Accessing results of indirect effects after SEM

[Roman Mostazir]

Dear all,

I am trying to format the results after Structural Equation Model (SEM). The results of direct effects can easily be called by programs like 'estout' or 'esttab' or outreg2 and be formatted according to the taste. The problem I am facing is after I run 'teffects' for indirect and total effects. The information stored after 'teffects' command are the followings:

return list

scalars:
           r(N_groups) =  1

matrices:
            r(V_total) :  35 x 35
              r(total) :  1 x 35
         r(V_indirect) :  35 x 35
           r(indirect) :  1 x 35
           r(V_direct) :  35 x 35
             r(direct) :  1 x 35
               r(nobs) :  1 x 1

The usual programmes like estout, outreg2, do not recognise these matrices as they read matrices from the original model. Can anyone suggest, after 'teffects' command, how to access the indirect effect coefficient, 95% CI and 'p' values and be formated as per requirements.

Your help is much appreciated.
Regards,
Roman

[David Sikkink]

Roman,

Were you able to find an answer to this question?

Also, do you know anything about how to get estout to handle results after sem multiple group analysis?

Thanks.

--David

[David Sikkink]

I should add: when running estout after sem multiple group analysis, I get this error:

. estout test, cells(b(star fmt(3)) se(par fmt(2)))
1bn: operator invalid

"1bn:" is the prefix that labels the coefficients for group 1 in the multiple group analysis.

Thanks for any help anyone can offer.

--David

[David Sikkink]

It seems that estadd should place the r matrices from the estat teffects postestimation command into the stored results of e (which are stored after an sem multiple group command). I don't get an error when I try to do this, but I haven't been able to get estout to include the added stats in the table as I would like.

Ideally, I would place the total and indirect coefficients in columns adjacent to the direct effects. Estout handles the direct effects from the sem regression just fine (sorry for the earlier email on this -- I found out that I had an old verison of estout installed and after upgrading to the new version, estout handled the results from an sem multiple group command very well). So I have one column with all the variables and coefficients for each model and for each of the two groups in my sem analysis. So far so good.

But now: how to place the coefficients for the total and direct effects, which are available from the estat teffects postestimation command, as columns right next to the appropriate direct effect.

It is true that estat teffects does print the direct effects as well as the indirect and total effects. So perhaps rather than "merging" e and r class results, I should try to get all the r class results into something estout can read.

Any guidance would be greatly appreciated.

--David



But somehow need to inform estout what these new

[Richard Williams]

This may be clunkier than necessary but I think it is in the ballpark. This could all be put in a do file and run and you could just keep changing the sem command, or you can break the te_ programs into separate ado files. Note that you could add things to the ereturns, e.g. the model chi-square values if you wanted to. Since such things are redundant you may just want to store them with the direct results.

capture program drop te_direct
program te_direct, eclass
    quietly estat teffects
    mat b = r(direct)
    mat V = r(V_direct)
    local N = e(N)
    ereturn post b V, obs(`N')
    ereturn local cmd te_direct 
end

capture program drop te_indirect
program te_indirect, eclass
    quietly estat teffects
    mat b = r(indirect)
    mat V = r(V_indirect)
    ereturn post b V
    ereturn local cmd te_indirect 
end

capture program drop te_total
program te_total, eclass
    quietly estat teffects
    mat b = r(total)
    mat V = r(V_total)
    ereturn post b V
    ereturn local cmd te_total 
end


webuse nhanes2f, clear
sem (height black female -> weight)(female black -> height)
est store main
te_direct
est store direct
est restore main
te_indirect
est store indirect
est restore main
te_total
est store total
esttab direct indirect total, mtitles(direct indirect total)

. esttab direct indirect total, mtitles(direct indirect total)

------------------------------------------------------------
                      (1)             (2)             (3)   
                   direct        indirect           total   
------------------------------------------------------------
height                                                      
black               0.398               0           0.398   
                   (1.80)             (.)          (1.80)   

female             -13.52***            0          -13.52***
                 (-99.32)             (.)        (-99.32)   
------------------------------------------------------------
weight                                                      
height              0.662***            0           0.662***
                  (34.69)             (.)         (34.69)   

black               3.490***        0.264           3.754***
                   (8.11)          (1.79)          (8.26)   

female             -2.664***       -8.955***       -11.62***
                  (-7.21)        (-32.75)        (-41.62)   
------------------------------------------------------------
N                   10337                                   
------------------------------------------------------------

[David Sikkink]

That works perfectly. In an age of divisiveness, Prof Williams has gone the extra mile to bring the e class and the r class closer together. That is a beautiful thing.

