4. The need for further studies
Previous studies have used different measures for both the dependent and independent variables. Where
the adoption or non-adoption ofABCsystems has been used as the dependent variable the terms ‘adoption’
and ‘non-adoption’ have been subject to different interpretations with some studies defining adoption as
actual ABC implementation and others defining it as actual implementation or a desire to implement it.
The studies have also generally allowed the respondents to self-specify whether their organization operate
an ABC system despite the fact that there is also some disagreement as to whether systems described
by survey respondents as ABC are really ABC systems (Dugdale and Jones, 1997; Innes and Mitchell,1997). Inconsistent measures have also been used for measuring the independent variables. Most of the
studies have used measures derived from a single question for obtaining non-factual data rather than using
composite scores derived from multiple questions (Foster and Swenson, 1997).
Apart from the studies by Gosselin and Krumwiede, the ABC studies based on a selection approach
to fit have used bivariate statistics to examine whether the difference between adopters and non-adopters
were statistically significant. Where the contextual variables are related to each other there is a danger that
spurious relationships may be reported. There is a need for tests to be undertaken using higher powered
multiple and logistic regression statistical tests that express the unique contribution of each variable by
systematically controlling for the impact of other variables in the model.
This research therefore seeks to remedy the above deficiencies. In particular, prior research has adopted
a much too simplistic approach to product cost system design. Instead of using only the adoption or nonadoption
of ABC systems as a measure of product cost system design this research uses four different
measures of cost system sophistication to capture the attributes of the product costing systems. This allows
for a more robust test of the relations among the predictor variables and cost system sophistication