The influences on Activity-Based Costing adoption as an optimal costing system design: Evidence from Saudi Arabia

Research Question: Under which conditions is the adoption of activity-based costing (ABC) beneficial to the business-unit and, hence, represents the optimal costing system design (CSD)? Motivation: Identifying the situations under which ABC is beneficial with regard to the performance of the business-unit and, thus, represents the optimal CSD is important in order to make a correct decision to invest in ABC. This necessitates the adoption of the matching form of fit when examining the relationship between the influential factors and ABC adoption as an optimal CSD due to its ability to reflect when ABC is beneficial and, therefore, indicate the optimal CSD. This study builds on Ittner et al. (2002) to examine this relationship from the standpoint of the matching form of fit, and also extends this line of research by examining the effect of the most key influential factors, using a sufficiently comprehensive multi-dimensional production complexity (PC) measure. Idea: This paper examines the role of the key influential factors on ABC adoption as an optimal CSD from the standpoint of the matching form of fit, using a sufficiently comprehensive multi-dimensional PC measure. Data: The data were collected from 200 Saudi manufacturing business-units. Tools: The questionnaire survey strategy was used to collect the data, which were then analysed using the residual analysis (RA) technique. Findings: Unexpectedly, the results of the RA showed that none of the examined key factors of competition, indirect costs, PC and information technology quality (IT quality) affects ABC adoption as an optimal CSD from the viewpoint of the matching form of fit. Contribution: This research contributes to the ABC adoption research by examining the situations under which ABC is beneficial and, hence, represents the optimal CSD through examining the impact of the key influential factors on ABC adoption as an optimal CSD from the standpoint of the matching form of fit.

between one influential factor, which was the plant operational characteristics related to PC, and ABC adoption as an optimal CSD from the standpoint of the matching form of fit. Nevertheless, the study suffered from two limitations. First, it examined the effect of only a single factor despite the existence of other key influential factors, such as the level of competition, indirect costs, and IT quality. Second, it employed a limited measure of PC, although more comprehensive measures can be utilised.
Given: (1) the importance of identifying the situations under which ABC is beneficial and, hence, represents the optimal CSD to make a correct decision regarding investing in ABC; and (2) the limitations of Ittner et al. (2002) regarding the number of examined contingency factors and measurement of PC, the aim of this paper is to examine the influence of the key contingency factors, namely the level of competition, the level of indirect costs, PC and IT quality, on ABC adoption as an optimal CSD from the standpoint of the matching form of fit, using a sufficiently comprehensive multi-dimensional PC measure. Using questionnaire data collected from Saudi manufacturing business-units, the residual analysis (RA) results showed that none of the examined key factors affects ABC adoption as an optimal CSD from the viewpoint of the matching form of fit. This research contributes to the existing literature on ABC adoption research by examining the situations under which ABC is beneficial and, hence, represents the optimal CSD through testing the effect of the key influential factors on ABC adoption as an optimal CSD from the standpoint of the matching form of fit, using a sufficiently comprehensive multi-dimensional PC measure.
The remainder of this paper is organised as follows. Section 2 reviews the relevant literature and highlights the research gaps and opportunities. Section 3 develops hypotheses related to the effect of the cost of error-related and the cost of measurement-related factors on ABC adoption as an optimal CSD from the standpoint of the matching form of fit. Section 4 explains the research methodology and methods. Section 5 provides the results of this research. Section 6 discusses these results and concludes this paper.

