Professional Documents
Culture Documents
Laurie J. Kirsch
Katz Graduate School of Business, University of Pittsburgh, Pittsburgh, Pennsylvania 15260
T his manuscript provides needed clarification o1 a number of key constructs in the area of
managerial control, a much discussed but inadequately studied phenomenon within the organi- zation
sci‹ nces. Further, the carefully designed field study provides clear evidence of the importance
of both b‹ havioral and self c‹›ntrol in the management of complex organizatic›nal tasks.
Robert Zmud
Abstract Introduction
Control theory attempts to explain how one person or group
Control in organizations has long been a topic of
in an organization can ensure that antither person or group
interest for researchers and practitioners alike, who
works toward and attains a set or organiz‹itii Anal goals. Prior
empirical work investigai ing control theory has shown that generally recognize that control mechanisms are
characteristics of the task and of the organizational environ- needed t‹a help ensure organizations achieve their goals.
ment predict the use or various types of ct introl. However, While control can be viewed from various perspectives,
this paper argues that w hen control theory is applied to a it iS viewed here in a hehauiorul sense, that is, as
complex, nonroutine task such as the management of infor- attempts to ensure that individuals working on organi-
mation systems development, the theory of control is incom- zational projects act according to an agreed-up‹in strat-
plete. In particular, it prt poses that knowletlge of the task is egy t‹i achieve desired objectives (Jaworsk i 1988,
a key determinant of typi of control. Merchant I 9h8). This definition implies that when a
t-tour modes ot contrc›l (behavior, outcome, clan, and self) controller exercises control over a controllee, she iS
arc identified from the ‹ organizational literature; each high-
taking some action in order to regulate or adjust the
lights different aspects ‹›i control in organizations. Building
hehaoior ref the controllee. The purpose of this study is
on prior empirical work, this paper integrales the dil‘ferent
the‹iretical perspectives a nd predicts the circumstances under to empirically test a model that predicts the circum-
w'hich each type of ctintrol will be implemented. Survey stances under which various modes of control will be
rcsp‹inses from 96 partic ipants of 32 systems development exercised in the context of information systems devel-
efft›rts suggest that the c tent to which behaviors arc moni- torccl opment (ISI3) Jaroject management. ISD is the process
interacts with the prtaject sponsor’s ieveI of systems of identi in k * tl siness objectives for a computerized
development knowledge io determine the amount of behav- information system; designing, building, and testing the
ior contra I; that t›utcomc control is rt functi‹›n of the extent system; and implementing and maintaining it. ISD pro-
tc› which behaviors are m ›nitored and outcorries are measur- ject management involves planning, organizing re-
able; and that self-conti ol is dependent on the extent to sources, :ind measuring progress against plans with an
which outcomes are mea* urable and the levy l ‹it the project overall gr›al oi achieving a set of project objectives
spc›nsor’s kntiwledge ab‹›ut systems development activities.
(Page-.fones 1985).
No relationship between clan control and t he independent
vari:ibles was found. Controlling complex organizational tasks is an elu-
(Monagrment I“ontrol. €."ontrol Theory; .Software Development) sive prcoci:ss. one that is not well understood (Flamholtz
et al. 198a, Merchant 1958, Snell 1992). Because the
management of ISD is nonroutine and diNicult (Slevin
and Pinto 1957, Stuckenbruck 1988), it provides a
classic example of the issues associated with structuring integrate the separate knowledge bases and indepen-
control mechanisms for complex tasks. Much of this dently created products into a coherent whole (Walz
complexity stems from the communication and coordi- et al. 1993).
nation problems inherent in tasks that span organiza- This paper proceeds as follows. In the next section, a
tional departments. These problems are manifested in theoretical mode1 of control is developed and hypothe-
several ways. First, each stakeholder involved in ISD ses are generated. The research methodology is de-
has a set of goals, which may include a quality end- scribed in the following section. Next, the data analysis
product, timeliness of implementation, client satisfac-
Downloaded from informs.org by [131.94.16.10] on 28 August 2015, at 18:31 . For personal use only, all rights reserved.
