Accuracy is a numerical expression that gauges how much information went into the calculation of an animal’s EPD, and therefore how much confidence we can place in the estimate. Each EPD has an associated accuracy that ranges from 0 to 1. To make the most effective use of accuracy, users need to understand the concept of possible change—in both groups and individual animals.
Possible change is the range ± an animal’s EPD that, 67 percent of the time, we expect the animal’s true genetic value to fall within. If we extend the range to 2 and 3 possible change units ± an animal’s EPD, its true value is expected to fall within the range 95 and 99 percent of the time, respectively. With these percentages in mind, we can make some assumptions; first, in a group of 100 bulls, it is expected that 33 (100 - 67), 5 (100 – 95) and 1 (100 – 99) of them have true genetic values outside a 1, 2 and 3 possible change unit range, respectively, from their EPD for a particular trait; second, when considering multiple traits, the number of instances in which true values fall outside possible change ranges increases by a multiple of the number of traits. For example, if we considerin looking at 15 traits (the number currently evaluated in our Sire Summary) on our sample of 100 bulls, we expect 495 (15 x 33), 75 (15 x 5) and 15 (15 x 1) instances where sires have true values more than 1, 2 and 3 possible change units, respectively, from their current EPDs.
For example, a bull calf with a 100 YW EPD and a corresponding 0.30 accuracy, the PC range for this calf’s EPD is ± 18. If he turns into an AI sire, eventually developing a YW accuracy of .99 (i.e., his EPD is essentially his true genetic value), there is a 67, 95 and 99 percent chance that his .99 accuracy EPD will fall between 82-118 (100 ± 18), 64-136 (100 ± 2(18)) and 46-154 (100 ± 3(18)), respectively. As you can see, it would be fairly common (33 percent of the time) for the calf to end up with an EPD over 118 or under 82, a result that would fairly categorize him as either a high- or low-growth bull. Furthermore, it wouldn’t be that extraordinary (1 percent of the time) for this middle-of-the-road YW calf to end up being on the very extreme ends of the spectrum (from 154 to 46). If we expand the array of traits to 15 for this calf, it would hardly be remarkable for one of his .99 accuracy EPDs to end up 3 PC units from where he started; it should happen 15 percent of the time.
See this video from the University of Missouri which further explains EPD Accuracy.