The reliability of genetic evaluations across separate management units (e.g. flocks) depends on the extent of genetic links or connections among animals in these units. Where poor connectedness exists, comparisons of estimated breeding values (EBV) across units may be biased. The objectives of this study were to identify breeding strategies to increase connectedness among units, to evaluate statistics that may reflect the reduction in bias as connectedness increases in such strategies, and to assess levels of connectedness in Suffolk and Targhee flocks participating in the National Sheep Improvement Program.

Expectations of bias when production units have different genetic means were derived for a simple sire model. These expectations were applied to data involving two flocks of animals with three different types of connections: sharing of a common reference sire or use of either a full- or half-sibling sire in each flock. Bias decreased as numbers of progeny in each flock increased for all methods. Linking through a reference sire was most effective and was the only method that eliminated bias as progeny numbers became infinite.

Pedigree and performance data on a single trait with heritability 0.25 were then simulated for 15 flocks with 40 to 140 ewes per flock. Each flock was simulated with a different founder genetic mean to introduce bias into the genetic evaluation. Flocks participated in sire referencing schemes by artificial insemination, with varying levels of participation, or by natural service. With sire referencing genetic gain was higher and inbreeding was lower than without, and bias was rapidly reduced to near-zero levels. Discontinuing the schemes led to lower genetic gain, but bias was not reintroduced. The prediction error correlation of flock genetic means was proposed as a connectedness measure because it was strongly associated with bias. Benchmarks of 0.05 and 0.10 for "good" and "superior" connectedness were established.

Targhee flocks have increased connectedness across the breed by actively exchanging rams over 15 yr. In the Suffolk breed, connectedness has only improved within segregated clusters of flocks. Suffolk breeders need to engage in active ram exchange to decrease risk of biased across-flock EBV comparisons.