Attached is a Beginner's Guide to Structural Equation Modeling. Use it as a reference and go through the table of content for section/chapter relevant to your case.

The link is a downloadable book for statistical test. Find the relevant tests that are applicable to your model. Cheers.

http://fcc-statistics.wikispaces.com/file/view/100+Statistical+Tests.pdf

Hi Kegar,

the aim of your analysis looks very interesting to me. At a first reading, you should also control for crossed-effects between markets and months. Other than the useful references provided by Paul, I found very interesting the book by Rabe-Hesketh and Skrondal "Multilevel and Longitudinal Modeling Using Stata", which I consider as a very interesting guide for both specification and estimation of linear mixed models.

Below is the link to the Stata home page where you can find the subject index, and also datasets and program files for the solution of selected exercises:

http://www.stata.com/bookstore/multilevel-longitudinal-modeling-stata/

A more methodological contribution by the same authors is "Generalized Latent  Variable Modelling: Multilevel, Longitudinal and Structural Equation Models", (Chapman and Hall).

I hope it could help.

Best

Carmine

I got the answer: A factor can either we considered as a random effect or fixed effect; not both. The choice depends on the objective: if objective is to estimate variance component associated with the predictor, then random effect; and if the objective is to estimate the differences between factors, then fixed effect. If there are only very limited levels of factor variable (say 2), then fixed effect (variance calculation requires at least three levels).

Based on this, the conceptual model is:

gsem (wo i.s i.mn i.mk i.s#i.mk i.s#i.mn i.mn#i.mk i.mn#i.s#i.mk M2[c] M1[w]-> wl, family(gaussian) link(identity)) ( i.s i.mn i.mk  i.s#i.mk i.s#i.mn i.mn#i.mk i.mn#i.s#i.mk M2[c]  M1[w]-> wo, family(gaussian) link(identity))

Customer (c) and Week (w) are random effects, species (s), market (mk), month (mn) are fixed effects.

wo- weight order, wl- weight loss, mk- market, mn-month, s-fish species, C - customer, w- week, #-interaction, M...[] multilevel variables.

Suggestions/Comments welcome. :)

Yes, the random-effects model can be used to make inferences

regarding the population of clusters, but at the cost of requiring a sufficient number of clusters (at least >2), and making additional assumptions regarding the random-intercept distribution. An advantage of the random-effects model is that it can be used to estimate the effects of cluster-level covariates, in contrast to the FE model. In general, concerning the choice of fixed vs. random effects, in a preliminary stage of analysis you could also perform a Durbin-Wu-Hausman test statistic for endogeneity. If you reject the null hypothesis, then you should abandon the random-intercept model in favor of a fixed-effects model (Rabe-Hesketh and Skrondal, MLM, ch.3).