence for mixed emotions of happiness and sadness may not completely overcome the limitations of others. Consider how much more of the richness of pristine emotional experience can be gleaned from qualitative approaches like descriptive experience sampling (Heavey, Hurlburt, & Lefforge, 2012; Heavey, Lefforge, Lapping-Carr, & Hurlburt, 2017) than from Larsen and McGraw’s (2011) button press task. Though we must beware of venturing even further from pristine emotional experience by abandoning self-reports altogether (see Heavey et al., 2017), I believe that much can be learned by supplementing self-report measures with indirect measures of emotion, which refer broadly to measures in which participants do not directly answer questions about their emotional experience.

A variety of indirect measures can provide separate indices of positive and negative affect (e.g., the implicit positive and negative affect task, Quirin, Kazén, & Kuhl, 2009; functional magnetic resonance imaging, Grabenhorst, Rolls, Margot, da Silva, & Velazco, 2007), but it is unclear whether any of them have enough temporal resolution to shed light on whether people can simultaneously feel happy and sad. Facial expressions of emotion, which can last as little as one third of a second (Sayette, Cohn, Wertz, Perrott, & Parrott, 2001), are more promising. In fact, Ekman and Friesen (1975) suggested that mixed emotions may yield what Harris and Alvarado (2005) termed mixed smiles (i.e., genuine, Duchenne smiles accompanied by expressions of negative affect such as furrowed brows).

Harris and Alvarado (2005) investigated whether being tickled elicits mixed emotions. Just as participants reported that being tickled was both amusing and unpleasant, analysis of their facial expressions revealed that being tickled elicited just as many mixed smiles as pure Duchenne smiles. Nonetheless, Harris and Alvarado raised the possibility that subjects may have smiled not to express positive affect but to mask their negative affect. A closely related possibility is that tickle elicited miserable smiles (Ekman & Friesen, 1982), which people express to convey that they can endure some stressful experience. Supportive evidence for both of these possibilities comes from the finding that a cold pressor task elicited proportionally even more mixed smiles than did tickle. In another study, Griffin and Sayette (2008) showed heavy smokers a lit cigarette for 30 sec and found that nearly one quarter of them displayed mixed smiles for at least 300 msec. It is possible, however, that smokers’ mixed smiles were merely masking smiles or miserable smiles.

In sum, mixed smiles have been documented, but it is not clear whether they reflect mixed emotions. My lab is extending earlier work by having people report how happy and sad they feel during bittersweet film clips from Life Is Beautiful, which are presumably less likely to prompt socially desirable responding than tickles and cigarettes are. We further reduce social desirability concerns by leaving people alone while they watch the clips and only tell them that we videotaped them after the study. We are also asking naïve judges to watch each participant’s videos and provide moment-to-moment measures of the participants’ positive and negative affect. The question is whether judges will detect more mixed emotions on participants’ faces when participants were watching bittersweet scenes compared with scenes that are neutral, exclusively pleasant, and exclusively unpleasant.

Such findings would provide particularly compelling evidence that people can experience simultaneously mixed emotions, at least when they are being asked to report whether they are experiencing positive and negative emotions on a moment-to-moment basis. We are ultimately far more interested in how people experience emotions when psychologists are not asking them questions about those emotions. Collecting self-report measures may influence the very emotions that we are trying to measure by, for instance, calling people’s attention to the ambivalent aspects of the stimulus (Larsen & McGraw, 2011). We have addressed this reactive measurement hypothesis in prior studies by showing people evocative films without mentioning our interest in emotion beforehand. After the film, we asked participants open-ended questions about their emotions (e.g., “How do you feel right now?”) and found that people are more likely to report more mixed emotions after watching bittersweet clips than after watching control clips. If judges detect mixed emotions in the faces of participants who had been rating their emotions while watching bittersweet films, one follow-up question will be whether they detect mixed emotions in the faces of participants who had been unaware of our interest in emotion.

The Specter of Unfalsifiability

Scientific hypotheses are most useful if they are falsifiable, but psychological theories can be notoriously unfalsifiable (Meehl, 1978). In stark contrast, Russell and Carroll (1999) were daring enough to state that, “Bipolarity says that when you are happy, you are not sad and that when you are sad, you are not happy” (p. 25). Not only did they make a point prediction about the incidence of mixed emotions, they predicted that the incidence was precisely 0. As it turns out, falsifying this simple hypothesis has been anything but simple. To the extent that even the most sophisticated measures of happiness and sadness fail to provide valid indices of happiness and sadness, the bipolarity hypothesis may ultimately prove unfalsifiable. On the other hand, the bivariate hypothesis may also be unfalsifiable because the absence of evidence is not evidence of absence. Proving that no one can ever experience mixed emotions in any situation would require observing everyone in all situations, which is no more possible than confirming that all the world’s ravens are black.

Fortunately, the cumulative nature of the research program serves as a corrective. It can be difficult to make inferences about any particular study viewed in isolation, but a broader view allows stronger inferences. During a train ride with the experimental physicist Paul Dirac, theoretical physicist Wolfgang Pauli reputedly saw some sheep and noted that, “It looks like the sheep have been freshly shorn.” Dirac looked at the sheep and replied, “At least on this side” (Capri, 2007). Despite his sardonic reply, Dirac knew full well that the sheep were shorn on both sides. We have limited numbers of imprecise observations, but we have considerable ability to come up with reasonable inferences from the balance of those imprecise observations. I will not be so bold as to make a point prediction about how falsifiable the bipolarity and bivariate hypotheses are, but they are falsifiable enough to continue pushing on.

