2. Literature Review and Hypotheses

Although Williamson (1963) embeds managerial discretion in an economic theory of the firm using the notion of expense preference, he does not offer a definition of the term. However, it is implicit in his work that managerial discretion is viewed as the latitude of managers to pursue their own (non-profit-maximising) objectives, particularly in terms of channelling the firm’s monopoly profits to discretionary expenses that benefit the management, such as top management compensation. Williamson (1963) finds some empirical evidence that is consistent with this view, i.e. that given opportunities for high discretion (e.g. high entry barriers and high internal representation on the board of directors), discretionary expenses tend to be higher.

Some studies on principal-agent theory explicitly mention discretion and define it in a way consistent with Hambrick and Finkelstein’s (1987, pp. 371–378) definition, such as defining discretion ‘as managers’ decision-making latitude’ (Chang and Wong 2003, p. 2) or as ‘control rights’ (Shleifer and Vishny 1997, p. 742).

Managerial discretion is sometimes abbreviated by the term ‘discretion’ in the present study. The term ‘middle management discretion’ used herein therefore refers to the managerial discretion of middle management.

Glaister et al. (2003) find empirical evidence that the managerial discretion a manager perceives for himself/herself may differ from the discretion that his/her superiors perceive. In particular, in their sample of UK-European joint ventures, perceptions of managerial discretion of the joint venture management differ between the joint venture management itself and the parent firms as well as between each of the parent firms.

As explained in Sects. 2.3.1 and 2.3.2, Finkelstein and Peteraf (2007, pp. 237–243) incorporate the assumption of post-contractual asymmetric information (i.e. hidden action) from principal-agent theory (e.g. Eisenhardt 1989, p. 59; Jensen and Murphy 1990, p. 226; Khanchel 2009, p. 97; Levinthal 1988, p. 153; Spremann 1987, p. 3; Van Slyke 2007, p. 162; Werner and Tosi 1995, p. 1673) into managerial discretion theory. They argue that different characteristics of managerial activities affect the ability of key stakeholders (i.e. the powerful parties) to pre-specify and monitor the manager’s work, thus creating or constraining discretion. Asymmetric information (i.e. the inability to monitor the manager’s actions) therefore widens the ‘zone of acceptance of powerful parties’ (Hambrick and Finkelstein 1987, p. 378). E.g. if top management at corporate headquarters in China could not properly monitor the plant manager’s actions, the plant manager might undertake capital investments out of self-serving interests that reduced performance, which top management might have to accept due to their inability to monitor the plant manager’s action (Spremann 1987, p. 10).

In addition to discretion having a potentially important impact on performance (see above), it has been empirically demonstrated that discretion may significantly affect managerial power (Carpenter and Golden 1997), managerial compensation (Finkelstein and Boyd 1998; Magnan and St-Onge 1997; Rajagopalan and Finkelstein 1992; Werner and Tosi 1995; Wright and Kroll 2002; Zhang and Xie 2008), workers’ incentives (Groves et al. 1994), a successor chief executive officer’s age (Wang 2009), top management team tenure, trust (Perrone et al. 2003), strategic attention (Abrahamson and Hambrick 1997), environmental commitment (Aragon-Correa et al. 2004), pricing (Cameron 2000), organisational knowledge creation (Oh 2002), and research and development (Zhang et al. 2006a, b).

Decentralisation of decisions rights from top management at corporate headquarters in China to the plant manager of the present study is related to the extent to which the plant manager possesses decision rights (and thus multiple courses of action) across various domains that are acceptable to top management (i.e. powerful parties). While decentralisation is therefore closely related to managerial discretion and autonomy in the present study, it tends to emphasise the objective delegation of decision rights to the plant manager rather than the perceived latitude of managerial action that defines the managerial discretion of the plant manager.

This discontinuity motivates the study’s first hypothesis (see below) and the integration of the various existing theories into a new single model for the impact of managerial discretion on performance (see Chap. 4).

