Organizational control and project performance

Abstract

The theory of organisation control developed by Ouchi and others focuses on the managers’ choice of individual control modes. Studies on the normative effects of control modes assumes the implicit normative effects of the Ouchi model and largely ignore the choice of a combination or portfolio of control modes and its effects on performance.

The emerging stream on the “balance” of control modes emphasize the importance of employing multiple control modes simultaneously but fails to partition the total effects into the aggregate effects on performance of individual modes and the effects of interactions among those modes. It, therefore, does not formally model the effect on performance of balance, defined in terms of interactions among control modes.

Here, a more complete and complex normative model of organization control is developed and validated. The model includes interactions among control modes and the effects on performance of those interactions. Surveying Australian construction contractors and information services companies, eight of the nine hypotheses comprising the normative model are supported with important implications for both theory and practice.

Keywords: Organisational Control, Control Modes, Output, Input and Behavior Control, Balance, Exploitative and Exploratory Learning, Organization Performance

Introduction

Organizational control theory (Ouchi, 1977, 1979; Eisenhardt, 1985; Snell, 1992) has frequently been applied in the context of both in-house information systems development (ISD) projects (Kirsch 1996; 1997; Henderson and Lee, 1992) and outsourced ISD projects (Kirsch, Sambamurthy, Ko, & Purvis, 2002; Choudhury & Sabherwal, 2003; Rustagi, King, & Kirsch, 2008). However, the dominant control framework suffers from a number of limitations and, as a result, could have constrained our understanding of the effects of control modes on performance.

The purpose of this paper is to identify the limitations of the control theory, develop a new framework that predicts the effect of control modes on performance, and, finally, elaborate how the framework can increase our understanding of how to effectively control ISD projects.

In practice, various control modes are typically implemented through organisation-wide processes such as PRINCE II and the Capability Maturity Model (CMM) developed by the Software Engineering Institute (SEI). Drawing on the literature of the effects of CMM implementation on IS project performance, we show that the new normative framework explains the findings better than the dominant control theory framework with important implications for practice.

Organizational Control Theory

The dominant model of organization control was developed by Ouchi and his colleagues in a series of papers beginning with Ouchi and McGuire (1975) and ending with Snell (1992). Ouchi and McGuire began by analyzing the conditions that govern the use of output and behavior control by managers. Ouchi (1977) developed the model and presented it in the familiar 2x2 matrix form, with dimensions of availability of output measures and knowledge of transformation processes. The model predicted the use of behavior or output control in the high/high cell and the use of ritual in the low/low cell. Ritual is replaced with clan control in Ouchi (1979, 1980) and with input control in Snell (1992). The resultant, frequently cited model is presented in Figure 1.

Inspecting Figure 1 highlights two issues. One is that the model proposes a contingent framework for managers’ choices of control modes, rather than of the effects of those choices on performance. The other is that the model specifies either a single mode of control or, in the high/high cell, a choice between two modes. Combinations of modes are not included in the model. These two issues are the basis of the two questions framing this research.

Figure 1. Choice of Organizational Control Mode.

First, the model presented in Figure 1 predicts the choices of control modes by managers. Its logic is elegantly simple. If a task is well understood, it is possible to specify the behaviors necessary to achieve the planned result. Thus, behavior control is appropriate. If it is easy to measure outputs, then outputs can be monitored and controlled to deliver the planned result. Thus, output control is appropriate. If neither of these conditions holds, input control is the default mode.

Although Ouchi (1977) claimed that the model is normative, it is only normative in the limited sense that rational managers would adopt an appropriate control mode given the context. Subsequent research on the model in Figure 1 and its derivatives typically assumes, rather than formally tests, the effects on performance of the control modes chosen. The limited empirical research reports results that are sometimes inconsistent with Ouchi's implicit normative framework. For example, while Cardinal (2001) reported, consistent with Ouchi's framework, that input control enhances innovation, she also reported, inconsistent with Ouchi's framework, that output control enhances innovation in R&D project management, where task uncertainty is high. Other studies show that overreliance on outcome control leads to gaming behavior, misrepresentation of performance (Bevan & Hood, 2006; Heinrich, 2002, 2007), and thus negative performance effects. These findings, combined with the importance of organization control, motivated the first of the two questions framing this research: How do control modes affect organization performance?

Second, combinations of control modes are not included in Ouchi's framework. Ouchi (1980) argues that markets, bureaucracies, and clans are three distinct mechanisms that operate independently of each other. However, the research stream based on the model in Figure 1 frequently comments that different control modes are applied in combination (see, for example, Henderson & Lee, 1992; Kirsch, 1996; Nidumolu & Subramani, 2003).

For example, Kirsch (1996) wrote: “While multiple types of control were measured in this research, this study was based on a tradition in the control literature of examining the relationship between a set of environmental characteristics and each mode of control. Consequently, the focus was not on how controls are used in combination in organizations” (Italics added for emphasis). She went on to comment that the correlations among control modes are significant and concluded: “These correlations suggest that, when control mechanisms are put in place, it is likely that multiple types of mechanisms are implemented.” This and similar speculations about combinations of control modes motivated the second question guiding this research: How do combinations of control modes influence organization performance?

