• Tidak ada hasil yang ditemukan

Directory UMM :Data Elmu:jurnal:A:Accounting, Organizations and Society:Vol24.Issue7.Oct1999:

N/A
N/A
Protected

Academic year: 2017

Membagikan "Directory UMM :Data Elmu:jurnal:A:Accounting, Organizations and Society:Vol24.Issue7.Oct1999:"

Copied!
35
0
0

Teks penuh

(1)

The impact of contextual and process factors on the

evaluation of activity-based costing systems

$

Shannon W. Anderson

a,

*, S. Mark Young

b

aAccounting Department, University of Michigan Business School, Ann Arbor, MI 48109, USA bLeventhal School of Accounting, University of Southern California, Los Angeles, CA 90089, USA

Abstract

This paper investigates associations between evaluations of activity based costing (ABC) systems, contextual factors, and factors related to the ABC implementation process using interview and survey data from 21 ®eld research sites of two ®rms. Structural equation modeling is used to investigate the ®t of a model of organizational change with the data. The results support the proposed model; however, the signi®cance of speci®c factors is sensitive to the evaluation cri-terion. The model is stable across ®rms and respondents, but is sensitive to the maturity of the ABC system.#1999 Elsevier Science Ltd. All rights reserved.

1. Introduction

Early proponents of ABC claimed superiority over traditional costing methods that stemmed from employing causally related ``cost drivers'' to assign common costs to business activities, pro-ducts and services (Cooper, 1988; Cooper & Kaplan, 1988). Later studies argued that a judi-ciously designed ABC system provides e€ective behavioral control (Cooper & Kaplan, 1991; Cooper & Turney, 1990; Foster & Gupta, 1990.

Evidence of ABC implementation failures1 has

caused researchers to suggest that achieving either objective depends critically on organizational and technical factors (Anderson, 1995; Malmi, 1997) and recent empirical evidence supports this view

(Chenhall & Lang®eld-Smith, 1998; Foster & Swenson, 1997; Gosselin, 1997; Innes & Mitchell, 1995; Krumwiede, 1998; McGowan & Klammer, 1997; Shields, 1995). This paper combines the results of previous studies with theory on organi-zational change to propose a structural model of the relation between evaluations of ABC systems, contextual factors and factors related to the ABC implementation process.

We evaluate the model's descriptive validity using survey data from managers and system developers associated with 21 implementation projects in two automobile manufacturers. Struc-tural equation modeling (SEM) is used to investi-gate the in¯uence of the contextual environment and the implementation process on evaluations of

0361-3682/99/$ - see front matter#1999 Elsevier Science Ltd. All rights reserved. P I I : S 0 3 6 1 - 3 6 8 2 ( 9 9 ) 0 0 0 1 8 - 5

www.elsevier.com/locate/aos

$ The authors are not permitted to redistribute the data of

this study without permission of the participating ®rms. * Corresponding author. Tel.: 734-647-3308; fax: +1-734-936-8716.

E-mail address:swanders@umich.edu (S.W. Anderson)

(2)

the ABC system and the in¯uence of the con-textual environment on the ABC implementation process. The data are consistent with the proposed structural model. After establishing a plausible model structure, we investigate its applicability to di€erent evaluation criteria and its stability across di€erent sub-groups in our sample. Thus, the research contributes a uni®ed investigation of

three aspects of ABC implementation: model

structure; variable de®nition and measurement; and,

model stability.

We test astructural relation between contextual

variables, process variables and ABC system eva-luations that is hypothesized in ``process theories'' of ABC implementation (Anderson, 1995; Argyris & Kaplan, 1994; Kaplan, 1990; Shields & Young, 1989). By separating the in¯uence of the con-textual environment on the ABC implementation project from its in¯uence on the evaluation of the ABC system, we extend previous studies that document correlation between ABC project out-comes and contextual and process factors. In a case study of an ABC project that did not sur-vive, Malmi (1997) posits that implementation failures are related more to exogenous contextual factors than to the process of implementation Ð that even good implementation processes fail on barren ground. Our results support this observa-tion and point to speci®c exogenous factors that make ABC less suitable and that are unlikely to be remedied by improving the implementation process.

The study contributes to the area of variable

de®nition and measurement through ®eld research and the use of multiple data collection methods. Multiple modes of data collection (e.g. surveys and personal interviews) provide the opportunity to address the question: What is meant by ``suc-cess'' in ABC implementation? A danger of asking managers to rate ABC implementation success without specifying the de®nition of success is fail-ure to detect cases in which individuals hold dif-ferent views on the de®nition of success but share views on attainment of a particular dimension of success. In light of evidence that success in ABC implementation is multi-dimensional (Cooper, Kaplan, Maisel, Morrissey & Oehm, 1992) with each dimension having somewhat di€erent

corre-lates (Foster & Swenson, 1997), it is appropriate to ask what criteria respondents use in evaluating ABC. Content analysis of interviews conducted with survey respondents reveals two dominant views of what de®nes an e€ective ABC system: whether ABC data are used in product cost reduction or process improvement; and, whether ABC data are more accurate than data from the traditional cost system. Further investigation indicates that the respondent's job is a predictor of which view is held. Previous studies explore deter-minants of di€erent evaluation measures (Foster & Swenson, 1997) and document respondent-e€ects on evaluation levels (McGowan & Klammer, 1997). This study investigates whether di€erent evaluation measures correspond to di€erent views on appro-priate evaluation criteria. Then, using survey data we estimate simultaneously the contextual and process determinants of respondents' evaluations of the ABC systems according to the operative criteria. Simultaneous estimation methods allow us to consider whether the factors that in¯uence di€erent evaluation measures are common Ð (suggesting that all criteria may be achieved) Ð or mutually exclusive Ð (suggesting that success along one dimension is achieved at the expense of another).

A ®nal contribution of this research is investi-gation of thestabilityof the relation between eva-luations of ABC implementation and factors related to context and the implementation process. The research design and sampling plan permit investigation of three potential sources of model instability: company e€ects, respondent e€ects and e€ects of ABC system maturity. Company e€ects are examined in the spirit of sensitivity analysis, to explore the degree to which the results generalize. We can not address the degree to which results generalize to di€erent industry settings; however, we ®nd few di€erences between the ®rms and reject the hypothesis that a ®rm-speci®c model is warranted. The exploration of respondent e€ects, speci®cally of di€erences between man-agers and ABC system developers, is a unique contribution of this study.2Previous research relies

(3)

responsibility for ABC implementation, or as a result of the degree to which ABC threatens or rein-forces their professional standing. The results indi-cate di€erences in what determines evaluations of the ABC system between managers and ABC system developers; however, statistical tests reject the hypothesis that a model that distinguishes between respondent types improves model ®t. Exploration of e€ects of ABC system maturity on model stability continues the investigation of ``stages'' of ABC implementation in Anderson (1995) and Krum-wiede (1998). The results indicate signi®cant di€er-ences in determinants of respondents' evaluation of ABC as a function of ABC system maturity. Thus, the maturity-speci®c model ®ts the data better than a model that omits this factor.

Section 2 reviews the literature and develops the

research questions. Section 3 describes the

research sites and data collection methods.

Variable measures and descriptive statistics are presented in Section 4. Evidence on the relation between ABC system evaluations and contextual and process variables is presented in Section 5. The stability of the model between di€erent companies, respondents and ABC systems of di€ering maturity is investigated in Section 6. Section 7 summarizes the contributions of the research and discusses avenues for future investi-gation.

