Lecture 8 · Interactive explanation
What does a hazard ratio scale with?
The lecture’s model reports HR 1.04 for a one-unit NO₂ contrast. Under its log-linear form, the ratio for a contrast Δ is 1.04Δ.
This is a model-implied comparison, assuming the fitted relationship is appropriate for this contrast. It is not an absolute probability change and it is not an independent causal result.
What information is missing for absolute risk?
You would need baseline survival/hazard over a specified time and relevant population/model information. HR 1.04 alone does not tell you what percentage of people will die. A hazard concerns the event rate among people still at risk, not cumulative probability.
Keep the two modelling jobs separate
Land-use regression predicts concentration at addresses. The Cox model relates assigned exposure to mortality and includes adjustment variables. Uncertainty and confounding can enter both the measurements and the interpretation.