Literature review
The underlying assumption of the matching form of fit is that the relationship between CSD and the outcome representing its optimality is curvilinear and affected by the contingency factors (Burkert et al., 2014;Chenhall & Chapman, 2006). This assumption indicates that, at any level of the contingency factor, there is only one CSD level that provides the maximum outcome level -i.e., the optimal CSD level -and that any deviation -i.e., a misfit -from this CSD level reduces the outcome (Schoonhoven, 1981;Donaldson, 2001;Chenhall & Chapman, 2006;Meilich, 2006;Gerdin & Greve, 2008;Burkert et al., 2014). In other words, a fit is achieved and the outcome is maximised when there exists a match between the levels of the contingency factor and the CSD (Klaas & Donaldson, 2009;Burkert et al., 2014). In contrast, a misfit is attained and the outcome is reduced when there exists a mismatch between the two (Chenhall & Chapman, 2006;Klaas & Donaldson, 2009;Burkert et al., 2014). Misfit can be in the form of an over-fit, where the CSD level exceeds the required level by the contingency factor, or an under-fit, where the CSD level is below the level required by the contingency factor (Donaldson, 2001;Meilich, 2006;Klaas & Donaldson, 2009). Given the above, examining the relationship between the contingency factor and the optimal CSD from the perspective of the matching form of fit entails examining: (1) the direction of the association between the contingency factor and optimal CSD; and (2) the magnitude and sign of the impact of a misfit between both on the outcome. When both: (1) the direction of the association between the contingency factor and the optimal CSD; and (2) the magnitude and sign of the impact of the misfit on the outcome conform to expectations, there exists support for the hypothesised effect of the contingency factor on the optimal CSD from the perspective of the matching form of fit.
As mentioned in Section 1, it is crucial to investigate the relationship between the influential factors and ABC adoption as an optimal CSD from the standpoint of the matching form of fit to identify the situations under which ABC is beneficial and represents the optimal CSD, which afterwards assists when making decisions regarding investment in ABC. Despite the importance of utilising the matching form of fit, most studies have utilised the selection form of fit (e.g., Bjørnenak, 1997;Gosselin, 1997;Malmi, 1999;Al-Mulhem, 2002;Brown et al., 2004;Kallunki & Silvola, 2008;Schoute, 2011;Jusoh & Miryazdi, 2016), which models the relationship between the contingency factor and ABC adoption as an optimal CSD as an association between the contingency factors and ABC adoption without incorporating any outcome variable that represents the optimality of ABC (Gerdin & Greve, 2004;Hartmann, 2005). This is known as the equilibrium assumption (Chenhall & Chapman, 2006;Meilich, 2006;Burkert et al., 2014). Other studies have used the moderation form of fit (e.g., Frey & Gordon, 1999;Cagwin & Bouwman, 2002;Elhamma, 2012;Elhamma & Zhang, 2013;Krumwiede & Charles, 2014;Maiga et al., 2014), which models the relationship between the contingency factors and ABC adoption as an optimal CSD as a linear rather than a curvilinear association between CSD and the outcome representing its optimality, with the contingency factors affecting this linear relationship (Meilich, 2006;Gerdin & Greve, 2008;Burkert et al., 2014). 2 Only Ittner et al. (2002) examined the relationship between one influential factor, which was the plant operational characteristics related to PC, on ABC adoption as an optimal CSD from the standpoint of the matching form of fit. The study found that plant operational characteristics related to PC had a weak positive effect on ABC adoption as an optimal CSD from the viewpoint of the matching form of fit. In particular, Ittner et al. (2002) found that the plant operational characteristics related to PC factor is positively associated with ABC adoption as an optimal CSD, but that only an over-fit has a weak negative effect on financial performance that represents the optimality of ABC. Despite the study's originality, it has two limitations. First, it examined the effect of only a single cost of error-related factor, namely plant operational characteristics related to PC, on ABC adoption as an optimal CSD from the perspective of the matching form of fit. It is necessary to identify how other predominant cost of error-related factors -namely the level of competition and indirect costs -and the main cost of measurement-related factor, namely IT quality, affect ABC adoption as an optimal CSD from the perspective of the matching form of fit. This is because the prior literature emphasised that, in addition to PC, these factors also influence ABC adoption as an optimal CSD (Cooper, 1988b;Kaplan & Cooper, 1998). Examining the impact of these factors will provide a clearer insight regarding the situations under which adopting ABC is beneficial. Second, Ittner et al. (2002) used a limited measure of PC. Specifically, the study did not use a sufficiently comprehensive multi-dimensional PC measure. Rather, Ittner et al. (2002) used separate constructs that, in total, covered the three PC dimensions of production type, product diversity and the frequency of introducing new products. Prior research suggested that there are many PC dimensions, such as product customisation (e.g., Bjørnenak, 1997;Brierley, 2011), product complexity (Swenson, 1998), product diversity (Cooper, 1988b), the frequency of introducing new products (Nguyen & Brooks, 1997), the frequency of making changes to products and manufacturing processes (Cagwin & Bouwman, 2002) and the production period of products (Duh et al., 2009). Using a sufficiently comprehensive multi-dimensional PC measure is crucial in order to identify correctly the true effect of PC on ABC adoption as an optimal CSD from the standpoint of the matching form of fit.
Both (1) the importance of identifying the situations under which ABC is beneficial and, hence, represents the optimal CSD to make a correct decision in investing in ABC; and (2) the limitations of Ittner et al. (2002) encourage further research to examine the impact of the key contingency factors, namely the level of competition, the level of indirect costs, PC and IT quality, on ABC adoption as an optimal CSD from the standpoint of the matching form of fit, using a sufficiently comprehensive multi-dimensional PC measure.