Empirical studies have modelled unidimensional discretion constructs by measuring one or several proxies related to e.g. ratings of managerial power, internal representation on the board of directors, managerial stock ownership, and financial ratios (e.g. Huiyuan Chen 2006; Khanchel 2009; Yougen Li and Zhao 2004; Zhang and Li 2008b; Zhang et al. 2006a, b) as well as multiple antecedents drawn from mostly the task environment (e.g. Agarwal et al. 2009; Berman et al. 2005; Cameron 2000; Finkelstein and Boyd 1998; Finkelstein and Hambrick 1990; He et al. 2009; Magnan and St-Onge 1997; Rajagopalan and Finkelstein 1992; Williamson 1963). In addition, industry-level discretion has been frequently proxied in existing studies (e.g. Abrahamson and Hambrick 1997; Datta et al. 2003; Finkelstein and Hambrick 1990; Hambrick and Abrahamson 1995; Hambrick et al. 1993; Keegan 2006; Keegan and Kabanoff 2008; Thomas and Peyrefitte 1996).

The present study reviews over 80 empirical studies on managerial discretion and related phenomena, e.g. Abrahamson and Hambrick (1997), Acemoglu et al. (2007), Adams et al. (2005), Agarwal et al. (2009), Agrawal and Knoeber (1996), Aragon-Correa et al. (2004), Barnabas and Mekoth (2010), Baysinger and Butler (1985), Berger et al. (1997), Berman et al. (2005), Bloom et al. (2008), Bowen et al. (2008), Brush et al. (2000), Burkart et al. (1997), Zhang and Li (2008b), Zhang and Xie (2008), Zhang et al. (2006a, b), Cameron (2000), Carpenter and Golden (1997), Caza (2007), Caza (2011), Chaganti et al. (1985), Chang and Wong (2003), Chang and Wong (2004), Chen (2006), Cheng et al. (2006), Colombo and Delmastro (2004), Crossland and Hambrick (2007), Datta et al. (2003), Demsetz and Lehn (1985), Denis and Denis (1993), Denis et al. (1997), Donaldson and Davis (1991), Finkelstein and Boyd (1998), Finkelstein and Hambrick (1990), Gammelgaard et al. (2010), Glaister et al. (2003), Groves et al. (1994), Wang (2009), Haleblian and Finkelstein (1993), Hambrick and Abrahamson (1995), Hambrick et al. (1993), He et al. (2009), Heinecke (2011), Hutzschenreuter and Kleindienst (2007), Kayhan (2008), Keegan and Kabanoff (2008), Keegan (2006), Kesner (1987), Khanchel (2009), Lang et al. (1995), Lieberson and O’Connor (1972), López-Navarro and Camisón-Zornoza (2003), Mackey (2008), Magnan and St-Onge (1997), Manner (2010), Marin and Verdier (2006), Misangyi (2002), Oh (2002), Ongore (2011), Palmer (1973), Perrone et al. (2003), Quigley and Hambrick (2009), Rajagopalan and Finkelstein (1992), Stano (1976), Wang et al. (2008), Tang (2008), Thomas and Peyrefitte (1996), Venaik (1999), Verhoest (2003), Zhang (1997), Walters (1995), Werner and Tosi (1995), Williamson (1963), Wright and Kroll (2002), Li (2007), Xu et al. (2005), Li and Zhao (2004), Yan et al. (2010), Zahra and Stanton (1988), Zhao et al. (2010), and Zheng (2007).

As explained below, the impact of managerial discretion on performance can be decomposed into a direct effect and moderating effects, which are denoted by ‘Direct’ and ‘Moderating’ in Table 2.4, respectively. Moreover, as the review of the literature on the impact of discretion on performance has implications for the choice of the unit of analysis of the present study, Table 2.4 includes the ‘Unit of analysis’ as an additional row.

Similar research design definitions can be found in the literature (e.g. Chui 2007, p. 66; Punch 2005, p. 62).

The unit of analysis of the present study (i.e. the plant manager in China) is discussed in detail in Chap.​ 3.

The measures of discretion and performance in the present study are described in Sect.​ 4.​2 and are demonstrated to exhibit high reliability and construct validity in Sects.​ 5.​2 and 5.​3, respectively.

A discussion of alternative multivariate analysis techniques is provided in Sect.​ 5.​1.​1, which develops its own decision-tree logic for choosing an appropriate modelling methodology for the present study.

