Top Banner
Available online at http://www.idealibrary.com on doi: 10.1006/mare.2001.0174 Management Accounting Research, 2002, 13, 41–69 Demonstrating the effect of the strategic dialogue: participation in designing the management control system 1 Marco de Haas*and Jen A. Algera* Stimulation of goal congruent behaviors through the coordinated allocation of human resources—i.e. time, energy and attention—requires the mental models of organizational actors to be convergent. The Strategic Dialogue is advocated for the purpose of convergent mental modeling. In relation to the need for convergence, the concept of Goal Coherence is introduced, which is measured as the amount of consensus on goal priorities within and, moreover, between goal interdependent groups. The relation between Strategic Dialogue and Goal Coherence represents the instrumental hypothesis of the research. Support for this hypothesis is empirically demonstrated in a two-group pre-test/post-test design, using a specific measure of association. This design has been applied twice in a practical case study on the participatory designing of a management control system. c 2002 Elsevier Science Ltd. All rights reserved. Key words: strategic dialogue; goal coherence; performance measurement; mental model; human behavior; human resources; participation; consensus; intersubjectivity; association. 1. Introduction In this paper, the organization is primarily approached as a social system of human elements. This approach has a fundamental repercussion for the process of management. The emphasis on the human factor of organization takes in the 1 The first draft of this paper was presented at the 4th Int. Seminar on Manufacturing Accounting Research in Kolding (Denmark), June 1999. *Human Performance Management Group, Department of Technology Management (PAV-U13), Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands. E-mail: [email protected] Received 1 December 1999; accepted 4 October 2001. 1044–5005/02/$ - see front matter c 2002 Elsevier Science Ltd. All rights reserved.
29

(2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Nov 28, 2015

Download

Documents

Ennayojarav
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Available online at http://www.idealibrary.com ondoi: 10.1006/mare.2001.0174Management Accounting Research, 2002, 13, 41–69

Demonstrating the effect of the strategicdialogue: participation in designing themanagement control system1

Marco de Haas*† and Jen A. Algera*

Stimulation of goal congruent behaviors through the coordinated allocation of humanresources—i.e. time, energy and attention—requires the mental models of organizationalactors to be convergent. The Strategic Dialogue is advocated for the purpose ofconvergent mental modeling. In relation to the need for convergence, the concept ofGoal Coherence is introduced, which is measured as the amount of consensus ongoal priorities within and, moreover, between goal interdependent groups. The relationbetween Strategic Dialogue and Goal Coherence represents the instrumental hypothesisof the research. Support for this hypothesis is empirically demonstrated in a two-grouppre-test/post-test design, using a specific measure of association. This design has beenapplied twice in a practical case study on the participatory designing of a managementcontrol system.

c© 2002 Elsevier Science Ltd. All rights reserved.

Key words: strategic dialogue; goal coherence; performance measurement; mental model;human behavior; human resources; participation; consensus; intersubjectivity; association.

1. Introduction

In this paper, the organization is primarily approached as a social system ofhuman elements. This approach has a fundamental repercussion for the processof management. The emphasis on the human factor of organization takes in the

1The first draft of this paper was presented at the 4th Int. Seminar on Manufacturing Accounting Researchin Kolding (Denmark), June 1999.

*Human Performance Management Group, Department of Technology Management (PAV-U13),Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands.†E-mail: [email protected]

Received 1 December 1999; accepted 4 October 2001.

1044–5005/02/$ - see front matter c© 2002 Elsevier Science Ltd. All rights reserved.

Page 2: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

42 M. de Haas and J. A. Algera

simple recognition that ‘the only way organizational goals are going to be obtainedis through the behavior of organizational actors’ (Steers, 1977).

Human behavior is based on a construction of reality in the human mind (Vennix,1996). There is convincing scientific evidence (e.g. Neisser, 1967) that the humanmind actively constructs external reality rather than passively stores and recallsinformation, which is received from the environment through the senses. The activeand deliberate construction of reality takes place through processes of selectiveperception and selective recollection. Stated differently, the human mind is biased ininformation selection and recollection of past events. The construction of the externalreality in the human mind is referred to as a mental model. These models are dynamicin nature and will change over time due to newly gained insights in the functioningof reality.

A mental model is an individual’s cognitive representation of a system (e.g. anorganization) and the individual’s interaction with the system (i.e. behavior), withparticular focus on how the individual’s interaction with the system causes outcomesof interest (i.e. goal attainment). The notion of belief in causality is of major interest:beliefs represent the basic conceptual building blocks of mental models (Hinsz, 1995).In short, people build mental models of their environment and in turn base theirbehavior on these mental models, thereby creating situations which are subsequentlyinterpreted as reality (Vennix, 1996).

Given the way human beings selectively process information, a holistic view ofreality is the exception rather than the rule. A number of cognitive limitations inducepeople to perceive and recollect selectively and thus to focus on the parts rather thanthe whole:

• Limited systems thinking capability: people have difficulty in identifying inter-connections and thinking in causal nets (Dörner, 1980);

• Limited information processing capacity or ‘bounded rationality’ (Simon, 1948):people tend to (un)consciously reduce complexity in order to prevent informationoverload and to reduce mental effort (Hogarth, 1987). Miller (1956) was one of thefirst to empirically demonstrate this phenomenon. He pointed out that in generalpeople can only hold seven (plus or minus two) pieces of information in theirshort-term memory;

• Limited span of attention: ‘. . . before information can be used by the deliberativemind, however, it must proceed through the bottleneck of attention—a serial, notparallel, process whose information capacity is exceedingly small’ (Simon, 1985).

Due to these cognitive limitations, people make mental models that are bydefinition incomplete. Since people tend to look for information which confirms theirview of the world rather than to look for evidence which might refute it (Hogarth,1987), the existing and incomplete mental models in turn feed the processes ofselective perception and recollection. In addition, and perhaps far more important,people make mental models that are idiosyncratic (Hinsz, 1995). People differ dueto differences in background, personality, experience, learning, etc., and will thusselect differently. As a consequence, individuals interpret reality in their own uniqueways. Everyday life thus presents itself to the individual as a subjective reality: thereis no question of one single and objective reality perceived similarly by multipleindividuals.

Page 3: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 43

The point we are trying to make is that there is ample opportunity for differenthuman beings to construct and maintain different mental models of the ‘same’ reality.What are the chances that people with idiosyncratic views of the world will select thesame chunks of information from their environment and subsequently construct thesame mental representation of the perceived reality? Quite likely, these chances arenegligible. Ickes and Gonzales (1994) use the term divergence to indicate the degreeof dissimilarity between mental models.

In this paper, we stress the above point from a management perspective. Inorganizations, managers and employees also have mental models. These modelscontain—if at all!—ideas and opinions regarding the relevance of local goals atthe individual and group level, the way these goals mutually relate and the waythese goals jointly contribute to the attainment of overall goals at the organizationallevel. Subsequently, the allocation of (scarce) human resources—i.e. time, energy andattention—to alternative courses of action is implicitly or explicitly decided uponduring the daily hassle of organizational life. All else being equal, the larger thediscrepancies between managers’ and/or employees’ mental models, the greaterthe lack of shared vision, the more the divergence in behavior and the higher thedispersion of organizational energy. According to Vennix (1996), these discrepancieswill impede the effective operation of the organization, because it will induce a lackof co-operation.

