Viewpoint Expert opinions on trends and controversies in autism research.
Opinion / Viewpoint

Why inferring autism’s causes from epidemiology is dangerous

by  /  29 July 2014

Led astray: A study of autism in Africa looked only at people of high socioeconomic status, but some researchers mistakenly assumed that only this group is susceptible to the disorder.

In the past few years, I’ve seen several reports that suggest that health disparities in autism are related to social factors such as culture, race, ethnicity and socioeconomic status.

For example, claims abound that a mother’s ethnicity is associated with genetic differences that predispose her children to autism. Or that her country of birth means she was exposed to harmful environmental factors that may lead to autism in her children. Or even that the stress associated with immigration increases her risk of having a child with autism.

Most recently, a study published earlier this month reported that in Los Angeles County, children born to mothers who immigrated from Central or South America, the Philippines or Vietnam have an elevated risk of autism. It also concludes that African-American and Hispanic children have higher rates of autism than do Caucasian children1.

At first glance, these ‘just-so’ stories resemble well-established associations between some single-gene disorders and certain races or ethnic groups — for example, the link between Tay-Sachs disease and the Jewish population. But it is far more challenging to identify and explain such disparities in conditions such as autism, which result from multiple genetic and environmental risk factors.

While working on a World Health Organization-commissioned review of the global prevalence of autism, I came across many other just-so stories linking factors such as ethnicity, nativity and race to the prevalence of the condition2. But association does not imply causality, and prevalence data cannot be used to infer underlying genetic, biological or environmental differences.

Out of Africa:

I was particularly struck by a claim in the review that autism is a rare or even nonexistent condition in Africa. I was even more surprised to find the origins of this claim in a misinterpretation of Victor Lotter’s pioneering and insightful case series in Africa3.

A well-known epidemiologist in the 1970s, Lotter traveled to a number of African countries and described cases of autism that looked remarkably similar to what he had seen in his home country, the U.K.

Lotter is often misquoted as suggesting that there is a lower prevalence of autism in Africa than in the U.K., or that autism is associated with high socioeconomic status. He was in fact open about the fact that he had looked at only a small group of people of high socioeconomic status. He attributed this to the fact that these families were probably more likely than others to seek help in urban clinics.

As weput together the puzzle pieces of race, culture and biology as risk factors for autism, it is worth making sure that we aren’t perpetuating this type of misunderstanding. If we do, we risk repeating the past mistakes of social Darwinism, which attributes individual differences in intelligence, personality and cultural and social characteristics to a genetic basis.

For example, African-Americans as a group may score consistently lower on tests of intelligence than other ethnic groups, even after controlling for a wide range of social and economic confounds4. This has led some people to attribute these results to innate (genetic) differences between races. But prominent critics of this perspective have challenged not only the quality of the evidence on which these claims are based but also their underlying assumptions5.

“Mistakenly attributing racial differences [in autism risk] to biology offers a convenient excuse for political apathy.”

Complex social phenomena cannot be reduced to measurable concepts such as intelligence quotients. And the association of these measures with race does not imply causality. The lower test scores could be a result instead of biases in the tests themselves, which in the case of IQ tend to reflect the person’s ability to take tests in general. What’s more, the skills measured in these tests are by no means culturally universal.

Mistakenly attributing racial differences to biology offers a convenient excuse for political apathy, in lieu of sustained efforts to eliminate health and social disparities where possible.

In autism, many studies have similarly tried to link autism risk or prevalence to race, ethnicity and country of origin.

The stories often lead in two curious directions: Mothers take the lion’s share of the blame. And the increase in autism risk and severity is seen mostly in non-Caucasians, or those who are born outside the U.S. or Northern Europe.

In the past decade, social and advocacy pressures have revealed many disparities in autism prevalence. Indeed, in our global review we found that prevalence estimates are highly variable across geography and culture. Others’ findings suggest differences in severity of symptoms or in functioning across racial groups.

