Consequences of measurement error. Psychology 588: Covariance structure and factor models

Size: px

Start display at page:

Download "Consequences of measurement error. Psychology 588: Covariance structure and factor models"

Jewel Stevens
5 years ago
Views:

1 Consequences of measurement error Psychology 588: Covariance structure and factor models

2 Scaling indeterminacy of latent variables Scale of a latent variable is arbitrary and determined by a convention for convenience Typically set to variance of one (factor analysis convention) or to be identical to an arbitrarily chosen indicator s scale By centering indicator variables, we set latent variables means to zero Consider the following transformation: x j J a b b j j j j,,...,,, 0 a j j j j b b

3 If all J indicators are considered simultaneously, vector notation is more convenient: x νλ δ, a b a ν λ λ δ b b meaning that the linear transformation of ξ can be exactly compensated in the accordingly transformed ν = ν λa/b and λ = λ/b, leaving the errors δ unchanged (i.e., same fit)

4 What s great about measurement errors in equation 4 Regression weights and correlations are interpreted, implicitly assuming that the operationally defined variables involve no measurement error --- hardly realized for theoretical constructs (e.g., self esteem, IQ, etc.) Ignoring the measurement error will lead to inconsistent estimates We will see consequences of ignoring measurement errors

5 Univariate consequences 5 Consider a mean-included equation for X (hours worked per week) to indicate ξ (achievement motivation): X, E, E 0, E 0 E X X var X var Given only one indicator per latent variable, the intercept and loading (i.e., weight) are simply scaling constants for ξ However, if the ξ scale is set comparable to the X scale (i.e., λ = ), we see that var(x) is an over-estimation of ϕ = var(ξ) if δ is not included in the equation

6 Bivariate relation and simple regression 6 True data structure: x y η: job satisfaction y: satisfaction scale xi x d gamma eta y e zeta cov(x, y) is unbiased estimate of cov(ξ, η) with λ = λ =, since no other variables (δ and ε) can explain cov(x, y) cov, cov, cov xy, cov,

7 From the previous equations, and by analogy with y = γ x + ζ if measurement errors are ignored, cov, cov xy, var x var The parenthesized ratio (reliability) becomes only with no measurement error; otherwise, γ is an attenuated estimate of γ and s s is an inconsistent estimator of γ, ˆ xy xx If the bias of regression weight has an additional factor as --- but such scaling is unusual xx when there is only one indicator per latent variable xx

8 Correlations: cov, var var xy cov xy, var x var y var x var y var var x var y var xx yy which shows an attenuation of the true correlation due to measurement error, with the familiar correction formula: xy xx yy

9 Consequences in multiple regression 9 True data structure: γξ x ξδ y with Λ x = I and λ y = d d d3 x x x3 xi xi xi3 g g g3 eta zeta y e Ignoring measurement errors: y γ x σ cov ξ, cov ξ, ξγ Φγ σ cov x, y cov ξδ, ξγ Φγ xy

10 γ Φ σ and by analogy with γ x, γ Σxxσ xy ΣxxΦγ Φ Θ Φγ y Without measurement error (Θ δ = 0), γ γ ; otherwise, γ γ Alternatively written: since --- where i Σ Σ xx x γ is the OLS estimator of B in i.e., regression weights for prediction of ξ by x Σ Σ γ xx x Σ Φ x ξ Bx e, Again, without measurement error, Σ Σ xx x I Note: in Bollen (pp ), are meant to be respectively, for the multiple regression model ΓΣ,, Σ γσ,,, xy σxy

11 As a very simplified case, suppose x is the only fallible as: x x, i,..., q i i with the true and estimated regression equations: x q q q In this special case, the regression weight matrix has a simple multiplicative form of bias (hint: use ): Σ Σ Φ Θ Φ I c 0 xx x q q c q Φ Φ Θ c 0 I,

12 Consequently, resulting bias factors are: i b x q b,,, i q i x i Bias-factor for x is less than in absolute value ( without measurement error), and so is biased toward the bias factor indicates regression weight b in ξ = b 0 + b x + b ξ + + b q ξ q ~ i b x q Consequences for x i, i =,,q are additive, depending on relationships between ξ and ξ i holding all other IVs constant, and γ

13 So far all reasoning is based on rather unrealistic assumptions: Only single indicator per latent variable, and so its loading becomes simply scaling constant Only one fallible IV Without such assumptions (e.g., all IVs fallible), consequences of measurement error become too complicated and hard to simplify algebraically --- no particular simple form of aσ Σ One clear conclusion: all estimates are inconsistent --- systematically different from what they meant to be xx x

14 Consequence in standardization: standardized ii var i i i var var Consequence in SMC is similar to the bivariate case: plim R plimr What should we do with essentially omnipresent measurement error? Use SEM which allows for measurement errors in the model --- though we are limited in certain models regarding the model identification (e.g., Table 5., p. 64)

15 Correlated errors of measurement 5 Consequence in regression weights further complicated: γ Σ Σ γ Σ σ xx x xx For simple regression: Now, is not necessarily < γ cov, xx var x If correlated measurement errors are only within IVs (i.e., σ δε = 0, Σ xx = Φ + Θ σ where Θ σ is not diagonal), γ Σ Σ γ still holds (but the bias factor will have a more complicated form, also involving off-diagonal entries of Θ σ ) xx x

16 With multi-equations 6 In path models with sequential causal paths, consequences of measurement errors very hard to simply generalize --- see the union sentiment (Fig. 5., p. 69) and SES (Fig. 5.4, p. 73) examples If reliabilities are known, the corresponding error variances can be constrained; if unknown, the error variances may be modeled as free parameters provided that they are identifiable To keep in mind: we need more than one indicator per latent variable for identifiability and statistical testing --- leading to measurement models with multiple indicators or CFA

General structural model Part 1: Covariance structure and identification. Psychology 588: Covariance structure and factor models

General structural model Part 1: Covariance structure and identification Psychology 588: Covariance structure and factor models Latent variables 2 Interchangeably used: constructs --- substantively defined