Assuming that management is about coordinating individual and group effortsto produce organized action, its purpose is thus to share and align multiple mentalmodels in order to foster concerted action. Rather than devoting one’s time to theconstruction of strategic plans, the focus must instead be on the process of changingmental models (Vennix, 1996). In the words of Checkland and Scholes (1990): ‘Whatis in short supply in organizations is an organized sharing of perceptions sufficientlyintense that concerted action gets taken corporately’. Striving for convergent orshared mental models, i.e. for an intersubjective perception of reality throughout theorganization, seems thus essential for organizational effectiveness.

2. Research question

In a previous issue of this journal, de Haas and Kleingeld (1999) proposed a practicalintervention for the purpose of aligning mental models. This intervention is calledthe Strategic Dialogue. The intervention prescribes an interactive and multilevelgoal setting process, producing shared goal definitions for goal interdependentgroups. During the course of the dialogue, groups acting at multiple organizationallevels—ranging from the macro or strategic level to the micro or operational level—participate in the designing of relevant content for their respective performancemeasurement systems.

According to systems theorists, the working of any system is explained by the co-working of its parts. We principally view an organization as a system, moreoveras a social system of human elements. At the group level, an organization thusrepresents a network of multiple groups.2 In light of the common goals the

2In turn, a group is a social system itself, since it represents a network of multiple individuals. In thispaper, however, the individual level is disregarded.

Page 4: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

44 M. de Haas and J. A. Algera

organizational system as a whole is pursuing, its parts—i.e. groups—are by definitiongoal interdependent. A goal interdependence relation between a principal and anagent is either vertical or horizontal in nature. Vertical goal interdependence existsbetween two groups acting at adjacent organizational levels. Think, for example, of amanagement team at the factory level in relation to an assembly team at the shopfloor level. Horizontal goal interdependence exists between two groups acting atadjacent stages of the organization’s goods or services flow. Think, for example, ofthe same assembly team in relation to the planning department. The notion of goalinterdependence has a fundamental repercussion for the Strategic Dialogue. Groupsnot only participate in their own performance measurement system design, but inthat of goal interdependent principal and/or agent groups as well.

Each group participating in the Strategic Dialogue explicates a means–end relationby answering two fundamental, interrelated questions:

1. As part of an organizational system, what should we as a group contribute to?

2. Given our stated contributions, how can we as a group achieve them?

For control purposes, the what-to-achieve question is translated into result-orientedperformance indicators (in short: result indicators) and the how-to-achieve ques-tion into process-oriented performance indicators (in short: process indicators). Con-cretely, a group goal next corresponds with a prioritized target for such a definedperformance indicator.

What to contribute to for each and every group ultimately is a function of thecommon goals the organizational system as a whole is pursuing. Therefore, the firstgroup that participates in the Strategic Dialogue is the (top) management team atthe macro level, invited to translate the business strategy into an overall means–end relation. Next, this principal group invites agent groups at the next lowerorganizational level to discuss the validity of the overall means–end relation. It isimportant to emphasize the invitation to bottom-up validation, since people’s mentalmodels will not change as a consequence of the top–down imposition of strategicinitiatives. After consensus has been reached, the agreed-upon means become therelevant ends for the agent groups, producing shared goals. Depending on thesteepness of the organizational hierarchy, these agent groups might in turn functionas principals themselves for one or more lower-level agent groups. The process ofinteractively deploying means–end relations and consequently producing sharedgoals, continues until operational groups at the micro level have explicated localmeans–end relations.

For a further explanation and illustration of the Strategic Dialogue, the reader isreferred to de Haas and Kleingeld (1999) and to de Haas (2000). The remainder of thispaper focuses on the first empirical findings that have resulted from action research,in an attempt to demonstrate the effect of the Strategic Dialogue.

Defined goal interdependence ‘on paper’—the tangible outcome of the StrategicDialogue—is only instrumental to perceived goal interdependence ‘in the mind’.To empirically demonstrate the effect of the Strategic Dialogue on the amountof convergence in mental modeling, we propose the concept of Goal Coherence.Goal Coherence has strong connotations with goal congruence, which is definedby Vancouver et al. (1994) as ‘the agreement among organizational employees onthe importance of the goals the organization could be pursuing’. Goal congruence,

Page 5: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 45

however, only regards the common goals for the organization as a whole. Itdisregards deployed goals at lower levels in the organization. From a workmotivation perspective, it is important to deploy overall goals into local goals in orderto make people accountable for results they can control and thus achieve (Van Tuijlet al., 1995). The available concept of goal congruence insufficiently covers the issueof varying degrees of controllability.

Therefore, we define Goal Coherence as ‘the degree of consensus on constituencygoal priorities’. Our definition is derived from the constituency level conceptsdefined by Vancouver et al. (1994). We prefer to speak of constituencies instead ofgroups from now on, to indicate that any group within the organization has a raisond’être at stake and therefore at least implicitly pursues certain goals. In analogy withVancouver et al. (1994), we distinguish the between-constituency Goal Coherenceconcept and the within-constituency Goal Coherence concept:

• Between-constituency Goal Coherence: ‘the degree of inter-consensus on sharedgoal priorities’ (i.e. consensus between two goal interdependent constituencies);

• Within-constituency Goal Coherence: ‘the degree of intra-consensus on sharedgoal priorities’ (i.e. consensus within one constituency).

Both concept equivalents are logically related in the sense that the degree of GoalCoherence between constituencies is a function of the degree of Goal Coherencewithin constituencies: there can be no between-constituency Goal Coherence if thewithin-constituency equivalent is lacking.

An empirical demonstration of consensus on goal priorities indicates that theunderlying mental models of organizational actors have become more convergent,i.e. similar. Hence, we use the concept of Goal Coherence to operationalize the degreeof congruent mental modeling. The relation between the intervention of StrategicDialogue and the degree of Goal Coherence describes the instrumental hypothesisof our research. The research question thus reads: ‘Does the intervention of StrategicDialogue positively affect degrees of Goal Coherence within and, moreover, betweengoal interdependent constituencies?’

3. Research design and method

To address the research question, we conducted a case study in the practical field. Thecase refers to a business unit of the Corus Corporation situated in The Netherlands,called Corus IJmuiden Long Products. Corus, a global manufacturer and supplier ofsteel and aluminum products, is the result of the recent merger between the BritishSteel Corporation and the Dutch Koninklijke Hoogovens Corporation. During thecase study, we invited goal interdependent constituencies to participate in a StrategicDialogue. This dialogue focussed on the contents of the performance measurementsystem for the business unit management team. Typically, such intervening inpractice for the purpose of doing research is called action research.

Two-group pre-test/post-test designParticipation of organizational actors in the Strategic Dialogue was organized in atwo-group pre-test/post-test design. The two-group pre-test/post-test design is anextended variant of Cook and Campbell’s (1979) one-group pre-test/post-test design.

Page 6: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

46 M. de Haas and J. A. Algera

Xp O1,pXp+a

O1,a

O2,p

O2,aXa

Figure 1. Two-group pre-test/post-test design.

