Hello,
I am new to the Stata forum (let me know if I have not correctly followed the forum's rules). I am conducting mediation analysis using panel data and would like to ask what command would be appropriate to run the analysis - taking into account the structure of my data - and what statistical method would be the best approach. I'd appreciate if you can give me recommendations on them.
To give you a brief background, 1) there are two separate mediators (which I would like to run them separately). One of them is continuous and another one is binary. Ex) the mediator is child's nutrition and I would like to test this mediator by using two types of variables -- food insecurity (binary) variable and nutritional intake variable (continuous). The dependent variable is continuous and the independent variable is binary. 2) The data is longitudinal in nature (two waves) and the unit of analysis is at the individual level. I would like to add fixed effects in my model. The data is survey data, and thus I need to use the "svy" command to take this into account.
I read several Mackinnon's papers - and found that "products-of-coefficients approach" using bias-corrected bootstrapping methods has advantages over other methods. After reviewing other posts, I figured SEM command can conduct mediation analysis using bias-corrected bootstrapping (https://stats.idre.ucla.edu/stata/fa...e-sem-command/). However, it seems like SEM cannot be used in longitudinal data. Then, I saw many posts talking about gsem / xtdpdml commands being used in panel data. However, I couldn't figure out whether either of the commands is able to take into account fixed effects, not random effects. Stata manual explains how gsem can be used in multi-level modelling (https://www.stata.com/manuals13/sem.pdf) but does not specifically explain whether it also allows to add fixed effects.
Under this consequence, I want to ask if anyone knows what command would allow me to conduct bias-corrected bootstrapping for mediation analysis using survey data ("svy"), fixed effects, and binary/continuous mediators. Also, I would appreciate if any of you can give me advice on methods for mediation analysis, if you think other methods (e.g. sobel test) would be better in my case than the bias-corrected bootstrapping.
Thank you,
Stephanie
I am new to the Stata forum (let me know if I have not correctly followed the forum's rules). I am conducting mediation analysis using panel data and would like to ask what command would be appropriate to run the analysis - taking into account the structure of my data - and what statistical method would be the best approach. I'd appreciate if you can give me recommendations on them.
To give you a brief background, 1) there are two separate mediators (which I would like to run them separately). One of them is continuous and another one is binary. Ex) the mediator is child's nutrition and I would like to test this mediator by using two types of variables -- food insecurity (binary) variable and nutritional intake variable (continuous). The dependent variable is continuous and the independent variable is binary. 2) The data is longitudinal in nature (two waves) and the unit of analysis is at the individual level. I would like to add fixed effects in my model. The data is survey data, and thus I need to use the "svy" command to take this into account.
I read several Mackinnon's papers - and found that "products-of-coefficients approach" using bias-corrected bootstrapping methods has advantages over other methods. After reviewing other posts, I figured SEM command can conduct mediation analysis using bias-corrected bootstrapping (https://stats.idre.ucla.edu/stata/fa...e-sem-command/). However, it seems like SEM cannot be used in longitudinal data. Then, I saw many posts talking about gsem / xtdpdml commands being used in panel data. However, I couldn't figure out whether either of the commands is able to take into account fixed effects, not random effects. Stata manual explains how gsem can be used in multi-level modelling (https://www.stata.com/manuals13/sem.pdf) but does not specifically explain whether it also allows to add fixed effects.
Under this consequence, I want to ask if anyone knows what command would allow me to conduct bias-corrected bootstrapping for mediation analysis using survey data ("svy"), fixed effects, and binary/continuous mediators. Also, I would appreciate if any of you can give me advice on methods for mediation analysis, if you think other methods (e.g. sobel test) would be better in my case than the bias-corrected bootstrapping.
Thank you,
Stephanie
Comment