Statalist


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

Re: st: Generalized Sign Test


From   Steven Samuels <sjhsamuels@earthlink.net>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Generalized Sign Test
Date   Tue, 7 Oct 2008 18:50:42 -0400

-
I should add that "sign test" and "generalized sign test" are not proper terms for what Mai wants to do. Mai wants to test the hypothesis in binomial data that the true proportion P = P0, a specified value, against H1: P ≠ P0. As I stated, Stata's -bitest- is designed to do this. I should have added that -ci- will provide a confidence interval for the proportion, which would be a useful complement to a p-value.

The sign test is a test for location with continuous, not categorical, data; it happens to use the binomial hypothesis test for inference. For example, the sign test may be used to test that the median of a distribution is equal to a certain value. It counts the number of observations which exceed the hypothesized median and ignores ties; thus, in contrast to Mai's problem, the test sample size may be less than the number of observations. The sign test can also test the equality of distributions for paired (X,Y) data, by testing the hypothesis that P(X>Y) = 1/2; form Z = X - Y and count the number of times Z exceeds 0. This version also ignores ties. The sign test is relatively simple to do because of the connection to the binomial distribution. However the same hypotheses can be tested more powerfully with Wilcoxon's signed rank sum test. See: P. Armitage: Statistical Methods in Medical Research, Wiley, 1971, pp 395-397.

Different questions: What are Mai's data and how is a null value to be "calculated from the estimation period"?

-Steve



The sign test is a nonparametric test applied to continuous data
-bitest-

On Oct 7, 2008, at 3:15 PM, mai7777 wrote:

Hi,
Is there a way in Stata to perform a generalized sign test which
allows the null hypothesis to be different from 0.5. I am using it for
an event study and I would like the null to be calculated from the
estimation period rather than a standard 0.5.
Thanks
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index