Published on 15 September 2026
Parental wealth and youth mental disorders: what survives a comparison between siblings?
In brief
A Norwegian registry cohort followed 1,414,670 young people aged 7 to 24 between 2006 and 2023, which amounts to 13,046,418 observation-years. Four indicators of parental social position entered the models together: gross wealth, income, educational attainment, occupational prestige. Wealth is the one that remains most strongly associated with the prevalence of mental disorders in adolescents and young adults, on a par with educational attainment in children. The predictive value of income is substantially attenuated, and in many categories eliminated, as soon as wealth enters the model. In adolescents, the prevalence ratio between the lowest and the highest wealth percentile is 1.83 (95% CI 1.76 to 1.90) after adjustment for the other three indicators. The comparison between siblings, which cancels out what a family shares, brings that ratio down from 2.11 (2.07 to 2.15) at population level without adjustment to 1.34 (1.26 to 1.42) within families. What is established here: a graded wealth gradient, precisely estimated, present in almost every diagnostic category, and one that does not disappear once shared familial confounding is controlled for. What remains suggested: that wealth counts in itself rather than standing in for something else. Two reservations weigh heavily. A disorder exists in these data only if it has met the healthcare system, and Norway combines access to care that depends little on household means with wealth inequality among the highest in the OECD countries, a combination found nowhere else in the same form.
The context
That the social position of a family is associated with the mental health of its children is one of the best replicated facts in psychiatric epidemiology. In consultation, that fact usually shrinks to two questions: what the parents do for a living, what the household earns. Both describe a flow, what comes in each month. Neither says anything about a stock, what the family owns or owes: savings, housing, debts, the capacity to absorb an unforeseen expense.
The distinction is not rhetorical. Two households on the same salary do not have the same margin when a parent stops working, when a move becomes unavoidable, when a bill lands. The study discussed here tests whether that margin of wealth carries the bulk of the association with mental disorders in the young, once income, education and occupational status are held constant. It tests it in the only kind of setting where this is possible: a country that records the taxable wealth of households and the diagnoses issued by the healthcare system, and that allows the two to be linked across several generations of births.
The mechanism
What is measured, and what is not
The pathways by which low family wealth might weigh on a child’s mental health are plausible and have been studied elsewhere: lasting material insecurity and the burden of stress that goes with it, instability of housing and schooling, the availability and the health of the parents themselves, social comparison in adolescence, unequal access to structuring activities. None of these pathways is measured in this study.
The work links two ends of a chain, a wealth position recorded by the tax administration and a diagnosis recorded by the health reimbursement system, and leaves the middle empty. This is a limitation of design rather than an oversight: an administrative registry contains neither perceived stress, nor the family climate, nor life events. The consequence for the reader is direct. Any sentence explaining how wealth might act belongs to hypothesis, not to results.
One point has to be set down before the figures. Parental wealth is not a neutral variable: it is itself associated with the mental health of the parents, with their schooling, with their occupational trajectory, all characteristics transmitted to the child by pathways that are at once genetic and environmental. That is exactly what the comparison between siblings seeks to cancel out, and the interest of the paper depends on how that contrast is read.
