[R] Different results on running Wilcoxon Rank Sum test in R and SPSS

John Fox j|ox @end|ng |rom mcm@@ter@c@
Thu Jan 21 00:10:39 CET 2021 

Dear Bharat Rawlley,

On 2021-01-20 1:45 p.m., bharat rawlley via R-help wrote:
>   Dear Professor John,
> Thank you very much for your reply!
> I agree with you that the non-parametric tests I mentioned in my previous email (Moods median test and Median test) do not make sense in this situation as they treat PFD_n and drug_code as different groups. As you correctly said, I want to use PFD_n as a vector of scores and drug_code to make two groups out of it. This is exactly what the Independent samples median test does in SPSS. I wish to perform the same test in R and am unable to do so.
> Simply put, I am asking how to perform the Independent samples median test in R just like it is performed in SPSS?

I'm afraid that I'm the wrong person to ask, since I haven't used SPSS 
in perhaps 30 years and have no idea what it does to test for 
differences in medians. A Google search for "independent samples median 
test in R" turns up a number of hits.

> 
> Secondly, for the question you are asking about the test statistic, I have not performed the Wilcoxon Rank sum test in SPSS for the PFD_n and drug_code data. I have said something to the contrary in my first email, I apologize for that.

For continuous data, the Wilcoxon test is, I believe, a reasonable 
choice, but not when there are so many ties. If SPSS doesn't perform a 
Wilcoxon test for a difference in medians, then there's of course no 
reason to expect that the p-values would be the same.

Best,
  John

> Thank you very much for your time!
> Yours sincerelyBharat Rawlley    On Wednesday, 20 January, 2021, 04:47:21 am IST, John Fox <jfox using mcmaster.ca> wrote:
>   
>   Dear Bharat Rawlley,
> 
> What you tried to do appears to be nonsense. That is, you're treating
> PFD_n and drug_code as if they were scores for two different groups.
> 
> I assume that what you really want to do is to treat PFD_n as a vector
> of scores and drug_code as defining two groups. If that's correct, and
> with your data into Data, you can try the following:
> 
> ------snip ------
> 
>   > wilcox.test(PFD_n ~ drug_code, data=Data, conf.int=TRUE)
> 
>      Wilcoxon rank sum test with continuity correction
> 
> data:  PFD_n by drug_code
> W = 197, p-value = 0.05563
> alternative hypothesis: true location shift is not equal to 0
> 95 percent confidence interval:
>    -2.000014e+00  5.037654e-05
> sample estimates:
> difference in location
>                -1.000019
> 
> Warning messages:
> 1: In wilcox.test.default(x = c(27, 26, 20, 24, 28, 28, 27, 27, 26,  :
>    cannot compute exact p-value with ties
> 2: In wilcox.test.default(x = c(27, 26, 20, 24, 28, 28, 27, 27, 26,  :
>    cannot compute exact confidence intervals with ties
> 
> ------snip ------
> 
> You can get an approximate confidence interval by specifying exact=FALSE:
> 
> ------snip ------
> 
>   > wilcox.test(PFD_n ~ drug_code, data=Data, conf.int=TRUE, exact=FALSE)
> 
>      Wilcoxon rank sum test with continuity correction
> 
> data:  PFD_n by drug_code
> W = 197, p-value = 0.05563
> alternative hypothesis: true location shift is not equal to 0
> 95 percent confidence interval:
>    -2.000014e+00  5.037654e-05
> sample estimates:
> difference in location
>                -1.000019
> 
> ------snip ------
> 
> As it turns out, your data are highly discrete and have a lot of ties
> (see in particular PFD_n = 28):
> 
> ------snip ------
> 
>   > xtabs(~ PFD_n + drug_code, data=Data)
> 
>        drug_code
> PFD_n  0  1
>      0  2  0
>      16  1  1
>      18  0  1
>      19  0  1
>      20  2  0
>      22  0  1
>      24  2  0
>      25  1  2
>      26  5  2
>      27  4  2
>      28  5 13
>      30  1  2
> 
> ------snip ------
> 
> I'm no expert in nonparametric inference, but I doubt whether the
> approximate p-value will be very accurate for data like these.
> 
> I don't know why wilcox.test() (correctly used) and SPSS are giving you
> slightly different results -- assuming that you're actually doing the
> same thing in both cases. I couldn't help but notice that most of your
> data are missing. Are you getting the same value of the test statistic
> and different p-values, or is the test statistic different as well?
> 
> I hope this helps,
>    John
> 
> John Fox, Professor Emeritus
> McMaster University
> Hamilton, Ontario, Canada
> web: https://socialsciences.mcmaster.ca/jfox/
> 
> On 2021-01-19 5:46 a.m., bharat rawlley via R-help wrote:
>>    Thank you for the reply and suggestion, Michael!
>> I used dput() and this is the output I can share with you. Simply explained, I have 3 columns namely, drug_code, freq4w_n and PFD_n. Each column has 132 values (including NA). The problem with the Wilcoxon Rank Sum test has been described in my first email.
>> Please do let me know if you need any further clarification from my side! Thanks a lot for your time!
>> structure(list(drug_code = c(0, 0, 1, 1, 1, 0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 1, 0, 1, 1, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, 0, 0, 1, 1, 0, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 0, 0, 1, 1, 0, 0, 0, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 1, 1, 0, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0), freq4w_n = c(1, NA, NA, 0, NA, 4, NA, 10, NA, 0, 6, NA, NA, NA, NA, NA, 10, NA, 0, NA, NA, NA, NA, 0, NA, 0, NA, NA, NA, 0, NA, 0, NA, NA, NA, NA, NA, NA, NA, NA, 0, 0, 12, 0, NA, 1, 2, 1, 2, 2, NA, 28, 0, NA, 4, NA, 1, NA, NA, NA, NA, NA, 0, 3, 1, NA, NA, NA, NA, 4, 28, NA, NA, 0, 2, 12, 0, NA, NA, NA, 0, NA, 0, NA, NA, NA, NA, NA, NA, NA, NA, NA, 3, NA, NA, NA, NA, NA, NA, 6, 1, NA, NA, NA, 0, NA, NA, NA, 0, 0, NA, 0, NA, 2, 8, 3, NA, NA, NA, 0, NA, NA, NA, 9, NA, NA, NA, NA, NA, NA, NA, NA), PFD_n = c(27, NA, NA, 28, NA, 26, NA, 20, NA, 30, 24, NA, NA, NA, NA, NA, 18, NA, 28, NA, NA, NA, NA, 28, NA, 28, NA, NA, NA, 28, NA, 28, NA, NA, NA, NA, NA, NA, NA, NA, 28, 28, 16, 28, NA, 27, 26, 27, 26, 26, NA, 0, 30, NA, 24, NA, 27, NA, NA, NA, NA, NA, 28, 25, 27, NA, NA, NA, NA, 26, 0, NA, NA, 28, 26, 16, 28, NA, NA, NA, 28, NA, 28, NA, NA, NA, NA, NA, NA, NA, NA, NA, 25, NA, NA, NA, NA, NA, NA, 22, 27, NA, NA, NA, 28, NA, NA, NA, 28, 28, NA, 28, NA, 26, 20, 25, NA, NA, NA, 30, NA, NA, NA, 19, NA, NA, NA, NA, NA, NA, NA, NA)), row.names = c(NA, -132L), class = c("tbl_df", "tbl", "data.frame"))
>>
>> Yours sincerely Bharat Rawlley    On Tuesday, 19 January, 2021, 03:53:27 pm IST, Michael Dewey <lists using dewey.myzen.co.uk> wrote:
>>    
>>    Unfortunately your data did not come through. Try using dput() and then
>> pasting that into the body of your e-mail message.
>>
>> On 18/01/2021 17:26, bharat rawlley via R-help wrote:
>>> Hello,
>>> On running the Wilcoxon Rank Sum test in R and SPSS, I am getting the following discrepancies which I am unable to explain.
>>> Q1 In the attached data set, I was trying to compare freq4w_n in those with drug_code 0 vs 1. SPSS gives a P value 0.031 vs R gives a P value 0.001779.
>>> The code I used in R is as follows -
>>> wilcox.test(freq4w_n, drug_code, conf.int = T)
>>>
>>>
>>> Q2 Similarly, in the same data set, when trying to compare PFD_n in those with drug_code 0 vs 1, SPSS gives a P value 0.038 vs R gives a P value < 2.2e-16.
>>> The code I used in R is as follows -
>>> wilcox.test(PFD_n, drug_code, mu = 0, alternative = "two.sided", correct = TRUE, paired = FALSE, conf.int = TRUE)
>>>
>>>
>>> I have tried searching on Google and watching some Youtube tutorials, I cannot find an answer, Any help will be really appreciated, Thank you!
>>> ______________________________________________
>>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
>>> 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.
>>>
>>
>    
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.
> 

-- 
John Fox, Professor Emeritus
McMaster University
Hamilton, Ontario, Canada
web: https://socialsciences.mcmaster.ca/jfox/

More information about the R-help mailing list