Testing mediation using 'medsem' package in Stata

This video provides a demonstration of the 'medsem' package which can be used when testing for mediation in Stata.
A copy of the data can be downloaded here: drive.google.com/file/d/1OlDL...
A copy of the Do-file referenced in the video can be downloaded here:
drive.google.com/file/d/1EutD...
A copy of the Powerpoint referenced in the video can be downloaded here: drive.google.com/file/d/1I3-5...
For additional videos and resources on multivariate statistics, check out: sites.google.com/view/statist...

[@minahpark2153]

Super helpful!! Among many stata related videos, I found this one the most well-created and explained. Thank you so much!

[@Zane_Zaminsky]

Thank you, Dr. Mike! Love your Stata videos!!!

[@mdirtejaislam5930]

Hi Dr Crowson, It really helped me a lot. Much appreciated :)

[@benardmusekesewabukala2300]

Super illustration. Thank you very much for this clip.

[@HuzMusFinanceData]

Thank you so much for contributing Useful Video and Notes.... Highly appreciated...

[@IrfanKhan-yg8zu]

Thank you so much for this valuable analysis. Please, upload one video for DID? Thanks

[@happytree23256]

Hello Dr.Crowson! thank you for this amazing video! I just have one question. For example 1, which only has 3 variables, should I report the goodness of fit(chi-square)? I did it and the dof is 0 and p>chi2 was just a dot (.). I assumed dof was because I had 3 pathways having literally all possible paths :) Thank you so much in advance!

[@anilkshatriya8209]

Fantastic, Mike. Thanks. Can you please share a video/link where you demonstrate how to use graphics to plot the results.

[@sidseldomazet4898]

Hi Mark. Great video! How do I obtain confidence intervals (CIs) for the direct and total effects after medsem? Medsem only provides CIs for the indirect effect. However, I cannot use the estimates and CIs from the sem model, because the medsem posthic estimation provides estimates and CIs from Monte Carlo simulations.

[@FelipeDias-wo5bw]

Hi, Dr. Crowson. great video! I have a question about your medium command. If you have a moderated mediation situation, in which, say the independent variable is mastery, the moderating variable is anxiety, the mediator is interest, and the dependent variable is achieve, then would you say that the "indep" category in medsem command is the interaction term of mastery*anxiety. So, medsem, indep(mastery*anxiety) med(interest) dep(achieve) mcreps(500) rit rid Is this correct?

[@margheritabelgioioso2283]

Thank you for this! I have a question on interpretation I hope you can help me with: How do I interpret a complete negative mediation where path a is negative and path b positive? STEP 1 - (X -> M) with B=-0.608 and p=0.000 STEP 2 - (M -> Y) with B=6.826 and p=0.004 STEP 3 - (X -> Y) with B=4.715 and p=0.443 As STEP 1, STEP 2 and the Sobel's test above are significant and STEP 3 is not significant the mediation is complete!

[@msmreview2221]

Thanks, Dr. Crowson. This is EXTREMELY helpful. I would say your video is better than a 10-week lecture in school. A follow-up question, sometimes I see researchers will control for covariates in the mediation model. For example, when examining the mediating role of interest between mastery and achieve, I want to control age, education, and income level (variable names: age, edu, income) . How should I do this? Thanks in advance!

[@mikecrowson2462]

Hi there. Thanks for your question. I noticed that I did not include any covariates in my example, but I would expect it to work the same by adding in the covariates when using the sem command. I just gave it a run through and it worked. Try: sem (achieve

[@msmreview2221]

@@mikecrowson2462 Thank you so much, Dr. Crowson!

[@a.k.yarkin3666]

Hello Dr Crowson, Thank you very very much for your unmatchable videos (with powerpoint and data) especially the ones on Hayes' PROCESS for SPSS. I was wondering if it is in your schedule doing videos on Montoya's MEMORE Macro for SPSS. I have been struggling using SEM for analyzing mediated models with repeated measures. Thaaanks again!!

[@mikecrowson2462]

Hi there! Thanks for following my work! I've run across this macro. But I'll need to do some learning on my own before attempting a video on medmore. But it is on my radar to eventually do a video on it. Maybe over summer when I can have more time for studying up on it. Best wishes!

[@tutunmukherjee2]

Thank you for your nice presentation. Sir, could you please guide us on the incorporation of control variables in medsem package?

[@ElPerreke007]

Did you solve it? I would like to know how to deal with it too :( please

[@anjo6248]

If your model contains control variables, i.e., covariates, you must include these in each of the sem equations. Thus, your sem model will look something like this: sem (MV

[@user-hg7ns4cm7u]

Hi Mike Thank you for the helpful videos Do you know if there is an option to calculate mediation with a logit model (binary DV) but is also multilevel ?

[@mohameddjellouli2223]

Hi dear Dr, thank you for sharing this video, I have a question about sample, so can I do path analysis on stata if I have 3 variables from 2000 to 2018???

[@ahmedadel822]

Hi Dr Crowson, I am just wondering could I use this command with continuous variables not from survey data? Thanks.

