Digital emotion measurement is incoherent, incomplete and potentially misleading

In the first installment of this series, we examined some of the most common current methods by which people seek to express and understand emotions in various digital media, and outlined some of the limitations of those methods.  Those limitations lead to potential confusion and frustration at the local, individual level, as we saw, but also at more macro levels, where scientists, employers, and brand development officers, and others  hope to derive insights into human nature and behavior.

Because of a lack of adequate, clear, and coherent means of conveying and understanding feelings, those interested in how individuals and groups of people feel and behave, lack adequate tools to capture and measure emotion. This is true particularly with respect to specific types of feelings and their intensities.

That isn’t to say that no one is trying to generate and utilize good data. For example, one can count numbers of “likes” and generate ratios of thumbs-up sentiments as contrasted to thumbs-downs.  Many well-established businesses do. But these are often just measures of engagement, buzz, or summary impressions. These metrics are not without value, but generally they hardly suffice to capture, much less provide for the measurement of human emotions.

Alternatively several businesses have developed ostensibly powerful means of scraping social media posts and other written statements to assess whether a given post or statement is positive or negative in valance. However, these efforts always depend on several a priori assumptions as to which words should be considered positive, which negative, which serve as modifiers of other words, and how closely-appearing those words must be in order to qualify as qualifiers. And that’s difficult enough for straightforward, non-idiomatic writing.

photo credit: idibon.com

Unfortunately there are more opportunities for errors in this process than most want to admit to their clients, and aside from some small-sample validation efforts, most of these business seem to neglect to collect any verification from the originators of posts as to what feelings they actually were trying to convey, much less the intensities thereof.

photo credit: 1stslice.com

Interestingly, in light of the growing popularity of emoji and stickers, some have made limited attempts at applying quantification to emoji in surveys, but these have thus far been extremely limited in scope.  They generally lack a grounding in emotion theory, and quantification for data generation purposes is typically rather simplistic with respect to specific emotions and their intensities.  Admittedly, creating order out of the creative organic use of these visual representations is a daunting task, particularly outside of traditional survey methodologies. For example, should we all agree that six poop emoji is three times worse in severity than two poop emoji?

 

 

And which of these conveys greater joy?

 

or

attachment-570cf182b6aa60e8ae969b3e

 

Even the popular Likert and Likert-type response format faces challenges of reliability and validity, owing to the fact that their anchors often include such terms as “extreme” and “severe” and “totally.” What is extreme for one person is not necessarily extreme to another, or even for that same person from one point in time to another. The pain you rated as a 10 on a 0-to-10 scale when you accidentally hammered your finger might merit a lesser value after your first round of natural childbirth.

This means, in effect, that though scientists and brand managers and other quants using these methods might be able to acquire and catalog large quantities of data, the veracity of those data are greatly in question, often to a greater extent than even those well-educated or otherwise well-informed investigators realize. In fairness, some of these folks do realize that their data are error-strewn, limited in utility, potentially misleading, and poorly representative of actual emotions. But what can they do?  These types of data are what they have available, and they need to feed their families. The people they report to aren’t necessarily schooled in these matters or inclined to look any more closely than they must in order to finish their own reports to their superiors, so they make do with large, if not necessarily useful, data piles and statistical break-outs.  This does not diminish the fact  that emotions are varied, both with respect to type and intensity.  Those emotions can be represented as data, and the method by which this is accomplished can matter greatly, because emotions are critical foundation material underlying motivation, decision-making, and behavior.  Enriched understandings of emotion types and intensities can matter in understanding and predicting what people value, and how they might behave.

photo credit: giphy.com

An Example

Is it sufficient to know what is the ratio of satisfied consumers to dissatisfied consumers for a given product or service, provided you can gather enough data to make meaningful predictions as to how the two groups will behave about your brand in the future?

No, it isn’t.

Because dissatisfied people aren’t homogeneous. Of course, they’re individuals, members of various gender groups, racial groups, age groups, etc. But more to the point, they differ as individuals in how they feel, and in how their feelings are likely to prompt them to behave, particularly when we take into account the type and intensity of emotions they feel.  More intensely felt emotions usually are more compelling, and can lead to different varieties of behavior within a given class of actions. Slight anger/frustration might lead to avoidance of a business or paltry tip; more intense frustration might compel a social media diatribe to all one’s friends and followers.

Disappointed people are at their core sad, and although they’re more likely to describe their subjective experience as “disappointed” rather than “sad” the fact remains that they simply want that sadness to dissipate and to be replaced with relief or joy. They don’t want much else. They’re relatively easy to win back — just remove the disappointment and its attendant core of sadness.

Frustrated people, on the other hand, have a core emotional experience of anger. They, too, want relief, but they also want something more — an apology, sometimes compensation, and possibly even a measure of vengeance, or at least what they perceive to be justice. And they’ll take that pound of compensation out of your brand’s hide in their communications with their friends or, even worse, on social media, as they ventilate and seek to humiliate you. They’ll take a feeling in place of other satisfactions, owing to what is frequently referred to as the frustration-aggression phenomenon.

And some dissatisfied people have an experience of contempt, which has at its core the emotion of disgust — a feeling so visceral, so powerful, and so durable, that it will take a truly remarkable effort on your part to turn it around to something more pleasant. Just think for a moment about how you can’t bring yourself to eat that certain food again after that bout of food poisoning. Disgust is a powerful opponent to affinity.

The Bottom Line

The type and intensity of a felt emotion can have important implications for behavior, and many appreciate this.  But they are hampered by most of the common extant methods for gathering and measuring emotion.  And so their insights are limited, often in ways they do not understand or, if they do, they do not yet see pathways to wisdom.  Part of the reason for this is the current unavailability of easily-implemented and consistent methods for people to express their feelings across various digital media and platforms.  In part 3 of this series, we’ll examine some of the challenges developers face when designing or attempting to incorporate methods for emotional expression and measurement.

Comments are closed.