8.4: Accuracy of the breeding value: the basic concept

Also in case of a high regression coefficient there are still some animals that have an EBV that is higher or lower than the TBV. If we would be able to estimate the breeding value with 100% accuracy, the EBV and the TBV would be the same value. If we would plot the TBV against the EBV then all data points would be perfectly in line. The less the data points are in line, the less certain you are that the EBV indeed is representing the true breeding value: the estimations are not accurate. A measure for data points being in line, and thus the accuracy of the breeding value estimation, is the correlation. If the correlation between estimated and true breeding values is 1, then you have managed to create the perfect estimates. The further away from 1 (i.e. the more they form a cloud), the less accurate the estimated breeding values are.

This is illustrated in figure 3. On the left you see a cloud of data points: some EBV resemble the true breeding value, but some estimates are also way off the true breeding value. The correlation between the EBV and TBV in this figure is 0.76, the EBV not resembles the TBV for all animals. For example, there are two animals with an EBV of 4, whereas their true breeding values are different: 3 and 5. In real life we cannot produce a graph like in this figure because we do not know the true breeding value. But what we can do is estimate the accuracy of the estimated breeding value: the correlation between the phenotypic information and the true breeding value. So how much is the EBV in line with the true breeding value.

Open

Plots of TBV on EBV with the perfect regression line if EBV = TBV. On the left is an example of inaccurate EBV, indicated by the cloud of data points with correlation between EBV and TBV of 0.76. On the right the EBV are estimated much more accurately and are almost in line with the TBV, resulting in a correlation between EBV and TBV of 0.98.