- Email: [email protected]
Perceived semantic expressiveness of accounting systems and task accuracy effects
International Journal of Accounting Information Systems 1 (2000) 79–87
Perceived semantic expressiveness of accounting systems and task accuracy effects Cheryl L. Dunna,*, Severin V. Grabskib a Department of Accounting, Florida State University, Tallahassee, FL 32306-1110 USA Department of Accounting, Michigan State University East Lansing, MI 48824-1121 USA
b
Abstract Semantic expressiveness refers to how well a model reflects the underlying reality the model represents. Prior research has claimed the REA (Resources-Events-Agents) accounting model is more semantically expressive than is the traditional DCA (Debit-Credit-Account) accounting model. This research demonstrates experimentally that users perceive REA as more semantically expressive than DCA. This study also demonstrates that, controlling for cognitive fit, accounting knowledge, and field dependence, higher perceived semantic expressiveness is associated with higher task accuracy. © 2000 Elsevier Science Inc. All rights reserved. Keywords: Semantic expressiveness; REA accounting model; DCA accounting model; Cognitive fit.
1. Semantic expressiveness Semantic expressiveness is a term that refers to how well a model reflects the underlying reality the model represents. Computer scientists have long advocated the integration of semantics (real-world meaning) into data models and into technologies centered on those models (e.g., Abrial, 1974; Brodie, 1984; Hammer and McLeod, 1981). In accounting, McCarthy (1982) claimed that semantic expressiveness is an important advantage of the Resource-Event-Agent (REA) accounting model over the traditional DebitCredit-Account (DCA) accounting model. Dunn and McCarthy (1997) reiterated McCarthy’s (1982) position that accounting systems that use real world business phenomena as primitives are more semantically expressive than are accounting systems that use double-entry artifacts as primitives. They identified benefits of a semantically expressive accounting system as including easier integration of accounting phenomena with descriptions of non-accounting phenomena, and a better understanding by users of the system. To our knowledge, no research has attempted
to verify McCarthy’s claim that the REA model is more semantically expressive than the DCA accounting model, nor have any studies attempted to link semantic expressiveness with task accuracy. Two approaches can be taken to evaluate the semantic expressiveness of alternative accounting models. One approach is to identify as many features of the underlying reality as possible, and then to determine which of those features can be captured by and represented in each of the alternative accounting models. Such an approach (which could be referred to as an ontological approach) assumes that each and every user will agree on the underlying reality and the representational model, and will interpret the model in exactly the same manner. This approach ignores perceptions of users as to how well the model helps them to understand the underlying reality. A second approach is to assume that the degree of semantic expressiveness of a model (or a system based on the model) is indicated by user perceptions as to how well the model represents the underlying reality. This study takes the latter approach, having users of systems based on
* Corresponding Author. Tel.: 1-850-644-7878. E-mail: [email protected] 1467-0895/00/$ – see front matter © 2000 Elsevier Science Inc. All rights reserved. PII: S1 467-0895(00)00 0 0 4 - X
80
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
alternative accounting models evaluate how well those systems represented the underlying reality. This study also examines whether there is any demonstrable benefit associated with semantic expressiveness. Even if there is a difference between systems based on semantic expressiveness, if there is no associated performance difference there is no economic incentive to incur costs to change from a less semantically expressive system to a more semantically expressive system. Consequently, this study investigates whether an accounting system with a higher degree of perceived semantic expressiveness results in greater task accuracy. The remainder of this article is structured as follows: Section 2 provides the hypothesis development; Section 3 describes the method used to test the hypotheses; Section 4 presents the statistical tests and results of those tests; and Section 5 provides concluding discussion. 2. Hypothesis development For many years the traditional DCA accounting model was the only commonly accepted basis for accounting information systems (AIS) design. The REA accounting model, proposed by McCarthy (1982) has slowly but surely gained acceptance as an alternative basis for AIS design. This acceptance is evidenced by continued research in this area and by its inclusion in AIS textbooks (e.g., Hollander et al., 1996; Romney et al., 1997). Very few studies have attempted to compare this model with alternatives. Most REA research has been normative and focused on extending the model or on applying it to difficult accounting problems (e.g., Andros et al., 1992; Geerts and McCarthy, 1994; Grabski and Marsh, 1994). In reaction to the limited empirical research focused on evaluating the REA model, Dunn and McCarthy (1997)and David et al. (1999) called for more research to develop and test theories about the benefits of REA compared to other accounting models. Dunn and McCarthy (1997) claim that for an accounting model to be semantically expressive its components must reflect real world phenomena and should not use double-entry artifacts such as debits, credits, and accounts as declarative primitives. That is not to say that a semantically expressive system cannot include debits and credits in any form. To use them as declarative primitives is to make them an integral part of the foundation of the system—declarative primitives, which are the lowest level building blocks upon which everything else is built. Hollander et al. (1996) discuss the fact that an events architecture based on the REA model can include a chart of accounts, debits and credits, and a general ledger.
They explain that the fundamental difference between these types of systems and traditional DCA systems is that in the REA systems the debits, credits, and accounts are produced procedurally as user views, whereas they are primitives in traditional DCA systems. The distinction between producing the general ledger procedurally versus maintaining it as a declarative element of an accounting system is less important than is the distinction between including the accounting artifacts as primitives versus including them as report options. The difference is that when debits, credits, and accounts are used as base objects they become the filter which determines what details enter the system and thus limit the semantic expressiveness of the system. Use of business event details (such as resources, events, and agents) as system primitives is consistent with the recommendation of Hammer and McLeod (1981) to model primitives of a problem domain directly rather than translating them into artificial specification constructs (such as debits and credits) to enhance semantic expressiveness. While traditional DCA systems have some degree of semantic meaning through the chart of accounts coding and hierarchical organization, if a user doesn’t have specific training as to the coding and organization of the chart of accounts, that user would probably not be able to understand the system. Anecdotal evidence from our own experience with college students suggests that many students opt for careers in other business fields instead of in accounting because they have difficulty understanding the primitives of debits and credits. One design objective of the REA model is semantic expressiveness: it models business primitives directly, with terms familiar to users in all departments of a business. Traditional DCA systems translate these business primitives into debits and credits, thus obscuring the meaning to users in non-accounting departments of a business. Traditional DCA systems have also been criticized for not capturing several types of phenomena (e.g., core business activities that do not have an immediate effect on the assets, liabilities, and equity; therefore, they are not accounting transactions) that the REA model specifically models (McCarthy, 1982; Hollander et al., 1996). If these criticisms are valid, they lend support to the notion that REA-based systems include more features of the underlying reality than do DCA-based systems and are thereby more semantically expressive. The literature suggests REA-based systems are, as claimed by McCarthy (1982) and by Dunn and McCarthy (1997), more semantically expressive than are DCA-based systems. If that were the case, then one would expect users to perceive REA-based systems
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
as more semantically expressive than DCA-based systems. Hypothesis 1 is thus proposed, H1: REA-based accounting systems are perceived by users as more semantically expressive than DCA-based accounting systems. Even if one system is perceived as more semantically expressive than another system, the question must be raised as to whether or not it makes a difference. If the use of a more semantically expressive system results in improved user performance, then incentives would exist for the implementation and use of the more semantically expressive system (assuming performance improvements exceeded implementation costs). A more semantically expressive system, by definition, allows a better representation of the underlying reality than the less semantically expressive system. Therefore, users should be less likely to make mistakes because they can more easily refer back to the “reality” of the situation to verify that they are performing the task correctly. That is, users of more semantically expressive systems should be more accurate in task performance than users of less semantically expressive systems. As a result, it is important to show that a system perceived as more expressive than another results in some benefit, such as higher quality task performance; however, factors other than the system itself must be considered in such an analysis. Prior accounting and information systems research has demonstrated that user characteristics, task characteristics, and system or environmental characteristics all contribute to task accuracy (e.g., Libby and Luft 1993; Jih et al., 1989). Accounting knowledge and
81
field independence are two user characteristics that have been demonstrated to affect task accuracy using the REA accounting model (Dunn, 1994; Dunn and Grabski, 1998). Prior literature also demonstrated that one of the most important considerations with regard to tasks is whether or not the problem representation matches the task to be performed. Vessey (1991) introduced the model of cognitive fit, which occurs if the problem representation fits the task well, that is, if the problem solving steps required for a task are supported by the format of the problem representation. Dunn and Grabski (1997) found that cognitive fit affects accuracy of users of the two alternative accounting models in this study. The research model in Fig. 1 is thus proposed, and Hypothesis 2 is derived from this model. H2: The more users perceive an AIS to be semantically expressive, the greater will be their task accuracy, controlling for cognitive fit and user characteristics of accounting ability and field dependence. 3. Method 3.1. Overview A laboratory experiment was used to test the hypotheses. Participants were 73 students enrolled in an undergraduate AIS course. These students had previously completed an average of six undergraduate accounting courses. Demographics are summarized in Table 1. A within-subjects design was used for task accuracy and perceived semantic expressiveness, between-subjects for cognitive fit, and user characteris-
Fig. 1. Research model for Hypothesis 2.
