The impact of departmental interdependencies and management accounting system use on subunit performance: A comment

Click to increase image sizeClick to decrease image size

Notes

¹The literature has used different labels for these two kinds of ‘fit’. For example, the interaction concept of ‘fit’ may also be labelled the moderation concept of ‘fit’; the selection concept of ‘fit’ could also be called the adaptation or congruence concept of ‘fit’ (see also, e.g. Hartmann and Moers, 1999; Luft and Shields, 2003).

²Alternatively, one could argue that since the selection concept of ‘fit’ does not rest on a (testable) assumption of optimisation, but constitutes an (untestable) axiom, any testing whether selection fit results in superior performance is futile.

³This seems the backside of the paradox, associated with the use of moderation models, between the ex ante assumption that managers do not adjust their MAS use to departmental interdependencies and the ex post assumption that they should (cf. Hartmann and Moers, 2003). Note that the objections raised in this comment are based on the fact that cross-sectional (rather than longitudinal) performance is the object of explanation.

⁴An interesting avenue is to explore differences in speed of adapting MAS to the context. If such differences in speed are predicted to be caused by context as well, the appropriate selection model would include a context×context interaction.

⁵H3 attempts to establish that performance is no direct result of interdependence. The corroboration of H3 only illustrates that companies with higher interdependence, all adapt their MAS in the same direction. This provides additional support for the existence of selection fit.

⁶Note that the difficulty of comparing selection and interaction models is not restricted to the area of MAS research. In the psychology literature there is much debate about the possibilities and impossibilities of matching mediation and moderation models (see, e.g. Holmbeck, 1997; Shrout and Bolger, 2002), and strong pleas to test mediation models through moderation statistics (see, e.g. MacKinnon et al., 2002). Note, however, that in this literature many constructs involved are traits, rather than variables, which hinders comparison with MAS-related contingency work, and that studies more often rely on longitudinal data (see, e.g. Marjoribanks, 2003, who combines mediation and moderation elements in one causal model, using longitudinal and cross-sectional data).