An investigation of the factors influencing cost system functionality using decision trees, support vector machines and logistic regression

• Published date: 04 March 2019
• Date: 04 March 2019
• Pages: 27-55
• DOI: https://doi.org/10.1108/IJAIM-04-2017-0052
• Author: Cemil Kuzey,Ali Uyar,Dursun Delen
• Subject Matter: Accounting & Finance,Accounting/accountancy,Accounting methods/systems

An investigation of the factors

inf‌luencing cost system

functionality using decision trees,

support vector machines and

logistic regression

Cemil Kuzey

Arthur J. Bauernfeind College of Business, Murray State University,

Murray, Kentucky, USA

Ali Uyar

La Rochelle Business School, Excelia Group, La Rochelle, France, and

Dursun Delen

Department of Management Science and Information Systems,

Spears School of Business, Oklahoma State University, Stillwater, Oklahoma, USA

Abstract

Purpose –The paper aims to identify and critically analyze the factors inf‌luencing cost system

functionality (CSF) using several machine learning techniques including decision trees, support vector

machinesand logistic regression.

Design/methodology/approach –The study used a self-administered survey method to collect the

necessary data from companiesconducting business in Turkey. Several prediction models are developedand

tested; a series of sensitivity analysesis performed on the developed prediction models to assess the ranked

importanceof factors/variables.

Findings –Certain factors/variables inf‌luence CSF much more than others. The f‌indings of the study

suggest that utilization of management accounting practices require a functional cost system, which

is supported by a comprehensive cost data management process (i.e. acquisition, storage and

utilization).

Research limitations/implications –The underlying data were collected using a questionnaire

survey; thus, it is subjectivewhich ref‌lects the perceptions of the respondents. Ideally, it is expectedto ref‌lect

the objective of the practicesof the f‌irms. Second, the authors have measured CSF it on a “Yes”or “No”basis

which does notallow survey respondents reply in between them; thus,it might have limited the choices of the

respondents.Third, the Likert scales adopted in the measurement of the otherconstructs might be limiting the

answers of the respondents.

Practical implications –Information technology plays a very important role for the success of CSF

practices. That is, successfulimplementation of a functional cost system relies heavily on a fully integrated

informationinfrastructure capable of constantlyfeeding CSF with accurate, relevant and timelydata.

Originality/value –In addition to providing evidence regarding the factors underlying CSF based on a

broad range of industries interesting f‌inding, this study also illustrates the viability of machine learning

methodsas a research framework to critically analyze domainspecif‌ic data.

Keywords Sensitivity analysis, Machine learning, Support vector machines, Predictive analytics,

Decision trees, Cost system functionality

Paper type Research paper

Cost system

functionality

27

Received28 April 2017

Revised18 July 2017

Accepted15 August 2017

InternationalJournal of

Accounting& Information

Management

Vol.27 No. 1, 2019

pp. 27-55

© Emerald Publishing Limited

1834-7649

DOI 10.1108/IJAIM-04-2017-0052

The current issue and full text archive of this journal is available on Emerald Insight at:

www.emeraldinsight.com/1834-7649.htm

1. Introduction

Increasing competition –local, regionaland/or global –and decreasing prof‌it margins (due

largely to competitive product/service pricing) are forcing f‌irms to use more accurate

product costing/pricing methods and to implement more sophisticated management

accounting practices (MAPs) (Chanand Suk-Yee Lee, 2003;Cardinaels et al.,2004;Uyar and

Kuzey, 2016a;Uyar and Kuzey, 2016b).To succeed, these implementations need to be based

on a carefully designed, domain specif‌ic, comprehensive (i.e. inclusive of all relevant

attributes) functional cost system within the organization. In the literature, the most

commonly cited attributes of a functionalcost system include the level of detail information

provided, the better classif‌ication of costs according to behavior, the frequency of cost

information reported, the accuracy of cost data and the extent of variances calculated

(Pizzini, 2006;Pavlatos and Paggios, 2009). Effective and comprehensive design and

execution of the cost system is a criticalinput to the management of business organizations

and, if done properly, can lead to accurate and timely managerial decisions and enhanced

f‌irm performance (Copperand Kaplan, 1992;Foong and Teruki, 2009). Indeed, a recent study

proved that internal and externalstakeholders’information needs affects cost systemdesign

in public organizations(Schoute and Budding, 2017).

Most of the earlier studies have used a contingency-based approach in analyzing MAPs

across various industriesand countries (Abdel-Kader and Luther, 2008;Cadez and Guilding,

2008;King et al., 2010;Albuand Albu, 2012;Christ and Burritt, 2013;Al-Sayed and Dugdale,

2016;Otley, 2016). They mainlytried to identify the contingent factors that are likely to have

a signif‌icant impact on MAPs used by the organizations. According to contingency theory,

an appropriate cost system relies heavily on the organization’s specif‌ic circumstances,

which should be adaptive to the changing circumstances (Pavlatos and Paggios, 2009). In

other words, there is no unique cost system design that f‌its uniformly toall organizations or

to the same organization all the time.While the scope of a cost system is rather small/limited

for one organization, it might be quite large/comprehensive for another. For example, for a

small f‌irm, establishing an overly comprehensive functional cost system might be

unnecessary, prohibitively costly and overly luxurious. Similarly, if a f‌irm does not use

sophisticated MAPs, it may not need an extensive cost system. The wealth of f‌inancial

sources and the suff‌iciency of other capabilities of the f‌irm, as well as the organizational

needs shape the size and structureof the cost system functionality (CSF).

This study extends the relevant literature in severaldimensions. The previous literature

showed that CSF studies have usually been in unique settings. For instance, while

Cardinaels et al. (2004) and Pizzini (2006) investigated the determinants and the relevance

and usefulness of cost data in the health-care industry, Pavlatos and Paggios (2009) carried

out a study regarding the impact of contingent factors on the design of cost systems and

functionality within the hospitality industry. A recent study investigated how information

needs of decision makers impact cost system design in local municipalities (Schoute and

Budding, 2017). Studies like theseused a homogenous sample of data coming from a single-

industry, which has the potential of limiting the ability to generalize the results to a cross-

industry setting (Pizzini, 2006). To differentiate and further the extent literature, our study

used data collected from several industries to capture the factors and theirinteractions in a

cross-industry context. In addition, these prior studies recommended future research

directions to include additional variables (Pavlatos and Paggios, 2009;Pavlatos, 2012), and

to carry out similar study on a wider sample(Cohen and Kaimenaki, 2011); hence, this study

aimed to fulf‌ill that purpose by includingand operationalizing eight additional variables on

a wider sample. Finally, thisstudy used a machine learning-based analytic approach,which

is different than the traditional regression-based analysis used in most of the previous

IJAIM

27,1

28