Chauvenet's criterion is a rule for identifying a possible outlier among 
observations under the assumption that their errors follow a normal
distribution. For an observation 
, let
 |
where 
and 
are the sample arithmetic mean and standard
deviation. The observation is rejected when
![NPr[|Z|>=z]<1/2,](/images/equations/ChauvenetsCriterion/NumberedEquation2.svg) |
where 
has the standard normal distribution.
Thus the expected number of observations at least as far from the arithmetic
mean is less than one half. The rule depends on the normal-error assumption and
is ordinarily applied to at most one observation at a time, with the statistics recomputed
after a rejection.
