From Group-Level Science to the Individual in Front of Us
Introduction
Something has been bugging me for decades.
I have spent most of my professional life trying to understand people scientifically. Along the way, behavioral science has given us an enormous amount of valuable knowledge about human behavior. We know a great deal about emotion, personality, culture, cognition, communication, nonverbal behavior, relationships, and many other aspects of human functioning.
But throughout my career, I have repeatedly encountered the same problem. What science tells us reliably about people in general is not necessarily sufficient for understanding the particular individual standing in front of us.
That distinction has become increasingly important to me, especially during the last couple of decades as more of my work has involved applying behavioral science to real people in real situations.
The more I have thought about the problem, the more I have come to believe that understanding people is fundamentally an inference problem.
We do not directly observe another person’s thoughts, intentions, motives, emotions, or beliefs. We do not directly observe trust, deception, confidence, fear, empathy, or intent. What we observe are behaviors and communications—words, facial movements, voice, actions, timing, decisions, interactions, and other forms of behavior.
From those observations, we evaluate what they might mean and construct provisional explanations from necessarily incomplete evidence.
And that is much harder than it sounds.
People—and Behavior—Are Extraordinarily Complex
Start with the individual. Human beings come into the world with what I think of as a universal psychological toolkit: capacities for emotion, personality, language, cognition, social interaction, and other psychological functions. But we do not all come into the world with exactly the same amounts or configurations of these capacities. Anyone who has spent time around infants knows that children differ from one another from very early in life.
Then development begins. From birth—and probably before birth—we interact with our environments. Those interactions produce experiences and histories that are unique to each individual. Our initial capacities interact with development, relationships, circumstances, opportunities, challenges, and countless other experiences throughout infancy, childhood, adolescence, adulthood, and beyond.
By the time we encounter another person, we are therefore encountering the product of an extraordinarily complex developmental history.
Behavior reflects that complexity. At any particular moment, a person’s behavior may be influenced simultaneously by biology, personality, development, goals, emotions, relationships, culture, organizational demands, the immediate situation, and personal history. No single factor adequately explains most human behavior.
Consider something I know a little about: the face. Facial behavior is produced by a complex neuromuscular system. The many muscles of the face can be activated in different combinations, at different intensities, with different degrees of symmetry, and across different temporal patterns. Those movements can signal emotion, but emotion signaling is only one thing the face does. Facial muscles are also involved in speech articulation, conversational regulation, cognitive signaling, physical activity, and many other functions.
So even within one behavioral channel—the face—we are dealing with an extraordinarily complex signal system. And then we try to figure out what it means.
A Problem That Bothered Me as a Researcher
I first encountered another side of this problem relatively early in my career. I went to graduate school in clinical psychology. Although I ultimately decided that becoming a practicing clinician was not the path I wanted to pursue, I remain extremely grateful for that training.
Clinical psychology taught me to think about the individual. A clinician is not treating an average. There is an individual sitting in front of you. Perhaps it is a couple or a family, but ultimately you are dealing with particular human beings.
At the same time, I was being trained as a researcher. And research generally operates differently. Most psychological research examines group-level effects. We compare groups, or we examine associations among variables. We collect observations from many individuals and use statistics to identify systematic patterns in those observations.
Much of my own early research involved cross-cultural comparisons. I might compare Americans and Japanese on emotional experience, display rules, judgments of emotion, or another psychological variable. We would collect data, conduct statistical analyses, find a significant difference, and write a paper explaining how the groups differed. There is nothing wrong with that. That is how science identifies reliable regularities.
But something about those analyses kept bothering me. Suppose we compare two groups and find a statistically significant difference between their averages. That difference may be highly reliable. But when we look at the distributions of the individual people within those groups, we usually see enormous variability and substantial overlap.
A commonly known statistic that tests differences between groups is Analysis of Variance, also known as ANOVA. ANOVA makes the issue I’m discussing here especially visible. It separates the variability in the data associated with differences between groups and the variability among individuals within those groups. As a young researcher, I was trained to focus on the significant group effect. That was the finding. That was what went into the title, the abstract, and the discussion section.
But I kept looking at everything else. And what I invariably found was that the differences among individuals within the groups were enormous (Figure 1).
