Statistical Methods
Report Circular Dispersion in the Chosen Angular Units
Define the task
Angular dispersion needs a circular definition. Ordinary standard deviation can become large simply because values straddle the reporting boundary. Specify degrees or radians, the complete-turn interval and whether data represent directions or axes. Circular standard deviation summarizes concentration through the mean resultant length R; it does not describe every feature of a multimodal angular distribution. Pair it with a plot or a tabulation of directions, sample size and the missing-value policy.
Carry out the check
For a conventional full turn in radians, circular standard deviation is sqrt(−2 log R). Convert the result to degrees only after using dimensionless trigonometric arguments. Software options that normalize the result change its interpretation, so record those options. R=1 gives zero dispersion; as R approaches zero the formula grows without bound. This is not a maximum angular separation and should not be read as a conventional interval around the mean. Near cancellation, a mean direction can be undefined even when a software routine returns a number.
Worked hypothetical example
For hypothetical directions 330 and 30 degrees, R=cos(30 degrees)=sqrt(3)/2. The circular standard deviation is approximately 0.536360 radians, or 30.7312 degrees. Their maximum separation is 60 degrees, which is a different quantity. For 0 and 180 degrees the resultant cancels; report the degeneracy rather than an apparently precise finite summary caused by floating-point rounding. Recalculate the resultant independently before reporting a dispersion statistic.
Method checked 16 November 2026 against SciPy documentation: circstd.