HSCI 330 Lecture Notes - Lecture 9: Relative Risk, Confounding, Data Set
16 Jan 2019
School
Department
Course
Professor
Document Summary
Interaction considers what happens after we control for another variable. Interaction is present if the estimate of the measure of association differs at different levels of a variable being controlled. When assessing confounding and interaction in the same study, it is possible to find one with or without the other. In the presence of strong interaction, the assessment of confounding may be irrelevant or misleading. Confounding: compares the estimated effects before and after control. Dataset 1 there is clearly interaction, because the estimate for stratum 1 indicates no association but the estimate for stratum 2 indicates a reasonably strong association. There is clearly confounding, because any weighted average of the values 1. 02 and 3. 50 will be meaningfully different from the crude estimate of 6. 0. Data set 2 again there is clearly interaction, as in data set 1. However, it is not clear whether or not there is confounding.
