Scaling Response Processes on Personality Items Using Unfolding and Dominance Models

Abstract

Abstract. Stark, Chernyshenko, Drasgow, and Williams (2006) and Chernyshenko, Stark, Drasgow, and Roberts (2007) suggested that unfolding item response theory (IRT) models are important alternatives to dominance IRT models to describe the response processes on self-report personality inventories. To obtain more insight into the structure of personality data, we investigated whether dominance or unfolding IRT models are a better description of the response processes on personality trait inventories constructed using dominance response processes or ideal-point response processes. Data from 866 adolescents on a Dutch personality inventory, the NPV-J ( Luteijn, van Dijk, & Barelds, 2005 ), and from 704 adolescents on a Dutch translation of an Order scale ( Chernyshenko et al., 2007 ) were used. Results from Stark et al. (2006) and Chernyshenko et al. (2007) were partly supported. The self-report inventory that was constructed using dominance response processes (NPV-J) consisted mostly of items with monotonically increasing item response functions (IRFs), but some IRFs were single-peaked. The Order scale (constructed on the basis of ideal-point response processes) consisted of items with monotonically increasing, decreasing, and single-peaked IRFs. Further implications for personality test construction are discussed.

Uncovering curvilinear relationships between conscientiousness and job performance: How theoretically appropriate measurement makes an empirical difference.

“…That is, items were developed and retained by these firms based on analyses such as item-total correlations, factor analysis, and/or conventional IRT modeling, all of which carry dominance assumptions. Although the studies reviewed above (e.g., Carter & Dalal, 2010;Stark et al, 2006;Tay et al, 2009;Weekers & Meijer, 2008) show that the ideal point model fits data from dominance scales better than dominance models, measures specifically developed under ideal point assumptions (e.g., Chernyshenko et al, 2007) may show even greater gains in variance explained for curvilinear relations that those found here. Future research should explore this possibility as well.…”

“…Moreover, they showed the possibility of considerable differences in the top-down rank-order of individuals between dominance and ideal-point IRT scores. Weekers and Meijer (2008) demonstrated that ideal point models showed better fit than dominance models when applied to responses from a Dutch personality inventory as well as responses from a Dutch translation of Chernyshenko et al's (2007) Orderliness scale. In addition, Weekers and Meijer found the correlation between dominance model-estimated person standing on the latent trait continuum (i.e., estimates) and GGUM based estimates to be .98 and .99 for the two scales, respectively.…”

Item Response Theory and Clinical Measurement

“…This feature of the model produces a bell-shaped IRC. Stark et al (2006) fit both dominanceand ideal-point models to scales from a wellknown personality inventory (see also Weekers & Meijer 2008). Results indicated that several items operated more consistently with an idealpoint representation than a dominance representation.…”

Ideal Point Modeling of Non-cognitive Constructs: Review and Recommendations for Research

“…For non-cognitive constructs, the construct continuum is better understood as various locations , since the “correct” response is simply that which most accurately reflects one’s position along the latent trait. This is reflected in the references to “item location parameters” when IRT is applied to non-cognitive constructs (Reise et al., 2001; Reise and Haviland, 2005; Vittersø et al., 2005; Gomez, 2008; Weekers and Meijer, 2008; Zampetakis, 2010; Reise et al., 2011).…”