How people move toward each other and away again is not a matter of character. It follows a small number of patterns that were first measured in children and later found in adults. This chapter draws the four patterns, what happens when each meets each, the loop two of them fall into, and the popular model of the narcissist and the empath, which is marked as a model because that is what it is.
John Bowlby's idea was simple and, in the 1950s, unwelcome: a child's need for closeness is not a by-product of being fed, it is a system in its own right, with the same standing as hunger. Mary Ainsworth then made it measurable. In her strange situation a small child is separated from its mother for a few minutes and then reunited, and what turned out to matter was not the crying during the separation but what the child did at the reunion.
Three patterns appeared, and later a fourth. Some children sought contact, were soothed and went back to playing. Some protested loudly and then could not be settled, pushing the mother away and clinging at once. Some behaved as though nothing had happened and turned back to the toys, while their heart rate said otherwise. A fourth group showed no strategy at all, freezing, or approaching backwards.
In 1987 Cindy Hazan and Phillip Shaver asked whether adults in love do the same thing, and the answer was that in broad terms they do. Bartholomew and Horowitz then sorted the adult version into the two axes used here: how much fear there is of being left, and how much distance is kept to prevent it from mattering. Everything on this page rests on those two lines.
Click a field. The axes are continuous, the boxes are a reading aid.
Almost nobody sits in a corner. The two axes are measured on scales, and most people land somewhere in the middle with a leaning. Reading yourself into one box and staying there is the most common mistake made with this model.
Two studies, two methods, two answers. The spread is the point.
Hazan and Shaver ran their survey as a quiz in a newspaper, so the sample chose itself. The National Comorbidity Survey used a representative sample of the United States. Both used three categories rather than four, so what is called avoidant here contains both avoidant groups of the four-field model. Numbers vary by instrument, and any table that gives one figure to the decimal place is overselling.
Pick a pairing. On a phone, choose your own pattern first and then the other one.
Friction is shaded from light to dark. It is not a forecast: two anxious people can be happy for forty years and two secure people can run out of reasons to stay. What the shading shows is how much work the pairing needs before it settles.
Six steps, and each one is a reasonable answer to the step before it.
Nobody in this loop is behaving unreasonably from the inside. Pursuit is an attempt at repair and withdrawal is an attempt at regulation. What makes it a trap is that each of them solves their own problem by creating the other's.
Six rounds of the same conversation. Nothing new is added; only the threshold falls.
Attachment pattern, trait, diagnosis, label. Most arguments about these words are arguments about the level.
How a person regulates closeness and distance. Two axes, four fields, everything in between occupied. Changes slowly across years. This is the level of the first three tabs.
What someone is like in general, outside relationships too. Five broad dimensions cover most of it: openness, conscientiousness, extraversion, agreeableness and emotional instability. Narcissism as a trait belongs here, distributed like height, with no line between has and has not.
Descriptions with criteria, made by a clinician who has met the person. Only at this level does the sentence “he has X” mean anything, and at this level X is rare. Narcissistic personality disorder sits here.
Empath, energy vampire, toxic, sociopath in everyday use. Widespread, often well meant, but with no definition, no boundary and no way to check. Useful for sorting your own experience, useless as a statement about another person.
What each one is, and what it usually gets confused with.
Strategic use of other people, without the grandiosity of narcissism. Those high on it plan long term, withhold information, and treat morality as an instrument. Together with narcissism and psychopathy it forms the dark triad, whose shared core is low agreeableness.
Telling it apart: Confused with narcissism. The difference: one wants to be admired, the other wants to win and is indifferent to the applause.
Impulsivity, low fear, little remorse, high appetite for risk. Distributed across the population as a trait and not the same thing as the diagnosis.
Telling it apart: Confused with coldness. Avoidant attachment looks similar but runs on a completely different engine, protection rather than indifference.
Strong fear of abandonment, rapid swings in how others are valued, impulsivity, often self-harm. Described as borderline personality disorder.
Telling it apart: Very often confused with vulnerable narcissism, because both look wounded and intense. The dividing line is what everything turns on: being left, or the image of oneself.
