Resilience: Ordinary Recovery Rather Than a Rare Psychological Talent

Introduction

Almost every study that founded resilience research was designed to find damage.

The Kauai project followed a birth cohort from before birth into middle age, built to document what poverty, family instability and perinatal stress do to a child [1]. Emmy Werner and Ruth Smith called their book-length report Overcoming the Odds, which tells you what they found rather than what they went looking for [2].

Norman Garmezy's Project Competence sat inside developmental psychopathology, a field named after the thing it expected to see.

The 1984 paper Garmezy wrote with Ann Masten and Auke Tellegen proposed studying stress and competence together, and found the two related in children without either determining the other [3]. Michael Rutter came from the other direction, arguing in 1985 that the striking thing about children at high psychiatric risk was how many did not become ill [4].

None set out to find that people are durable. They kept finding it anyway.

That accident is why this field exists, and why it has spent twenty-five years arguing with itself. Once you notice that most people exposed to something terrible are functioning normally a year later, two questions follow. How many, exactly. And what is it about them.

The first question is far harder than it looks. There is a number you have almost certainly seen, and it comes from a real study with thousands of people in it. There is also a group of researchers who took the same kind of data, changed two lines in the statistical model, and watched that number invert. Both sides are still publishing.

The second question has a cleaner answer, and it is the one in the title.

Ann Masten called it ordinary magic in 2001, and the phrase gets quoted without the argument attached [5].

Her claim was not that resilience is heartwarming. It was that resilience runs on unremarkable equipment, and that what wrecks children is what breaks the equipment.

This article works through both in the order the field met them, naming who is on each side where the matter is contested. At the end there is a randomised trial that should change how you help someone after a disaster, because the people who got the help were worse off three years later than the people who did not.

The Number Everyone Quotes

Start with the study that gave you the figure.

In the six months after the September 11 attacks, George Bonanno and colleagues surveyed 2,752 adults across the New York area by random-digit dialling, a probability sample rather than a convenience sample of volunteers [6].

They defined resilience narrowly and unforgivingly: zero or one symptom of post-traumatic stress disorder. Not "coping well". Not "feeling positive". One symptom at most.

By that definition 65.1 percent were resilient.

The figure fell as exposure rose, which is what you would expect.

But even in the exposure groups with the largest increases in PTSD, the proportion never dropped below one third [6]. A follow-up analysis of the same 2,752 people found resilience predicted independently by nine things at once: gender, age, race and ethnicity, education, exposure level, income change, social support, frequency of chronic disease, and both recent and past life stressors [7]. Nine predictors, none of them dominant.

Hold that list in your head. It matters later.

Twelve years later, Isaac Galatzer-Levy, Sandy Huang and Bonanno pooled the trajectory literature: 54 studies producing 67 separate cases [8].

Four patterns came up over and over. A resilient group whose functioning barely moves. A recovery group that drops and climbs back. A chronic group that stays down. And a smaller delayed group that looks fine at first and gets worse later.

Averaged across populations, the resilient trajectory accounted for 65.7 percent, with a 95 percent confidence interval from 61.6 to 69.8 [8].

Recovery took 20.8 percent, chronic difficulty 10.6 percent, and delayed onset 8.9 percent. The resilient pattern appeared in 63 of the 67 cases. The chronic pattern appeared in 47 of them.

That last sentence deserves as much attention as the first. One person in ten in these samples is on a trajectory that does not come back on its own. If something happened to you and you are not fine, you are not a statistical failure. You are in a group that appears in nearly every study ever run on this, and the size of the large group is not a verdict on the small one.

Resilience Is Not Recovery, and the Difference Is the Whole Argument

Before 2004 the field mostly did not separate those first two groups, and that is why resilience looked rare.

Bonanno's 2004 paper in American Psychologist made the distinction that everything since has been built on [9].

Recovery, he wrote, means functioning drops below normal and comes back, sometimes quickly and sometimes over one or two years. Resilience is different in kind, not degree: it reflects the ability to maintain a stable equilibrium throughout. Lump the two together, as trauma researchers routinely had, and you assume that everyone who ends up all right must have gone through the same process to get there.

