Dear Statalisters,
I am trying to calculate a mediation in a logistic regression. My IV is categorical (ordered / Likert scale), my MV is categorical (ordered / Likert scale) and my IV is binary. I used the following command:
sgmediation DV, iv(xx) mv(xy) cv(control 1-9)
bootstrap r(ind_eff), reps(10000) level(95) nodots: sgmediation DV, iv(xx) mv(xy) cv(control 1-9)
estat bootstrap, percentile
Enclosed my questions and the results:
- Is it possible that the results of the mediation are negative (see proportion of total effect that is mediated)?
- Do I need to standardize my IV/MV before the mediation?
- Is the mediation still significant, since the bootstrapping results show p>0.1?
Thank you very much in advance for your help in interpreting these results - I'm new to mediation analysis and not sure how these methods can be applied in logistic regression.
Best regards
Christina
Sobel-Goodman Mediation Tests
Coef Std Err Z P>|Z|
Sobel -.00643056 .00408734 -1.573 .11565284
Goodman-1 (Aroian) -.00643056 .00426169 -1.509 .13131891
Goodman-2 -.00643056 .00390522 -1.647 .09962865
Coef Std Err Z P>|Z|
a coefficient = .114706 .060355 1.90051 .057366
b coefficient = -.056061 .01999 -2.80448 .00504
Indirect effect = -.006431 .004087 -1.57329 .115653
Direct effect = .070538 .020426 3.45342 .000554
Total effect = .064107 .020541 3.12093 .001803
Proportion of total effect that is mediated: -.10030904
Ratio of indirect to direct effect: -.09116443
Ratio of total to direct effect: .90883557
However, with bootstrapping, the results do not seem to be significant anymore:
Observed Bootstrap Normal based
Coef. Std. Err. z P>z [95% Conf. Interval]
_bs_1 -.0064306 .0043352 -1.48 0.138 -.0149274 .0020663
Results from command "Estat bootstrap, percentile":
Observed Bootstrap
Coef. Bias Std. Err. [95% Conf. Interval]
_bs_1 .00643056 -.0002737 .00433522 -.0167019 -.0001188 (P)
I am trying to calculate a mediation in a logistic regression. My IV is categorical (ordered / Likert scale), my MV is categorical (ordered / Likert scale) and my IV is binary. I used the following command:
sgmediation DV, iv(xx) mv(xy) cv(control 1-9)
bootstrap r(ind_eff), reps(10000) level(95) nodots: sgmediation DV, iv(xx) mv(xy) cv(control 1-9)
estat bootstrap, percentile
Enclosed my questions and the results:
- Is it possible that the results of the mediation are negative (see proportion of total effect that is mediated)?
- Do I need to standardize my IV/MV before the mediation?
- Is the mediation still significant, since the bootstrapping results show p>0.1?
Thank you very much in advance for your help in interpreting these results - I'm new to mediation analysis and not sure how these methods can be applied in logistic regression.
Best regards
Christina
Sobel-Goodman Mediation Tests
Coef Std Err Z P>|Z|
Sobel -.00643056 .00408734 -1.573 .11565284
Goodman-1 (Aroian) -.00643056 .00426169 -1.509 .13131891
Goodman-2 -.00643056 .00390522 -1.647 .09962865
Coef Std Err Z P>|Z|
a coefficient = .114706 .060355 1.90051 .057366
b coefficient = -.056061 .01999 -2.80448 .00504
Indirect effect = -.006431 .004087 -1.57329 .115653
Direct effect = .070538 .020426 3.45342 .000554
Total effect = .064107 .020541 3.12093 .001803
Proportion of total effect that is mediated: -.10030904
Ratio of indirect to direct effect: -.09116443
Ratio of total to direct effect: .90883557
However, with bootstrapping, the results do not seem to be significant anymore:
Observed Bootstrap Normal based
Coef. Std. Err. z P>z [95% Conf. Interval]
_bs_1 -.0064306 .0043352 -1.48 0.138 -.0149274 .0020663
Results from command "Estat bootstrap, percentile":
Observed Bootstrap
Coef. Bias Std. Err. [95% Conf. Interval]
_bs_1 .00643056 -.0002737 .00433522 -.0167019 -.0001188 (P)
