Unidimensionality and interpretability of psychological instruments.

Abstract

One of the fundamental ideas in the constmction of psychological measurement instruments is that each instrument should be homogenous and measure one attribute only. The idea of unidimensionality is a central assumption of most models within both classical test theory and modem test theory (e.g., Gulliksen, 1950;Lord, 1980;McDonald, 1999). There are good statistical reasons for favoring one-dimensional models to solve measurement problems. Reasons of interpretation also speak in favor of a focus on unidimensionality, because if multiple attributes are measured, researchers will not know which attribute to invoke to account for a particular score.However, many observations in the literature have suggested that the unidimensionality requirement may have negative effects on the interpretability and usefulness ofthe resulting measure. One problem is that this requirement causes measures to focus on narrow aspects of phenomena. For example, Humphreys (1962) observed that the principle of unidimensionality caused the construct of intelligence to splinter into a large set of measures of narrowly defined cognitive abilities, causing the broad construct of intelligence to fall out of focus for a long time.Another indication that unidimensionality need not be a necessary characteristic of psychological instruments is that many instruments that have been proven to be highly useful for theoretical, diagnostic, and predictive purposes do not fulfill the unidimensionality requirement. For example, intelligence test batteries, such as the Wechsler series, are certainly not unidimensional but are considered to be extremely useful for purposes of diagnosis and prediction. In virtually any field of psychological measurement, there are numerous other examples of instruments that consist of different subtests aggregated into a composite score.The emphasis on unidimensionality is based on the idea that a variable should be unitary and express one characteristic only. However, there are situations in which variables are not seen as unitary. In a multiple regression analysis, for example, the independent variables are typically regarded as unitary, but the dependent variable is not. Instead, the main aim of a multiple regression analysis is to decompose the dependent variable into different components of 97

Becoming more specific: Measuring and modeling teachers' perceived usefulness of ICT in the context of teaching and learning

“…Following this procedure, we specified nested factor models with and without cross-loadings to sort out the general and specific components of the perceived usefulness construct and to test the existence of construct overlaps after controlling for general perceived usefulness (Models 3 and 4, Fig. 1 c, d; Gustafsson & Åberg-Bengtsson, 2010;Muth en & Asparouhov, 2012b). These steps were undertaken to obtain evidence on factorial validity (Kane, 2013).…”

Interdisciplinary Research in History Education

“…nicht überprüfb are Voraussetzungen wie Unidimensionalität, Intervallskalenniveau der Daten; Moosbrugger & Kelava, 2012) (Wright & Masters, 1982) aufgedeckt werden (Wu & Adams, 2013). Besonders relevant für die Beurteilung des Itemfi ts ist die sogenannte weighted-mean-square-oder "Infi t"-Statistik (Wright & Masters, 1982 (Gustafsson & Åberg-Bengtsson, 2010).…”

“…Allgemeiner spricht man hier auch von Nested-Factor-Modellen (Gustafsson & Åberg-Bengtsson, 2010) (Zumbo, 1999). Von einem Impact hingegen spricht man, wenn die Itemlösung von einem zusätzlichen relevanten Merkmal beeinfl usst wird (Zumbo, 1999), also etwa der Verfügbarkeit bestimmter Konzepte (z.B.…”

Mathematics Performance and Cognition (MPAC) Interview

“…Where ω h exceeded α (i.e., NF and ESRS), variability was present in the general factor loadings, but group factor loadings were relatively small, indicating that lumpiness in the scale was not attributable to multidimensionality. In every case, ω h met or exceeded the conventional minimum value of .70, suggesting that composite scores can be interpreted as reflecting a single common source of variance in spite of evidence for some within-scale multidimensionality (Gustafsson & Aberg-Bengtsson, 2010). Inspection of the 2-pl UIRT TIC in Figure 10 reveals the information curve for the Grade 2 MPAC interview to exceed 2.33 (reliability of .70) for the ability range of approximately -3.1 through 1.9.…”