Ouchi’s conceptualization by positing that the choice of’ into a group, self-control stems from individual objec-
a control strategy is also dependent on “behavior ob- tives and standards (Jaworski 1988). The difference is
scrvability”: the extent to which the controller has subtle. With self-control, an individual monitors his
ac:cess to informati‹›n systems that reveal the con- behavior .ind rewards car sanctions himself accordingly.
trollee’s actions. These information systems, among With clan control, monitoring, rewarding and sanction-
‹other mechanisms, include boards of directors, ac- ing .me a functit›n of the group. Consider the following
counting systems, personal observation, and evaluative example ui' self-control. An IS programmer, because he
Downloaded from informs.org by [131.94.16.10] on 28 August 2015, at 18:31 . For personal use only, all rights reserved.
trollers are ‹ibout ISD, the more likely they will im-
I igui c 2, is a retincincnt ‹a1 Govindar: jan and Fisher’s plerrient behavior-based control. This ‹argument is
( I “J9ti) model. It is ink portant to note that the depiction consists:nt with Snell’s ( 1992) finding that, among other
‹›t ‹i certain mocle c›f ct›ntrol in cach cell is nr›t meant things, the use of behavior control is dependent on
It i imply that mocJes of control ‹occur independently. “hat ing complete cause-effect knowledge”(p. 317), and
ltathcr. the intent i• tr› predict the rcl‹itive extent to suggests that behavior control will be implemented in
w hich the various m‹odes of control arc present. While cells 1 antl 3.
t he model :ippears t‹ be descriptive, C ovindarajan and t“untrol theories also suggest that the controllers in
I is he r ( l9'i0) point ‹but that il is normative ‹is well, the situations described by cells 1 and 5 have two
implying I h‹it fr›r el lective pcrli›rmance, an organiza- options. l3eciiuse outcome measurability is high, one
i it›n should use the I ‹arm of control specified” (p. 261) opti‹›n is to implement outcome-based control. While
in cach cell. The rai ionale for the hvpr›theses gener-
this approach is feasible, agency theory suggests that it
‹ited from Figure 2 i explained next.
place:s risk ‹›n a risk-‹iverse controllee. In the ISD
Co nsider first cells 1 and 3. Agencv theory suggests
c‹intext, thcrc would also seem t‹i be a bias against the
th‹tt when 1 ehavior ‹ observability is high, behavior use •f ‹outcome contra 1: with outcome control, errors
con- trol will he used: t lie more contrt›llcrs monitor the
may not he detected until the system is completed;
howeve r, the costs of correcting the errors are much
greater at this point than during the early phases of
ISU (Pressm:in 1992, Walz et al. 1993). According to
agency thcorv, a second option in cells 1 and 5 is to
Figure 2 Research Model purchase inf‹›rmation about the controllee’s behavior
Predictions in cells 1, 4, 5, 7, and 8 differ from and structure a behavioral contract. This makes sense
Govindarajan and Fisher's (19901 model, where the i{ the i-ontroller understands the transformation pro-
respective t›redictions are Outcome or Behavior; cess we11 enough to recognize which behaviors are
Clan; Behavior Behavior; and Clan appropriate (cell 1). For example, to ensure that the
Controller's Knowledge of the Transformation Process
users’ require ments are met, the controller can monitor
the ‹:oritr‹i1lee in a number of ways, including partici-
pating i n walk-throughs to review project progress, and
nt›ting the way in which the project leader incorporates
Behavior Ou‹ come a devel‹ipment method‹ilogy to structure the ISD task.
observability
lf t he t ‹introller is knowledgeable about ISD, she can
observe these types of activities, evaluate whether they
measurability
are .ipJ›ropriate, and reward or sanction the project
leader accor‹1ingly. However, if the controller is not
w B v ha yGOF
observability knt›wle‹âgc able, it is unlikely that she can make sense
o1 and ev‹iluiite what she observes; thus, it is unlikely
Cell 4 Cell 7 she will clepcnd on behavior control mechanisms to
High behavior
regulate the behavior ot the project leader. Thus, when
Behavior Clan
observability behavior ribservability is high, that is, when information
IAWOUtOOm
about the ISD task is available, the controller will
Cell 4 implement behavior control only if she is knowledge-
able about the systems develc›pment process (cells 1
Self
anti 3). Thus. Figure 2 suggests that behavior observ-
‹ bservab›Iiry
ability antl controller’s knowledge of the tr‹insforma-
tion process interact to determine the level of behavior heavy reliance on controller—controllee interaction that
control: leads to a sense of trust and commitment between the
HYPOTHESIs 1. Behavior observability and controller’s parties. Thus, Figure 2 predicts that the use of clan
knowledge of the transformation process haue a positive control increases as behavior observability increases, as
interaCtiue effect on the leuel of’ behauior control. outcome measurability decreases, and as controller’s
knowledge decreases. The model suggests that these
Since the controller in cell 5 is not knowledgeable three independent variables are not only related to
Downloaded from informs.org by [131.94.16.10] on 28 August 2015, at 18:31 . For personal use only, all rights reserved.