2. Associations between context, process and evaluations of ABC systems

2.1. Summary of prior research

Practitioner accounts of ABC projects and case study research on determinants of project outcomes have long associated technical and behavioral factors with ABC implementation success (Beaujon & Singhal, 1990; Foster & Gupta, 1990; Cooper,

1990; Cooper et al., 1992; Drumheller, 1993; Eiler & Campi, 1990; Foster & Gupta, 1990; Haedicke & Feil, 1991; Jones, 1991; Kleinsorge & Tanner, 1991; Richards, 1987; Shields & Young, 1989; Stokes & Lawrimore, 1989). Anderson (1995) sur-veys the literature on ABC and information

tech-nology implementation and compiles ®ve

categories of 22 variables that are implicated in ABC project outcomes. More recent empirical

studies provide evidence on the correlation

between these factors and ABC implementation e€ectiveness and introduce ®ve new variables (Chenhall & Lang®eld-Smith, 1998; Foster & Swenson, 1997; Gosselin, 1997; Innes & Mitchell, 1995; Krumwiede, 1998; Malmi, 1997; McGowan & Klammer, 1997; Shields, 1995). Table 1 updates Anderson's (1995) list of variables hypothesized to in¯uence ABC system evaluations (column 1) and summarizes the statistical relations documented in these studies (column 3).

A diculty in specifying the hypothesized e€ect of each factor on evaluations of ABC systems is that each study is di€erent in ways that draw into question comparability of results. For example, Anderson (1995), Gosselin (1997) and Krumwiede (1998) distinguish di€erent stages of ABC imple-mentation and ®nd evidence that di€erent factors in¯uence ``success'' at each stage. Foster and Swenson (1997), McGowan and Klammer (1997) and Malmi (1997) study speci®c ABC imple-mentation projects, while other studies gather data at the ®rm level. Moreover, Foster and Swenson document somewhat di€erent correlates for four di€erent measures of ABC implementation suc-cess. Finally, McGowan and Klammer collect data from di€erent informants at an ABC imple-mentation site and ®nd evidence of a shift in the mean level of evaluation that is associated with respondents' involvement with the ABC project. Other studies rely almost exclusively on informants from accounting functions (of the ®rm or the ABC site). In sum, comparing empirical studies of the determinants of ABC implementation e€ectiveness requires strong assumptions about invariance of these relations across time, respondent groups, measures of e€ectiveness and units of analysis.

In spite of caveats concerning comparability of prior studies, the research ®ndings are remarkably

(4)

Table 1

Candidate variables for analysis of determinants of evaluations of ABC implementation: contextual factors and process factors identi®ed in the research literature

Candidate variables Literature

sources(s)a

Hypothesized e€ect on evaluation of ABCb

Contextual factors Process factors Research design or variable mappingd

Individual factors

Organizational factors

ABC project management

Team processc

Individual characteristics

Disposed to change A,C,E,F +,+,+,+ X ± ± X CHANGE

Production process knowledge A + X ± ± X interviews

Role involvement A,E +,+, X ± ± X COMMIT,

VALUES, AND M vs D

Individual received ABC training G 0 X ± ± ± M vs D

Organizational factors

Centralization A,D +,+ ± X ± X C1 vs C2

Functional specialization A,B ÿ,0 ± X ± X C1 vs C2

Formalization/job standardization D + ± X ± X C1 vs C2

Vertical di€erentiation D + ± X ± C1 vs C2

Formal support in accounting function B,C 0,+ X X C1 vs C2

Support A,B,C,E,F,G,H +,+,+,+,+,0,+ ± ± X X

Top management support ± ± ± ± ± MSUPPORT,

Local management support ± ± ± ± ± MINVOLVE,

Local union support ± ± ± ± ± USUPPORT

Internal communications A + ± X ± X C1 vs C2

Extrinsic reward systems A,B,E,G +,+,+,+ X X REWARD

ABC training investments A,B,E,G,H +,+,+,+,+ ± ± X X Team

Technological factors

Complexity for users A,C ±,± ± ± ± X Team

Compatibility with existing systems A,B,C,I +,0,+,+ ± X ± X N/A

Relative improvements over existing system (accuracy and timeliness)

A,C,H +,+,0 ± X ± X INFOQUAL

Relevance to managers' decisions and compatibility with ®rm strategy

A,B,C,E,H,I +,+,+,+,+ ± X ± X IMPCOST

Task characteristics

Uncertainity/lack of goal clarity A,B,E,H ÿ,0,ÿ,0 ± ± ± X Team

Variety A + ± ± X Team

Worker autonomy A + ± ± ± X Team

Worker responsibility/personal risk A ÿ ± ± ± X Team

Resource adequacy B,C,E +,+,+ ± ± X ± RESOURCES

Availability of ABC software B,C 0,+ ± ± X ± N/A

S.W.

Anderson

,

S.M.

Young

/

Accountin

g,

Organizations

and

Society

24

(1999)

(5)

Heterogeneity of demands A,C,D,H,I +,+,+,+,0 ± X ± ± TURB

Competition A,C,H +,+,+ ± X ± ± COMPETE

Environmental uncertainity A,C,D,F,I, ‹,ÿ,+,+,+ ± X ± ±

Likelihood of layo€s ± ± ± ± ± LAYOFF

Growth opportunities ± ± ± ± ± NOGROW

Labor relations ± ± ± ± ± LABOR

Importance of site to company ± ± ± ± ± IMPPLT

External communications/external experts

A,B, +,0 ± X ± X C1 vs C2 and

Team

a Source legend: A: Anderson (1995). Anderson constructs a list of factors previously implicated in ABC implementation outcomes from the research literature and

from practitioner accounts of implementations and provides evidence on how these factors in¯uence one ®rm's adoption of ABC. Recent empirical research: B: Shields (1995); C: Innes and Mitchell (1995); D: Gosselin (1997); E: Foster and Swenson (1997); F: Malmi (1997); G: McGowan and Klammer (1997); H: Krumwiede (1998); I: Chenhall and Lang®eld-Smith (1998)

b Foster and Swenson (1997), McGowan and Klammer (1997) and Malmi (1997) study speci®c ABC implementation sites rather than ®rm-level implementation.

Anderson (1995), Krumwiede (1998) and Gosselin (1997) distinguish correlates of di€erent stages of implementation. Because we study plants that implemented ABC after the formal corporate adoption of ABC, we focus on factors that stage-speci®c studies ®nd to in¯uence the ``adoption'' and ``adaptation'' stages. For studies that do not distinguish stage-speci®c e€ects or that employ multiple measures of ABC project outcomes, we report the sign of the correlation between of the variable and overall evaluations of the ABC project.

c Variables associated with intrateam processes of the ABC design team (e.g. group cohesion, team leadership) are not examined in this study.

d Variable treatment legend:C1 vs C2indicates a variable that di€ers between but not within companies.M vs Dindicates a variable that di€ers between managers and

ABC developers but not within each group. The designationN/Aindicates that the variable is not examined in this study because there is neither within ®rm nor between ®rm variation (e.g. all sites use a common software package and no site integrated ABC with other information systems prior to the end of the study).Teamindicates a variable that re¯ects intrateam processes of the ABC development team. The e€ects of intrateam processes on ABC system evaluations are not considered in this paper (see Anderson et al., 1999). Words in UPPER CASE are variable names of factors that are examined in this paper. ``Interviews'' designates an association that is examined in this paper using interview data but not survey data.

S.W.

Anderson

,

S.M.

Young

/

Accountin

g,

Organizations

and

Society

24

(1999)

525±559

(6)

consistent. Although some studies fail to docu-ment a statistically signi®cant e€ect (denoted ``0''), in only one case (environmental uncertainty) do studies ®nd signi®cant but con¯icting e€ects. In this case, Innes and Mitchell (1995) investigate uncertainty associated with the likelihood of ABC data threatening respondents' employment (e.g. cost reduction through layo€s), while other studies focus on the potential for ABC data mitigating informational uncertainties and promoting better decision-making. Thus, this is a case of di€erent de®nitions of uncertainty rather than substantive disagreement between the studies' results. In sum-mary, there is widespread agreement in the aca-demic literature on broad correlates of ABC implementation e€ectiveness. Sources of instability in the relation are less well understood. These observations are the departure point for this study, which tests a structural framework among corre-lates of ABC system e€ectiveness and examines the stability of the structural model.

