Hi Emma, You are very welcome. I am delighted to hear that I was able to help someone else out there. It motivates me to post more videos like these. Cheers!
Hi Jocelyn, thank you for making my life so much easier :) Been trying to get my head around this all day. I'm so happy there are people like you making these awesome videos! Em xx
Hey Jocelyn... i would to thank you so much for this very informative video... you literally saved my life.... i was so confused about performing mediation analysis for my research... may god bless you with all good things in your life.... can't thank you enough..... lots of love from Malaysia~
Thank you for your question Emma! If the A and B paths are significant, but the indirect effect is not, it is possible that you lack statistical power to detect the effect. If possible, try to increase the sample size. My guess is that your confidence interval is very close to being significant. Have you tried running the analysis again using 5000 samples? There are some small variations among bootstraps and that might help you get the confidence interval that you need.
HI! Thank you for asking this important question. There are many ways of getting standardized betas. You could (1) run Multiple Regression Analyses for every path or (2) standardize your variables before running the analysis. The latter solution will provide you with standardized coefficients, however the bootstrap confidence interval would not be those of the standardized effects. Remember also that standardized effects of dichotomous predictors do not yield very substantive interpretation.
Hi there! Thank you for your comment. Traditionally, Baron and Kenny (1986) made the argument that the C-path is necessary to test mediation. However, in a more recent paper (see Kenny, Kashy, Bolger, 1998), Kenny suggested that the C-path was no longer necessary in order to test for mediation. The latter argument has caught on and it is not uncommon that researchers test the ab path even if the c-path is not significant. After all, mediation is primarily about the ab path, not the c-path.
Hi Hazman, Thank you for your question! I am much surprised by Preacher's comment. I am pretty sure he meant the C-path was optional (i.e., only the ab path is mandatory). I personally cannot conceive how mediation is possible without the a-path. At least something needs to be significant! ;)
Hi Ye, including control variables is very simple. In the INDIRECT SPSS window, insert your control variables (e.g., demographics) in the Covariate(s) box. Et voilà!
What an awesome video. Thanks a lot! This thorough explanation is helping me a lot for my thesis and made me understand the entire process way faster. I have a model IV: network configurations DV: performance M: entrepreneurial orientation the IV was originally built up from two constructs, namely network size and network diversity network size makes for a great mediation model network diversity certainly not should I do both, explain the issue, then drop diversity and just use only size?
Hi! The idea is that if the Confidence Interval includes 0 then the indirect effect is not significant. This is the case with your CI. A simple trick: If the confidence interval includes two different signs + and -, then it is not significant. If the CI has two similar signs e.g., - and -, or + and +, then the indirect effect is significant. Hope that helps!
Hi Emma It is possible that 80 cases is too little. Although I have seen significant indirect effect with fewer cases. It all boils down to the notion of effect size. I would recommend increasing the sample size if the effect your examining is low in magnitude. Good luck!
hi Jocelyn thank you for this video. please i have a question. how can i conduct mediation when i have 2 independent variables, 1 dependent variable and 1 mediator variable?
Hi Jocelyn, I find the video a good addition to my materials for my students. I have one question though. Recently, hayes mentioned in his workshop on mediation that significance of the a path is not necessary to determine mediation. Like to hear your comments on this. My opinion is that if the requirement of a significance a path is not required, then the mediation could just be viewed as another independent variable.
Hey Jocelyn, thank you so much for the reply! I wanted to see whether there were indirect effects separately for males and females so split the sample, and then there was only 80 cases for males- perhaps that's too small a sample? This was with 5000 bootstrap re-samples too. CI was -.07 - .56 so I don't know what's going on... Can't wait for the regression ;-) Thanks again, Em x
Hi Jocelyn, your show is very helpful to my research, many thanks!!! However, How to add in some control variables in the mediation effects, such as demographics?