Thanks again to my now even more highly esteemed colleague!

--David

[Richard Williams]

Getting back to Roman's original Q, you might want to tweak the te_ programs to include a replay option:

capture program drop te_direct
program te_direct, eclass
    if replay() {
        ereturn display
        exit
    }
    quietly estat teffects
    mat b = r(direct)
    mat V = r(V_direct)
    local N = e(N)
    ereturn post b V, obs(`N')
    ereturn local cmd te_direct
    est store direct
end

capture program drop te_indirect
program te_indirect, eclass
    if replay() {
        ereturn display
        exit
    }
    quietly estat teffects
    mat b = r(indirect)
    mat V = r(V_indirect)
    ereturn post b V
    ereturn local cmd te_indirect
    est store indirect
end

capture program drop te_total
program te_total, eclass
    if replay() {
        ereturn display
        exit
    }
    quietly estat teffects
    mat b = r(total)
    mat V = r(V_total)
    ereturn post b V
    ereturn local cmd te_total
    est store total
end


webuse nhanes2f, clear
sem (height black female -> weight)(female black -> height)
est store main
te_direct

est restore main
te_indirect

est restore main
te_total

esttab direct indirect total, mtitles(direct indirect total)

est replay direct
est replay indirect
est replay total

. est replay direct

-------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model direct
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
height       |
       black |   .3979381   .2216452     1.80   0.073    -.0364786    .8323548
      female |  -13.51765   .1360978   -99.32   0.000     -13.7844   -13.25091
-------------+----------------------------------------------------------------
weight       |
      height |   .6624414   .0190954    34.69   0.000     .6250151    .6998677
       black |   3.490113   .4303801     8.11   0.000     2.646583    4.333642
      female |  -2.664485   .3693839    -7.21   0.000    -3.388464   -1.940505
------------------------------------------------------------------------------

. est replay indirect

-------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model indirect
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
height       |
       black |          0  (omitted)
      female |          0  (omitted)
-------------+----------------------------------------------------------------
weight       |
      height |          0  (omitted)
       black |   .2636107   .1470235     1.79   0.073    -.0245501    .5517714
      female |  -8.954654   .2734169   -32.75   0.000    -9.490541   -8.418767
------------------------------------------------------------------------------

. est replay total

-------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model total
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
height       |
       black |   .3979381   .2216452     1.80   0.073    -.0364786    .8323548
      female |  -13.51765   .1360978   -99.32   0.000     -13.7844   -13.25091
-------------+----------------------------------------------------------------
weight       |
      height |   .6624414   .0190954    34.69   0.000     .6250151    .6998677
       black |   3.753723   .4546729     8.26   0.000     2.862581    4.644866
      female |  -11.61914   .2791849   -41.62   0.000    -12.16633   -11.07195
------------------------------------------------------------------------------

[Roman Mostazir]

Many thanks for the solution Richard. I actually forgot how I went about solving this problem when I made this post one and half year back. But your solution definitely is a relief for the future.
All the best.

[Richard Williams]

Just be careful of ereturned results you create this way. It may not be legitimate for using them for much of anything besides displaying results, e.g. I don't know if you could use test commands or margins or predict or whatever. It might depend on whether you have ereturned enough information.

[Nicola Babamov]

Thank you very much, Richard Williams!

Is there a way to modify the code so that it shows the standardized coefficients? I tried adding the option 'standardized' after the 'quietly estat teffects'-command and than adding the option 'beta' after the 'esttab'-command but the output gave out a red error message.

