Abstract
Association is not causation. This is a well-known statement. In this connection, there is a definition: “Biases are present when there is a difference between association and causation”. The counterfactual model and directed acyclic graph (DAG), presented during Lesson 3, provide very useful tools for understanding and arranging biases. It is advisable to view confounding biases as those arising from common causes in DAG and selection biases as those arising from adjustment of common results in DAG. So far as information biases are concerned, understanding of non-differential misclassification is useful. When results are to be interpreted, it is essential to know the direction and extent to which a given bias causes deviation of the “estimated value” from the “true value”, that is, to examine whether a given bias involves overestimation (away from the null) or underestimation (toward the null), making use of 2 × 2 cross-tables.