2.2. The proposed structural framework

Although the empirical literature has progressed from case studies and anecdotes to systematic evidence on correlates of ABC project outcomes,

there is little correspondence between empirical studies and studies that propose process theories of ABC implementation (Anderson, 1995; Argyris

& Kaplan, 1994; Kaplan, 1990; Shields &

Young,1989). Process theories hypothesize that project outcomes depend critically on the imple-mentation process and on contextual factors related to the external environment. Process theories of ABC implementation are similar to Rogers' (1962, 1983) model of organizational change and innova-tion. In Rogers' model, managers' consideration of an innovation is motivated or constrained by cir-cumstances in the ®rm's external and internal envir-onment and by characteristics of the individual evaluating the innovation Ð what we refer to col-lectively as contextual factors. Subsequent

evalua-tions of the innovation are in¯uenced by

comparison between the innovation, the status quo, and alternative innovations, and by factors related

to the innovation experience Ð what we term

pro-cessfactors. Contextual factors in¯uence the process of implementation and the evaluation of the result-ing ABC system. Process factors only in¯uence eva-luations of the ABC system. This ``structure'' that process theories hypothesize is depicted by arrows that connect the boxes in Fig. 1 (ignore for now relations among process factors).

(7)

Process theories distinguish from the list of cor-relates in Table 1 those that relate to the context in which the evaluation of ABC is conducted Ð including factors related to organizational context and factors related to the individual asked to ren-der an evaluation Ð and those that relate to the process of implementing ABC. This distinction is re¯ected in columns 4±7 of Table 1. The categories are not perfectly separableÐfor example, organi-zational norms with respect to functional speciali-zation may be mirrored in how the ABC project is managed (e.g. as an accounting project or as a

multi-disciplinary project). Nonetheless, the

separation is reasonably straightforward. Con-textual factors include those related to the organi-zation (column 4) and those related to the individual asked to evaluate the ABC system (col-umn 5). Implementation process factors are also divided into two types: those that re¯ect interac-tions between the ABC project team and the organization (column 6), and those that re¯ect the internal functioning of the ABC project team (column 7, e.g. communications and goal clarity among team members). The organizational litera-ture discusses the relation between project outcomes and internal team processes (e.g. Bettenhausen, 1991; Cohen, 1993; Cohen & Bailey, 1997). How-ever, because these research questions are not clo-sely related to the accounting literature that we extend, are motivated by di€erent organizational theories and demand a di€erent unit of analysis, we consider the variables in column 7 outside the

scope of this paper.3 We focus instead on the

structure and stability of relationships between the variables in columns 4±6 and evaluations of ABC systems.

2.3. Research questions

Two sets of related research questions are con-sidered. The ®rst research questions are related to the objective of testing the descriptive validity of process theories of ABC implementation. Speci®cally, we investigate whether there are associations between:

1.1 evaluations of ABC systems and contextual variables that represent individual and organizational circumstances;

1.2 evaluations of ABC systems and the imple-mentation process; and,

1.3 the ABC implementation process and con-textual variables that represent individual and organizational circumstances.

These research questions are depicted in Fig. 1 as arrows between three groups of variables: con-textual and process factors and implementation outcomes. Questions 1.1 and 1.2 have been con-sidered in previous empirical research on

corre-lates of ABC implementation e€ectiveness.

Question 1.3, which links process theories of ABC implementation and theories of organization change to the existing empirical literature, is a unique contribution of this study.

A second set of research questions is motivated by the earlier observation that the third column of Table 1 can not be constructed without assuming very strong forms of model stability. We explore the validity of these assumptions in two ways. First, we use ®eld research to explore criteria that our respondents employ in evaluating ABC sys-tems. We then investigate how the model of ABC evaluation is a€ected by the evaluation criterion. Second, our research design and sample selection permits us to test for forms of model stability suggested by previous studies. Thus, we explore:

2.1 what criteria are used by managers and ABC system developers to evaluate ABC implementation e€ectiveness;

2.2 do the associations examined in research questions 1.1±1.3 di€er for di€erent evalua-tion criteria; and,

2.3 are the associations examined in research questions 1.1±1.3 stable across ®rms, across respondents with di€ering involvement in the ABC project, and across sites with ABC systems that are of di€erent maturity?

In summary, this paper attempts to link empirical studies of correlates of ABC implementation with process theories of ABC implementation (questions 1.1±1.3) and provides evidence on model stability across a number of dimensions (questions 2.1±2.3).

(8)

3. Research design

The research design is de®ned by three choices: selection of ®rms and speci®c ABC implementation projects for study and selection of quali®ed respon-dents for each project. Evaluating process models of ABC implementation necessitates an understanding of the ®rm's approach to implementing ABC and access to several ABC implementation projects at each ®rm. A research design that employs ®eld-based research o€ers this level of understanding and access; however, costs of ®eld research and the time to develop relationships with corporate partners limit the sample size. We study two automobile manufacturing ®rms, both with mature corporate ABC programs and many ABC implementation sites. In the sections that follow, we discuss the ®rms' programs for implementing ABC and our approach for selecting sites for study.

3.1. ABC implementation in two automobile manufacturers

The US auto industry is an appropriate setting for studying the applicability of models of organi-zational change to ABC implementation because ®rm-wide implementation demands developing ABC models for many of remote locations and because in 1995 ABC was a mature technology for two of the ®rms. Previous research that examines ABC adoption by the ®rm ®nds that large organi-zations with hierarchical structures, centralized decision-making and signi®cant job standardiza-tion are more likely to adopt ABC (Gosselin, 1997). Moreover, ABC is attractive to ®rms in

competitive environments that demand

con-tinuous cost reduction (Chenhall & Lang®eld-Smith, 1998), particularly when existing cost sys-tems fail to support decisions related to cost reduction. In the early 1980s, the then ``Big 3'' US auto manufacturers ®t this characterization. Con-sequently, it is not surprising that when ABC became visible in the practitioner literature (e.g. Cooper, 1988), at least two ®rms began experi-menting with it and adopted it by 1991. Anderson (1995) investigates the correspondence of a model of innovation with one company's eight year experience of moving from ``problem awareness'', to

experimentation and evaluation of alternative cost

systems, and ®nally, to adoption of ABC.4 This

paper continues the exploration, adding a second ®rm that adopted ABC shortly after the ®rst ®rm, and shifting the unit of analysis to individuals involved in 21 ABC implementation projects.

After the ®rms adopted ABC as a corporate initiative, corporate ABC groups were charged with supporting implementation at all manufacturing sites. Nonmanufacturing sites were to follow, as the corporate group gained implementation expertise. The process of implementing ABC at a site di€ers somewhat between the two companies. Although both companies espouse a theory of activity based management Ð in which process or activity costs are as important as product costs Ð Company 2 has made this more central to the objective of ABC implementation than has Company 1. Neither com-pany used ABC in budgeting or performance eva-luation at the time of our study, although both were experimenting with these possibilities. Company 2 used an outside consulting ®rm to oversee site-level ABC implementation projects and augmented site teams with corporate ABC group members. Com-pany 1 relied solely on employees at the site, although divisional liaisons were available to assist the team. Assistance most often took the form of computer software technical support.

In spite of these di€erences, there are striking similarities in the ®rms' approach to ABC imple-mentation. Both ®rms had a corporate mandate to implement ABC that allowed plants to implement ABC within a three to ®ve year window of local management's choosing. Both ®rms used what Lindquist and Mauriel (1989) term a ``depth strategy'' of implementing ABC fully in a few sites and adding sites over time, rather than a ``breadth strategy'' of simultaneously implementing a more limited version of ABC across all sites. Neither ®rm used ABC data for inventory valuation or in performance evaluations of managers (e.g. calcu-lating ABC-based cost variances) at the close of our study. Indeed both ®rms were advised by their (di€erent) external auditors to delay using ABC

(9)

for inventory valuation until all sites installed ABC systems. Both ®rms standardized development and maintenance of ABC models over time and used the same PC-based software. Both ®rms managed ABC implementation from a corporate group that reports through the ®nance function, but attempted to gain support from the operations function. Both ®rms asked that the local ABC team be ``multi-dis-ciplinary''; however, most teams included at least one accountant or budget analyst and the team always reported to the plant controller. Finally, both ®rms introduce the ABC project with an executive awareness session for functional managers at the site, training for ABC developers, and sub-sequent management reviews of project milestones.

