Can we beat the mean? Bringing parsimony to the field of emotion dynamics.
In our latest publication featured in Nature Human Behaviour we put the added value of complex emotion dynamic measures into perspective, and illustrate that caution is warranted in developing novel and complex dynamic measures for the prediction or explanation of psychological well-being and emotion disorder.
(Poster picture made by Janne Adolf!)
Feelings come and go with the ebb and flow of everyday life. This is an indisputable truth, and a classic opening line to start a paper on the temporal dynamics of people’s emotional lives. Indeed, we hardly ever feel the same: one moment we may experience intense feelings of joy and excitement because we finally got that job promotion, at other times we feel overwhelmed by deep sadness because our dearest friend just passed away. Our emotions are thought to fluctuate adaptively according to situational demands, signalling relevant changes in our environment that pose a threat or opportunity to our well-being, and prepare us to cope with these changes in an appropriate manner. In a way, our emotional lives only have meaning because they change.
Research on emotion dynamics typically studies the patterns and regularities with which feelings and emotions change over time, such as how unstable, inert, or differentiated people’s feelings are, and how this plays a role in our well-being. However, in our latest publication featured in Nature Human Behaviour, we show that these patterns of emotional change (typically summarized in different emotion dynamic measures) are not that independently informative for people’s psychological well-being as we previously thought. That is, many measures are interrelated considerably, which illustrates that a single emotion dynamic measure often does not convey a lot of unique information about a person’s emotional life. Second, and more importantly, once average levels of positive and negative emotion are taken into account (generally considered the first and most parsimonious summaries of a person’s emotional life), more complex emotion dynamic measures show little incremental value in the prediction of psychological well-being. In this blogpost, I (Egon Dejonckheere) share my personal insights into the origin and process of this research, as well as its potential impact for the field of emotion dynamics.
The seed for this paper was planted somewhere in March 2018 and originally started as a side project Merijn Mestdagh and I would work on. Stimulated by a shared impression that many new emotion dynamic measures are introduced to the field in isolation, yet claim to be of crucial and unique importance to understand psychological well-being, we decided to put the added value of these measures to the test. A quick glance at the field of emotion dynamics taught us that a plethora of measures have been proposed. In the last decade, studies established significant associations between psychological (mal)adjustment and emotional variability, instability, inertia, emotion-network density, emotional switching, differentiation, determinism and recurrence, emodiversity, emotional fragmentation, flux, pulse and spin. Of note, Merijn and I (not to mention our supervisors Peter Kuppens and Francis Tuerlinckx) have both contributed to this enthusiastic accumulation of new measures ourselves. While Merijn had shared his relative variability index with the field to disentangle mean and variance in bounded affect ratings, I studied affective bipolarity in relation to depressive symptoms. In a way, we regarded this project also as a critical evaluation of our own work. What is the possible redundancy of these measures, and do they all truly add to the prediction of individual differences in psychological well-being?
Our plan of action was pretty straightforward:
- Combine all ESM and daily diary data on emotion that were at our disposal;
- Compute commonly studied emotion dynamic measures from these time series;
- Relate these measures to three prominent indicators of well-being;
At least, that was the scenario we had in mind. In reality, our initial results were rather surprising, not to say somewhat disappointing. Surprising to the extent we asked our co-author Isa Rutten to independently carry out parts of the analyses, to ensure no mistakes had messed up Merijn’s code. Disappointing because these findings seemed to jeopardize a substantial amount of work conducted in our research group (for example a recent meta-analysis carried out in our team by our colleague Marlies Houben), but also the work from other labs interested in the relation between emotion dynamics and psychological adjustment. We demonstrated that once we controlled for basic differences in average levels of positive and negative affect, many emotion dynamic measures no longer showed an independent relation with people’s psychological well-being. In other words, the more complex emotion dynamic measures in our study could not beat the mean.
From a scientific perspective, it was obvious we should and would share these findings with the broader community. From a human perspective, however, we felt torn. The last thing we wanted was to act as annoying kill-joys who are out to discredit an entire field of research, a field we even helped to establish ourselves, and in which we have a large network of collaborators! Although I admit this initial feeling was somewhat dramatic, and may overestimate the true impact of our scientific contribution, I still consider it much more rewarding and satisfying to contribute positively to a field, instead of criticizing its common practices.
So where do we go from here? Do our findings completely renounce the importance of emotion dynamics in psychological well-being? The answer is no. Our results merely place serious doubt on whether conventional emotion dynamic research is currently able to demonstrate meaningful and independent relations between affect dynamics and psychological well-being, and invite researchers to be more cautious when introducing new dynamic measures and spectacular conclusions. Similarly, in clinical applications, where clinicians increasingly rely on data harvested in daily life to inform patient treatment, our findings advocate for great caution in the use of perhaps intuitively appealing, but in fact not always such predictively valid summaries of these data.
In sum, I firmly believe that these null-findings bring about new opportunities, as many advances can be made to optimize the current modus operandi of our discipline. When researchers and clinicians adopt a parsimonious attitude, this will foster cumulative science and may counteract the fragmentation of a research field.