about ISD, behavior control is not an option. Control clan control, but interact to determine the level of clan
theories argue that when outcomes are measurable, control:
controllers can use outcome-based control, regardless HYPOTHESis 3. Behavior observability, outcome
of their knowledge of the transformation process. For mea- surability, and controller’s knowledge of the
example, a controller does not have to understand the transforma- tion proces.s haue an interactive effect on
intricacies of ISD to know whether the system was the leuel of clan control. In particular, high levels of
implemented on time. Thus, in cell 5, the controller is behavior obseruahil- ity, low levels of outcome
more likely to use her ability to measure outcomes and measurability, and low leuel.s of controller's knowledge
structure an outcome-based contract, passing on some of the transformation process lead to increased use of
risk to the project leader. Further, consistent with clan control.
control theory predictions and previous empirical work
(e.g., Snell 1992), the controllers in cells 2 and 6 will Finally, consider cells 4 and S. Prior work predicts
most likely use outcome control because outcomes are the use of clan controls; however, it seems more likely
measurable. Based on these arguments, Figure 2 sug- that the controller will induce the project leader to use
gests that outcome control will be used in those cells self-control. Since behaviors are not observable and
where outcomes are highly measurable, except where outcomes are not measurable, behavior and outcome
both behaviors are highly observable and controllers controls are ruled out. Further, the low level of behav-
are very knowledgeable, in which case behavior control ior observability suggests the controller is not closely
will be used and unnecessary risk will not be passed to supervising or interacting with the controllee, resulting
controllees (i.e., in cells 2, 5, and 6, but not in cell 1): in less opportunity for clan-type conditions to evolve. It
also suggests that the controllee has much more discre-
HYPOTHESIS 2. Behavior observability and controller’s tion or freedom in choosing and managing his work
knowledge of the transformation process haue a negative (Alavi et al. 1986, Manz and Angle 1986), implying the
interactive effect on the leuel of outcome Control. presence of self-control. The absence of observable
outcomes means that a set of precise project goals has
Next consider cell 7. Outcome control is no longer
not been articulated. This, in turn, suggests that the
an option since outcomes are not measurable. While
controllee may self-set goals, again implying the pres-
behaviors are observable, they are not very meaningful;
ence of self-control. Thus, in cells 4 and 8, the con-
thus, behavior control is not an option. However, be-
troller is likely to rely on the controllee’s ability to
cause behavior observability is high, it suggests that the
manage his own work. Specifically, the model in Figure
controller has invested in information systems such as
2 predicts that self-control is a function of behavior
reports, meetings, and personal observation to follow
observability and outcome measurability. The model
the progress of the project. Given that behavior and
also implies that outcome measurability and behavior
outcome controls are not possible, this high level of
observability have an interactive effect on self-control,
behavior observability provides the controller with an
suggesting, for example, that low behavior observability
opportunity to implement clan control: she can use her
results in an increase in self-control only when out-
time with the controllee to explain group values, norms,
come measurability is low:
and problem-solving approaches. For example, when a
user works with the project leader to identify system HYPOTHESIS 4. Behavior obseroubility and outcome
requirements, to lay out screen and report formats, and measurability haue a negative tnteractiue effect on the
when she attends project meetings, she can sensitize level of self-control.
the project leader to the goals of the user community
and build a sense of trust between the user and IS. Methodology'
This is consistent with Beath’s (1987) argument that Research Design
successful social contracting is exercised through a While there are a number of stakeholders involved in
ISD, in this research, two key relationships are the
6
O RG AN1ZATI O N S Cl KN Cu/ Vol. 7, No. 1, January—February 199f›
target of study. the relationship between the systems participants. Each participant was contacted by the
development manage r and the project leader, and be- rcse arc’her, who conducted (on average) a 15-minute
tween the user contact and the project leader. The user teleyiht›nc interview. The purpose of the phone inter-
cc›ntact is the key lia isr›n person in the client area, who view w‹is primarily to obtain the person’s commitment
,resists in arious phases of the development process. to the research. since the questionnaires were quite
for example, require ments specification, testing, and long: 12 pages for the controllee, and 9 for the con-
implementation. The development m anager and user
troller. At the same time, relevant background infor-
Downloaded from informs.org by [131.94.16.10] on 28 August 2015, at 18:31 . For personal use only, all rights reserved.
tion systems development efforts prior to the current or clan controls are operating: with formal control,
effort. goals are typically set by a supervisor or management;
with clan control, goals evolve in concert with the clan’s
Instruments value set. Additional constructs from the sclf-control
New scales were developed to measure behavior ob- conceptual literature include: controllee self-monitor-
servability, outcome measurability, and controller’s ing; controllee self-evaluation, intrinsic motivatic n of
knowledge of the transformation process. In addition, controllee; and degree to which the controller rewards
Downloaded from informs.org by [131.94.16.10] on 28 August 2015, at 18:31 . For personal use only, all rights reserved.