Hypotheses development
The hypotheses developed in this section are related to the effect of the cost of error-related factors -namely the level of competition, indirect costs and PC -and the cost of measurement-related factor, namely IT quality, on ABC adoption as an optimal CSD from the perspective of the matching form of fit. With respect to the cost of error-related factors, the ABC adoption research has suggested that the level of competition, the level of indirect costs and PC positively affect ABC adoption (e.g., Cooper, 1988b;Kaplan & Cooper, 1998;Bjørnenak, 1997;Malmi, 1999;Al-Omiri & Drury, 2007). This is due to the association between these factors, the cost of error and CSD (Cooper, 1988b). The risk of competitors being able to exploit any errors made by the business-unit and, hence, the danger of reducing the business-unit's profitability is enhanced when the level of competition is high (Cooper, 1988b;Cooper & Kaplan, 1991). Thus, in highly competitive environments, ABC is needed, owing to its ability to assign indirect costs to products more precisely than traditional costing systems (TCSs) and, therefore, decrease the number of errors (Cooper & Kaplan, 1991;Bjørnenak, 1997;Malmi, 1999). Ultimately, ABC adoption in highly competitive environments contributes towards reducing the competitors' opportunity to benefit from erroneous decisions made by the business-unit based on inaccurate costs (ibid).
When the levels of indirect costs and PC are high and TCS is being used, the possibility of reporting distorted product costs and, thus, the cost of error are likely to be high (Cooper, 1988b;Kaplan & Cooper, 1998;Booth & Giacobbe, 1998;Nguyen & Brooks, 1997). High levels of indirect costs increase the chance of having costs at various levels -i.e., the unit, batch, product and facility levelsmost of which cannot be accurately assigned to products by TCS (Cooper, 1990). High levels of PC suggest that the amount of resources consumption is highly likely to vary between products and that the amount of batch-and product-level costs is likely to increase (Cooper, 1988a;1988b;1989a;Estrin et al., 1994;Kaplan & Cooper, 1998;Swenson, 1998;Al-Omiri & Drury, 2007;Schoute, 2011). Accordingly, ABC would be the optimum choice in high indirect cost and PC environments, given that ABC uses a large number of activity-based cost pools and volume-and non-volume-based cost drivers and, thus, is capable of accounting for resource consumption variations and accurately assigning most levels of indirect costs to products. In contrast, when the levels of indirect costs and PC are low, using TCS is not harmful because TCS causes minimal or no product cost distortion and therefore low or no errors, respectively (Nguyen & Brooks, 1997;Booth & Giacobbe, 1998;Kaplan & Cooper, 1998;Brierley et al., 2001;Charaf & Bescos, 2013). Thus, adopting ABC in such environments is less likely to add value and may, when the levels of indirect costs and PC are extremely low, be harmful because of the over-investment in the CSD (Kaplan & Cooper, 1998;Stuart, 2013).
The empirical evidence has provided inconsistent results regarding the effect of the level of competition, indirect costs and PC on ABC adoption. Several studies have found, as suggested by ABC theory, positive effects of these factors on ABC adoption, while others have not or even found negative effects (for example, see the mixed results of Bjørnenak, 1997;Nguyen & Brooks, 1997;Booth & Giacobbe, 1998;Malmi, 1999;Chen et al., 2001;Cohen et al., 2005;Al-Omiri & Drury, 2007;Brierley, 2011;Jusoh & Miryazdi, 2016;Pokorná, 2015). These inconsistent findings may have been due to, for example, the differences between studies in relation to the definitions utilised for ABC adoption and nonadoption, the measurements used for the factors, the adopted statistical analysis techniques and the selected samples. Nevertheless, the theory asserts the positive influences of these factors on ABC adoption. The positive association between each of the level of competition, indirect costs and PC and ABC adoption indicates the fit between each contingency factor and ABC adoption that provides the highest level of the outcomes reflecting CSD optimality. This suggests a positive relationship between each factor and ABC adoption as an optimal CSD. Any misfit between the two is expected to have a negative influence on the outcomes.
Specifically, although using ABC at low levels of competition, indirect costs and PC is associated with a low cost of error due to ABC's provision of more accurate product costs that can assist in making informed product-related decisions, using ABC in such situations is associated with a high cost of measurement (Cooper, 1988b;1989b;Cooper & Kaplan, 1991;Stuart, 2013;Drury, 2018). The cost of measurement of using ABC in such cases will probably exceed ABC's benefits in terms of reducing the cost of error, i.e., the benefit of ABC in relation to the provision of more accurate product costs that can assist in making informed product-related decisions. Therefore, it is uncertain that using ABC is beneficial at competition, indirect costs and PC levels that are below those required by ABC.
On the other hand, despite the association with the low cost of measurement, using TCS rather than ABC at high levels of competition, indirect costs and PC is associated with a high cost of error due to the provision of inaccurate product costs that leads to the making of inferior product-related decisions (Cooper, 1988b;1989b;Cooper & Kaplan, 1991;Stuart, 2013;Drury, 2018). The cost of error when using TCS rather than ABC in such cases will probably outweigh TCS's benefits in terms of reducing the cost of measurement. Accordingly, it is unlikely that TCS is considered beneficial at competition, indirect costs and PC levels higher than those required by TCS. Based on the above discussion, the following three hypotheses will be tested:

H1.
From the perspective of the matching form of fit, the level of competition is expected to have a positive influence on ABC adoption as an optimal CSD that generates the highest levels of outcomes. This positive influence includes two elements; namely, a positive association between the level of competition and ABC adoption as an optimal CSD and a negative impact of the misfit between the level of competition and ABC adoption as an optimal CSD on the outcomes.

H2. From the perspective of the matching form of fit, the level of indirect costs is expected to have a positive influence on ABC adoption as an optimal CSD that generates the highest levels of outcomes. This positive influence includes two elements; namely, a positive association between the level of indirect costs and ABC adoption as an optimal CSD and a negative impact of the misfit between the level of indirect costs and ABC adoption as an optimal CSD on the outcomes.
H3. From the perspective of the matching form of fit, PC is expected to have a positive influence on ABC adoption as an optimal CSD that generates the highest levels of outcomes. This positive influence includes two elements; namely, a positive association between PC and ABC adoption as an optimal CSD and a negative impact of the misfit between PC and ABC adoption as an optimal CSD on the outcomes.
Regarding the cost of measurement-related factor, the ABC adoption research suggests that IT quality positively impacts on ABC adoption (Cooper, 1988b;Cooper, 1989b;Al-Omiri & Drury, 2007;Stuart, 2013;Krumwiede, 1998). High IT quality indicates that a large amount of information concerning various functions -e.g., purchasing and manufacturing -is available in the information system. This, in turn, reduces the cost of measurement of complex costing systems, such as ABC, because the required detailed information about resources, products and production processes and, hence, activities and cost drivers can be obtained at almost no cost (Cooper, 1988b;Cooper, 1989b;Al-Omiri & Drury, 2007;Stuart, 2013;Krumwiede, 1998;Baxendale & Jama, 2003;Kaplan & Cooper, 1998). Another argument suggests that, when the IT quality is high, it is unnecessary to adopt ABC (Krumwiede, 1998). This is due to the satisfaction with the information provided by high-quality information systems, which reduces the desire to invest resources in applying ABC (ibid). The empirical evidence regarding the effect of IT quality is mixed. Al-Omiri and Drury (2007) found no effect of IT quality on ABC adoption, while Krumwiede (1998) found that IT quality can both encourage and discourage the adoption of ABC. These mixed results can be attributed to one or more of the reasons for the inconsistent results mentioned earlier in this section. Despite these mixed results, the theory supports the positive effect of the IT quality factor on ABC adoption. This positive relationship between IT quality and ABC adoption indicates the fit between the two that delivers the highest level of outcomes, reflecting CSD optimality. This suggests a positive relationship between IT quality and ABC adoption as an optimal CSD. Any misfit between the two is anticipated to have a negative influence on the outcomes.
In particular, although using ABC at low levels of IT quality is linked with a low cost of error because of ABC's provision of more accurate product costs that can assist in making informed product-related decisions, using ABC in such a situation is associated with a high cost of measurement (Cooper, 1988b;1989b;Cooper & Kaplan, 1991;Stuart, 2013;Drury, 2018). The cost of measurement of using ABC in such a case will probably exceed ABC's benefits in terms of reducing the cost of error, i.e., the benefit of ABC in relation to the provision of more accurate product costs that can assist in making informed product-related decisions. Thus, it is doubtful that using ABC is beneficial at IT quality levels lower than those suitable for ABC.
In contrast, notwithstanding the association with the low cost of measurement and the possibility of having high or low costs of error, using TCS rather than ABC at high levels of IT quality can decrease the chance of having more accurate product costs. There is a high possibility that the cost of errors when using TCS rather than ABC at high levels of IT quality is higher than TCS's benefits regarding reducing the cost of measurement. This, in turn, may reduce the possibility that TCS is considered beneficial at IT quality levels higher than those compatible with TCS. Based on the above, the following hypothesis will be tested:

