The Eddington bias is a selection bias applicable to surveys from measurement errors and population characteristics. If common items are at times mistaken for rare items, the count of the rare items can be significantly overestimated. For example, brighter stars are rarer, so if estimates of brightness mistake some percentage of dimmer stars as being brighter, that causes a larger overestimate than if the same percentage of brighter stars are mistaken as being dimmer. Even if the measurement errors are random (a normal distribution), a sample-set's randomness can still be skewed.
This is in contrast to another type of observation bias, the Malmquist bias (See Malmquist bias for the difference).