The study at a glance
| Population, exposure, comparator, outcome | |
|---|---|
| Population (P) | |
| Detail All individuals born in Norway between 1993 and 2011, observed from age 7 to age 24. N = 1,414,670, that is 13,046,418 observation-years over the period 2006 to 2023, each subject being observed six times, once a year, in each stratum they belong to. Three age strata, one subject being able to contribute successively to several of them: children aged 7 to 12 (N = 837,693), adolescents aged 13 to 18 (N = 850,445), young adults aged 19 to 24 (N = 486,265). The sum of the strata exceeds the total, which confirms that they are not mutually exclusive. 3,904 individuals were removed for missing health data, following emigration, death or another reason. The authors report no restriction based on the country of birth of the parents. | |
| Exposure (I) | |
| Detail Parental gross wealth in the low percentiles. The measure is the average of the two parents over the child’s ages 0 to 6, converted into a percentile rank within the child’s birth year so that the four indicators and the cohorts remain comparable. Gross wealth is defined as the sum of real capital and estimated financial capital, with no deduction of debts. The first percentile of wealth and of income was excluded, predicted probabilities there diverging from the rest of the distribution, which the authors attribute to possible tax evasion. Source: variables constructed by Statistics Norway from tax returns, the tax register and the unsecured debt register. | |
| Comparator (C) | |
| Detail The comparator changes with the analysis, which governs how the figures read. Descriptive prevalences: highest wealth quintile. Multivariable models and sibling comparison: highest wealth percentile. Ratios by diagnostic category: third quintile taken as the reference. The multivariable models adjust simultaneously for three other indicators of parental social position: income, educational attainment, occupational prestige. | |
| Outcome (O) | |
| Detail One-year prevalence of 17 categories of mental disorders, coded in ICD-10 and ICPC-2, drawn from the national health reimbursement registry (KUHR). This is a prevalence and not an incidence: the ratios reported are prevalence ratios, not relative risks of onset. | |
| Design and analysis | |
| Detail Longitudinal national registry cohort, level of evidence CEBM 2b. Generalised estimating equations with a log link and robust covariance estimation for the population models. For the sibling comparison, multilevel regression with a logit link and random family intercepts, siblings being identified by the shared mother in the population registry; odds ratios were converted into prevalence ratios from the baseline risk. Trends by birth cohort were tested through an interaction term between wealth quintile and birth year, model fit being compared with chi-square tests and the Akaike information criterion. Analyses stratified by age and by sex. R software, glmtoolbox and lme4 packages. |
Quality control
| Domain | Judgement |
|---|---|
| Comparison between siblings | Robust |
| Finding Multilevel models with random family intercepts cancel out what siblings have in common: part of the genetic background, the family environment, most often the neighbourhood and the school. With no possible randomisation of wealth, this is the best equipped observational tool for the question. It does not turn it into an experimental question, and the authors state its limits themselves: siblings identified through the mother alone, so half-siblings with different fathers are included, restriction to families with at least two children (252,687 families among children, 256,341 among adolescents, 112,653 among young adults), and a very high correlation of wealth rank between siblings, 0.86 to 0.92, which leaves little variation to exploit. | |
| Power and precision | Maximal |
| Finding 13,046,418 observation-years. Confidence intervals stay narrow down to the strata. The counterpart deserves saying: at this power, a minimal difference becomes significant. The useful question is no longer the statistical threshold, it is the magnitude and its clinical relevance. | |
| Four social indicators in the same model | Rare and relevant |
| Finding Wealth, income, education and occupational prestige are modelled jointly, which the literature rarely does. That is what allows indicators to be compared with one another. Collinearity remains an issue, but the authors place it at a moderate level: correlations between the four indicators run from 0.27 to 0.70, and the correlation between wealth and income among Norwegian parents is given as 0.37 in earlier work by the same team. | |
| Measurement of wealth | Partial |
| Finding The authors acknowledge three blind spots of the tax registry: informal assets, the value of unlisted companies, which is underestimated, and the wealth of individuals who practise tax avoidance, the exclusion of the first percentile addressing that last point only in part. To this is added the choice of gross wealth, which does not deduct household debt: two families with the same gross wealth do not have the same real margin. The exposure is therefore measured with error. If that error is independent of the diagnosis, it attenuates the associations rather than exaggerating them. | |
| Ascertainment of disorders | Detection bias |
| Finding Only disorders that gave rise to a recorded contact with a clinician are visible. If the least resourced families consult less, the real gap is underestimated. The symmetry must be maintained: for other diagnoses, more frequent use of services in well-off families can create an apparent gap in the opposite direction. This point is decisive for interpreting the exception of eating disorders. | |