Decisions are handed to others, submission out of fear of rejection, great fear of being alone. Described as dependent personality disorder.
Telling it apart: This is the clinical counterpart to part of what is popularly called the empath, and therefore the most important entry in this list.
Closeness is wanted and avoided out of fear of rejection. Described as avoidant personality disorder.
Telling it apart: Confused with the avoidant attachment pattern. There closeness is unwanted; here it is wanted and not reached.
Comes from addiction work and describes caring for someone at your own expense while carrying their behaviour along. Not a diagnostic term and inconsistently measured.
Telling it apart: The closest neighbour of the empath, and the more honest word, because it names your own part instead of ennobling it.
A temperament characteristic with a published scale: deeper processing, faster overstimulation, stronger response to subtle things. One study found roughly a third low, two fifths medium and just under a third high; the older estimate was about a fifth.
Telling it apart: Equated with empathy. It is a different thing: sensitivity to stimulation, not orientation toward other people.
Narcissism as a trait. Distributed across the population like any other trait, with scales that separate an agentic form (dominance, self-promotion, boldness) from an antagonistic one (exploitation, devaluing others) and a neurotic or vulnerable one (hypersensitivity, defensiveness, a fragile self-image). The popular picture usually mixes the three.
Narcissistic personality disorder. A diagnosis with criteria, made by a clinician who has met the person. Grandiose and vulnerable presentations are both recognised.
Empathy. Measured as several separate things, not one: perspective taking, empathic concern, and personal distress. Someone high in concern and high in personal distress reads other people fast and has trouble telling the other's state from their own. That is the closest measurable thing to what the popular literature calls an empath.
Sensitivity. Sensory processing sensitivity is a temperament trait with a published scale, and it correlates with the above without being the same thing.
Two fixed types who find each other. One who takes and one who gives, drawn together as if by physics. Neither half of that pair is a category in any diagnostic manual, and no validated instrument sorts people into them.
A single villain. The model puts all the agency on one side. That is what makes it satisfying and also what makes it unreliable: it is used most often after the fact, by the person who was hurt, about someone who is not present to be assessed.
A fixed sequence of phases. Many people recognise the sequence, which is why it spread. A review in 2025 found that it has almost no presence in the academic literature and argued it should be defined and tested rather than assumed.
Each dot is one adult. The filled dots are the highest published lifetime figure for the diagnosed disorder.
The largest epidemiological survey to look for it, the second wave of the American NESARC study with 34,653 face-to-face interviews, found a lifetime prevalence of 6.2 per cent, 7.7 for men and 4.8 for women. DSM-5 gives a range of 0 to 6.2 per cent across community samples. That is the ceiling, and it is a ceiling for a diagnosis, not for the trait. Ignore that base rate and you will overestimate: most partners who were self-centred, hurtful or exhausting were not disordered, and the label explains less than it appears to.
Patterns over time, not single moments. Everybody does some of these sometimes.
Called agentic and antagonistic in the research. This is the picture most people mean.
Wherever it starts, after two moves it is about him again. Not as an interruption but as a friendly detour. Anyone keeping count notices that their own story rarely gets finished.
Praise does not look like pleasure, it looks like refuelling. When it fails to arrive the mood turns faster than the occasion warrants.
Waiting, agreements and promises become negotiable the moment they are in the way, and the justification arrives instantly and even sounds reasonable.
Others land strikingly fast in useful or unimportant, and the sorting changes with their usefulness, not with the person.
“That hurt me” becomes “you are too sensitive” or “you started it” within one sentence. The subject changes owner.
In every account of the past he is the unrecognised, the overlooked or the superior one. All previous partners were difficult.
Called neurotic or vulnerable in the research. From outside it does not look like grandeur but like woundedness, which is why it is often missed.
Instead of grandiosity: silence, withdrawal, a sense of being constantly passed over. The effect on the other person is the same, except that it asks for redress rather than admiration.
What others have comes up often, usually as a question of fairness. His own part in it rarely does.
First it was the rescue, then it was too little, too late or not sincerely meant.
Apologising buys no resolution, only a new invoice. An argument does not end, it gets deferred.
