Wilcoxon Signed Rank Test with two Series?

[Deninhoo]

Hello,

I have two series. If I group them I can only do a simple t-test or a Wilcoxon Mann Whitney Test.
I want to test whether the difference between the means of two series is significantly different or not.

Is it possible to substract series X from series Y or the other way round and then with just one series do the wilcoxon signed rank test?

thanks you very much!

regards
Deni

[EViews Glenn]

I don't believe so. The unpaired form Is a slightly different form for the test as it's testing for median 0. To be honest, I think the easiest way is just to compute it directly.

[Deninhoo]

Hi Glenn,

thanks for your reply.
What do you mean with "compute it directly"?

I have another question: I want to estimate a regression where one independent variable is constant (total expense ratio). If I estimate the regression EViews shows a matrix singular error.. Do you know if it is possible to estimate a regression with a constant indep. variable?

Thanks!!

[EViews Glenn]

It's a pretty simply calculation.

Off the top of my head. Compute the difference series as you've described. Rank the non-zero absoulte differences. Create a new series which is equal to +/- 1 where the sign matches the difference series. Compute the absolute value of the sign weighted ranks. For a full sample calculation, something like

Code: Select all

smpl @all series x = y1 - y2 ' compute the difference smpl if x<>0 ' use the non-zero pairs series r = @ranks(@abs(x)) ' rank by absolute difference series sgn = @recode(x>0, 1, -1) ' obtain the signs series rsgn = r*sgn ' compute the sign weighted ranks scalar ranksum = @abs(@sum(rsgn)) ' take the absolute sum

You'll then have to compare to the values in a table. You should be able to find critical values on the net.

As to your other question. Only if you don't also have an intercept.