[@sysysysysysysysys482]

Thanks for your kind explanation! Your channel helped me a lot in doing my own research. I'm wondering how can I put two different mediators in the model, which means how can I test two different mediating pathways in one model. Again, thank you so much!

[@joboldenziel3632]

did you ever find the answer to this?

[@alinik4737]

Can we test the mediation's significance using bootstraping with this command? Or is the monte carlo simulation part is same as bootstraping?

[@shaunasimon4452]

Hi Dr.Crowson, I found a paper that used the medsem package with a categorical IV. Do you know if this is correct? Thanks!

[@will74lsn]

great video! Does something similar exist for gsem (i.e multilevel mediation)?

[@nima3915]

In the last example, there are 2 parallel mediators (interest and anxiety) however, medsem only includes choosing "interest" as the mediator. How can we put 2 mediators there?

[@tranthu3450]

Hello Dr Crowson. Thank you for this video! I have a question about data. Can I use Medsem to process panel data?

[@jingnanli7325]

have u found the answer. he did not show us the data. i also have panel data.

[@davida.jimenezj.7208]

I have the same question.

[@jingnanli7325]

thank u, how about using panel data? the same procedure? where do I include control variables?

[@jacobeberharter6342]

Hallo, thanks for this very helpful video. However, I have a question: Is it possible to use this stata package with panel data? If yes do the codes change?

[@diary_m281]

thank you!!!!!!!!!! :) I hava a question Is it not available in logistic regression analysis? If possible, can you tell me how to do it?

[@mikecrowson2462]

Hi there. Thanks for your question. There is no way to test a mediation analysis where you have a binary DV or mediator using the medsem package. I've written a do-file that would allow you to test for mediation where the mediator is continuous but the dependent variable (Y) is binary. If you go to the video here (kzbin.info/www/bejne/moTLo6CCob9gjsU), you can download the do-file and the video walks you through its use. Apart from that, I don't know of any other packages or do file (not saying they aren't out there, just that I don't know of any). Your only other option is to write your own code to accomplish this. That's kind of what I ended up doing to create my do-file. There is actually a more efficient way to do it than how I did (I was trying to mimic Hayes' Process macro through SPSS) and that is to run a generalized SEM and then use the stored regression coefficients to compute the indirect effect. You'd also bootstrap the indirect effect to generate the confidence interval. There are a couple of nice tutorials here (stats.idre.ucla.edu/stata/faq/how-can-i-do-mediation-analysis-with-the-sem-command/) and here (stats.idre.ucla.edu/stata/faq/how-can-i-do-moderated-mediation-in-stata/) that I pivoted off of when writing my code. You could probably just extend this from sem (which is what they demonstrate) to generalized sem. This is probably way more than you wanted to know. Good luck

[@rubenvanpampus7400]

Great video! Is it possible to do the medsem analysis for subgroups?

[@mikecrowson2462]

Hi there. It's been awhile since I have worked with this package, but I don't recall there being anything related to subgroups analysis. Here's some documentation on the package I found on the web: ntnuopen.ntnu.no/ntnu-xmlui/bitstream/handle/11250/2589997/offprint-medsem+a+Stata+package+for+statistical+mediation.pdf?sequence=1 If your aim is to test a model in say Group A and then in Group B, you could subset out your data for each group and then run your model on each using the medsem package. If your question is about invariance between groups, then that's another matter entirely as that is clearly not what medsem is designed to get at. I hope this helps!

[@heathersky8218]

Thank you for the video, may i ask whether medsem can be used with binary dependent variable?

[@mikecrowson2462]

Hi there. Thanks for your question. It really isn't for use with a binary DV. Best wishes!

[@heathersky8218]

@@mikecrowson2462 thank you so much for your response

[@anwarhariyono3659]

Thanks. Is it possible for secondary data, in the field of accounting where the data is from financial reports. Can I be helped?

[@safiasuleman4591]

Can you please tell me that if we have panel data then for mediation this method can be used ?

[@jingnanli7325]

have u found the answer. he did not show us the data. i also have panel data.

[@safiasuleman4591]

@@jingnanli7325 yes we can but i will suggest use stata software for panel data

[@IrfanKhan-yg8zu]

Dear sir, How to get access to the contents on your website? because it doesn't show the materials. thanks

[@mikecrowson2462]

Hello. The links are provided under the video description. Best wishes

[@piyumiseneviratne2465]

I recently used this medsem package and when I ran the first command I got an error message as unrecognized command: sem . Can you please help me to solve this problem

[@jagopenelitian]

How about R² or R - Square sir? what is command to show R² ? Thank you sir

[@gerieallccaco256]

direct and indirect effects and the total effects ? how?

[@youlicho9275]

Thank you for this great video. My understanding is that SEM typically requires a larger sample size, ideally over 200 observations, but the 'medsem' package could handle smaller sample sizes? Could anyone direct me to academic sources that support this?"

[@youlicho9275]

Another quesiton is... what is "delta" column for? and why is it almost always the same as sobel test?