In the following sections, we review previous findings regarding African-American and White group mean score differences on the methods of interviews, biodata, assessment centres, simulations and situational judgement tests. It should be noted that racial-ethnic group mean score differences on each method largely depend on the saturation of various constructs, most notably cognitive ability (Bobko & Roth, 2013). Thus, while estimates of the Black-White ds exist for the various methods, these differences are ostensibly a function of the constructs that the methods aim to assess.
Interviews, the most commonly used selection method, have been previously proffered as an alternative to cognitive ability tests due to supposed lower racial-ethnic group mean score differences. Huffcutt and Roth (1998) provided one of the first meta-analyses investigating these differences on employment interviews (see also Hough et al., 2001). These authors reported an overall mean d of 0.25 favouring Whites (k = 31, N = 10,476). However, they identified a number of moderators, including the level of structure, job complexity, cognitive ability saturation and minority representation in the sample (all ds favouring Whites unless otherwise noted). When interviews had more structure, the effect size tended to decrease. Interviews with lower structure demonstrated a d of 0.32 (k = 10, N = 1,659) as compared with a d of 0.23 (k=21, N = 8,817) for interviews with a higher degree of structure. (Potosky, Bobko and Roth, 2005, later applied a statistical correction for range restriction to the high-structure d estimate of 0.23, increasing this estimate to d = 0.31.) Interviews for low-, medium- and high-complexity jobs demonstrated respective ds of 0.43 (k = 12, N = 5,148), 0.22 (k = 13, N = 4,093) and 0.09 (favouring African- Americans; k = 5, N = 768). Interviews with a lower saturation of cognitive ability demonstrated a smaller effect size than interviews with a higher saturation of cognitive ability, with d decreasing from 0.45 (k=7, N = 1,213) to 0.26 (k = 17, N = 6,118). Finally, samples with a lower ratio of Black participants (<30%) demonstrated a lower d as compared with samples with a higher ratio of Black participants, with d decreasing from 0.41 (k = 12, N = 4,797) to 0.15 (k = 19, N= 5,679).
As indicated above, the effect sizes for interviews vary depending on the constructs measured (Huffcutt, Conway, Roth & Stone, 2001). Huffcutt and colleagues’ meta-analysis provided a taxonomy of constructs for interviews that included cognitive constructs (e.g., general intelligence, job knowledge and skills), personality constructs (e.g., Big Five personality factors), social skills, occupational interests, organizational fit and physical attributes. Separate effect sizes were reported for all the identified constructs and by low versus high structure (though many estimates were obtained from a small number of studies and small sample sizes; e.g., k = 1 and N = 103 for interviews measuring job knowledge and skills). Effect sizes demonstrated a considerable variation depending on the construct and degree of structure. For example, the mean d for high-structure interviews measuring the organizational fit construct of ‘values and moral standards’ was 0.12 favouring African- Americans (k = 3, N = 568), while the mean d for low-structure interviews measuring the mental capability construct of ‘general intelligence’ was 0.58 favouring Whites (k = 5, N = 1,564). As another example, high-structure interviews assessing personality constructs tended to demonstrate more modest effect sizes as compared with low-structure interviews assessing the same traits. To demonstrate, Huffcutt and colleagues reported a d of 0.17 (k=7, N = 2,889) for high-structured interviews measuring conscientiousness compared with d = 0.41 for low-structure interviews measuring conscientiousness (k = 8, N = 2,554).
Finally, Roth, van Iddekinge, Huffcutt, Eidson and Bobko (2002) highlighted that ds for interviews are likely larger than those reported in earlier studies due to range restriction (see also Potosky et al., 2005). For example, Huffcutt and Roth (1998) reported a mean d of 0.10 (favouring Whites) in their meta-analysis for high-structured interviews that were behavioural in nature (k=6, N= 1,614). Roth and colleagues (2002) reported ds for two forms of a behavioural interview of 0.34 and 0.54, with corrected estimates increasing to 0.36 and 0.56, respectively. Note that this was a single primary study, and as Roth and colleagues (2002) highlighted in the limitations of their study, much more work needs to be done to understand how range restriction affects Black-White group differences in interviews, whether behavioural or otherwise. Overall, given the variability in the constructs measured, degree of structure and range restriction in studies using interviews, Bobko and Roth (2013) suggest a tentative range for d of 0.31-0.46.