Warmth triggers suspicion first. The question is not whether it feels good but what it will cost.
Not a diagnostic term. What can be measured is high empathic concern, high personal distress, and usually an anxious attachment pattern alongside them.
You know how the other person is before you know how you are. That is a real skill and at the same time the point at which your own perception drops out.
You agree before consulting your own calendar. The no comes later and then costs explanations.
For moods, for conflicts, including ones you had no part in. The quiet assumption is that when something is uncomfortable it must be down to you.
Someone else's suffering produces not only compassion but your own arousal, which is hard to switch off. This is exactly where research separates two things that everyday language calls empathy.
After company you are empty, without anything bad having happened.
A no does not feel like self-protection, it feels like betrayal. Which is why it is so rarely said.
Sorting the other person out from a distance is unreliable, even when you know them well. Your own experience is the better source because you can observe it directly. Four signs that, in this order, point fairly reliably to an unhealthy dynamic regardless of which label eventually fits:
First. You sort through conversations afterwards. You run them again trying to work out what just happened. In healthy relationships a conversation is over when it is over.
Second. You tell friends a milder version. Not out of loyalty, but because the real one would sound too clear once said out loud.
Third. Your own perception becomes negotiable. You start sentences with “maybe I am imagining it”, including about things you heard.
Fourth. You get smaller in your own account. What you are good at comes up less, what you get wrong comes up more. Noticing that in yourself is a finding, and it does not depend on whether the other person would qualify for a diagnosis.
Confidence, success, a polished appearance, or liking to be photographed. Those are traits, not patterns.
That someone hurt you. People hurt each other out of fear, carelessness, overload and cowardice far more often than out of a personality disorder.
That someone says little, shows little, or finds closeness hard to bear. That describes an avoidant attachment pattern, which is a different thing.
A list on the internet that fits a former partner. Such lists are written so that they fit almost anyone; it is the same mechanism the birth chart chapter demonstrates with the Forer effect.
And the other way round: recognising yourself here as the empath proves no virtue. Having no boundaries is not kindness, and the role of the counterpart is comfortable because it makes your own share in the dynamic invisible.
The assumption that narcissists cannot read other people does not survive testing. In three experiments (N = 282, 95 and 88) the negative link between narcissism and empathy disappeared once participants were explicitly asked to take the other person's perspective, and not only on the questionnaire but in heart rate as well. The authors read that as a problem of motivation rather than of capacity.
A more recent study (N = 87) separates it further. On a questionnaire, people high in narcissism rated their own perspective taking as no worse than anyone else's; on a task where the accuracy was actually checked, they did worse. On feeling with someone the shortfall showed up across every kind of measure.
Read together that is a less comfortable picture than the popular one: the ability is there and is overestimated, the feeling with is lower, and both can be shifted in the short term. For everyday purposes one thing follows: “he can do it when he wants to” does not mean he will keep doing it.
Seven meetings of the same small group. One line is the group's rating, the other is his own.
Drawn to show the shape of two findings, not measured curves. In a classic study, people high in self-enhancement were rated positively by a group of five strangers after the first meeting and negatively after the seventh. A later field study of 126 first-year students separated the two engines: the agentic side, the self-presentation, builds popularity and keeps building it, while the antagonistic side, the devaluing of others, drives the long-term decline. Both run in the same person, which is why the first impression is so good and the fourth month so bad.
The pattern has one property that separates it from ordinary inconsiderateness: it consumes exactly what it lives on. Confirmation only works when it comes from someone whose judgement counts. Whoever counts gets close enough to see the cracks. Whoever sees the cracks becomes a threat and has to be made smaller. And the devaluation removes precisely the person whose confirmation would have worked.
That is not a moral verdict but the description of a circuit that looks inevitable from inside. From outside you can see it using up its own supply. From inside you only see that people have once again turned out to be disappointing.
Distance, then, is not a by-product but the protection. Let nobody close enough and you never have to devalue anybody. The price is that the confirmation has to come from further and further away and wears off faster each time.