We and others have attributed this variability to a range of social factors that influence the measurement of autism prevalence. These include broadening of the diagnostic criteria of the condition, the rise in awareness, improved identification and stronger advocacy, alongside the many methodological differences in prevalence studies.

If differences in prevalence across diverse groups were truly the result of genetics, they would be immutable. Instead, prevalence estimates are highly variable and amount to snapshots within a given community at a certain time period.

False assumptions:

The most comprehensive evidence from the U.S. Centers for Disease Control and Prevention confirms that prevalence across racial groups in the U.S. is a moving target. The pattern of change suggests a ‘catch up’ in diagnosis in groups of individuals who were initially underdiagnosed. This makes prevalence estimates powerful advocacy tools to signal the unmet needs of various subgroups.

Further problems with claims about differences in prevalence relate to the validity of their constructs. For example, how is ‘foreign birth’ a biologically meaningful construct? Who or what is the person foreign to? Where did she come from? Did she choose to leave her home country or was she driven out by natural or political circumstance?

Similarly, U.S.-defined race categories are limited in capturing the complexity of individual differences both in biology and in culture. For example, ‘black’ encompasses African-Americans alongside immigrants from African countries who are a socioculturally distinct group. Neighboring Canada has dozens of government-recognized ethnic categories that are not collapsible into the U.S. categories, despite the overlap in ethnic origins between the populations of the two countries.

More often than not, ‘U.S.-born Caucasian’ is used as a reference group of convenience, rather than one that is logically or statistically justified. Rather than measuring individual differences, we tend to measure how different everyone else is from this ‘prototypical’ Caucasian Anglo-American or European group.

What’s more, disparities in access to care and clinician biases are documented phenomena that may be driving the reports of differences in prevalence. And the lack of a typical comparison group in most studies leaves open the possibility that tests used to measure IQ or language skills underestimate these skills in children from certain racial and ethnic groups.

If we accept that race and ethnicity are signs of biological differences in autism, this also opens to the door to using a child’s skin color to decide his or her prognosis or treatment. We should instead focus on making access to care more equitable for all children.

Rather than relying on frequency counts, studies investigating questions of culture and ethnicity need to formulate solid hypotheses based on well-grounded assumptions and the highest-quality data. For now, evidence suggests that where these disparities exist, they are unlikely to relate to underlying causes.

As with social Darwinism, our epidemiological just-so stories may inadvertently have socially and ethically questionable implications. The reality is that what underlies these stories is probably as complex as life itself.


1: Becerra T.A. et al. Pediatrics 134, e63-71 (2014) PubMed

2: Elsabbagh M. et al. Autism Res. 5, 160-179 (2012) PubMed

3: Lotter V. J. Child Psychol. Psychiatry 19, 231-244 (1978) PubMed

4: Herrnstein R. and C. Murray (1994) The Bell Curve: Intelligence and Class Structure in American Life New York: Free Press

5: Gould S.J. (1981) The Mismeasure of Man New York: W.W. Norton

8 responses to “Why inferring autism’s causes from epidemiology is dangerous”

  1. Kozmo says:

    ” …Mothers take the lion’s share of the blame… ”

    Isn’t this a rehash of the ‘Refrigerator Mother’ canard?

  2. passionlessDrone says:

    *Rather than relying on frequency counts, studies investigating questions of culture and ethnicity need to formulate solid hypotheses based on well-grounded assumptions and the highest-quality data.*

    At this time, we have neither well grounded assumptions nor C grade quality data, which in turn, leads me to wonder, why not spend our dollars and researcher time on a bioinformatic approach, instead of questions of culture and ethnicity? I know that such approaches give soft scientists the ability to get grants and all, but such studies do pretty much nothing if our goals are to help people with autism today, understand how to help tomorrows generation of children with autism tomorrow, and gain insight into what amount of the observed increase is real. Those are the questions we should be working towards.

    There problem with the mantras of ‘greater awareness’ and diagnostic shifting is that they have an impossible to disprove ceiling; it doesn’t matter if autism rates are found to be 1 in 100 or 1 in 4, the argument for ‘greater awareness’ can be trotted out without losing any of it’s ability to explain the entirety of the increase.