82
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
Table 1 Descriptive statistics, mean (standard deviation)
Overall GPA Age Intermediate accounting grade* Number of accounting classes taken Gender Male Female
Field Dependent, N ⫽ 35
Field Independent, N ⫽ 38
Total, N ⫽ 73
3.13 (.47) 23.31 (5.67) 2.62 (.93) (N ⫽ 33)** 6.03 (2.50)
3.18 (.50) 22.16 (2.61) 3.08 (.67) 5.84 (1.91)
3.16 (.48) 22.72 (4.39) 2.87 (.85) (N ⫽ 71) 5.93 (2.20)
21 14
17 21
38 35
* Significantly different at p less than .009. ** Intermediate accounting grades were not available for two participants, both field dependent, who had taken intermediate accounting at different universities.
tics were controlled as covariates. Participants used both the REA and DCA-based accounting systems and were each given two tasks to complete with the REA-based accounting system and two tasks to complete with the DCA-based accounting system. Each participant completed all four tasks. All tasks asked the participants to explain how they would obtain some information from the system with which they were provided. The Appendix lists the four information retrieval tasks the participants were asked to complete. Participants were randomly assigned as to which system they used first (REA or DCA) and as to which tasks they completed first. The tasks were completed as an in-class exercise worth 3 percent of each participant’s course grade and the participants understood that the number of points they earned on the exercise depended on their accuracy. 3.2. Alternative accounting systems’ documentation The REA-based system documentation consisted of REA diagrams in entity-relationship format and the corresponding relational table structures. The REA diagrams serve as the system’s abstraction from reality; the relational table structures provide the implementation of that abstraction. The DCA-based system documentation consisted of a chart of accounts, sample journals (general and special), and sample ledger accounts (general and subsidiary). The chart of accounts serves as the system’s abstraction from reality1; the journals and ledgers provide the
1 The notion of a chart of accounts providing a description of the underlying reality is evidenced by the new company setup features of many general ledger software packages. Often one of the first questions asked is something like, “What type of business does the company engage in?” The suggested chart of accounts provided (which may be modified) is different for different types of businesses. Thus, if
implementation of that abstraction. Neither set of system documentation included any type of system flowchart or data flow diagram describing how processing occurs within the system. The models and the system documentation based on those models described only the architecture of the data storage mechanism. The purpose of a document/system flowchart is to describe the information processing steps, and has nothing to do with information storage structure. Because participants were asked to identify how they would obtain some information from within the accounting system and not to perform information processing steps, flowcharts, data flow diagrams, and other processing documentation were not provided, nor were they necessary to complete the tasks. Participants were provided with introductory background information about the organization so they could place the DCA and REA system documentation into context. In addition, both sets of system documentation included sample source documents (e.g., sales invoice, purchase order, receiving report, etc.) to clarify the types of activities in which the experimental company was involved. Although some would debate whether source documents are part of a system or whether they are outside a system, participants were asked to use the source documents as if they were part of both systems (the source documents could be viewed as either paper-based in a manual system or they could be input screens in a computer-based system). Familiarization questions were given for each system prior to the experimental tasks; those familiarization questions required particthe business is a funeral parlor, the suggested chart of accounts will include different accounts than if the business is a distributor. Similarly, the Encyclopedia of Accounting Systems (Pescow, 1976) provides various sets of charts of accounts and special and general journals for use with different organization types.