Thank you very much for an excellent & well explained tutorial! May I ask a question regarding reporting confidence intervals please? I have been given a handout with mediation analysis assumption checking & bootstrapping, but at BOTH 95% & 99% CI to do an APA write-up on. I cannot work out which CI I should report. Looking at the indirect effect of x on y, both the LLCI & ULCI for 95% & 99% are negative values (so don't include zero). The sample size is 120. Any advice please? Thanks
I'd love you to post more videos like these, on hierarchical regression please XD Could I be cheeky and ask a question? I've just conducted the analysis in the vid, and the a and b paths are significant, but 0 is included in the confidence intervals. So I was wondering how you would report this? Do you say that the mediation model is non-significant, even though the a and b paths are significant? Would really appreciate any help :) Thanks, Em x
Very nice tutorial Jocelyn, thanks for posting this. In the path modelling, issues mostly lie in interpreting the results (as usual). For example, how would you interpret this outcome: A-path significant, B-path non-significant, C-path (total effect) significant, and C1-path (direct effect of IV on DV), non significant?
In that case you would simply say that the A and C paths are significant, but not the b-path, which suggests that the proposed mediator is not the mechanism at play.
Hi! I'm glad you found the video helpful. Could you please send me one of the paper you are referring to? I'd be happy to take a look at it to help you out.
Thanks Jocelyn, was a great help ! The only problem I have is that the coefficients differ with the coefficients that I get when I do a regression analysis on the individual paths. Do you have any idea how that's possible? Regards Omar
Hi! I'm glad I could help. This a difficult question to answer. Perhaps it is of interest to demonstrate that only network size is related to performance, but ultimately it depends on your theory and your hypotheses. Could you explain this difference? Do you have extra information to support that claim? Welcome to the world of academic writing. Convince me! :)
Hi Jocelyn, this is a very good video. However, following Justin question, could I ask if the procedure for bootstrapping is the same when there are 3 levels of IV. My independent variable is level of helpfulness with 3 different level. My mediator is negative affect and there is a curvilinear trend. Will this method still work?
Hi Jocelyn, Your explanation was very clear and understandable! Much appreciated. I am currently in the process of determining what meditational analyses to use for my dissertation. My independent variable is racial identification. I plan to incorporate three racial groups. Is bootstrapping still applicable when utilizing 3 groups? Thanks again for the video!
HI! Thank you for your question. There is nothing preventing you from testing three different models, but my preference would be a model including all three independent variables simultaneously, especially if they are correlated. This approach could be tested using path analysis or Structural Equation Modeling with more advanced statistical packages such as AMOS, Lisrel, or EQS.
Hi Alyssa, I would suggest you look into the following reference: Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior research methods, 40(3), 879-891.
could i use 90% interval instead of 95%. the C' path would be significant in that case. i mean 90& of accuracy for a small sample of 125 is good isnt it?
Hi Mahdi, You could use a 90% interval instead of a 95% one. It's still pretty good. However, conventionally, scientific publication (at least in Psychology) requires a 95% interval.
The betas from the indirect dialog and standard regression analyses in SPSS should be identical. I suspect you have missing values and that your regression analyses have different degrees of freedom.
Hi Emma,
You are very welcome. I am delighted to hear that I was able to help someone else out there. It motivates me to post more videos like these.
Cheers!
Thank you so much for this video! You covered every parts that someone who is looking for this video for sure has to know and do.
Hi Jocelyn, thank you for making my life so much easier :) Been trying to get my head around this all day. I'm so happy there are people like you making these awesome videos!
Em xx
Hey Jocelyn... i would to thank you so much for this very informative video... you literally saved my life.... i was so confused about performing mediation analysis for my research... may god bless you with all good things in your life.... can't thank you enough..... lots of love from Malaysia~
Thank you for your question Emma! If the A and B paths are significant, but the indirect effect is not, it is possible that you lack statistical power to detect the effect. If possible, try to increase the sample size. My guess is that your confidence interval is very close to being significant. Have you tried running the analysis again using 5000 samples? There are some small variations among bootstraps and that might help you get the confidence interval that you need.
Hi Jocelyn,
Thank you for the wonderful explanation and video clip. It's very information and well organized.
HI!
Thank you for asking this important question. There are many ways of getting standardized betas. You could (1) run Multiple Regression Analyses for every path or (2) standardize your variables before running the analysis. The latter solution will provide you with standardized coefficients, however the bootstrap confidence interval would not be those of the standardized effects. Remember also that standardized effects of dichotomous predictors do not yield very substantive interpretation.