| Funding and declared interests | No industry ties |
| Finding European academic funding: European Research Council consolidator grant no. 101045526, GeoGen project. Data access and management costs are covered by the Research Council of Norway and by other European Research Council grants. The funder took no part in the design, the collection, the analysis or the writing. No conflict of interest declared, open access publication under a Creative Commons licence, analysis code publicly deposited. The risk of allegiance is not nil for all that: a team whose programme is about social inequality publishes a result that confirms the weight of social inequality. | |
The findings
| Result | Estimate |
|---|---|
| Crude difference, lowest against highest quintile | + 87% (95% CI 78 to 96%) |
| Reading One-year prevalence, averaged over cohorts and over both sexes, with no adjustment at all. The authors note that the widest jump sits consistently between the lowest quintile and the second, which points to a threshold effect rather than a regular slope. This figure aggregates everything that accompanies wealth in the life of a family. It describes a social reality, it does not read as the share attributable to wealth. | |
| Children aged 7 to 12, multivariable model | PR 1.53 (1.46 to 1.60) |
| Reading Adjusted for the other three social indicators. A clear and precise association from school age onwards. This is the only stratum where wealth does not stand out: parental education is equivalent there, PR 1.52 (1.45 to 1.60), and the authors present the two indicators as jointly the most predictive in children. | |
| Adolescents aged 13 to 18, multivariable model | PR 1.83 (1.76 to 1.90) |
| Reading The point estimate is higher than at school age, but the authors describe wealth gradients that are on the contrary similar from one age stratum to the next. The progression of the multivariable estimates is visible; the study does not present it as a result and tests no explanation for it. | |
| Young adults aged 19 to 24, multivariable model | PR 1.92 (1.80 to 2.04) |
| Reading The highest point estimate of the three strata, but on the smallest sample and for an exposure measured thirteen to twenty-four years earlier. The distance from the adolescent stratum is modest given the intervals. | |
| Comparison between siblings | PR 1.57 then 1.34 then 1.42 |
| Reading Children 1.57 (1.44 to 1.71), adolescents 1.34 (1.26 to 1.42), young adults 1.42 (1.31 to 1.54). The comparator chosen by the authors is the population estimate without adjustment, respectively 2.04 (1.98 to 2.10), 2.11 (2.07 to 2.15) and 2.01 (1.97 to 2.04). The attenuation is therefore clear in all three strata, the strongest being observed in adolescents. Part of the association did rest on what siblings share, and part of it resists. | |
| Income adjusted for wealth | Independent contribution strongly attenuated |
| Reading The authors phrase it cautiously: the predictive value of income is “substantially attenuated or eliminated” in many cases when wealth, education and occupation appear in the same model. This is not a general disappearance. Nor does it demonstrate that income does not count: the four indicators correlate between 0.27 and 0.70 and compete for part of the same variance. The authors themselves offer a statistical explanation, the far greater variability of wealth, and show that in standard deviation units rather than in ranks, the slope for wealth is clearly the steepest only in young adults. The result is compatible with wealth being more informative than income, it does not establish that income has no effect, and nothing here invites anyone to drop income from the social assessment. | |
| Categories with the steepest gradient | Conduct, substance use, attempted suicide |
| Reading Consistent with the literature on externalising disorders and suicidality. In adolescents, against the third quintile taken as the reference, the lowest wealth quintile gives a ratio of 2.17 (1.93 to 2.45) for conduct disorder, 2.45 (2.08 to 2.87) for substance use and 2.23 (2.05 to 2.44) for attempted suicide. These three categories are also the ones whose contact with the healthcare system is most often compelled, through emergency care or reporting, and therefore the least dependent on a voluntary step taken by the family. | |
| Exception, eating disorders | Gradient absent or reversed |
| Reading The authors describe “either no gradient or a positive wealth gradient”, and elsewhere in the discussion attach this category to the absence of a socioeconomic gradient. The magnitude is small: in adolescents, against the third quintile, the lowest quintile gives 0.92 (0.84 to 1.01) and the highest 1.14 (1.05 to 1.24). Two explanations remain open and the study does not settle between them: a genuinely higher frequency in better resourced families, or earlier and more frequent ascertainment where people consult more. The second hypothesis deserves all the more consideration in that detection bias runs the other way for the remaining disorders. | |
| Results stratified by sex | Gradient present in both sexes |
| Reading In the cohort born in 1999, boys have the higher prevalence up to age 14 and girls beyond it, and this holds in every wealth quintile through age 24. The wealth gradient remains significant in both sexes at all ages, except at 7 and 8 years in girls. Estimates by sex and by diagnostic category appear in the source article; they are not reproduced here in full, which is a choice of presentation and not a finding that no difference exists. | |
Critical appraisal
| Domain | Risk |
|---|---|
| Residual confounding | Moderate |