Effects of Control Modes on Performance

To develop a more complete and complex normative model of the effects on performance of control modes than is implicit in Figure 1, this study draws from three research streams. The first includes the goal setting theory (Locke & Latham, 1990), management by objective (MBO) (Tosi & Carroll, 1968; Greenwood, 1981; Rodgers & Hunter, 1992), and outcome-based performance management (Bevan & Hood, 2006; Heinrich, 2002, 2007) to explain the effects on performance of output control. The second includes the organization learning theory (March, 1991) and process management (Benner & Tushman, 2003) to explain the effect on performance of exploitative and exploratory processes. The third is the literature on human resource management (HRM), which explains the effects on performance of input control (Snell, 1992; Delaney & Huselid, 1996; Evans & Davis, 2005).

First, goal setting affects performance in two ways (Locke & Latham, 1990). When task uncertainty is low, goal setting increases individuals’ motivation to perform, affecting the direction, persistence, and level of an individual's effort. This is called the goal motivation effect. In contrast, when task uncertainty is high, goal setting stimulates individuals to look for innovative solutions. This is called the goal strategy effect. Goal setting theory is one of the most rigorous and extensive research streams in management theory. Although its primary level of analysis is individual behavior, the MBO and outcome-based performance management literatures extend the goal setting approach to organization behavior and performance.

Second, organization learning theory also specifies different mechanisms by which processes affect performance when task uncertainty is high versus when it is low (March, 1991; Benner & Tushman, 2000). Exploitative behavior delivers high performance when task uncertainty is low and exploratory behavior delivers high performance when task uncertainty is high. The learning approach an organization adopts influences the choices and effects of control modes.

Third, the HRM literature examines various approaches to the development of human capital and their effects on organization performance (Evans & Davis, 2005; Delaney & Huselid, 1996). Essentially, its focus is on the link between input control and performance. This studies draws from a subset from this stream to develop the hypotheses on the link between input control and performance.

Low Task Uncertainty

In goal setting theory, when task uncertainty is low, performance is a positive function of the goal setting motivational effect on the level of effort, and its direction and persistence. Goals affect the intensity of effort an individual expends on a task. They also affect its direction and duration, motivating individuals to persist in their actions until the goal is reached. Goals motivate individuals to adopt goalrelevant actions, while ignoring non-goal-relevant activities (Locke & Latham, 1990). At the organization level of analysis, the management by objective (MBO) literature (Tosi & Carroll, 1968; Greenwood, 1981; Drucker, 1954; Rodgers & Hunter, 1992) and the literature on outcome-based performance management (Bevan & Hood, 2006; Heinrich, 2002, 2007) support the general assumption that setting goals increases incremental, continuous improvements in organization performance. Formally,

Hypothesis 1: When task uncertainty is low, performance is a positive function of output control.

In organization learning theory when task uncertainty is low, performance is improved through the exploitation of knowledge about existing practices (March, 1991). Adjustments to existing behavior improve mean performance and reduce performance variance (Benner & Tushman, 2003). This occurs through single loop learning. This learning leads to improvement in business processes and reduction of costs through standardization and routinization, i.e., behavior control (March, 1991; McGrath, 2001).

This exploitation effect on performance of process improvements relies on comprehensive knowledge, conditional on low task uncertainty, to enable the specification of effective design processes (Fredrickson, 1984; Pisano, 1994). In such situations, organization learning theory predicts that exploitative learning through comprehensive process management generates positive organization outcomes (Benner & Tushman, 2002, 2003; March, 1991; McGrath, 2001). Formally,

Hypothesis 2: When task uncertainty is low, performance is a positive function of behavior control.

Input control regulates the human resource-based antecedent conditions of performance, specifically, the knowledge, skills, abilities, values, and motives of employees (Snell, 1992). Approaches for enhancing employees’ skills include selection, training, and development activities (Delaney & Huselid, 1996; Evans & Davis, 2005). Various conceptual frameworks, including agency theory, resource-based view, institutional theory, to name a few, have been used to explain the link between HRM practices and organization performance (Delaney & Huselid, 1996). Empirical results show that staff selectivity is related to organization level performance (Becker & Huselid, 1992, Schmidt, Hunter, McKenzie, & Muldrow, 1979). Training and development activities improve organization performance (Bartel, 1994; Knoke & Kalleberg, 1994). Formally,

Hypothesis 3: When task uncertainty is low, performance is a positive function of input control.

Now, consider the interactions between those control modes. First, examine the interaction between behavior and output controls. When behavior control is weak, processes are not monitored and reinforced by management, non-optimal behavior develops. In those circumstances, increases in output-based motivation, the goal setting motivation effect, have a limited positive effect on performance. Instead, the increased motivation reinforces the emergent non-optimal behavior. Thus, when behavior control is weak, the effect on performance of output control is limited.

