Loss of power by degree of heaping in covariables
Heaping or digit preference is frequently occurring in health research data as well as in deliberate data masking. We investigate the associated loss of power associated by degree of heaping by simulation and illustrate the effect with a real dataset. Moderate heaping leads to little power loss and may be tolerable. However, heaping in broad categories induces major power loss and should be avoided.
Authors
- Willi Sauerbrei (ORCID: https://orcid.org/0000-0002-6792-4123)
- Heiko Becher (ORCID: https://orcid.org/0000-0002-8808-6667)
- Matthias Becher (ORCID: https://orcid.org/0009-0007-1267-8854)
Institutions
- University of Freiburg (DE)
- University Hospital Heidelberg (DE)
- Humboldt-Universität zu Berlin (DE)
- Heidelberg Institute of Global Health (DE)
- Charité - Universitätsmedizin Berlin (DE)
Publication Details
- Journal
- BMC Medical Research Methodology
- Published
- 2026-10-17
- DOI
- https://doi.org/10.1186/s12874-026-03043-4
- Primary Topic
- Statistical Methods in Epidemiology
- Type
- article
- Field-Weighted Citation Impact
- 0.00