They had not, and you can see the consequences in how people were treated.

Bereavement research shows this most sharply. Theorists treated the absence of visible grief as suspicious, reasoning that anyone not devastated was in denial and would pay for it later.

Bonanno points out that chronic depression and distress after a death occur in only 10 to 15 percent of bereaved people [9].

The rest are not all fine, but the large group is not the distressed one.

Then a study came along that could settle the denial question, because it had data from before the death. Participants had been assessed on average three years before the spouse died.

Forty-six percent of them had low levels of depression both before the loss and through 18 months of bereavement, with few grief symptoms [9].

These were the people the theory said were cold, avoidant and heading for a delayed collapse. Their pre-loss assessments showed nothing of the kind. Interviewers had not rated them as emotionally distant, they had not reported difficulties in their marriages, and no unequivocal evidence of delayed grief turned up.

One detail in that group is the single most useful sentence in this whole literature.

Most of that 46 percent still reported yearning and emotional pangs, and virtually all of them reported intrusive thoughts and rumination early after the loss [9].

Resilience is not the absence of distress. It never was. It describes where somebody's functioning ends up over eighteen months, not what they felt on the way. Any version of this idea that promises you will not hurt is selling something the research does not contain.

Ordinary Magic Is an Argument About Machinery

Masten's 2001 paper is quoted more than it is read, and the part that gets dropped is the part that carries the weight.

Her summary of the developmental literature is that the most surprising conclusion is the ordinariness of resilience, that it usually arises from the normative functions of human adaptational systems, and that the greatest threats to human development are those which compromise those protective systems [5].

Read that third clause again. It is a claim about damage, not about virtue.

If resilience runs on ordinary systems, a child who does badly is usually not a child who lacked a rare gift. They are a child whose ordinary systems broke. A caregiver who disappeared. A school that stopped working. A neighbourhood that came apart. The question stops being what is special about the survivors and becomes what got wrecked for everybody else.

Masten and Coatsworth had already worked this out in 1998, comparing competence in favourable and unfavourable environments and finding the same protective machinery running in both [10]. Garmezy's 1991 work on children in poverty pointed the same way, reporting that the children who did well were not distinguished by anything exotic [11].

Rutter's framing is the crispest of the lot.

What protects a child is a process, not a possession [12]. Masten, Best and Garmezy had argued in 1990 that studying children who overcome adversity was a route into normal development rather than an exotic side road [13], and Masten's 2018 review describes the field reaching that position over four successive waves of research [14].

The modern version of this is explicitly about systems outside the person.

Michael Ungar and Linda Theron argued in Lancet Psychiatry in 2020 that resilience depends just as much on the culturally relevant resources available in someone's social, built and natural environment as it does on their own thoughts and behaviour [15]. A 2023 World Psychiatry review of protective and compensatory childhood experiences makes the same move in the other direction, asking what has to be present rather than what has to be absent [16].

A 2021 review in Child Development Perspectives pushes it one step further.

Supportive relationships, it argues, do not simply buffer a child against events; they change how safe and how stressed the child perceives their situation to be, regardless of which events they were actually exposed to [17]. Masten was still making this case in 2026, in a debate piece asking whether young people face too much chaos or have too little resilience, and coming down on the side of the chaos [18].

This is where resilience stops being an individual trait, and it reads nothing like the version you meet in a workplace training deck. It also sits directly opposite an idea you probably already know. Where learned helplessness describes an animal or a person who stops trying because trying stopped working, ordinary magic describes the far more common case where the machinery holds and trying keeps working.

The Same Four Shapes, In Places That Should Not Match

If those patterns were an artefact of American undergraduates or of one instrument, they would not survive being moved.

They survive it. Bonanno, Samuel Ho and Jane Chan followed 997 people hospitalised for SARS in Hong Kong, interviewing them at 6, 12 and 18 months after discharge [19].

Latent class analysis produced four groups: chronic dysfunction, delayed dysfunction, recovery and resilience. The authors note that an Asian sample facing a major health threat produced trajectories closely resembling those in Western trauma studies.

