[Mediation-information] mediation clarification

[Kosuke Imai]

I'm not sure what you mean when you say "a model like that" but I think
this discrepancy comes from the nonlinearity of the model.  That is, your
mediation outcome model is probit.  In that case, the implied outcome model
given the treatment is not necessarily probit.  You could do a more
non-parametric approach as we discuss in our statistical science paper:
http://imai.princeton.edu/research/mediation.html  If you decide to pursue
that approach, one of my collaborators may be able to share his code with
you.  I'm ccing them.

Best,
Kosuke

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


On Wed, Mar 5, 2014 at 5:58 AM, Pietro Ferrari <FerrariP at iarc.fr> wrote:

>  Dear Kosuke,
>
>  am reading with great interest your work on mediation analysis,
> including the command *mediate* that you made available in R. In an
> effort to fully understand the different steps of your reasoning, I played
> around with the 'framing' data, using a simplified version, ie. omitting
> the adjusting factors (age + educ + gender + income).
>
>  What bothers me a bit is tat I do not fully follow the expression for
> tau_i described at page 769 of Imai et al., APSR, 2011:
>
>  tau_i = Y_i(1,M(1)) - Y_i(0,M(0)).
>
>  I was under the impression that a total effect was the one estimated in
> a model with the variable treat as the only predictor. When a model like
> that is fitted, the parameter estimate is equal to 0.288, while the output
> of the *mediate* analysis suggests that the total effect is 0.09986.
>
>  It would great if you could provide some insights to clarify where my
> mistake resides.
> You find below the basic analytical steps I followed.
>
>  Thank you in advance for your help. I really appreciate it.
>
>  All the best, Pietro
>
>
>  &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
>
>  install.packages("mediation")
> install.packages("sandwich")
> library("mediation")
> library("MASS")
> data("framing")
>
>  med.fit <- lm(emo ~ treat , data = framing)
> summary(med.fit)
>
>  out.fit <- glm(cong_mesg ~ treat + emo,
>                data = framing, family = binomial("probit"))
> summary(out.fit)
>
>  library("sandwich")
> med.out <- mediate(med.fit, out.fit, treat = "treat", mediator = "emo",
>                    robustSE = TRUE, sims = 100)
>
>  summary(med.out)
>
>
>
>
>
>
>
>
>
>
>
>
>
> *
> ------------------------------------------------------------------------------------------------
> This message and its attachments are strictly confidential. If you are not
> the intended recipient of this message, please immediately notify the
> sender and delete it. Since its integrity cannot be guaranteed, its content
> cannot involve the sender's responsibility. Any misuse, any disclosure or
> publication of its content, either whole or partial, is prohibited,
> exception made of formally approved use.
> ------------------------------------------------------------------------------------------------*
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/mediation-information/attachments/20140305/2f0a146c/attachment.html>