The Person Is the System

Psychology has two traditions for explaining what people do: social psychology and personality psychology. How do they differ, and why isn't either one enough?

Two sub-disciplines of psychology try to understand and predict human behavior in the world, and they come at it from different angles.

Social psychology has traditionally been interested in the effect of situations on behavior. Personality psychology has been interested in internal dispositions, the tendencies and the individually unique aspects of the psyche that lead to differences between people in how they behave. When you look at the empirical data, neither one on its own gives you a comprehensive framework, because some individuals behave in particular ways in particular situations, and that is really their disposition. Without understanding those intersections, we cannot predict human behavior.

What does that mean for anyone trying to model a person?

It has a direct consequence. Personality cannot be modeled in isolation from situations, and situations cannot be modeled apart from the person who construes them. What has to be discovered is the latent structure that connects the two, the stable if-then relationships between situations as a person experiences them and the responses those situations activate. A generative model of a person is a model of that system. It is not a list of traits, and it is not a record of past behavior.

A generative model of a person is a model of that system. It is not a list of traits, and it is not a record of past behavior.

Personality is supposed to be about consistency. Are people actually consistent?

Personality as a construct depends on an assumption. The qualities that distinguish one person from another should be consistent across time and situations. Yet both common intuition and empirical data show that although average levels of behavior are fairly stable over time, people's behavior changes quite a bit from one situation to the next (Mischel 1968). Even a generally caring and generous person passes up opportunities to help the homeless or is insensitive to someone else's plight. Recent studies find that within-person variability in behavior is as large as, if not larger than, between-person variability. So should we ignore that variability as random noise, or take it seriously as an additional source of consistency in personality? The cognitive-affective processing system, or CAPS (Mischel and Shoda 1995), is an effort to do the latter.

How do you show that variability is systematic rather than noise?

The key challenge for CAPS from the start was to show that there is stability in the way behavior changes from one situation to another. If the variability is not systematic, situation-behavior contingencies are noise. If it is systematic, they can be thought of as if-then signatures of personality. Testing this required repeated observations of the same people across time and contexts, so early evidence was sparse. With mobile phones and experience-sampling designs, a substantial amount of data has accumulated, and it shows that if-then profiles are relatively stable.

So if stability isn't in behavior itself, where is it?

CAPS offers a complementary view of where stability lies. Trait mean levels are stable when averaged across enough observations, and they predict important life outcomes, but they capture only part of the picture. CAPS also locates stability in the relationship between situations and behavior. If a person reliably does A in situation X and B in situation Y, that if-then pattern is itself stable and serves as a signature of who they are. Situations here are psychological: they are defined by how the person construes them, not by nominal features like being at school.

What links a situation to a person's response?

What links the ifs to the thens is a system of five kinds of cognitive and affective units. How a person reads the situation. What they expect and believe. What they feel. What they want and value. And how well they can regulate themselves. A feature of a situation activates a few of these units, activation spreads to the units connected to them, and a behavior comes out. Which units a person has, and how they are connected, comes from their own learning history. So for one person, seeing a situation as threatening leads to anger. For another, the same appraisal leads to withdrawal. The behaviors differ, but each person's link is stable. What is stable is the system itself, and that system is personality.

Couldn't you shortcut this by identifying the handful of situations that matter most, say the five biggest threats?

Explanations of personality can sit at different points along a continuum from broad and abstract to concrete and specific. One could identify a set number of situations that are normatively threatening, but what is gained in parsimony is lost in specificity. To really understand the psychological system that generates behavior coherent across time and situations, we need to get more concrete about what different situations mean to the person.

Dispositional approaches sit toward the broad, abstract end of this continuum. CAPS sits somewhere in the middle, because it seeks to identify the if-then signatures that characterize groups of people. In doing so, it combines an idiographic focus on within-person patterns with a nomothetic one. This difference shows up in how we measure each construct. A trait inventory asks how often a person generally feels or acts a certain way, and a single Big Five inventory is used for everyone.

A CAPS-informed measure instead asks in which situations a person feels a certain way, what it is about those situations that triggers the feeling, and how they then respond. Phobias offer a useful analogy. Knowing that someone is generally anxious tells us little about what will trigger intense fear in them. To understand a phobia, we need to identify the specific stimulus that produces excessive anxiety for a particular group of people. People with a snake phobia may not fear spiders, and vice versa. Each group shares a distinctive if-then pattern that a global anxiety score would miss.

Is this an argument against traits?

None of this is an argument against traits. Traits and CAPS simply answer different questions. Traits ask what a person generally tends to do. Because they are conceptualized at a broad level, they capture a lot of variance and are good at predicting broad trends in behavior and well-being. CAPS asks what a particular event means to the person, and how that meaning sets off a cascade of expectations, feelings, goals and regulation that produces a response. Traits describe general tendencies; CAPS explains behavior in a specific situation.