In contrast, when behavior control is effective, increases in output-based motivation reinforce the effective behavior and, therefore, have strong positive affects on performance. This is because process improvement typically involves the use of process effectiveness measures and statistical methods based on goals or output control for variation reduction in processes and outputs (Garvin, 1995; Harry & Schroeder, 2000). This is consistent with Benner and Tushman (2003), who, implicitly but not explicitly, modeled the effect of exploitative behavior to include an interaction between, rather than simply an additive function of, output and behavior control.

Specifically, Benner and Tushman (2003) argued that business process management approaches typically involve three practices: process mapping, process improvement, and adhering to the improved processes. During the mapping process, the functions of a business entity are defined, the responsible person is identified, and the expected standard and success criteria, or goals, for a business process are specified (Deming, 1982; Juran, 1988). The mapped and improved processes need to be adhered to reap potential business benefits from repeated processes, e.g., reliable measurement data for and continuous improvement in outputs (Mukherjee, Lapre, & Van Wassenhove, 1998). Formally,

Hypothesis 4: When task uncertainty is low, behavior control moderates the effect of output control on performance.

Second, consider the interaction between output and input control. From the goal setting theory, output control increases performance through its effect on extrinsic motivation (Hypothesis 1). Input control, by selecting for and improving employees’ ability to learn and work independently, increases performance through its effect on intrinsic motivation (Hypothesis 3). Drawing on cognitive evaluation theory (Anderson & Oliver, 1987), total motivation is less than the sum of the expected independent intrinsic and extrinsic effects.

Deci and Ryan argued that high extrinsic motivation erodes high intrinsic motivation because certain forms of output control, including negative performance feedback and deadlines, decrease intrinsic motivation (Deci, 1972; Ryan & Deci, 2000). So, when the effect of input control on intrinsic motivation is combined with the effect of output control on extrinsic motivation, the interaction of the two modes of control has less effect on performance than the sum of their expected individual effects. Formally,

Hypothesis 5: When task uncertainty is low, output control moderates the effect of input control on performance.

In our review of the literature, we did not identify a theoretical model of or compelling speculation for an interaction between behavior and input control, and the effect of that interaction on organization performance. So, no hypothesis is developed and presented here.

High Task Uncertainty

Now, consider the context when task uncertainty is high. The basic argument for the use of input control is well stated by Snell and Youndt (1995). Drawing on the work of researchers such as Dimmick and Murray (1978), Zajack (1990) and Russell et al. (1985), an HRM control system based on input control, including rigorous staffing, training, and induction, has a positive effect on organization performance. Specifically, Cardinal (2001) found that input control enhances innovation for the typically uncertain R&D projects.

Snell and Youndt (1995) noted that the positive effect of input control, when task uncertainty is high, is consistent with the arguments in the knowledge management literature on the transformation to flexible, knowledge-based organizations (see, for example, Nonaka, 1991). This is particularly the case in high task uncertainty contexts, where, unable to rely on past successful behavior, high performance depends on an organization's knowledge base (Eisenhardt & Bourgeois, 1988). Similarly, Nonaka and Takeuchi (1995) argued that it is the redundant and overlapping information sets in a knowledge base that support organization innovation and variety. Input control, through selection, training, and induction, helps develop those knowledge sets and supports access to them by developing and reinforcing values that support increased communication and interaction (Flamholtz, Das, & Tsui, 1985; Turner & Makhija, 2006).

Increased interaction gives managers access to diverse knowledge (Hoopes & Postrel, 1999). This knowledge is typically tacit rather than explicit. The ability to exchange that knowledge facilitates its recombination, generating new insights (Galunic & Rodan, 1998). Those insights, generating innovative actions, are the basis of exploratory learning (March, 1991; Turner & Makhija, 2006).

Organization learning theory involves four distinct processes, intuiting, interpreting, integrating, and institutionalizing, originating from the individual level and subsequently progressing to the group and organization levels, respectively (Crossan, Lane, & White, 1999). First, an individual perceives patterns and possibilities (intuition). There are two types of intuition—expert (past pattern recognition) and entrepreneurial (making novel connections and discerning new possibilities). The former relies on past experience and deliberate thought, and supports exploitative learning, whereas the latter is largely a subconscious process and conducive to exploratory learning. Second, the individual clarifies and defines the intuitive insight and develops cognitive maps through an interpretive process (Weick & Van Orden, 1990). Third, through continued dialogue and practice, a shared understanding of the insight develops in the community of practice. Finally, the knowledge becomes embedded in the way the organization conducts its business, i.e., reflected in its routines, structure, and culture.

Under conditions of high uncertainty, experiential learning becomes a crucial part of the development process such as on-the-job training, job rotation, and mentoring (Swap, Leonard, Sheilds, & Abrams, 2001). In contrast to low uncertainty conditions, where the development is likely to be focused on learning past experience and the application of known, stable work processes, development in high task uncertainty situations, that supports exploratory learning and facilitate entrepreneurial intuition, e.g., experiencing and understanding work processes, problem solving, and work languages, impact positively on organization performance.