Thanks for any help anyone can offer.

[Attaullah Shah]

asdocx offers a convenient way to export the table of total, direct, and indirect effects to MS Word, Excel, LaTeX, or HTML. asdocx compiles various statistics from the stored matrices and creates a complete table of coefficient, standard errors, z-statistics, and p-values. Here is one example (More examples and details of the options are discussed on this page https://fintechprofessor.com/asdocx/...l-latex-stata/ )

* Load another dataset
use https://stats.idre.ucla.edu/stat/data/hsbdemo, clear

sem (read <- write math ses socst ) ///
    (science <- read math awards ) ///
    (write <- math read awards )

asdocx estat teffects, effects(te de ie) tzok replace

Two more tables produced (total effects and direct effects). However, they are not reported here. The above example does not follow any theory for model building. It just shows the use of asdocx with estat teffects.

[Yasushi Hazama]

Dear Prof. Richard Williams,

I always benefit tremendously from your valuable advice.
Recently I have been running the exact commands posted 24 Oct 2016, 09:59.
However, in the end I have the error message "estimation result direct not found".
Do you think that this is because of some changes in STATA settings or problems with my PC?
Below I have copied and pasted the results of my running the commands posted 24 Oct 2016, 09:59.
If you see them, strangely, only direct effects have been produced; the indirect and total effects are the same as the direct effects.
I would appreciate it very much if you could show me the way out.
Best regards,
Yasushi

. capture program drop te_direct

. program te_direct, eclass
1. if replay() {
2. ereturn display
3. exit
4. }
5. quietly estat teffects
6. mat b = r(direct)
7. mat V = r(V_direct)
8. local N = e(N)
9. ereturn post b V, obs(`N')
10. ereturn local cmd te_direct
11. est store direct
12. end

.
. capture program drop te_indirect

. program te_indirect, eclass
1. if replay() {
2. ereturn display
3. exit
4. }
5. quietly estat teffects
6. mat b = r(indirect)
7. mat V = r(V_indirect)
8. ereturn post b V
9. ereturn local cmd te_indirect
10. est store indirect
11. end

.
. capture program drop te_total

. program te_total, eclass
1. if replay() {
2. ereturn display
3. exit
4. }
5. quietly estat teffects
6. mat b = r(total)
7. mat V = r(V_total)
8. ereturn post b V
9. ereturn local cmd te_total
10. est store total
11. end

.
.
.
. sem (height black female -> weight)(female black -> height)

Endogenous variables
Observed: height weight

Exogenous variables
Observed: black female

Fitting target model:
Iteration 0: log likelihood = -86090.901
Iteration 1: log likelihood = -86090.901

Structural equation model Number of obs = 10,337
Estimation method: ml

Log likelihood = -86090.901

------------------------------------------------------------------------------
| OIM
| Coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
Structural |
height |
black | .3979381 .2216452 1.80 0.073 -.0364786 .8323548
female | -13.51765 .1360978 -99.32 0.000 -13.7844 -13.25091
_cons | 174.7076 .1011681 1726.90 0.000 174.5093 174.9059
-----------+----------------------------------------------------------------
weight |
height | .6624414 .0190954 34.69 0.000 .6250151 .6998677
black | 3.490113 .4303801 8.11 0.000 2.646583 4.333642
female | -2.664485 .3693839 -7.21 0.000 -3.388464 -1.940505
_cons | -38.12579 3.341889 -11.41 0.000 -44.67577 -31.57581
-------------+----------------------------------------------------------------
var(e.height)| 47.7417 .6640727 46.45773 49.06117
var(e.weight)| 179.9492 2.50304 175.1096 184.9226
------------------------------------------------------------------------------
LR test of model vs. saturated: chi2(0) = 0.00 Prob > chi2 = .

