It seems to me that this raises two large questions:
1. Whether there is some non-parametric method for
two-way ANOVA that handles interaction. In particular,
my understanding is that Friedman's procedure does not
extend to interactions.
2. Assuning an answer to 1, whether there is some
Stata implementation. I am not aware of one.
Rich Goldstein did a Friedman program in STB-3
in 1991. It still works, except that it assumes that
you have no variables whose names begin with "r".
If you have, the result will be wrong and they
will get -drop-ped.
I modernised the code as below. This has not been
tested much at all. But it doesn't do interactions
either.
*! 2.0.0 NJC 3 April 2004
*! 1.01 Richard Goldstein 17 July 1991
program friedman8
version 8
syntax varlist(min=2 numeric) [if] [in]
quietly {
marksample touse
count if `touse'
if r(N) == 0 error 2000
else local n = r(N)
local k : word count `varlist'
tempvar ranksum rankss
gen `ranksum' = 0 if `touse'
foreach v of local varlist {
tempvar rank
egen `rank' = rank(`v') if `touse'
replace `ranksum' = `ranksum' + `rank'
drop `rank'
}
gen `rankss' = `ranksum'^2
su `rankss', meanonly
local S = `r(sum)' - (((`k'^2) * `n' * (`n' + 1)^2) / 4)
local F = (12 * `S') / (`k' * `n' * (`n' + 1))
local W = (12 * `S') / ((`k'^2) * `n' * (`n'^2 - 1))
local prob = `k' * (`n' - 1) * `W'
}
di "{txt}Friedman = {res}" %8.4f `F'
di "{txt}Kendall = {res}" %8.4f `W'
di "{txt}P-value = {res}" %8.4f chiprob(`n' - 1, `prob')
end
Nick
[email protected]
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of
> [email protected]
> Sent: 03 April 2004 04:38
> To: stata group mail
> Subject: st: ...non parametric alternative to ANOVA...
>
>
> Dear all,
> someone can tell me something about the non parametric
> alternatives to two-way
> ANOVA in Stata. I have to discover/investigate if a
> distribution is influenced
> by the interaction of two factors, I had just applied the
> kwallis test to
> the distribution with a combined factor of the two factor
> (combined factor=
> group (factor1 factor2) with good results but I think that
> this test doesn't
> go where I wish to go.
> Every opinion will be welcome.
> Many thanks.
>
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
