A study finds that two traits are negatively associated among selected participants. It may be tempting to conclude that one trait undermines the other. First ask how people entered the sample.
Berkson’s bias is a classic example of selection creating a misleading association. The research article Berkson’s bias, selection bias, and missing data explains its relationship to conditioning on a common effect, often called a collider.
A fictional admissions example
Imagine a workshop that admits applicants who score highly in either drawing or coding. In the full applicant pool, suppose the two skills are unrelated. Among admitted applicants, someone with a low drawing score must have a high coding score to qualify; someone with a high drawing score can qualify with a wider range of coding scores.
The selection rule can therefore make the skills appear negatively associated within the admitted group. It does not show that learning to draw makes a person worse at coding. The story is invented to explain the mechanism, not evidence about real applicants or either skill.
A simple causal sketch has two arrows pointing toward admission: drawing influences admission, and coding influences admission. Restricting analysis to people admitted can create a relationship that was not present in the wider pool.
Read the entry rule as part of the evidence
When a news report describes a study, identify who was eligible, who participated and what conditions determined inclusion. Ask whether the variables being compared could both influence entry into the dataset.
This does not mean every selected sample produces the same bias, or that a larger sample automatically fixes it. The problem concerns the selection mechanism and the causal question.
Look for the researchers’ explanation of those issues and the limits they place on interpretation. A careful headline describes the population actually studied while avoiding a causal story that the selection process may have manufactured.