a measure of how important the project goal is to the the controllee for exhibiting self-management. Thus,
controller (i.e., meeting user requirements; delivering measures for each of these four aspects, along with a
the system on-time; keeping costs within budget; ad- measure of the extent to which the controllee is the
hering to standards) was also developed. This construct source of the project goal, are combined ter indicate
was added after analyzing the pre-test data, since those how strongly self-control is operating.
preliminary results suggested that if the project goal Recall that what is of interest is the strength of
was unimportant to the controller, then she did not behavior, outcome, clan, and self-controls put in place
exert control at all. by each of the two controllers to ensure that the
Determining how to measure mode of control (be- project leader delivers a system that meets the four
havior, outcome, clan, and self) was especially trouble- project goals. Because the focus of the research is on
some since much of the prior research measures it in a multiple controller—controllee relationships and multi-
somewhat simplistic approach, often conceptualizing ple-project goals, the design of the questionnaires is
control as a unidimensional construct and measuring it complex. In particular, each construct is measured
with a single question. (See, for example, Ouchi multiple times by modifying the question for each goal
(1978) and Govindarajan and Fisher (1990).) In a and/or for each controller, as appropriate. To reduce
manner similar to Snell (1992), the approach taken threats from common source variance, the dependent
here was to recognize the multidimensional nature of variables were measured on the project leader (control-
control, con- structing composite measures. In lee) questionnaire, while the independent variables and
particular, for each mode of control, key conditions goal importance were measured on the controller (IS
were identified, the existence of which would strongly manager and user contact) questionnaire. Where possi-
suggest that a partic- ular approach was being used by ble, existing instruments were used; however, all items
the controller to regulate controllee behaviors. First, were modified to fit the ISD context. Specific items
since formal con- trols are written and management- and their source are in the Appendix.
initiated (Jaworski 1988), the measures of formal
behavior and outcome controls include an indication of Instrument Validation
the extent to which behaviors and outcomes are known
and prespecified by someone other than the controllee, Internal Consistency. Internal consistency was mea-
as well as an as- sessment of the degree to which sured by computing Cronbach’s alpha fer each scale.
organizational rewards are based on following The conventional cut-off for acceptable reliability is
prescribed behaviors or produc- ing desired outcomes 0.70 (Nunnally 1978). In Table 2, each scale is listed,
(i.e., the link between rewards and following behaviors along with the number of items measuring the con-
or producing outcomes). Sec- ond, when clan control struct, the sample size of the dataset used to compute
is operating, organizational goals are typically reliability, and the reliability. In addition, descriptive
unknown at the start of a project, but instead evolve statistics are included. Each item was measured on a
over time (Ouchi 1979). Further, the clan exhibits scale of one to seven.
shared values that are cooperative in nature (Birnberg Because self-monitoring (both feedback from the
and Snodgrass 1988). Given the simi- lar value set and task and feedback from agents) and self-evaluation
the desire to work together, an individual’s high level have reliabilities less than 0.70, these scales were
of commitment to the group is an indication that clan dropped. Conceptually, dropping these scales does not
controls are operating. Thus, the measure of clan mean that the essence of the self-control measure has
control combines measures for goal instability and been lost, as there was some overlap in the constructs
commitment to the clan. Third, when self-controls are being measured. In particular, if a person is highly
operating, the individual sets his own goals (Manz et intrinsically motivated (a scale that was retained), he is
al. 1987), which may or may not be formally rewarding or sanctioning himself. In order to reward or
documented. This notion of self-set goals con- trasts sanction himself, that person has to monitor and evalu-
sharply with the approaches taken when formal
’ ! he sample size depenr›s on how the variable was measured (e g once per pro ect, once per controller, once per goal) For some variables,
tne \›oss/D/e sample sizt is 35 projects x 4 goals x 2 controllers or 260 F‹›r ethers, it is 35 pro ents x 4 goals - 140, or 35 pro ects x 2
controllers 70 Actual s imole sizes differ slightly due t‹› trussing aata
error terms may be correlated, violating an assumption allows for the possibility that other mechanisms may be
underlying regression analysis. In particular, for this implemented to control for other project goals. Note
data, some of an individual’s responses with respect to that the argument here is not that project responsibtlities
the four goals within a single project are highly corre- are split along different potential goals, only that con-
lated, while others are not. The ranges of correlations trollers may implement different control strategte.s for
are: behavior observability: 0.12-0.65; outcome mea- the different project goals.
surability: — 0.01—0.34; controller’s knowledge of the The second reason for not analyzing the data at the
transformation process: 0.68—0.88; behavior control:
Downloaded from informs.org by [131.94.16.10] on 28 August 2015, at 18:31 . For personal use only, all rights reserved.