A study of 233 people with spinal cord injuries assessed them within six weeks of injury and again at three months, one year and two years [20].

The same classes appeared for depression, and the majority of patients were in the resilient one. The predictors are the interesting part. Resilient patients reported more challenge appraisals and fewer threat appraisals, greater acceptance and fighting spirit, and less coping through social reliance and behavioural disengagement.

Fran Norris, Melissa Tracy and Sandro Galea took two population-based datasets, 561 people after the 1999 Mexican floods and 1,267 after September 11, and tested six possible trajectories rather than four [21]. Four of the six were prevalent in both samples. Neither sample produced a relapsing and remitting pattern, and only New York produced a delayed one. A ten-year study assessed 1,357 adolescents who survived the Wenchuan earthquake at 6, 12, 18 and 24 months, then reached 744 of them a decade later [22].

Three resilience trajectories came out of that early data: 19.8 percent low, 58.7 percent moderate and 21.5 percent high.

Four shapes, across a terrorist attack, an epidemic, a flood, an earthquake and a spinal injury, in three continents. That part of the story is as replicated as anything in this corner of psychology.

Now the part that is not, and it is the part nobody told you.

Then Somebody Checked the Statistics

Here is what a trajectory study actually does, because it explains everything that follows.

You measure the same people several times. You hand the measurements to a growth mixture model, a procedure that sorts people into a small number of latent classes by the shape of their change over time. The model tells you how many classes there are and how big each is. Then you name them, and the biggest gets called resilient.

The number of classes the model finds depends on the assumptions you give it.

In 2016 Frank Infurna and Suniya Luthar published a paper in Perspectives on Psychological Science with a title that left little room for interpretation: resilience to major life stressors is not as common as thought [23].

They took the same longitudinal dataset that had been used to argue the opposite, across three events: spousal loss, divorce and unemployment.

Run with identical model specifications to the earlier work, the resilient trajectory came out most common, exactly as before. Then they relaxed two assumptions. The first was that all classes have similar variability in how people are adjusting afterwards. The second was that there is no variation between individuals in how much they change. Both are convenience assumptions, made because they simplify the estimation.

With those two relaxed, the resilience class was the least common one.

Their conclusion is not that resilience is rare. It is narrower and harder to argue with.

They write that the results underscore the errors inherent in any categorical statement about rates of resilience among people confronted with major life stressors [23]. The shapes are real. The percentages are outputs.

The obvious next question is how many published studies made those two assumptions.

Infurna and Luthar answered it in 2018 by reviewing 77 empirical growth mixture studies of resilience in adulthood [24].

Sixty-six of the 77, or 86 percent, assumed homogeneity of variance across trajectories. Fifty-two, or 68 percent, set slope variances to zero. And in the minority of studies that did not impose the homogeneity assumption, the resilient trajectory was among the smallest.

A companion methods paper shows the mechanism. Infurna and Kevin Grimm applied growth mixture models to spousal loss data from two national panels, once constraining variances to be equal across classes and once letting them differ [25].

Constraining them produced systematic overextraction of classes. Assuming normally distributed data when the data were not normal did the same thing. Their verdict on the method as commonly applied is blunt. The assumptions typically underlying it, they write, "are not tenable", and the result is that they are "misinforming conceptual models of resilience".

The Same 421 People, Five Different Answers

The 2018 review turned up a third problem, easier to grasp than the statistical ones and probably more damaging.

Sixty-three of those 77 studies, 82 percent of them, awarded the label resilient on the basis of a single outcome measure [24].

One questionnaire. One dimension of a person's life.

Infurna and Luthar tested what happens when you use more than one.

They took 421 people from the Australian HILDA panel who lost a spouse during the study and modelled five indicators separately: life satisfaction, negative affect, positive affect, self-rated general health and physical functioning [26].

Same bereaved people. Same statistical method. Same eighteen months.

The proportion who look resilient runs from 66 percent down to 8 percent depending entirely on which question you decide counts [26]. Twenty percent were non-resilient on all five.

You can see why a headline percentage is hard to defend. Measure life satisfaction and you report two thirds. Require the whole person to be doing well and you report one in twelve. Neither is wrong. They are answers to different questions wearing the same word.

