Oh god, Cremiuex, you get so into the statistical weeds here! I had a pretty solid education in basic social science stats, but it did not cover a number of things you talk about here: E-value, propensity weighting, probabilistic sensitivity analysis, negative control analysis,"stabilized inverse probability of treatment weighting. I could look up each of these terms and follow your train of thought, but life's too short. I asked GPT to summarize your train of thought without the use of arcane statistical terms, and the result is here: https://chatgpt.com/share/68e5b82a-9be4-8008-81f5-a158ee07131a
The text of what it wrote is below. It also included a nice graph at the end of the decline of the estimate of autism risk from tylenol step by step, starting with the baseline results and declining step by step to 0 as you applied de-confounding stats one by one. You should maybe check the text to see whether you think it's correct. If it is, maybe consider including something like this if a post is very technical.
***************
1. Study Basics
Researchers in Japan studied over 180,000 mothers and more than 200,000 of their children.
They looked at whether acetaminophen (Tylenol) use during pregnancy was linked to later autism diagnoses in the children.
Acetaminophen use was identified from prescription records, which in Japan cover most use since pregnant women rarely take over-the-counter (OTC) medications.
2. Initial (“Crude”) Results
At first glance—without adjusting for anything—mothers who used Tylenol during pregnancy seemed more likely to have children later diagnosed with:
ADHD: +34% risk
Autism: +17% risk
Intellectual disability: +20% risk
All these differences were statistically significant at that first step.
3. Adjusting for Confounding Factors
Once the researchers took into account other differences between mothers who did and did not use Tylenol—such as their health history, medication habits, and pregnancy characteristics—the apparent effect shrank sharply:
Autism: only +8% increased risk, and still statistically significant but small.
In other words, after accounting for known background differences, most of the earlier apparent risk diminished.
4. Deeper Control Methods
The authors then applied two stronger statistical balancing methods:
Propensity-score matching (PSM)
This method pairs each Tylenol-using mother with a very similar non-user (same age, health profile, etc.), as if creating “twins” in different exposure groups.
Result: autism risk only +7%, not statistically significant.
Inverse probability of treatment weighting (IPTW)
This technique gives more weight to people who are under-represented in either group so that, overall, the exposed and unexposed groups become balanced on all measured characteristics.
Result: autism risk +9%, barely significant.
So across these methods, the autism link kept weakening and usually lost statistical significance.
5. Checking Whether Small Biases Could Explain Everything
The authors then tested how strong any unmeasured confounding would have to be to create a false appearance of risk.
They found that:
Even a very small hidden bias (roughly a correlation of 0.06–0.09 between Tylenol use and autism) could fully explain the tiny remaining effect.
That means the apparent link could easily vanish if even a minor unnoticed factor (such as cautious mothers avoiding all drugs, or doctors’ differing prescribing habits) were accounted for.
6. “Negative Control” Checks
They compared autism risk when mothers used similar drugs:
Before pregnancy: Tylenol use was actually linked to lower autism risk (–7%).
After pregnancy: slightly higher risk (+5%), not significant.
Other pain relievers (NSAIDs, aspirin): small or inconsistent increases.
Because these patterns didn’t make biological sense (Tylenol before pregnancy couldn’t cause autism), they suggested that the remaining differences were due to bias, not causation—probably differences in which mothers take or avoid medications at various times.
7. Misclassification of Exposure
They also noted that many small errors in measuring Tylenol use could falsely inflate the apparent risk.
For instance, some pregnant women might take OTC Tylenol without it appearing in prescription data, blurring the line between “users” and “non-users.”
Simulations showed that even modest amounts of such misclassification would make the risk seem higher than it really is.
8. Sibling Comparisons
Finally, they compared siblings—children from the same mother, where one was exposed to Tylenol in utero and the other was not.
This design cancels out all family-level factors (genes, home environment).
Result: the autism link disappeared or reversed direction, confirming no real effect of Tylenol itself.
9. Overall Reasoning and Conclusion
All methods converged on the same picture:
Stage of Analysis Apparent Autism Risk Interpretation
Crude comparison +17% Looked risky, but unadjusted
Adjusted for basic factors +8% Small, borderline
Matched/weighted samples +7–9% Not statistically significant
Sibling comparison 0% or reversed No causal link
Because:
even tiny hidden biases could erase the effect,
results were inconsistent across timing and control drugs,
and sibling comparisons showed nothing,
the author concludes that Tylenol use during pregnancy does not increase autism risk at all—and earlier studies suggesting otherwise were probably misled by small biases and measurement errors.
The putrid drug tylenol is being used as the autism scapegoat. There is no way they want anyone to believe or tie the use of treacherous vaccines into any debilitating illnesses. The ghoulish vaccine empire will be protected at all costs...even at the expense of millions of children who they don't want around anyway in the name of depopulation.
You can tell Cremieux is smart and skeptical and doesn’t give a shit what he’s *supposed to* think, right? Ok, so the same smart no-nonsense guy who wrote the present article also posted this, about vaccines: https://x.com/cremieuxrecueil/status/1915828316969329139
I wouldn't refer to X (reduced space) when you can refer to substack. Here's the substack review paper. I find it very reasonable. Finally, I saw the comparison in the arrythmia rates for vaccines vs the infection with Wuhan Corona.
Thanks for the conformation and clarification.
