In summary: Own performance is valuable for trait with a high heritability. Phenotypes collected on half sibs are more valuable than those on full sibs. Even more valuable are phenotypes collected on half sib progeny: no common environment with the animal, generally larger potential numbers available, and offspring receive the true half of the genetic potential of the animal. This all allows for the most accurately estimated breeding value.
Now you know how to calculate the accuracy of selection. But why would you? Because the higher the accuracy of the EBV, the lower the risk of ranking the animals in the wrong order, and thus the lower the risk of selecting the wrong animals for breeding. Figure 4 illustrates the meaning of accuracy. In the figure are three normal distributions, each representing an EBV of 50, but with various levels of accuracy. The most accurate EBV has a 95% confidence interval between 45 and 55. In other words: the best estimate of the breeding value is 50, but with some level of inaccuracy around that estimate. However, it is 95% certain that the true breeding value lays between 45 and 55. The least accurate EBV also indicates that the best estimate is 50, but the 95% confidence interval lays between 35 and 65. That is much larger than that of the most accurate estimate. It means that the risk of ranking the animals in the wrong way is larger.
This difference in accuracy may be a result of a different heritability. The higher the heritability, the more accurate the EBV. But it can also be a result of suboptimal quality of information source, for example only a limited number of offspring, that could be used to determine the regression coefficient. In that case, in a next round of breeding value estimation it is likely that more information has become available. More information usually means that the regression coefficient can be better determined, and thus that the breeding value can be estimated more accurately. This can have consequences for the size of the best estimate, and thus for the ranking of the animals!! This is exactly why EBV, especially of the young animals with little information available, may change with new runs of breeding value estimation. If they do, it is uncertain in which direction they will change: up or down is equally likely.
Thus:
The lower the accuracy of an EBV, the larger the risk of a change in EBV when new information sources (e.g. offspring) become available.