Bonanno's Group Replies, and They Are Not Wrong Either

This is where a lot of popular writing would pick a side. The evidence does not support picking one.

Galatzer-Levy and Bonanno replied in the same 2016 issue, under the title "It's Not So Easy to Make Resilience Go Away" [27]. Infurna and Luthar answered again in that issue, arguing that resilience has been and always will be, but that declared rates are inevitably suspect [28]. Three papers, one issue, no resolution.

Two years later the 54-study review gave Bonanno's side its best evidence [8].

Across all those samples the four trajectories showed up in consistent proportions, and the authors concluded that variation in the estimates came mostly from genuine population differences rather than bias. Bias appeared specifically where prospective data was missing. That is a real finding, and pointing at model assumptions does not answer it.

Bonanno's own 2011 review with Maren Westphal and Anthony Mancini had already conceded the part that matters most, which is that resilience comes from many independent predictors rather than a few dominant ones [29]. Nobody in this argument thinks they have found the resilience gene. A 2024 review of person-centred studies in young people finds the heterogeneity problem unresolved [30], and a 2025 analysis shows how much the picture depends on how adversity itself is graded [31].

The honest summary is two sentences. Most people exposed to a potentially traumatic event do not develop a lasting disorder, and every party to this dispute accepts it. The specific percentage is a model output that moves when the model changes, and nobody should quote it as a constant.

The Measurement Problem Underneath All of It

This field argues about numbers more than most because the thing being measured has never been pinned down.

The field said so itself in 2000. Luthar, Dante Cicchetti and Bronwyn Becker published a critical appraisal in Child Development listing the charges against their own construct: ambiguity in definitions and central terminology, heterogeneity in the risks people faced and the competence they achieved, instability of the phenomenon over time, and doubts about whether resilience is useful as a theoretical construct at all [32].

They defended the construct, but they printed the charges first.

Twenty-five years on this is not fixed. David Fletcher and Mustafa Sarkar reviewed definitions and theory in 2013 and found the same conceptual disagreement running through the literature [33].

The instruments make it concrete. The Connor-Davidson Resilience Scale, published in 2003, asks people to rate 25 statements about themselves [34]. The Brief Resilience Scale, from 2008, gets the same construct down to six items about bouncing back [35].

Neither has settled the matter. A 2026 psychometric synthesis of the Connor-Davidson scale is still working through how its properties change across populations [36].

Here is the thing worth holding onto. A trajectory study and a questionnaire study both report on resilience, and they are not measuring the same object. One is a shape in longitudinal data. The other is what somebody says about themselves on a Tuesday.

And the problem reaches the adversity side too.

Jessie Baldwin and colleagues meta-analysed 16 studies that had both prospective and retrospective measures of childhood maltreatment in the same people, and found the two identify substantially different individuals [37].

Researchers had been using them interchangeably. If you cannot agree on who was harmed, you will not agree on who recovered.

The Search for the Thing That Makes People Durable

Given all that, the obvious move is to look inside the person for the mechanism. That search has a cautionary history.

In 2003 Avshalom Caspi and colleagues reported in Science that a polymorphism in the serotonin transporter gene moderated the effect of stressful life events on depression, so that people carrying the short allele were more likely to become depressed after the same adversity [38].

It became one of the most cited findings in psychiatry and launched an enormous candidate-gene literature.

In 2009 Neil Risch and colleagues meta-analysed 14 studies of that interaction in JAMA, covering 14,250 participants and obtaining individual-level original data from 10 of the samples [39].

The number of stressful life events predicted depression, with an odds ratio of 1.41 and a confidence interval from 1.25 to 1.57. The gene on its own gave 1.05, with an interval from 0.98 to 1.13. The interaction between them gave 1.01. The effect that had launched a decade of research was not there.

Adriana Feder, Eric Nestler and Dennis Charney's 2009 review remains the standard account of the neurobiology, and is careful about how much of it is animal work [40]. A 2025 open-science review of neuroimaging correlates of resilience is still working on replicability rather than reporting a settled circuit [41].