This quasi-experimental design is one of the more frequently used research designsin the social sciences. It constitutes the recording of pre-test observations O1 on asingle group of individuals, who later receive a treatment X, after which post-testobservations O2 are made. In this research, the treatment concerns the interventionof Strategic Dialogue, in which organizational actors participate during design teammeetings and management approval meetings (see de Haas and Kleingeld, 1999).The pre-test and post-test observations relate to the empirical measurement of GoalCoherence before and after the dialogue.

We did not record observations on a single group, but on two vertically goalinterdependent groups at a time: a principal- and an agent-constituency. Therefore,we speak of a two-group pre-test/post-test design. This design is diagrammed inFigure 1, with Xp and Xa the design team meetings of the principal constituencyand the agent constituency respectively, O1,p and O1,a the pre-test measurementsof consensus on goal priorities among principal constituency members and amongagent constituency members respectively, Xp+a the management approval meetingsand O2,p and O2,a the post-test measurement of consensus on goal prioritiesamong principal constituency members and among agent constituency membersrespectively.

During design team meetings, the principal constituency and the agent con-stituency autonomously design the ingredients of a means–end relation (i.e. result-and process-oriented performance indicators, indicator targets and target priorities).The principal constituency does so first, since the identified means for the principalare to be deployed as the relevant ends for the vertically goal interdependent agent.Next, the agent constituency discusses the validity of the deployed ends and in turnidentifies relevant—and controllable!—means in relation to those ends. Thus far,there has been no interaction between the two constituencies; there has only beeninteraction within each.

Interaction between constituencies is facilitated during the management approvalmeeting. In this meeting, the agent constituency presents its means–end relationdesign, which the principal constituency has to approve of. To further explainthe difference between the design team meeting and the management approvalmeeting: the former is a within-constituency event whereas the latter is a between-constituency event. Especially the management approval meeting incorporates avital moment of constructive controversy (Tjosvold, 1985). Convergent mentalmodels within either constituency, developed during the preceding design teammeetings, might turn out to diverge between the two constituencies during thefollowing management approval meeting.

In the current paper, the two-group pre-test/post-test design at Corus IJmuidenLong Products is reported upon twice in the design of:

Page 7: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 47

XM O1,M

XM+Q

O1,Q

O2,M

O2,QXQ

Figure 2. M-Q design: treated ‘experimental’ group.

O1,M

XM+L

O1,L

O2,M

O2,LXL

XM

Figure 3. M-L design: untreated ‘control’ group.

1. the Management constituency (seven members) and the Quality constituency(eight members);

2. the Management constituency (again, seven members) and the Logisticsconstituency (seven members).

There is no specific relation between the two designs: they have been executedautonomously. The thing both designs have in common is the composition of theprincipal constituency: the same managers acted in the dialogue with the Qualityand Logistics constituencies. These two agent constituencies were cross-functionallycomposed of staff members from the business unit’s three plants (the Steelworks,the Rolling Mill and the Finishing Center) and of staff members from the businessunit’s Sales, Logistic Planning and Quality Assurance departments. The first design isshown in Figure 2, where M stands for Management constituency and Q for Qualityconstituency.

The design of Figure 2 turned out to represent a treated ‘experimental’ group. As aconsequence of a turbulent business environment, we were not able to apply a propertreatment in the second design. This design is shown in Figure 3, where M stands forManagement constituency and L for Logistics constituency. Since we were not able tofacilitate a proper management approval meeting, the design of Figure 3 comparedto the design of Figure 2 represents an untreated ‘control’ group.

Shortly after the merger, Corus’ Corporate Board of Directors decided negativelyon the continuity of the Corus IJmuiden Long Products business unit within thenew corporate strategy. In the past, this decision had been postponed several timesby the Corporate Board of the former Dutch Koninklijke Hoogovens corporation,but was to be made definitively by the new Corporate Board. At the time of themanagement approval meeting in the M-L design, it was two more weeks before thisdecision was made. Like the sword of Damocles, this decision was hanging over theheads of the organizational actors participating in the dialogue on strategic priorities,which naturally undermined its relevance. Due to this pseudo-management approvalmeeting, we speak of an untreated condition in the M-L design.

Page 8: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

48 M. de Haas and J. A. Algera

As a consequence of the different design conditions, we refer to an ‘experimental’group that received the treatment and a ‘control’ group that did not. The analogywith Cook and Campbell’s (1979) classification types is hyphenated though, sincebeforehand a deliberately designed experiment in the true senses of the word wasnever intended. The dramatic decision described above was an unforeseen eventduring the action research. Nonetheless, different design conditions are there and,thus, we will have to interpret possible effects from an ‘experimental’ perspective.We will stress this delicate point again in our concluding section.

Data collectionIn order to measure the amount of between-constituency and within-constituencyGoal Coherence in the M-Q and M-L designs, we asked constituency membersto prioritize their shared goal definitions. A simple scaling procedure enabled theprioritizing of shared goal definitions. Such a procedure constitutes an ordinal levelof measurement and consequently produces categorical (i.e. ranking) data. Part ofthe scaling instruction was that constituency members had to do the ranking exerciseindividually, i.e. the ranking was not part of the interactive group processes duringthe design team meetings and the management approval meetings. The reason forthis was to avoid the possibility of mock consensus. Given the pre-test/post-testfeature of the research design, goal interdependent constituency members had toorder the shared goal definitions twice: before and after the between-constituencyinteraction of the management approval meeting.

The ranking exercise is an integral part of the Strategic Dialogue, since it aimsat the clarification of multiple, seemingly conflicting goals. Multiple goals shouldnot be thought of as being in conflict, but rather that they are all important for thesuccess of the organization: the real issue is to apply organizational resources tomeet multiple goals in proportion to their importance. Even more so, the applicationof organizational resources and, thus, the prioritizing of shared goals, should becoordinated between goal interdependent constituencies in order to avoid optimizedperformance of local parts to result in sub-optimized performance of the integralsystem.

Data analysis: categorical principal components analysisA respondent’s ranking data in fact makes explicit a simplified, one-dimensionalmental model: the prioritized goals can be thought of as being positioned equidis-tantly along a straight line. The degree to which these one-dimensional mentalmodels are similar over multiple respondents expresses the degree of consensus ongoal priorities. To quantify and visualize consensus, we used a categorical principalcomponents analysis, as implemented in the program CATPCA in SPSS Categories10.0 (Meulman and Heiser, 1999). CATPCA stands for Categorical Principal Com-ponents Analysis with optimal scaling. A few words are devoted to this techniquehere. For further background and references, the interested reader is referred to deHaas (2000).

The CATPCA technique can be thought of as a method of dimension reduction: itsimultaneously quantifies categorical (i.e. qualitative) variables, while reducing thedimensionality of the data with minimal loss of the information contained in theoriginal variables—i.e. with minimal loss of variance accounted for (VAF). Stateddifferently, the technique reduces a set of categorical variables into a smaller set of

Page 9: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 49

uncorrelated principal components. For an uncomplicated example, the reader isreferred to the Appendix.

