My outcome variable is binary and my IV is also binary. My mediation is continuous. Do I do the same thing as you did but for the b path, I conduct a binary logistic regression? Is that correct? Thank you.
You cannot use Sobel Test on logistic regression. The Sobel Test is only applicable to continuous X, Y and M connected by linear regression models, while logistic regression models are nonlinear. Therefore, the math does not work out.
@@MINGJIPHD Hi, yes, the paper is called "The unresponsive bystander: Are bystanders more responsive in dangerous emergencies?" It's from 2006. You will find at the end, that they did a few mediations and the final one was with a binary outcome. They do say that they found partial mediation but to me it seems that it should have been full mediation (the IV became non-significant after the mediator was added). Can you have a look please?
@@valamoss I will try to find this paper. But you have a PDF file of it, could you email it to me at qxsr@yahoo.com? I will review it and give you my input. Thanks!
Not completely by itself. The test only examines the mediation path and not the c' path from X to Y. For a complete mediation, X and Y should be independent when conditioned on M. This characterization is superior to the simple mediation model used in the Sobel Test.
@@MINGJIPHD Thank you. Now I understand that the Sobel test only shows if there is a significant mediation. I did my mediation analysis in PROCESS in spss but I came back to the old time of doing a mediation when things were simpler.
@@MINGJIPHD My PROCESS results are as follows: In the first model, the IV (X) significantly predicts the mediator (M). In the second model, the mediator significantly predicts the outcome variable (Y) but the IV (X) does not predict the outcome variable (Y). In the third model, the direct effect of X on Y is non-significant. The indirect effect of X on Y is significant. I know that this means that I have full mediation but I want to be able to say that the effect of the IV decreased after adding the mediator. PROCESS does not show this first IV and outcome relationship like regression would. In the old times of the Sobel Test, the first regression model would show a significant relationship between the IV and the outcome. Then, in the next model, when the mediator was added, it could be seen if the IV's effect decreased as a result of this and whether its effect was still significant. This would show us whether there was mediation and if it was partial or complete. PROCESS also does not compute a total effect when the outcome is binary. So I went back to the old times and ran a logistic regression for this relationship. I ran a linear regression for the IV to mediator relationship. And then I ran another logistic regression adding both the IV and the mediator as predictors of the binary outcome. Then I did a Sobel Test.
Just wanted to let you know that I appreciate the point you made about things being simpler. It got me thinking about the mediation model for the Soble Test, which happens to be the simplest mediation model. I haven't used PROCESS before, but I checked out Hayes' website and it seems like there are different mediation models available. I'm not sure which one would be suitable for your analysis though. To be honest, I find mediation analysis to be too ambitious. Not only does it claim causal effects, but it also makes statements about causal pathways. It's worth keeping in mind that to confirm one causal relationship - say between a drug and a disease outcome - we need to run a Randomized Controlled Trial (RCT). In a mediation analysis, we're only testing or estimating associations. If your Dependent Variable (DV), Independent Variable (IV), or Mediator is binary, the Sobel Test might be too simple to apply. It's possible that more research needs to be done to find or even create a more appropriate analytic approach. I did a quick search and came across a paper related to mediation analysis using binary outcomes, which you might find helpful. www.ncbi.nlm.nih.gov/pmc/articles/PMC6544488/ Good luck with your exploration of mediation analysis! Let me know if you have any further questions.
thanks for short video but easy to understand
My outcome variable is binary and my IV is also binary. My mediation is continuous. Do I do the same thing as you did but for the b path, I conduct a binary logistic regression? Is that correct? Thank you.
You cannot use Sobel Test on logistic regression.
The Sobel Test is only applicable to continuous X, Y and M connected by linear regression models, while logistic regression models are nonlinear. Therefore, the math does not work out.
@@MINGJIPHD Hi, I saw it used on a binary outcome (Y) in a published paper so it is possible.