3.2. Selection of ABC implementation sites and critical informants

Anderson (1995) observed that, after the ®rm's decision to adopt ABC, continued corporate sup-port of the ABC initiative did not depend upon success of speci®c ABC projects. Rather, following a period of controlled experimentation when pro-ject success was critical, the corporate initiative acquired a life of its own. This pattern is con-sistent with research on innovation adoption and illustrates a critical distinction between individual and organizational adoption of innovations:

Because organizations are complex hierarchical systems, contradictory part-whole relations are often produced when system-wide innovations are introduced. An organization-wide innova-tion of change developed by one organizainnova-tional unit often represents an externally imposed mandate to adopt the innovation to other, often lower-level, organizational units. Thus. . .

top management . . . may express euphoria

about the innovation it developed for the entire organization, while frustrations and opposi-tions to that same innovation are expressed by the a€ected organizational units (Van de Ven, 1993, p. 285).

Studies of ABC implementation that rely on a single senior manager's evaluation of a ®rm's ABC system may re¯ect an average assessment of

widely di€ering project outcomes or biases of top management. To address these concerns, we gather data from several ABC projects within each ®rm and use local informants for each project.

ABC sites are selected using three criteria. First, we select sites that initiated ABC developmentafter

the corporate decision to implement ABC but dur-ing di€erent periods of the ®rm's ABC implementa-tion history. We exclude experimental or prototype projects to avoid confounding routine implementa-tions with those that were linked to the organiza-tional decision to adopt ABC.5Inclusion of projects

from di€erent periods permits investigation of tem-poral in¯uences on ABC system evaluations.

A second factor in site selection is an attempt to include all auto manufacturing production pro-cesses. Approximately half of the sites from each ®rm produce components that are unlikely to be outsourced (termed ``core'' components), including: major metal stampings, foundry castings, engines, and transmissions. The remaining sites produce peripheral components and face external competi-tion. Selecting core and peripheral components sites maximizes variation in the external environment factors of Table 1 (subject to the limitation that the study occurs within a single industry), allowing investigation of the relation between these con-textual factors and ABC system evaluations. In addition to traditional manufacturing sites, the ser-vice parts distribution groups of both ®rms are included. We were unable to match a contiguous stamping and assembly plant that was included for one ®rm; thus, the sample includes eleven sites from one ®rm and ten sites from the other ®rm.

A ®nal factor in selecting sites is the perceived success of the ABC project. We wish to study sites that represent the full range of implementation outcomes. One ®rm adopted ABC in 1991 and consequently had fewer ABC projects from which to choose. Meeting the ®rst two criteria for site selection virtually exhausted the population of ABC sites; thus, it is unlikely that we were directed toward exceptional projects. The second ®rm adopted ABC in 1989 and had over 150 ABC models at the inception of our study. To guard

(10)

against the ®rm directing us to relatively successful projects, we examined an independent assessment of ABC project success. In a related study we conducted a survey of eight division-level ABC managers; assessments of 50 ABC implementation projects. Seven of the proposed sites were covered by the survey. Responses to questions related to ABC project outcomes indicate that the seven sites span the full range of evaluations.

Selection of quali®ed informants about local ABC implementation projects is guided by the literature on organizational change and evidence on ``respon-dent e€ects'' in ABC system evaluation. McGowan and Klammer's (1997) ®nding that managers charged with using the ABC system hold di€erent opinions about the system compared to those who develop the ABC data suggests two groups of respondents. The organizational literature goes fur-ther, arguing that organizational change is a process of changing the beliefs and behaviors of individuals (Marcus & Weber, 1989). Although key individuals may play a disproportionate role in convincing oth-ers to adopt an innovation, it is rare that a single individual can unilaterally adopt innovations on behalf of an organization. Moreover, the beliefs and behaviors of individuals toward a particular inno-vation are shaped by their unique, individual cir-cumstances within the organization.6We attempt to

obtain full participation from two populations of critical informants: ABC system developers and the site's management team (e.g. the plant manager and functional managers who report to the plant man-ager), and measure ``individual characteristics'' (Table 1) that are hypothesized to in¯uence these

respondents evaluations of ABC7

3.3. Data collection

Each research site was visited for two days between March and November of 1995. Surveys were mailed to the site one week before the visit and respondents brought the completed survey to the interview, or, if they were not available for an interview, mailed it to the researchers. Similar surveys were administered to ABC developers and managers. Survey questions about individual motivation, organizational circumstances and the ABC project were developed from established scales in the organizational and information sys-tems literatures (e.g. Davis, 1989; Davis, Bagozzi & Warshaw, 1989; Jaworski & Young, 1992; McLennan, 1989; Robinson, Shaver & Wrights-man, 1991; Seashore, Lawler, Mirvis & Cammann, 1983; Van de Ven & Ferry, 1980). At times wording changes were necessary to ®t the organizational context and the speci®cs of ABC system develop-ment. Questions dealing with the implementation process were based on information gathered in the ®rm-level study of ABC implementation and, where applicable, by using question structures that are similar to related scales (e.g. for involvement in sys-tem design, scales for ``task involvement'' were used). A survey pre-test administered to ten corpo-rate and divisional ABC employees at each ®rm, all with experience implementing ABC, was used to revise the survey questions. We received 265 surveys Ð 176 management surveys and 89 ABC developer surveys (Table 2).8

The pro®les of the respondents do not di€er visibly between ®rms. Company 2 uses slightly less experienced people as ABC system developers

6 See Pedhazur and Schmelkin (1991) for a discussion of respondent e€ects associated with factors such as age, gender, and educational attainment, as well as a variety of personality traits. Weick (1995) describes individual processes of interpret-ing, or ``making sense'' of the environment.

7 Although some have argued that e€ective use of ABC requires involvement of employees at the lowest level of the organization, the ®rms of this study have not disseminated ABC data to this audience. Thus we have no basis for investi-gating the associations between workers' attitudes and ABC system outcomes and were discouraged from doing so by both ®rms. Production workers are occasionally included in the sur-vey population as ABC system developers when they served on development teams.

(11)

than does Company 1 Ð evidenced by fewer years on average with the company and in the depart-ment from which they joined the ABC project. There are no di€erences between the ®rms in edu-cational attainment or ABC training Ð developers typically receive 30 hours of training in ABC and managers typically receive no ABC training. Managers have longer tenure with the ®rms than do ABC developers; however, by virtue of pro-motions and company transfers, tenure at the site is not appreciably di€erent between managers and ABC developers.

Interviews were conducted with 236 of the 265 survey respondents. Although most interviewees returned their surveys at the start of the interview, they did not provide their names in the body of the survey and speci®c survey responses were not discussed during the interview. Interviews fol-lowed a loose structure aimed at supplementing the survey data. On average 10 interviews were conducted at each plant, ranging in duration from approximately 30 minutes, for functional man-agers with little awareness of the ABC project, to 4 hours for system developers with years of system design and maintenance experience. With two exceptions, all interviews were taped and transcribed for content analysis.

4. Measurement of variables and process structure

4.1. Contextual variables

A limitation of having only two ®rms is that ``organizational factors'' that re¯ect ®rm character-istics (e.g. centralization, functional specialization)

are confounded (e.g. Table 1, column 8, ``C1 vs

C2''). We investigate the stability of our model

across ®rms; however, the research design does not permit us to distinguish among alternative explanations for ®rm e€ects. Studies that use a large number of ®rms, each providing data on several implementation sites are needed to investi-gate the e€ect of ®rm characteristics on imple-mentation project outcomes. This study focuses on organizational factors that are more ``local'' Ð as a result of di€erent products, processes and people at the sites.