A more complete version of Equation (2.1) is developed from the literature in Sect.​ 4.​3 as part of the present study’s new empirical discretion model. This equation resembles Equation (2.1) when expressed in vector/matrix notation (e.g. Gentle 2007, pp. 479–491; Harville 2008, pp. 1–10; Knapp 2007, pp. xxi–xxiv), but it is sufficiently general that it can be disaggregated into an arbitrary number of discretion dimensions, controls, and moderators. In contrast to the simplified Equation (2.1), the equation in Sect.​ 4.​3 fulfils the requirements of state-of-the-art methodological research such as that all the components of the product term \( \left( {D\cdot M} \right) \) must be included in the equation in direct form (Carte and Russell 2003, pp. 480–495; Cohen 1978; Cronbach 1987; Henseler and Fassott 2010, pp. 718–719; Irwin and McClelland 2001, p. 105). Applying this requirement to Equation (2.1), it would read \( P=d\cdot D+c\cdot C+m\cdot \left( {D\cdot M} \right)+q\cdot M+\varepsilon \). The intercept term is normalised to zero.

A discussion of direct effects, control effects, mediating effects, and moderating effects is presented in Sect.​ 4.​3. Moreover, a discussion of control variables is included in Sect.​ 5.​4.​1 on the threat to internal validity of excluding associated variables.

While there are differences in the equations estimated across existing studies other than in the number of controls (\( C \)) and moderators (\( M \)), the overall structure of the equations remains similar to Equation (2.1). Some studies add such terms to Equation (2.1) as intercept terms [which are normalised to zero in Equation (2.1)], quadratic terms so as to model non-linear effects (e.g. Huiyuan Chen 2006; Zhang and Li 2008b, p. 122) or lagged terms when time series are involved (e.g. Cheng et al. 2006; Groves et al. 1994). Likewise, instead of modelling moderating effects with the product term approach in Equation (2.1) (see Sect.​ 4.​3.​1; e.g. Carte and Russell 2003, pp. 480–495; Chin et al. 2003, pp. 196–200; Irwin and McClelland 2001, p. 105), certain studies exclude the product term \( \left( {D\cdot M} \right) \) and apply the multi-group comparison approach for testing moderating effects (e.g. Arnold 1982; Henseler and Fassott 2010, pp. 719–721; Rigdon et al. 1998, p. 1; Venkatraman 1989, p. 426). Yet even these studies can be conceptually represented by Equation (2.1), since these two approaches to modelling moderating effects resemble each other when the moderator can be sensibly dichotomised (Henseler and Fassott 2010, p. 721; Qureshi and Compeau 2009, p. 199).

As explained in Sect. 2.2.1, when moderators deviate from zero, the impact of discretion on performance deviates from the direct effect of discretion (\( d \)) by moderating effects, which are reviewed in Sect. 2.2.4.

In order to achieve a broad coverage of existing empirical evidence, studies are also included if they denote managerial discretion by alternative comparable terms, such as managerial autonomy (see Sect. 2.1.1).

It should further be noted that in addition to empirical studies that explicitly aim to measure managerial discretion or autonomy (exemplified by the references above), the contradictory evidence extends to studies that implicitly measure constructs potentially related to discretion, such as diffusion of ownership, managerial stock ownership, internal representation on the board, and other measures of board composition. Chang and Wong (2003, p. 7) view such studies as evidence of an inconclusive relationship between discretion and performance, with the empirical findings of e.g. Donaldson and Davis (1991) and Kesner (1987) supporting a positive, those of Chaganti et al. (1985), Demsetz and Lehn (1985), and Zahra and Stanton (1988) supporting a neutral, and those of Baysinger and Butler (1985) and Palmer (1973) supporting a negative relationship.

The scatter plot format in Fig. 2.1 is applied in Sects. 2.3 and 2.4 below in order to link the theories to the empirical evidence on the impact of discretion on performance and derive the present study’s Hypothesis 4.

Non-zero values of a control variable (\( C\ne 0 \)) simply shift the dotted lines in Fig. 2.1 upwards or downwards (provided there is a non-zero control effect, i.e. \( c\ne 0 \)) without affecting the slopes of the lines. By contrast, non-zero values of a moderator variable (\( M\ne 0 \)) alter the slope and therefore tilt the dotted lines (if \( m\ne 0 \)).

It should be noted that due to their observational cross-sectional designs, the existing empirical studies cannot unequivocally demonstrate that additional discretion causes an increase or decrease in performance but rather only make statements regarding association that may be consistent with causality (e.g. Caza 2007, p. 46; Finkelstein and Hambrick 1990, p. 500; Granger 1969; Sánchez 2008, p. 5; Simon 1954, pp. 477–478; Wagner 2002, pp. 287–292; see Sect.​ 5.​4.​1).

For example, if two different studies adopt two different measures of discretion, each of which taps into a distinct dimension of discretion, then one study might find a positive and the other a negative impact of discretion on performance, provided the distinct discretion dimensions have different performance impacts.