The current proposals are less about having the right equipment and more about switching.

Bonanno, Shuquan Chen and Galatzer-Levy set out regulatory flexibility in Nature Reviews Psychology in 2023, arguing that what predicts good outcomes is not any particular coping strategy but the ability to read a situation and change strategy when it changes [42]. A 2026 paper in Trends in Cognitive Sciences makes a parallel case for positive appraisal style, where the candidate mechanism is how automatically a person evaluates a threat rather than what they then do about it [43].

Notice what happened to the tidier explanations you have probably met.

Bonanno's 2004 paper listed several routes to resilience including hardiness, self-enhancement, repressive coping and positive emotion, and pointed out that some are traits psychology normally treats as undesirable [9].

Self-enhancement is a bias. Repressive coping is avoidance. Both were associated with better adjustment in bereaved samples.

A 2026 meta-analysis pooling 28 studies of 11,622 children with attention-deficit hyperactivity disorder found the same shape: six factors inside the child, including emotional regulation and social skills, and four outside them, contributing to good outcomes [44].

Ten weak contributors is what this literature keeps producing, and it never resolves into one strong one.

The predictor with the longest track record is not inside the person at all.

Sheldon Cohen and Thomas Wills laid out the buffering hypothesis in 1985, arguing that social support protects by intervening between a stressful event and the response to it [45]. Forty years later it is still the finding that holds up best, though a 2023 re-analysis of the meta-analytic evidence found that depending on which statistical model you fit, social support can be shown to decrease, increase or have no prospective effect on later post-traumatic stress [46].

The lesson of the 5-HTTLPR episode applies to everything in this section. A finding that explains why some people are tougher than others is exactly the finding a field wants to be true, which is precisely when replication matters most. The same caution belongs on the neuroimaging work now. It is also why the amygdala appearing in a press release about resilience should be read as a starting point rather than a conclusion, and why what stress hormones actually do to memory is better understood than what they predict about long-term outcome.

The Pandemic Ran the Experiment Nobody Could Have Designed

In early 2020 an adverse event hit essentially everyone at once, and a large number of ongoing cohort studies already had baseline measurements on their participants.

Eric Robinson, Angelina Sutin and Michael Daly meta-analysed 65 longitudinal cohorts that measured mental health in the same people before and during 2020 [47].

That is a far stronger design than the cross-sectional surveys which dominated pandemic coverage, because each person acts as their own control.

Symptoms did rise in March and April, with a standardised mean change of 0.102 and a confidence interval from 0.026 to 0.192. By May to July that increase had declined and was no longer statistically significant, at 0.067 with an interval crossing zero. Depression and mood disorder symptoms behaved differently and stayed elevated at 0.20. General mental health did not move at all.

The most counterintuitive result is easy to miss.

There was no evidence of any change in symptoms among samples that already had a mental health condition [47]. The group everyone worried about most did not, on average, get worse.

A latent class analysis of 19,763 adults in a UK national probability panel found five distinct mental health trajectories through the same period rather than one population-wide response [48].

That is the trajectory picture again, at national scale, in real time.

Be careful about what this does and does not show you. A small average rise that fades is not the same as nobody suffering. It is a statement about a population, and populations contain the chronic ten percent. What it establishes is that the expected population-level collapse did not happen, and that the recovery pattern is not an artefact of studying unusual events in unusual samples.

Does Resilience Training Work?

If most people already follow a recovery path, and the protective factors are many and weak, what should a programme designed to build resilience achieve?

Roughly what we get, a small effect.

Aaron Leppin and colleagues meta-analysed 25 randomised trials of resiliency training programmes and found a pooled standardised mean difference of 0.37, with a 95 percent confidence interval from 0.18 to 0.57, on resilience measured within three months of follow-up [49]. They describe the trials as small and at moderate to high risk of bias, and rate confidence in the evidence as low. Sadhbh Joyce and colleagues screened 437 citations, included 17 studies and meta-analysed 11 randomised controlled trials, reporting a pooled effect of 0.44 with an interval from 0.23 to 0.64 [50].