To explore research question 1.1, we examine the signi®cance of direct associations between twelve contextual factors (Table 1, column 8). Three variables hypothesized to in¯uence indivi-duals' evaluations of ABC are considered: (1) the extent to which the individual believes that change is warranted (CHANGE) Ð what the organiza-tional psychology literature terms ``felt need for

Table 2

Distribution of survey respondents by site and company

Number of respondents Number of respondents Plant processes Total ABC developers Plant management

Company 1 Company 2 Company 1 Company 2

Core manufacturing

Major stampings 22 3 2 10 7

Contiguous stamping and assembly 14 0 4 0 10

Foundry 28 4 2 10 12

Engine manufacture 31 5 9 9 8

Engine manufacture 24 2 5 10 7

Transmission assembly 40 7 5 15 13

Secondary manufacturing

Parts assembly 19 3 2 7 7

Parts machining and assembly 19 4 4 5 6

Parts machining and assembly 18 2 3 8 5

Electronics assembly 27 6 2 12 7

Other: service parts distribution 23 10 5 5 3

(12)

change''; (2) individual commitment to the orga-nization (COMMIT); and (3) the extent to which the individual identi®es with the values of the organization (VALUES). Of the four variables identi®ed as individual contextual factors in Table 1, two are captured in these measures. A third factor, ABC training received, was measured; however, because most managers receive no train-ing and most developers receive 30±40 hours of training, this ``variable'' is confounded with other factors that di€er between managers and ABC developers (e.g. role involvement in the ABC sys-tem). A fourth factor Ð the individual's knowl-edge of production processes and job experience, was explored in interviews only.

Eight organizational contextual factors are hypothesized to in¯uence evaluations of the ABC system and management involvement in the implementation process: (1) the extent to which individual performance is linked to rewards

(REWARD)9 (2) the competitive environment

(COMPETE); (3) the quality of existing informa-tion systems (INFOQUAL); (4) environmental turbulence (TURB); (5) the likelihood of employee layo€s (LAYOFF); (6) impediments to plant growth (NOGROW); (7) the perceived importance of the plant to the company (IMPPLT); and (8) the perceived importance of cost reduction to the plant (IMPCOST). Advocates of ABC claim that new cost data are most valuable when competition or limited growth prospects cause ®rms to focus on cost reduction. COMPETE, NOGROW and IMPCOST measure these motivations for adopt-ing ABC. Cost reduction e€orts often produce reductions in employment (Innes and Mitchell, 1995). In a unionized environment with con-tractual employment guarantees, managers are often reluctant or unable to reduce employee headcount. As a result, many of the prospective bene®ts of ABC systems may be unrealizable. A measure of the likelihood of layo€s (LAYOFF) is

used to capture this potential deterrent to ABC implementation. In the same vein, the quality of historical management±labor relations (LABOR) is a ninth contextual factor that is hypothesized to in¯uence the degree to which the local union

sup-ports the ABC implementation project (but not

managers' evaluations of the ABC system). Even in ®rms that face heightened competition, some operations are less threatened than others. Selec-tion of core and peripheral component plants maximizes the observed range of competition within each ®rm. The perceived importance of the site (IMPPLT) to the ®rm is included as a poten-tial mitigating factor to external competition. Organizational theorists argue that there are limits to the amount of change that an organization can absorb. Site-speci®c turbulence (TURB) is a mea-sure of concomitant changes that compete with ABC for management attention. Finally, within a ®rm, sites often have diverse ``legacy'' information systems that are more or less e€ective in meeting managers' information needs. Adoption of new information technology such as ABC systems depends upon ®t with and incremental improve-ment upon existing systems (Kwon & Zmud, 1987). A measure of respondents' beliefs about the quality of existing information systems (INFOQ-UAL) is used to examine how the current

infor-mation environment in¯uences managers'

commitment to or evaluation of the ABC system.

4.2. Process variables related to ABC implementation

Appropriate process variables for analysis and the relation between process variables depend upon ®rm-speci®c ABC implementation strategies. Previous studies that consider the relation between process variables and ABC project outcomes test for correlation between a wide range of possible process variables without knowledge ex ante of ®rms' implementation processes. As a result, tests that pool ®rms with di€erent implementation strategies can not distinguish between relevant process factors that are not statistically correlated with ABC outcomes and irrelevant process fac-tors. The ABC implementation process variables that we consider are those identi®ed in Table 1,

(13)

column 6. The hypothesized structure between the process variables (Fig. 1) was developed prior to data collection based on interviews at the corporate level and is described below.

The general structure of the ®rms' ABC imple-mentation process is depicted inside of the box labeled ``ABC Implementation Process Factors'' in Fig. 1. Speci®cally, ®rm-level managers committed the company to ABC implementation prior to the ABC project at the research sites. Respondents' opinions about the strength of this commitment are measured by the variable, MSUPPORT. The management awareness sessions that each ®rm used to introduce ABC to local managers was intended to increase managers' knowledge of ABC and to secure their involvement in the project. Respondents' beliefs about whether local commit-ment was achieved are measured by the variable, MINVOLVE. The proposed relation between MSUPPORT and MINVOLVE re¯ects the direc-tion of in¯uence described above. Both ®rms assigned responsibility for implementing ABC to local management. Although corporate resources were available to augment or support the team, team members were selected and freed from other responsibilities (or not) by local managers. Local managers were also given discretion in the deci-sion to involve local union leaders in the project. The adequacy of resources committed to the ABC project is measured with the variable, RESOUR-CES. The degree to which the union was aware of and involved in the ABC project is measured by the variable, USUPPORT. The proposed relation between MINVOLVE and both RESOURCES and USUPPORT re¯ects the gatekeeper role that local managers are assigned. The proposed rela-tion between MSUPPORT and USUPPORT re¯ects a possible ¯ow of in¯uence from top man-agers of the ®rm, to the local union. Although top managers did not intervene directly in local union a€airs, at least one ®rm noted that the ®rm's decision to implement ABC had become a nego-tiating point with labor at the national level. The proposed relation between MSUPPORT and RESOURCES re¯ects the support that the corporate ABC group provided development teams after local management initiated an ABC project.

The relationships described above re¯ect ®rm-speci®c implementation processes that have been ignored in previous research. However, ®nding that these relations hold is simply evidence of face validity of the data rather than evidence on the proposed research questions. To explore research question 1.2 we examine the e€ects of each process variable on evaluations of the ABC system. Because implementing ABC is a local management decision, we explore research question 1.3 by examining the association between contextual fac-tors and local managers' involvement in the ABC project (MINVOLVE).

4.3. Evaluation measures of the ABC system

As in previous studies we employ an overall evaluation of the ABC system (OVERALL) that allows respondents to self-de®ne the evaluation criteria. However, as research question 2.1 indi-cates, this raises the question: ``What criteria are embedded in overall assessments of ABC sys-tems?'' Content analysis of 236 taped interviews is

used to explore this question.10 Full text

tran-scripts of the 236 interviews were searched for key words related to ABC system evaluations. Search

results yielded 129 respondents'' opinions.11 The

129 respondents represent 20 of the 21 sites and are split in approximately the same proportion as the survey respondents (Table 2) between the ®rms and between ABC developers and managers. After identifying discussions of ABC system evaluation, the transcripts were read by a researcher and a

research assistant. Three evaluation criteria

emerged: (1) use of ABC data for cost reduction; (2) use of ABC data for process improvements; and, (3) improved accuracy of product cost infor-mation relative to the traditional cost system. The

10 An analysis software package designed for non-numerical, unstructured data (NU.DIST1) was used. The researcher codes

interview passages related to constructs of interest and attri-butes of the interviewee. The software uses these codes to create a structured index that is useful for investigating systematic response patterns.

