Matching, Standardization, Propensity Scores & Sensitivity Analysis
Matching ensures comparison groups are balanced on known confounders before data collection. Essential in case-control studies.
Match each Case to 1+ Controls with identical confounder values (e.g., 50yo Male Smoker matched to 50yo Male Smoker).
Requires Conditional Logistic Regression!
Ensure the distribution is similar at the group level (e.g., 30% Cases and 30% Controls are aged 50-60).
Comparing rates between populations with different age/sex structures.
| Method | starting Point | Output |
|---|---|---|
| Direct | Study population rates | Age-adjusted Rate |
| Indirect | Standard population rates | SMR (Observed/Expected) |
The Propensity Score (PS) is the probability of being exposed given measured confounders. It collapses high-dimensional data into one scalar.
Match individuals with similar probabilities. Standard caliper: 0.1 - 0.25 SD of logit PS.
Inverse Probability Treatment Weighting. Reweights sample to create a pseudo-population where exposure is independent of covariates.
"How strong would an unmeasured confounder need to be to explain away my observed effect?"
Example: If observed RR = 2.5, E-value = 4.44. A hidden confounder needs RR ≥ 4.44 with both exposure and outcome to nullify the result.