Effects of 0.37 and 0.44 on a self-report resilience questionnaire are not nothing. They are also not the transformation these programmes are usually sold as, and they are measured on the instrument the programme spent weeks teaching people to think in terms of.

The flagship youth programme has the most informative evidence, because it is the one with the most trials.

Steven Brunwasser, Jane Gillham and Eric Kim reviewed 17 controlled evaluations of the Penn Resiliency Program covering 2,498 young people [51].

Effect sizes on depressive symptoms ran from 0.11 to 0.21, and were still detectable a year after the intervention. Then comes the sentence that should be in every article about resilience training and is in almost none of them: limited data showed no evidence that the programme is superior to active control conditions.

An active control is another activity of the same length and attention as the one you are testing. Beating a waitlist tells you doing something beats doing nothing. Beating an active control tells you the content matters. That second test has not been passed.

This is not marginal, because the Penn programme was the basis for Comprehensive Soldier Fitness, rolled out across the United States Army.

The programme was presented in American Psychologist in 2011 by Rhonda Cornum, Michael Matthews and Martin Seligman as resilience building at scale [52]. In the same journal and the same year of 2011, Roy Eidelson, Marc Pilisuk and Stephen Soldz published a critique arguing that the evidence base did not support deployment on that scale [53].

That disagreement, in the field's flagship journal, rarely appears in coverage of resilience training.

School programmes tell a similar story. A systematic review of universal school-based resilience interventions aimed at reducing adolescent substance use identified 19 eligible randomised trials from over 16,000 records and did not find the consistent benefits the approach promised [54], and a 2025 review of 38 randomised trials covering 15,730 participants continues to report modest effects from school-based programmes [55].

There is a comparison worth making, because you have probably met it. Grit was presented as a distinct trait that predicts success where talent does not.

Marcus Credé, Michael Tynan and Peter Harms meta-analysed 584 effect sizes from 88 samples covering 66,807 people [56].

The higher-order structure was not confirmed. Grit correlated only moderately with performance and very strongly with conscientiousness, meaning it was largely an existing trait under a new name, and the authors concluded that interventions to enhance it may have only weak effects. Popular individual-difference constructs shrink when someone pools all the studies, and knowing that pattern is worth more than any single result. The same caution applies to how impostor syndrome is discussed, and to the difference between high standards and self-punishment.

The Intervention That Made People Worse

Now the study that should change what you do.

Psychological debriefing is the intuitive response to trauma. Within a day or two you sit the person down and walk them through what happened, in detail, with their reactions normalised by someone trained to do it. It was built for emergency responders and spread to civilians, because it feels obviously kind.

The trial randomised road traffic accident victims admitted to hospital to receive it or not, and at four months there was no benefit.

Mayou, Anke Ehlers and Hobbs then followed the same patients to three years [57].

The intervention group had a significantly worse outcome. Not on one incidental measure, but on general psychiatric symptoms, on travel anxiety as a passenger, on pain, on physical problems, on overall level of functioning and on financial problems.

The detail that matters most is what happened to the people worst off at the start.

Patients who initially had high intrusion and avoidance symptoms remained symptomatic if they had received the intervention, and recovered if they had not [57].

The authors' conclusion runs to one sentence. Psychological debriefing, they write, "is ineffective and has adverse long-term effects" and "is not an appropriate treatment for trauma victims".

A 2002 Cochrane review pooled eight randomised trials of single-session debriefing [58].

It found no reduction in the risk of PTSD at three to five months, with a pooled odds ratio of 1.0 and a confidence interval from 0.6 to 1.8. At one year, one trial reported a significantly increased risk of PTSD in those who had been debriefed, with an odds ratio of 2.9 and an interval from 1.1 to 7.5.

Bonanno makes the connection to the trajectory work explicitly, and it is the argument this whole article has been building towards [9].

If most people exposed to a potentially traumatic event are already on a resilient or recovering path, then a universal intervention is administered overwhelmingly to people who were going to be all right. Best case, it does nothing for them. Worst case, it interrupts a process that was already working, by keeping the event vivid and rehearsed at exactly the point when ordinary systems were beginning to file it away.