H4.
From the perspective of the matching form of fit, IT quality is expected to have a positive influence on ABC adoption as an optimal CSD that generates the highest levels of outcomes. This positive influence includes two elements; namely, a positive association between IT quality and ABC adoption as an optimal CSD and a negative impact of the misfit between IT quality and ABC adoption as an optimal CSD on the outcomes.

Strategy, context, and non-response bias assessment
In order to achieve the research aim, it is necessary to collect quantitative data from a large number of cases and analyse the data using statistical analysis techniques. Hence, the questionnaire survey strategy was utilised, as it satisfies these needs (Ryan et al., 2002;Collis & Hussey, 2014). 3 The questionnaire was designed, constructed, pre-tested and administrated following Dillman et al. (2014)'s guidelines in order to maximise the benefits and minimise the limitations of the questionnaire data collection method. The data for this research were collected from Saudi manufacturing business-units. Saudi Arabia was selected as the research context mainly due to the many significant changes that have been After evaluating the suitability of many databases, including the MCI, SIDF, the Saudi Industrial Property Authority (MODON) and the Royal Commission for Jubail and Yanbu (RCJY) databases, the MODON and RCJY databases were selected because: (1) these two databases do not overlap and, together, cover most of the Saudi industrial cities; (2) the MCI database was being updated; and (3) the SIDF database contained only manufacturing business-unit that obtained support from the fund. Despite being extensive and updated, the MODON and RCJY databases lack important contact details, which made it necessary to make the initial contact with the business-units through visits or phone calls. Making the initial contact in this way necessitated decreasing the sample size of this research to avoid high costs and a lengthy time period. However, it enhanced the validity of the results, since it made it possible to identify accurately the person responsible for assigning the indirect costs, with a solid knowledge about the business-unit which, to the author's knowledge, was not the case in the prior research on ABC adoption, apart from Hoque (2000). Initially, the questionnaire was sent to 368 contacted business-units, after which multiple phone and e-mail reminders were directed to non-respondents. This process resulted in obtaining 233 responses, representing an initial response rate of 63.3%. Performing checks in relation to missing data, inconsistent questionnaire answers, outliers and normality decreased the number of respondents and response rate to 200 and 54.3%, respectively (Tabachnick & Fidell, 2007;Hair et al., 2010;Hair et al., 2017). The non-response bias was checked through comparing early respondents (total = 120) with late ones (total = 80) regarding the examined influential factors on ABC adoption (Armstrong & Overton, 1977;Sax et al., 2003;Van der Stede et al., 2005), and no significant differences were found between the two groups.

Construct measurement
The constructs representing the influential factors, ABC adoption and the outcomes involved in this research were measured based on the prior literature or the researcher's development. Specifically, the level of competition (COMP) was measured using a five-point Likert-scale question adapted from Khandwalla (1972), Drury and Tayles (2000) and Brierley (2007). The level of indirect costs was measured as the percentage of the indirect manufacturing costs to total manufacturing costs (CostStructure-MANUFACTURING) (Brierley, 2007). 4 Production complexity (PC) was measured using two five-point Likert-scale questions. These questions were developed based on the prior literature (e.g., Estrin et al., 1994;Brierley, 2011;Nguyen & Brooks, 1997;Swenson, 1998;Cagwin & Bouwman, 2002;Duh et al., 2009) and the findings of exploratory interviews conducted by the researcher involving eight Saudi manufacturing business-units. The first question concerned the product customisation PC dimension (e.g., Bjørnenak, 1997;Brierley, 2011), while the second covered the other five PC dimensions; namely, product complexity (Swenson, 1998), product diversity (Cooper, 1988b), the frequency of introducing new products (Nguyen & Brooks, 1997), the frequency of making changes to products and manufacturing processes (Cagwin & Bouwman, 2002) and the production period for products (Duh et al., 2009).
The level of IT quality was measured using a five-point Likert-scale question that attempts to assess the extent to which the business-unit adopts various advanced manufacturing technologies (AMT). These technologies include: (1) product design technologies, e.g., computer-aided design and engineering; (2) process technologies, e.g., flexible manufacturing systems; (3) logistic and planning technologies, e.g., production scheduling systems; and (4) information exchange technologies, e.g., common databases. A question related to AMT was included to measure IT quality due to the association between AMT adoption, IT quality and the cost of measurement (Cooper, 1988b;Kaplan & Cooper, 1998). Specifically, the adoption of various AMTs increases the amount of available information about the resources, products and production processes and, hence, increases the level of IT quality and decreases the cost of measurement. The AMT question was adapted from Swamidass and Kotha (1998) and Kotha and Swamidass (2000).
ABC adoption was measured using a multiple-choice question regarding different experiences of ABC (Brierley, 2011). As shown in Table 1, ABC adoption was operationalised as a dichotomous variable where ABC adoption includes only ABC users, while ABC non-adoption encompasses all other groups of non-users (Ittner et al., 2002;Banker et al., 2008;Maiga et al., 2014).