The construct you've worked with most is rejection sensitivity. What is it?

It is the tendency to expect rejection, to fear it, and to see it where other people do not. It is learned. If you were rejected in the past, you carry the lesson forward. I am not likable. I will probably be rejected. Now put that person in a new relationship. The person they are attracted to looks distracted. How does that look to them? To a rejection-sensitive person, it looks like intentional rejection. Someone who is not rejection sensitive says, I don't know, they're daydreaming. Same situation, different meaning. That difference in meaning is where personality lives.

How do you measure it?

Because the whole thing starts with that fearful expectation, that is what we measure, and we measure it in the situations where it comes up. Our questionnaire presents people with eighteen everyday situations in which rejection is possible: asking your parents for money, asking a classmate for their notes, asking someone you barely know on a date. In each one, we ask how worried you would be about being rejected and how likely you think rejection is. That is a very different question from “Are you anxious most of the time?” And because which situations carry the possibility of rejection depends on a person's social world, the measure has to be adapted to different groups. For members of stigmatized groups, some of the most potent rejection cues are tied to their group identity rather than to personal relationships. A Latinx driver, for example, may feel intense anxiety when pulled over on the highway, yet feel little anxiety when asking their parents for money. Versions of the questionnaire for these groups therefore sample situations in which a person might be rejected or mistreated because of their group membership.

Is it rejection specifically that sets people off, or negativity in general?

The trigger is specific. Rejection-sensitive people flinch harder, measured by the startle reflex, when they look at images of rejection and loneliness, but not when they look at images that are merely unpleasant (Downey et al. 2004). Rejection sets them off, not negativity in general.

And how do they respond once it's triggered?

The response is specific too, and it has two faces. Before any rejection happens, rejection-sensitive people work hard to prevent it. They hold back thoughts and feelings that might upset a partner. Rejection-sensitive girls and women report being more willing to do things they know are wrong to keep a partner happy. Then, when they feel rejected, the response flips. They pick fights. They lash out at the person who rejected them, in the lab and at home. Others turn it inward. Rejection-sensitive women became more depressed after a partner left them, but not after a breakup they chose themselves, and not after failing academically (Ayduk, Downey, and Kim 2001).

What does that look like across people's lives?

Zoom out to a whole population and you can see the consequences. Partners of rejection-sensitive women get angrier during conflict and think more often about leaving, and it is the women's own hostility that produces this. Their relationships are shorter. The fear of rejection helps bring about the rejection. In another domain, among African American college students, sensitivity to rejection because of race predicts weaker relationships with professors, lower attendance, and lower grades (Mendoza-Denton et al. 2002).

So are rejection-sensitive people simply more aggressive?

No, and this is the point that matters most for anyone building a model of people. On average, rejection-sensitive people are not more aggressive than anyone else. When they are not being rejected, they are if anything more accommodating. Aggressive if rejected, accommodating if not. Average their behavior and the pattern disappears.

Aggressive if rejected, accommodating if not. Average their behavior and the pattern disappears.

Is that pattern fixed?

The pattern can be changed by other parts of the system. Children who could wait longer for a treat at age four and who grew up rejection sensitive had, at twenty-five, self-esteem no different from people who were not rejection sensitive (Ayduk et al. 2000). Same disposition, better self-control, different life.

Does rejection sensitivity look the same for everyone?

Rejection can come for who you are, or for a group you belong to, and the triggers are different. For an African American student, the trigger might be leaving a store as the alarm goes off and a security guard walks over. For an Asian American student, it might be a salesperson mumbling an answer and having to ask them to repeat it. The same cascade runs in both cases. What differs is the situation that starts it, because that comes from the world each person learned in. So we measure the disposition in the situations that matter for that group. There is no questionnaire written for one individual. But the individual is built into the measurement through the situations, and the same logic goes all the way down to the person.

What does CAPS imply for people trying to simulate humans?

A representation built from trait scores and demographic labels does not contain enough information to identify a person's conditional response structure. It can tell you what people like this tend to do. It cannot tell you what this event means to this person, or what their system does with that meaning. Ask it how this person will respond to this message, this policy, this partner looking distracted, and it hands you a base rate, not an individual.

What should a model aim for instead?

CAPS suggests a different target. Predicting a person across situations requires discovering their stable if-then structure, the map of which psychologically meaningful situations activate which responses. Once that structure is known, behavior can be generalized to novel situations that trigger the same underlying processes. The goal is not to memorize how a person behaved in previous situations. It is to recover the system that generates those behaviors.

The goal is not to memorize how a person behaved in previous situations. It is to recover the system that generates those behaviors.

You've described this work as “CAPS AI.” Where does that name come from?

When the team at Heura Lab told me they wanted more sensitive modeling of emotion, what I heard was that they want to understand the underlying system that generates behavior. I called it CAPS AI. That captures my vision of CAPS, and of what they are trying to do, better than emotion alone.