Have you considered Monte Carlo?
Invent a test, then Monte Carlo it to get a p-value.
Something similar to what you are asking is provided by Monte
Carlo confidence bounds on a QQ plot, discussed on this list last June.
To find it, go to www.r-project.org -> Search -> "R site search"
->
"confidence bounds on QQ plot". I got 9 hits from this just now. The
first was an answer to a question I asked then on this issue, giving two
references.
hope this helps.
spencer graves
Yong Chao wrote:
>I tried to use shapiro.test or ks.test to check the
>normality of some data, the problem is, the
>distribution function is a mixture of a Gaussian and
>some other distributions at the tails. The hypothesis
>is that if the tails are excluded, the distribution is
>perfect Gaussian, and I want to test that.
>
>But I cannot simply cut the tails off and do a
>normality test on the truncated data, as shown in the
>following example, this will fail.
>
>
>So that question is: how can I test whether the middle
>chunk of the distribution is Gaussian?
>
>Thanks!
>
>Yong
>
>
>
>>r<-rnorm(1000)
>>r.trunc<-r[which(abs(r)<1.5)]
>>shapiro.test(r.trunc)
>>
>>
>
> Shapiro-Wilk normality test
>
>data: r.trunc
>W = 0.9855, p-value = 1.237e-07
>
>
>
>>ks.test(r.trunc, "pnorm")
>>
>>
>
> One-sample Kolmogorov-Smirnov test
>
>data: r.trunc
>D = 0.0873, p-value = 3.116e-06
>alternative hypothesis: two.sided
>
>
>
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://stat.ethz.ch/mailman/listinfo/r-help
>PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
>
>