(14)

researcher and research assistant independently coded the responses as belonging primarily to one of the three categories. The codes were compared and di€erences were discussed and resolved. Although respondents occasionally combined the ®rst and second criteria in their discussion, very few respondents discussed the third criterion in conjunction with either of the others. Conse-quently, we combine the ®rst and second groups to form two groups who di€er in the way that they de®ne an e€ective ABC system. In the ®rst group are 67 respondents who de®ne e€ectiveness as use of ABC data in cost reduction or process improve-ments. In the second group are 47 respondents who believe that increased cost accuracy de®nes an

e€ective ABC system.12

Contingency tables are used to evaluate the association between opinions about evaluation criteria and: the respondents' company; whether the respondent worked in a plant that faced sig-ni®cant external competition; whether the respon-dent was an ABC system developer; whether the respondent claimed to have received training in ABC; and the respondent's job title. Small sample sizes cause some contingency tables to have observed and expected counts that violate asymp-totic assumptions for normal chi-square estima-tions. Consequently, we use exact test procedures that do not impose distributional assumptions. Comparing di€erences between respondents who evaluated ABC system e€ectiveness based on use in cost reduction or process improvement with those who base their opinion on improved accu-racy of cost data, respondents' job title is the only signi®cant (p<0.01) determinant of which view is held. The same result is obtained in a multivariate analysis. When a logit model is estimated for the probability of holding the ``accuracy'' view, the respondent's job title is the only signi®cant

expla-natory variable.13Respondents with jobs that are

linked to production (e.g. engineering, materials

management, production operations, quality con-trol) are signi®cantly more likely to evaluate the ABC system based on whether it provides more accurate costs than are respondents in adminis-trative roles (e.g. plant manager, ®nance and accounting, information systems, human resource management). Di€erences in opinions are unre-lated to the respondent's ®rm, whether the respondent works in a site that faces competition, whether the respondent is a system developer or whether the respondent received ABC training.

The content analysis is consistent with evidence (Cooper et al., 1992) that ®rms typically adopt ABC with one of two objectives: increased pro-duct cost accuracy (e.g. for inventory valuation) or use in process improvement (e.g. activity based management). Cooper et al. hypothesize that dif-ferent ABC project objectives require di€erent implementation processes. Foster and Swenson (1997) ®nd that di€erent evaluation measures are correlated with di€erent implementation process variables. We explore these issues (research ques-tion 2.2), by substituting for the OVERALL mea-sure of success in Fig. 1, meamea-sures of ABC data accuracy (ACCURACY) and use (USE) developed from the survey data. Simultaneous estimation of both aspects of ABC system e€ectiveness allow us to provide evidence on whether the two objectives are compatible or are mutually exclusive outcomes.

4.4. Descriptive statistics

Responses to 61 survey questions are used as indicators of 16 latent contextual (12) and imple-mentation process (4) variables and of measures of ABC implementation e€ectiveness (3). Table 3 provides the full text of the survey questions developed for each latent variable, as well as summary descriptive statistics and Cronbach's (1951) measure of construct validity. Responses to 58 of the survey items generated responses that spanned the ®ve point Likert-type scale. With two exceptions the constructs are adequately identi-®ed, with Cronbach's alpha exceeding the 0.6 level used in exploratory research. Item-level response rates from the 265 respondents ranges from 190 to 265. There are typically two reasons for non-response: (1) inadequate knowledge for forming

12 Ten respondents (7.8%) o€ered opinions that were su-ciently unusual that they were discarded. Five respondents o€ered valid but very di€erent responses and thus are not included in subsequent analysis.

(15)

Summary statistics for manifest variables and latent constructsa

Latent construct Survey items Item Item Std. Cronbach's

(R=reverse coded item) 5=strongly agree, 1=strongly disagree N Mean Dev. std. alpha

Measures of ABC implementation success

OVERALL: overall value of ABC 0.93

1 Despite the implementation challenges, I am convinced that ABC is the right tool for helping us manage costs in this company

245 3.8 0.75

2 Overall, the bene®ts of ABC data outweigh the costs of installing a new system

220 3.6 0.87

3 Supporting ABC is the right thing to do in this company 240 3.9 0.79 4 If I were asked to decide whether this company should continue

implementing ABC, I would vote to continue

240 3.9 0.92

5 In general ABC is a good thing for this company 244 4.0 0.70

ACCURACY: perceived accuracy of ABC data 0.65

1 (R) The ABC costs do not seem reasonable to me based on what I know about this plant

223 3.8 0.81

2 The results from the ABC model matched my intuition about costs of production

233 3.6 0.77

3 Data from the ABC model provides an accurate assessment of costs in this plant

240 3.7 0.72

USE: perceived use of ABC data 0.70

1 Information from the ABC model has had a noticeable positive impact on this plant

228 2.9 0.83

2 (R) I am reluctant to use ABC data in place of costs from the traditional cost system

232 3.4 0.96

3 (R) The ABC model has not been used and has been `gathering dust' since it was completed

233 3.5 0.97

4 Data from the ABC model are used for special costs studies 229 3.2 1.0

ABC implementation process variables

MSUPPORT: top management support 0.77

1 This company's top managers have provided visible support for the ABC initiative

252 3.4 1.1

2 Support for implementing ABC in this company comes from both the manufacturing operations and ®nance groups

254 3.2 1.0

3 Support for implementing ABC in this company is widespread 250 2.9 1.0

MINVOLVE: local management knowledge of and involvement in ABC 0.72

1 The managers of this plant are knowledgeable about the theory of ABC 243 3.4 0.80 2 Most managers of this plant are capable of using ABC data to reduce costs 238 3.1 1.0

(continued on next page)

S.W.

Anderson

,

S.M.

Young

/

Accountin

g,

Organizations

and

Society

24

(1999)

525±559

(16)

Table 3 (continued)

Latent construct Survey items Item Item Std. Cronbach's

(R=reverse coded item) 5=strongly agree, 1=strongly disagree N Mean Dev. std. alpha

3 Most managers of the plant were involved in determining how thei

departmental expenses were allocated to activities and products

228 3.5 1.0

4 When the developers of the ABC model met with local managers,

they received suggestions from the managers

205 3.4 0.91

USUPPORT: union support 0.77

1 The local union is receptive to the concept of ABC 190 2.9 0.87

2 The local union was involved in decisions that a€ected the ABC model 192 2.6. 0.99

RESOURCES: adequacy of resources for the ABC development project 0.63

1 The people who developed the ABC model had the equipment and

materials needed to do their job

214 3.9 0.63

2 The people who developed the ABC model had access to the people

from whom they needed to get information

224 4.0 0.60

3 The ABC development project was adequately sta€ed to insure

completion of the task in the time allotted

210 3.3 0.90

Contextual variables

COMPETE: competitive environment 0.54

1 Future demand for the products that this plant produces is uncertain 264 2.6 1.0

2 Competitive pressures could cause this plant to close 264 3.3 1.0

3 This plant has had a lot of management turnover in recent years 263 3.3 0.98

4 This plant faces competition from other plants in this company for business 241 3.3 1.2

5 This plant faces sti€ competition from outside companies for business 264 3.8 0.97

INFOQUAL: quality of other information systems 0.76

1 Most of the data required for a good ABC model are readily available

in this plant

222 3.0 1.1

2 The plant's information systems generally provide data that are

accurate and up to date

246 3.0 1.0

3 (R) The information systems of this plant contain many data errors 238 2.8 1.0

TURB: environmental turbulence 0.61

1 The working environment at this plant changes constantly 264 2.9 0.88

2 Manufacturing processes at this plant change all the time 261 3.0 0.96

3 New management programs are introduced all the time in this plant 265 3.5 0.82

LAYOFF: history of employee layo€s 0.70

1 (R) The threat of layo€s or cutbacks to hourly workers is low 264 2.4 0.99

2 This plant has had major cutbacks and layo€s in recent years 264 2.3 1.1

NOGROW: impediments to plant growth 0.71

1 Getting authorization to hire new employees for this plant is dicult 263 4.0 0.93

S.W.