    Also, did I notice a comment get deleted? Of course, the comment was bunk and utility free, but I didn’t think wholesale comment removal was the SFARI way.

    • gregboustead says:

      Hi passionlessDrone,

      Thanks for your comment. As far as deleting comments here on, we do occasionally remove comments that are inflammatory, personal attacks, or simply obvious flaming or trolling attempts that don’t move the scientific conversation forward. Also flagged, edited, or removed are remarks that promote unsafe treatments or otherwise violate the ‘standards of behavior’ outlined in our Terms and Conditions.

      For our complete guidelines see:

      This is not something we enjoy doing, but we feel it’s important for promoting a constructive place where readers like yourself and scientists feel safe commenting.


    • Matt Carey says:

      “but such studies do pretty much nothing if our goals are to help people with autism today”

      If large populations of autistics are unidentified today, they are most likely not being supported appropriately. If we understand not only which populations are likely to be underserved, but also why, we can correct that.

      Also, should one wish to do the sort of epidemiology that could point to autism causes, one needs to have more accurate data. Are you giving up on that? Just because we don’t have that quality of data today doesn’t mean we won’t in the future.

      “why not spend our dollars and researcher time on a bioinformatic approach”

      Sounds like a false dichotomy. It would be interesting to see someone comment on whether we have bioinformatic data to work from that are of as good or better quality than the prevalence data of today.

      When did greater awareness and the rest become a “mantra”. Calling something a “mantra” doesn’t make it less true.

      If one takes the CDC MMWR data as an example, each one includes a sizable fraction of previously undiagnosed autistics. Recently about 30% of the estimate is previously undiagnosed. Few people are saying these factors “explain the entirety” of the increase, but it is clear and real that social changes are very real and still very much at play. That’s not “mantra”. That’s data.

  3. Matt Carey says:

    I am reminded of a webpage noting that the prevalence of autism in Kenya was officially zero.

    This was on a Kenyan autism group’s web page. The true prevalence wasn’t zero (hence the existence of an autism organization). There were no people in country qualified to diagnose.

  4. ASD Dad says:

    “culture, race, ethnicity and socioeconomic status”

    It is a reminder as is the rest of this article that Autism is a human condition observed by humans that have differing perspectives.

    How do we overcome such inadequate observations – by including the observations of others and of other perspectives. Thus a greater and more complex picture is begun and slowly a ‘truth’ may be arrived at for the benefit of all.

    Understanding and translating the complex aspects of Autism interactions even within the genetic, physiological and neurological data we have at this time demands at least a more than fundamental knowledge of a variety of scientific and medical disciplines.

    Even then a person needs to translate that knowledge with the help of a variety of sociological insights.

    For myself I am blessed to be part of diverse family of differing gender “culture, race , ethnicity and socioeconomic status” which is further diversified by genetics, physiology and neurology.

    Perhaps that accounts for my having a more flexible perspective and more empathic approach on some fundamental and controversial issues in autism.

  5. Anonymous says:

    I run a research support agency in Uganda. Last year, I was involved in a study that sought to identify disabled girl children in Kampala (and the nature of disability)- and children with autism numbered quite high. And as Matt Carey correctly asserts above, the ‘experts’ that were brought to train data collectors on identifying the different disabilities (the data collectors had no medical background), it soon downed on me that other than a few physical ‘signs’, they didn’t seem to have the depth of knowledge required of an expert. The inclusion of such people in studies unfortunately results in flawed findings, which is a shame really, given the high number of competent professionals in the field.

  6. ASD Dad says:

    Western protocols in mental health are not necessarily able to be exported to other cultures and societies. Even within Europe there is a raft of epidemiological information driven by differing strategies , protocols and administration some driven by an incorrect understanding of autism and the beliefs that flow from that philosophical position.

Leave a Reply

Your email address will not be published. Required fields are marked *