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
ipants to locate items in the sample source documents as well as locating items in the REA diagrams, relational tables, chart of accounts, journals, and ledgers. Each of the four experimental tasks required participants to explain how they would retrieve certain information from their system based on the system documentation (including the source documents). 3.3. Perceived semantic expressiveness Perceived semantic expressiveness was measured as the participants’ response to a seven-point Likert scale question: “The documentation I received provided me with a realistic representation of the accounting information flows of the business” (agree ⫽ 1, disagree ⫽ 7). This question was answered after completing two information retrieval tasks with respect to their use of one of the accounting system’s documentation; then it was answered again after completing the other two tasks with respect to their use of the other system’s documentation. Although students were assured of confidentiality from their instructor as to their response to this question, a demand effect in the answers to this question is possible; however, if participants did not answer according to their actual perception we would not expect to find support for H2, that the perceived semantic expressiveness contributes to task accuracy. Answers were reverse-coded in the data analysis so that higher numbers corresponded to greater perceived semantic expressiveness. 3.4. Cognitive fit Two of the tasks were designed to be easier to complete using a traditional DCA-based system than with a REA-based system. One asked how to determine the company’s accounts receivable balance as of the end of a month; the other asked how to determine the company’s total current liabilities as of the end of a month. The chart of accounts contains accounts receivable and also contains classifications of all liability accounts as either current or non-current. In the REA-based system, multiple diagrams must be consulted and the user must understand how the economic events and the business processes fit together in order to explain how these balances could be calculated. The other two tasks were designed to be easier to complete using a REA-based system than using a traditional DCA-based system. One asked how to determine the company’s gross sales by inventory stock number. This information is not stored at the appropriate level of aggregation in a traditional DCAbased system; users must examine the source documents (sales invoices). In the REA-based system this
83
information is available in the tables that represent sales, inventory, and the relationship between them. The second REA-based task asked how to determine whether the company makes partial payments on its general and administrative service acquisitions. This can be answered easily by looking at the structural constraints on the REA diagram for the general and administrative service acquisition business process, whereas in the DCA system the information could only be obtained by comparing the source documents (general and administrative service orders and the corresponding checks). Cognitive fit was operationalized as a betweensubjects variable. Half of the participants were randomly assigned to complete the REA facilitated tasks using the REA-based system documentation and to complete the DCA facilitated tasks using the DCAbased system documentation. These participants thus were classified as the cognitive fit group. The other half of the participants were randomly assigned to complete the REA facilitated tasks using the DCAbased system and to complete the DCA facilitated tasks using the REA-based system. These participants were classified as the group that did not have cognitive fit. 3.5. User characteristics Two user characteristics were controlled for in this study, accounting knowledge and field dependence. Accounting knowledge was operationalized as the participants’ grades in a junior-level financial accounting and reporting course. Knowledge of accounting concepts should affect accuracy on information retrieval tasks using a DCA-based system. Dunn (1994) found accounting knowledge to be a significant determinant of information retrieval task accuracy using a REA-based system. The intermediate accounting course grade was determined to be the most adequate available surrogate for accounting knowledge. Several participants had taken introductory accounting at other colleges or universities, making the use of grades in those other courses, or overall accounting grade point average less comparable. Field dependence was included in the research model based on Dunn and Grabski (1998), who found that field-independent undergraduate students performed significantly better on conceptual modeling design problems than field-dependent undergraduate students. Field dependence was measured using the Group Embedded Figures Test (GEFT) developed by Witkin et al. (1971). Students were classified as field dependent if they scored twelve or less (n ⫽ 35), and were considered to be field independent if they scored thirteen or greater (n ⫽ 38). This catego-
84
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
rization is based on Witkin et al. (1971) and adjusted to include business school females’ performance equal with males’ based on DeSanctis and Dunikoski (1983). The GEFT was administered on the first day of class. 3.6. Accuracy Accuracy was measured by independent verification of the processes specified by the participants to obtain the needed information for each task. A model solution was developed and each participant’s solution was evaluated relative to the model solution by two independent graders. The scores were then compared and any differences reconciled. Intergrader reliability before reconciliation of differences as indicated by Cronbach’s alpha was .834. 4. Statistical tests and results The mean value for perceived semantic expressiveness (higher ⫽ more semantically expressive) for REA system users was 5.329 (S.D. 1.259). The mean value for perceived semantic expressiveness for DCA system users was 4.164 (S.D. 1.581). A paired samples t-test reveals the difference between these mean values is significant, with t ⫽ 4.591, p ⬍ .000. There was no order effect, that is, it did not matter
which system documentation the participant used first. Users of the REA system documentation perceived the system to be significantly more semantically expressive than did users of the DCA system documentation, therefore H1 is supported. To alleviate concern about alternative explanations associated with test results for H1, further analysis was performed. One concern is that by including tasks for which DCA system users must refer to source documents, DCA is unfairly disadvantaged. Therefore a second t-test was run to compare the perceived semantic expressiveness of users for the REA and DCA models for only the DCA-facilitated tasks. Average perceived semantic expressiveness for REA users completing DCA-facilitated tasks was 5.11 (S.D. 1.26). The mean value of perceived semantic expressiveness for DCA users completing DCAfacilitated tasks was 4.19 (S.D. 1.70). REA users perceived the system as significantly more semantically expressive (t ⫽ 2.671, p ⬍ .011) than did DCA users, even for the DCA-facilitated tasks. Table 2 presents the mean perceived semantic expressiveness broken down by cognitive fit categories, system documentation, and field dependence of users. This breakdown illustrates what is shown by the two t-tests—that differences in perceived semantic expressiveness in this study are associated with the
Table 2 Perceived semantic expressiveness, cell means (standard deviations)
REA documentation users Consistent tasks (cognitive fit) Inconsistent taks (no cognitive fit) Total DCA documentation users Consistent tasks (cognitive fit) Inconsistent tasks (no cognitive fit) Total Grand mean for REA and DCA users
Field Independent Users
Field Dependent Users
Total
5.70 (1.174) 4.944 (1.211) 5.342 (1.236)
5.353 (1.320) 5.278 (1.319) 5.314 (1.301)
5.540 (1.238) 5.111b (1.260) 5.329a (1.259)
4.250 (1.618) 4.167 (1.383) 4.210 (1.492) 4.776d (1.475)
4.118 (1.833) 4.111 (1.605) 4.114 (1.694) 4.714d (1.616)
4.189b (1.697) 4.139 (1.476) 4.164a (1.581) 4.745 (1.539)
Grand Mean (standard deviation) for Consistent Tasks ⫽ 4.865c (1.625). Grand Mean (standard deviation) for Inconsistent Tasks ⫽ 4.625c (1.448). a Significantly different at p ⬍ .000. b Significantly different at p ⬍ .008. c Not significantly different, p ⬍ .809. d Not significantly different, p ⬍ .413.
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
two different accounting systems. It also reveals that the grand mean for all tasks for which users had cognitive fit (4.86) is not significantly different from the grand mean for all tasks for which users did not have cognitive fit (4.63) (t ⫽ 0.24, p ⬍ .81). Similarly, the grand mean for field dependent users (4.71) and the grand mean for field independent users (4.78) are not significantly different (t ⫽ 0.82, p ⬍ .41).2 The second hypothesis, which relates semantic expressiveness to task performance (accuracy), was tested using regression, with two observations for each participant. The regression equation tested is: Accuracy ⫽ ␣ ⫹ o Perceived Semantic Expressiveness ⫹ 1Cognitive Fit ⫹ 2Accounting Knowledge ⫹ 3Field Independence ⫹ ε Table 3 shows the model summary, coefficients, and correlations. Accuracy is significantly correlated with all variables (the only other significant correlation is between field dependence and accounting knowledge). Results indicate support for the research model, with adjusted R-square of .26. All VIF factors were between 1.020 and 1.138 indicating a lack of multicollinearity within the model tested. Standardized Beta weights reveal that cognitive fit is the most significant determinant of accuracy (Beta ⫽ .428, p ⬍ .0001). Perceived semantic expressiveness (Beta ⫽ .159, p ⬍ .031) and accounting knowledge (Beta ⫽ .174, p ⬍ .025) are also significant determinants of accuracy. Field independence (Beta ⫽ .139, p ⬍ .075) is slightly less important in determining accuracy. Overall, H2 is supported, that higher perceived semantic expressiveness is correlated with improved accuracy. It is important to note that H2 and its results do not claim simply that REA users are more accurate than DCA users. Rather, they suggest that if DCA users perceive the system to be more semantically expressive than REA, they will be more accurate with the DCA system than with the REA system. Likewise if REA users perceive the system to be more semantically expressive than DCA, they will be more accurate with the REA system than with the DCA system.
2 A repeated measures ANOVA found a main effect for semantic expressiveness (p ⬍ .001) and an interaction between semantic expressiveness and field dependence (p ⬍ .029). No effect was found for semantic expressiveness by cognitive fit (p ⬍ .516) nor was there any effect associated with semantic expressiveness by cognitive fit by field dependence (p ⬍ .343).
