Reliability (r²)
Each EBV comes with a reliability index (r²) that measures the confidence we can place in the estimate. The higher the reliability, the more accurate the EBV.
What Is Reliability?
Reliability r² is a number between 0 and 1 (or 0% and 100%). It is calculated from the PEV (Prediction Error Variance) and the genetic variance:
r² = 1 - PEV / σ²_a
- r² close to 0: little data available, the EBV is a provisional estimate
- r² close to 1: abundant data, the EBV is highly reliable
Reliability increases with the number of direct evaluations, the number of tested daughters, and the quality of pedigree information.
Interpretation
| r² | Interpretation | Typical data |
|---|---|---|
| < 0.20 | Low — provisional value | Few or no direct evaluations |
| 0.20 - 0.40 | Moderate | Some direct evaluations |
| 0.40 - 0.60 | Good | Direct evaluations + evaluated offspring |
| > 0.60 | High | Large evaluated offspring from multiple testers |
A queen with r² = 0.50 and EBV = 115 is more trustworthy than one with r² = 0.15 and EBV = 125. The second queen's EBV could shift significantly with new data, while the first is already well established.
Sicherheit (BeeBreed)
BeePass displays reliability in the same format as BeeBreed's Sicherheit (SI), the reference platform for genetic evaluation in German-speaking beekeeping. This enables direct comparison between the two systems.
Sicherheit is expressed as a percentage (e.g., SI = 45% corresponds to r² = 0.45). On BeePass, you'll find this value next to each EBV on the queen's detail page.
How to Improve Reliability
Several strategies can increase the reliability of your queens' EBV:
- More testers — Each tester who evaluates an F1 daughter adds information. The more independent testers in varied environments, the better.
- Complete pedigree — A well-documented pedigree (dam, grand-dam, sire group) allows the model to better connect queens to one another.
- Accurate measurements — For the hygiene test, using BeePass's built-in timer (status
exact) gives maximum weight in the computation. Times recalled from memory (declared) receive reduced weight. - Connected network — Using drones from known lineages (instrumental insemination) strengthens genealogical links and improves reliability across the entire network.
HYG 6h and HYG 24h traits are particularly sensitive to measurement timing quality. A time_quality = unknown (time not recorded) excludes the data point from genetic computation entirely. Time your readings to maximise their impact on reliability.
See also:
- Reading Your EBV — Interpreting breeding values
- BLUP — Overview — The computation pipeline
- Inbreeding — Genetic diversity and the relationship matrix