By reducing the dimensionality of the data, interpretation can be restricted toa few principal components rather than a large number of variables. Principalcomponents, common sources of variance and reduced dimensions are thus termswith a similar meaning. Ultimately, dimension reduction reveals the dimensions thatrepresent the major sources of variance in the original data. As said, dimensionreduction reveals the common sources of variance that the categorical variablesshare. The transformation of the original, categorical variables into metric variablesis underpinned by monotonically increasing transformation functions.

The principal component loadings are correlations between the variables andthe principal components, and they give coordinates to graphically represent thevariables as vectors in the principal component space. The squared principalcomponent loading gives the VAF by each dimension; the summation of squaredloadings over variables gives the total VAF, which is equal to the eigenvalueassociated with each principal component. The eigenvalues are related to Cronbach’sα, where α = m(λ − 1)/(m − 1)λ, with λ the eigenvalue and m the number ofvariables (Heiser and Meulman, 1994). Optimal scaling in CATPCA implies that thetotal proportion of VAF is as large as possible (given the ordinal information), andthus that Cronbach’s α is maximized (Heiser and Meulman, 1994).

In a traditional principal component analysis, the subjects (here, constituencymembers) are considered to be the units and the options (here, shared goaldefinitions) the variables which order or classify the units. In the analysis of rankingdata, where the respondents have ordered the options, the reversed data matrixshould be analyzed, with subjects as variables and options as units (Cronbach andGleser, 1953). In the case of such a subject-oriented multivariate analysis, resultingmeasures of correlation are interpreted as measures of association, quantifying theamount of intersubjective perception among respondents. Intersubjective measuresare commonly applied in social science research due to a lack of objective measures.In particular, Q-methodology (Stephenson, 1953) has evolved as a science of thesubjective. A common measure of association is represented by Cohen’s Kappa(Cohen, 1960). The application of CATPCA in our study produces a measure ofassociation that operationalizes the Goal Coherence concept.

4. Empirical findings

InterventionDuring design team meetings of the Management constituency, strategic considera-tions regarding business success in the Automotive market for Forging Steel productswere made explicit in terms of an overall means–end relation. The overall ends—i.e.the organizational goals—were identified in terms of Profitability and Growth, asoperationalized by the six Strategic Result Indicators of Table 1.

In relation to the organizational goals of Profitability and Growth, the Managementconstituency identified the overall means—i.e. the organization’s critical successfactors—in terms of Production Capacity, Customer Satisfaction and Low Cost, asoperationalized by the seven Strategic Process Indicators of Table 2.

Page 10: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

50 M. de Haas and J. A. Algera

Table 1Strategic result indicators

Profitability • Return on invested capital (ROIC)• Profit margin• Productmix composition

Growth • Market share• Geographic distribution• Top-5 position

Table 2Strategic process indicators

Production capacity • Throughput volume• Material yield• Stock levels

Customer satisfaction • Quality complaints• Delivery reliability• Product development

Low cost • Unit cost

For the year 1999, the following targets were set for the Strategic Process Indicators:

• increase Throughput Volume from 140 to 220 kilotonnes;• increase Material Yield from 80.8% to 81.7%;• decrease Stock Levels from 29.2 to 28 kilotonnes;• decrease Quality Complaints from 56 to 50 complaints;• increase Delivery Reliability from 70% to 90%;• increase Product Development (i.e. creation of round bars of forging steel) from

15 to 40 kilotonnes;• decrease Unit Cost from € 308 to € 292.

Jointly, the performance indicators of Tables 1 and 2 constituted the heart of thebusiness strategy and thus of the management control system. The identified means–end relation represents management’s vision of what the organization should begood at for achieving its overall goals. Provided that the assumed relation betweenmeans and ends is valid, the Management constituency should urge lower-levelagent constituencies to buy-in and allocate their resources to Production Capacity,Customer Satisfaction and Low Cost. This requires the Strategic Process Indicators(see Table 2) to become part of the Quality and Logistics constituency members’mental models. Hence, these organizational actors were invited to discuss therelevance of these performance indicators as part of a Strategic Dialogue. It should benoted that management’s indicator design was communicated as a prototype designapt for adjustment in case of relevant bottom-up input.

Data collectionAn example of an ordinal data set is presented in Table 3. This table contains thepre-test ranking data of the Quality constituency in the M-Q design of Figure 2.The columns in Table 3 represent the respondents—i.e. the Quality constituency

Page 11: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 51

Table 3Categorical data O1,Q: pre-test of the quality constituency

VRIE1 DBIE1 HAMO1 EIJD1 OVER1 MENS1 JONK1 TEMM1

Increase throughput volume (TV) 4 2 1 5 6 7 3 5Increase delivery reliability (DR) 3 5 1 7 1 2 5 5Decrease stock levels (SL) 2 2 1 1 1 1 1 4Decrease quality complaints (QC) 6 7 7 7 7 6 7 6Decrease unit cost (UC) 1 6 7 7 6 3 6 4Increase material yield (MY) 5 6 7 4 6 5 2 7Increase product development (PD) 7 6 7 3 5 4 4 6

members; the number that is added to the abbreviated names indicates the pre-testnature of the data. A similar, post-test data set was collected afterwards. Similar datasets were also collected for the Management constituency. Each constituency memberwas asked individually to arrange the Strategic Process Indicators after the designteam meetings (pre-test) and again after the management approval meeting (post-test). The instruction was to arrange the indicators in order of importance for theorganization’s overall wellbeing, using a 7-point scale; the score of 7 representing thehighest importance, while the score of 1 represented the lowest importance. In theM-L design of Figure 3, similar data sets were collected. For the full data sets, thereader is referred to de Haas (2000).

Data analysisAn example of a non-metrically transformed ranking data set is presented in Table 4.This table is the result of the CATPCA dimension reduction of the pre-test data ofthe Quality constituency shown in Table 3. The result is presented in terms of vectorcoordinates (representing the subjects, i.e. respondents in the analysis and, moreover,their loadings upon the reduced dimensions) and point coordinates (representingthe options, i.e. the shared goal definitions in the analysis that define the reduceddimensions). For the full data analysis of both the M-Q and the M-L designs, thereader is referred to de Haas (2000). We will first explain the left-hand side of Table 4for the purpose of measuring within-constituency Goal Coherence; later, the right-hand side will be explained for the purpose of measuring between-constituency GoalCoherence.

In general, the number of dimensions in a CATPCA procedure is optional. Inour analyses, the number of dimensions was set at a default value of 2, whichconsequently allowed for two-dimensional graphic representation. The reductionto two dimensions was allowed for since the sum of VAF, which is a measure ofmodel fit, was largely sufficient in all the analyses. The left-hand side of Table 4 isgraphically displayed in Figure 4 (vectors represent the subjects; points representthe options). This figure shows pre-test within-constituency Goal Coherence of theQuality constituency.

