Re: st: Analyzing multiple mediators

[Steve Samuels <sjsamuels@gmail.com>]

Kim:

To answer your original question:  The error arose because you applied
the -if- expression only to -sureg- , not to the -bootmm- command.
-jackknife-, called by the "bca" option, assumed that every observation
was in the estimation sample.

The fix:
  1. After the -syntax- statement in your program,
     add the statement:  "marksample touse".
  2. In place of "if NMV==1", substitute "if `touse'".
  3. In the -bootstrap- statement, after "bootmm", add "if NMV==1".

I agree with Ariel that -khb- offers a more comprehensive approach.
The -mediation- package by Hicks and Tingley (at SSC) is also worth
considering, because it does a sensitivity analysis of an important
assumption.


Reference:
Hicks, R., and D. Tingley. 2011. Causal mediation analysis. Stata
Journal 11, no. 4: 605. A preprint is as
http://scholar.harvard.edu/dtingley/software/mediation-stata.

(You can often find preprints with Google Scholar).


Steve

On Nov 22, 2012, at 9:46 AM, Ariel Linden, DrPH wrote:

Kim, a much better approach for you to consider is -khb- ( a user-written
program : -findit khb-). You should read the accompanying Stata Journal
article since it gives nice examples of a multi-mediator scenario:

http://www.stata-journal.com/article.html?article=st0236

Ariel

Date: Wed, 21 Nov 2012 09:50:15 -0500
From: "Kim, Isok" <isokkim@buffalo.edu>
Subject: st: Analyzing multiple mediators

Hi,

I'm trying to analyze a model with multiple mediators, as described in ULCA
stata FAQ site (http://www.ats.ucla.edu/stat/stata/faq/mulmediation.htm).
The analyses are based on Preacher & Hayes (2008) method and uses
combination of -sureg- and -nlcom- commands to get the coefficients and
indirect effects coefficients, respectively. Then to get the standard
errors, it demonstrates use of ado-program called 'bootmm' and subsequent
bootstrap post-estimation command.

Where I'm having trouble with this combination of analyses is at the last
stage with bootstrap command. The error msg reads, "insufficient
observations to compute jackknife standard errors; no results will be saved"
r(2000).

There are 397 observations included in the analyses and -sureg- command runs
fine.  I'm not sure where the error is coming from.  Below is the syntax I
used. Any suggestion or insight into resolving this problem would be greatly
appreciated!

Thanks,
isok

*---------------------------------------------------------------------------
*/
/* H2a: Mediation Test IVs(PRD) MV(GC3E GC3D RC3E RC3D) DV(CESD)
CV(controls)*/
/* surge (mv1 iv cv)(mv2 iv2 cv)(dv mv1 mv2 iv cv)
*/
/* nlcom [mv1]_b[iv]*[dv]_b[mv1]
*/
/*--------------------------------------------------------------------------
-*/

/*-sureg-&-nlcom- methods*/

sureg (GC3E PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
     (GC3D PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
	  (RC3E PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
	  (RC3D PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
	  (CESD GC3E GC3D RC3E RC3D PRD i.SEX AGE EDU i.UNEMP i.MARSTS
i.NATIVITY) if NMV==1

//Indirect via GC3E
nlcom [GC3E]_b[PRD]*[CESD]_b[GC3E]
//Indirect via GC3D
nlcom [GC3D]_b[PRD]*[CESD]_b[GC3D]
//Indirect via GC3E
nlcom [RC3E]_b[PRD]*[CESD]_b[RC3E]
//Indirect via GC3D
nlcom [RC3D]_b[PRD]*[CESD]_b[RC3D]

//Total indirect
nlcom [GC3E]_b[PRD]*[CESD]_b[GC3E]+[GC3D]_b[PRD]*[CESD]_b[GC3D] ///
    +[RC3E]_b[PRD]*[CESD]_b[RC3E]+[RC3D]_b[PRD]*[CESD]_b[RC3D]

capture program drop bootmm
program bootmm, rclass
	syntax [if] [in]
	sureg (GC3E PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
         (GC3D PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
	      (RC3E PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
	      (RC3D PRD i.SEX AGE EDU i.UNEMP i.MARSTS i.NATIVITY) ///
	      (CESD GC3E GC3D RC3E RC3D PRD i.SEX AGE EDU i.UNEMP i.MARSTS
i.NATIVITY) if NMV==1
	return scalar indGC3E = [GC3E]_b[PRD]*[CESD]_b[GC3E]
	return scalar indGC3D = [GC3D]_b[PRD]*[CESD]_b[GC3D]
	return scalar indRC3E = [RC3E]_b[PRD]*[CESD]_b[RC3E]
	return scalar indRC3D = [RC3D]_b[PRD]*[CESD]_b[RC3D]
	return scalar indtotal =
[GC3E]_b[PRD]*[CESD]_b[GC3E]+[GC3D]_b[PRD]*[CESD]_b[GC3D] ///

+[RC3E]_b[PRD]*[CESD]_b[RC3E]+[RC3D]_b[PRD]*[CESD]_b[RC3D]
end

bootstrap r(indGC3E) r(indGC3D) r(indRC3E) r(indRC3D) r(indtotal), bca
reps(1000) nodots: bootmm
estat boot, percentile bc bca


*


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/