Hi, thank you for the helpful video. I was wondering please if you have any advice for what to do if there is no prior research to estimate the effect size for one of the paths? For example, I have coefficents from prior research for all paths except path b1 in the serial mediation analysis
One possible alternative would be to ask: What is the smallest meaningful effect size? What effect size would be relevant in the real world? (That, of course, depends on your area of research).
@@RegorzStatistik Thank you for your quick response! If you have easily to hand, would you please be able to point me in the direction of a reference for using the smallest meaningful effect size, when there is no effect size from prior research available? Even if the reference is not specific to my area (psychology) it would still be useful for me to read. I had a look myself but couldn't really find anything.
Technisch kann man mit N = 80 eine Mediation rechnen, aber die Power ist halt recht klein, so dass man nur einen relativ großen Effekt mit diesem Design finden kann.
Do you have any suggestions for calculating the sample size for a parallel mediation with four mediators? It seems this app only has the capacity for up to three mediators.
Hello Regorz, many thanks for all the really helpful videos! I was wondering about how to best approach a post hoc power analysis of a serial mediation model which was conducted in PROCESS. Using the standardized coefficients provided by the macro yields very different power results (.82) compared to using the correlations between variables (.55). Unfortunately, I did not find any literature/lectures by Schoemann et al. on the use of standardized coefficients. What approach would you recommend? Further, is it possible to use the joint significance approach you explain in a different video to assess the power of a1db2? Best
Unfortunately, I don't know where the difference comes from. And I know of no literature applying the joint significance approach to a serial mediation, so there I'd have to answer "I don't know", too.
Hi, in my master's thesis, I am investigating a parallel mediation model. In another thesis, which also examines a parallel mediation model, the sample size was first calculated using a one-tailed bivariate correlation analysis for the relationship between the independent variable and the dependent variable with G*Power. Subsequently, the sample size for the mediation analysis was determined using the MARlab simulation tool. Now I am wondering whether the first step is necessary at all or what its purpose is. Could you please help me with this? Thank you very much!
Hi, thank you for you video. Once we press "calculate power" and receive the results, could you please tell me what is the meaning of the lines "difference" ? Thank you
Hi, thanks for the great video. However, I have 2 predictors,2 mediators, and 3 outcome variables. I plan to run serial mediation analysis using process macro. How do I determine the sample number? Would your method here apply to me?
hi, thanks for the information. However, I have 2 predictors,2 mediators, and 1outcome variables. I plan to run serial mediation analysis using AMOS. How do I determine the sample size? Would your method here apply to me? the total population is 804, how can I take the representative sample size from these population?
I think for that you would have to run a power analysis for a path model since the tool works only with one predictor, I believe. There is an R package for that, semPower, but I haven't worked with that, yet.
Determining the effect sizes is the hardest part in power analysis. Possible sorces for that information: 1 Research for similar hypotheses 2 What is the smallest effect that is practically important?
Super helpful thanks for sharing!
thank you, you are an angel
Hi, thank you for the helpful video. I was wondering please if you have any advice for what to do if there is no prior research to estimate the effect size for one of the paths? For example, I have coefficents from prior research for all paths except path b1 in the serial mediation analysis
One possible alternative would be to ask: What is the smallest meaningful effect size? What effect size would be relevant in the real world? (That, of course, depends on your area of research).
@@RegorzStatistik Thank you for your quick response! If you have easily to hand, would you please be able to point me in the direction of a reference for using the smallest meaningful effect size, when there is no effect size from prior research available? Even if the reference is not specific to my area (psychology) it would still be useful for me to read. I had a look myself but couldn't really find anything.
@@dw6873 Unfortunately, I don't have a specific source for that.
was tun wenn ich nur n 80 habe jedoch eine Mediation berechnen will geht das dann nicht ?
Technisch kann man mit N = 80 eine Mediation rechnen, aber die Power ist halt recht klein, so dass man nur einen relativ großen Effekt mit diesem Design finden kann.
Do you have any suggestions for calculating the sample size for a parallel mediation with four mediators? It seems this app only has the capacity for up to three mediators.
Unfortunately, no, I don't have any recommendations for that.
Hello Regorz, many thanks for all the really helpful videos! I was wondering about how to best approach a post hoc power analysis of a serial mediation model which was conducted in PROCESS. Using the standardized coefficients provided by the macro yields very different power results (.82) compared to using the correlations between variables (.55). Unfortunately, I did not find any literature/lectures by Schoemann et al. on the use of standardized coefficients. What approach would you recommend? Further, is it possible to use the joint significance approach you explain in a different video to assess the power of a1db2? Best
Unfortunately, I don't know where the difference comes from. And I know of no literature applying the joint significance approach to a serial mediation, so there I'd have to answer "I don't know", too.
@@RegorzStatistik Thanks a lot for the swift reply!
Hi, in my master's thesis, I am investigating a parallel mediation model. In another thesis, which also examines a parallel mediation model, the sample size was first calculated using a one-tailed bivariate correlation analysis for the relationship between the independent variable and the dependent variable with G*Power. Subsequently, the sample size for the mediation analysis was determined using the MARlab simulation tool. Now I am wondering whether the first step is necessary at all or what its purpose is. Could you please help me with this? Thank you very much!
I don't know the purpose of that step - I only know how to calculate the power using the tool I described in the video.
Hi, thank you for you video. Once we press "calculate power" and receive the results, could you please tell me what is the meaning of the lines "difference" ? Thank you
It's the difference between the two indirect effects (if your hypothesis is, that one indirect effect is stronger than the other).
Hi, thanks for the great video. However, I have 2 predictors,2 mediators, and 3 outcome variables. I plan to run serial mediation analysis using process macro. How do I determine the sample number? Would your method here apply to me?
I don't think so, because that tool is not made for more than one predictor (or for more than one outcome variable).
hi, thanks for the information. However, I have 2 predictors,2 mediators, and 1outcome variables. I plan to run serial mediation analysis using AMOS. How do I determine the sample size? Would your method here apply to me? the total population is 804, how can I take the representative sample size from these population?
I think for that you would have to run a power analysis for a path model since the tool works only with one predictor, I believe.
There is an R package for that, semPower, but I haven't worked with that, yet.
Hi, how can I determine the standard coefficient for a1, b1.... to be 0.3 or 0.5?
Determining the effect sizes is the hardest part in power analysis. Possible sorces for that information:
1 Research for similar hypotheses
2 What is the smallest effect that is practically important?
Eckhart Tolle does statistics
What does it mean that the sigma is not positive definite?
I don't know.