Longitudinal Modelling

[Hernando]

Hello Dear Smart PLS community,

I am trying to make a longitudinal analysis from data comprising two years (both with the same questions for the measurement model). The situation is this:

When I model separately, the output provided by Smart PLS present reliable and valid results for both years. However this is not the case when I model latent variables corresponding both years into one single model. The results obtained seem doubtful. I have seen your previous postings about multi-group comparison ,but I am not sure if this is the case here. Therefore I want to know:

1)What is to be considered when modelling longitudinal data with Smart PLS?

2) Is there any literature concerning longitudinal data related to PLS? ( Most of the articles I found about longitudinal data regarding Structural Equation Modelling refer to Covariance Structure Models).

Thanks a lot.

[koppara]

nobody seems to answer. i´m also interested in longitudinal modeling, especially with regard to moderator analysis with pls. could somebody help with sources of literature or articles, please?

[schroer]

Dear all,

this is an interesting issue. I believe there are some thoughts on this in Wold (1982) [1], but I don't know of any other references.

Some time ago, there was a small discussion of this on SEMNET (archives available at http://www.bama.ua.edu/cgi-bin/wa?A0=SEMNET&X=-, July 3). I'll just quote Cameron McIntosh's posting below:

Hi Andrea,
Interesting issues. In theory I am fairly sure this could be done - but I am also completely sure that it has not been implemented. Perhaps a hopeful avenue is GSCA (Generalized Structured Components Analysis) and the program VisualGSCA, developed by Heungsun Hwang. It would be excellent if complex survey features made their way into it someday. However, VisualGSCA does do bootstrapping (default = 100 replications) automatically in order to get standard errors. The bootstrapping is not design-based, but I imagine it would break down the intracluster correlation and your standard errors would be more appropriate. I guess the main issue is whether the substantive theory/research questions dictate a model- versus design-based approach for dealing with the complex sampling plan.
HTH,
Cam

Cameron N. McIntosh, MA
Analyst / Analyste
Health Analysis and Measurement Group / Groupe d'analyse et de mesure de la santé
Statistics Canada / Statistique Canada
24-I R.H. Coats Building
>100 Tunney's Pasture Driveway
Ottawa, ON
K1A 0T6
Phone: (613) 951-3725
Fax: (613) 951-3959

BTW, the reference she mentions is [2]. If anyone knows another reference, I'd be most interested. :-)

HTH,

Joachim


[1] Wold, H. (1982). Soft modeling: The basic design and some extensions. In K. G. Jöreskog & H. Wold (Eds.), Systems under indirect observation: Causality, structure, prediction. Part II (pp 1-54). Amsterdam: North-Holland.

[2] Hwang, H. & Takane, Y. (2004). Generalized structured component analysis. Psychometrika, 69, 81-99.
http://www.psych.mcgill.ca/perpg/fac/hw ... lGSCA.html

[viswadatta]

If the theory proposed is true and valid, it has to satisfy criterion validity. Criterion validity has the following types
1. Concurrent validity
2. Predictive validity

If the model is truly predictive, it should work across a time frame. Only then can it be generalised.

[Hernando]

Well I finally managed to make the models... I find out what the problem was...however the litterature related to PLS is really scarce...I found this article where PLS has been applied if someone is interested:

Shea and Howell (2000), “Efficacy Performance Spirals: An Empirical Test”, Journal of Management, 26.