Re: [R] A questionb about the Wilcoxon signed rank test

[Thomas Lumley]

The problem is that your data contains ties, which mess up the nice theory and 
result in different people using different approximations.

I don't know where your z-statistic formula comes from, but you can find the 
one R uses by looking at the source code in stats:::wilcox.test.default.

To see that R's z-statistic approximation is better than yours, try breaking 
the ties randomly and using exact=TRUE.
    
wilcox.test(endprice0+rnorm(length(endprice0),s=1e-10),endprice1,paired=TRUE,exact=TRUE)
You will find that the p-values agree fairly well with R's 0.88.

        -thomas


On Mon, 5 Apr 2010, hix li wrote:

Hi guys,
?
I have two data sets of prices: endprice0, endprice1
?
I use the Wilcox test:
?
wilcox.test(endprice0, endprice1, paired = TRUE, alternative = "two.sided",? 
conf.int = T, conf.level = 0.9)
?
The result is with V = 1819, p-value = 0.8812.
?
Then I calculated the z-value of the test: z-value = -2.661263. The 
corresponding p-value is: p-value = 0.003892, which is different from the 
p-value computed in the Wilcox test, I am using the following steps to compute 
the z-value:
?
diff = c(endprice0 - endprice1)
diffNew = diff[diff !=0]
diffNew.rank = rank(abs(diffNew))
diffNew.rank.sign <-? diffNew.rank? *? sign(diffNew)
ranks.pos <- sum(diffNew.rank.sign[diffNew.rank.sign >0]) = 1819
ranks.neg <- -sum(diffNew.rank.sign[diffNew.rank.sign<0]) = 1751
?
v = ranks.neg
n = 100
z= (v - n *(n+1)/4)/sqrt(n*(n+1)*(2*n+1)/24) = -2.661263


Which p-value should I take for the Wilcox test then?
?
Hix
?
the data sets used in my?test?are:

endprice0 = c(136.3800, 134.8500, 350.7500, 18.8400, 0.0000, 0.0600, 159.1900, 
242.5600, 0.0400, 289.9000, 0.0000, 42.6100, 275.9500, 76.6200, 36.6400, 
0.0000, 81.5900, 179.3600, 86.2200, 210.8000, 118.7200, 45.5800, 98.1900, 
137.0300, 47.7900, 123.7700, 23.2400, 0.0400, 130.2300, 0.0400, 0.0000, 
130.3800, 150.7600, 0.5900, 277.3000, 166.0100, 0.0400, 71.9400, 80.1300, 
162.8800, 85.0500, 125.4400, 138.0600, 0.0600, 140.6300, 100.9700, 0.0000, 
0.0400, 213.7300, 86.9200, 294.8200, 0.0400, 0.0000, 239.2100, 0.0000, 13.7700, 
95.5300, 0.0400, 146.7200, 0.0000, 0.00, 121.57, 68.23, 5.31, 0.04, 96.31, 
206.02, 313.39, 92.34, 31.64, 118.71, 499.6, 0, 129.04, 106.88, 183.92, 50.42, 
0, 0.04, 0.04, 1.57, 355.56, 81.19, 327.17, 151.18, 0, 0, 125.03, 0, 0.04, 
132.01, 0, 0, 11.49, 23, 13.46, 326.64, 198.19, 114.22, 79.53)
?
endprice1 = c(138.9300, 131.9700, 300.4700, 0.0000, 0.0000, 0.2200, 159.6300, 
277.9100, 0.0000, 328.9700, 0.0000, 40.5100, 270.1000, 52.8000, 39.3800, 
0.0400, 79.7100, 110.5600, 41.1600, 224.6600, 123.8800, 53.2700, 96.1500, 
67.2800, 40.7300, 99.4900, 20.4900, 0.0400, 126.1000, 0.0000, 1.3700, 140.6500, 
165.7200, 0.0000, 314.4200, 207.7400, 0.0400, 76.9300, 75.8000, 184.9100, 
83.3700, 139.5300, 157.0500, 0.0000, 147.5900, 105.2800, 0.0000, 0.0000, 
207.3000, 74.1100, 288.3900, 0.0400, 0.0000, 213.7200, 0.0400, 14.8300, 
53.7000, 0.0400, 150.0800, 0.0000, 0, 123.73, 68.01, 9.52, 0, 111.86, 249.69, 
354.18, 98, 31.3, 117.54, 455.32, 1.06, 127.92, 114.51, 173.85, 53.22, 0, 0, 0, 
0.31, 376.69, 69.43, 278.8, 147.11, 0.04, 0, 120.05, 0, 0.04, 132.97, 0, 0, 
9.98, 28.85, 13.77, 295.17, 191.54, 126.44, 84.83)


?


     __________________________________________________________________
Make your browsing faster, safer, and easier with the new Internet 
Explorer[[elided Yahoo spam]]
com/ca/internetexplorer/
        [[alternative HTML version deleted]]

Thomas Lumley                   Assoc. Professor, Biostatistics
tlum...@u.washington.edu        University of Washington, Seattle

______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.