That the movement continues has been measured, and the reason is more precise than the expectation. A study of some five hundred people in relationships found a negative association between narcissism and commitment. But it was not down to lower satisfaction and not to less investment in the relationship. It was down to perceived alternatives: people high in narcissism see more and better options elsewhere and pay more attention to them.
That explains an observation that otherwise stays puzzling. The movement does not start when things get bad, it starts when something else becomes available. And it explains why effort on the other side achieves so little: it improves the existing relationship without changing a calculation in which the existing relationship is set against every possible other one.
What looks like weakness from outside is usually the effect of the alternation between warm and cold. Irregular reward produces the most persistent behaviour that learning psychology knows of, and it does so regardless of education, intelligence or confidence. Someone who never knows when the warmth will return does not stop trying.
A second effect weighs more. When your own perception is repeatedly contested, it becomes unusable as a tool. You stop relying on your own judgement, and that is precisely the state in which people no longer leave. Which is why the question “why did you stay” is put wrongly: it assumes the judgement whose damage is the actual problem.
Part of the loop does come from the other side, and that belongs here without assigning blame. An anxious attachment pattern supplies exactly what the arrangement needs: a great deal of attention, little boundary, and a readiness to put the other person's state before one's own. The tab Who meets whom draws the same pairing with no diagnosis at all.
Three parts of the usual story do not survive testing, and those matter more than the parts that do.
The parasite image. Draining energy, vampire, host: these are pictures, not mechanisms. What gets measured is a regulation loop that needs confirmation from outside. The picture turns that into a kind of creature, and therefore into something that cannot change. Which is exactly why it is so popular: it ends the question.
That he does not understand other people exist. The ability to take the other's perspective is there, as the previous section shows. What is lower is the feeling with and the priority. That is less comfortable and more accurate, and it has a practical consequence: appealing to his understanding achieves nothing, because understanding is not what is missing.
The lonely old age. It is the ending that makes the story just, and it is not supported. A study of 200 middle-aged and older people found less narcissism and more loneliness among the older group, with narcissism acting as a buffer rather than an accelerant. For the vulnerable form it runs the other way, where more narcissism goes with more loneliness. Both findings rest on small samples. What is well documented is the course within one relationship history, not the balance of a whole life.
The uncomfortable consequence: expecting it to even out in the end is a story about justice, not a finding. Anyone staying because they are waiting for that settlement is waiting for something the numbers do not promise.
The reading. The pairing is described so consistently because the two strategies fit each other like a lock and a key. One person regulates their self-image through the attention of others and therefore needs someone who supplies attention and does not walk away. The other reads the state of the room before their own and treats another person's need as the more urgent fact. Each gets exactly what their strategy requires, which is why the arrangement is stable long after it stopped being good.
What speaks against it. The same course can be described without any diagnosis. An anxious pattern meeting a dismissing-avoidant one produces the same idealisation, the same conditional warmth, the same intermittent reward and the same abrupt end, and that version is supported by decades of measurement. Adding a disorder to the description explains nothing extra; it supplies a villain. It also travels badly: once someone has been named a narcissist, everything they do afterwards confirms it, which is the definition of a claim that cannot be tested.
What both readings agree on. The mechanism that holds the second person in place is not love and not weakness. It is irregular reward, and that is one of the most reliably documented effects in psychology.
A large part of the vocabulary that turns up in these conversations does not come from research but from a single source: the self-published writing of an anonymous author who says he is a diagnosed narcissist and who has, since the twenty-tens, built an extensive private terminology. The super empath, the golden period, the fuel and the supernova all come from there. The terms are internally consistent, they are used consistently, and they rest on not a single study.
Only the core is older. The idea that a fragile self-image needs a continuous supply of confirmation from outside comes from psychoanalysis in the nineteen thirties and forties. Everything added since is embellishment.
That is no reason to look down on the vocabulary. Someone who has lived through something for which there was no language reaches for the language on offer, and this one is vivid, accurate in tone and free. It is only a reason not to mistake it for a finding.