Consistent with the extant control theory (Cardinal, 2001; Ouchi, 1979; Snell, 1992) and the HRM literature (Becker & Huselid, 1992; Schmidt et al., 1979; Bartel, 1994; Knoke & Kalleberg, 1994; Russell et al., 1985), the selection, development, and training of employees has a positive effect on firm performance. Formally,

Hypothesis 6: When task uncertainty is high, performance is a positive function of input control.

It is implicit in the dominant model of organizational control theory in Figure 1 that, under conditions of high task uncertainty, both output and behavioral controls have weak or no effects on performance. Now, while Hypothesis 1 is based on the goal motivation effect when task uncertainty is low, when task uncertainty is high, performance is a positive function of the goal strategy effect. Specifically, challenging goals stimulate the search for innovative ways to meet the targets (Locke & Latham, 1990, p. 96).

The discovery of new task strategies is also consistent with exploratory leaning processes, which involve risk taking, experimentation, and innovation (Benner & Tushman, 2003; March, 1991). Adaptation through exploratory learning processes fits high task uncertainty contexts, where changes are unpredictable (Eisenhardt & Martin, 2000). In those contexts, successful performance depends upon generating a sufficient number of novel solutions or options so that some are successful (McGrath, 2001). In this way, specific, challenging goals stimulate the process of exploratory learning to identify options. In addition, when performance is below their aspiration levels, decision makers take on more risks (Abrahamson, 1996; Abrahamson & Fairchild, 1999; Tversky & Kahneman, 1986), engaging in exploratory search processes (March, 1991) and experimenting with new processes, technologies, and strategies (Baum, Rowley, Shipilov, & Chuang, 2005).

Empirical evidence from studies of software development (Choudhury & Sabherwal, 2003; Nidumolu & Subramani, 2003) and pharmaceutical R&D projects (Cardinal, 2001) provides further support that output control has a positive effect on performance when task uncertainty is high. The literature on software project risk management also reports that output control has a positive effect on performance (Barki, Rivard, S. & Talbot, 2001; Rustagi et al., 2008). Both studies report that the use of output controls, including PERT and CPM to monitor project status, is associated with improved performance. Formally,

Hypothesis 7: When task uncertainty is high, performance is a positive function of output control.

Now, consider the effect of behavior control on performance. Essentially, behavior control is the basis of standard operating procedures (SOPs). The effective application of SOPs requires that organizations have complete understanding of the interdependences among the tasks to deliver high performance (Eisenhardt, 1985; Snell, 1992; Turner & Makhija, 2006). Although this condition is typically satisfied when task uncertainty is low, it is unlikely to be satisfied when task uncertainty is high.

Consistent with this, Ittner and Larcker (1997) reported a positive effect of behavior control on performance in the low task uncertainty auto industry of the ‘90s (see Hypothesis 2) but no relationship in the dynamic, high task uncertainty, computer industry. This absence of a positive effect of behavior control on performance when task uncertainty is high is consistent with its implicit null effect in the dominant organizational presented control model in Figure 1. Formally,

Hypothesis 8: When task uncertainty is high, performance is independent of behavior control.

Finally, consider the interaction between input and output controls when task uncertainty is high. As argued in Hypothesis 7, when task uncertainty is high, output control increases performance by improving the task strategy selected. This is the goal strategy effect (Locke & Latham, 1990) rather than the goal motivation effect that underpins Hypothesis 1.

Therefore, when task uncertainty is high, unlike the case when task uncertainty is low (Hypothesis 5), there is no interaction between input and output controls on performance. This is because the intrinsic motivation effect of input control on performance is independent of the strategy selection effect of output control on performance. Therefore, the effects of input and output controls on performance are additive. Formally,

Hypothesis 9: When task uncertainty is high, input and output controls have independent effects on performance.

Research Design

The hypotheses are tested using survey data collected from senior managers in the Australian construction and information services industries. Both industries have engineering as a common functional background; they conduct projects for other organizations on a contract basis; and they are comparable in relation to their practices in managing projects. Project-based business makes a good testing ground for organizational control theory because projects reflect organizational rather than individual behavior but are sufficiently discrete that performance is visible.

The two industries were chosen because of the wide variance in expected levels of project performance (Sauer, Liu, & Johnston, 2001) and the differences in the contextual conditions under which they operate (Shenhar & Dvir, 2007). A similar justification is presented by Ittner and Larcker (1997) in their comparison of the auto and computer industries.

The survey instrument was customized to each industry, with minor variations introduced to accommodate differences in terminology. The questionnaires were pilot-tested with six senior managers from each industry. The feedback from the respondents was used to improve the face validity and the relevance of the questions.

Unit of Analysis

There are recognizable differences in management control systems and information control capabilities between construction and information services business units. Conversely, the differences are small in management control systems across projects within business units (Sauer et al., 2001). The business unit was, therefore, selected as the unit of analysis to test the hypotheses.