| Finding The comparison between siblings cancels out what a family shares, not what separates its members. And what makes parental wealth move between two births in the same family is anything but trivial: job loss, separation, illness, an accident of life. These events have their own association with the mental health of the child. The design removes one familial confounder and lets in another, a temporal one. It moves closer to causal inference, it does not establish it. | |
| Selection of subjects | Low at population level |
| Finding An exhaustive national birth cohort, with no volunteering and no loss to follow-up in the usual sense of clinical cohorts; only 3,904 individuals are removed for missing health data. Selection shifts into the sibling analysis, which by construction excludes only children: the authors note that one-child families have a lower socioeconomic position than families with at least two children, and they themselves call this a selection bias. The within-family estimate therefore does not apply to the same population as the population estimate. | |
| Measurement of the exposure | Moderate |
| Finding Gross taxable wealth approaches real wealth without covering it, and a percentile rank is not an amount: the authors count among their limitations the fact that ranking erases the real material spacing and probably underestimates the weight of large wealth gaps. The order of magnitude can still be read for one cohort: in the cohort born in 1999, mean parental wealth at child ages 0 to 6 runs from 45,168 Norwegian kroner in the first quintile to 1,052,918 in the fifth, the three middle quintiles sitting between 137,480 and 340,948. The distance between the first four quintiles is therefore small next to the break represented by the fifth. This measurement error is most likely non-differential, and therefore conservative. | |
| Measurement of the outcome | High |
| Finding This is the main weakness. The outcome is not the mental disorder, it is the mental disorder diagnosed and recorded. Use of services itself depends on social position. Misclassification is therefore probably differential, that is, linked to the exposure, and its direction may vary with the diagnosis. It most likely attenuates the gradients of the quieter disorders and could on its own produce the reversed gradient observed for eating disorders. | |
| Missing data | Partly documented |
| Finding The authors report registry completeness of 74.6 to 96.8% for wealth, income and education, available since 1993, and lower completeness for occupational prestige, whose codes go back only to 2000. The detail of that completeness and the handling of missing values sit in the supplementary appendix, which could not be consulted. The matter is therefore not closed: completeness of 74.6% on an exposure indicator is not trivial, and what is reported here is an incompleteness on our side, not a finding that there is no problem. | |
| Multiplicity of analyses | No correction described |
| Finding Seventeen diagnostic categories, three age strata, two sexes, four social indicators, two families of models. The number of estimates produced is high. The published text describes no correction for multiple comparisons; the supplementary appendix could not be consulted and might mention one. With intervals this narrow, the issue bears less on the main results than on isolated signals, of which the eating disorder exception is one. | |
| Causal inference | Association only |
| Finding No randomisation, no exogenous wealth shock exploited as a natural experiment. The authors claim no causal relation and write that definitive causal inference requires triangulation across multiple approaches. Their vocabulary is nonetheless more committed than that reservation: they speak of a fundamental social gradient, suggest that parental wealth “might play a role” in the later development of mental disorder, and call for policies targeting wealth inequality. Caution must therefore be maintained downstream: observing that wealth is associated with disorders does not license the claim that raising wealth would reduce their frequency, a question that belongs to another type of study. | |
| External validity | Transpose with caution |
| Finding The reasoning is worth taking exactly as the authors set it out, because it is counterintuitive. Norway combines dense social protection, access to care that depends little on household means and low income inequality, yet it ranks among the OECD countries with the highest wealth inequality. The authors first put forward their estimates as a lower bound for other countries, then immediately qualify this: that high wealth inequality and the existence of a fixed copayment for mental health services make the lower bound status unclear. Transposing the numerical magnitude to a health system where out-of-pocket costs weigh more heavily therefore has no empirical support. | |
| Independence and transparency | Low |
| Finding European public funding, no declared conflict of interest, open access publication, data sources described, registry access procedures set out, analysis code publicly deposited, ethical approval referenced. No product and no market is at stake in this result. The publication reports no pre-registration of the analysis plan, which is common practice in registry work of this kind but leaves open the question of how many analyses were run before those reported. | |
Level of evidence
What is demonstrated. In an entire national population observed from 2006 to 2023, there is a wealth gradient in the prevalence of diagnosed mental disorders in young people. The gradient is graded, it is found in almost every diagnostic category, and it persists, attenuated but clearly present, once what siblings share has been cancelled out. The precision of the estimates makes chance a very improbable explanation.
