Computing Standard Deviations: Accuracy

Four algorithms for the numerical computation
of the standard deviation of (unweighted) sampled data
are analyzed.  Two of the algorithms are well-known in the statistical
and computational literature; the other two are new algorithms
specifically intended for automatic computation.  Our discussion is 
expository, with emphasis on reaching a suitable definition of
"accuracy."  Each of the four algorithms is analyzed for the conditions
under which it will be accurate.  We conclude that all four
algorithms will provide accurate answers for many problems, but two
of the algorithms, one new, one old, are substantially more accurate
on difficult problems than are the other two.

CACM September, 1979

Chan, T.
Lewis, J.

Mean, standard deviation, least squares, updating
estimates, rounding error analysis, condition number.

5.5 5.11 5.19

CA790907 DB January 16, 1980  4:23 PM

1619	4	3166
2731	4	3166
3159	4	3166
3159	4	3166
3166	4	3166
3166	4	3166
3166	4	3166
3166	4	3166
3166	4	3166
3167	4	3166
3167	4	3166
1411	5	3166
2731	5	3166
2799	5	3166
3159	5	3166
3166	5	3166
3166	5	3166
3166	5	3166
3167	5	3166
2731	6	3166
2799	6	3166
3166	6	3166