A vector coordinate corresponds with the correlation between a subject and adimension. The calculation of correlations is actually possible after the transforma-tion of categorical data into metric data. In factor analysis (or principal componentanalysis) terms, the correlation between a subject and a dimension is interpreted asthe loading of a subject upon a dimension. The higher a specific subject loads upon

Page 12: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

52 M. de Haas and J. A. Algera

Table 4Dimension loadings and scores: pre-test of the quality constituency

Active variable DIM1 DIM2 Suppl. variable DIM1 DIM2(Quality) loading loading (Management) loading loading

VRIE1 0.779 −0.536 STRI1 −0.005 0.553DBIE1 0.930 0.223 DORT1 0.336 0.841HAMO1 0.809 −0.070 GEER1 0.127 0.504EIJD1 0.252 0.922 BROE1 0.166 0.691OVER1 0.912 0.167 WOUD1 0.499 0.497MENS1 0.799 −0.224 BREE1 0.709 0.336JONK1 0.717 0.683 SCHI1 0.654 0.673TEMM1 0.756 −0.567

Eigenvalue 4.745 2.056 Eigenvalue 1.335 2.558Cronbach’s α 0.902 0.587 Cronbach’s α 0.293 0.710VAF (6 = .850) 0.593 0.257 VAF (6 = .556) 0.191 0.365Association 0.593 0.091 Association 0.191 0.365

Unit (i.e. option) DIM1 DIM2score score

Throughput volume (TV) −0.394 −0.388Delivery reliability (DR) −0.851 0.814Stock levels (SL) −1.558 −0.185Quality complaints (QC) 1.779 0.881Unit cost (UC) 0.001 1.441Material yield (MY) 0.538 −1.364Product development (PD) 0.486 −1.199

a specific dimension, the more the specific ordering of options by that subject isexplained by that specific dimension.

Of major importance is the interpretation of dimensions. In general, the dimensionsthat result from reduction with the CATPCA procedure are anonymous: they merelydefine a multi-dimensional space. However, since we apply the procedure to trans-posed data matrices in which the subjects are the variables, the dimensions attain aspecific meaning. The perpendicular projection of the options onto either dimension,indicated by the dotted lines in Figure 4, produces a metric approximation of theordinal goal priorities found in Table 4. In other words, DIM1 and DIM2 each repre-sent a one-dimensional mental model: goals arranged along a straight line, albeit nolonger equidistantly due to the non-metric transformation that is part of the CATPCAanalysis. Now recall our definition of Goal Coherence: inter- and intra-consensus onshared goal priorities. Consensus in terms of CATPCA can thus be identified by thesubjects in the analysis loading highly (i.e. loadings close to +1 or −1) upon the samedimension. In graphic representation, consensus will consequently result in all vec-tors lying in the same direction on or near this shared dimension. But how can weidentify the dimension that is shared most among group members? For this purpose,we calculate the VAF by each dimension.

The eigenvalues in Table 4 represent a measure of how much variance is accountedfor by the dimensions that resulted from dimension reduction. The eigenvalue perdimension is calculated as the SSQ (i.e. sum of squares) of the loadings. In order todetermine how much variance the reduced dimensions account for, the calculatedeigenvalues after dimension reduction are divided by the sum of eigenvalues in

Page 13: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 53

-2 ,0

-1 ,5

-1 ,0

-0 ,5

0 ,0

0 ,5

1 ,0

1 ,5

2 ,0

-2 ,0 -1 ,5 -1 ,0 -0 ,5 0 ,0 0 ,5 1 ,0 1 ,5 2 ,0

dim1

throughput volume

stock levels

delivery reliability

unit cost

EIJD1 quality complaints

JONK1

DBIE1

OVER1

HAMO1

MENS1

VRIE1

TEMM1

product development

material yield

dim2

Figure 4. Pre-test of the quality constituency.

the original data sets. In the original data set, the sum of eigenvalues equals theminimum of the number of variables (i.e. subjects) and the number of options, minusone. The dimension with the highest VAF is dominant in explaining individualordering preferences among respondents. In other words, the dominant dimensionrepresents a specific ordering of goals that is shared most among subjects, i.e.the dominant dimension represents the strongest intersubjective perception of goalpriorities by the group. By definition, the CATPCA procedure labels the dominantdimension DIM1, the second dominant dimension DIM2, etc.

In addition to Figure 4, post-test within-constituency Goal Coherence of the Qualityconstituency is shown in Figure 5. Both pre-test and post-test measures are alsoshown for the Management constituency in Figures 6 and 7. When comparingFigure 5 to 4 and Figure 7 to 6, the converging vectors clearly illustrate an increase inwithin-constituency Goal Coherence in the M-Q design. For similar displays in theM-L design, the reader is referred to de Haas (2000).

Page 14: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

54 M. de Haas and J. A. Algera

-1 ,0

-0 ,5

0 ,0

0 ,5

1 ,0

1 ,5

2 ,0

2 ,5

-1 ,0 -0 ,5 0 ,0 0 ,5 1 ,0 1 ,5 0 22 , ,5

delivery reliability

quality complaints

JONK2

HAMO2 dim1MENS2

TEMM2

DBIE2VRIE2

OVER2

unit cost

dim2material yield

product developmentthroughput volume

stock levels

Figure 5. Post-test of the quality constituency.

From the interpretation of a dimension as the metric equivalent of ordinal goalpriorities and the interpretation of the dominant dimension as the specific order ofgoals that is shared most among constituency members, we can now calculate theamount of association as the intersubjective measure of within-constituency GoalCoherence. This measure corresponds with the calculation of VAF for the dominantdimension, corrected for opposite loading signs. Association is hence calculated asthe difference between the SSQ of positive loadings and the SSQ of negative loadingson the dominant dimension, divided by m (the number of variables, i.e. subjects inthe analysis). The numerator of this quotient is by definition zero or larger, sincethe CATPCA procedure attributes positive signs to the majority of similar loadings.Hence, degrees of within-constituency Goal Coherence vary between 0 and 1. Informula notation

association=SSQ(loadingpos) − SSQ(loadingneg)

m

Page 15: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 55

-2 ,5

-2 ,0

-1 ,5

-1 ,0

-0 ,5

0 ,0

0 ,5

1 ,0

1 ,5

-2 ,0 -1 ,5 -1 ,0 -0 ,5 0 ,0 0 ,5 1 ,0 1 ,5 2 ,0

throughput volume

WOUD1

delivery reliability

quality complaints

dim1

unit cost

dim2

material yield

product development

stock levels

STRI1

BREE1

GEER1

DORT1

SCHI1

BROE1

productdevelopment

Figure 6. Pre-test of the management constituency.

The right-hand side of Table 4 is graphically displayed in Figure 8, whichbrings us to between-constituency Goal Coherence. As the within-constituency GoalCoherence concept relates to the sharing of goal priorities among members of asingle constituency, the between-constituency Goal Coherence concept relates to thesharing of goal priorities among members of two goal interdependent constituencies.Thus, the degree of similarity of within-constituency Goal Coherence between goalinterdependent constituencies is a measure of between-constituency Goal Coherence.

