Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Re: st: Ranksum test for many events

From   "Airey, David C" <>
To   "<>" <>
Subject   Re: Re: st: Ranksum test for many events
Date   Thu, 19 Dec 2013 13:32:58 +0000


Agreed. Thanks.

Reading more closely, he mentions two groups, before and after,
so time is involved, but there appears to be only 2 time points.

To use -ranksum-, you'd have to form the differences of before 
and after, to get a single value for each firm, and then apply

If you have more than 2 time values per firm, you'll need to use
another approach, that handles multiple time points but that also
focuses on the median.


> If the raw data are time series, standard P-values are probably
> worthless as they rest on an assumption of independence.
> Nick
> On 19 December 2013 12:47, Airey, David C <> wrote:
> > .
> >
> > You can use -statsby- with -ranksum- to collect p-values by group
> > and then apply a multiple test correction procedure using -qqvalue-
> > from SSC on the resulting p-values.
> >
> >
> >> Dear All
> >> I want to run a ranksum test to compare the median changes in a figure
> >> between two group of firms before and after some 100 events. The
> >> number and combination of firms in each group changes from  event to
> >> event. I was wondering if I need to run the test for each event
> >> separately or there is a way to do it for all events at once. As far
> >> as I understand, one can only introduce a dummy group variable in
> >> ranksum command in Stata, and I don't see how 'in' or 'if' can be used
> >> to run the test for all events at once.
> >> Best
> >> Nima
> >

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index