Date of Award

2000

Degree Type

Dissertation

Degree Name

Doctor of Philosophy in Psychology

Department

Psychology

First Advisor

Wayne F. Velicer

Abstract

When testing confirmatory factor analysis (CF A) models, the researcher may either utilize individual items as the manifest variables for the model, or use item parcels or composites, i.e., the sum or average of a set of individual items. Utilizing parcels instead of items may affect the chance of obtaining a proper solution, the fit of the model, and model parameter estimates. Unfortunately there is little previous research to guide the decision to choose between a parcel-level analysis and an item-level analysis. This investigation examined the current status of parceling in the literature and the impacts of parceling through three data simulation studies. In Study 1, three psychology journals were reviewed to determine the status of parceling in applied research. Parceling was utilized extensively, typically without justification or with a test of parcel unidimensionality. In Study 2, item-level solutions were compared with two-item parcel-level solutions across four general model structures, four levels of sample size, two levels of saturation and three levels of kurtosis. Percent of proper converging solutions was greater in item-level models, but only when the number of indicators per factor (p/f) was reduced to two in the parcel-level analyses. Mean item saturation and four measure of goodness of fit (GFI, CFI, NNFI and RMSEA) were improved in the parcel-level solutions. Study 3 extended the results found in Study 2 to four- and eight-item parcel models. Percent of proper converging solutions was decreased when p/f was decreased to two. Mean indicator saturation and goodness of fit indices improved as the number of items per parcel increased. Study 4 examined eleven mis-specified parcel model structures, most of which included non-unidimensional parcels. In general, the mis-specified parcel models fit just as well or better than the correctly specified item-level models. This demonstrates both the advantages of and some very serious problems with using parcels. Utilizing parcels can result in violations of intra-parcel unidimensionality, hidden correlated errors, fit index bias and incorrect model confirmation. It is recommended that measurement models should be evaluated only at the item level and parcels should be employed only after extensive preliminary testing.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.