As noted in Chap.​ 3, a unit of analysis is also defined in terms of time (Northrop and Arsenault 2007, p. 214).

As explained in Chap.​ 3, there are tens of thousands of plant managers in China alone (Guojia tongji ju [National Bureau of Statistics] 2007, 14–1, 14–2, 14–18). With plant managers being but one example of middle managers, this translates into an even larger number of middle managers in organisations worldwide.

Relating these measures of discretion to the dimensions of discretion is an important step in the present study’s attempt to work towards resolving the discretion puzzle and leads to the formulation of Hypothesis 1.

Acemoglu et al. (2007) measure autonomy for British firms only in the domain of employment decisions and autonomy for French firms only in the domain of investment decisions in two out of their three datasets.

For instance, Caza (2007, pp. 14–16) combines indicators on the manager’s discretion in training resources, hiring, firing, and assigning specific tasks into a unidimensional discretion construct to investigate its impact on performance, although he later finds that discretion covers multiple dimensions (pp. 26–82).

For example, Agarwal et al. (2009, p. 2221) proxy managerial discretion by the length of lockup, notice, and redemption periods for the hedge funds under investigation.

Thomas and Peyrefitte (1996) investigate the impact of discretion on performance using Finkelstein and Hambrick’s (1990) industry-level discretion, which has been further advanced in subsequent studies (e.g. Abrahamson and Hambrick 1997; Datta et al. 2003; Hambrick and Abrahamson 1995; Hambrick et al. 1993; Keegan 2006; Keegan and Kabanoff 2008).

The value of the review of existing measures of discretion in the literature (summarised in Table 2.5) extends beyond demonstrating that the measures of discretion have differed starkly in the literature. Relating these measures of discretion to the dimensions of discretion is an important step in the present study’s attempt to work towards resolving the discretion puzzle and leads to the formulation of Hypothesis 1 in Sect. 2.2.4. Based on the test results of Hypothesis 1 (see Sect.​ 6.​2.​1), Sect.​ 7.​3.​1 generates implications on how these measures of discretion in Table 2.5 might be flawed and partly responsible for the discretion puzzle. This leads to recommendations on how discretion measures should be built and interpreted in future research.

Section 2.2.1 provides an explanation of these two aspects of the modelling methodology.

For instance, Zhang and Li (2008b, p. 122) as well as Chen (2006) include a quadratic term so as to model non-linear effects, i.e. an inversed U-shaped relationship between discretion and performance, with discretion initially increasing and eventually decreasing performance.

Most of the studies above differ from each other in terms of both their unit of analysis and their discretion measure, and in some cases further differ in terms of their performance measure and modelling methodology. When attempting to hold the unit of analysis constant by comparing only studies on top management in China, it is found that the contradictory empirical results remain—with Chang and Wong (2003, 2004) and Zhang (1997) finding a positive, Li and Zhao (2004) finding a neutral, and Xu et al. (2005) finding a negative direct effect of managerial discretion on performance. While this hints that controlling for the unit of analysis in terms of level of management (i.e. top management) and geography (i.e. China) does not (on its own) account for the literature’s contradictory results, such inferences are hindered by the fact that the studies’ research designs still differ in a number of other ways (e.g. discretion measures and modelling methodologies).

It should be noted that as explained in Box 1.1 in Sect.​ 1.​2 on the delimitations of the research objective, the present study’s approach is designed as a proof-by-counter-example and is subject to the caveat of observational cross-sectional studies in terms of demonstrating causality (e.g. Caza 2007, p. 46; Finkelstein and Hambrick 1990, p. 500; Granger 1969; Sánchez 2008, p. 5; Simon 1954, pp. 477–478; Wagner 2002, pp. 287–292; see Sect.​ 5.​4.​1). In simple terms, if it is found in the instance of the present study that discretion is multidimensional, this could constitute a proof-by-counter-example that discretion is not unidimensional in every case. While this does not prove that a unidimensional measure of discretion is flawed in the existing studies and responsible for the discretion puzzle, it would offer evidence that this is a possibility and warrants further investigation.