85
5. Discussion Results of this study indicate that users perceive the REA accounting model to be more semantically expressive than the DCA accounting model, thus providing evidence to support McCarthy’s (1982) claim that REA is more semantically expressive. Results of this study also indicate that perceived semantic expressiveness of accounting information systems is important to users. The Beta weights indicate that semantic expressiveness is as important as accounting knowledge in determining task accuracy. These results were obtained with undergraduate accounting majors, who were trained to understand the DCA accounting model (having had an average of six accounting courses) and were thus familiar with the semantic meaning inherent in the chart of accounts coding and organization. These students had less than a semester of training with the REA accounting model. With users who are more familiar with the REA accounting model and/or who are not as familiar with the DCA accounting model, results may be even more pronounced. Future research should be conducted to gain more insight into the generalizability of this study’s results to other types of users, particularly non-student users. The finding in this study of cognitive fit as a significant determinant of accuracy is consistent with Dunn and Grabski (1997). The finding of accounting knowledge as a significant determinant of accuracy for an information retrieval task is consistent with Dunn (1994). The finding of field independence as marginally significant is not consistent with Dunn and Grabski (1998); however, that study examined accuracy with respect to design tasks as opposed to information retrieval tasks. It is possible that field independence is more important for design tasks than for information retrieval tasks. This is a question for future research. An important implication of this study is that evidence suggests perceived semantic expressiveness is a pervasive characteristic of an information system; that is, it will not be affected by a user’s characteristics or by cognitive fit. This evidence, combined with the association of perceived semantic expressiveness with increased task accuracy, indicates that system designers should consider the likely perceived semantic expressiveness of systems as they are developing them. They should also consider cognitive fit, as it has a relatively stronger association with task accuracy; however, cognitive fit will vary from task to task. Thus it may be more meaningful to focus on developing systems that represent real world phenomena as closely as possible and to provide facilities for users to construct data representations that match the task they desire to accomplish.
86
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
Table 3 Regression of accuracy on perceived semantic expressiveness, cognitive fit, accounting knowledge, and field independence Correlations Pearson Correlation Accuracy Field dependence Acct. Knowledge Sem. Express. Cognitive fit Significance (1 tailed) Accuracy Field dependence Acct. Knowledge Sem. Express. Cognitive fit
Accuracy
Field Dependence
Accounting Knowledge
Semantic Expressiveness
Cognitive fit
1.000 .216 .184 .185 .445
1.000 .322 ⫺.085 .081
1.000 ⫺.039 ⫺.066
1.000 .103
1.000
.005 .014 .014 .000
.000 .158 .169
.323 .216
.111
ANOVA
Sum of Squares
df
Mean Square
F
Sig.
Regression Residual Total
46104.798 117389.9 163494.7
4 137 141
11526.199 856.861
13.452
.000
Unstandardized Coefficients
Standardized Coefficients
Coefficients
Beta
Std. Error
Beta
t
Sig.
(Constant) Semantic expressiveness Cognitive fit Accounting knowledge Field dependence
⫺15.434 3.519 29.082 8.252 1.147
13.988 1.613 4.984 3.646 .639
.159 .428 .174 .139
⫺1.103 2.182 5.835 2.263 1.797
.272 .031 .000 .025 .075
Adjusted R Square ⫽ .261.
This study is the first to examine perceived semantic expressiveness of accounting systems, and while the evidence suggests perceived semantic expressiveness is greater in REA systems than in DCA systems and that greater perceived semantic expressiveness is associated with greater task accuracy, no definitive conclusions can be drawn. Future research is required to develop better measures for perceived semantic expressiveness and to determine whether results would differ for other types of users (especially non-accounting and non-student users) and for other types of tasks (other than information retrieval). Future research can also examine whether other benefits result from systems perceived as more semantically expressive than others (such as increased user satisfaction and increased system use). Appendix 1 Information retrieval tasks How would you determine what accounts receivable balance RSW had as of April 30, 1997?
How would you determine gross sales (sales before returns, allowances, and discounts) by inventory stock number for the period January 1—April 30, 1997? (RSW’s fiscal year begins January 1 and ends December 31.) How would you determine RSW’s total current liabilities as of June 30, 1997? How would you determine whether RSW makes partial payments for general and administrative services they acquire?
References Abrial JR . Data semantics. In: Klimbie JW, Koffeman, KL, editors. Data base management. Amsterdam: North Holland Publishing Company; New York: American Elsevier, 1974. p. 1–60. Andros DP, Cherrington JO, Denna EL. Reengineer your accounting, the IBM way. Financ Exec 1992;8:28–31. Brodie ML. On the development of data models. In: Brodie ML, Mylopolous J, Schmidt JW, editors. On conceptual modeling. New York: Springer-Verlag, 1984. p. 19–47. David JS, Dunn CL, Poston R, McCarthy, WE. The re-
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87 search pyramid: A framework for accounting information systems research. J Inform Syst 1999;13:7–30. DeSanctis G, Dunikoski, R. Group Embedded-Figures Test: Psychometric data for a sample of business students. Perceptual and Motor Skills 1983;56:707–10. Dunn CL. An investigation of abstraction in events-based accounting systems. Unpublished doctoral dissertation, Michigan State University, 1994. Dunn CL, Grabski, SV. Cognitive fit and a comparison of two accounting models. Paper presented to the American Accounting Association national meeting, Dallas, TX. August, 1997. Dunn CL, Grabski, SV. The effect of field dependence on conceptual modeling performance. Adv Account Inform Syst 1998;6:65–77. Dunn CL, McCarthy WE. The REA accounting model: Intellectual heritage and prospects for progress. J Inform Syst 1997;11:31–51. Geerts GL, McCarthy WE. The economic and strategic structure of REA accounting systems. Paper presented to the 300th Anniversary Program, Martin Luther University, Halle-Wittenberg, Germany. September 1994. Grabski SV, Marsh R. Integrating accounting and manufacturing information systems: An ABC and REA-based approach. J Inform Syst 1994;8:61–80. Hammer M, McLeod D. Database description with SDM: A
87
semantic database model. ACM Trans Database Syst 1981;6:351–86. Hollander AS, Denna EL, Cherrington JO. Accounting, information technology, and business solutions. Chicago: Irwin, 1996. Jih WJK, Bradbard DA, Snyder CA, Thompson NGA. The effects of relational and entity-relationship data models on query performance of end users. Int J Man-Mach Stud 1989;31:257–67. Libby R, Luft J. Determinants of judgment performance in accounting settings: Ability, knowledge, motivation, and environment. Account Org Soc 1993;18:425–50. McCarthy WE. The REA accounting model: A generalized framework for accounting systems in a shared data environment. Account Rev 1982;57:554–78. Pescow JK. Encyclopedia of accounting systems 2nd Ed. Englewood Cliffs, NJ: Prentice-Hall, 1976. Romney MB, Steinbart PJ, Cushing BE. 1997. Accounting information systems (Seventh edition). Reading: Addison-Wesley, 1997. Vessey I. Cognitive fit: A theory-based analysis of the graph versus tables literature. Decision Sciences 1991;22:219–40. Witkin HA, Oltman PK, Raskin E, Kar, SA. A manual for the Embedded Figures Test. Palo Alto, CA: Consulting Psychologists Press, 1971.
