Review: RCTs

Observational study


Association is not causation

Aside from smoking, how else might smokers and nonsmokers be different?
  • US adults who are smokers: 18.6% of men, 14.2% of women
  • Men are more likely to get heart disease for other reasons
  • Gender (or possibly sex) is a confounding variable
How can we fix this problem in the study?
  • Compare smokers who are (wo)men to nonsmokers who are (wo)men
  • This is called “controlling for” the confounding variable

Berkeley graduate admissions sex bias


More examples


Clofibrate drug to prevent heart attacks, 5-year death rate

Treatment Placebo
Adherers 15% 15%
Non-adherers 25% 28%
Total 20% 21%
Do RCTs protect from confounding? How?