Sample Selection and Data Collection

Two hundred and thirty-two (232) Australian construction companies were identified, including general, heavy construction, and special trade contractors, including Plumbing, Heating and Air Conditioning, Electrical, Masonry and Insulation Work, Roofing, Siding and Sheet Metal Work, Concrete Work, Structural Steel Erection, Excavation Work, Wrecking and Demolition Work, within the definition of the Standard Industry Classification (SIC) code, 1521-1799, with annual turnover exceeding US$30 million. Similarly, 224 Australian information services companies were identified within the definition of SIC 7371-7379, with annual turnover exceeding US$7.5 million.

A fax was sent to each CEO seeking a list of senior managers who could potentially participate in the survey. Questionnaires were sent to the 67 senior managers identified in the construction industry. Fifty-seven of the 67 completed and returned their questionnaires. Fifty-two senior managers were identified in the IS services industry and were sent questionnaires. Thirty-nine completed and returned their questionnaires. Multiple responses from the same business unit were aggregated, using the mean of the responses, resulting in a usable sample of 90, with 54 responses in the construction sub-sample and 36 in the IS sub-sample.

To evaluate the potential response bias validity threat, the representativeness of each sample was examined. To do this, the populations defined by the relevant SIC codes were identified using Dun & Bradstreet's “Business Who's Who of Australia” database. The distribution of the variable—number of employees—was determined for each population. The populations were then partitioned into deciles and the observed and expected frequencies in each decile for the samples and the populations were compared using a Chi-Square test. There is no evidence of a non-response bias for either sample (Construction: χ² = 1.35, df = 9, Asymp. Sig. = 0.99; Information Services: χ² = 0.55, df = 9, Asymp. Sig. = 1.00).

Instrument Design and Validation

Project performance reflects a business unit's capability to satisfy expectations concerning cost, time, and quality. It is measured in terms of the respondent's perceptions of overall performance, relative performance against competitors, and client satisfaction on a 9-item, 5-point Likert scale (see Appendix 1). The measure of project performance exhibits acceptable reliability (α = 0.85).

Output control is operationalized as an emphasis on targets (Cardinal, 2001; Snell, 1992), reward-outcome links (Cardinal, 2001; Kirsch, 1996) and pre-specified targets (Kirsch, 1996). Each of these three perspectives is measured in this study, where output control is measured as the use of specific output targets, reporting performance results, attention to measurable results, and holding subordinate managers accountable. The items have been adapted to management at the business unit level, with output control measured on a 4-item, 5-point Likert scale with acceptable reliability (α = 0.72).

Behavior control is operationalized as pre-specified behavior (Kirsch, 1996), reward-behavior link (Cardinal, 2001; Kirsch, 1996) and corporate procedures (Cardinal, 2001; Snell, 1992). Each of these dimensions is measured with the wording adapted to management at a business unit level. Behavior control is measured on a 4-item, 5-point Likert scale with acceptable reliability (α = 0.75).

Input control is operationalized as staff selection and skill development (Cardinal, 2001; Snell, 1992). Following Sauer et al. (2001), project manager development is included as contributing to input control. For example, in successful construction firms, potential project managers are carefully nurtured. The business units provide on-the-job-training for their project managers; project managers are routinely rotated through different roles; they are assigned a mentor; their responsibility is gradually increased with experience; and project managers’ appointments and responsibilities are typically based on past performance. Non-performers are selected out. Such initiatives are commonly recommended as part of project manager development (Graham & Englund, 1997).

The pre-test of the survey instrument identified that different processes were followed in the two industries to implement input control. So, input control [construction] is measured on a 5-item, 5-point Likert scale with low and marginally acceptable reliability (Straub, Boudreau, & Gefen, 2004) for an exploratory study (α = 0.59); and input control [information services] is measured on a 4-item, 5-point Likert scale with acceptable reliability (α = 0.74). The question concerning on-the-job-training was deleted for the IS sample, increasing a from 0.55 to 0.74. The potential construct validity threat contingent on the reliability of the measure of input control in the construction industry is addressed in the Discussion section.

Analysis

In this study subgroup analysis is chosen in preference to MRA in the combined sample for four reasons. First, the data structure is fundamentally industry or subgroup based, with random sampling from the information services and construction industries. The two industries were selected to maximize the difference between the levels of task uncertainty in the two subsamples.

Second, when using MRA, the power to detect a moderator effect decreases with increases in the relative sizes of the two subsamples (Stone-Romero, Alliger, & Aguinis, 1994). In this study, there is a significant difference in the relative sizes of the two subsamples, with sample sizes of 54 and 36 for the construction and information services industries, respectively.

Third, adopting subgroup analysis over MRA increases the predictive power when the error variances in the subsamples are more homogeneous than in the combined sample (Prescott, 1986). MRA also has low power to detect a moderating effect when the moderator co-varies with the error term. In this case, the measure of performance is a function of the level of task uncertainty and there is a significant difference in performance between the two subgroups. So, again, subgroup analysis is preferred over MRA.

