Having a Fit: Impact of Number of Items and Distribution of Data on Traditional Criteria for Assessing IRT's Unidimensionality Assumption

Purpose Confirmatory factor analysis fit criteria typically are used to evaluate the unidimensionality of item banks. This study explored the degree to which the values of these statistics are affected by two characteristics of item banks developed to measure health outcomes: large numbers of items...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Quality of life research 2009-05, Vol.18 (4), p.447-460
Hauptverfasser: Cook, Karon F., Kallen, Michael A., Amtmann, Dagmar
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Purpose Confirmatory factor analysis fit criteria typically are used to evaluate the unidimensionality of item banks. This study explored the degree to which the values of these statistics are affected by two characteristics of item banks developed to measure health outcomes: large numbers of items and nonnormal data. Methods Analyses were conducted on simulated and observed data. Observed data were responses to the Patient-Reported Outcome Measurement Information System (PROMIS) Pain Impact Item Bank. Simulated data fit the graded response model and conformed to a normal distribution or mirrored the distribution of the observed data. Confirmatory factor analyses (CFA), parallel analysis, and bifactor analysis were conducted. Results CFA fit values were found to be sensitive to data distribution and number of items. In some instances impact of distribution and item number was quite large. Conclusions We concluded that using traditional cutoffs and standards for CFA fit statistics is not recommended for establishing unidimensionality of item banks. An investigative approach is favored over reliance on published criteria. We found bifactor analysis to be appealing in this regard because it allows evaluation of the relative impact of secondary dimensions. In addition to these methodological conclusions, we judged the items of the PROMIS Pain Impact bank to be sufficiently unidimensional for item response theory (IRT) modeling.
ISSN:0962-9343
1573-2649
DOI:10.1007/s11136-009-9464-4