Dunn's post hoc test

I think I misunderstood your follow-up question.  Try this:

averank<-sort(sample(1:100,10,replace=TRUE))
x=matrix(nrow=length(averank),ncol=length(averank))
for (i in 1:length(averank)){

+ for (j in 1:length(averank)){
+ x[i,j] <- averank[i] - averank[j]
+ }}

x

      [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
 [1,]    0  -14  -15  -16  -16  -23  -37  -51  -67   -79
 [2,]   14    0   -1   -2   -2   -9  -23  -37  -53   -65
 [3,]   15    1    0   -1   -1   -8  -22  -36  -52   -64
 [4,]   16    2    1    0    0   -7  -21  -35  -51   -63
 [5,]   16    2    1    0    0   -7  -21  -35  -51   -63
 [6,]   23    9    8    7    7    0  -14  -28  -44   -56
 [7,]   37   23   22   21   21   14    0  -14  -30   -42
 [8,]   51   37   36   35   35   28   14    0  -16   -28
 [9,]   67   53   52   51   51   44   30   16    0   -12
[10,]   79   65   64   63   63   56   42   28   12     0

averank

 [1] 15 29 30 31 31 38 52 66 82 94



----- Original Message -----
From: "IAIN GALLAGHER" <iaingallagher at btopenworld.com>
To: "Martin Henry H. Stevens" <HStevens at MUOhio.edu>
Cc: <r-help at stat.math.ethz.ch>
Sent: Monday, October 17, 2005 2:29 PM
Subject: Re: [R] Dunn's post hoc test

Thanks for your reply Hank. It's not really what I'm
after (though it's good to know).

For the test ( as described in Statistics for the
Biosciences by W. Gardiner. Prentice Hall, 1997) I
have to rank my groups, calculate the average rank,
then subtratc each average rank from every other. Any
value greater than the test statistic is significant.

eg average rank difference table:

   2     5     8    9
---|------------------
2  -     3     6    7
   |
5  -     -     3    4
   |
8  -     -     -    1
   |
9  -     -     -    -
   |

I can't get my head around writing an algorithm for
this if I have a vector of average ranks eg averank<-
c(2,5,8,9).

I know I can address the vector by index and that this
is probably the correct route but I can't get the
indexing algorithm right!

I'm sure someone will point out somethng simple and
I'll kick myself but the help would be appreciated.

Thanks again.

Iain Gallagher

--- "Martin Henry H. Stevens" <HStevens at MUOhio.edu>
wrote:

I don't know Dunn's rank test, but the following
substracts each of
the sums of averanks from the next rank.

cumsum(averank)[-length(averank)] - averank[-1]

Hank

On Oct 17, 2005, at 4:30 AM, Iain Gallagher wrote:

Hi Everyone.

I am rather new to R and I've been trying to

implement a function to

carry out the above test. For a couple of days now

I've been stuck on

how to generate average rank differences.

Say I have a vector of average ranks:

averank<- c(2,5,9,12)

I would like to subtract averank[1] from

averank[2], averank[1] and

averank[2] from averank[3] and averank[1],

averank[2] and averank[3]

from averank[4] etc (I know the syntax is wrong

here... it's just for

illustration) but I can't work out how to do it.

Ideally I would like to generate an array showing

the differences

between the average ranks so I could tell at a

glance which ones were

greater than my critical value

I've been looking at loops etc but it's a little

beyond me at the

moment. Thanks for any suggestions.

Iain Gallagher
IIIR
Edinburgh University