Desktop version

Home arrow Psychology arrow The Wiley Blackwell handbook of the psychology of recruitment, selection and employee retention


Black-White group mean score differences on biodata inventories have also been explored, albeit to a much lesser extent than interviews. According to Bobko and Roth (2013), initial estimates of d for biodata inventories were about 0.33 (e.g., Bobko, Roth & Potosky, 1999). These authors noted that updating this finding was difficult given the limited research in this area. However, two studies offer some guidance. Potosky and colleagues (2005) combined and corrected for range restriction the results of a few primary studies to d = 0.57 (k=2, N = 6,115). More recently, Becton, Matthews, Hartley and Whitaker (2009), in a large study exploring the predictive validity of a biodata inventory for a wide variety of healthcare jobs, reported a Black-White d of 0.31 (uncorrected; N = 13,301). Combining these results, Bobko and Roth (2013) suggested an updated average d of about 0.39, noting that this estimate varies depending on construct saturation.

Assessment centres

Early in the literature it was thought that Black-White group mean score differences for assessment centres was small (Dean, Roth & Bobko, 2008), though the range reported by Bobko and Roth (2013) was quite wide (d = 0.03 to 0.60, favouring Whites). In the only meta-analysis on this topic to date, Dean and colleagues (2008) estimated an overall d of 0.52 favouring Whites (k = 17, N = 8,210). They found range restriction to be an important moderator, with d = 0.56 among job applicants (k = 10, N = 3,682) as compared with d = 0.32 among job incumbents (k = 6, N = 1,689). As Bobko and Roth note, assessment centres can measure a wide range of constructs, and cognitive ability can increase estimates of d. Interpreting these differences is problematic, though, as assessment centres generally comprise multiple exercises (e.g., simulations such as role-plays and in-basket exercises, personality inventories, cognitive ability tests, interviews) that assess multiple constructs. Thus, it is unclear whether most of the variance in these effect sizes is a function of exercises, constructs or even formats.

< Prev   CONTENTS   Source   Next >

Related topics