How to model observed variables in plssem

[Il Schubert]

HI all,

I am trying to model a path diagram (with formative and reflective constructs) using plssem. I don't seem to be able to work out how to include observed variables in the structural part of the PLS. If I include them (in capital letters or in small - ZSNINDEXNP ztalkto - in the example below) they simply do not appear in my output (see the code and the attached output picture). I have searched, read the paper by the authors of the plssem (i.e., Venturini & Mehmetoglu), but I can not find an answer. Can anyone help? (I have to also say that I am new to stata so might have made an obvious mistake).

#d ;
plssem (BENEFIT < zbp1 zbp2 zbp3 zbp4)
(ACCEPT < za1 za3 za4 za5)
(RISK > zrp1_r zrp2_r zrp3 zrp4),
structural (ACCEPT RISK BENEFIT, RISK BENEFIT ZSNINDEXNP ztalkto, BENEFIT ZSNINDEXNP ztalkt )
stats correlate (lv)
;
#d cr

Thanks a million for your help!

[Il Schubert]

HI All,

me again, I am curious if nobody is using the plssem command to model PLS and that is why I havent had any answers? Alternatively I might have not made my post title clear enough? Any help would be very much appreciated.

Best Il

[Il Schubert]

Hi Statalist users,

For all that are interested in this issue of modeling observed variables in plssem. I emailed my question to Sergio Venturini, the co-author of the plssem programme, and here is what he replied:

Dear Il

thanks for your interest in our package. as you correctly pointed out, currently plssem is not able to account for observed variables in the structural equations, that’s why you don’t see them when you estimate your model.

however, there is a trick that you can use to bypass this issue: you can define a new (fake) latent variable which loads only on a single item, that is the observed variable you want to include in the structural model and use them in the structural part of the model instead of the observed variables. using your model this corresponds to:

plssem (BENEFIT < zbp1 zbp2 zbp3 zbp4) (ACCEPT < za1 za3 za4 za5) (RISK > zrp1_r zrp2_r zrp3 zrp4) ///
(ZSNINDEXNP_lv <ZSNINDEXNP) (ZTALKTO_lv <ztalkto), ///
structural(ACCEPT RISK BENEFIT, RISK BENEFIT ZSNINDEXNP_lv ZTALKTO_lv, ///
BENEFIT ZSNINDEXNP_lv ZTALKTO_lv) ///
stats correlate(lv)

clearly, all latent variables are standardized, so the corresponding coefficients should be interpreted as standardized coefficients.

i hope this suggestion helps.

keep in touch because we are continuously updating and revising the package to improve it and let it become a reference for PLS-SEM.

best,
sergio

***Thank you again Sergio for your quick and extremely helpful respons. The model now works.