Table 1. ABC adoption measurements ABC adoption definition
ABC non-adoption definition "Currently using ABC" "Intending to use ABC", "Currently investigating using ABC", "Intending to investigate using ABC", "Implemented ABC and subsequently decided to abandon it", "Investigated using ABC and decided to reject it", "Investigated using ABC and decided to reject it. However, the company established a system of activity analysis (AA) or cost driver analysis (CDA)", "Rejected ABC, but never considered using it" The optimality of CSD with respect to balancing the cost of error and measurement was indicated using two subjective outcome measures; namely, the respondents' perception of the usefulness and accuracy of cost information (USEFULNESS) and the extent of cost information usage in decision-making (USAGE) (e.g., Pizzini, 2006;Schoute, 2009). These two subjective outcomes were used due to the lack of objective outcome measures that could precisely identify the extent to which the CSD is optimal in relation to balancing the cost of error and measurement. USEFULNESS was measured using a five-point Likert-scale question adapted from Drury and Tayles (2000), Pizzini (2006) and Brierley (2008b). USAGE was measured using two five-point Likert-scale questions (Brierley et al., 2006;Schoute, 2009), the first of which concerned the extent of cost information usage in decision-making, while the second the importance of cost information in decisionmaking. USAGE was measured as a weighted-average using the two questions, with importance determining the weight (for an application, see Govindarajan, 1984;Govindarajan & Gupta, 1985;Chenhall & Langfield-Smith, 1998;Guenther & Heinicke, 2018). Table 2 shows the research constructs and their definitions. The level of competition the business-unit faces.

CostStructure-MANUFACTURING
The level of indirect costs represented by the percentage of indirect manufacturing costs to total manufacturing costs. PC The level of production complexity.

IT quality
The level of information technology quality.

ABC adoption
A dichotomous constructs that shows whether the business-unit adopts ABC or not.

USEFULNESS
The respondents' perception of the usefulness and accuracy of cost information.

USAGE
The extent of cost information usage in decision-making.
Before assessing the quality of the various latent or multi-indicator constructs, it is important to determine whether the latent construct is reflective or formative. There are many different criteria for differentiating between reflective and formative latent constructs (for example, see Diamantopoulos & Winklhofer, 2001;Jarvis et al., 2003;Bisbe et al., 2007;Petter et al., 2007;Hair et al., 2010;Hair et al., 2017). 5 For example, the indicators of the reflective latent constructs should be highly correlated with each other and interchangeable, whereas the opposite is true for those of the formative latent constructs (Hair et al., 2017). Accordingly, removing any indicator affects the nature of the formative latent constructs rather than the reflective ones. After evaluating the constructs based on established criteria to differentiate between reflective and formative latent constructs, COMP and USEFULNESS were classified as reflective latent constructs, whereas PC and IT quality were classified as formative latent constructs. The remaining constructs, namely CostStructure-MANUFACTURING, ABC adoption and USAGE, do not fall into either type of latent construct because they are single-indicator constructs.
The quality of the reflective latent constructs was evaluated using confirmatory factor analysis (Hurley et al., 1997), using partial least squares (PLS) as the estimation method (Tenenhaus & Hanafi, 2010). The evaluation of the constructs included the internal consistency reliability and convergent and discriminant validity (Henseler et al., 2009;Hair et al., 2017). Table 3 provide the results, which indicate that the reflective latent constructs meet the required quality standards.
The quality of the formative latent constructs was evaluated with respect to content validity and the existence of collinearity among the formative indicators (Diamantopoulos & Winklhofer, 2001;Petter et al., 2007;Hair et al., 2014;Hair et al., 2017). The content validity can be established through either adopting a construct from the prior literature or conducting the following steps: (1) determining the conceptual domain of the formative construct; (2) reviewing the literature extensively to identify a comprehensive list of indicators that covers the construct's conceptual domain; and (3) obtaining expert feedback in relation to the suitability of the chosen indicators (Diamantopoulos & Winklhofer, 2001;Petter et al., 2007;Hair et al., 2017). The content validity of IT quality was established through adapting a construct from the prior literature (Swamidass & Kotha, 1998;Kotha & Swamidass, 2000), while that of PC was established using the abovementioned steps. First, the conceptual domain of PC was determined as any production-related cause for increasing the complexity of the production environment and product cost distortion if TCS rather than ABC was used. Second, the literature review identified 11 possible PC dimensions which, after conducting exploratory interviews with production and cost accounting staff in eight Saudi manufacturing business-units, were reduced to six PC dimensions. Lastly, the indicators of the identified PC dimensions were selected. The second criterion for evaluating the quality of the formative latent constructs, i.e., the existence of collinearity issues, was met, since the variance inflation factor (VIF) of all of the formative indicators was below the threshold values of five (Hair et al., 2011;Hair et al., 2017). For the PC indicators, the VIF values ranged from 1.23 to 2.32, while those for IT quality ranged from 1.50 to 2.39. The remaining constructs, namely CostStructure-MANUFACTURING, ABC adoption and USAGE, are singleindicator constructs, which means that they have a relationship of 1 with their single indicator and, hence, the constructs and indicators have identical values (Hair et al., 2017).