The story goes like this. When a narcissist meets an ordinary empath he uses them up and moves on. When he meets a super empath, someone who gives just as much but who fights back at the end and leaves of their own accord, the relation tips over. That one person becomes the benchmark. He looks for her in every one who follows, does not find her, and never gets over it.
It is immediately clear why this story is needed. It reverses the power exactly: the person who was used up becomes the person who could not be replaced. It gives the ending a justice it otherwise does not have, and the experience a meaning. As consolation it works. That does not make it true.
Three things, and none of them has been measured.
First, super empaths would have to exist as a distinct group. Empathy is measured as several continuous characteristics. There is no threshold, no category, and no instrument that could place somebody in one.
Second, the loss would have to have lasting effect. What has actually been measured after an insult is something else. In two studies with 260 and 280 people, narcissism predicted aggressive behaviour after an insult while self-esteem predicted nothing at all, and the aggression was aimed specifically at the person who had given the insult rather than at uninvolved third parties. The documented response to injury is attack, not longing.
Third, the search for what was lost would have to steer the behaviour. The opposite was measured. Lower commitment traces back to perceived alternatives, that is, to the next person looking better, not to the previous one staying out of reach.
Something does remain, and it can be explained without anyone having to be unique.
An episode broken off abruptly stays in memory better than one carried to an end. That holds for everybody, and it holds for tasks as much as for relationships. An ending with no explanation is the sharpest form of it.
And the idealisation phase sets a benchmark against which everything later is measured. That works on both sides. Whoever lived through the golden period compares against it afterwards, and so does the person who produced it.
Together those explain why someone says later that they never had anything comparable again. They do not explain why that should have been down to one particular person.
Marked as an assessment. What follows puts the findings of this chapter together and goes beyond them in doing so. That is why it sits here and not in the evidence ladder.
The answer is yes, for three reasons that reinforce one another.
First, production. The pattern manufactures the conditions of a difficult relationship itself. Conditional warmth, alternating attention, criticism turned around: someone treated that way starts checking, asking, distrusting and explaining. After a year that looks like two difficult people, and a substantial part of it is the effect of one on the other. Anyone who looks only late can hardly tell cause from consequence.
Second, selection on one side. People with firm boundaries leave early, usually the first time attention is made conditional. They barely appear in the relationship history at all. What remains are the people who stay, and those who stay usually have a reason that itself has to do with attachment.
Third, selection on the other side. The opening phase is extraordinarily attractive to an anxious pattern, because it delivers exactly what is permanently missing there. So both move toward each other, and neither of them is doing anything unreasonable.
What speaks against it. No study has measured the sentence “narcissists have more difficult relationships” directly. Parts of it are measured: shorter relationships, lower commitment, more conflict, declining popularity over time. Putting those parts together is precisely the step that makes an assessment an assessment.
And the thought cuts both ways. If the same kind of relationship over and over points to the common denominator, that holds for anyone looking back on three similar relationships. That is uncomfortable and still the better use of it: the thought serves self-examination better than it serves sorting somebody else.
Click a phase. The shaded area is the swing between the lowest and the highest thing felt at the time; the two lines run on top of it. Drawn to show the shape the model describes, not measured averages.
The most striking line in the chart is not a line but a distance: the area between the highest and the lowest thing felt in the same week. At the start it is narrow, because everything is up. In the devaluation it widens, because the lows arrive and the highs remain. In the alternation phase it is at its widest, and that is the point. It is not the height that produces the strong feeling, it is the distance travelled.
First, contrast. The same warmth registers as larger after cold than after warmth. Someone who has not known where they stand for three days experiences an ordinary friendly message as relief. What is registered is not the state but the change, and here the change is enormous.
Second, arousal. A body under constant tension reads the situation as significant. The usual evidence for this is a 1974 experiment on a suspension bridge, and that evidence is shakier than it is usually quoted: the method has been criticised, replications came out mixed, and the effects are small to moderate. The current reading is that arousal amplifies a tendency that is already there rather than being mistaken for attraction. In practice both come to the same thing: an agitated body makes everything larger, including the good parts.
Third, memory. What remains of an episode is governed by its highest point and its end, not by its average. A relationship with enormous peaks is therefore remembered by its peaks, even when most of the days were bad. That explains the sentence almost everyone says afterwards: it was the most intense relationship of my life.