Perceived semantic expressiveness of accounting systems and task accuracy effects Cheryl L. Dunna,*, Severin V. Grabskib a Department of Accounting, Florida State University, Tallahassee, FL 32306-1110 USA Department of Accounting, Michigan State University East Lansing, MI 48824-1121 USA
b
Abstract Semantic expressiveness refers to how well a model reflects the underlying reality the model represents. Prior research has claimed the REA (Resources-Events-Agents) accounting model is more semantically expressive than is the traditional DCA (Debit-Credit-Account) accounting model. This research demonstrates experimentally that users perceive REA as more semantically expressive than DCA. This study also demonstrates that, controlling for cognitive fit, accounting knowledge, and field dependence, higher perceived semantic expressiveness is associated with higher task accuracy. © 2000 Elsevier Science Inc. All rights reserved. Keywords: Semantic expressiveness; REA accounting model; DCA accounting model; Cognitive fit.
1. Semantic expressiveness Semantic expressiveness is a term that refers to how well a model reflects the underlying reality the model represents. Computer scientists have long advocated the integration of semantics (real-world meaning) into data models and into technologies centered on those models (e.g., Abrial, 1974; Brodie, 1984; Hammer and McLeod, 1981). In accounting, McCarthy (1982) claimed that semantic expressiveness is an important advantage of the Resource-Event-Agent (REA) accounting model over the traditional DebitCredit-Account (DCA) accounting model. Dunn and McCarthy (1997) reiterated McCarthy’s (1982) position that accounting systems that use real world business phenomena as primitives are more semantically expressive than are accounting systems that use double-entry artifacts as primitives. They identified benefits of a semantically expressive accounting system as including easier integration of accounting phenomena with descriptions of non-accounting phenomena, and a better understanding by users of the system. To our knowledge, no research has attempted
to verify McCarthy’s claim that the REA model is more semantically expressive than the DCA accounting model, nor have any studies attempted to link semantic expressiveness with task accuracy. Two approaches can be taken to evaluate the semantic expressiveness of alternative accounting models. One approach is to identify as many features of the underlying reality as possible, and then to determine which of those features can be captured by and represented in each of the alternative accounting models. Such an approach (which could be referred to as an ontological approach) assumes that each and every user will agree on the underlying reality and the representational model, and will interpret the model in exactly the same manner. This approach ignores perceptions of users as to how well the model helps them to understand the underlying reality. A second approach is to assume that the degree of semantic expressiveness of a model (or a system based on the model) is indicated by user perceptions as to how well the model represents the underlying reality. This study takes the latter approach, having users of systems based on
* Corresponding Author. Tel.: 1-850-644-7878. E-mail: [email protected] 1467-0895/00/$ – see front matter © 2000 Elsevier Science Inc. All rights reserved. PII: S1 467-0895(00)00 0 0 4 - X
80
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
alternative accounting models evaluate how well those systems represented the underlying reality. This study also examines whether there is any demonstrable benefit associated with semantic expressiveness. Even if there is a difference between systems based on semantic expressiveness, if there is no associated performance difference there is no economic incentive to incur costs to change from a less semantically expressive system to a more semantically expressive system. Consequently, this study investigates whether an accounting system with a higher degree of perceived semantic expressiveness results in greater task accuracy. The remainder of this article is structured as follows: Section 2 provides the hypothesis development; Section 3 describes the method used to test the hypotheses; Section 4 presents the statistical tests and results of those tests; and Section 5 provides concluding discussion. 2. Hypothesis development For many years the traditional DCA accounting model was the only commonly accepted basis for accounting information systems (AIS) design. The REA accounting model, proposed by McCarthy (1982) has slowly but surely gained acceptance as an alternative basis for AIS design. This acceptance is evidenced by continued research in this area and by its inclusion in AIS textbooks (e.g., Hollander et al., 1996; Romney et al., 1997). Very few studies have attempted to compare this model with alternatives. Most REA research has been normative and focused on extending the model or on applying it to difficult accounting problems (e.g., Andros et al., 1992; Geerts and McCarthy, 1994; Grabski and Marsh, 1994). In reaction to the limited empirical research focused on evaluating the REA model, Dunn and McCarthy (1997)and David et al. (1999) called for more research to develop and test theories about the benefits of REA compared to other accounting models. Dunn and McCarthy (1997) claim that for an accounting model to be semantically expressive its components must reflect real world phenomena and should not use double-entry artifacts such as debits, credits, and accounts as declarative primitives. That is not to say that a semantically expressive system cannot include debits and credits in any form. To use them as declarative primitives is to make them an integral part of the foundation of the system—declarative primitives, which are the lowest level building blocks upon which everything else is built. Hollander et al. (1996) discuss the fact that an events architecture based on the REA model can include a chart of accounts, debits and credits, and a general ledger.
They explain that the fundamental difference between these types of systems and traditional DCA systems is that in the REA systems the debits, credits, and accounts are produced procedurally as user views, whereas they are primitives in traditional DCA systems. The distinction between producing the general ledger procedurally versus maintaining it as a declarative element of an accounting system is less important than is the distinction between including the accounting artifacts as primitives versus including them as report options. The difference is that when debits, credits, and accounts are used as base objects they become the filter which determines what details enter the system and thus limit the semantic expressiveness of the system. Use of business event details (such as resources, events, and agents) as system primitives is consistent with the recommendation of Hammer and McLeod (1981) to model primitives of a problem domain directly rather than translating them into artificial specification constructs (such as debits and credits) to enhance semantic expressiveness. While traditional DCA systems have some degree of semantic meaning through the chart of accounts coding and hierarchical organization, if a user doesn’t have specific training as to the coding and organization of the chart of accounts, that user would probably not be able to understand the system. Anecdotal evidence from our own experience with college students suggests that many students opt for careers in other business fields instead of in accounting because they have difficulty understanding the primitives of debits and credits. One design objective of the REA model is semantic expressiveness: it models business primitives directly, with terms familiar to users in all departments of a business. Traditional DCA systems translate these business primitives into debits and credits, thus obscuring the meaning to users in non-accounting departments of a business. Traditional DCA systems have also been criticized for not capturing several types of phenomena (e.g., core business activities that do not have an immediate effect on the assets, liabilities, and equity; therefore, they are not accounting transactions) that the REA model specifically models (McCarthy, 1982; Hollander et al., 1996). If these criticisms are valid, they lend support to the notion that REA-based systems include more features of the underlying reality than do DCA-based systems and are thereby more semantically expressive. The literature suggests REA-based systems are, as claimed by McCarthy (1982) and by Dunn and McCarthy (1997), more semantically expressive than are DCA-based systems. If that were the case, then one would expect users to perceive REA-based systems
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
as more semantically expressive than DCA-based systems. Hypothesis 1 is thus proposed, H1: REA-based accounting systems are perceived by users as more semantically expressive than DCA-based accounting systems. Even if one system is perceived as more semantically expressive than another system, the question must be raised as to whether or not it makes a difference. If the use of a more semantically expressive system results in improved user performance, then incentives would exist for the implementation and use of the more semantically expressive system (assuming performance improvements exceeded implementation costs). A more semantically expressive system, by definition, allows a better representation of the underlying reality than the less semantically expressive system. Therefore, users should be less likely to make mistakes because they can more easily refer back to the “reality” of the situation to verify that they are performing the task correctly. That is, users of more semantically expressive systems should be more accurate in task performance than users of less semantically expressive systems. As a result, it is important to show that a system perceived as more expressive than another results in some benefit, such as higher quality task performance; however, factors other than the system itself must be considered in such an analysis. Prior accounting and information systems research has demonstrated that user characteristics, task characteristics, and system or environmental characteristics all contribute to task accuracy (e.g., Libby and Luft 1993; Jih et al., 1989). Accounting knowledge and
81
field independence are two user characteristics that have been demonstrated to affect task accuracy using the REA accounting model (Dunn, 1994; Dunn and Grabski, 1998). Prior literature also demonstrated that one of the most important considerations with regard to tasks is whether or not the problem representation matches the task to be performed. Vessey (1991) introduced the model of cognitive fit, which occurs if the problem representation fits the task well, that is, if the problem solving steps required for a task are supported by the format of the problem representation. Dunn and Grabski (1997) found that cognitive fit affects accuracy of users of the two alternative accounting models in this study. The research model in Fig. 1 is thus proposed, and Hypothesis 2 is derived from this model. H2: The more users perceive an AIS to be semantically expressive, the greater will be their task accuracy, controlling for cognitive fit and user characteristics of accounting ability and field dependence. 3. Method 3.1. Overview A laboratory experiment was used to test the hypotheses. Participants were 73 students enrolled in an undergraduate AIS course. These students had previously completed an average of six undergraduate accounting courses. Demographics are summarized in Table 1. A within-subjects design was used for task accuracy and perceived semantic expressiveness, between-subjects for cognitive fit, and user characteris-
Fig. 1. Research model for Hypothesis 2.
