lecture4_s10 observed

# lecture4_s10 observed - SEM with observed variables...

This preview shows pages 1–8. Sign up to view the full content.

SEM with observed variables: parameterization and identification Psychology 588: Covariance structure and factor models Feb 3, 2010

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Limitations of SEM as a causal modeling 2 • If an SEM model reflects the reality, the data will be consistent with the model, given that measurement errors are tolerable, all assumptions made are tenable, etc.; but the reverse is not true (e.g., see Fig. 3.9, p. 70) • Theory- vs. data-driven • Cause vs. effect indicators ¾ simultaneous reciprocal relation (e.g., financial health and stock price of companies) --- really concurrent? • To be cautious about goodness (mostly badness) of fit testing
Fundamental equation with observed variables 3 () ( ) 1 =+ =− + + y IB ζ By Γ Γ x x ζ • Dependent variables y are modeled as linear combinations of (a subset of) y and x as Hypothesized explanation for the covariances of y unexplained (but allowed to exist) in the y variation • From the measurement perspective, y and x may be considered as “single-indicator latent variables” with no measurement errors Demographic variables are often considered so (e.g., sex, age)

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Fundamental equation with observed variables 4 • Alternatively, y is modeled as linear combinations of a partitioned vector of y , x and ζ as [] ⎡⎤ ⎢⎥ == =+ + ⎣ ⎦ y yB y B Γ Ix B y Γ x ζ ζ ± ± • This is essentially the approach taken by RAM (reticular action model), though RAM incorporates all latent variables, ξ and η --- to be discussed later
Two conditions for recursive models 5 B should be a lower triangular matrix, and so no feedback loop of causal paths • All error terms ζ are not correlated with one another Correlated errors themselves don’t result in a feedback loop; instead, they lead to inconsistent estimates due to errors correlated with explanatory variables ¾ Note: any exogenous variables (including error terms) are allowed to be inter-correlated while no endogenous variables are allowed so (instead, correlated through exogenous vars) x1 y1 zeta1 y2 zeta2 1 1

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Implied covariance matrix 6 • Basic hypothesis with ideal measurement: Σ = Σ ( θ ) --- population covariance matrix Σ is a function of free model parameters θ • Basic hypothesis in reality: --- sample covariance matrix S is a function of estimates of model parameters • Discrepancy in the data ( Σ vs. S ) is due to sampling errors while the discrepancy in the parameters ( θ vs. ) is due to not only sampling errors but also any violated assumptions made for a particular way of finding the estimates (e.g., ML) () ˆ = S Σθ ˆ θ ˆ θ
Implied covariance matrix 7 () [] ( ) ()() ( ) 1 11 1 1 E EE −− ⎛⎡ ⎤ ′′ == + ⎜⎟ ⎢⎥ ⎝⎣ ⎦ ⎡⎤ = ⎣⎦ ⎡′ −+ = y Σθ yx y IB Γ x ζ x yy xy xx ΓΦΓ Ψ ΓΦ ΦΓ Φ

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

## This note was uploaded on 06/11/2010 for the course PSYC 588 taught by Professor Sunjinghong during the Spring '10 term at University of Illinois at Urbana–Champaign.

### Page1 / 24

lecture4_s10 observed - SEM with observed variables...

This preview shows document pages 1 - 8. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online