Skip to content

fligner.test (ctest) (PR#6739)

2 messages · karel.zvara@mff.cuni.cz, Peter Dalgaard

karel.zvara@mff.cuni.cz
# 
Full_Name: Karel Zvara
Version: 1.8.1
OS: MS Winows 2000
Submission from: (NULL) (195.113.30.163)


The test statistics of the fligner.test (ctest package) depends on the order of
cases:
Fligner-Killeen test for homogeneity of variances

data:  count by spray 
Fligner-Killeen:med chi-squared = 14.4828, df = 5, p-value =
0.01282
Fligner-Killeen test for homogeneity of variances

data:  count by spray 
Fligner-Killeen:med chi-squared = 5.4527, df = 5, p-value =
0.3632
Fligner-Killeen test for homogeneity of variances

data:  count by spray 
Fligner-Killeen:med chi-squared = 2.7349, df = 5, p-value =
0.7408

The problem lies, I think, in the ordering of centered values x:

 x <- unlist(tapply(x, g, function(u) u - median(u)))
Peter Dalgaard
# 
karel.zvara@mff.cuni.cz writes:
Right. x is not in the same order as g is after that operation, so
the two lines that come immediately after are incoherent:

    a <- qnorm((1 + rank(abs(x))/(n + 1))/2)
    STATISTIC <- sum(tapply(a, g, "sum")^2/tapply(a, g, "length"))


I think the unlist construction wants to be replaced with

  x <- x - tapply(x,g,median)[g]