Descriptive analysis results
In relation to ABC adoption, Table 4 displays the numbers and percentages of respondents for each of the various experiences of ABC. The results show that ABC is used by 19 (9.5%) business-units. In addition, the results indicate that ABC has the potential to be used by 63 (31.5%) business-units, as 34 (17%) businessunits are intending to use ABC, 17 (8.5%) are investigating its use and 12 (6%) are intending to investigate its use. Furthermore, the results suggest that the majority of business-units (n = 101, 50.5%) have never considered using ABC. Moreover, the results show that 17 (8.5%) business-units are unlikely to use the system, given that one (0.5%) business-unit had implemented it and subsequently decided to abandon it, while a further 16 (8%) had investigated its use and decided to reject it, with 11 of these 16 business-units deciding to confine its use to the pre-stages of ABC; namely, the activity analysis (AA) or cost driver analysis (CDA). Lastly, the results show that none (n = 0, 0 %) of the business-units rejected ABC without investigating its use.

Experience with ABC Number of cases Percentage
Currently using ABC 19 9.5 Intending to use ABC 34 17 Currently investigating using ABC 17 8.5 Intending to investigate using ABC 12 6 Implemented ABC and subsequently decided to abandon it 1 0.5 Investigated using ABC and decided to reject it 5 2.5 Investigated using ABC and decided to reject it. However, the company established a system of activity analysis (AA) or cost driver analysis (CDA) 11 5.5 Rejected ABC, but never considered using it 0 0 Never considered using ABC 101 50.5 Total 200 100 In relation to the influential factors on ABC adoption and the outcomes of ABC adoption, Table 5 provides the descriptive statistics. The results show that the level of COMP (mean = 3.86, median = 4.00) is high, suggesting that the Saudi market faces high levels of competition. However, the results show that the level of each of CostStructure-MANUFACTURING (mean = 23.00%, median = 21.00%), PC (mean = 2.72, median = 2.69) and IT quality (mean = 2.69, median = 2.75) is low, indicating that direct manufacturing costs represent the highest percentage of total manufacturing costs, that the overall Saudi production environment is not that complex and that IT in Saudi manufacturing business-units experiences low levels of quality, respectively. Regarding the outcomes, the results show that the level of each of USEFULNESS (mean = 4.09, median = 4.00) and USAGE (mean = 3.80, median = 3.90) is high, indicating a high degree of respondents' perception of the usefulness and accuracy of the implemented costing systems and extensive use of cost information in decision-making, respectively.  Table 6 shows the Pearson correlations of all of the constructs involved in this research. It indicates that none of the contingency factors or outcomes has a significant correlation with ABC adoption, which provides no initial support for the hypotheses. It also shows that USEFULNESS has a significant positive correlation with USAGE, indicating that both outcomes are indicative of the same concept; namely, the optimality of CSD.

Table 6. The Pearson correlations of all constructs
The influences on Activity-Based Costing adoption as an optimal costing system design: Evidence from Saudi Arabia

Hypotheses testing results
The hypotheses were tested using the residual analysis (RA) (Duncan & Moores, 1989;Dewar & Werbel, 1979;Ittner & Larcker, 2001;Ittner et al., 2002). The RA was deemed appropriate to test the matching form of fit's hypotheses, given that the CSD variable, i.e., ABC adoption, is measured using a dichotomous variable (Burkert et al., 2014). The RA was performed using logistic regression analysis that handles dichotomous dependent variables (Pallant, 2013;Field, 2013). The RA entails following two steps to test the matching form of fit's hypotheses. With the assumption that business-units, on average, operate with optimal CSD, the first step is to determine the optimal CSD levels as a function of the influential variables using regression analysis (Ittner et al., 2002;Chen & Jermias, 2014). The residuals of this regression model, either positive (indicating an over-fit) or negative (indicating an under-fit), represent the distance of each business-unit from the estimated optimal CSD level, i.e., the extent of the misfit (Pizzini, 2006). The first step provides information regarding the first element of the matching form of fit, i.e., the direction of the association between the influential factor and optimal CSD (see Section 1 and hypotheses 1 to 4). The second step of the RA is to regress the outcomes indicating CSD optimality on misfit, represented by two constructs; namely, the absolute values of positive residuals (i.e., over-fit) and the absolute values of negative residuals (i.e., under-fit) (Ittner et al., 2003;Van der Stede et al., 2006;Burkert et al., 2014). The second step furnishes information with respect to the second element of the matching form of fit (i.e., the magnitude and sign of the impact of the misfit on the outcomes) (see Section 1 and hypotheses 1 to 4).

Results of the first step of the RA
The results of first step of the RA involved estimating four regression models, one for each of the examined influential factors of COMP, CostStructure-MANUFACTURING, PC and IT quality. The matching form of fit requires using a separate regression model for each influential factor in order to identify the independent effect of each factor rather than the combined effect of all factors on ABC adoption as an optimal CSD, i.e., the effect from the standpoint of the system form of fit (see Endnote 2). To illustrate, estimating a separate model for each factor in the first step of the RA results in a misfit value for each, which can be subsequently used in the second step of the RA to identify the independent effect of the factor on ABC adoption as an optimal CSD. However, if a single regression model, that includes all of the influential factors, is estimated in the first step of the RA, a single misfit value will be obtained and then used in the second step of the RA to identify the combined effect of all of the factors on ABC adoption as an optimal CSD. Table 7 shows the results of the first step, which indicate that each of the four models was insignificant, as indicated by the significance values of the Chi-square goodness of fit test. These results suggest that none of the factors was able to differentiate between ABC adopters and non-adopters. The insignificance of the models is further illustrated by the low Cox and Snell R square and Nagelkerke R square values, which represent the amount of variation explained in the ABC adoption dichotomous variable. Furthermore, the insignificance of the models is further supported by the insignificant coefficients of the effects of COMP, CostStructure-MANUFACTURING, PC and IT quality on ABC adoption.