From an empirical point of view, differences in the measure of performance across studies could likewise offer a potential explanation for the divergent empirical results on the impact of discretion on performance. Some of the existing literature can be interpreted in favour of the hypothesis that differences in measures of performance may help explain differences in empirical results. For example, Khanchel (2009) finds a significantly positive impact of discretion on return on assets (ROA) but insignificant results when measuring performance by Tobin’s Q. Likewise, Wang et al. (2008) find a positive impact of discretion on sales efficiency but a negative impact on strategic partnership with the global customers. While this potential explanation lies beyond the scope of the present study, future studies might extend the present study’s approach of distinguishing measures in a more granular way and formulate a hypothesis analogous to Hypothesis 1 for testing whether differences in performance measures help explain the discretion puzzle.

As explained in Sect. 2.1.2, most existing studies have adopted unidimensional measures of discretion, meaning they have not exhibited sufficient granularity to allow for a multidimensional measure of discretion (e.g. Agarwal et al. 2009; Barnabas and Mekoth 2010; Berman et al. 2005; Bloom et al. 2009a, b; Cameron 2000; Caza 2007; 2011; Chang and Wong 2003; Chen 2006; Cheng et al. 2006; Finkelstein and Boyd 1998; Finkelstein and Hambrick 1990; Gammelgaard et al. 2010; He et al. 2009; Khanchel 2009; Li and Zhao 2004; Magnan and St-Onge 1997; Marin and Verdier 2006; Rajagopalan and Finkelstein 1992; Williamson 1963; Zhang and Li 2008b; Zhang et al. 2006a, b).

As explained above, the reviewed empirical studies have also differed from each other in terms of aspects of the modelling methodology other than moderator variables, such as multivariate analysis techniques. Future studies could investigate these differences as potential explanations of the discretion puzzle.

This equivalence can be understood by drawing on Chap.​ 4. With the multi-group comparison approach for testing moderating effects (see Sect.​ 4.​3.​1; e.g. Arnold 1982; Henseler and Fassott 2010, pp. 719–721; Rigdon et al. 1998, p. 1; Venkatraman 1989, p. 426), Hypothesis 2 is tested by separately estimating an equation similar to Equation (2.1) in Sect. 2.2.1 for the two firm type groups, which yields a separate direct effect of discretion on performance for the unit of analysis of Chinese firms (\( {d^{Chinese }} \)) and a separate one for multinationals (\( {d^{Multi. }} \)). Testing whether firm type affects the impact of discretion on performance then tests whether the difference between \( {d^{Chinese }} \) and \( {d^{Multi. }} \) is significant, and this difference is the moderating effect of firm type (see Sect.​ 4.​3.​2). In principle, the product term approach for testing moderating effects (see Sect.​ 4.​3.​1; e.g. Carte and Russell 2003, pp. 480–495; Chin et al. 2003, pp. 196–200; Irwin and McClelland 2001, p. 105) can also be applied for testing Hypothesis 2, whereby firm type is coded as a dichotomous categorical variable by using a dummy moderator (Henseler and Fassott 2010, p. 721; Qureshi and Compeau 2009, p. 199). With reference to Equation (2.1) in Sect. 2.2.1, \( M=0 \) could represent multinationals and \( M=1 \) Chinese firms, so the moderating effect \( m \) would measure the difference in the impact of discretion on performance between a Chinese firm and a multinational, and testing for \( m\ne 0 \) would test Hypothesis 2.

Based on the discussion in this section, the existing empirical evidence on moderating effects (\( m \)) can now be interpreted as showing that the way in which a manager uses his/her discretion can depend on various influences on managers—in terms of such statistically significant moderators as managerial experience (Caza 2007), managerial incentives (Cheng et al. 2006), top management pay gap (Zhang and Li 2008b), corporate control via performance monitoring, incentive systems, and social integration (Wang et al. 2008), export joint venture group composition (López-Navarro and Camisón-Zornoza 2003), and market competition (Zhao et al. 2010). For some of the moderators tested in the literature, moderating effects were found to be insignificant, such as managerial education, managerial commitment, and the number of similar units (Caza 2011), ownership concentration (Yougen Li and Zhao 2004), and the organisational type of state-owned enterprises in China (Xu et al. 2005), i.e. employee-owned stock cooperatives, limited liability companies, and limited liability stock companies.

The theories can be applied to study a wide variety of topics other than the impact of managerial discretion on performance, such as a range of topics in economics and finance for principal-agent theory. However, as these applications do not contribute to fulfilling the present study’s research objective, they are not included in this literature review. Instead, the theories are reviewed only with respect to what is relevant to this study.