And the uncomfortable conclusion. Intensity is not depth. A nervous system cannot distinguish “this matters infinitely to me” from “this is unresolved and dangerous”; both feel the same. The sense of specialness is real as a feeling and is still not evidence about the quality of the relationship.
It is also why a calm relationship afterwards can feel like too little. The comparison is not with quality but with amplitude. A measure of how well you are doing is the height of the curve, not its swing, and that has to be said to oneself deliberately for a while.
What people report, and what can be measured.
Drawn to show the shape of the finding, not measured curves. In the study the reported growth rose while the separate measure of actual change did not move with it. Both lines are real: one is what the person knows about themselves, the other is what an instrument could pick up. They are not the same quantity, and the second one takes longer.
People who have come out of a long, difficult relationship almost always report growth: clearer boundaries, better judgement, a firmer sense of what they will not do again. A prospective study with 599 students, a hundred of whom went through a break-up during the study, put that report next to a separate measure of actual change. The reported growth was substantial. The measured change was unrelated to distress, unrelated to the reported growth, and unrelated to whether a break-up had happened at all. Optimism predicted the report, not the change.
That is not a reason to dismiss what people feel. It is a reason to expect the real movement later and from a different source: not from insight into what happened, but from repeated new experience that contradicts the old expectation.
Attachment is stable, not fixed. Rank order over time sits around r = 0.40, which means people keep their relative position more often than not, while the average drifts slowly toward security across adulthood. A new relationship with someone reliable raises security. Reaching a goal you set yourself raises it. Frequent small positive events between partners raise it. Job loss produces a lasting rise in attachment anxiety. Repeated arguments raise both anxiety and avoidance.
Emotion-focused couple therapy lowers both dimensions measurably, and simply wanting to be more secure predicts later security. The change is real, modest, and measured in years.
Narcissism declines on average across a life. A meta-analysis of 51 longitudinal samples with 37,247 people found decreases from age 8 to 77 on all three forms: agentic d = −0.28, antagonistic d = −0.41, neurotic d = −0.55. Those are small to medium effects across seventy years. The rank order over an average of eleven years, by contrast, stayed between 0.60 and 0.73, which is high.
Read together: the level comes down, the position in the ranking barely moves. Waiting for someone to become a different person is not supported by the numbers. Clinical samples declined more steeply than others, which is the one place the data leave room for treatment doing something.
Everything on this page, sorted by how much sits behind it. Figures and terms with a dotted underline lead to the tab The numbers, where each one is explained.
A pattern is not a diagnosis, and nobody makes a diagnosis from a distance or about a person who is not there to be asked.
The model explains a dynamic. It does not excuse behaviour. Control, threats, humiliation and violence are not an attachment style, and they are not answered by understanding them better.
The useful part is not sorting the other person. It is knowing your own threshold: what a short reply does to you, and what an insistent question does to you.
A label for a former partner ends the thinking instead of starting it. The question worth keeping is not what they were, but what you did with what they did.