Finally, and most importantly, the theoretical forms of the relationships between, for example, output control and performance, when task uncertainty is high versus low, are different. The relationship in the former case is based on the identification and selection of an effective strategy to manage projects. In contrast, the relationship in the latter case is based on changes in the motivation to manage projects. To adopt MRA would combine apples with oranges.

For the four previously discussed reasons, subgroup analysis is adopted and is conducted in two stages. First, Hypotheses 1, 2, 3, 6, 7, and 8, the individual effects of output, behavior and input control modes on performance, are tested. For example, the effect of output control on performance when task uncertainty is low (Hypothesis 1) is tested by estimating Equation 1 for the construction subsample. A significant positive ₁ supports Hypothesis 1: When task uncertainty is low, performance is a positive function of output control.

The moderating effects (Hypotheses 4, 5, and 9) are tested by estimating Equation 2 in each of the construction and information services subsamples (Zedeck, 1971; Sharma, Durand, & Gur-Arie, 1981), where X₁*X₂ is the interaction term modelling the moderator effect of X₁ on the relationship between X₂ and performance.

For example, see Hypothesis 4: When task uncertainty is low, behavior control moderates the effect of output control on performance. A significant positive ₃ indicates that X₁, behavior control, moderates the effect of X₂, output control, on performance.

The tests for Hypotheses 4, 5, and 9 are subject to potential internal validity threats from the presence of multicollinearity if the interaction term X1*X2 is highly correlated with either or both X1 and X2. In that case, the standard errors for the estimates of the affected regression coefficients are inflated, supporting invalid conclusions with respect to the hypothesized relationships (Chatterjee, Hadi, & Price, 2000). Tolerance and the Variance Inflation Factor (VIF) are the two most frequently employed diagnostic statistics for the presence of multicollinearity. For simplicity, this study reports only VIF with VIF => 4 used as the cutoff point above which a regression model is considered to be subject to a significant internal validity threat. Some researchers suggest using higher cutoffs such as 10.0 (Chatterjee et al., 2000).

Many authors recommend mean centering variables to mitigate potential multicollinearity problems (Jaccard, Turrisi, & Wan, 1990). In this study, all variables used to estimate Equation 2 are standardized and the interaction term is the product of the two standardized independent variables specified in the regression model. Standardizing the variables also enables the comparison of results across the two contexts. The critical assumption underlying the subgroup analysis is that industry differences are a surrogate variable for differences in task uncertainty. The findings are, therefore, subject to a potential construct validity threat, which is reviewed in the Discussion section.

Results

For ease of presentation, the results are reported in two steps. First, the findings for the individual effects on performance of control modes are presented. Second, the analysis testing the effects on performance of interactions among control modes is presented.

Individual Effects of Control Modes on Performance

Table 1 reports the findings for the effects of individual control modes on performance. First, consider the main effects of output control on performance. Hypothesis 1 is supported: When task uncertainty is low, performance is a positive function of output control (construction subsample: = 0.47, p<= 0.01). Hypothesis 7 is also supported: When task uncertainty is high, performance is a positive function of output control (information services subsample: = 0.45, p<= 0.01).

Second, consider the main effects of behavior control on performance. Hypothesis 2 is supported: When task uncertainty is low, performance is a positive function of behavior control (construction subsample: = 0.32, p<= 0.05). Hypothesis 8 is also supported: When task uncertainty is high, performance is independent of behavior control (information services subsample: = 0.14, n.s.).

Third, consider the main effects of input control on performance. Hypothesis 3 is supported: When task uncertainty is low, performance is a positive function of input control (construction subsample: = 0.31, p<= 0.05). Hypothesis 6 is also supported: When task uncertainty is high, performance is a positive function of input control, (information services subsample: = 0.42, p<= 0.01).

Table 1: Main Effects of Control Modes on Performance.

Interactions Between Control Modes

The results for Hypotheses 4, 5, and 9, which specify the effects on performance of interactions among control modes, are presented in Table 2. Hypothesis 4 is supported: When task uncertainty is low, behavior control moderates the effect of output control on performance (construction subsample: (BC*OC) = 0.24, p<=0.06). The VIF values for the three independent variables range from 1.00 to 1.32. Therefore, there is no internal validity threat to the findings from multicollinearity.

Hypothesis 5 is also supported: When task uncertainty is low, output control moderates the effect of input control on project performance (construction subsample: (OC*IC) = -0.28, p<=0.05). The VIF values for the three independent variables range from 1.03 to 1.10. Therefore, there is no internal validity threat to the findings.

Table 2: The Moderator Effects of Control Modes on Performance.

Hypothesis 4		Output Control	Behavior Control	OC*BC
		0.46 (p<=0.01)	0.03 (n.s.)	0.24 (p<=0.06)
	VIF	1.32	1.32	1.00
Hypothesis 5		Output Control	Input Control	OC*IC
		0.37 (p<=0.01)	0.24 (p<=0.05)	-0.28 (p<=0.05)
	VIF	1.10	1.10	1.03
Hypothesis 9		Output Control	Input Control	OC*IC
		0.27 (n.s.)	0.21 (n.s.)	-0.16 (n.s.)
	VIF	1.30	1.70	1.38

Note: The significance level reported here is based on two-tailed t test.