What is suggested. That wealth is a better marker of social position than income for this question. The argument rests on the attenuation of one association in the presence of a correlated variable, an argument that is statistically fragile by construction, and the authors half concede it by showing that the superiority of wealth owes much to its greater variability and that in standard deviation units its slope is clearly the steepest only in young adults. That the gradient steepens from childhood towards young adulthood: the point estimates rise, but the authors describe similar gradients between strata, and no explanation is tested. That eating disorders follow a reversed slope: the signal is small in magnitude and at least as compatible with differential access to diagnosis as with a real difference in frequency.
What is opinion. Our reading is that the within-family estimate, 1.34 in adolescents, is the most defensible figure in the paper, and that it is the one to quote in preference to the 1.83 of the multivariable population model. It is less striking. It stands up better to the most obvious objections. This ranking of what to quote is an editorial judgement, not a result of the study.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ One million four hundred thousand young people, eighteen years of registries, and it is the parents’ wealth that comes out, not their salary, including when you compare two children in the same family. Fine. But all I see there are the young people who consulted, and the families who consult least are precisely the ones we are talking about. So I take the direction, not the figure. And above all I note that the question I have been asking for fifteen years, what the parents do for a living, may not have been the most useful one. ”
What this means in practice: the social history gains one question, about the household’s reserves and its capacity to absorb an unforeseen expense. It does not gain an individual prognostic factor. Nothing here is calculated at a patient’s bedside, and no figure in this article applies to a particular case.
What you can do with this
- Add to the socioeconomic history a question about reserves rather than about income alone, phrased without intrusion: can the household meet an unforeseen expense, are there debts that weigh, is the housing stable. Declared income says nothing about that margin. The transposition here goes beyond what the study measures: its variable is gross wealth, which does not deduct household debt.
- Keep a full exploration of conduct problems, substance use and suicidality in young people from households without reserves. The point is not to announce a risk to a family, but not to cut those three areas of the interview short.
- Do not reverse the rule for eating disorders. The reversed gradient observed here is fragile and could rest in part on the use of services. An adolescent from a modest background is not protected, they may simply be identified less often.
- Bear in mind that the exposure is measured in early childhood, between 0 and 6 years, while the disorders are recorded up to twenty years later: the past situation of a family remains clinical information, even if it has improved since.
- Change no therapeutic indication on the basis of this work. It shifts a question in the interview, it touches neither prescriptions nor diagnostic criteria.
Frequently asked questions
Is a prevalence ratio of 1.34 between siblings a lot?
At the scale of one individual it is modest, and it supports no prediction for a given patient. At the scale of a population, an excess of 34% of diagnosed disorders across whole age bands represents a considerable volume of suffering and of demand for care. Both readings are true at the same time, and confusing them is the commonest error with this type of result.
Does the comparison between siblings prove causality?
No. It removes what siblings share, that is, part of the genetic background and the common family environment, which is already a great deal. It says nothing about what separates two children in the same family. They are born at different moments in their parents’ economic trajectory, and whatever moved that trajectory, a job loss or a separation for instance, also acts on them directly. This design controls shared familial confounding, it does not establish a cause and effect relation.
Why do eating disorders run the other way?
The study observes it, it does not explain it. Two hypotheses remain equally open: a genuinely higher frequency in well-off backgrounds, or faster ascertainment where people consult earlier and more readily. Since detection bias acts in the opposite direction for every other disorder, the second hypothesis deserves at least as much attention as the first.
Do these figures hold outside Norway?