Anderson

,

S.M.

Young

/

Accountin

g,

Organizations

and

Society

24

(1999)

(17)

employees who retire or leave this plant

LABOR: quality of labor relations 0.87

1 At this plant the union and management have similar goals 261 2.9 1.1

2 (R) Relations between labor and management need to be improved at this plant 261 2.4 0.92

3 Labor and management in this plant work well together 262 3.2 0.98

IMPPLT: importance of the plant to company 0.78

1 This plant is one of the most important manufacturing sites of this company 241 4.0 0.87 2 This plant produces products that have a major in¯uence on this company's

pro®tability

264 4.4 0.70

3 This plant is critical for the success of this company 262 4.0 1.0

IMPCOST: importance of cost reduction to plant 0.40

1 This plant's cost reduction e€orts are important to the company 264 4.3 0.53

2 Cost reduction is the most important objective in this plant 264 3.0 0.94

3 (R) Cost reduction is not a major concern in this plant 263 4.2 0.82

CHANGE: felt need for change 0.58

1 Changes in the way we work in this plant are needed 265 3.8 0.85

2 (R) There is no need for this plant to change the way it does things 265 4.2 0.73

3 I would like to see changes in plant policies and procedures 264 3.6 0.71

COMMIT: commitment to the organization 0.73

1 I am proud to work for this company 264 4.3 0.65

2 (R) I feel very little loyalty to this company 264 4.2 1.0

3 I am proud to work for this plant 264 4.2 0.77

4 (R) I feel very little loyalty to this plant 265 4.1 1.1

VALUES: shared organizational values 0.75

1 My values and the values of this company are quite similar 264 3.7 0.69

2 My values and the values at this plant are quite similar 263 3.6 0.82

REWARD: reward expectancy 0.84

1 In this plant, high quality work increases my chances for a raise,

a bonus, or a promotion

264 3.6 1.0

2 In this plant, ®nancial rewards are tied directly to performance 264 3.1 1.0

3 The ability to reduce costs is rewarded in this plant 264 3.3 0.95

4 In this plant, high performance is recognized and reward 263 3.3 0.92

a This table provides the text of the survey items and descriptive statistics for each of the items that comprise the latent constructs used in subsequent analysis. Item

response indicates the number of respondents that completed the question out of a possible sample of 265 respondents. Descriptive statistics and scale reliability (Cron-bach's alpha) are reported for the reduced sub-sample of 112 respondents who completed all questions (e.g. with listwise deletion of missing observations).

S.W.

Anderson

,

S.M.

Young

/

Accountin

g,

Organizations

and

Society

24

(1999)

525±559

(18)

an assessment, and (2) apprehension about respondent anonymity. Understandably, items with the highest non-response are those related to management-union relations, a politically sensitive topic. Higher response rates for environmental context variables, which depend on general plant knowledge alone, compared to ABC implementa-tion process variables, which require knowledge of the ABC project, suggests that inadequate knowl-edge may also explain di€erential response rates.14

Earlier we argued that individuals are the appropriate unit of analysis for modeling the determinants of evaluations of ABC systems because organizational theory suggests signi®cant within-site variation in perceptions of contextual and process factors likely to in¯uence these assessments. The counter argument is that surveys completed by individuals who observe the same ABC system should be identical and that variation simply re¯ects measurement error. Table 4 pre-sents evidence on the existence of respondent e€ects. Analysis of variance is used to estimate a ®xed e€ects, hierarchical linear model with ®rm, site, and respondent e€ects (Bryk & Raudenbush, 1992). The hierarchical model nests respondents within sites and sites within ®rms. JointF-tests on

Table 4

Analysis of company, plant and respondent e€ectsa

Variable p-Value onF-statistic on ®rm e€ects

p-Value ofF-statistic on plant e€ects within ®rms

p-Value ofF-statistic on respondent e€ects within plants, within ®rms

Model adjusted (R2)

OVERALL 0.000 0.000 0.000 0.67

ACCURACY 0.016 0.000 0.000 0.45

USE 0.000 0.000 0.000 0.33

MSUPPORT 0.000 0.000 0.000 0.49

MINVOLVE 0.000 0.000 0.000 0.39

USUPPORT 0.844 0.000 0.000 0.65

RESOURCES 0.016 0.000 0.000 0.20

COMPETE 0.001 0.000 0.043 0.10

INFOQUAL 0.017 0.000 0.000 0.46

TURB 0.263 0.002 0.000 0.21

LAYOFF 0.124 0.000 0.000 0.49

NOGROW 0.407 0.000 0.000 0.48

LABOR 0.000 0.000 0.000 0.48

IMPPLT 0.423 0.000 0.000 0.53

IMPCOST 0.637 0.018 0.887 0.00

CHANGE 0.101 0.000 0.000 0.27

COMMIT 0.693 0.223 0.000 0.38

VALUES 0.005 0.000 0.000 0.67

REWARD 0.559 0.000 0.000 0.52

a For each dependent and independent variable, thep-values ofF-statistics from an ANOVA model of nested ®rm, site and respondent e€ects are reported. The estimated ®xed e€ects model treats respondents as nested within sites, which are in turn nested within ®rms. The model estimated for responses to items that comprise each construct (X) is:

Xˆintercept‡1…Firm† ‡2…Site Firm‰ Š† ‡3…Respondent Site Firm‰ ‰ ŠŠ† ‡"

The results indicate signi®cant respondent e€ects for all but one variable. This supports the claim that evaluation of ABC systems is signi®cantly related to individual perceptions and that individuals are an appropriate unit of analysis (Bryk & Raudenbush, 1992; Goldstein, 1987; Hannan, 1991).

(19)

the signi®cance of each e€ect indicates signi®cant ®rm, plant and respondent e€ects for all three mea-sures of ABC implementation e€ectiveness. Simi-larly, three of the four process variables di€er signi®cantly at all three levels.F-tests of union sup-port evince plant and respondent e€ects, but not ®rm e€ects. Eleven of the 12 contextual variables show strong evidence of respondent e€ects. There is signi®cant agreement about the importance of cost (IMPCOST) within a site; however, the importance of cost di€ers signi®cantly between sites of a ®rm. After site e€ects are included, there is no di€erence between the ®rms in the importance attributed to cost. The overwhelming signi®cance of respondent e€ects for these constructs is consistent with orga-nizational research that documents the importance of unique, individual circumstances in shaping individual beliefs about and behaviors toward a particular innovation. The data support using indi-viduals as the unit of observation and argue against data reduction (e.g. averaging survey responses of a site), which would be appropriate if random mea-surement error was the primary source of with-in site disagreement (Goldstein, 1987; Hannan, 1991; Seidler, 1974; Rousseau, 1985).

5. The relation between evaluations of ABC systems and contextual and process variables

Structural equation modeling (SEM) is used to investigate the correspondence of the data with Fig. 1. SEM is a method of assessing the correspondence of a proposed set of associations and implied var-iance and covarvar-iance relationships with observed sample variances and covariances (Bollen, 1989). A maximum likelihood ®tting function is the basis for

assessing goodness of ®t.15 Maximum likelihood

estimates are presented for the coecients of the

measurement model, which relates the survey items to latent variables, and the structural model, which relates latent variables to one another. The relation between a latent variable,i, and a survey item that comprises it,xj, is: xjˆliji‡ij, wherelij is the loading, or degree of association between the latent variable and the manifest variable, andij is mea-surement error associated with the survey item. The latent variable approach mitigates measurement error in individual survey items and provides better estimates of the relation between latent variables. None of the con®dence intervals for correlations of the latent variables with one another includes the value of one; a necessary but insucient condition for the constructs to be distinct from one another (discriminate validity).