Sources. Bowlby, Attachment and Loss (1969 to 1980) · Ainsworth et al., Patterns of Attachment (1978) · Hazan & Shaver, Romantic love conceptualized as an attachment process, JPSP (1987) · Bartholomew & Horowitz, four-category model, JPSP (1991) · Mickelson, Kessler & Shaver, adult attachment in a nationally representative sample, JPSP (1997) · Candel & Turliuc, insecure attachment and relationship satisfaction, 132 papers, N = 71,011, Personality and Individual Differences (2019) · Li & Chan, anxious and avoidant attachment and relationship quality, 73 studies, European Journal of Social Psychology (2012) · Schrodt, Witt & Shimkowski, meta-analysis of the demand/withdraw pattern, 74 studies, Communication Monographs (2014) · Roisman et al., the Adult Attachment Interview and self-reports, mean r = .09, JPSP (2007) · Chopik, Weidmann & Oh, attachment security and how to get it, Social and Personality Psychology Compass (2024) · Stinson et al., prevalence of DSM-IV narcissistic personality disorder, NESARC wave 2, N = 34,653, J Clin Psychiatry (2008) · StatPearls on narcissistic personality disorder · development of narcissism across the life span, 51 longitudinal samples, N = 37,247, Psychological Bulletin (2025) · Ameen et al., the narcissistic abuse cycle deserves clinical and research attention, Indian Journal of Psychological Medicine (2025) · Dutton & Painter, traumatic bonding (1981) · Owenz & Fowers, perceived post-traumatic growth may not reflect actual positive change, Journal of Social and Personal Relationships (2019) · Aron & Aron, sensory processing sensitivity, JPSP (1997) · Davis, Interpersonal Reactivity Index, JPSP (1983). · Paulhus, trait self-enhancement, a mixed blessing, JPSP (1998) · Leckelt et al., behavioural processes underlying the decline of narcissists' popularity, JPSP (2015), and a longitudinal field investigation, N = 126, PSPB (2020) · Campbell & Foster, narcissism and commitment, an investment model analysis, N = 531, PSPB (2002) · Hepper, Hart & Sedikides, moving Narcissus, can narcissists be moved to care, PSPB (2014) · Shahri et al., lower affective empathy in narcissism, Frontiers in Psychology (2024) · Carter & Douglass, the aging Narcissus, narcissism moderates the age to loneliness relationship, N = 200, Frontiers in Psychology (2018) · Finch et al., loneliness in narcissistic vulnerability, Personality and Mental Health (2024). · Dutton & Aron, heightened attraction under high anxiety, JPSP (1974), with the methodological criticism and the mixed replications noted in the text · Kahneman, Fredrickson, Schreiber & Redelmeier, when more pain is preferred to less, Psychological Science (1993), on the peak and the end of an episode · Funder & Ozer, evaluating effect size in psychological research, Advances in Methods and Practices in Psychological Science (2019), source of the everyday correlations · Lionetti et al., dandelions, tulips and orchids, Translational Psychiatry (2018). · Bushman & Baumeister, threatened egotism, narcissism, self-esteem and aggression, two studies with N = 260 and 280, JPSP (1998). The super empath, the golden period, the fuel and the supernova come from the self-published writing of an anonymous author and from the online community around it; they are named here because they are in wide use, and they are deliberately not cited as a source.
Read this once and every figure in this chapter becomes readable.
Take the strongest number in this chapter. Avoidance and relationship satisfaction go together at r = −0.44. That does not mean: whoever is avoidant is dissatisfied.
It means: if two people stand in front of you and all you know is which of them is the more avoidant, and you had to guess which is less satisfied in their relationship, you would be right more often than wrong. Not always. More often.
With no information at all you would be at fifty-fifty. With this one piece of information you are better off, and you are still wrong in a substantial share of cases. That is exactly the difference between a bet and a rule, and every number in this chapter is a bet.
Picture a hundred couples who differ in how satisfied they are: some very, some not at all, most somewhere in between. The question is where those differences come from.
At r = −0.44 about nineteen parts in a hundred of those differences go together with avoidance. The remaining eighty-one come from everything else: money, health, work, children, how long they have been together, how the other person is built, what happened this year, and a great deal of chance.
That is not a criticism of the number. Nineteen per cent is a lot in psychology. It is the statement of how much you know once you know this one thing: about a fifth.
Two figures from outside this chapter, for calibration. Aspirin after a heart attack lowers the risk of the next one at a correlation of about 0.03. That is almost nothing, and it is still a standard recommendation, because the event is severe, the measure costs almost nothing, and it concerns very many people.
The other way round, height and weight in adults go together at about 0.44, exactly as strongly as the strongest number in this chapter. And yet everyone knows tall thin people and short heavy ones, and nobody would think of reading weight off height.
Together those are the actual lesson. A small number can matter a great deal when many people or severe consequences are involved. A large number still says nothing about the one person in front of you.
The old rule of thumb comes from Cohen: 0.10 small, 0.30 medium, 0.50 large. A 2019 recommendation sets the bar lower, because large samples have shown how rare high values really are: 0.05 very small, 0.10 small, 0.20 medium, 0.30 large, and from 0.40 very large. The authors note that a value above 0.40 is usually an overestimate in large samples.
