Unrealistically low p-values from ranksum

I am unable to replicate SPSS results for the Mann-Whitney U test in MATLAB (via ranksum). My dataset has a relatively large number of cases: n1 = 152, n2 = 76. The data are clutch sizes, thus integer values from 1-3.

For one of my datasets, the ranksum output is as follows:

 >> [p,h,stats]=ranksum(x,y,'method','approximate')

 p =
  1.8996e-012

 h =
     1

 stats = 
       zval: -7.0417
    ranksum: 18282

And

 >> [p,h,stats]=ranksum(x,y,'method','exact')

 p =
     0

 h =
     1

 stats = 
    ranksum: 18282

Running the same data in SPSS returns:

              Habitat     N     Mean Rank     Sum of Ranks
 Clutch Size     1.00    152       120.28         18282.00
                 2.00     76       102.95          7824.00

 Mann-Whitney U           4898.000
 Wilcoxon W               7824.000
 Z                          -2.451
 Asymp. Sig. (2-tailed)       0.14

Is anyone familiar with both programs, and can you comment on why the z-scores are so different? The p-values from SPSS seem much more realistic for my data.

I am happy to provide the actual data.

[Akash Swamy]

I have the exact same problem! The p value i got = 4.5104e-11.

Internet please help!

[the cyclist]

It's probably too late to help the original poster, but Akash can you post

• your data
• the exact command you ran in MATLAB, including all parameters
• the result you got in MATLAB
• the exact command you ran in SPSS
• the result you got in SPSS

FYI, these sorts of questions are almost always resolved by figuring out the differing assumptions between the two statistical software packages. It's almost never "MATLAB is wrong" (or the other package being wrong).