Oh god, Cremiuex, you get so into the statistical weeds here! I had a pretty solid education in basic social science stats, but it did not cover a number of things you talk about here: E-value, propensity weighting, probabilistic sensitivity analysis, negative control analysis,"stabilized inverse probability of treatment weighting. I could look up each of these terms and follow your train of thought, but life's too short. I asked GPT to summarize your train of thought without the use of arcane statistical terms, and the result is here: https://chatgpt.com/share/68e5b82a-9be4-8008-81f5-a158ee07131a
The text of what it wrote is below. It also included a nice graph at the end of the decline of the estimate of autism risk from tylenol step by step, starting with the baseline results and declining step by step to 0 as you applied de-confounding stats one by one. You should maybe check the text to see whether you think it's correct. If it is, maybe consider including something like this if a post is very technical.
***************
1. Study Basics
Researchers in Japan studied over 180,000 mothers and more than 200,000 of their children.
They looked at whether acetaminophen (Tylenol) use during pregnancy was linked to later autism diagnoses in the children.
Acetaminophen use was identified from prescription records, which in Japan cover most use since pregnant women rarely take over-the-counter (OTC) medications.
2. Initial (“Crude”) Results
At first glance—without adjusting for anything—mothers who used Tylenol during pregnancy seemed more likely to have children later diagnosed with:
ADHD: +34% risk
Autism: +17% risk
Intellectual disability: +20% risk
All these differences were statistically significant at that first step.
3. Adjusting for Confounding Factors
Once the researchers took into account other differences between mothers who did and did not use Tylenol—such as their health history, medication habits, and pregnancy characteristics—the apparent effect shrank sharply:
Autism: only +8% increased risk, and still statistically significant but small.
In other words, after accounting for known background differences, most of the earlier apparent risk diminished.
4. Deeper Control Methods
The authors then applied two stronger statistical balancing methods:
Propensity-score matching (PSM)
This method pairs each Tylenol-using mother with a very similar non-user (same age, health profile, etc.), as if creating “twins” in different exposure groups.
Result: autism risk only +7%, not statistically significant.
Inverse probability of treatment weighting (IPTW)
This technique gives more weight to people who are under-represented in either group so that, overall, the exposed and unexposed groups become balanced on all measured characteristics.
Result: autism risk +9%, barely significant.
So across these methods, the autism link kept weakening and usually lost statistical significance.
5. Checking Whether Small Biases Could Explain Everything
The authors then tested how strong any unmeasured confounding would have to be to create a false appearance of risk.
They found that:
Even a very small hidden bias (roughly a correlation of 0.06–0.09 between Tylenol use and autism) could fully explain the tiny remaining effect.
That means the apparent link could easily vanish if even a minor unnoticed factor (such as cautious mothers avoiding all drugs, or doctors’ differing prescribing habits) were accounted for.
6. “Negative Control” Checks
They compared autism risk when mothers used similar drugs:
Before pregnancy: Tylenol use was actually linked to lower autism risk (–7%).
After pregnancy: slightly higher risk (+5%), not significant.
Other pain relievers (NSAIDs, aspirin): small or inconsistent increases.
Because these patterns didn’t make biological sense (Tylenol before pregnancy couldn’t cause autism), they suggested that the remaining differences were due to bias, not causation—probably differences in which mothers take or avoid medications at various times.
7. Misclassification of Exposure
They also noted that many small errors in measuring Tylenol use could falsely inflate the apparent risk.
For instance, some pregnant women might take OTC Tylenol without it appearing in prescription data, blurring the line between “users” and “non-users.”
Simulations showed that even modest amounts of such misclassification would make the risk seem higher than it really is.
8. Sibling Comparisons
Finally, they compared siblings—children from the same mother, where one was exposed to Tylenol in utero and the other was not.
This design cancels out all family-level factors (genes, home environment).
Result: the autism link disappeared or reversed direction, confirming no real effect of Tylenol itself.
9. Overall Reasoning and Conclusion
All methods converged on the same picture:
Stage of Analysis Apparent Autism Risk Interpretation
Crude comparison +17% Looked risky, but unadjusted
Adjusted for basic factors +8% Small, borderline
Matched/weighted samples +7–9% Not statistically significant
Sibling comparison 0% or reversed No causal link
Because:
even tiny hidden biases could erase the effect,
results were inconsistent across timing and control drugs,
and sibling comparisons showed nothing,
the author concludes that Tylenol use during pregnancy does not increase autism risk at all—and earlier studies suggesting otherwise were probably misled by small biases and measurement errors.
Wow! This is exactly how science should be, reproducing results across continents.
Poor Harvard Public Health Epidemology School - another Claudine Gay?
The putrid drug tylenol is being used as the autism scapegoat. There is no way they want anyone to believe or tie the use of treacherous vaccines into any debilitating illnesses. The ghoulish vaccine empire will be protected at all costs...even at the expense of millions of children who they don't want around anyway in the name of depopulation.
You can tell Cremieux is smart and skeptical and doesn’t give a shit what he’s *supposed to* think, right? Ok, so the same smart no-nonsense guy who wrote the present article also posted this, about vaccines: https://x.com/cremieuxrecueil/status/1915828316969329139
I wouldn't refer to X (reduced space) when you can refer to substack. Here's the substack review paper. I find it very reasonable. Finally, I saw the comparison in the arrythmia rates for vaccines vs the infection with Wuhan Corona.
https://www.cremieux.xyz/p/covid-vaccines-and-arrhythmia-surely
"why would you do controls for something like this anyway?"
To attempt to eliminate confounding.
"hen even with controls there's a statistically significant correlation between Acetaminophen use during pregnancy and autism + ADHD."
Not in the causal models.
"Yet somehow you're posting this as proving the opposite?"
Every causal model's central estimate was below 1.
"And I've read the whole post and can't tell why you're coming to this obviously wrong conclusion from data that disagrees with you?"
It does not. You have no idea what you're talking about.
"Is this the same guy that plagiarized dynomight and never admitted to it?"
I've never plagiarized anything, but because you came into this comment section and lied, you are banned for life.