From gp366 at scarletmail.rutgers.edu Fri Aug 11 17:17:01 2017 From: gp366 at scarletmail.rutgers.edu (GREG PORUMBESCU) Date: Fri, 11 Aug 2017 10:17:01 -0500 Subject: [Mediation-information] Trouble with mediate.ped function in Mediation for R Message-ID: Dear All, My colleagues and I are trying to analyze data coming from a parallel encouragement design using the mediate.ped function in the mediate R package. Our encouragement, time, has three levels: 1 (long time), -1 (short time) and 0 (no time limit). We have two questions: 1) We have followed the instructions provided in Tingley et al. 2014, but when we run the syntax, the lower and upper bound confidence intervals on all of the ACME report 0. These estimates hold even when we used different (binary) mediators. The code we are using and the output is as follows: > ped<- mediate.ped("PERFORMANCE_binary", "sum_obund_new_binary", "negframe", "time", DFC_coded) > summary(ped) Design-Based Causal Mediation Analysis Parallel Encouragement Design Lower Bound Upper Bound Population ACME (control) 0 0 Complier ACME (control) 0 0 Population ACME (treated) 0 0 Complier ACME (treated) 0 0 Sample Size Used: 610 Would you have any idea what is causing our estimates to behave this way? Any advice on how to resolve this issue? 2) In our code, we transform our outcome variable, PERFORMANCE, into a binary variable in order to run mediate.ped, as the instructions indicate. However, is there any way to run mediate.ped with a continuous outcome variable? If not, are there any plans to allow for this in the future? Many thanks, Greg Best wishes, Gregory A. Porumbescu Assistant Professor School of Public Affairs and Administration Rutgers University Newark https://spaa.newark.rutgers.edu/gregory-porumbescu https://scholar.google.com/scholar?hl=en&q=gregory+porumbescu&btnG=&as_sdt=1%2C14&as_sdtp=&oq=gre ? -------------- next part -------------- An HTML attachment was scrubbed... URL: From teppei at mit.edu Sat Aug 12 07:38:21 2017 From: teppei at mit.edu (Teppei Yamamoto) Date: Fri, 11 Aug 2017 22:38:21 -0700 Subject: [Mediation-information] Trouble with mediate.ped function in Mediation for R In-Reply-To: References: Message-ID: <3a7dc5a5-3a6d-24ab-74d3-c1d2d7ce054d@mit.edu> Hi Greg, The function can produce those values for the bounds when your data are not consistent with the identification assumptions, such as the consistency assumption or the exclusion restrictions. It could also be due to some mistake in the data (e.g. miscoding in a variable). If you send me your code and the dataset (maybe a subset that reproduces the same problem, if you cannot send the whole thing), I'll be happy to look into it to the extent I can. We unfortunately don't have a plan to extend it to a continuous outcome variable, primarily because our theoretical framework doesn't simply generalize to such a setup. There are alternative approaches that could potentially achieve what we need, though, so we (or someone else) might someday get to it -- not on our immediate agenda unfortunately! Best, Teppei ------------------------------------------ Teppei Yamamoto Associate Professor of Political Science Alfred Henry and Jean Morrison Hayes Chair Massachusetts Institute of Technology http://web.mit.edu/teppei/www/ ------------------------------------------ On 8/11/17 08:17, GREG PORUMBESCU wrote: > Dear All, > > My colleagues and I are trying to analyze data coming from a parallel > encouragement design using the mediate.ped function in the mediate R > package. Our encouragement, time, has three levels: 1 (long time), -1 > (short time) and 0 (no time limit). We have two questions: > > 1) We have followed the instructions provided in Tingley et al. 2014, > but when we run the syntax, the lower and upper bound confidence > intervals on all of the ACME report 0. These estimates hold even when we > used different (binary) mediators. The code we are using and the output > is as follows: > > > ped<- mediate.ped("PERFORMANCE_binary", "sum_obund_new_binary", > "negframe", "time", DFC_coded) > > summary(ped) > > Design-Based Causal Mediation Analysis > > Parallel Encouragement Design > > Lower Bound Upper Bound > Population ACME (control) 0 0 > Complier ACME (control) 0 0 > Population ACME (treated) 0 0 > Complier ACME (treated) 0 0 > > Sample Size Used: 610 > > > Would you have any idea what is causing our estimates to behave this > way? Any advice on how to resolve this issue? > > 2) In our code, we transform our outcome variable, PERFORMANCE, into a > binary variable in order to run mediate.ped, as the instructions > indicate. However, is there any way