@@valamoss Could you point me to where I can find that paper? It may be possible but not using Sobel Test.
@@MINGJIPHD Hi, yes, the paper is called "The unresponsive bystander: Are bystanders more responsive in dangerous emergencies?" It's from 2006. You will find at the end, that they did a few mediations and the final one was with a binary outcome. They do say that they found partial mediation but to me it seems that it should have been full mediation (the IV became non-significant after the mediator was added). Can you have a look please?
@@valamoss I will try to find this paper. But you have a PDF file of it, could you email it to me at qxsr@yahoo.com? I will review it and give you my input. Thanks!
Thanks, very informative video. Can we use it in a model with 2 Independents, 1 Mediator, and 1 Depedents?
That require more complicated path models. Sobel's test needs to be generalized.
Hi, thanks for the informative video on sobel test. May I know what is the test statistics indicating? Thanks!
It's a t statistics
Does the Sobel test show whether the mediation is partial or complete?
Not completely by itself. The test only examines the mediation path and not the c' path from X to Y. For a complete mediation, X and Y should be independent when conditioned on M. This characterization is superior to the simple mediation model used in the Sobel Test.
@@MINGJIPHD Thank you. Now I understand that the Sobel test only shows if there is a significant mediation. I did my mediation analysis in PROCESS in spss but I came back to the old time of doing a mediation when things were simpler.
@@MINGJIPHD My PROCESS results are as follows:
In the first model, the IV (X) significantly predicts the mediator (M).
In the second model, the mediator significantly predicts the outcome variable (Y) but the IV (X) does not predict the outcome variable (Y).
In the third model, the direct effect of X on Y is non-significant. The indirect effect of X on Y is significant.
I know that this means that I have full mediation but I want to be able to say that the effect of the IV decreased after adding the mediator. PROCESS does not show this first IV and outcome relationship like regression would.
In the old times of the Sobel Test, the first regression model would show a significant relationship between the IV and the outcome. Then, in the next model, when the mediator was added, it could be seen if the IV's effect decreased as a result of this and whether its effect was still significant. This would show us whether there was mediation and if it was partial or complete.
PROCESS also does not compute a total effect when the outcome is binary.
So I went back to the old times and ran a logistic regression for this relationship. I ran a linear regression for the IV to mediator relationship. And then I ran another logistic regression adding both the IV and the mediator as predictors of the binary outcome. Then I did a Sobel Test.
Just wanted to let you know that I appreciate the point you made about things being simpler. It got me thinking about the mediation model for the Soble Test, which happens to be the simplest mediation model. I haven't used PROCESS before, but I checked out Hayes' website and it seems like there are different mediation models available. I'm not sure which one would be suitable for your analysis though.
To be honest, I find mediation analysis to be too ambitious. Not only does it claim causal effects, but it also makes statements about causal pathways. It's worth keeping in mind that to confirm one causal relationship - say between a drug and a disease outcome - we need to run a Randomized Controlled Trial (RCT). In a mediation analysis, we're only testing or estimating associations.
If your Dependent Variable (DV), Independent Variable (IV), or Mediator is binary, the Sobel Test might be too simple to apply. It's possible that more research needs to be done to find or even create a more appropriate analytic approach. I did a quick search and came across a paper related to mediation analysis using binary outcomes, which you might find helpful.
www.ncbi.nlm.nih.gov/pmc/articles/PMC6544488/
Good luck with your exploration of mediation analysis! Let me know if you have any further questions.
While doing the test my p-value is coming to 0.0. What does it mean?
It is not actually 0. Computers have limits on how many digits they can store for a very small number.
Can we use this for panel data and how to add control variables while doing mediation analysis?
Same question please answer this can we use this for panel data? And how can we use control variable in mediation analysis?
@@safiasuleman4591 yes we can use it for panel data
@@safiasuleman4591 adding controls in mediation is similar as we do in simple regression analysis
@@riffatshaheen1970 thanks for your reply
Not really