82
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
Table 1 Descriptive statistics, mean (standard deviation)
Overall GPA Age Intermediate accounting grade* Number of accounting classes taken Gender Male Female
Field Dependent, N ⫽ 35
Field Independent, N ⫽ 38
Total, N ⫽ 73
3.13 (.47) 23.31 (5.67) 2.62 (.93) (N ⫽ 33)** 6.03 (2.50)
3.18 (.50) 22.16 (2.61) 3.08 (.67) 5.84 (1.91)
3.16 (.48) 22.72 (4.39) 2.87 (.85) (N ⫽ 71) 5.93 (2.20)
21 14
17 21
38 35
* Significantly different at p less than .009. ** Intermediate accounting grades were not available for two participants, both field dependent, who had taken intermediate accounting at different universities.
tics were controlled as covariates. Participants used both the REA and DCA-based accounting systems and were each given two tasks to complete with the REA-based accounting system and two tasks to complete with the DCA-based accounting system. Each participant completed all four tasks. All tasks asked the participants to explain how they would obtain some information from the system with which they were provided. The Appendix lists the four information retrieval tasks the participants were asked to complete. Participants were randomly assigned as to which system they used first (REA or DCA) and as to which tasks they completed first. The tasks were completed as an in-class exercise worth 3 percent of each participant’s course grade and the participants understood that the number of points they earned on the exercise depended on their accuracy. 3.2. Alternative accounting systems’ documentation The REA-based system documentation consisted of REA diagrams in entity-relationship format and the corresponding relational table structures. The REA diagrams serve as the system’s abstraction from reality; the relational table structures provide the implementation of that abstraction. The DCA-based system documentation consisted of a chart of accounts, sample journals (general and special), and sample ledger accounts (general and subsidiary). The chart of accounts serves as the system’s abstraction from reality1; the journals and ledgers provide the
1 The notion of a chart of accounts providing a description of the underlying reality is evidenced by the new company setup features of many general ledger software packages. Often one of the first questions asked is something like, “What type of business does the company engage in?” The suggested chart of accounts provided (which may be modified) is different for different types of businesses. Thus, if
implementation of that abstraction. Neither set of system documentation included any type of system flowchart or data flow diagram describing how processing occurs within the system. The models and the system documentation based on those models described only the architecture of the data storage mechanism. The purpose of a document/system flowchart is to describe the information processing steps, and has nothing to do with information storage structure. Because participants were asked to identify how they would obtain some information from within the accounting system and not to perform information processing steps, flowcharts, data flow diagrams, and other processing documentation were not provided, nor were they necessary to complete the tasks. Participants were provided with introductory background information about the organization so they could place the DCA and REA system documentation into context. In addition, both sets of system documentation included sample source documents (e.g., sales invoice, purchase order, receiving report, etc.) to clarify the types of activities in which the experimental company was involved. Although some would debate whether source documents are part of a system or whether they are outside a system, participants were asked to use the source documents as if they were part of both systems (the source documents could be viewed as either paper-based in a manual system or they could be input screens in a computer-based system). Familiarization questions were given for each system prior to the experimental tasks; those familiarization questions required particthe business is a funeral parlor, the suggested chart of accounts will include different accounts than if the business is a distributor. Similarly, the Encyclopedia of Accounting Systems (Pescow, 1976) provides various sets of charts of accounts and special and general journals for use with different organization types.
