Parental Wealth and Mental Health

Does a change in parental income cause a change in the risk for offspring mental health?

Michel Nivard

Income and the mental health of your children

In the last couple of weeks two large and well executed studies looked at the (causal) relation between parental income and offspring mental health. These kinds of studies are a big deal if you are interested in preventing psychiatric disease and you view fiscal or other governmental policy, like Universal basic income (UBI) or raising the minimum wage, as vehicle to improve mental health in children. Obviously there are lots of reasons other then mental health to consider UBI or raising the minimum wage, this is not an evaluation of those proposed policies this is just a peak at whether we should expect those policies may have indirect positive effects on mental health. The studies I discuss are observational and natural experiments, direct experimental evidence (randomly assigning funds to parents to be, and waiting ±20 years to see whether their kids grow up mentally healthier) is expensive difficult and rare.

Both studies considered a large and representative population (most of Norway and most of Finland respectively) and both studies find a strong correlation between parental income and psychopathology in their children, both articles are even published in the same journal in the same month. However, the studies come to different conclusion as to whether the effect is causal, the Norwegian study suggests there is a small but meaningful causal effect of income on offspring mental health where the Finnish study concludes there is no evidence for a causal effect. I’ll go over both studies, their design their finding and try to discuss whether we can come to some sort of definitive conclusion as to whether parental income causally influences the risk of psychopathology in children. Ill then briefly discuss other older, but still solid, work that attempts to answer the same question.

Its important to note these studies consider income, so any conclusions should probably be drawn about income not other facets of social economic or cultural capital and their influence on offspring health

Parental income and mental disorders in children and adolescents: prospective register-based study

Jonas Minet Kinge and colleagues published their article, (read it here it’s open access), on May 11th 2021 in the international journal of epidemiology. Their sample consists of 1.35 Million Norwegian children aged 5 to 17 measured between 2008 and 2016, from this they select two samples to study 969K children form Norwegian born parents and a sample of 5698 international adoptees. Trough national registries (more on those later) of primary healthcare, hospital records and specialist clinics the identify children who suffer from mental disorders (and are treated), they look both at specific disorder (e.g. bipolar disorder, depression, conduct disorder) and at an overall metric of “any disorder”. They then use logistic regression to relate the disorder(s) to parental income percentile they also perform analyses in which they adjust for: “geographical indicator variables, mother’s age, father’s age, number of household members, mother and father employment, single-parent household, parental education and parental mental disorders.”. With these adjustments in place they find a strong relation between parental income and childhood diagnoses Their figure 1 (reproduced here) is fairly clear, and adjusting for reasonable confounders didn’t remove the effect.

Kinge Figure 1 observed prevalence of any disorder per income percentile

But, as always there are many unmeasured or simply unknown confounders that could play a role, not the least of which is the fact that parents and children are related, and that both mental disorder as well as success in school are heritable (and school in turn influences income…) and therefrom could be jointly transmitted from parent to child, without there being a causal relation of income. Kinge and colleagues did some clever follow up analyses, they study the relationship between parental income and childhood mental health international adoptees, this analysis bypasses the genetic relationship between parent and child, which could be a source of confounding. The effect persists in the adoptees, though it was strongly attenuated for adoptees every decile increase in income was related to a .25% decrease in diagnoses (any diagnosis) where for the biological children a decile increase in income was related to a 0.66% decrease in the risk for (any) diagnosis, the attenuation is clear in their Figure 7.

Kinge Figure 7: Mental health in adoptees and biological offspring conditional on income decile

They conclude their controls (parental mental disorders, genetic confounding, socio-demographic factors) did not explain the observed association, increasing the likelihood that parental income has a small causal effect on childhood mental health.

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

Sariaslan and colleagues come to a different conclusion, their article is open access as well so you can read it here. This study studies 650k Finish individuals born between 1986 and 1996 and folloewd up from their 15th birthday until the either migrate, died where diagnosed or reached the study period at the end of 2017, so these people between 15 and 35-45 years of age, which is a difference with the Norwegian study which studies 6-17 year olds. The outcomes considered are severe mental disorders, depression, anxiety, the study also considers substance abuse and arrest for drug related crime. Like the Norwegian study the authors find a correlation between parental income and offspring psychopathology a ±9% reduction in risk for severe mental health, depression or for each $15K increase in parental income at age 15 they consider income at various ages, or cumulative income with similar results. If they control for a set of measured family confounders their results are attenuated but still substantial (±4-5% risk reduction for 15K). However, as we discussed not all confounders are or can be measured! Sariaslan and colleagues proceed to consider sibling pairs where they regress mental health status on differences in income between different siblings born into a single family, keeping genetic confounding and other social processes share by siblings constant but leveraging natural fluctuations in income within a family as exposure. Within sibling regression suggests no effect of income on childhood mental health (I include their Figure 1).

