Hi Statalisters -
I am hoping that someone can help me understand a strange finding I am receiving while using the medeff procedure. I am seeking to identify whether the relationship between history of adverse reproductive health outcome (binary) and contraceptive method satisfaction (binary) is mediated by level of trust in provider (continuous). For the proportion total effect mediated, I was surprised to see that the 95% CI included values above/below 1.
My code is as follows:
medeff ///
(regress provtrust ipv age hisp black asianoth insmedicaid insprivate parous arm) ///
(logit bcrecomd_6mo_v1 ipv provtrust age hisp black asianoth insmedicaid insprivate parous arm decidetier2 decidetier3) ///
, treat(ipv) mediate(provtrust) sims(1000) vce(cluster blindsiteid)
}
And the output is as follows:
ipv
Using 0 and 1 as treatment values
Linear regression Number of obs = 1,062
F(9, 39) = 3.14
Prob > F = 0.0061
R-squared = 0.0209
Root MSE = .52319
(Std. err. adjusted for 40 clusters in blindsiteid)
------------------------------------------------------------------------------
| Robust
provtrust | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
ipv | -.0807202 .0356985 -2.26 0.029 -.1529272 -.0085132
age | .004571 .0080987 0.56 0.576 -.0118102 .0209521
hisp | -.068913 .033135 -2.08 0.044 -.1359349 -.0018911
black | -.1127631 .0677217 -1.67 0.104 -.2497432 .0242169
asianoth | -.0882541 .0584832 -1.51 0.139 -.2065476 .0300393
insmedicaid | -.0406858 .0477949 -0.85 0.400 -.1373602 .0559886
insprivate | -.0434941 .035422 -1.23 0.227 -.1151418 .0281536
parous | -.0859105 .0475882 -1.81 0.079 -.1821668 .0103458
arm | .0125716 .0398476 0.32 0.754 -.0680277 .093171
_cons | 3.200203 .1807392 17.71 0.000 2.834623 3.565782
------------------------------------------------------------------------------
Iteration 0: log pseudolikelihood = -623.61477
Iteration 1: log pseudolikelihood = -599.31587
Iteration 2: log pseudolikelihood = -598.98421
Iteration 3: log pseudolikelihood = -598.98401
Logistic regression Number of obs = 1062
Wald chi2(12) = 42.57
Prob > chi2 = 0.0000
Log pseudolikelihood = -598.98401 Pseudo R2 = 0.0395
(Std. err. adjusted for 40 clusters in blindsiteid)
------------------------------------------------------------------------------
| Robust
bcrecomd_6~1 | Coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
ipv | .0994067 .1666788 0.60 0.551 -.2272778 .4260911
provtrust | .5885296 .1325096 4.44 0.000 .3288156 .8482437
age | -.0059536 .0345329 -0.17 0.863 -.0736369 .0617297
hisp | .8431579 .182438 4.62 0.000 .485586 1.20073
black | .4811103 .2338404 2.06 0.040 .0227915 .9394292
asianoth | .3291927 .2872338 1.15 0.252 -.2337751 .8921606
insmedicaid | .6025237 .2304948 2.61 0.009 .1507621 1.054285
insprivate | .1713909 .1552445 1.10 0.270 -.1328827 .4756646
parous | -.5831224 .2186995 -2.67 0.008 -1.011766 -.1544792
arm | -.0735658 .1369856 -0.54 0.591 -.3420526 .1949211
decidetier2 | .0291987 .0994385 0.29 0.769 -.1656972 .2240947
decidetier3 | -.0472242 .0588196 -0.80 0.422 -.1625084 .06806
_cons | -1.108129 .9017834 -1.23 0.219 -2.875592 .6593345
------------------------------------------------------------------------------
(62 missing values generated)
(62 missing values generated)
(62 missing values generated)
------------------------------------------------------------------------------------
Effect | Mean [95% Conf. Interval]
-------------------------------+----------------------------------------------------
ACME1 | -.0086732 -.0181068 -.0012532
ACME0 | -.0090556 -.019245 -.0013109
Direct Effect 1 | .0193307 -.0398151 .0810223
Direct Effect 0 | .0189483 -.0390699 .0790609
Total Effect | .0102751 -.0498005 .071245
% of Total via ACME1 | -.2048171 -4.892001 3.403679
% of Total via ACME0 | -.213847 -5.10768 3.55374
Average Mediation | -.0088644 -.0188779 -.001286
Average Direct Effect | .0191395 -.0394425 .0800535
% of Tot Eff mediated | -.209332 -4.99984 3.478709
I have not used this procedure before, but based on what I've seen, the % of tot eff mediated is interpreted as a %, so I don't understand why would be negative and why would the 95% CI be so large (should it not max at 1?).
Thanks so much for any feedback you can provide.