It is also worth noting that even unfalsifiable hypotheses can be useful. The checkered history of cognitive dissonance theory, whose falsifiability has long been questioned (e.g., Greenwald, 1975), serves as one case in point. Even if it is not falsifiable, cognitive dissonance theory has been among the most generative theories in the history of social psychology. The debate over mixed emotions has also been generative. For example, basic questions about mixed emotions have raised tangible questions about the nature of the seemingly ineffable experience of complex emotions such as nostalgia (Wildschut, Sedikides, Arndt, & Routledge, 2006). In addition, East Asians’ better ability to accept contradiction has spurred research on cultural differences in mixed emotions (Goetz, Spencer-Rodgers, & Peng, 2008). My colleagues and I have also proposed that mixed emotions may foster healthy coping (Larsen, Hemenover, Norris, & Cacioppo, 2003). In sum, debates over mixed emotions have been generative even if the competing hypotheses involved are unfalsifiable. Finish lines are useful not only because they tell us where to stop. They also tell us where to go.

Looking Ahead

Where exactly am I going? Several metaphors present themselves. Just as Diogenes of Synope reputedly carried a lamp around town in broad daylight shedding light where none was needed, one possibility is that I have answered a question that has already been answered. When I ask undergraduates to raise their hands if they think people can feel happy and sad at the same time, most raise their hands within moments. When I then mention that I have spent more than a decade trying to answer this same question, I get quizzical looks. Of course, science would still be in the dark ages if we relied upon laypeople’s intuitions. Moreover, Greenwald (2012) identified the disagreement about the structure of affect as being one of 13 long-standing unanswered theoretical debates in the fields of social and cognitive psychology and others have been able to generate alternative interpretations as readily as I have (Barrett & Bliss-Moreau, 2009; Russell, 2003). Answers to questions about mixed emotions are not as plain as day.

Another possibility is that I am like one of those allegorical medieval scholars who argued about how many angels can dance on the head of a pin. Perhaps I seek answers to an unanswerable question. I concur with most emotion theorists (e.g., Russell, 2003) that emotions are better seen as theoretical constructs rather than natural kinds. In this sense, we cannot discover the properties of emotional experiences in the same way that we can discover the properties of ravens. This might make it foolish to ask questions about whether people can experience any two particular emotions (e.g., happiness, sadness) at a single moment. On the other hand, such questions are only very slightly removed from questions about whether people can experience any one particular emotion (e.g., happiness) at a single moment. If questions about the experience of mixed emotions are unanswerable, so too are questions about the experience of single emotions. In the event, we must discard the vast portion of affective science aimed at understanding the experience of emotion, which would certainly be foolish.

In one of Zeno’s paradoxes, a traveler can only get halfway closer to his destination with each step. One step brings him halfway, the next three quarters of the way, and the third seven eights of the way, but he will never arrive. I may be on the traveler’s road. I do not anticipate ever gathering a final piece of evidence that will allow me to conclude that people can feel happy and sad at the same time. A survey of the alternative mentioned interpretations suggests that we would need to observe mixed smiles from participants who are in the middle of mixed mystery moods but do not know that we have any interest in their emotions. This is unlikely to occur because only intense emotional experiences produce characteristic facial expressions (e.g., Ekman, 1994) and mystery moods are unlikely to be intense. Electromyographic recordings of activity over the muscles involved in facial expressions of emotion can be more sensitive than videographic analyses (Cacioppo, Martzke, Petty, & Tassinary, 1988), so mixed mystery moods might result in mixed smiles that are invisible to the naked eye (but see Larsen, Norris, & Cacioppo, 2003). I would anticipate such effects to be so small that conducting such a study would require a prohibitive number of participants.

My fate could be worse than that of Zeno’s traveler. The field’s progress toward answering questions about mixed emotions is more similar to that of Zeno’s traveler than to that of Sisyphus, who was forced to roll a boulder to the top of a hill in Hades. Every time he neared the top, the boulder would roll back to the bottom, whereupon he had to start all over again. Indeed, stronger tests of mixed emotions have generally provided stronger evidence that happiness and sadness can co-occur, rather than quashing the evidence that had come before. It is important to point out, however, that my task is to provide stronger tests of mixed emotions, not stronger evidence for mixed emotions. Scientists should not fight wars with each other to determine which side had it right and which side had it wrong (Lench, Bench, & Flores, 2013). Rather, we should collaborate with each other so that we can all get it right. I began to appreciate the distinction when I approached James Russell at a conference in 2000 and told him that the poster that I was presenting provided evidence against the hypothesis that happiness and sadness are mutually exclusive. He was eager to hear more and invited us to submit the article for a special section he was editing. Russell wants to get it right and so do I. Regardless of whether the evidence ultimately indicates that people can experience mixed emotions of happiness and sadness, it will shed considerable light on the structure of affect. ²

In the meantime, I try to avoid the hubristic pride that I once mistook for confidence. Bob Dylan wrote, “I was so much older then, I’m younger than that now.” The evidence for mixed emotions is more compelling now than it was 2015 years ago, but I suspect that I am less convinced now than I was then. Fortunately, looking for this white raven has always been fun and never boring. All the more reason to get back to work.