If you use this software tool in published work, please kindly cite it:
Author: Dongliang Ge, PhD
Each of the included methods in MetaP has its assumptions and conditions. In addition, each of the combined P values is based on one particular arbitrary distribution. Because of this, it is the author's view that the combined P values generated
here may be viewed as ROUGH PREDICTIONS, and can NOT be applied universally. Users are advised to study these methods before the interpretation of the results.
Whitlock, M. C. Combining probability from independent tests: the weighted Z-method is superior to Fisher's approach. J Evol Biol 18, 1368-73 (2005).
The author would like to thank Dr. Mike Weale at UCL and Dr. David B. Goldstein at Duke IGSP Center for Human Genome Variation for the help and support in the development of this software. The development of this software tool is also supported by Duke Institute for Genome Sciences & Policy, and Center for HIV/AIDS Vaccine Immunology. This software tool uses a JAVA library SSJ (Stochastic Simulation in Java) developed by a number of colleagues at
the Universite de Montreal.
For questions and comments, contact the author at:firstname.lastname@example.org
This software tool is free for public uses. The author keeps all the copyright.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
What does this program do?
MetaP performs a meta-analysis and combines the statistical association signals (P values) from independent studies or study populations,
taking account of the impacts of sample sizes and effect directions.
What is the system requirement to run this program?
MetaP is a JAVA applet. It needs your internet browser to be JAVA-enabled - which is usually (if not always) true in today's computer.
To test whether java is working in your browser, here is a useful link: http://www.java.com/en/download/help/testvm.xml
Which combined P value should I choose from the output?
MetaP outputs four different combined P values, using different levels of avilable information:
Conventional Fisher's: only considers P values;
Fisher's trend: considers P values and effect directions;
Stouffer's z: considers P values and sample sizes;
Stouffer's z trend: considers P values, sample sizes, and effect directions.
Geneally speaking you should look at the P value corrected for as many as factors as allowed by your data. That is, if you have both sample size and effect direction information available, the Stouffer's z trend is probably the most accurate one to look at.
See also: Combining probability from independent tests: the weighted Z-method is superior to Fisher's approach.