Figure 1.
(A) Schematic demonstrating an in-sample negative correlation (filled circles and dashed line) that does not exist in the underlying population (all circles) may be generated by selectively including individuals with lower levels of both exposure and outcome. (B) Simulated data showing how in-sample associations (dashed line) may not reflect population level associations (line) if selected individuals (crosses) are a specific subset of the population (triangles).