confounding
When a difference in outcomes between groups is caused by factors other than the treatment being studied because those factors also differ systematically between the groups.
As defined in Adam Cifu on Ending Medical Reversal
Where it comes up · 7
- Adam Cifu on Ending Medical Reversal3 claims
- Emily Oster on Pregnancy, Causation, and Expecting Better2 claims
- Robert Aronowitz on Risky Medicine1 claim
- Jonathan Skinner on Health Care Costs, Technology, and Rising Mortality1 claim
- John Ioannidis on Statistical Significance, Economics, and Replication1 claim
- Dr. Peter Attia - The Portfolio to Live Longer - [Invest Like the Best, EP.325]1 claim
- 165 | Kathryn Paige Harden on Genetics, Luck, and Fairness1 claim
Most important claims · 10 of 10Search all claims for “confounding” →
- Looking at the evidence is always a good idea, but the evidence is almost always murkier than advocates on either side concede; numbers rarely speak for themselves because of questions of interpretation, leaps of faith in measuring variables, and confounding effects from unmeasured variables—and it is hard to stay bias-free even when one prides oneself on skepticism.Robert Aronowitz on Risky Medicine
- Observational comparisons of smokers and nonsmokers cannot fully correct for confounders such as diet and exercise, so the decimal precision in hazard ratios is false and the figures should be rounded.Dr. Peter Attia - The Portfolio to Live Longer - [Invest Like the Best, EP.325]
- Controlling for confounders with statistics has a place, but you never know if you've fully adjusted for them; even sophisticated researchers controlling for a dozen-plus variables in the Nurses' Health Study couldn't capture how different the groups were, so observational results agree with RCTs only about 80% of the time and you can't tell when they're misleading.Adam Cifu on Ending Medical Reversal
- For questions where the noise is many orders of magnitude larger than the signal—such as single nutrients linked to disease outcomes—observational designs should be abandoned because they cannot recover such tiny effects; randomized or experimental trials minimizing confounding are needed, and continuing observational studies is largely a waste that may bias our thinking rather than provide preliminary insight.John Ioannidis on Statistical Significance, Economics, and Replication
- Observational studies are natural experiments where groups self-select or are selected into treatments, so the groups differ in many ways besides the intervention; in the estrogen case, women who took estrogen were younger, thinner, had better cholesterol and drank more, so their better outcomes likely reflected those traits rather than the estrogen itself.Adam Cifu on Ending Medical Reversal
- Estrogen replacement therapy was widely prescribed to postmenopausal women based on observational data (the Nurses' Health Study) and a plausible biochemical rationale, but a good randomized controlled trial later showed it does not reduce cardiovascular events and may actually increase event rates in the first couple of years.Adam Cifu on Ending Medical Reversal
- Because messy free-range humans make experiments hard and everything correlates with everything else, observed environmental correlations (e.g., families who eat dinner by 5:30pm have kids who do better in school) don't prove intervention would help; genetics offers another way to compare people similar on a measured capacity but in different environments, helping identify the most promising environmental levers for change while getting genetic confounds out of the way.165 | Kathryn Paige Harden on Genetics, Luck, and Fairness
- The observed correlation between coffee consumption and miscarriage is likely spurious because nausea is a sign of a healthy pregnancy and reduces coffee consumption; women who aren't nauseous both drink more coffee and miscarry more, so not-being-nauseous, not the caffeine, may drive the apparent effect.Emily Oster on Pregnancy, Causation, and Expecting Better
- The European moderate-drinking studies are credible precisely because women who drank in moderation look similar to abstainers on observables and because the same studies do detect worse outcomes among heavy drinkers, demonstrating the studies have enough power to find an effect when one exists.Emily Oster on Pregnancy, Causation, and Expecting Better
- Analyzing terabytes of Medicare claims data turns up associations—such as two drugs together correlating with bone fractures—but expert clinicians reveal these are not causal, because frail elderly patients are prescribed those drugs precisely because they are already likely to fall, so Big Data findings remain an art requiring clinical judgment to separate confounding from causation.Jonathan Skinner on Health Care Costs, Technology, and Rising Mortality