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
ipants to locate items in the sample source documents as well as locating items in the REA diagrams, relational tables, chart of accounts, journals, and ledgers. Each of the four experimental tasks required participants to explain how they would retrieve certain information from their system based on the system documentation (including the source documents). 3.3. Perceived semantic expressiveness Perceived semantic expressiveness was measured as the participants’ response to a seven-point Likert scale question: “The documentation I received provided me with a realistic representation of the accounting information flows of the business” (agree ⫽ 1, disagree ⫽ 7). This question was answered after completing two information retrieval tasks with respect to their use of one of the accounting system’s documentation; then it was answered again after completing the other two tasks with respect to their use of the other system’s documentation. Although students were assured of confidentiality from their instructor as to their response to this question, a demand effect in the answers to this question is possible; however, if participants did not answer according to their actual perception we would not expect to find support for H2, that the perceived semantic expressiveness contributes to task accuracy. Answers were reverse-coded in the data analysis so that higher numbers corresponded to greater perceived semantic expressiveness. 3.4. Cognitive fit Two of the tasks were designed to be easier to complete using a traditional DCA-based system than with a REA-based system. One asked how to determine the company’s accounts receivable balance as of the end of a month; the other asked how to determine the company’s total current liabilities as of the end of a month. The chart of accounts contains accounts receivable and also contains classifications of all liability accounts as either current or non-current. In the REA-based system, multiple diagrams must be consulted and the user must understand how the economic events and the business processes fit together in order to explain how these balances could be calculated. The other two tasks were designed to be easier to complete using a REA-based system than using a traditional DCA-based system. One asked how to determine the company’s gross sales by inventory stock number. This information is not stored at the appropriate level of aggregation in a traditional DCAbased system; users must examine the source documents (sales invoices). In the REA-based system this
83
information is available in the tables that represent sales, inventory, and the relationship between them. The second REA-based task asked how to determine whether the company makes partial payments on its general and administrative service acquisitions. This can be answered easily by looking at the structural constraints on the REA diagram for the general and administrative service acquisition business process, whereas in the DCA system the information could only be obtained by comparing the source documents (general and administrative service orders and the corresponding checks). Cognitive fit was operationalized as a betweensubjects variable. Half of the participants were randomly assigned to complete the REA facilitated tasks using the REA-based system documentation and to complete the DCA facilitated tasks using the DCAbased system documentation. These participants thus were classified as the cognitive fit group. The other half of the participants were randomly assigned to complete the REA facilitated tasks using the DCAbased system and to complete the DCA facilitated tasks using the REA-based system. These participants were classified as the group that did not have cognitive fit. 3.5. User characteristics Two user characteristics were controlled for in this study, accounting knowledge and field dependence. Accounting knowledge was operationalized as the participants’ grades in a junior-level financial accounting and reporting course. Knowledge of accounting concepts should affect accuracy on information retrieval tasks using a DCA-based system. Dunn (1994) found accounting knowledge to be a significant determinant of information retrieval task accuracy using a REA-based system. The intermediate accounting course grade was determined to be the most adequate available surrogate for accounting knowledge. Several participants had taken introductory accounting at other colleges or universities, making the use of grades in those other courses, or overall accounting grade point average less comparable. Field dependence was included in the research model based on Dunn and Grabski (1998), who found that field-independent undergraduate students performed significantly better on conceptual modeling design problems than field-dependent undergraduate students. Field dependence was measured using the Group Embedded Figures Test (GEFT) developed by Witkin et al. (1971). Students were classified as field dependent if they scored twelve or less (n ⫽ 35), and were considered to be field independent if they scored thirteen or greater (n ⫽ 38). This catego-
84
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
rization is based on Witkin et al. (1971) and adjusted to include business school females’ performance equal with males’ based on DeSanctis and Dunikoski (1983). The GEFT was administered on the first day of class. 3.6. Accuracy Accuracy was measured by independent verification of the processes specified by the participants to obtain the needed information for each task. A model solution was developed and each participant’s solution was evaluated relative to the model solution by two independent graders. The scores were then compared and any differences reconciled. Intergrader reliability before reconciliation of differences as indicated by Cronbach’s alpha was .834. 4. Statistical tests and results The mean value for perceived semantic expressiveness (higher ⫽ more semantically expressive) for REA system users was 5.329 (S.D. 1.259). The mean value for perceived semantic expressiveness for DCA system users was 4.164 (S.D. 1.581). A paired samples t-test reveals the difference between these mean values is significant, with t ⫽ 4.591, p ⬍ .000. There was no order effect, that is, it did not matter
which system documentation the participant used first. Users of the REA system documentation perceived the system to be significantly more semantically expressive than did users of the DCA system documentation, therefore H1 is supported. To alleviate concern about alternative explanations associated with test results for H1, further analysis was performed. One concern is that by including tasks for which DCA system users must refer to source documents, DCA is unfairly disadvantaged. Therefore a second t-test was run to compare the perceived semantic expressiveness of users for the REA and DCA models for only the DCA-facilitated tasks. Average perceived semantic expressiveness for REA users completing DCA-facilitated tasks was 5.11 (S.D. 1.26). The mean value of perceived semantic expressiveness for DCA users completing DCAfacilitated tasks was 4.19 (S.D. 1.70). REA users perceived the system as significantly more semantically expressive (t ⫽ 2.671, p ⬍ .011) than did DCA users, even for the DCA-facilitated tasks. Table 2 presents the mean perceived semantic expressiveness broken down by cognitive fit categories, system documentation, and field dependence of users. This breakdown illustrates what is shown by the two t-tests—that differences in perceived semantic expressiveness in this study are associated with the
Table 2 Perceived semantic expressiveness, cell means (standard deviations)
REA documentation users Consistent tasks (cognitive fit) Inconsistent taks (no cognitive fit) Total DCA documentation users Consistent tasks (cognitive fit) Inconsistent tasks (no cognitive fit) Total Grand mean for REA and DCA users
Field Independent Users
Field Dependent Users
Total
5.70 (1.174) 4.944 (1.211) 5.342 (1.236)
5.353 (1.320) 5.278 (1.319) 5.314 (1.301)
5.540 (1.238) 5.111b (1.260) 5.329a (1.259)
4.250 (1.618) 4.167 (1.383) 4.210 (1.492) 4.776d (1.475)
4.118 (1.833) 4.111 (1.605) 4.114 (1.694) 4.714d (1.616)
4.189b (1.697) 4.139 (1.476) 4.164a (1.581) 4.745 (1.539)
Grand Mean (standard deviation) for Consistent Tasks ⫽ 4.865c (1.625). Grand Mean (standard deviation) for Inconsistent Tasks ⫽ 4.625c (1.448). a Significantly different at p ⬍ .000. b Significantly different at p ⬍ .008. c Not significantly different, p ⬍ .809. d Not significantly different, p ⬍ .413.
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
two different accounting systems. It also reveals that the grand mean for all tasks for which users had cognitive fit (4.86) is not significantly different from the grand mean for all tasks for which users did not have cognitive fit (4.63) (t ⫽ 0.24, p ⬍ .81). Similarly, the grand mean for field dependent users (4.71) and the grand mean for field independent users (4.78) are not significantly different (t ⫽ 0.82, p ⬍ .41).2 The second hypothesis, which relates semantic expressiveness to task performance (accuracy), was tested using regression, with two observations for each participant. The regression equation tested is: Accuracy ⫽ ␣ ⫹ o Perceived Semantic Expressiveness ⫹ 1Cognitive Fit ⫹ 2Accounting Knowledge ⫹ 3Field Independence ⫹ ε Table 3 shows the model summary, coefficients, and correlations. Accuracy is significantly correlated with all variables (the only other significant correlation is between field dependence and accounting knowledge). Results indicate support for the research model, with adjusted R-square of .26. All VIF factors were between 1.020 and 1.138 indicating a lack of multicollinearity within the model tested. Standardized Beta weights reveal that cognitive fit is the most significant determinant of accuracy (Beta ⫽ .428, p ⬍ .0001). Perceived semantic expressiveness (Beta ⫽ .159, p ⬍ .031) and accounting knowledge (Beta ⫽ .174, p ⬍ .025) are also significant determinants of accuracy. Field independence (Beta ⫽ .139, p ⬍ .075) is slightly less important in determining accuracy. Overall, H2 is supported, that higher perceived semantic expressiveness is correlated with improved accuracy. It is important to note that H2 and its results do not claim simply that REA users are more accurate than DCA users. Rather, they suggest that if DCA users perceive the system to be more semantically expressive than REA, they will be more accurate with the DCA system than with the REA system. Likewise if REA users perceive the system to be more semantically expressive than DCA, they will be more accurate with the REA system than with the DCA system.
2 A repeated measures ANOVA found a main effect for semantic expressiveness (p ⬍ .001) and an interaction between semantic expressiveness and field dependence (p ⬍ .029). No effect was found for semantic expressiveness by cognitive fit (p ⬍ .516) nor was there any effect associated with semantic expressiveness by cognitive fit by field dependence (p ⬍ .343).