That is what the ordinariness claim is for. It is not reassurance for you. It is a warning about who you treat.

Targeted intervention is a different matter, and the same literature supports it.

Bonanno's 2004 paper endorses screening for the risk factors that predict chronic outcomes, prior trauma, low social support and hyperarousal, and intervening with those people [9]. Recent work is trying to make that screening accurate enough to act on in the days after an injury [59].

The distinction is not treatment versus no treatment. It is treating the ten percent versus treating everyone.

What the Word Does When It Leaves the Laboratory

There is a second criticism of resilience that has nothing to do with statistics, and it must be kept separate from the first. Merging them is the commonest mistake in writing on this subject.

The complaint is about usage. When an employer tells overloaded staff to be more resilient, or a policy document asks a community with no services to be resilient, the word is doing work the research never licensed.

A 2023 paper in the International Journal of Mental Health Nursing argues that the term has drifted into exactly this role, placing the burden on individuals for conditions they do not control [60]. Catherine Panter-Brick's 2014 review of the anthropological literature makes the related point that resilience as usually operationalised carries assumptions about what a good outcome looks like that do not travel between cultures [61]. The 2020 multisystem argument from Ungar and Theron is the constructive version of the same objection [15].

This is a critique of how the concept gets used, not evidence that recovery is uncommon. Both can be true. The data can show that most people recover, and the word can still be misused to imply that anyone who does not has failed personally.

The 2014 round-table in the European Journal of Psychotraumatology is the cleanest place to see the disagreements laid out together, because Steven Southwick, Bonanno, Masten, Panter-Brick and Rachel Yehuda state their positions side by side rather than through review articles [62].

They agree resilience is multidimensional. They do not agree on definition, on measurement, or on how much of it sits inside the individual.

One more comforting idea deserves the same treatment on the way out.

Post-traumatic growth, or PTG, introduced by Richard Tedeschi and Lawrence Calhoun in 2004, is the claim that some people are improved by what happened [63]. Adriel Boals reviewed the measurement evidence in 2023 and drew a distinction between perceived growth, genuine growth and illusory growth [64].

Perceived growth is extremely common, with over half of people exposed to a potentially traumatic event reporting moderate or greater levels of it. Boals argues most of those self-reports are greatly exaggerated, and he writes that he has come to "the bold conclusion that the occurrence of genuine PTG is very rare".

So no, on the current evidence, the bad thing did not make you stronger. Returning to who you were is the finding, and it is a better one.

What This Actually Means

Three things survive all of that, and they are the three worth keeping.

Most people exposed to something terrible do not develop a lasting disorder, and researchers who disagree about almost everything else agree on that. The four trajectory shapes replicate across a terrorist attack in New York, an epidemic in Hong Kong, a flood in Mexico, spinal injuries in a rehabilitation ward and a pandemic across 65 cohorts. Little else in this branch of psychology replicates that well.

The percentage attached to the largest group is not a fact about people.

It is an output of a statistical model, and it moves from most common to least common when two convenience assumptions are relaxed [23], and from 66 percent to 8 percent depending on how many parts of a life you insist on checking [26].

If you quote a single figure without that caveat, you have not read past the abstract.

And the mechanism is ordinary. Not a rare talent, not a gene that survived meta-analysis, not a course you can take. It is caregiving that held, relationships that stayed intact, a routine that kept going, the expectation that somebody would show up.

Masten's phrase for it is that resilience is made of ordinary rather than extraordinary processes [5], and the practical consequence of that is not that individuals should try harder.

It is that the systems around people are where the difference gets made, which also explains why the conditions that support motivation keep reappearing in this literature under different names.

The most useful thing here is the least comfortable thing you will read.

When something bad happens to someone you know, the instinct to sit them down and get them to process it is the instinct that produced worse outcomes at three years in a randomised trial [57].

What the evidence supports instead is unglamorous. Stay available. Keep the ordinary systems running. Watch for the minority who are not recovering, and get those people real help. And notice that what trauma does to memory in that minority is a genuinely different process from what most people go through.

That is the whole finding, and it is quieter than the version you were told.