Understanding Confounding Bias in Experiments
Author Information
Author(s): Bulbulia Joseph A.
Primary Institution: Victoria University of Wellington, Wellington, New Zealand
Hypothesis
Does randomisation eliminate all systematic causes of treatment assignment and outcome?
Conclusion
Confounding biases can still occur in randomised experiments even when randomisation is successful.
Supporting Evidence
- Randomisation does not prevent confounding bias from post-treatment adjustments.
- Causal directed acyclic graphs help identify sources of bias in experimental designs.
- Confounding biases can arise even in well-designed randomised controlled trials.
Takeaway
Even when experiments are set up correctly, there can still be hidden problems that affect the results. It's important to be careful about how we design and analyze these experiments.
Methodology
The paper uses causal directed acyclic graphs to illustrate sources of confounding bias in randomised controlled trials.
Potential Biases
Post-treatment selection and conditioning can introduce confounding bias.
Limitations
The examples presented do not cover all potential biases in causal inference.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website