Organizational identification: Development and testing of a conceptually grounded measure

Abstract

There is continuing debate in the literature as to how organizational identification (OID) should be conceptualized and operationalized. We present a new six-item measure of OID that includes both cognitive and affective components and that integrates the main dimensions of OID found in the literature. The new measure comprises three main subcomponents: self-categorization and labelling, sharing of organizational goals and values, and a sense of organizational belonging and membership. The measure was tested on two separate samples of over 600 employees working in the UK National Health Service (NHS) using Confirmatory Factor Analysis. The results provided support for the proposed three-component conceptualization of OID. However, the three subcomponents were highly intercorrelated and showed low discriminant validity. We therefore propose a single overall measure of OID. This six-item aggregate scale has acceptable psychometric properties and provides a theoretically meaningful, but parsimonious, measure of OID for use in field research.

Notes

¹In selecting these cutoff points we recognize that there is some debate in the literature about what constitutes a reasonable χ²/df ratio to identify a good fitting model (Bollen, 1989; Kelloway, 1998; Marsh & Hocevar, 1985; Mueller, 1996). Bollen (1989, p. 278), for example, notes that “there is no consensus on what represents a ‘good’ fit, with recommendations ranging from ratios of 3, 2 or less…to as high as 5”, and Kelloway (1998) suggests that finding an absolute cutoff value for this indicator is problematic as ratios below 2 could indicate an overfitting model (p. 28).

²Once again, there is debate in the literature about what are reasonable RMSEA and SRMR cutoffs for good fitting models (Hu & Bentler, 1999; Steiger, 2000). Robust cutoffs of 0.05 for the RMSEA combined with 0.06 for the SRMR have been suggested; these cutoffs however, sometimes risk rejecting models unnecessarily, so other combinations have been suggested (0.06 for RMSEA and 0.09 for SRMR) to reduce type II errors (Hu & Bentler, 1999). Indeed Steiger (2000) suggests that using an absolute cutoff of 0.05 risks being overly stringent.

³It has been suggested that when there are very small differences in the ECVI for two different models (for example a difference of around 0.04) then it might not be sensible to reject the worst performing model purely on the basis of this indicator, especially if other fit indices suggest good performance and this model is more parsimonious (Mueller, 1996).

Additional information

Notes on contributors

Martin R. Edwards

The authors are grateful to Rolf van Dick for his advice on an earlier version of this article.