Content area
Full Text
Keywords Teams, Reliability, Statistics
Abstract Previous research into the reliability of the "Belbin" team role self-perception inventory has concluded that internal consistency reliability is poor, raising concerns about the validity of the inventory as a selection and development tool. This paper argues that, in overcoming the problem of unequal scale lengths between respondents, researchers have tampered with the true error variance contained within item response sets, leading to a misapplication of Cronbach's alpha in this case. Using weighted average inter-item correlation coefficients in a formula unrelated to scale length, alternative estimates of the internal consistency reliability of the TRSPIs scales are derived For a large data set (n = 5,003), results show that the internal consistency of item responses is better than previous research suggests and tends towards the boundary at which internal consistency is considered acceptable in social and psychological research.
Introduction
The team role concept has become particularly important as organisations seek flatter structures, flexible working patterns and organisational networks to achieve competitive advantage. Although several typologies to describe team roles exist (Davis et al., 1992; Kirnan and Woodruff, 1994; Parker, 1990), Belbin's (1981, 1993a) schema enjoys good popularity, perhaps because it is supported by a team role self-perception inventory (TRSPI) that easily identifies a person's team role preferences (see Fisher et al, 2000; Senior, 1997, 1998). While results from the TRSPI should be used alongside observers' assessments of an individual's team role to get a more accurate picture (Senior and Swailes, 1998), it is important that the TRSPI demonstrates good reliability as a prerequisite to validity.
A number of papers have, however, questioned the internal consistency reliability of the TRSPI. Fisher et al. (1996) reported poor reliabilities for the eight-role TRSPI ranging from 0.01 to 0.66. Furnham et al. (1993a, b) also challenged the reliability and validity of the eight-role version after finding alphas ranging from 0.34 to 0.77 (see Belbin (1993b) for a rejoinder). The only study found reporting reliability for the later nine-role TRSPI in its original restricted-choice response format was by Broucek and Randall (1996), which also reported disappointing alpha values ranging from 0.25 to 0.52.
We aim to show, however, that previous reliability estimates have been heavily influenced by a data manipulation technique employed by researchers...