From kimai at Princeton.Edu  Sun Oct 12 18:24:26 2014
From: kimai at Princeton.Edu (Kosuke Imai)
Date: Sun, 12 Oct 2014 12:24:26 -0400
Subject: [Mediation-information] Question about mediate
In-Reply-To: <0B6C21CA5FBE1747BC94ABCCEF3E56B85AC1201D@XMAIL-MBX-BT1.AD.UCSD.EDU>
References: <0B6C21CA5FBE1747BC94ABCCEF3E56B85AC10CD2@XMAIL-MBX-BT1.AD.UCSD.EDU>
 <CAFiYZ5xsuV=XEbb2emTYR1XFvjD9dWtn8Y-4DHk-OjgyKTGXeA@mail.gmail.com>
 <0B6C21CA5FBE1747BC94ABCCEF3E56B85AC1201D@XMAIL-MBX-BT1.AD.UCSD.EDU>
Message-ID: <CAFiYZ5zxw4ekmy+=Zb+AS6XCp+pw3DgcCRtr-5gtxrdDc5OKvw@mail.gmail.com>

Dear John,

  yes, we have updated the software significantly since 2010 and now it
should incorporate the ordered mediator and outcome models.  The package
can also handle weights when fitting each of the models.

Best,
Kosuke


Kosuke Imai
Department of Politics
Princeton University
http://imai.princeton.edu

On Sun, Oct 12, 2014 at 9:36 AM, Evans, John <jhevans at ucsd.edu> wrote:

>  Dear Kosuke,
>
> Thanks for your quick response.  I would like to ask a follow up question
> please.  I am looking at your paper called ?causal mediation analysis using
> R? which is Chapter 8 of a book.  Table 8.1 on page 137 says that the R
> version can handle an ordered mediator, but not an ordered outcome.
> Moreover, if the mediator is ordered, the outcome must be treated as
> continuous.   But, in your forthcoming paper in the Journal of Statistical
> Software it says in Table 1 on page that the program can handled ordered
> mediator when matched with an ordered outcome.  If I am reading that right,
> can I presume you and your colleagues have improved the R version between
> 2010 and now?
>
>
>
> Before I read these papers on R more carefully, one final question: can
> the R package (of which I know nothing) handle population weights?  If not,
> that might be reason to stick to Stata, despite having to either use binary
> or continuous models.
>
>
>
> Thanks for your help.
>
> John
>
>
>
> John H. Evans
>
> Professor of Sociology
>
> University of California, San Diego
>
> http://sociology.ucsd.edu/faculty/JohnHEvans.shtml
>
>
>
> *From:* Kosuke Imai [mailto:kimai at Princeton.Edu]
> *Sent:* Saturday, October 11, 2014 8:12 AM
> *To:* Evans, John
> *Subject:* Re: Question about mediate
>
>
>
> Dear John,
>
>
>
>   Thanks for your email and interests in my work.  Unfortunately, there is
> no immediate plan to extend the Stata software.  However, we have been
> working on the development of our R package and we made it easy even for
> non-R users.   And so, I encourage you to use the R package if that is
> possible.  Otherwise, I think you have to go with a binary model or a
> linear model, neither of which is optimal but may still give you estimates
> that are close to what you get with more appropriate modeling strategies.
>
>
>
> Best,
>
> Kosuke
>
>
>
> ---------------------------------------------------------
> Kosuke Imai               Office: Corwin Hall 036
> Professor                 Phone: 609-258-6601
> Department of Politics    Fax: 609-258-1110
> Princeton University      Email: kimai at Princeton.Edu
> Princeton, NJ 08544-1012  http://imai.princeton.edu
> ---------------------------------------------------------
>
>
>
> On Sat, Oct 11, 2014 at 10:00 AM, Evans, John <jhevans at ucsd.edu> wrote:
>
>  Dear Professor Imai,
>
> I have not made claims in my career so far that require mediation models.
> However, I now realize that I might have to do so for a project I am
> working on.  I read your co-authored 2010 paper (in Psychological Methods)
> and the 2011 paper (in APSR) with interest.  I use Stata.
>
>
>
> All of the critical variables I would be using (X, M, Y) are ordered
> categorical, typically with 4 to 7 categories, from public opinion data,
> that need to have population weights.  In the APSR paper you imply that
> your method could be used in an ordered logistic (or ordered probit)
> environment.  But, the Stata program ?mediate? only accepts linear and
> logistic models.  Do you know if there are any plans underway to expand
> this program to ordered logistic models?  Is there an alternative?  My back
> up plan if I go the mediation route would be to dichotomize all of these
> variables, but that would be throwing away a lot of information.
>
>
>
> Thank you for any advice you may have.
>
>
>
> Best wishes,
>
> John
>
>
>
> John H. Evans
>
> Professor of Sociology
>
> University of California, San Diego
>
> http://sociology.ucsd.edu/faculty/JohnHEvans.shtml
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/mediation-information/attachments/20141012/0ed5b26f/attachment.html>