Larry Williams 2008-09-12

From OBTnotes

Measurement Models for Linking Latent Variables and Indicators

A Review of Alternatives for Organizational Researchers

• from VCU

Abstract

This lecture first reviews the various measurement model options for linking latent variables to indicators that are available to organizational researchers. A special emphasis is placed on the option of parcels, created by combining subsets of items to form indicators. Next, a review is presented of 27 articles from the major HRM journals that have used parcels, with a focus on conceptual/theoretical and empirical issues. Recommendations for organizational researchers for improved use of parcels are also provided.

Notes

• Ways to analyse items that relate to a latent variable
  • individually
  • aggregate into one scale
  • separate into parcels

• Methods for separating items into parcels
  • Random assignment
  • Item-to-Construct Balance - rank on factor loadings then do a draft to split them up
  • A priori construction - use information about the items to balance things (like reverse coded items) across parcells
  • Correlational Algorithm - match items my correlation into groups
  • Radial Algorithm
  • Correlated uniqueness
  • Internal consistency - figure out which items represent different facets of the latent variable and group by those facets
  • Domain representative approach - combine items across facets so that each parcel represents all the facets

• Recommendations
  • Consider dimensionality
    • Examine construct definitions and empirical findings from previous research
  • Consider broader focus on research
  • Examine information on distributions for all items
  • Examine correlation among items for each facet and across facets
  • Examine EFA and CFA results for item set associate with each latent variable
  • Shoot for 3 o 4 indicators/parcels for each latent variable