Hi there!
Thank you for your comment. Traditionally, Baron and Kenny (1986) made the argument that the C-path is necessary to test mediation. However, in a more recent paper (see Kenny, Kashy, Bolger, 1998), Kenny suggested that the C-path was no longer necessary in order to test for mediation. The latter argument has caught on and it is not uncommon that researchers test the ab path even if the c-path is not significant. After all, mediation is primarily about the ab path, not the c-path.
Hi Hazman,
Thank you for your question! I am much surprised by Preacher's comment. I am pretty sure he meant the C-path was optional (i.e., only the ab path is mandatory). I personally cannot conceive how mediation is possible without the a-path. At least something needs to be significant! ;)
The video was unbelievably helpful. I might actually finish my research by the deadline. Thank you!!!!!!!
Hi Justin,
Bootstrapping can be used if you are comparing groups, no problem!
Hi Ye,
including control variables is very simple. In the INDIRECT SPSS window, insert your control variables (e.g., demographics) in the Covariate(s) box. Et voilà!
Great video! Especially the tips about how to describe it in a thesis were very nice!
Hi Jocelyn! Very nice video...clear and useful, thanks a lot!
Thank you so much for this informative, considerable demonstration.
What an awesome video. Thanks a lot! This thorough explanation is helping me a lot for my thesis and made me understand the entire process way faster.
I have a model
IV: network configurations
DV: performance
M: entrepreneurial orientation
the IV was originally built up from two constructs, namely network size and network diversity
network size makes for a great mediation model
network diversity certainly not
should I do both, explain the issue, then drop diversity and just use only size?
Hi!
The idea is that if the Confidence Interval includes 0 then the indirect effect is not significant. This is the case with your CI. A simple trick: If the confidence interval includes two different signs + and -, then it is not significant. If the CI has two similar signs e.g., - and -, or + and +, then the indirect effect is significant. Hope that helps!
thanks again Jocelyn. i am using Smart PLS.
Hi Emma
It is possible that 80 cases is too little. Although I have seen significant indirect effect with fewer cases. It all boils down to the notion of effect size. I would recommend increasing the sample size if the effect your examining is low in magnitude. Good luck!
hi Jocelyn thank you for this video. please i have a question. how can i conduct mediation when i have 2 independent variables, 1 dependent variable and 1 mediator variable?
Thank you dear Jocelyn for this productive and informative lecture, I wish you all the best. Keep the good work up :)
Hi Jocelyn, I find the video a good addition to my materials for my students. I have one question though. Recently, hayes mentioned in his workshop on mediation that significance of the a path is not necessary to determine mediation. Like to hear your comments on this. My opinion is that if the requirement of a significance a path is not required, then the mediation could just be viewed as another independent variable.
Hey Jocelyn, thank you so much for the reply! I wanted to see whether there were indirect effects separately for males and females so split the sample, and then there was only 80 cases for males- perhaps that's too small a sample? This was with 5000 bootstrap re-samples too. CI was -.07 - .56 so I don't know what's going on...
Can't wait for the regression ;-)
Thanks again,
Em x
Hi Jocelyn, your show is very helpful to my research, many thanks!!!
However, How to add in some control variables in the mediation effects, such as demographics?
Thanks! You're video is very instructive.
Thank you very much for an excellent & well explained tutorial!
May I ask a question regarding reporting confidence intervals please?
I have been given a handout with mediation analysis assumption checking & bootstrapping, but at BOTH 95% & 99% CI to do an APA write-up on. I cannot work out which CI I should report. Looking at the indirect effect of x on y, both the LLCI & ULCI for 95% & 99% are negative values (so don't include zero). The sample size is 120. Any advice please? Thanks
I'd love you to post more videos like these, on hierarchical regression please XD
Could I be cheeky and ask a question? I've just conducted the analysis in the vid, and the a and b paths are significant, but 0 is included in the confidence intervals. So I was wondering how you would report this? Do you say that the mediation model is non-significant, even though the a and b paths are significant?
