Processing math: 100%

Doornik-Hansen test

The Doornik-Hansen test for multivariate normality (DOORNIK, J.A., and HANSEN, H. (2008)) is based on the skewness and kurtosis of multivariate data that is transformed to ensure independence. It is more powerful than the Shapiro-Wilk test for most tested multivariate distributions1.

The skewness and kurtosis are defined, respectively, as: s=m3m23 and k=m4m22, where: mi=1nni=1(xi¯x)i ¯x=1nni=1xi and n is a number of (non-missing) residuals.

The Doornik-Hansen test statistic derives from SHENTON, L.R., and BOWMAN, K.O. (1977) and uses transformed versions of skewness and kurtosis.

The transformation for the skewness s intoz1 is as in D'AGOSTINO, R.B. (1970):

β=3(n2+27n70)(n+1)(n+3)(n2)(n+5)(n+7)(n+9) [1]

ω2=1+2(β1) [2]

δ=1log(ω2) [3]

y=s(ω21)(n+1)(n+3)12(n2) [4]

z1=δlog(y+y21) [5]

The kurtosis k is transformed from a gamma distribution to χ2, which is then transformed into standard normal z2 using the Wilson-Hilferty cubed root transformation:

δ=(n3)(n+1)(n2+15n4) [6]

a=(n2)(n+5)(n+7)(n2+27n70)6δ [7]

c=(n7)(n+5)(n+7)(n2+2n5)6δ [8]

l=(n+5)(n+7)(n3+37n2+11n313)12δ [9]

α=a+c×s2 [10]

χ=2l(k1s2) [11]

z2=9α(19α1+3χ2α) [12]

Finally, the Doornik-Hansen test statistic is defined as the sum of squared transformations of the skewness and kurtosis. Approximately, the test statistic follows a χ2distribution, i.e.:

DH=z21+z22χ2(2) [13]

  1. The description of the test derives from DOORNIK, J.A., and HANSEN, H. (2008).