The direction of the gradient, in all likelihood. Its magnitude, no. Norway has access to care that depends little on household means and low income inequality, yet its wealth inequality is among the highest in the OECD countries. The authors put their estimates forward as a lower bound for other countries, while acknowledging that this lower bound status is unclear, precisely because of that wealth inequality and of a fixed copayment for mental health services. Transposing 1.83 or 1.34 to another population would be an extrapolation without empirical support.
Should a family’s wealth really be asked about in consultation?
Not as an inventory, and never as a check. The useful question is functional: can the household absorb an unforeseen expense, does a debt weigh on daily life, is the housing secure. It is contextual information, on the same footing as the composition of the family, and it is collected with the same care and the same restraint.
Annotated bibliography
Source study. Ebeltoft JC, Ystrøm E, Ayorech Z, Eilertsen EM. Parental wealth and mental disorders in Norway (2006-2023): a nationwide registry-based study of 1.4 million young people. The Lancet Regional Health – Europe 2026; 62: 101567. Published online on 29 December 2025. DOI 10.1016/j.lanepe.2025.101567. PMID 41542029. PMCID PMC12803840. Open access publication. The article number 101567 stands in for pagination, the journal not paginating its articles continuously.
Background references. Each is given with what it contributes and what it does not allow anyone to conclude. The identifiers were verified on PubMed.
- Reiss F. Socioeconomic inequalities and mental health problems in children and adolescents: a systematic review. Social Science and Medicine 2013; 90: 24-31. DOI 10.1016/j.socscimed.2013.04.026. PMID 23746605. Contribution: establishes the consistency of the social gradient in childhood from several dozen studies. Limitation: major heterogeneity of the measures of social position and a predominance of income and education indicators, that is, precisely what the Norwegian work calls into question.
- Lund C, Brooke-Sumner C, Baingana F, et al. Social determinants of mental disorders and the Sustainable Development Goals: a systematic review of reviews. The Lancet Psychiatry 2018; 5 (4): 357-369. DOI 10.1016/S2215-0366(18)30060-9. PMID 29580610. Contribution: a conceptual framework of social determinants and a survey of the interventions evaluated. Limitation: a very aggregated level of synthesis, few results transferable to the individual level, and little treatment of the wealth dimension.
- Costello EJ, Compton SN, Keeler G, Angold A. Relationships between poverty and psychopathology: a natural experiment. JAMA 2003; 290 (15): 2023-2029. DOI 10.1001/jama.290.15.2023. PMID 14559956. Contribution: one of the few quasi-experiments in the field, an income supplement allocated exogenously to families, with an observed reduction in externalising symptoms. Limitation: a small sample, a very particular population, no comparable result on internalising symptoms, and an exposure bearing on income and not on wealth.
- Sariaslan A, Mikkonen J, Aaltonen M, Hiilamo H, Martikainen P, Fazel S. No causal associations between childhood family income and subsequent psychiatric disorders, substance misuse and violent crime arrests: a nationwide Finnish study of >650 000 individuals and their siblings. International Journal of Epidemiology 2021; 50 (5): 1628-1638. DOI 10.1093/ije/dyab099. PMID 34050646. Contribution: a direct methodological precedent on Finnish registries, with strong attenuation of the associations once siblings are taken into account. Limitation: outcomes centred on psychiatric disorders, substance misuse and arrests for violence, and an exposure bearing on family income and not on wealth.
- Kinge JM, Øverland S, Flatø M, et al. Parental income and mental disorders in children and adolescents: prospective register-based study. International Journal of Epidemiology 2021; 50 (5): 1615-1627. DOI 10.1093/ije/dyab066. PMID 33975355. Contribution: a direct point of comparison, on the Norwegian national registries and on a very similar population. Limitation: an exposure centred on income with no modelling of wealth, which leaves the question of the hierarchy between indicators entirely open. The two preceding works also reach opposite conclusions on whether the association is causal, a divergence that the authors of the study discussed here attribute to differences in population, in the definition of income and in design.
What was consulted. References verified on 12 August 2026 against the published version. The supplementary material could not be consulted. This analysis underwent an independent double reading.