Dear Joris, thanks again for these very useful insights. Your point about PCA, sem, FA is clear to me now. And I understand what you say about it not being the point of sem to make corrections in a model and then try to "save the scores" for another analysis. But still I am wondering what would happen if you did? suppose I had: a measured variable A and a latent variable B with indicators B1-B10, and A and B are supposed theoretically to correlate strongly. Then suppose I do a FCA and extract one component B'. If I did a similar CFA in an sem package and added a couple of correlated errors which improve the model, and if I could "save the scores" of the latent variable, wouldn't I expect it to correlate better with A than B' does?
Best Wishes Steve www.promente.org | skype stevepowell99 | +387 61 215 997 On Wed, Jun 23, 2010 at 12:36 PM, Joris Meys <jorism...@gmail.com> wrote: > I should have specified: lavaan is not familiar to me. I'm also not > familiar enough with the sem package to tell you how to obtain the > scores, but all information regarding the fit is in the object. With > that, it shouldn't be too difficult to get the scores you want. This > paper might give you some more information, in case you didn't know it > yet : > > http://socserv.mcmaster.ca/jfox/Misc/sem/SEM-paper.pdf > > On a side note, sem with a single latent variable might be seen as a > factor analysis with one component, but definitely not as a PCA. A PCA > is constructed based on the total variance, rendering an orthogonal > space with as many dimensions as there ara variables. Not so for a FA, > as the matrix used to calculate the eigenvectors and eigenvalues is a > reduced matrix, in essence only taking into account part of the > variation in the data for calculation of the loadings. This makes PCA > absolutely defined, but FA only up to a rotation. > > On a second side note, using the saved scores in some subsequent > analysis is in my view completely against the idea behind sem. > Structural equation modelling combines those observed variables > exactly to be able to take the variation on the combined latent > variable into account. If you use those latent variables as input in a > second analysis, you lose the information regarding the variation. > > Cheers > Joris > > > > On Wed, Jun 23, 2010 at 9:53 AM, Steve Powell <st...@promente.net> wrote: >> Dear Joris, >> thanks for your reply - it is the sem case which interests me. Suppose >> for example I use sem to construct a CFA for a set of variables, with >> a single latent variable, then this could be equivalent to a PCA with >> a single component, couldn't it? From the PCA I could "save" the >> scores as new variables; is there an equivalent with sem? This would >> be particularly useful if e.g. in sem I let some of the errors covary >> and then wanted to use the "saved scores" in some subsequent analysis. >> >> By the way, lavaan is at cran.r-project.org/web/packages/lavaan/index.html >> >> Best Wishes >> Steve >> >> www.promente.org | skype stevepowell99 | +387 61 215 997 >> >> >> >> >> On Tue, Jun 22, 2010 at 7:08 PM, Joris Meys <jorism...@gmail.com> wrote: >>> PCA and factor analysis is implemented in the core R distribution, no >>> extra packages needed. When using princomp, they're in the object. >>> >>> pr.c <- princomp(USArrests,scale=T) >>> pr.c$scores # gives you the scores >>> >>> see ?princomp >>> >>> When using factanal, you can ask for regression scores or Bartlett >>> scorse. See ?factanal. >>> Mind you, you will get different -i.e. more correct- results than >>> those obtained by SPSS. >>> >>> I don't understand what you mean with scores in the context of >>> structural equation modelling. Lavaan is unknown to me. >>> >>> Cheers >>> Joris >>> >>> On Tue, Jun 22, 2010 at 3:11 PM, Steve Powell <st...@promente.net> wrote: >>>> Dear expeRts, >>>> sorry for such a newbie question - >>>> in PCA/factor analysis e.g. in SPSS it is possible to save scores from the >>>> factors. Is it analogously possible to "save" the implied scores from the >>>> latent variables in a measurement model or structural model e.g. using the >>>> sem or lavaan packages, to use in further analyses? >>>> Best wishes >>>> Steve Powell >>>> >>>> www.promente.org | skype stevepowell99 | +387 61 215 997 >>>> >>>> ______________________________________________ >>>> R-help@r-project.org mailing list >>>> https://stat.ethz.ch/mailman/listinfo/r-help >>>> PLEASE do read the posting guide >>>> http://www.R-project.org/posting-guide.html >>>> and provide commented, minimal, self-contained, reproducible code. >>>> >>> >>> >>> >>> -- >>> Joris Meys >>> Statistical consultant >>> >>> Ghent University >>> Faculty of Bioscience Engineering >>> Department of Applied mathematics, biometrics and process control >>> >>> tel : +32 9 264 59 87 >>> joris.m...@ugent.be >>> ------------------------------- >>> Disclaimer : http://helpdesk.ugent.be/e-maildisclaimer.php >>> >> > > > > -- > Joris Meys > Statistical consultant > > Ghent University > Faculty of Bioscience Engineering > Department of Applied mathematics, biometrics and process control > > tel : +32 9 264 59 87 > joris.m...@ugent.be > ------------------------------- > Disclaimer : http://helpdesk.ugent.be/e-maildisclaimer.php > ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.