Best,
Alison
I am hoping that someone can help me understand a strange finding I am receiving while using the medeff procedure. I am seeking to identify whether the relationship between history of adverse reproductive health outcome (binary) and contraceptive method satisfaction (binary) is mediated by level of trust in provider (continuous). For the proportion total effect mediated, I was surprised to see that the 95% CI included values above/below 1.
My code is as follows:
medeff ///
(regress provtrust ipv age hisp black asianoth insmedicaid insprivate parous arm) ///
(logit bcrecomd_6mo_v1 ipv provtrust age hisp black asianoth insmedicaid insprivate parous arm decidetier2 decidetier3) ///
, treat(ipv) mediate(provtrust) sims(1000) vce(cluster blindsiteid)
}
And the output is as follows:
ipv
Using 0 and 1 as treatment values
Linear regression Number of obs = 1,062
F(9, 39) = 3.14
Prob > F = 0.0061
R-squared = 0.0209
Root MSE = .52319
(Std. err. adjusted for 40 clusters in blindsiteid)
------------------------------------------------------------------------------
| Robust
provtrust | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
ipv | -.0807202 .0356985 -2.26 0.029 -.1529272 -.0085132
age | .004571 .0080987 0.56 0.576 -.0118102 .0209521
hisp | -.068913 .033135 -2.08 0.044 -.1359349 -.0018911
black | -.1127631 .0677217 -1.67 0.104 -.2497432 .0242169
asianoth | -.0882541 .0584832 -1.51 0.139 -.2065476 .0300393
insmedicaid | -.0406858 .0477949 -0.85 0.400 -.1373602 .0559886
insprivate | -.0434941 .035422 -1.23 0.227 -.1151418 .0281536
parous | -.0859105 .0475882 -1.81 0.079 -.1821668 .0103458
arm | .0125716 .0398476 0.32 0.754 -.0680277 .093171
_cons | 3.200203 .1807392 17.71 0.000 2.834623 3.565782
------------------------------------------------------------------------------
Iteration 0: log pseudolikelihood = -623.61477
Iteration 1: log pseudolikelihood = -599.31587
Iteration 2: log pseudolikelihood = -598.98421
Iteration 3: log pseudolikelihood = -598.98401
Logistic regression Number of obs = 1062
Wald chi2(12) = 42.57
Prob > chi2 = 0.0000
Log pseudolikelihood = -598.98401 Pseudo R2 = 0.0395
(Std. err. adjusted for 40 clusters in blindsiteid)
------------------------------------------------------------------------------
| Robust
bcrecomd_6~1 | Coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
ipv | .0994067 .1666788 0.60 0.551 -.2272778 .4260911
provtrust | .5885296 .1325096 4.44 0.000 .3288156 .8482437
age | -.0059536 .0345329 -0.17 0.863 -.0736369 .0617297
hisp | .8431579 .182438 4.62 0.000 .485586 1.20073
black | .4811103 .2338404 2.06 0.040 .0227915 .9394292
asianoth | .3291927 .2872338 1.15 0.252 -.2337751 .8921606
insmedicaid | .6025237 .2304948 2.61 0.009 .1507621 1.054285
insprivate | .1713909 .1552445 1.10 0.270 -.1328827 .4756646
parous | -.5831224 .2186995 -2.67 0.008 -1.011766 -.1544792
arm | -.0735658 .1369856 -0.54 0.591 -.3420526 .1949211
decidetier2 | .0291987 .0994385 0.29 0.769 -.1656972 .2240947
decidetier3 | -.0472242 .0588196 -0.80 0.422 -.1625084 .06806
_cons | -1.108129 .9017834 -1.23 0.219 -2.875592 .6593345
------------------------------------------------------------------------------
(62 missing values generated)
(62 missing values generated)
(62 missing values generated)
------------------------------------------------------------------------------------
Effect | Mean [95% Conf. Interval]
-------------------------------+----------------------------------------------------
ACME1 | -.0086732 -.0181068 -.0012532
ACME0 | -.0090556 -.019245 -.0013109
Direct Effect 1 | .0193307 -.0398151 .0810223
Direct Effect 0 | .0189483 -.0390699 .0790609
Total Effect | .0102751 -.0498005 .071245
% of Total via ACME1 | -.2048171 -4.892001 3.403679
% of Total via ACME0 | -.213847 -5.10768 3.55374
Average Mediation | -.0088644 -.0188779 -.001286
Average Direct Effect | .0191395 -.0394425 .0800535
% of Tot Eff mediated | -.209332 -4.99984 3.478709
I have not used this procedure before, but based on what I've seen, the % of tot eff mediated is interpreted as a %, so I don't understand why would be negative and why would the 95% CI be so large (should it not max at 1?).
Thanks so much for any feedback you can provide.
Best,
Alison