Finally, Hypothesis 9 is supported: When task uncertainty is high, input and output controls have independent effects on performance (information services subsample: = -0.16, n.s.). The VIF values for the three independent variables range from 1.30 to 1.70. Again, there is no internal validity threat to the findings.

Industry Differences

There is an industry effect on performance, with project performance in the construction industry significantly higher, as expected, than in the information services industry (Mean difference=0.57, t=4.7, p<=0.001). Project performance in the construction sample is about one standard deviation above that in the information services sample (sd=0.57 in the construction sample; sd=0.55 in the information services sample).

Discussion

The discussion begins with a description of the findings. These are subject to a number of potential validity threats. Four are assessed. Then the implications for theory and practice are reviewed. The former focuses on the adoption of multiple modes rather than a single mode of control. The latter, in addition to evaluating the adoption of a portfolio of controls, speculates about the effect on performance of reengineering the context.

Findings

Consistent with the dominant model of organizational control presented in Figure 1, the results reported in Table 1 show that output (Hypothesis 1) and behavior (Hypothesis 2) control have a positive effect on performance when task uncertainty is low; input control (Hypothesis 6) has a positive influence on performance when task uncertainty is high; and performance is independent of behavior control when task uncertainty is high (Hypothesis 8).

While supporting the implicit normative model of organization control in Figure 1, the results also show that the relationships between control modes and performance are more complex than specified in that model. Critically, the results show that there are significant effects of output (Hypothesis 7) and input (Hypothesis 3) control on performance when task uncertainty is high and low, respectively.

In addition to these previously un-investigated effects of individual control modes, the results report the effects on performance of interactions among control modes. Specifically, when task uncertainty is low, behavior control moderates the effect of output control on performance (Hypothesis 4) and output control moderates the effect of input control on performance (Hypothesis 5). Finally, output and input controls have independent effects on performance when task uncertainty is high (Hypothesis 9).

Validity Threats and Limitations

The results are subject to three potential validity threats. First, differences other than task uncertainty, which covary with the industry context, could account for the effect of context on performance. Second, results of the tests for Hypotheses 3, 5, and 6 are subject to a potential Type 2 validity threat due to the low Cronbach alpha for input control in the construction subsample. Third, the results reported for Hypothesis 9 in Table 2 raise a potential internal validity threat. Finally, there is a potential validity threat to generalizing the findings to non-project-based organizations.

First, subgroup analysis (Venkatraman, 1989) is employed to test the moderating effects of the context on the effects on performance of organization control modes. Therefore, model specification error is a potential construct validity threat to the findings, if differences between the two subgroups, other than differences in task uncertainty, moderate the effect on performance of organization control modes. This potential validity threat should be addressed in future research.

However, two factors mitigate this validity threat. Recall, the construction and information services industries were selected for two reasons. One was that prior evidence suggested that performance is different across the two industries, with performance expected to be higher in construction as compared with information services. This is confirmed by the findings reported above. The other reason was that the similarities between the two industries with their engineering background and focus on delivery of projects to clients control for other potential construct validity threats. With the consistent empirical support presented for the nine hypotheses, we conclude that both the framework developed and the findings reported here are robust against this potential construct validity threat because any alternative explanation needs to account for a complex set of findings.

Second, the Cronbach alpha of 0.59 for input control in the construction subsample almost satisfies Straub et al.'s (2004) recommended cut off of 0.6 for exploratory studies. The potential validity threat to Hypotheses 3, 5 and 6 is that a Type II error (a true relationship is rejected) is made. This is not the case here: All three hypotheses are supported.

Third, the business units, which are the subjects in this study, are strategic business units in construction and IS project-based organizations. It is possible that, while the results hold for other similarly structured, project-based organizations, they do not hold for organizations in general. For example, the findings for output control in low task uncertainty contexts may be more important in project-based organizations, with their focus on in-budget, on-time and to-specification, than for other forms of organizations. Against this argument, all organizations are increasingly using projects to deliver new products and services, and implement organizational change. However, this threat should also be the subject of future research.

Balance Across Control Modes

The effect on performance of interactions among control modes is the motivation for a critique of the concept of balance in the use of control modes. The limited extant research on the effect on performance of combinations of control modes is limited to simply aggregating their joint effects without specifying the functional form of those effects. Exploring the dynamics of control modes, Cardinal, Sitkin, and Long (2004) analyzed the effect of reducing the emphasis on output and input control, interpreted as unbalancing the control modes, in phase two of a start-up furniture removal business. The use of behavior control on its own resulted in a major drop in performance. Reintroducing output and input controls, interpreted as rebalancing the control modes, reversed this loss of performance.