In the M-Q design, we interpret between-constituency Goal Coherence as thedegree to which the Management constituency members share the goal priorities thatare shared most among the Quality constituency members, and vice versa. Thus, inCATPCA terms, between-constituency Goal Coherence is measured by the loadingsof the Management constituency members upon the dominant dimension in theanalysis of the Quality constituency, and vice versa. These loadings can be obtainedby adding Management constituency members as supplementary variables to theanalysis of the Quality constituency members, and vice versa. The resulting loadings

Page 16: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

56 M. de Haas and J. A. Algera

- 2 ,5

- 2 ,0

- 1 ,5

- 1 ,0

- 0,5

0 ,0

0 ,5

1 ,0

1 ,5

- 1,5 - 1,0 - 0,5 0 ,0 0,5 1 ,0 1,5 0 22 , ,5

quality complaints

dim1

unit cost

dim2

material yield

product development

throughput volume

delivery reliability

stock levels

WOUD2

STRI2

BREE2

GEER2DORT2

SCHI2

BROE2

Figure 7. Post-test of the management constituency.

from the supplementary analysis of the Management constituency (pre-test) areshown in Table 4. Supplementary variables are passively involved in the CATPCAdimension reduction procedure, i.e. these variables do not define the dominantdimensions: these dimensions are defined by the active variables. Hence, the optioncoordinates in Figure 8 correspond exactly with those in Figure 5. For the othersupplementary analyses, both in the M-Q design and in the M-L design, the reader isreferred to de Haas (2000).

In addition to Figure 8, post-test between-constituency Goal Coherence of theManagement constituency is shown in Figure 9. Both pre-test and post-test measuresare also shown for the Quality constituency in Figures 10 and 11. When comparingFigure 9 to 8 and Figure 11 to 10, the converging vectors clearly illustrate an increasein between-constituency Goal Coherence in the M-Q design. For similar displays inthe M-L design, the reader is referred to de Haas (2000).

By applying the previously presented formula, we can calculate the amountof association as the intersubjective measure of Goal Coherence between the

Page 17: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 57

-2 ,0

-1 ,5

-1 ,0

-0 ,5

0 ,0

0 ,5

1 ,0

1 ,5

2 ,0

-2 ,0 -1 ,5 -1 ,0 -0 ,5 0 ,0 0 ,5 1 ,0 1 ,5 2 ,0

quality complaints

dim1

unit cost

dim2

material yield

product development

throughput volume

delivery reliability

stock levels

WOUD1STRI1

BREE1

GEER1

DORT1

SCHI1BROE1

Figure 8. Supplementary pre-test of the management constituency.

Management and the Quality constituency. Note that in this case the numerator of theassociation quotient does not have a positive sign by definition. In a supplementaryanalysis, the CATPCA procedure cannot attribute positive signs to the majorityof similar loadings of the passive variables, since the dimensions have alreadybeen defined by the active variables. Hence, degrees of between-constituency GoalCoherence vary between −1 and +1.

All calculations of pre-test and post-test degrees of Goal Coherence in the M-Qand M-L designs are summarized in Table 5 and Table 6 respectively. An effect iscalculated as the difference between post-test and pre-test association. Since within-constituency Goal Coherence varies between 0 and +1, within-constituency effectscan vary between −1 and +1; since between-constituency Goal Coherence variesbetween −1 and +1, between-constituency effects can vary between −2 and +2.

Page 18: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

58 M. de Haas and J. A. Algera

- 1 ,0

- 0 ,5

0 ,0

0 ,5

1 ,0

1 ,5

2 ,0

2 ,5

- 1 ,0 - 0 ,50 ,

0,5 0 1,51,0 2 ,0 2 ,5

quality complaints

dim1

unit cost

dim2material yield

product developmentthroughput volume

delivery reliability

stock levels

WOUD2

STRI2

BREE2

GEER2DORT2

SCHI2

BROE2

Figure 9. Supplementary post-test of the management constituency.

Table 5Summary of findings in the M-Q design: treated ‘experimental’ group

Management constituency Quality constituencyGoal coherence pre-test post-test pre-test post-test

Within-constituency Association 0.507 0.756 0.593 0.930Effect 0.249 0.337

Between-constituency Association 0.191 0.720 0.163 0.871Effect 0.529 0.708

5. Conclusions

The positive effects demonstrated in the M-Q design (Table 5) compared to theneutral effects demonstrated in the M-L design (Table 6), provide support for theinstrumental hypothesis of our research. However, we must present this support with

Page 19: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 59

-2 ,5

-2 ,0

-1 ,5

-1 ,0

-0 ,5

0 ,0

0 ,5

1 ,0

1 ,5

-2 ,0 -1 ,5 -1 ,0 -0 ,5 0 ,0 0 ,5 1 ,0 1 ,5 2 ,0

TEMM1

HAMO1

DBIE1

OVER1

JONK1EIJD1

MENS1

VRIE1

quality complaints

dim1

unit cost

dim2

material yield

product development

throughput volume

delivery reliability

stock levels

Figure 10. Supplementary pre-test of the quality constituency.

Table 6Summary of findings in the M-L design: untreated ‘control’ group

Management constituency Logistics constituencyGoal coherence pre-test post-test pre-test post-test

Within-constituency Association 0.756 0.756 0.481 0.601Effect 0.000 0.120

Between-constituency Association 0.329 0.300 0.454 0.446Effect −0.029 −0.008

the greatest care. Recall that we did not design an experiment according to the exactrules of Cook and Campbell (1979). The lack of management approval meeting inthe M-L design was not planned for, but happened unintentionally in the course ofthe action research. We introduced the analogy of an experiment to be able to makeany sense of our findings. We referred to the M-Q design as a treated ‘experimental’

Page 20: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

60 M. de Haas and J. A. Algera

-2 ,5

-2 ,0

-1 ,5

-1 ,0

-0,5

0,0

0,5

1 ,0

1 ,5

-1,5 -1 ,0 -0,5 0, , 5 1, , 5 2,000 0 1 2 ,5

quality complaints

dim1

unit cost

dim2

material yield

product development

throughput volume

delivery reliability

stock levels

TEMM2HAMO2

DBIE2

OVER2

JONK2

MENS2VRIE2

Figure 11. Supplementary post-test of the quality constituency.

group, while we referred to the M-L design as an untreated ‘control’ group. From this‘experimental’ perspective, our findings are promising, though very preliminary.

The strong and positive effect on between-constituency Goal Coherence in the M-Qdesign is further illustrated in Figure 12. This figure shows the dominant dimensionsin the CATPCA analyses of the pre-test and post-test ranking data of the Managementconstituency and the Quality constituency. These dimensions represent a metricapproximation of ordinal goal priorities, found by the perpendicular projectionof the options upon the dominant dimension in the two-dimensional graphicrepresentations. Moreover, since these dimensions are dominant, they represent themetric approximations of ordinal goal priorities that are shared most within theManagement constituency and within the Quality constituency.

Recall that the degree of between-constituency Goal Coherence in fact correspondswith the degree of similarity of within-constituency Goal Coherence between goalinterdependent constituencies. The similarity in ordering preferences within separateconstituencies is graphically illustrated in Figure 12 by the dotted connection-

Page 21: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 61

UC

MYQC

DRT V

SL

PD

SL

DR

T V

UC

MYPD

QC

SL

PD

DR

UC

QC

MY

T V

PDSLT V

MYDR

QC

UC

- 2 ,0

- 1 ,5

- 1 ,0

- 0 ,5

0 ,0

0 ,5

1 ,0

1 ,5

2 ,0

2 ,5

Managementconstituency

Quality constituency

Management constituency

Quality constituency

pretest posttestm

etric

app

roxi

mat

ion

of o

rdin

al p

riorit

ies

Figure 12. Similarity of dominant dimensions in the M-Q design.

lines between shared goal definitions. Clearly, the preferred sequence of orderingby the Management constituency and the Quality constituency, especially withregard to the goal of decreasing unit cost (UC), decreasing quality complaints (QC)and increasing product development (PD), is more similar after than before theconstructive controversy of the management approval meeting. In comparison,hardly any developments have taken place in the M-L design, as is illustrated inFigure 13.