to run mediate.ped with a continuous > outcome variable? If not, are there any plans to allow for this in the > future? > > Many thanks, > > Greg > > Best wishes, > > Gregory A. Porumbescu > Assistant Professor > School of Public Affairs and Administration > Rutgers University Newark > https://spaa.newark.rutgers.edu/gregory-porumbescu > > https://scholar.google.com/scholar?hl=en&q=gregory+porumbescu&btnG=&as_sdt=1%2C14&as_sdtp=&oq=gre > > > > ? > > > _______________________________________________ > 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 > From kimai at princeton.edu Sun Aug 13 03:42:15 2017 From: kimai at princeton.edu (Kosuke Imai) Date: Sat, 12 Aug 2017 21:42:15 -0400 Subject: [Mediation-information] Trouble with mediate.ped function in Mediation for R In-Reply-To: <3a7dc5a5-3a6d-24ab-74d3-c1d2d7ce054d@mit.edu> References: <3a7dc5a5-3a6d-24ab-74d3-c1d2d7ce054d@mit.edu> Message-ID: You can also do a modeling approach based on two-stage least squares. In your case, this might be the easiest. Take a look at page at the end of page 783 and the beginning of page 784 of this paper for discussion: http://imai.princeton.edu/research/files/mediationP.pdf Kosuke Imai Professor, Department of Politics Center for Statistics and Machine Learning Princeton University http://imai.princeton.edu On Sat, Aug 12, 2017 at 1:38 AM, Teppei Yamamoto wrote: > Hi Greg, > > The function can produce those values for the bounds when your data are > not consistent with the identification assumptions, such as the consistency > assumption or the exclusion restrictions. It could also be due to some > mistake in the data (e.g. miscoding in a variable). If you send me your > code and the dataset (maybe a subset that reproduces the same problem, if > you cannot send the whole thing), I'll be happy to look into it to the > extent I can. > > We unfortunately don't have a plan to extend it to a continuous outcome > variable, primarily because our theoretical framework doesn't simply > generalize to such a setup. There are alternative approaches that could > potentially achieve what we need, though, so we (or someone else) might > someday get to it -- not on our immediate agenda unfortunately! > > Best, > Teppei > > ------------------------------------------ > Teppei Yamamoto > > Associate Professor of Political Science > Alfred Henry and Jean Morrison Hayes Chair > Massachusetts Institute of Technology > > http://web.mit.edu/teppei/www/ > ------------------------------------------ > > > On 8/11/17 08:17, GREG PORUMBESCU wrote: > >> Dear All, >> >> My colleagues and I are trying to analyze data coming from a parallel >> encouragement design using the mediate.ped function in the mediate R >> package. Our encouragement, time, has three levels: 1 (long time), -1 >> (short time) and 0 (no time limit). We have two questions: >> >> 1) We have followed the instructions provided in Tingley et al. 2014, but >> when we run the syntax, the lower and upper bound confidence intervals on >> all of the ACME report 0. These estimates hold even when we used different >> (binary) mediators. The code we are using and the output is as follows: >> >> > ped<- mediate.ped("PERFORMANCE_binary", "sum_obund_new_binary", >> "negframe", "time", DFC_coded) >> > summary(ped) >> >> Design-Based Causal Mediation Analysis >> >> Parallel Encouragement Design >> >> Lower Bound Upper Bound >> Population ACME (control) 0 0 >> Complier ACME (control) 0 0 >> Population ACME (treated) 0 0 >> Complier ACME (treated) 0 0 >> >> Sample Size Used: 610 >> >> >> Would you have any idea what is causing our estimates to behave this way? >> Any advice on how to resolve this issue? >> >> 2) In our code, we transform our outcome variable, PERFORMANCE, into a >> binary variable in order to run mediate.ped, as the instructions indicate. >> However, is there any way to run mediate.ped with a continuous outcome >> variable? If not, are there any plans to allow for this in the future? >> >> Many thanks, >> >> Greg >> >> Best wishes, >> >> Gregory A. Porumbescu >> Assistant Professor >> School of Public Affairs and Administration >> Rutgers University Newark >> https://spaa.newark.rutgers.edu/gregory-porumbescu >> >> https://scholar.google.com/scholar?hl=en&q=gregory+porumbesc >> u&btnG=&as_sdt=1%2C14&as_sdtp=&oq=gre >> >> >> >> ? >> >> >> _______________________________________________ >> 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 >> >> _______________________________________________ > 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 -------------- next part -------------- An HTML attachment was scrubbed... URL: