Error in computation of Mann-Whitney-U Test seems to be identified

[Dr. Oliver Walter]

Dear Prof. Nuszbaum,

you were right: There seems to be a problem and I think that I found it. The missing data subcommand /MISSING=ANALYSIS does not seem to be working properly. Although it should exclude cases with missing data for each analysis (hence, it is called ANALYSIS) it does not do it. It seems to include these cases in the U test analyses instead of excluding them. When I assigned ranks to cases using the RANK VARIABLES command the ranks PSPP built were the same as in SPSS. But in the U test analyses the ranks differ.

You may ask why this subcommand is important here, because you did not use it in your syntax. I think it is the default setting and has not to be written explicitely. Hence, PSPP uses it automatically.

When I excluded cases with missing data manually (I deleted them) and later by the subcommand /MISSING = LISTWISE the results of PSPP were the same as in SPSS and R (I checked R's results for variable I002_01 only).

Hence, PSPP developers should check the missing data subcommand /MISSING=ANALYSIS in the NPAR command and correct it.

In the meantime we - you, your students and others - can run U test analyses *** for each variable separately *** and use the subcommand /MISSING=LISTWISE. Then the results should be the same as in SPSS. I checked it using your data set and they were the same. The commands are

NPAR TESTS
/STATISTICS=DESCRIPTIVES
/MISSING=LISTWISE
/MANN-WHITNEY = I002_01  BY sum_HL (0 1) .

NPAR TESTS
/STATISTICS=DESCRIPTIVES
/MISSING=LISTWISE
/MANN-WHITNEY = I002_02  BY sum_HL (0 1) .

NPAR TESTS
/STATISTICS=DESCRIPTIVES
/MISSING=LISTWISE
/MANN-WHITNEY = I002_03  BY sum_HL (0 1) .

NPAR TESTS
/STATISTICS=DESCRIPTIVES
/MISSING=LISTWISE
/MANN-WHITNEY = I002_04  BY sum_HL (0 1) .

for your data set and analyses.

I attached the results in a pdf file (html output was strangely not working for this output) and you could compare them to your results of SPSS.

Kind regards,

Oliver Walter

Am 06.03.2017 um 13:10 schrieb Mandy Nuszbaum:

Dear Oliver Walter,

 

I attached a shortened data file and Output (Copy in Excel including syntax). The data base is not very large but results should be the same, or? Maybe it’s my fault by doing the analysis.

 

The independent Variables are I002_01 to I002_04. The group variable is sum_HL with groups 0 and 1.

 

John Darrington mentioned, that there was an error, which was fixed (“I suspect this is bug 47041 which has been fixed about a year ago”). So, I installed PSPP new and did the analysis again (see Excel file).

 

I have installed the version: pspp-20160927-daily-32bits-setup.exe

 

Hope you can help, because my students use pspp and I like to be sure that they get correct data output. The attached data is from such a student project.

 

Thanks for your fast response.

 

Kind regards,

Mandy Nuszbaum

 

 

Von: Dr. Walter Statistics [mailto:address@hidden]
Gesendet: Samstag, 4. März 2017 19:51
An: address@hidden; address@hidden
Betreff: Re: Error in computation of Mann-Whitney-U Test

 

Dear Prof. Nuszbaum,

your question would be easier to answer if you described it in more detail, e.g. by posting your outputs and / or your data sets (if the sample sizes are small, we could compare the results).

Kind regards

Oliver Walter

---
Dr. Walter Statistics
Gabelsberger Straße 27
24148 Kiel

Am 04.03.2017 um 19:20 schrieb Mandy Nuszbaum:

Dear all,

 

I computed Mann-Whitney-U using PSPP and SPSS and compared the results using different data sets. Results are completely different and led to wrong conclusions.

 

Can anyone fix the problem?

 

Kind regards,

mn

_______________________________________________

Pspp-users mailing list

address@hidden

https://lists.gnu.org/mailman/listinfo/pspp-users

 

_______________________________________________
Pspp-users mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/pspp-users

output_missings.pdf
Description: Adobe PDF document