Timeline for Wilcoxon rank sum test W=0 but p<0.05?
Current License: CC BY-SA 4.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 4 at 16:14 | history | edited | jginestet | CC BY-SA 4.0 |
Added details about why the subtraction
|
| Oct 4 at 13:13 | comment | added | Ser Huisman | Thank you! I will be very explicit in my methods section about where the W comes from. I agree that reporting the W does not add much, however, I am not a fan of only p-values. Ideally, I would report an effect size, I'll look into the options for that! PS I like that last sentence, haha | |
| Oct 2 at 20:02 | comment | added | jginestet | Rather than reporting any W, or U statistic (subtracted or not)), you may simply want to report none. Yes, any of the above are equivalent, but they are also completely redundant with the p-value. These statistic and the p-value are deterministacally related. So reporting their value does not add anything, does not re-inforce the p-value, does not confirm the significance. Rather than confusing your audience with statistics the computatiuons of which they may or not know, just use the p-value (at least most of your audience will think they know what it means). | |
| Oct 2 at 14:22 | comment | added | Christian Hennig | @SerHuisman As all these are equivalent, you can give out whatever of these values you want if you say what it actually is. You can even give your hand-calculated value as long as it is correct, with definition, and give the p-value from R. | |
| Oct 2 at 9:03 | comment | added | Ser Huisman | Thank you for the useful answer! And the addition about the literature. I agree, quite confusing; I have already read about 4 different ways now to calculate W. I would prefer R to just use Mann Whitney U and provide a U and/or z value for this test, but well. I agree especially with this strange outcome for my sample (the same happened in another sample with less extreme differences; 1 group had neg and pos values in there), I calculated it by hand and got your W's too... I am leaning towards reporting SPSS's U or z for this, but I am hesitant as R is considered the holy grail in my field ;) | |
| Oct 2 at 6:15 | comment | added | Roland | From the documentation: "The literature is not unanimous about the definitions of the Wilcoxon rank sum and Mann-Whitney tests. The two most common definitions correspond to the sum of the ranks of the first sample with the minimum value (m(m+1)/2 for a first sample of size m) subtracted or not: R subtracts. It seems Wilcoxon's original paper used the unadjusted sum of the ranks but subsequent tables subtracted the minimum." | |
| Oct 1 at 21:03 | history | edited | jginestet | CC BY-SA 4.0 |
edited typos in body
|
| Oct 1 at 19:41 | history | answered | jginestet | CC BY-SA 4.0 |