Results of the second step of the RA
As discussed in the previous section, none of the first-step regression models pertaining to the influential factors is significant. Hence, this suggests that the optimal CSD could not be identified and so performing the second step to identify the effect of the deviation from the optimal CSD is meaningless. In contrast to expectations, the overall results of the RA indicate that none of the examined factors affects ABC adoption as an optimal CSD from the perspective of the matching form of fit. Accordingly, hypotheses 1 to 4 are not supported.

Discussion and conclusion
Due to: (1) the importance of identifying the situations under which ABC is beneficial and, hence, represents the optimal CSD to make a correct decision about investing in ABC; and (2) the limitations of Ittner et al. (2002) regarding the number of the examined contingency factors and measurement of PC, this paper aimed to examine the key impacts on ABC adoption as an optimal CSD from the viewpoint of the matching form of fit, utilising a sufficiently comprehensive multidimensional PC measure. Specifically, instead of assessing the effect of a single cost of error-related factor, namely PC, this research examined the influence of three cost of error-related factors -i.e., the level of competition, indirect costs and PC -and the cost of measurement-related factor -i.e., IT quality -on ABC adoption as an optimal CSD from the standpoint of the matching form of fit. Prior ABC adoption research suggested that, besides PC, the remaining factors also affect ABC adoption as an optimal CSD (Cooper, 1988b;Kaplan & Cooper, 1998). Therefore, this examination will produce a more refined understanding of the situations under which ABC adoption is beneficial. In addition, instead of using a limited measure of PC, this research used a sufficiently comprehensive multidimensional PC measure that covers six PC dimensions identified from the prior literature and the findings of exploratory interviews with costing and production staff in eight Saudi manufacturing business-units. This helps to reveal the true effect of PC on ABC adoption as an optimal CSD from the perspective of the matching form of fit.
Using questionnaire data collected from Saudi manufacturing business-units, residual analysis (RA) results showed that none of the examined factors influences ABC adoption as an optimal CSD from the standpoint of the matching form of fit. As shown in Section 5.2.1, none of the logistic regression models was significant in the first step of the RA, which made it meaningless to proceed with the second step. These results were unexpected, given the solid theory behind the effect of the examined factors on ABC adoption as an optimal CSD from the perspective of the matching form of fit (Cooper, 1988b;Kaplan & Cooper, 1998;Stuart, 2013). Nevertheless, they are consistent with other studies that found no effect of the level of competition (e.g., Booth & Giacobbe, 1998;Chen et al., 2001;Chongruksut & Brooks, 2005;Cohen et al., 2005;Brierley, 2008a;, indirect costs (e.g., Nguyen & Brooks, 1997;Clarke et al., 1999;Malmi, 1999;Chen et al., 2001;Al-Mulhem, 2002;Khalid, 2005;Al-Omiri & Drury, 2007;Brierley, 2011;Nassar et al., 2009), PC (e.g., Clarke et al., 1999;Chen et al., 2001;Brown et al., 2004;Brierley, 2008a;Nassar et al., 2009) and IT quality (Al-Omiri & Drury, 2007) on ABC adoption in general. 6 There are many possible reasons for the unexpected findings regarding the impact of the examined factors on ABC adoption as an optimal CSD from the standpoint of the matching form of fit. Regarding the level of competition, it is possible that the COMP construct is limited, since it may need to be broadened by incorporating more dimensions of competition, such as product quality and variety (Khandwalla, 1972;Brierley, 2011). Another possible reason is that the effect of the level of competition on ABC adoption may be moderated by the ability of the business-unit to set prices. Specifically, the level of competition might affect ABC adoption when the business-unit is a price maker more than when it is a price taker. Price makers need more accurate cost information to set the price of their products (Drury & Tayles, 2006;Al-Omiri & Drury, 2013). However, price takers do not require as highly accurate costs because they use cost information for profitability analysis reasons to ensure that only profitable products are produced and provide attention-directing information about unprofitable products that need further assessment (ibid). With respect to the level of indirect costs, the CostStructure-MANUFACTURING measure might have been limited because it included both the unit-and-facility levels' indirect manufacturing costs, that do not affect ABC adoption, and batch-and product-level indirect manufacturing costs that do influence ABC adoption (Cooper, 1990;Drury & Tayles, 2005;Al-Omiri & Drury, 2007). Regarding PC, it is possible that the low level of task uncertainty in low PC environments, where mass production is applied, encourages the adoption of ABC more than in high PC environments, where job-shop production is utilised, that feature a high level of task uncertainty (Krumwiede, 1998). This can weaken the positive effect of PC on ABC adoption or even turn it into a negative one. Another possible reason for the insignificant effect of PC is that it is possible that The influences on Activity-Based Costing adoption as an optimal costing system design: Evidence from Saudi Arabia employees' resistance to highly complex costing systems, such as ABC, may emerge when the level of PC is high (e.g., Hiromoto, 1988;Merchant & Shields, 1993). A possible reason for this is that the extent of work in high PC environments is too high, making employees resistant to the additional work demands imposed by ABC's complexity. This may reduce or even eliminate any positive effect of PC on ABC adoption. Regarding IT quality, it is possible that the majority of business-units use high quality IT due to the decreased costs of these techniques in recent years (Al-Omiri & Drury, 2007). In addition, it is possible that the satisfaction with information provided by high-quality information systems makes investing resources in applying ABC less desirable (Krumwiede, 1998).
Besides the possible reasons related to each examined variable, there are other possible general reasons for our unexpected findings. The first possible general reason is the existence of organisational variables that execute a significant moderation role on the impact of the examined variables on ABC adoption as an optimal CSD from the perspective of the matching form of fit. For example, it is possible that the effect of the examined key variables on ABC adoption as an optimal CSD is only executed when the level of top management support for the costing system is high rather than low. The second possible general reason is the existence of context-related variables that change or eliminate, i.e., moderate, the effect of the examined key variables on ABC adoption as an optimal CSD from the viewpoint of the matching form of fit (Alawattage et al., 2007;Albu & Albu, 2012). For example, the national culture of Saudi Arabia, which is characterised by a wide power distance that appreciates hierarchies and a strong uncertainty avoidance that is careful and risk-averse to new ideas (Hofstede, 1984), may delay the adaption process of structure and systems, including the costing system, to the surrounding environment. Another example is the legal form of most businessunits in Saudi Arabia, which is family-owned (Joshi et al., 2011;Hanware, 2016;Schumpeter, 2016). Family-owned business-units are managed by family members who may base the investment decisions regarding the costing system on personal preferences rather than logic. For example, family members may reject investing in ABC, although the surrounding context requires it, because they wish to avoid extra costs that would reduce their short-term profits. In contrast, family members may encourage investment in ABC, even though the surrounding context does not require it, because they wish to show off to other family members in the same business-unit or other business-units in the same holding group.
As with all research, this research suffers from many limitations that should be accounted for in future research. First, the measurements of some of the constructs, such as COMP and CostStructure-MANUFACTURING, are limited and, hence, need to be improved by future research in the ways suggested above. Second, although the "Rejected ABC, but never considered using it" ABC experience is included in the definition of ABC non-adoption (see Table 1), in this research, the number of respondents belonging to this ABC experience was 0 (see Section 5.1), meaning that this ABC experience is not actually represented in the utilised definition of ABC non-adoption. Thus, the results should be interpreted with caution and future research should attempt to collect sufficient data from respondents representing all of the various experiences of ABC adoption. Third, this research assumed that the subjective outcome measures of USEFULNESS and USAGE are able to reflect the extent to which the CSD balances the cost of measurement and error, i.e., the optimality of CSD. Given the unexpected results, this assumption may be unrealistic and, therefore, future research should attempt to develop objective outcome measures for CSD optimality that can indicate the extent to which the CSD balances the cost of measurement and error. Given the difficulty of developing and applying such objective outcome measures when conducting empirical research, non-financial and financial performance seem to be reasonable outcome measures for CSD optimality. These outcome measures are also supported by the fact that the ultimate objective of having optimal CSD is to contribute positively towards overall business-unit performance. Nevertheless, the effect of known variables on performance should be controlled in order to identify the true effect of the fit between ABC adoption and its influential factors on the outcomes representing the optimality of ABC. Fourth, this research assumed a direct effect of the examined variables -e.g., PC and IT quality -on ABC adoption as an optimal CSD from the viewpoint of the matching form of fit. As mentioned above, this direct effect can be moderated by different organisational variables, such as top management support and organisational culture (Baird et al., 2004;Fullerton et al., 2013;Zhang et al., 2015), and context-related variables, such as national culture (Brewer, 1998). Accordingly, future research should examine the moderation role of these variables on the relationship between the examined variables -e.g., PC and IT quality -and ABC adoption as an optimal CSD from the standpoint of the matching form of fit. This would best be investigated by adopting the cross-sectional field studies strategy, which involves limited-depth studies of a reasonable number of cases, as it allows the uncovering of new insights about either hypothesised or new relationships (Lillis & Mundy, 2005). Despite these limitations, this research has contributed to the ABC adoption research by examining the situations under which ABC is beneficial and, therefore, represents the optimal CSD through testing the effect of the key cost of error-related and the cost of measurement-related factors on ABC adoption as an optimal CSD from the perspective of the matching form of fit, using a sufficiently comprehensive multidimensional PC measure.

