1 / 10

Factor model of Ordered -Categorical Measures: Measurement Invariance

Factor model of Ordered -Categorical Measures: Measurement Invariance. Michel Nivard Sanja Franic Dorret Boomsma Special thanks: Conor Dolan OpenMx “ team ”. Ordinal factor model. This model works for ordered categorical indicators.

hollandl
Télécharger la présentation

Factor model of Ordered -Categorical Measures: Measurement Invariance

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Factor model of Ordered-Categorical Measures: Measurement Invariance Michel Nivard Sanja Franic Dorret Boomsma Special thanks: Conor Dolan OpenMx “team”

  2. Ordinal factor model • This model works for ordered categorical indicators. • It assumes these indicators are indicators of an underlying continuous (and normally distributed) trait.

  3. Ordinal indicators • Q: do you get angry? • 0: Never • 1: Sometimes • 2: Often 0 1 2

  4. Ordinal indicators • The latent distribution is restricted too mean=0 and variance = 1. Given these constraints we can estimate the the threshold parameters indicated in red and blue here. Either: μ and Σ Or: the Thresholds are identified.

  5. Ordinal indicators factor model.

  6. Ordinal indicators factor model. • Expected (polychoric) covariance: • Cov(Xi) = Σ = i it+ i • Expected means/threshold: • E(Xi)= μ = τi + iκ • Restrictions for identification (alternatives, Millsap 2004) •  = 1 • diag(22t+ 2)=diag() • μ = 0 = τi + iκ •  = factor loadings •  = factor (co)variance matrix •  = residual variances ε1 to εn • τ = intercepts • κ = factor mean

  7. Model restrictions for measurement invariance. • Configural invariance: • Group 1: • 1= 11 1t+ 1 • Group 2: • 2= 22 2t+ 2 • Restrictions: • Thresholds are equal over groups. • diag(11t+ 1)=diag(22t+ 2)=diag() • Indicator means: Group1: 0 Group2: FREE • 2 = 1 = 1

  8. Model restrictions for measurement invariance. • metric invariance: • Group 1: • 1= 1t+ 1 • Group 2: • 2= 2t+ 2 • Restrictions: • Thresholds are equal over groups. • diag(11t+ 1)=diag(22t+ 2)=diag() • τ : group1: 0 group2: FREE • Κ: set to 0 • 1 = 1 (however 2 = FREE)

  9. Model restrictions for measurement invariance. • Strong invariance: • This concerns the means model. • E(Xi)= μ1 = 0 • E(Xi)= μ2 = κ and τ2 = 0 • Thresholds are equal over groups. • diag(11t+ 1)=diag(22t+ 2)=diag() • τ : group1: 0 group2: 0 • Κ: set to 0, free in group 2. • 1 = 1 (however 2 = FREE)

  10. Model restrictions for measurement invariance. • Strict invariance: • Can we equate the matrix  over groups. • However at this point  is not a free parameter: • diag(11t+ 1)=diag(22t+ 2)=diag() • So we must chose arbitrary values for  that are equal for both groups. • Replace the residuals algebra (diag(22t+ 2)=diag()) with a residuals matrix.

More Related