[R] fligner.test

[Stefan Th. Gries]

Hi all

I have a question regarding the Fligner-Killeen test. I am using

- a PC with Windows XP (Build 20600.xpsp080413-2111 (Service Pack 3);
- the following R version:
> sessionInfo()
R version 2.7.0 (2008-04-22)
i386-pc-mingw32


I have a vector LENGTH and a factor RELATION that are distributed like this:

> table(LENGTH, RELATION)
     RELATION
LENGTH object subject
   1       9      51
   2      25      18
   3      36      23
   4      23      14
   5      17      13
   6      12       7
   7      10       5
   8       3       7
   9       4       3
   10      3       5
   11      4       2
   12      2       0
   13      0       2
   16      2       1
   22      0       1
   49      1       0
   53      1       0

I wanted to run a Fligner test to see whether the lengths for the two
relations differ in terms of their variance. This is what I found,
though:

#########
> fligner.test(LENGTH[RELATION=="subject"], LENGTH[RELATION=="object"])
       Fligner-Killeen test of homogeneity of variances
data:  LENGTH[RELATION == "subject"] and LENGTH[RELATION == "object"]
Fligner-Killeen:med chi-squared = 18.3552, df = 14, p-value = 0.1911

> fligner.test(LENGTH[RELATION=="object"], LENGTH[RELATION=="subject"])
       Fligner-Killeen test of homogeneity of variances
data:  LENGTH[RELATION == "object"] and LENGTH[RELATION == "subject"]
Fligner-Killeen:med chi-squared = 16.8838, df = 13, p-value = 0.2047

> fligner.test(LENGTH~RELATION)
       Fligner-Killeen test of homogeneity of variances
data:  LENGTH by RELATION
Fligner-Killeen:med chi-squared = 0.626, df = 1, p-value = 0.4288
#########

The order of the vectors etc. changes the results??? Needless to say
this does not happen with var.test or ... Is this normal; if so, which
one is the one to report?

Thx,
STG