From law826 at gmail.com Mon Mar 11 20:39:40 2013 From: law826 at gmail.com (Lawrence Ngo) Date: Mon, 11 Mar 2013 15:39:40 -0400 Subject: [Mediation-information] Nonparametric Bootstrap Message-ID: Dear Dr. Imai, Thanks so much for putting together this useful mediation package. I have been having an issue re-running a nonparametric bootstrap mediation analysis. Comparing notes from the old and new analysis that I am doing on the same data (and both using 10000 simulations), it looks like the confidence intervals and p-values are different across the two analyses. I am fairly confident that nothing has changed in my data, since all of the parameter estimates are identical across old and new. Do you have any insight on why this may be the case? I ran the old analysis around 6 months ago. Thanks in advance for you help! Best, Lawrence -------------- next part -------------- An HTML attachment was scrubbed... URL: From teppei at MIT.EDU Tue Mar 12 14:19:02 2013 From: teppei at MIT.EDU (Teppei Yamamoto) Date: Tue, 12 Mar 2013 09:19:02 -0400 Subject: [Mediation-information] Nonparametric Bootstrap In-Reply-To: References: Message-ID: <513F2B46.1030907@mit.edu> Hi Lawrence, We've slightly changed the way p-values are calculated as of Version 4.1.1, which on my record was released on August 23, 2012. This change will generally not make big differences, except when your effect estimates are very close to zero (to the extent that R cannot tell the difference for some of your bootstrap draws) and/or when the bootstrap distributions are highly asymmetric. We didn't change the way confidence intervals are calculated, so I do not have a good sense of what could have gone wrong there. Best, Teppei (3/11/13 3:39 PM), Lawrence Ngo wrote: > Dear Dr. Imai, > > Thanks so much for putting together this useful mediation package. I > have been having an issue re-running a nonparametric bootstrap mediation > analysis. Comparing notes from the old and new analysis that I am doing > on the same data (and both using 10000 simulations), it looks like the > confidence intervals and p-values are different across the two analyses. > I am fairly confident that nothing has changed in my data, since all of > the parameter estimates are identical across old and new. > > Do you have any insight on why this may be the case? I ran the old > analysis around 6 months ago. Thanks in advance for you help! > > Best, > Lawrence > > > > _______________________________________________ > Mediation-information mailing list > Mediation-information at lists.r-forge.r-project.org > https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/mediation-information -- ==================================== Teppei Yamamoto Assistant Professor Department of Political Science Massachusetts Institute of Technology http://web.mit.edu/teppei/www/ ====================================