Continuity Correction at Wilcoxon Rank Sum Test

[Christopher]

Dear Eviews Team,

I am calculating the Wilcoxon Rank Sum Test and - as Eviews only offers p-values corrected for continuity - I wonder whether there are any size-related (or other) restrictions on continuity correction. For example, Newbold, Carlson, Thorne (1995), p. 201/202 suggest to correct for continuity when 5 < n * p (1-p) < 9, with p being the probability for success in a binomial distribution. Thus, if my sample is larger than 40 (assuming a 0.5 probability), would I achieve more accurate results not using a continuity correction factor? How would I deal with this in Eviews? I appreciate your thoughts.

Kind regards

Christopher

[EViews Glenn]

The continuity correction is a simply subtraction of .5 from numerator of the statistic. My quick glance at Sheskin tells me that the denominator is sqrt(n(n+1)(2n+1)/24). So it's effect on the final z-stat is **roughly** .5*sqrt(24)/n^1.5 = 2.45/n^1.5. So for a sample of 40, you've got a term of roughly .01 subtracted from the z-stat. Bottom line is that it probably won't matter for any reasonable hypothesis test you'd do unless you are right on the border of significance.

If you want, you could, however, add it back in then compute the p-value yourself. But we don't offer an option.

[Christopher]

Thank you very much, very helpful.