Future research is required to demonstrate further and more convincing supportfor the instrumental relation between Strategic Dialogue and Goal Coherence inparticular, and between participation and organizational effectiveness in general.

6. Discussion

To close, we return to our opening remarks. The statements made in the introductionimply a fundamental repercussion for the practice of and research in ManagementAccounting. Recall that human behavior, resulting in decisions and action, is largelybased on a person’s mental model. Furthermore, recall that a mental model feeds

Page 22: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

62 M. de Haas and J. A. Algera

S L

P D

T VDR

MY

QC

UC

T V

P D

MY

S L

UC

QC

DR

P D

QC

DR

MY

UC

T V

MYP D S L

T V

UC

QC

DR

S L

- 1 .5

- 1 .0

- 0 .5

0 .0

0 .5

1 .0

1 .5

2 .0

2 .5

Managementconstituency

Logisticsconstituency

Management constituency

Logistics constituency

pretest posttestm

etric

app

roxi

mat

ion

of o

rdin

al p

riorit

ies

Figure 13. Similarity of dominant dimensions in the M-L design.

the processes of selective perception and recollection of information. The purpose ofthe Management Accounting function in organizations is to provide decision-makerswith relevant information for control purposes: both accounting and non-accountinginformation—the latter among other things as a consequence of the relevance lostmovement. In general, information is periodically produced as part of the formalreporting structure. Today, expensive, tailor-made management dashboards, suchas Balanced Scorecards, provide formalized information, either on paper or via anintranet.

In light of the statements made in the introduction of this paper, an additionalinterpretation of relevance emerges: information is only relevant if it fits withinexisting mental models. If not, costly information that is periodically distributed aspart of the formal reporting structure is likely to be disregarded, because people tendto look for information that confirms rather than refutes their view of the world. Itis therefore of great importance to start linking the design of management controlsystems to peoples’ mental models. If this link is missing, people in organizationswill collect information from other sources. And in today’s organizations, thesealternative sources are in plenty, due to increasing investments in enterprise resourceplanning (ERP) systems, such as SAP, Peoplesoft, Oracle, JD Edwards and Baan. An

Page 23: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 63

ERP system is an organization-wide system that basically consists of one (relational)database in which transactional data are recorded. In combination with onlineanalytical processing (OLAP) tools, an ERP database has the potential to fulfillvirtually any information request that exists within the organization. In other words,it has the potential to feed and maintain existing divergence in peoples’ mentalmodels.

From this relevance perspective, the need for convergent mental modeling isfurthermore stressed. Without convergence, the relevance of management controlsystems remains lost. It should be a challenge for the Management Accountingresearcher to find ways to reconnect the purpose of information with the purposeof control. The current paper instigates this challenge.

Acknowledgements

The author gratefully acknowledges the helpful comments of Ed Vosselman, Harrievan Tuijl, Jacqueline Meulman, Jacques Theeuwes and two anonymous reviewers.

Appendix: CATPCA Example

This appendix illustrates the categorical principal component analysis (CATPCA)technique with an uncomplicated example. The example will clarify how CATPCAoptimally reduces the dimensionality of a categorical data set (ordinal level ofmeasurement) with minimal loss of information, using a non-metric transformationand optimal scaling of the categorical variables. Concretely, the example showsa reduction from three to two dimensions. Typically, the results of a CATPCAtransformation are graphically displayed.

In the example, we have three subjects and their ordering preferences regardingfour alternatives A, B, C and D. Rank number 4 indicates a high priority, 3 amoderately high priority, 2 a moderately low priority and 1 a low priority. Theranking (i.e. categorical) data are depicted in Table 7. This table is a transposed datamatrix in which the subjects are the variables and the options are the units. We use atransposed data matrix to illustrate CATPCA, since in our search for Goal Coherencewe also analyze transposed data matrices.

The categorical data of Table 7 can be graphically represented. Just as in a typicalCATPCA display, the variables (i.e. subjects) are represented by vectors and the units(i.e. options) by points. By using a three-dimensional space for Table 7, we are able

Table 7Categorical data set

Option SUBJECT1 SUBJECT2 SUBJECT3

A 1 1 2B 2 2 1C 3 4 4D 4 3 3

Page 24: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

64 M. de Haas and J. A. Algera

+

+

+

-

-

-

dim3

SUBJECT 3

option C

optionA option D

dim1

SUBJECT 1

option B

SUBJECT 2

dim2

Figure 14. Three-dimensional representation of categorical data.

to graphically represent the data with no loss of information. This is because thecategorical data set itself has three dimensions (i.e. sources of variance): the minimumof the number of variables and the number of options, minus one. In such a case,there is no distortion in the model and, consequently, all variance can be accountedfor. The graphical representation in three dimensions of the categorical data of Table 7is depicted in Figure 14.

The three-dimensional space is described by the mutual positions of the options,which in Figure 14 are indicated by points. Hence, the options define the dimensionsthrough the perpendicular projection of the points. A point projection on the positiveside of a dimension indicates a high priority for the corresponding option, on thenegative side a low priority and on the origin a moderate priority. Consequently,DIM1 corresponds with D-C-B-A as the equidistantly positioned and thus ordinallyarranged options in sequence of decreasing importance, DIM2 with C-D-B-A andDIM3 with C-D-A-B.

In Figure 14, vectors departing from the origin represent the subjects. Theperpendicular projection of the points on each of these vectors should produce,

Page 25: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 65

- 1 , 5

- 1 , 0

- 0 , 5

0 , 0

0 , 5

1 , 0

1 , 5

- 1 , 5 - 1 , 0 - 0 , 5 0 , 0 0 , 5 1 , 0 0 , 5

SUBJECT 3

option CoptionA

option D

dim1

SUBJECT 1

option B

SUBJECT 2

dim2

Figure 15. Two-dimensional representation of CATPCA data.

without any loss, the categorical information found in Table 7. In the three-dimensional space of Figure 14, the ordering preferences of SUBJECT1 are exactlyrepresented by the first dimension, i.e. DIM1, those of SUBJECT2 by DIM2 and those ofSUBJECT3 by DIM3. Hence, the vector that represents SUBJECT1 coincides completelyin the positive direction with DIM1, the SUBJECT2 vector with DIM2 and the SUBJECT3vector with DIM3.

By applying the CATPCA technique to Table 7, the same information can begraphically depicted in a two-dimensional space. By reducing the number ofdimensions, distortion is created and information is possibly lost. In order to accountfor as much variance as possible, CATPCA non-metrically transforms and optimallyscales the original, categorical data. The result of a CATPCA analysis is stated interms of 1) loadings per variable upon each of the reduced number of dimensions;and 2) scores per option, which relate the option to each of the reduced numberof dimensions. Loadings can actually be calculated since the data are metric afterdimension reduction. For graphical purposes, the loadings thus describe the vectorcoordinates, while the scores describe the point coordinates. The result of theCATPCA analysis of Table 7 is found in Table 8. The data of Table 8 are graphicallydisplayed in Figure 15.

