Critique of the Taylor et al. (2020) Study on Autism and Genetics

A study suffering from 8 fatal flaws is making the rounds.

Study Summary – What Did They Try to Do?

The authors wanted to know:

“Has the cause of autism changed over time?”

Specifically, they looked at whether genetic or environmental factors have become more or less important in explaining autism. To do this, they used data from twins born in Sweden between 1982 and 2008. The logic was this:

• Identical twins (monozygotic or MZ) share almost 100% of their DNA.

• Fraternal twins (dizygotic or DZ) share about 50% of their DNA, just like regular siblings.

So, if MZ twins are much more similar to each other on a trait (like autism) than DZ twins are, it suggests genes play a strong role.

They looked at:

• Diagnosed autism (based on medical records),

• Screening-level autism (based on a phone questionnaire with parents), and

• Autistic traits (measured continuously like a score, not just yes/no).

---

So What’s the Problem With This Study?

Below are eight major problems, explained clearly and with simple analogies or definitions.

---

1. They Confused Heritability With Cause

Heritability tells us how much of the variation in a trait (like autism scores) in a group of people can be statistically linked to genetics.

Important: Heritability does not mean how "genetic" a disease is.

Example: If everyone in a town eats the same food but only some people get sick (and those people are genetically more vulnerable), the illness will look highly heritable—even though the cause was environmental.

What They Did Wrong: The authors claim that because autism has a high heritability score (like 90%), that environmental causes can’t matter much. This is false logic.

---

2. They Used Oversimplified Diagnosis Categories

They turned autism into a yes/no variable—diagnosed or not.

That’s like asking: “Do you have anxiety—yes or no?” instead of rating your anxiety on a scale from 0 to 10.

Problem: This binary format hides important differences in severity, onset, and variation, and can’t pick up subtle effects of environment or time.

Even the authors admitted this:

“It would have been informative to examine the raw variance… which was not possible because of the use of a binary variable.” (p. 7)

---

3. They Downplayed Their Own Finding: Environmental Influence Increased

They actually found that environmental variance (difference caused by environment) went up over time in kids' autism-related traits.

But they buried this in the results and continued to argue that “genes matter most.”

That’s like seeing fire in the kitchen and saying, “Well, the house is mostly made of wood, so the match couldn’t have caused this.”

---

4. They Deleted Shared Environment From Their Model

In twin studies, there are usually three parts used to explain where differences come from:

• A = Additive genetics (heritable stuff),

• C = Common/shared environment (things twins experience together),

• E = Non-shared environment (unique experiences or measurement error).

They removed “C”, which includes things like:

• Pollution,

• Vaccines or medications,

• Home life,

• Prenatal exposures.

So they guaranteed they wouldn’t detect shared environmental effects.

This is a fatal flaw: if you remove the term that includes the most important environmental factors, how can you claim environment didn’t matter?

---

5. They Relied on a Weak Parent Phone Survey to Measure Traits

The screening test they used was a parent-reported phone interview. Its:

• Sensitivity was only 71% (it misses ~3 out of 10 kids with ASD),

• Specificity was better (96%).

Why it matters: Over time, parents may become more aware of autism or more likely to report traits, which can change results without any change in actual biology.

---

6. They Ignored Diagnostic Drift and Ascertainment Bias

Autism diagnosis rates have risen over the years. This is partly because:

• Definitions have broadened,

• Awareness has increased,

• More parents and doctors screen earlier.

But the study doesn’t control for that. It just treats the increase as if it needs no explanation.

That’s like measuring rising Google searches for “headache” and concluding headaches are increasing—without asking if people are just Googling more.

---

7. Their Models Were Built with Circular Logic

In some models, they forced the total variance to always equal 1.

That means they could only study relative proportions (like how much of the pie is genetic vs. environmental)—but not the absolute size of the pie.

So if the environmental “slice” grows, but the total pie also grows, you can miss real changes just because everything is still “in proportion.”

---

8. They Underreported Conflicts of Interest

One of the authors (Dr. Larsson) had personal ties to Shire/Takeda and Evolan—companies involved in psychiatric medication and neurodevelopmental disorders.

This wasn’t mentioned in the main text, only at the very end. When researchers make strong claims about what causes a condition, they should be fully transparent about industry ties.

---

So What Can We Learn?

This paper appears very technical and authoritative. It uses fancy models, big data, and a trusted journal name. But even well-designed studies can be misleading when they misuse tools, ignore context, or ask the wrong questions.

In short:

• They used a method that could never detect the environmental factors they claim don’t matter.

• They misinterpreted heritability.

• They buried evidence that contradicted their main point.

---

Key Terms to Remember

Final Thought

This study is often used to argue that “autism is genetic” and “environmental concerns are overblown.” But in truth, the study wasn’t designed to answer that question fairly, and it contains logical and statistical flaws that undermine its conclusions.

Science must stay humble—especially when lives, families, and children are at stake.

Taylor, M. J., Rosenqvist, M. A., Larsson, H., Gillberg, C., D’Onofrio, B. M., Lichtenstein, P., & Lundström, S. (2020). Etiology of autism spectrum disorders and autistic traits over time. JAMA Psychiatry, 77(9), 936–943. https://doi.org/10.1001/jamapsychiatry.2020.0680