Would really appreciate any help :)
Thanks,
Em x
Hi!
I've never tried with Smart PLS. However, if it is an issue, you can perform bootstrapping using the free SPSS trial version.
Very nice tutorial Jocelyn, thanks for posting this.
In the path modelling, issues mostly lie in interpreting the results (as usual). For example, how would you interpret this outcome: A-path significant, B-path non-significant, C-path (total effect) significant, and C1-path (direct effect of IV on DV), non significant?
In that case you would simply say that the A and C paths are significant, but not the b-path, which suggests that the proposed mediator is not the mechanism at play.
Hi! I'm glad you found the video helpful. Could you please send me one of the paper you are referring to? I'd be happy to take a look at it to help you out.
Thanks Jocelyn, was a great help ! The only problem I have is that the coefficients differ with the coefficients that I get when I do a regression analysis on the individual paths. Do you have any idea how that's possible? Regards Omar
Hi jocelyn,
Thank you very much
Hi!
I'm glad I could help. This a difficult question to answer. Perhaps it is of interest to demonstrate that only network size is related to performance, but ultimately it depends on your theory and your hypotheses. Could you explain this difference? Do you have extra information to support that claim? Welcome to the world of academic writing. Convince me! :)
Hi Jocelyn,
this is a very good video. However, following Justin question, could I ask if the procedure for bootstrapping is the same when there are 3 levels of IV. My independent variable is level of helpfulness with 3 different level. My mediator is negative affect and there is a curvilinear trend. Will this method still work?
Hi Jocelyn,
Your explanation was very clear and understandable! Much appreciated. I am currently in the process of determining what meditational analyses to use for my dissertation. My independent variable is racial identification. I plan to incorporate three racial groups. Is bootstrapping still applicable when utilizing 3 groups?
Thanks again for the video!
Thank you very much for this amazing video!!!!!!!!!!!
Jocelyn thank you very much for the video. I am using multiple IVs, one mediator and 1 DV, can I treat each IV-M-DV as separate simple mediation?
HI! Thank you for your question. There is nothing preventing you from testing three different models, but my preference would be a model including all three independent variables simultaneously, especially if they are correlated. This approach could be tested using path analysis or Structural Equation Modeling with more advanced statistical packages such as AMOS, Lisrel, or EQS.
hi Abdulrahman, i also have a model that is like your. with 2 IV, IM and IDV. please how did you eventually do your analysis?
You are very welcome!
outstanding job, thanks. keep up the good work!
simply you are the best!!!!
Hi,
The value 0 is included in the interval -.0057; .1556, therefore the indirect effect is not significant.
Great video many thanks
Thank you, thank you, thank you!
You are more than welcome!
what if c path is not significant? (p>.05)
helpful lesson! thank you!
Any idea on how to do it with categorical?? with dummies is not working.. and i don't know why
Hi, great video. You say to bootstrap to 5000. Why? What literature suggests this. I'm looking for a source, thanks :)
Hi Alyssa,
I would suggest you look into the following reference:
Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior research methods, 40(3), 879-891.
Jocelyn Bélanger Thanks!
So helpful! Thanks:)
could i use 90% interval instead of 95%. the C' path would be significant in that case. i mean 90& of accuracy for a small sample of 125 is good isnt it?
Hi Mahdi,
You could use a 90% interval instead of a 95% one. It's still pretty good. However, conventionally, scientific publication (at least in Psychology) requires a 95% interval.
Dear Jocelyn,
Thanks for the reply. which result is more accurate? results obtained from macros or hierarchical regression?
Bootstrapping results from macros are more accurate because they correct for bias.
Thanks
The betas from the indirect dialog and standard regression analyses in SPSS should be identical. I suspect you have missing values and that your regression analyses have different degrees of freedom.
Thanks a lot mate !!!
at 6:30 you say "unstandardized Beta coefficients" don't you mean "standardized" ? thanks!
Hi Sean,
Thanks for your comment. The betas are unstandardized, unless you centered your variables before running the analyses.
thank you!!
Re: Hierarchical Regression: Challenge Accepted!
Many Thanks for your work it was really useful. Will contact if I need your help in the process.
KRs.