The VAF by each dimension is calculated as the sum of squared loadings perdimension, divided by the number of variables. Apparently, no information is lost

Page 26: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

66 M. de Haas and J. A. Algera

- 1 , 5

- 1 , 0

- 0 , 5

0 , 0

0 , 5

1 , 0

1 , 5

- 2 , 0 - 1 , 5 - 1 , 0 - 0 , 5 0 , 0 0 , 5 1 , 0 1 , 5 2 , 0

option CoptionA

option D

SUBJECT 1

option B

Figure 16. Perpendicular point projections on vector subject1.

since the reduced dimensions jointly account for all the variance. In other words,Table 8 contains exactly the same information, without any distortion, as Table 7. Thereader should be aware of the fact that the transposed data presented here representan ideal situation, which will not be encountered in the analysis of empirical data.Generally, dimension reduction results in a loss of information, expressed by the sumof VAF being less than 1.

The perpendicular projection of the points on each of the three vectors (i.e.subjects) would have to generate a reflection of the initial categorical data ofTable 7. The perpendicular projections for SUBJECT1 are graphically demonstratedin Figure 16.

Table 8Non-metric transformation result

Variable DIM1 loading DIM2 loading Option DIM1 score DIM2 score

SUBJECT1 0.879 0.476 A −1.253 −1.003SUBJECT2 1.000 −0.017 B −0.670 1.054SUBJECT3 0.893 −0.451 C 1.220 −0.996

D 0.702 0.945

VAF (6 = 1.000) 0.857 0.143

Page 27: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 67

- 2 , 0

- 1 , 5

- 1 , 0

- 0 , 5

0 , 0

0 , 5

1 , 0

1 , 5

2 , 0

Per

pend

icul

ar p

oint

pro

ject

ions

SUBJECT 3

option C

optionA

option D

SUBJECT 1

option B

option C

optionA

option D

option B

option C

optionA

option D

option B

SUBJECT 2

Figure 17. Graphical representation of perpendicular point projections.

A perpendicular point projection in the direction of the vector head indicatesthe corresponding option to be of high importance for the corresponding subject,whereas a projection in the opposite direction indicates a low importance; projectionsnear the origin indicate a moderate importance. The distance from the intersection ofthe perpendicular line of point D with vector SUBJECT1 (or its extension) to the originis +1.068. This figure is positive since the intersection is situated in the directionof the vector head indicating a high importance for SUBJECT1. Similar calculationsfor point C result in +0.599 (indicating a moderately positive importance), for pointB in −0.087 (indicating a moderately low importance) and for point A in −1.579(indicating a low importance). All calculations, including those for SUBJECT2 andSUBJECT3, are summarized in Table 9. The data of Table 9 are graphically displayedin Figure 17.

Compared with Table 7, Figure 17 contains exactly the same information. However,the options are no longer equidistantly positioned after the non-metric transforma-

Page 28: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

68 M. de Haas and J. A. Algera

Table 9Perpendicular point projections on vectors

Option SUBJECT1 SUBJECT2 SUBJECT3

A −1.579 −1.236 −0.667B −0.087 −0.687 −1.072C 0.599 1.236 1.538D 1.068 0.687 0.201

Table 10Alternative representation of original, categorical data

Rank number SUBJECT1 SUBJECT2 SUBJECT3

4 D C C3 C D D2 B B A1 A A B

tion. For the interpretation of Figure 17 in terms of ordering preferences at an ordinallevel of measurement, this has no consequences. The similarity with Table 7 is all themore evident if we represent Table 1 alternatively, as is shown in Table 10.

References

Checkland, P. and Scholes, J., 1990. Soft Systems Methodology in Action, Chichester, Wiley.Cohen, J., 1960. A coefficient of agreement for nominal scales, Educational and Psychological

Measurement, 20, 37–46.Cook, T. D. and Campbell, D. T., 1979. Quasi-experimentation: Design and Analysis Issues for Field

Settings, Chicago, IL, Rand McNally.Cronbach, L. J. and Gleser, G. C., 1953. Assessing similarity between profiles, Psychological

Bulletin, 50, 456–473.Dörner, D., 1980. On the difficulties people have in dealing with complexity, Simulation and

Games, 11, 87–106.de Haas, M. and Kleingeld, P. A. M., 1999. Multilevel design of performance measurement

systems: enhancing strategic dialogue throughout the organization, Management AccountingResearch, 10, 233–261.

de Haas, M., 2000. Strategic dialogue: in search of goal coherence. Ph.D. Thesis, Department ofTechnology Management, Eindhoven University of Technology, The Netherlands.

Heiser, W. J. and Meulman, J. J., 1994. Homogeneity analysis: exploring the distribution of vari-ables and their nonlinear relationships, in M. Greenacre, J. Blasius (eds), Correspondence Anal-ysis in the Social Sciences: Recent Developments and Applications, New York, Academic Press.

Hinsz, V. B., 1995. Mental models of groups as social systems, Small Group Research, 26, 200–233.Hogarth, R., 1987. Judgement and Choice, Chichester, Wiley.Ickes, W. and Gonzales, R., 1994. “Social” cognition and social cognition: from the subjective to

the intersubjective, Small Group Research, 25, 294–315.Meulman, J. J. and Heiser, W. J., 1999. SPSS Categories 10.0, Chicago, IL, SPSS Inc.Miller, G. A., 1956. The magical number seven, plus or minus two: some limits on our capacity

for processing information, The Psychological Review, 63, 81–97.

Page 29: (2)Demonstrating the Effect of the Strategic Dialogue Participation in Designing the Management c

Demonstrating the Effect of the Strategic Dialogue 69

Neisser, U., 1967. Cognitive Psychology, New York, Appleton-Century-Crofts.Simon, H. A., 1948. Administrative Behavior: a Study of Decision-making Processes in Administrative

Organizations, New York, Macmillan.Simon, H. A., 1985. Human nature in politics: the dialogue of psychology with political science,

The American Political Science Review, 79, 293–304.Steers, R. M., 1977. Organizational Effectiveness: a Behavioral View, Santa Monica, CA, Goodyear

Publishing Company.Stephenson, W., 1953. The Study of Behavior: Q-technique and its Methodology, Chicago, IL,

University of Chicago Press.Tjosvold, D., 1985. Implications of controversy research for management, Journal of Manage-

ment, 11, 21–37.Van Tuijl, H. F. J. M., Kleingeld, P. A. M. and Algera, J. A., 1995. Prestatiemeting en beloning:

contextafhankelijk ontwerpen [Performance measurement and rewards: context dependentdesign], Gedrag en Organisatie, 8, 419–438.

Vancouver, J. B., Millsap, R. E. and Peters, P. A., 1994. Multilevel analysis of organizationalgoal congruence, Journal of Applied Psychology, 79, 666–679.

Vennix, J. A. M., 1996. Group Model Building: Facilitating Team Learning Using System Dynamics,Chichester, Wiley.