Why use a structural equation model?

Dan talks about three principal advantages of structural equation models (SEMs) relative to more traditional analytic techniques, like the linear regression model. These include...
(1) The ability to represent constructs as latent variables that are uncontaminated by measurement error
(2) Falsification tests and indices of fit with to evaluate the tenability of a proposed theoretical model
(3) Flexibility to allow researchers to specify statistical models that more closely match theory
Dan describes these advantages using a specific example on the factors that relate to young children's popularity with peers.
In addition to these three principal advantages of the SEM, there are many other ways that the model can be expanded and used to address interesting theoretical questions. For instance, a variety of SEMs exist for analyzing longitudinal data, including latent growth curve models and latent change score models. SEMs also provide a powerful framework within which to evaluate population heterogeneity, including differences over known groups (e.g., boys and girls) or latent groups (e.g., clusters of individuals for whom predictive relationships differ). For those interested in learning more, we offer summer training seminars on SEM and longitudinal SEM, see www.curranbauer.org/training/.

[@EBear519]

The best explanation I have found so far on SEM. Thank you!

[@johngalt5258]

This is incredibly helpful -- thank you for making these available.

[@justbecause9747]

Thank you for the wonderfully clear presentation.

[@segsb7085]

Wow! Just wow! This is the best explanation I've ever come across! Thanks so much!

[@SableLeaf]

Excellent explanation: very well summarised and - immediately - I subscribed.

[@teodorkadiyski6109]

went over my lecturer's explanation like 6 times before giving up and searching on youtube, very impressed at how quickly this video made it sink in. Great presenter.

[@marygracenueva6181]

Very clear explanations..thanks!

[@majrom]

Thank you! This is really helpful!

[@victorkankhokwe884]

Skilled presenter and facilitator , he knows the stuff!

[@fouaddjebbar4472]

Impressive, very simplistic and intuitive thank you so much

[@samueltirolegrahamschiller7965]

very, very helpful, thank you!

[@nachiketjangam4343]

Thank you , simple n precise..

[@denissekiwu5458]

The best explanation ever. Thank you professor

[@Suresh8848m]

Wonderful and Interesting presentation

[@Gergentine]

Very good explanation. Much easier to understand than other videos out there.

[@vrindavnair5125]

Made simple and really helpful

[@farrukhmahfooz1305]

Thank you for a clear presentation which was easy to understand

[@jayanthsaishiva]

Great video thanks

[@samiraabdeyazdan3629]

Thanks for you clear explanation

[@user-fm6ny4tp1z]

Thank you.

[@meddahabdelhadi3012]

Thank you

[@user-cq5qj7nz3v]

Thanks alot

[@zelalemmarkos8996]

Sir Thanks

[@BuffaloL100]

thanks

[@lewtenant_k]

Great presentation. Very concise and informative. What are the main limitations? And how does this technique handle reciprocal causality or feedback structures? e.g. What if popularity feeds back and increases pro-social behaviour, which then further increases popularity?
Thanks

[@eriangelino7800]

SEM is confirmatory factor analysis. In other words, before you commence your study, you ought ot jave certain theoretical framework you want to confirm. Therefore, the idea of reverse causality must also be included in the theory you have in mind.

[@ShafakTan]

Really sank in! Thanks for the regression comparison, that made it stick.

[@mateom6596]

well done, sir. Very clear and concise. I appreciate especially how you showed the three advantages relative to regression models and used a simple, but comprehensive example. A question arises for me though, can a latent variable be based on a total score on several measures? For example a latent variable of depression based on BDI-II and PHQ-9 total scores rather than the items on either or both of those measures - or would this reintroduce the problem of error?

[@centerstat]

Hi Mateo -- thanks for your nice words. Absolutely -- it's actually rather common to use a scale score (that might be the mean of several items) as one of several indicators on a latent factor. This is also sometimes called "item parceling" where you combine small numbers of items in parcels, and then fit the latent factor to the parcels. This is not without controversy, though -- some people like it and others hate it -- e.g.,
Marsh, H. W., Lüdtke, O., Nagengast, B., Morin, A. J., & Von Davier, M. (2013). Why item parcels are (almost) never appropriate: Two wrongs do not make a right-Camouflaging misspecification with item parcels in CFA models. Psychological methods, 18(3), 257-284.

[@mugomuiruri2313]

good

[@prateek89]

In this example, 5 items are used to represent the emotional regulation, is it necessary that all 5 items need to be answered by all the participants?

[@prateek89]

Thanks for clarification and providing the path forward. Also, the series on SEM is incredibly useful, keep up the great work.

[@mariosgeorgiou4990]

I can’t see the answer he gave to your question, has it gone?

[@anyasmiller806]

I recently run analysis of my data. Hypotheses look good, r square looks good; but model fits such as AGI, AGFI, RMSEA, etc are not good. Please what would you suggest for me to do in order to get my paper published

[@centerstat]

Thanks for your note. The r-squared value is not a measure of fit. It is simply a reflection of explained variance in the dependent variable. The other measures evaluate overall goodness-of-fit, and poor indices reflect that your hypothesized model is not adequately reproducing the characteristics of your sample data. This in turn increases the probability that you have bias in your parameter estimates. You must carefully consider modifying your model in some way, but be sure to stay close to theory and don't only make decisions based on your data. Good luck with your work.

[@ahadaway01]

Thank you.