Understanding Statistical Paradoxes in Epidemiology
Author Information
Author(s): Tu Yu-Kang, Gunnell David, Gilthorpe Mark S
Primary Institution: University of Leeds
Hypothesis
How do Simpson's Paradox, Lord's Paradox, and suppression effects manifest in epidemiological research?
Conclusion
The study illustrates that statistical paradoxes can lead to contradictory interpretations of epidemiological data.
Supporting Evidence
- Simpson's paradox shows that relationships can change when controlling for other variables.
- Lord's paradox indicates that adjusting for a covariate can reverse the relationship between variables.
- Suppression effects can increase model fit even when a covariate is uncorrelated with the outcome.
Takeaway
Sometimes, when you look at data in different ways, it can seem like two things are related when they really aren't, or the opposite can happen.
Methodology
The article uses hypothetical scenarios and statistical analyses performed within SPSS 15.0.
Potential Biases
Potential for unobserved confounders affecting the results.
Limitations
The study relies on hypothetical data and may not account for all real-world complexities.
Participant Demographics
1000 adult males in England.
Statistical Information
P-Value
p<0.05
Confidence Interval
95% CI: 1.12, 3.87
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website