PSYO 373: Structural Equation Modelling

What are the advantages in using SEM to assess measurement stability?

Stability estimates that come from the simple correlation between multiple assessments of a measured are biased - Extent or direction of bias is hard to discover due to joined attenuating effects of measurement error within, and because of inflation that is due to covariation between errors of measurement between occasions. Evaluation of stability is enhance by removing these sources of bias (unreliability and correlated errors of measurement) before estimating test-retest correlationn or consolidation by permitting detailed comparisons of measurement models for single construct at multiple points in time

Why is the term "causal modelling" potentially misleading?

• Causation can only be established in the experimental design • If performing SEM on a cross-sectional analysis, then you cannot determine causation

Describe correlated errors of measurement and specific effects, and how they are typically discovered and handled by researchers.

• Correlated errors of measurement can improve overall fit of the model o Can occur if there is something about the error of those two measurements which is not related to the other measurements (ex. those two measures were done by the same researcher whereas the others were not) o They are usually added after the fact and are therefore not a meaningful and not likely to replicate • Specific effects are a relation between a portion of unique variance in one measure to an observed or latent variable o Often added post-hoc and therefore subject to type 1 error • When they are discovered, researchers should attempt to replicate these findings

In a full SEM, relationships between variables can be examined "at the latent level". Explain.

• If you are only using a few items of a measure to define the latent variables you are not really on the "latent level" • To be on the "latent level" you must include everything that might influence that latent variables • Ex. including a few questions from the SWB measure does not put you at the latent level for happiness, you would need positive affect, negative affect, cognitive well-being, subjective well-being, self-report and observed indicator.

How is "isolation" at least partially achieved in SEM?

• Isolation is achieved through statistical means by including in the design variables that may produce an effect or jointly influence the putative cause and effect • Provides maximum flexibility in modelling effect of extraneous variables

In the most general sense, what does SEM entail?

• It is a comprehensive, flexible approach to modeling relations among variables • Model covariances, which entails proposing a set of relations and evaluating their consistency with the relations manifest in an observed covariance matrix

What are measurement models? Latent variables?

• Measurement models: relations between observed and latent variables o Hypotheses about the relations between a set of observed variables, and the unobserved variables o Variants of the factor analysis model which often are used to justify or create variables, are measurement models • Latent variables are things that cannot be measured or observed o We can only estimate them with other observed variables

What are the differences between SEM and simple path analysis?

• SEM can include multiple equations which are estimated simultaneously

What is the structural model?

• Structural model corresponds to aspects of a structural equation model that concern the relations among independent and dependent variables • No restrictions of relations between variables • Includes multiple equations that are estimated simultaneously