We might conclude that Americans tend to be one way and Japanese tend to be another way, and that statement could accurately describe the average difference. But it certainly did not mean that every American was one way and every Japanese person was another. The distributions overlapped.
That distinction matters because a scientifically valid group difference can become almost stereotypic when we move too easily from an average group tendency to an assumption about a particular individual.
The same issue appears throughout behavioral science. Large-scale reviews of social-psychological research have shown that many scientifically reliable effects account for a relatively modest portion of the overall variability in human behavior. That does not make the effects unimportant. Far from it. Reliable scientific findings give us essential information.
But they leave a great deal unexplained. That realization led me to a principle that has stayed with me:
A reliable scientific effect can be real, replicable, useful, and important without being a complete explanation of the individual in front of us.
My clinical training kept reminding me of the person. My research training kept giving me findings about groups. The tension between those two levels of understanding never really went away.
No Person Lives in a Vacuum
There is another complication. No person is an island, and nobody lives in an abstract vacuum. Every individual exists within multiple nested contexts.
At any moment, a person occupies a particular physical setting. They may be interacting with another person with whom they have a particular relationship. That interaction may occur within an organization—a workplace, family, school, military unit, police department, or another institution. Those organizations exist within communities and societies, which themselves exist within larger social and historical contexts.
Culture operates throughout these nested systems. For that reason, I do not think culture should be treated as an add-on to understanding behavior. Culture shapes the contexts in which behavior occurs. It influences expectations, meanings, relationships, norms, and behavior itself. And importantly, culture also shapes the observer who is trying to interpret that behavior.
This means that the same observable behavior can have different meanings in different contexts. Take behavior out of context and we may change its meaning.
Understanding another person therefore requires more than observing what that person does. We also need to understand where the behavior is occurring, with whom, under what circumstances, within what relationship, and within what larger cultural and social systems.
We Don’t Observe What We Most Want to Know
This brings us back to the fundamental inference problem. What do we usually want to know when we say that we want to understand another person?
We want to know what the person is thinking. What they are feeling. Whether they trust us. Whether they are being deceptive. What they intend to do. What motivates them.
But none of those things is directly observable. What we actually observe are words, facial movements, voice, gestures, actions, timing, decisions, and interactions.
Some of those behaviors may provide meaningful information about underlying mental states. Others may not. Which creates another problem.
We Make Inferences from a Mess
A lot of human behavior is noise. Some behavior is meaningful signal. But separating meaningful signals from noise is difficult, and people often do not do it very well. Even worse, real human interaction does not conveniently present us with one behavior at a time.
We receive words, facial movements, vocal characteristics, gestures, posture, timing, actions, context, and interaction history simultaneously. Some may be meaningful. Some may not. Some may converge. Others may contradict one another.
It is a mess. That is my technical term for it. And yet we make inferences from this mess every day.
Sometimes we observe one behavior and immediately believe we know what it means. Someone looks away, crosses their arms, hesitates before answering, changes their voice, smiles, frowns, or makes some other movement—and we jump from that observation to a conclusion about what is occurring inside that person’s mind.
But behavior does not work that simply.
When a Good Theory Gets the Individual Wrong
Another experience early in my career reinforced this point for me. In the 1980s, I worked with a prominent scholar studying appraisal theories of emotion. Appraisal theories generally propose that when people encounter events, their minds evaluate those events in particular ways, and those evaluations contribute to the emotions they experience.
I agree with the general premise. Something has to occur psychologically that produces anger rather than fear, sadness rather than happiness, or one emotional response rather than another.
Researchers developed elegant models of these appraisal processes, and many of the predicted relationships were supported across studies and cultures.
At one point, one of the major researchers in this area developed a computer program that essentially attempted to reverse the process. The program would ask a person to think about an emotional experience without saying which emotion they had experienced. It would then ask questions corresponding to the appraisal dimensions identified in the research. Based on the person’s answers, the program would infer which emotion that person had experienced.
I watched people use it. And the program was often wrong. A person could answer the questions and have the program conclude, in effect, “You were experiencing this emotion,” only for the person to respond, “No. That wasn’t the emotion I was experiencing.”
That experience stayed with me.
