Heavily mathematical treatment of SEMs, with little attempt at catering to beginners or non-mathematicians; all implementation details are purged from the text and segregated to a website (which was less than helpful when I got around to reading it during a period of no Internet). It's hard to see who the formalism would be useful for: who is so intimately familiar with the math & notation & linear algebra that they could follow this text's exposition but don't know anything about SEMs? I didn't get much out of it except a sense that ML estimation really struggles with any remotely complex model that I could fit without a second thought in JAGS.

And despite usually taking a high-level mathematical approach, it devotes a peculiar amount of space to depicting convergence plots and noting how many samples it took the MCMC to converge, which is something that is usually of the least importance especially since none of the models seem to take more than a few thousand iterations to converge, which was a trivial amount of computation in 2007 and is even more trivial now. Some other aspects of the modeling struck me as questionable: repeatedly the models are analyzed by a peculiar procedure in which noninformative priors are used to analyze 10% and then the posterior is reused as the prior for the other 90% of the data - which should be completely pointless a procedure since it should deliver exactly the same final result, but the author apparently thinks it's worth doing.