5.1. The relation between overall evaluations of ABC systems and process and contextual variables

Sample size limitations preclude estimating the full model depicted in Fig. 1 simultaneously. Consequently, the analysis proceeds in three steps. In the ®rst step the relation between contextual variables and overall evaluation of the ABC

sys-tem (OVERALL) is examined. Untabulated

results indicate that the only contextual variables that exhibit signi®cant (p <0.05) direct associa-tions with OVERALL are the reward environment (REWARD) of the site and the quality of existing information systems (INFOQUAL). Beliefs that individual performance is rewarded at the site (high reward expectancy) are positively associated with evaluations of the ABC system. Beliefs that existing information systems provide adequate, accurate information are negatively associated with evaluations of the ABC system, suggesting that ABC adds little in environments with good information systems. In the second step of analysis the relation between contextual variables and local management involvement (MINVOLVE) in the ABC project is examined. The only contextual variable that exhibits a signi®cant (p<0.05) rela-tion with MINVOLVE is the reward environment; high reward expectancy is positively related to local management involvement in the ABC project.

In the ®nal step, the ®rst two steps are com-bined, retaining only those contextual variables

(20)

that are associated with evaluations of the ABC system or local managers' involvement in the ABC

project in earlier steps (INFOQUAL and

REWARD). An additional contextual variable, the quality of historical labor-management relations (LABOR), is considered a potential antecedent of

union support of the ABC project.16Table 5

pre-sents coecient estimates of the measurement model (Panel A) and structural equation model (Panel B) for the relation between ABC imple-mentation process variables, the limited set of contextual variables and respondents' overall eva-luation of the ABC system. Reported results are based on analysis of the covariance matrix for the survey items and a sample in which observations with missing values are excluded (listwise dele-tion). Alternative treatments of missing values (e.g. pairwise deletion or imputation of missing values) do not alter the qualitative results. Boot-strapping methods (500 samples with replacement)

are used to generate approximatep-values for the

signi®cance of total e€ects of the contextual and process variables on overall evaluations of the ABC system (Stine, 1989).

The standardized estimated loadings of survey items on latent constructs in Panel A are sig-ni®cant at the p<0.01 level (two tailed test), and are large enough (greater than 0.4) to provide con®dence that they measure common latent con-structs. Overall model ®t is excellent, as indicated by two goodness of ®t measures: the comparative ®t index (CFI) developed in Bentler (1989), and the root mean square error of approximation

(RMSEA) with its 90% con®dence interval.17

Considering sub-model ®t, 56% of the variation in respondents' overall evaluation of the ABC system is explained by process variables. The remaining

endogenous process variables are also well

explained, suggesting that the survey data are valid representations of the ®rms' implementation process.

The results indicate that three process variables positively in¯uence respondents' overall evalua-tion of the ABC system: top management and union support of the ABC project and the ade-quacy of resources devoted to the project. The signi®cant associations of the reward environment and the quality of existing information systems on overall evaluations that were discovered in the ®rst step of the analysis persist. However, when total e€ects are considered, the positive indirect e€ect of high quality information systems acting through local managers' involvement in the ABC project o€sets the negative direct e€ect on the overall eva-luation of the ABC system. Also, the relation between local management involvement and the reward environment (step 2, above) disappears when the direct in¯uence of the reward environment on system evaluations is estimated simultaneously. Local management involvement is positively asso-ciated with top management support and with the quality of existing information systems. Union sup-port of the ABC project is associated positively with local management support and with the quality of historical labor relations. The adequacy of resources devoted to the ABC project appears to be in¯uenced more by top management support than by local management support of the ABC project. We might expect this result to di€er between companies, since one company systematically assisted the sites in stang ABC projects with consultants and corpo-rate employees; however, as we will see later, the result holds for both companies.

The results provide face validity of the proposed structure of the ®rms' ABC implementation pro-cess. Top management support is directly asso-ciated with local management awareness of and involvement in the ABC implementation project. Local management plays a gatekeeper role in assuring local union support and there is no evi-dence that top management intervenes in this relation. In contrast, local managers' knowledge of and involvement with the ABC project has little relation to whether the development team is pro-vided adequate resources. Whether the project is believed to have an adequate resource endowment is related to top management support.

Comparing previous research results (Table 1) to the total e€ects of Table 5, the sign of

coe-16 Recall from Section 4.1 that historical labor relations (LABOR) are not hypothesized to in¯uence management involvement in ABC implementation.

(21)

Maximum likelihood estimation of the relation between contextual and process variables and overall ABC implementation success Panel A: standardized coecients for the measurement modelsa

Survey item OVERALL MSUPPORT MINVOLVE USUPPORT RESOURCES INFOQUAL REWARD LABOR

1 0.90 0.91 0.64 0.72 0.61 0.57 0.80 0.79

2 0.76 0.64 0.60 0.88 0.71 0.81 0.67 0.74

3 0.88 0.68 0.60 0.46 0.79 0.66 0.95

4 0.90 0.69 0.89

5 0.86

Panel B: maximum likelihood estimates of the coecients of the structural modelt-statistics (in parentheses)b

Estimated direct e€ects Total e€ects (approximatep-value)

OVERALL MINVOLVE USUPPORT RESOURCES Predicted e€ect Estimated e€ect

Independent variables

MSUPPORT 0.20 0.30 ÿ0.19 0.17 + 0.29

(1.68) * (4.02)*** (1.63) (2.33)** (0.012)**

MINVOLVE ÿ0.35 ± 0.86 0.21 + 0.10

(1.21) ± (3.08)*** (1.47) (0.735)

USUPPORT 0.34 ± ± ± + 0.34

(2.50)** ± ± ± (0.016)**

RESOURCES 0.76 ± ± ± + 0.76

(2.58)*** (0.042)**

INFOQUAL ÿ0.25 0.18 ± ± ÿ ÿ0.23

(2.08)** (1.92)* ± ± (0.132)

REWARD 0.19 0.07 ± ± + 0.19

(1.96)** (0.87) (0.042)**

LABOR ± ± 0.17 ± 0.06

± ± (2.00)** ± (0.030)**

Model ®t statistics R2=0.56 R2=0.55 R2=0.39 R2=0.44 CFI=0.932 RMSEA=0.054, 90% C.I.=[0.039,0.068]

a All loadings are signi®cant at 0.01 (two-tail) level.

b The sample (N=112) includes only observations with no missing values. The results are qualitatively similar when data imputation methods are employed. Two

measures of overall model ®t are provided: Bentler's (1989) Comparative Fit Index (CFI) and the Root Mean Square Error of Approximation (RMSEA) with its 90 percent con®dence interval. Bootstrapping methods (500 samples with replacement) are used to generate approximatep-values (two-tail) for the signi®cance of the total e€ects of contextual and procedural variables on OVERALL. Thep-values are approximations because bootstrapping methods do not make distributional assumptions.

***,**,* Statistically signi®cant at thep<0.01, 0.05 or 0.10 (two-tail) level.

S.W.

Anderson

,

S.M.

Young

/

Accountin

g,

Organizations

and

Society

24

(1999)

525±559

Referensi

Dokumen terkait

Berdasarkan pada Berita Acara Pembuktian kualifikasi Nomor: 087/ULP-Pokja-I- JK/APBD-P/2015 tanggal 22 September 2015, pekerjaan Penyusunan Feasibility Studi Techno

Berdasar kajian teoritik pada bab-II, beberapa faktor basis yang dapat dijadikan butir pertimbangan untuk mendekati pencarian indikator-indikator yang memiliki potensi daya

the governance index of Yogyakarta in 2016 was also higher than the national average in 2012. The increase in increasing Yogyakarta governance index is very interesting for

hubungan antara perilaku diet anak dengan pengalaman ECC pada anak usia 37-71. bulan di Kecamatan

dipergun akan dalam kegiatan m otorik di dalam ruan gan , sedan gkan keteram pilan koordin asi otot kasar dilaksan akan di luar ruan gan karen a m en cakup kegiatan gerak

Cara mengonsumsi makanan / minuman merupakan salah satu faktor yang juga berperan dalam proses terjadinya ECC. Salah satu contoh ialah mengonsumsi gula sebelum tidur. Menurunnya

[r]

[r]