Cardinal et al. (2004) defined balance as “a state where an organization exhibits a harmonious use of multiple forms of control.” They offer no definition of “harmonious” independent of the effect on performance of the combination of output, behaviour, and input controls adopted in phase three. Nor do they consider in what context that combination of controls would not be harmonious or, at least, would not be successful.

The theory presented in Figures 2a and 2b offers insights into both of those issues. Figure 2a, when task uncertainty is low, models performance as a function of output and input controls, and of the interactions between output and behavior controls, and input and output controls. Balance requires the effective combination of output and behavior control. A comparison of performance against the goals provides feedback to incrementally refine the behavior controls and capture the benefits of exploitative behavior (Benner & Tushman, 2003). At the same time, the focus on output control should not be too strong; otherwise it would risk a loss of intrinsic motivation from the interaction of output and input controls.

The definition of balance is specific to the low task uncertainty context. Balance in a high uncertainty context is simply the additive effects on performance of output and input controls in Figure 2b. These findings are consistent with the Nidumolu and Subramani (2003) and Choudhury and Sabherwal (2003) findings.

Nidumolu and Subramani (2003) reported no effect on performance of standardization of methods, behavior control, but a strong effect on performance of standardization of performance criteria, output control. They also reported a positive effect on performance of decentralization of methods, dependent on input control.

Similarly, Choudhury and Sabherwal (2003) reported the importance of output and input (selection of vendor capability) control on outsourcing software development performance. When performance was unsatisfactory in the five projects they researched, both forms of control were increased. Behavior control was not considered to be an effective control mode in this context because behavior observability was low.

Implications for Practice

Organization performance can be improved by changing the portfolio of control modes or reengineering the context. We draw four messages for practicing managers from the results reported above. First, under conditions of high task uncertainty, behavior control may confer the illusion of control in a challenging and uncertain environment. However, it does not improve performance. Large IT projects represent the classic instance of this false assumption. Methodologies, specifying detailed sets of procedures, are rigorously followed with frequently disappointing results. Basing control theory on a foundation of goal-setting and learning theories helps explain this lesson.

Second, the performance achieved through reliance on a single mode of control can always be improved upon by the adoption of other control modes, with the exception of behavior control under high uncertainty. In particular, organizations that rely solely on aggressive output targets or bureaucracies that rely exclusively on behavior controls should be encouraged to adopt a portfolio of controls appropriate to the context.

Third, it is important to recognize that different mixes of controls support different strategies in different contexts. This requires fine tuning the mix of controls to the context, capturing the benefits of exploitation in low task uncertainty contexts and exploration in high uncertainty contexts. Therefore, when attempting to realize the benefits of exploitation, it is appropriate to implement strong behavior controls, but the output targets should not be so challenging as to shift the strategy from one of exploitation to one of exploration. Conversely, in the case of an exploration strategy under high uncertainty, conservative targets may fail to adequately motivate the search for innovation. In goal setting terms, exploitation sets high but achievable goals, consistent with processes including TQM and six sigma. Exploration sets stretch goals.

Finally, to reiterate a point made earlier, exploration and exploitation need to be managed under distinct control regimes. To attempt to achieve both using the same blend of controls is counterproductive. Abernathy (1978) presented an early critique of the idea that organizations should focus on both cost reductions and innovation.

Likewise, most organization theorists consider that exploration and exploitation are competing rather than compatible organizational processes (Benner & Tushman, 2003; March, 1991; Teece, Pisano, & Shuen, 1997). As Benner and Tushman (2003) concluded: “To create dual organization structures, senior teams must develop techniques that permit them to be consistently inconsistent as they steer a balance between the need to be small and large, centralized and decentralized, and focused on both the short and the long term, simultaneously.” The different portfolios of control in Figures 2a and 2b help explain why these strategies are incompatible rather than complementary.

The results reported above also show that, in the short term, managers should adapt their portfolio of controls to fit the context. In the longer term, an alternative is to transform a high task uncertainty context into a low task uncertainty context. This is the transition typically followed as an industry matures. The benefit of such a shift is the higher performance achievable in a low task uncertainty context compared with a high task uncertainty context.

In addition, organization contexts are characterized by cycles of technological variation, alternating between periods of incremental change and periods of rapid innovation (Tushman & Anderson, 1986; Uzumeri & Sanderson, 1995). A period of discontinuous innovation ends when a new dominant design emerges (Abernathy, 1978; Anderson & Tushman, 1990; Benner & Tushman, 2003).

This describes a cycle from low task uncertainty to high task uncertainty, and back to low task uncertainty, as the market moves from a dominant design through a period of innovation until a new dominant design emerges. This suggests the interesting question of whether it is more effective to retain the mix of control modes appropriate for a low task uncertainty context, while the market shifts to a high task uncertainty context. Then, when the new dominant design is identified, adopt a fast follower strategy. Alternatively, the organization could set up a subsidiary with controls appropriate for a high task uncertainty context and roll it back into the parent organization as the new market matures. This should be the subject of future research.

© 2010 Project Management Institute