85
5. Discussion Results of this study indicate that users perceive the REA accounting model to be more semantically expressive than the DCA accounting model, thus providing evidence to support McCarthy’s (1982) claim that REA is more semantically expressive. Results of this study also indicate that perceived semantic expressiveness of accounting information systems is important to users. The Beta weights indicate that semantic expressiveness is as important as accounting knowledge in determining task accuracy. These results were obtained with undergraduate accounting majors, who were trained to understand the DCA accounting model (having had an average of six accounting courses) and were thus familiar with the semantic meaning inherent in the chart of accounts coding and organization. These students had less than a semester of training with the REA accounting model. With users who are more familiar with the REA accounting model and/or who are not as familiar with the DCA accounting model, results may be even more pronounced. Future research should be conducted to gain more insight into the generalizability of this study’s results to other types of users, particularly non-student users. The finding in this study of cognitive fit as a significant determinant of accuracy is consistent with Dunn and Grabski (1997). The finding of accounting knowledge as a significant determinant of accuracy for an information retrieval task is consistent with Dunn (1994). The finding of field independence as marginally significant is not consistent with Dunn and Grabski (1998); however, that study examined accuracy with respect to design tasks as opposed to information retrieval tasks. It is possible that field independence is more important for design tasks than for information retrieval tasks. This is a question for future research. An important implication of this study is that evidence suggests perceived semantic expressiveness is a pervasive characteristic of an information system; that is, it will not be affected by a user’s characteristics or by cognitive fit. This evidence, combined with the association of perceived semantic expressiveness with increased task accuracy, indicates that system designers should consider the likely perceived semantic expressiveness of systems as they are developing them. They should also consider cognitive fit, as it has a relatively stronger association with task accuracy; however, cognitive fit will vary from task to task. Thus it may be more meaningful to focus on developing systems that represent real world phenomena as closely as possible and to provide facilities for users to construct data representations that match the task they desire to accomplish.
86
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87
Table 3 Regression of accuracy on perceived semantic expressiveness, cognitive fit, accounting knowledge, and field independence Correlations Pearson Correlation Accuracy Field dependence Acct. Knowledge Sem. Express. Cognitive fit Significance (1 tailed) Accuracy Field dependence Acct. Knowledge Sem. Express. Cognitive fit
Accuracy
Field Dependence
Accounting Knowledge
Semantic Expressiveness
Cognitive fit
1.000 .216 .184 .185 .445
1.000 .322 ⫺.085 .081
1.000 ⫺.039 ⫺.066
1.000 .103
1.000
.005 .014 .014 .000
.000 .158 .169
.323 .216
.111
ANOVA
Sum of Squares
df
Mean Square
F
Sig.
Regression Residual Total
46104.798 117389.9 163494.7
4 137 141
11526.199 856.861
13.452
.000
Unstandardized Coefficients
Standardized Coefficients
Coefficients
Beta
Std. Error
Beta
t
Sig.
(Constant) Semantic expressiveness Cognitive fit Accounting knowledge Field dependence
⫺15.434 3.519 29.082 8.252 1.147
13.988 1.613 4.984 3.646 .639
.159 .428 .174 .139
⫺1.103 2.182 5.835 2.263 1.797
.272 .031 .000 .025 .075
Adjusted R Square ⫽ .261.
This study is the first to examine perceived semantic expressiveness of accounting systems, and while the evidence suggests perceived semantic expressiveness is greater in REA systems than in DCA systems and that greater perceived semantic expressiveness is associated with greater task accuracy, no definitive conclusions can be drawn. Future research is required to develop better measures for perceived semantic expressiveness and to determine whether results would differ for other types of users (especially non-accounting and non-student users) and for other types of tasks (other than information retrieval). Future research can also examine whether other benefits result from systems perceived as more semantically expressive than others (such as increased user satisfaction and increased system use). Appendix 1 Information retrieval tasks How would you determine what accounts receivable balance RSW had as of April 30, 1997?
How would you determine gross sales (sales before returns, allowances, and discounts) by inventory stock number for the period January 1—April 30, 1997? (RSW’s fiscal year begins January 1 and ends December 31.) How would you determine RSW’s total current liabilities as of June 30, 1997? How would you determine whether RSW makes partial payments for general and administrative services they acquire?
References Abrial JR . Data semantics. In: Klimbie JW, Koffeman, KL, editors. Data base management. Amsterdam: North Holland Publishing Company; New York: American Elsevier, 1974. p. 1–60. Andros DP, Cherrington JO, Denna EL. Reengineer your accounting, the IBM way. Financ Exec 1992;8:28–31. Brodie ML. On the development of data models. In: Brodie ML, Mylopolous J, Schmidt JW, editors. On conceptual modeling. New York: Springer-Verlag, 1984. p. 19–47. David JS, Dunn CL, Poston R, McCarthy, WE. The re-
C.L. Dunn, S.V. Grabski / International Journal of Accounting Information Systems 1 (2000) 79–87 search pyramid: A framework for accounting information systems research. J Inform Syst 1999;13:7–30. DeSanctis G, Dunikoski, R. Group Embedded-Figures Test: Psychometric data for a sample of business students. Perceptual and Motor Skills 1983;56:707–10. Dunn CL. An investigation of abstraction in events-based accounting systems. Unpublished doctoral dissertation, Michigan State University, 1994. Dunn CL, Grabski, SV. Cognitive fit and a comparison of two accounting models. Paper presented to the American Accounting Association national meeting, Dallas, TX. August, 1997. Dunn CL, Grabski, SV. The effect of field dependence on conceptual modeling performance. Adv Account Inform Syst 1998;6:65–77. Dunn CL, McCarthy WE. The REA accounting model: Intellectual heritage and prospects for progress. J Inform Syst 1997;11:31–51. Geerts GL, McCarthy WE. The economic and strategic structure of REA accounting systems. Paper presented to the 300th Anniversary Program, Martin Luther University, Halle-Wittenberg, Germany. September 1994. Grabski SV, Marsh R. Integrating accounting and manufacturing information systems: An ABC and REA-based approach. J Inform Syst 1994;8:61–80. Hammer M, McLeod D. Database description with SDM: A
87
semantic database model. ACM Trans Database Syst 1981;6:351–86. Hollander AS, Denna EL, Cherrington JO. Accounting, information technology, and business solutions. Chicago: Irwin, 1996. Jih WJK, Bradbard DA, Snyder CA, Thompson NGA. The effects of relational and entity-relationship data models on query performance of end users. Int J Man-Mach Stud 1989;31:257–67. Libby R, Luft J. Determinants of judgment performance in accounting settings: Ability, knowledge, motivation, and environment. Account Org Soc 1993;18:425–50. McCarthy WE. The REA accounting model: A generalized framework for accounting systems in a shared data environment. Account Rev 1982;57:554–78. Pescow JK. Encyclopedia of accounting systems 2nd Ed. Englewood Cliffs, NJ: Prentice-Hall, 1976. Romney MB, Steinbart PJ, Cushing BE. 1997. Accounting information systems (Seventh edition). Reading: Addison-Wesley, 1997. Vessey I. Cognitive fit: A theory-based analysis of the graph versus tables literature. Decision Sciences 1991;22:219–40. Witkin HA, Oltman PK, Raskin E, Kar, SA. A manual for the Embedded Figures Test. Palo Alto, CA: Consulting Psychologists Press, 1971.