Trying to separate the wheat from the chaff: Construct- and faking-related variance on the Implicit Association Test (IAT)

Jessica Röhner, Torsten Ewers
  • Behavior Research Methods, February 2015, Springer Science + Business Media
  • DOI: 10.3758/s13428-015-0568-1

Using diffusion model analyses to separate construct-related and faking-related variance on the IAT

Photo by Patrick Fore on Unsplash

Photo by Patrick Fore on Unsplash

What is it about?

By reanalyzing a data set containing faked and unfaked IAT effects (Röhner et al., 2013), we investigated the possibility of using diffusion model analyses (Ratcliff, 1978; Voss & Voss, 2007) to separate construct-related from faking-related variance on the IAT.

Why is it important?

A multitude of research has shown that the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998) can be faked (e.g., De Houwer, Beckers, & Moors, 2007; Fiedler & Bluemke, 2005; McDaniel, Beier, Perkins, Goggin, & Frankel, 2009; Röhner, Schröder-Abé, & Schütz, 2011; Steffens 2004). Recent research has additionally highlighted that identifying IAT fakers is much harder than previously thought (Fiedler & Bluemke, 2005; Röhner, Schröder-Abé, and Schütz, 2013). Nevertheless, the traditional IAT effect (i.e., D measure; see Greenwald, Nosek, & Banaji, 2003a, 2003b) does not allow the researcher to separate construct- and faking-related variance from each other. Hence, in most cases, all variance will be treated as construct-related variance. Research by Klauer, Voss, Schmitz, and Teige-Mocigemba (2007) indicated that diffusion model analyses (Ratcliff, 1978; Ratcliff, Gomez, & McGoon, 2004; Ratcliff & Rouder, 1998; Voss & Voss, 2007) allow users to decompose the traditional IAT effect and thus may be useful for separating different sources of variance from each other. Klauer et al. (2007) found three dissociable IAT effects: IATv, which is suggested to comprise the construct-related variance of the traditional IAT effect, and IATa and IATt0, both of which may indicate faking (Klauer et al., 2007). It has been shown already that IATv is actually significantly associated with construct-related variance (Klauer et al., 2007). However, until now, research had not yet investigated whether and how faking-related variance might affect these newly developed IAT effects. Therefore, we tried to answer the question: Might it be possible to disentangle faking-related variance from construct-related variance by using diffusion model analyses? Our results complement previous findings by showing that diffusion model analyses might provide a helpful tool for correctly interpreting IAT effects.

Read Publication

http://dx.doi.org/10.3758/s13428-015-0568-1

The following have contributed to this page: Dr Jessica Röhner