Here was an elegant theoretical system supported by statistically significant, replicable, cross-cultural findings. Yet when we reversed the problem and asked the model to tell us what was happening inside one particular person, accuracy became much more difficult.
The lesson was essentially the same one I had encountered in my statistical work:
A theory can correctly describe an average pattern and still be wrong about the person sitting in front of you.
Not All Signals Are Equally Diagnostic
There is still another complication. Even after we successfully distinguish meaningful signals from noise, not all signals are equally informative. Some signals are weak and ambiguous. Others have stronger relationships with particular psychological states. Different observations therefore carry different evidentiary value.
That means professionals trying to understand another person need to distinguish weaker evidence from stronger evidence and recognize when additional information is required. This has become an important principle in my applied work:
A behavioral signal should be treated as an indicator or pointer—not as one-to-one proof of a mental state.
A signal may tell us that something is happening that deserves attention. It may suggest a hypothesis. It may tell us that we should ask another question, gather additional information, or look for evidence in another behavioral channel.
But an indicator is not proof. That distinction is crucial.
Multiple channels and sources of information may need to be integrated. Converging evidence can strengthen an interpretation. Diverging evidence can be equally important because it tells us that our current explanation may be incomplete or incorrect.
Understanding people therefore requires more than knowing cues. It requires knowing how to reason about evidence.
Every Inference Contains Uncertainty
Once we put all these pieces together, an important conclusion follows. Every inference about another person contains some degree of uncertainty.
People are complex. Behavior is multiply determined. Group-level findings do not completely determine individuals. People exist within multiple nested contexts. Mental states are not directly observable. Behavior contains signal and noise. And even meaningful signals vary in diagnostic value.
For any one behavior, multiple explanations may remain plausible. Good judgment therefore requires more than selecting the first explanation that comes to mind.
We should consider alternative hypotheses. We should test and retest our interpretations against converging, disconfirming, and newly emerging information. When possible, we should integrate other sources of evidence. We should calibrate our confidence to the strength of that evidence.
And perhaps most importantly, we should be willing to change our minds.
New information may strengthen our original interpretation. It may weaken it. It may require us to revise the interpretation substantially. Sometimes it may require us to discard it altogether. That is not a failure of understanding. That is how good understanding should work.
Inferential Humility
All of this has led me to a concept I have been thinking about recently: inferential humility. By inferential humility, I mean: the discipline of calibrating what we conclude about a particular individual to what the available evidence actually justifies.
Researchers need inferential humility. Perhaps we all need more inferential humility.
We should be careful about how broadly we generalize statistically reliable findings and about what those findings allow us to conclude about particular individuals.
But consumers and practitioners of behavioral science need inferential humility as well. When we observe a behavior, we should resist the temptation to believe immediately that we know what it means. Confidence in an interpretation should be proportionate to the quality and quantity of the evidence supporting it.
This does not mean refusing to make judgments. Professionals have to make judgments all the time. Interviewers, clinicians, investigators, managers, negotiators, sales professionals, physicians, educators, and countless others have to interpret people and sometimes act on those interpretations.
Inferential humility means making those judgments with appropriate discipline. It means remembering the difference between what I observed and what I think that observation means.
It means recognizing alternative explanations. It means knowing when I need more information. And it means being willing to update.
A Method, Not More Tricks
This brings me to the question that has increasingly occupied my thinking. Professionals can learn many useful techniques for understanding people. We can learn to read facial expressions. We can study voice. We can study body language. We can learn about personality, culture, interviewing, negotiation, deception, emotion, and many other aspects of human behavior.
All those areas can provide valuable knowledge. But perhaps the larger problem is not that we need more isolated techniques. Perhaps we need a disciplined way of thinking about how all of those pieces fit together. And that brings me to the question I increasingly believe we should be asking. Perhaps the most important question isn’t:
How good are we at reading people?
Perhaps the better question is:
Do we have a disciplined way of reasoning about people?
These questions have been weighing on my mind for much of my career—through clinical training, basic research, decades of cross-cultural work, and especially during the years in which I have tried to apply behavioral science to real individuals in real professional settings.
I don’t pretend that I have solved the problem. But I increasingly believe that understanding people better begins with recognizing just how difficult the task really is.
And, I admit, after finally getting all of that off my chest, I feel a little better.