. est store main

. te_direct
-------------------------------------------------------------------------------
| OIM
| Coefficient std. err. z P>|z| [95% conf. interval]
--------------+----------------------------------------------------------------
height |
black | .3979381 .2216452 1.80 0.073 -.0364786 .8323548
female | -13.51765 .1360978 -99.32 0.000 -13.7844 -13.25091
_cons | 174.7076 .1011681 1726.90 0.000 174.5093 174.9059
--------------+----------------------------------------------------------------
weight |
height | .6624414 .0190954 34.69 0.000 .6250151 .6998677
black | 3.490113 .4303801 8.11 0.000 2.646583 4.333642
female | -2.664485 .3693839 -7.21 0.000 -3.388464 -1.940505
_cons | -38.12579 3.341889 -11.41 0.000 -44.67577 -31.57581
--------------+----------------------------------------------------------------
var(e.height)| 47.7417 .6640727 71.89 0.000 46.44014 49.04326
var(e.weight)| 179.9492 2.50304 71.89 0.000 175.0434 184.8551
-------------------------------------------------------------------------------

.
. est restore main
(results main are active now)

. te_indirect
-------------------------------------------------------------------------------
| OIM
| Coefficient std. err. z P>|z| [95% conf. interval]
--------------+----------------------------------------------------------------
height |
black | .3979381 .2216452 1.80 0.073 -.0364786 .8323548
female | -13.51765 .1360978 -99.32 0.000 -13.7844 -13.25091
_cons | 174.7076 .1011681 1726.90 0.000 174.5093 174.9059
--------------+----------------------------------------------------------------
weight |
height | .6624414 .0190954 34.69 0.000 .6250151 .6998677
black | 3.490113 .4303801 8.11 0.000 2.646583 4.333642
female | -2.664485 .3693839 -7.21 0.000 -3.388464 -1.940505
_cons | -38.12579 3.341889 -11.41 0.000 -44.67577 -31.57581
--------------+----------------------------------------------------------------
var(e.height)| 47.7417 .6640727 71.89 0.000 46.44014 49.04326
var(e.weight)| 179.9492 2.50304 71.89 0.000 175.0434 184.8551
-------------------------------------------------------------------------------

.
. est restore main
(results main are active now)

. te_total
-------------------------------------------------------------------------------
| OIM
| Coefficient std. err. z P>|z| [95% conf. interval]
--------------+----------------------------------------------------------------
height |
black | .3979381 .2216452 1.80 0.073 -.0364786 .8323548
female | -13.51765 .1360978 -99.32 0.000 -13.7844 -13.25091
_cons | 174.7076 .1011681 1726.90 0.000 174.5093 174.9059
--------------+----------------------------------------------------------------
weight |
height | .6624414 .0190954 34.69 0.000 .6250151 .6998677
black | 3.490113 .4303801 8.11 0.000 2.646583 4.333642
female | -2.664485 .3693839 -7.21 0.000 -3.388464 -1.940505
_cons | -38.12579 3.341889 -11.41 0.000 -44.67577 -31.57581
--------------+----------------------------------------------------------------
var(e.height)| 47.7417 .6640727 71.89 0.000 46.44014 49.04326
var(e.weight)| 179.9492 2.50304 71.89 0.000 175.0434 184.8551
-------------------------------------------------------------------------------

.
. esttab direct indirect total, mtitles(direct indirect total)
estimation result direct not found
r(111);

end of do-file

r(111);

.

[Richard Williams]

I'd completely forgotten about this thread, let alone recall that I had contributed to it!

My code in post #5 continues to work fine for me. Try it instead of the code that had the replay option in it.

NOTE: Your output would be much easier to read if you used code tags.

[Yasushi Hazama]

Dear Prof. Richard Williams,

Thank you very much for your generous and valuable help, as always.
As you suggested, I tried your code in post #5 that had the replay option in it, and then it worked perfectly.
I am sorry for bothering you with my lack of understanding.
Best regards,
Yasushi

[Richard Williams]

I wish I could edit posts from 8 years ago! If future generations see this thread, use my code from post #5, not post #7. Or, maybe the strategy outlined in post # 11 is best — I haven’t tried it but it sounds promising.