
02369985
j
2008b.00452
Mohr, Donna
Xu, Yong
Shopping for efficient confidence intervals in structural equation models.
InterStat, No. 7, 15 p. (2006).
2006
,
EN
K85
structural equation models
loglikelihood
latent variables
parameter estimates
confidence intervals
Summary: Most users of Structural Equation Models are aware that Waldtype standard errors for parameter estimates can vary remarkably depending on the arbitrary choice of how the scale is identified. When the focus is on $Ho$: coefficient is 0, tests based on a likelihoodratio are invariant to the scale identification. However, a simple example shows that confidence intervals based on the likelihood ratio exhibit the same problem. A series of examples suggests that shopping for the scale identification that gives the best relative precision for a certain parameter has very little negative impact on coverage probabilities, and can yield substantially tighter confidence intervals.