Real Statistics Release 9.2.1

I want to thank everyone who has made suggestions or has identified errors in the website or software. Your help has improved the utility and accuracy of Real Statistics.

The following are the new features in this release:

Weighted Variance, Standard Deviation and Covariance

The following new worksheet functions have been added. See Weighted Variance for more information.

WVAR(R1, Rw, rel) = the unbiased weighted variance for the data in column array R1 and weights in column array Rw. If rel (default TRUE), then it is assumed that reliability weights are used; otherwise frequency weights are used.

WSTDEV(R1, Rw, rel) = the unbiased weighted standard deviation for the data in column array R1 and weights in column array Rw = SQRT(WVAR(R1, Rw, rel)).

WCOV(R1, Rw, rel) = the unbiased weighted covariance matrix for the data in the n × k array R1 and weights in the n × 1 column array Rw.

Changes to the Gradient Descent and BGFS functions

The Gradient Descent worksheet functions, MGRADIENT and MGRADIENTX, introduced in Rel 9.2. have been revised to add more flexibility. These changes are described in Gradient Descent.

Similar changes have been made to the BGFS function, which also finds the minima of a multivariate function.

Enhancement to the Trinomial Testing functions

The TRINOM_TEST and TrinomTest worksheet functions have been enhanced to support data sets with more than 170 elements.

Bug Fix

A bug that produces a value that is .5 too large has been corrected in the BNORMSDIST(x, y, rho, cdf, don) function when x = -y; x ≠ 0; rho is not 0, 1, or -1; cdf = TRUE; and don = TRUE. This bug also influenced the BNORMDIST and BNORMRECT functions. These have now been corrected.

An error was also produced when x = 0 or y = 0 but not both; rho is not 0, 1, or -1; and don = TRUE. This bug has also been fixed.

The default for the don argument has also been changed from TRUE to FALSE. This means that the Genz algorithm is now used by default instead of the Donnelly algorithm when estimating the cdf of the bivariate normal distribution.