As described in Sect. 2.3.1, managerial discretion theory was originally developed as a reconciliation of population ecology (e.g. Aldrich 1979; Baum 1996; Baum and Amburgey 2002; Carroll 1988; Freeman et al. 1983; Hannan and Freeman 1977, 1984; Singh and Lumsden 1990; Tushman and Romanelli 1985; Zohar and Luria 2005) and strategic choice theory (e.g. Child 1972, 1997, 2002; Child et al. 2003; Elbanna and Child 2007; Hitt and Tyler 1991; Hrebiniak and Joyce 1985; Judge and Zeithaml 1992; Marlin et al. 1994; Miles and Snow 1978; Stienstra et al. 2004).

As explained in Sect. 2.3.2, it is possible to assume that the principal is less performance-maximising than the agent, in which case managerial discretion may positively affect performance (Chang and Wong 2003, pp. 1–7). However, this assumption is rarely made in principal-agent theory (Thomsen and Pedersen 2000, p. 690). The literature on principal-agent theory therefore generally contends that managerial discretion has a negative direct effect on performance (e.g. Caza 2007, p. 10; Caza 2011; Chang and Wong 2003, p. 7; Davis et al. 1997b, p. 38; Hutzschenreuter and Kleindienst 2007, p. 4; Jensen and Murphy 1990; Zhao et al. 2010).

For instance, Hannan and Freeman (1977, p. 957) postulate that ‘there are very strong inertial pressures on structure arising both from internal arrangements (e.g., internal politics) and the environment (e.g., public legitimation of organizational activity).’

These contingencies affecting strategic choices and thus organisational outcomes are investigated in upper-echelon theory. The theory postulates that strategic choices affecting organisational outcomes are reflections of the cognitive bases and values (i.e. knowledge, ordering, and preferences) of influential top managers, since under the assumption of bounded rationality (Simon 1957b) the managers’ cognitive bases and values are pivotal for processing the complex information from environmental and organisational stimuli (Aragon-Correa et al. 2004; Galavan 2005; Galavan et al. 2009; Hambrick et al. 1993; Hambrick and Mason 1984; Hutzschenreuter and Kleindienst 2007; Manner 2010; Pegels et al. 2000). It is argued that these cognitive bases and values can be measured by observable managerial background characteristics (e.g. age, education, and socioeconomic roots), and indeed it has been confirmed that many of these demographics are empirically related to strategic choices and performance (see Carpenter et al. 2004; Hambrick 2007).

Outside the scope of managerial discretion theory, notions similar to managerial discretion have been applied to non-managerial units of analysis. For example, the job design literature (e.g. Hackman et al. 1975) is concerned with the autonomy of workers from a motivational perspective (i.e. their discretion relative to management). Moreover, the Nobel Laureates Kydland and Prescott (1977) have derived far-reaching implications for the design of macroeconomic policy from their analysis of the discretion of policymakers.

In addition to such harder organisational antecedents as organisational size and age, softer antecedents have been put forward in the literature. For example, Cennamo et al. (2009) contend that top management may enlarge their managerial discretion by pursuing a broader stakeholder management orientation.

As defined by the research objective in Sect.​ 1.​2, the dynamics of discretion (i.e. the analysis of discretion over time, such as differentiating between short-term and long-term effects) are still at an early stage of research in the literature (e.g. Finkelstein and Peteraf 2007, pp. 243–245; Hutzschenreuter and Kleindienst 2007, p. 1; Kayhan 2008, pp. 1–6) and, in line with the present study’s observational cross-sectional research design, are not addressed.

Finkelstein and Hambrick (1990, p. 488) define ‘the extent to which a firm’s performance is similar to the average for the industry’ as performance conformity, with the opposite denoted by performance extremeness (Quigley and Hambrick 2009, p. 3) or performance variability (Misangyi 2002, pp. 36–37). For instance, for top management, chief executives with the discretion to shape their company’s strategy might make radical changes to strategies and thereby impact on performance more strongly than chief executives whose discretion is highly constrained, as in regulated industries. For middle management, a plant manager whose discretion is constrained to making capital investments of only say 1,000 RMB has less latitude to affect performance through capital investments than a plant manager whose discretion is set to 1,000,000 RMB.

In addition to performance, it has been demonstrated that discretion may exhibit significant relations to such consequences as managerial power, managerial compensation, workers’ incentives, a successor chief executive officer’s age, top management team tenure, trust, strategic attention, environmental commitment, pricing, organisational knowledge creation, and research and development (see references above).