Sariaslan Figure 1: attenuated effect of income on mental health within sibling pairs

Obviously, the variation in income within a family is smaller than within the population, and even temporary income changes of considerable magnitude could be smoother out by savings or other factors (Scandinavian countries have good welfare systems). The authors acknowledge this and consider cousin pairs, where the parents are related, partly controlling for genetic confounding and other unmeasured shared influences, in the cousin analysis they observe a small but significant effect (±4-5% risk reduction for each 15K increase in parental income). The one additional analysis I would have loved to see here is cousin analysis adjusted for measured parental confounders, since both attenuate the relation partially, do both combined attenuate it fully?

What have we learned?

Based on these two (excellent, large and representative) studies we could conclude there is a small, or no effect of parental income on childhood mental health. Both studies attempt to mitigate the role of confounding factors, genetic or otherwise, the sibling analysis probably does so most comprehensively but also restrict the range of income variation most severely. I’d love to see a cousin analysis where the parents of the cousin pair are monozygotic twins, which would increase control for genetic confounding. The adoption design is clever, but the number of adoptees (5500) is far smaller sample, and adoption usually has some conditions (though these may have been lax in the past) which may mean a house visit by a social worker or other checks and procedures. Lower income adoptive parents may have to have other qualities that convince authorities they are able to adopt, this would introduce a form of selection that influences results. Finally adoptees in general have a far higher burden of mental health issues possibly related to their early childhood or to the adjustment/adoption itself, it’s hard to fully think trough how that would impact results.

A further difference between the studies is the different age range in which offspring diagnoses are measured, the Norwegian sample is diagnosed during childhood, while the finish sample is diagnosed after age 15, perhaps parental income has an influence that wanes with age Other than the technical aspects we should consider the nature of the sample these results will likely be generalizable to other strong Scandinavian style welfare states but a low income in Scandinavia is considerably higher than in other countries and social services exist to mitigate risks for low income families where in other countries they do not. The Finish sample considers anyone born in Finland the Norwegian sample further restricts thee sample to parents being born in Norway (possibly due to data availability?), which means the results are likely not representative of newly immigrated Norwegians.

Further evidence

Various people with whom I discussed the paper on twitter (Pietro Biroli, Edwin Dahlmaijer and Peter Barr) suggested further reading so I’d like to briefly highlight two other studies, again one positive finding and a null finding. Cesarini and colleagues studied and wrote about the children of Swedish (again with the great data….) lottery winners. Lottery winners are a selected bunch but a reasonable natural experiment for a sudden single income boost. They find suggestive effects on offspring obesity but no effect on mental health. The sample size is again rather large but by far the largest group win a modest amount (< 10K) but then again the policy changes to income like minimum wage increase or UBI won’t make anyone a millionaire either. The design is a good complement to the other two studies, but I can’t shake the feeling we are learning only about Scandinavia (which means we need to improve our data elsewhere…). So let me finally discuss work by Costello and colleagues they study a sample of 1420 rural children that were psychiatrically assessed annually, the sample was 25% American Indian and 75% white, midway through the measurement of these children a casino opened that provided American Indian parents with an income supplement which moved 14% of American Indian families out of poverty but had no effect on white families. Psychiatric symptoms dropped among those who were lifted out of poverty but remained unchanged among those that were not, effects where limited to conduct and oppositional deviant disorders. Obviously, this design isn’t a perfect experiment either (someone should do one) and considers different outcomes and a subtly different exposure but it may have one thing going for it, it specifically studies the group you would want to target with a policy change, those that are currently poor.

Where do we go from here?

Collectively the current body of evidence shows the population average causal effect of income (or winnings) on offspring mental heath are far smaller than the correlation between the two and modest in magnitude at most. This shouldn’t be a reason to reject policy initiatives that boost income, as income can have plenty of other highly valued causal consequences. Now because this is a blog and not a comprehensive review you should definitely consider the underlying papers carefully and seek out further evidence before you reach any consequential conclusions.

Ideally we’d see adversarial collaboration between the authors of the recent papers and i’d hope they would specifically look into changes in income that lift people out of poverty as I don’t believe income changes are expected to have the same effect at the lower and higher end of the income distribution. Raising income to low income families may do more then raising income for high income families, and I’d really like to see a careful attempt at getting the conditional causal effect of income change across various initial income levels. Ideally, I’d love to see other creative within family design such as the within cousin effect of timing of inheritance (before after the kids leave the home) on childhood mental health.