Under the newer rule the numbers in this chapter are large to very large; under the older one, medium. What they mean in practice does not change. So the scale below carries the old yardstick, and next to it figures from everyday life that are easier to picture than any word.
Every correlation used in this chapter on one scale, with the share of the differences it actually explains.
Dark bars come from this chapter, light ones from everyday life, so that the numbers can be placed. The dashed marks are Cohen's old rule of thumb. The column on the right is the share of the differences between people that the correlation accounts for, and it is the one worth remembering. Note where avoidance and satisfaction sits: exactly as strong as height and weight in adults. And note aspirin at the top, at almost nothing, which is nevertheless a standard recommendation. Everyday figures after Funder and Ozer (2019).
Every number in this chapter, in one place, with the tab it appears in.
How common the attachment patterns are, by two methods. · The four patterns
How strongly attachment anxiety and avoidance go with satisfaction. · Who meets whom, and What holds up
How well documented it is that pursuit and withdrawal do harm. · The trap
The ceiling for the lifetime prevalence of narcissistic personality disorder. · Narcissist and empath
The three experiments on whether empathy is missing or merely not called up. · Narcissist and empath
What lower commitment is down to: perceived alternatives. · Narcissist and empath, The dilemma
How much narcissism declines across a life and how stable the ordering stays regardless. · Afterwards
How little questionnaire and interview on attachment have to do with each other. · What holds up
How stable an attachment pattern is over time. · Afterwards, and What holds up
Every figure quoted here, and what it does and does not mean.
A measure of how strongly two quantities move together. It runs from −1 to +1. At 0 there is no relationship, at +1 both rise together, at −1 one rises as the other falls. The sign gives the direction, the magnitude gives the strength. It is an average across many people and says nothing about any one of them.
The square of the correlation, and the more honest number. It says what share of the differences between people is accounted for by the other quantity. A respectable-sounding r = 0.44 becomes 19 per cent. The rest lies elsewhere.
A difference between two groups or two points in time, measured in standard deviations. d = 1 means the groups sit a full spread apart. Used for change rather than for association.
A study of studies. Many individual results are pooled and weighted by size. That is steadier than any single study and inherits its weaknesses at the same time, above all publication bias: work that found nothing gets published less often, so the average looks a little clearer than it is.
The share of a population something applies to. Point prevalence means right now, lifetime prevalence means at some point in a life. The second is always the larger number, and reading it as a present-day figure overstates how common something is.
How common something is before you look at an individual case. Rare things stay rare even when a description fits well: with six in a hundred affected, among the people a rough description fits there are still far more unaffected than affected. Ignoring the base rate is the most common error in labelling.
An r comparing the same people at two points in time. It says whether the order is preserved, not whether the level changes. Both at once is possible: a trait can fall in everybody and still leave the ranking almost untouched. That is exactly what narcissism does across a lifespan.
Prospective means it was measured before the event. Retrospective means people were asked afterwards. Retrospective reports are coloured by what has happened since, and that is why reported growth after a break-up comes out so much stronger than measured growth.
Two ways of collecting the same thing. A questionnaire asks how you see yourself. A structured interview scores how someone talks about their history, independently of what they think about it. In attachment research the two routes give almost independent results, which is why “what is my attachment style” cannot be answered without naming the instrument.
The probability of getting a result at least this clear if there were in fact no relationship. p below 0.05 means: hard to explain by chance alone. It says nothing about the size of the effect. With 71,011 people almost everything becomes significant, including the trivial.
The number of people studied. Large beats small, but how they were selected matters more than how many there were. A magazine survey reaches only those who feel addressed and reply; a representative survey draws its participants by a procedure. That is why this chapter always puts both figures next to each other when it gives frequencies.
A construct is something you cannot observe directly and measure by detour, attachment anxiety for example. Validity means the instrument really captures what it claims to capture, and that this has been checked. For the empath no such instrument exists, which is why it sits at the bottom of the evidence ladder.