Abstract
Recent literature on parcel indicators in measurement models used in covariance structural modelling has mainly been concerned with statistical properties of parameter estimates. Less attention has been paid to measurement properties for inferring the assumed latent construct. The present study illustrates a two‐facet measurement model that considers items and parcels simultaneously to test whether the aggregated item parcel information can replace the disaggregated item information in representing the same latent construct. In particular, the two‐facet model tests the existence of unidimensionality in a two‐facet measurement design where items are nested within parcels. The present methodology emphasises the need for conceptually based parcels to facilitate interpretation of the latent construct. The methodology is applied to a multifaceted construct of political knowledge measured in a sample of secondary school students.
Notes
1. For a discussion of the measurement of political knowledge see Niemi and Junn (Citation1998) and Carpini and Keeter (Citation1993).
2. An anonymous reviewer suggested reporting the fit indices using maximum likelihood estimation with Satorra and Bentler (Citation1988) statistics in addition to the WLS‐based indices. For the original model we obtained χ2 344 = 868.48; p = .00 and RMSEA = .030 (.027; .032); p (close fit) = 1.0.
3. A LISREL script of the model is available on request from the first author.