The Check Mark Pattern
Abstract
We exhibit two studies (one epidemiological and one clinical), both with apparently paradoxical findings characterized by group (index versus control) similarity on the dependent (health) variable (Y) means, a significant group difference on the independent variable (X) means (index mean greater than the control mean) and a positive correlation between Y and X in the index group, causing index subjects with low values of X to have a lower Y mean than the controls and index subjects having high values of X to have a higher Y mean than the controls. This pattern has been called the "check mark" pattern. We predict this pattern using a linear model and use the model to estimate exposure effects in the epidemiologic study. Additionally, we show that a previously published study of the check mark pattern suggesting reverse causation is incorrect.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2006
- Accession Number
- ADA451631
Entities
People
- Gary Henriksen
- I. J. Russell
- Joel E. Michalek
- Pandu Kulkarni
- Ram C. Tripathi
- Suojin Wang
Organizations
- Air Force Research Laboratory