Production complexity (PC)
Which of the following best describes the percentage of the customised and standardised products produced by your business unit? (Please circle the appropriate number.) 1. At least 95% of products are standardised.
2. Approximately 75% of products are standardised, and 25% of products are customised. 3. Approximately 50% of products are standardised and 50% of products are customised. 4. Approximately 75% of products are customised, and 25% of products are standardised. 5. At least 95% of products are customised.
To what extent do each of the following statements apply or do not apply to your business unit's products and production environment? ( To a very great extent process, product and logistic technologies (e.g., Electronic Data Interchange (EDI)).

The extent of cost information usage in decision-making (USAGE)
To what extent, if at all, does your business unit use product cost information in the following decisions? ( The explanation of the forms of fit is drawn from various literature, yet it will be discussed in the context of CSD because it is the focus of this paper. 2 Besides the selection, moderation and matching forms of fit, there are the mediation and system forms of fit. Modelling the relationship between the contingency factors and ABC adoption as an optimal CSD from the viewpoint of the mediation form of fit entails examining the indirect effect of the contingency factor on an outcome through ABC adoption (Chenhall, 2003;Gerdin & Greve, 2004;Gerdin, 2005a;2005b;Hartmann, 2005;Burkert et al., 2014), while modelling it from the standpoint of the system form of fit involves examining the combined effect of multiple fits between multiple contingency factors and ABC adoption on an outcome (Donaldson, 2001). To the author's knowledge, the utilisation of the mediation and system forms of fit in ABC adoption research is limited. 3 Relevant extracts from the questionnaire are included in the appendix. 4 CostStructure-MANUFACTURING was transformed using square root transformation to reduce the impact of univariate outliers (Tabachnick & Fidell, 2007;Hair et al., 2010). 5 The explanation of formative constructs assumes composite-formative constructs rather than causal-formative constructs, as the former is more in line with social science research because it considers measurements to represent an approximation rather than a complete reflection of the theoretical concept (Hair et al., 2017). 6 These studies have used the selection form of fit.