New institutional economics, including principal-agent theory (see above) and transaction cost economics (Argyres and Liebeskind 1999; Argyres and Mayer 2007; Bercovitz et al. 2006; Coase 1937, 1960; Williamson 1985, 1991, 1996), evolved as a critique of neoclassical economics by relaxing first-best assumptions, such as regarding information (i.e. asymmetric information in principal-agent theory and information impactedness in transaction cost economics) or bounded rationality (Pascha and Storz 2005, p. 16).

While the assumptions of principal-agent theory can differ in detail between differing models in the literature (e.g. with assumptions at times also including bounded rationality and risk aversion), assumptions 1 and 2 are central to principal-agent theory and tend to be common across differing models (Eisenhardt 1989, pp. 58–59; Levinthal 1988, p. 153; Spremann 1987, p. 3). In addition to these two central assumptions, Sect. 2.3.2.2 discusses two further assumptions commonly made by principal-agent theorists when predicting the impact of managerial discretion on performance (i.e. assumption 3: the agent’s managerial action is rational in terms of tending to produce the outcomes that the agent intends; assumption 4: the agent’s interests are less aligned with performance maximisation than the principal’s interests; e.g. Albanese et al. 1997, p. 610; Chang and Wong 2003; Davis et al. 1997b, pp. 20–22; Eisenhardt 1989; Jensen 1986; Jensen and Meckling 1976; Jensen and Murphy 1990; Shleifer and Vishny 1997; Spremann 1987; Thomsen and Pedersen 2000; Werner and Tosi 1995, p. 1673).

For example, in Spremann’s (1987) principal-agent model total output and thus the principal’s welfare are dependent on the agent’s hidden efforts as well as some exogenous risk. This implies that the principal cannot fully distinguish between low effort on the part of the agent and bad luck on the part of exogenous risk, so the asymmetric information (hidden effort) gives the agent managerial discretion to pursue his/her own interests.

Although the likelihood that the principal’s and agent’s interests diverge is substantial, the interests may be aligned in some cases, whereby the agency problem is avoided and control mechanisms become unnecessary (Davis et al. 1997b, p. 22). Nevertheless, as the interests of the agent are difficult for the principal to judge ex ante (Williamson 1985), principal-agent theory prudently recommends imposing control mechanisms to limit potential losses. Internal control mechanisms are in general preferable to external control mechanisms, since external control mechanisms such as acquisitions and divestures tend to come at a higher expense to the principal’s utility (Walsh and Seward 1990, pp. 444–445). The extent to which agents fail to experience discipline from this full range of control mechanisms has been termed entrenchment (Berger et al. 1997, p. 1411).

As noted above, in Spremann’s (1987) principal-agent model, low output (i.e. low performance) can be caused by either low effort on the part of the agent or bad luck due to exogenous risk. As the principal cannot fully monitor the agent’s effort and thus cannot fully hold the agent accountable for his/her actions, this asymmetric information (hidden effort) gives the agent managerial discretion to pursue his/her own interests.

For example, suppose top management at corporate headquarters in China (principal) could initially not fully monitor the plant manager’s (agent’s) managerial activities, such as his/her capital investments, perhaps due to their complex, uncertain or unobservable nature. The plant manager might then have the discretion to undertake certain capital investments out of self-serving interests and against the interests of top management. The nature of the plant manager’s managerial activities would thus constitute an antecedent of discretion. Now suppose that top management imposed stricter monitoring control mechanisms, such as audits and performance evaluations that made it easier to monitor the plant manager’s activities. Top management would then be more able to hold the plant manager accountable for the effects of his/her capital investments and not every capital investment would fall into their zone of acceptance (Hambrick and Finkelstein 1987, p. 378). The monitoring control mechanisms would thus reduce the choices available to the plant manager for making investments that are acceptable to top management in China—i.e. they would reduce the manager’s discretion.

For example, when top management at corporate headquarters in China (i.e. the principal) is more interested in high performance than the plant manager (i.e. the agent), then the plant manager might use the additional discretion to make capital investments that suit his/her personal agenda at the expense of performance.

Even when accepting a close alignment of the principal’s and agent’s interests, the impact of discretion on performance may not be insignificant due to the additional theoretical mechanism described in Sect. 2.3.3.

As explained in Sect. 2.3.3, stewardship theory proposes an additional theoretical mechanism which can help explain a positive impact of discretion on performance. After discussing this mechanism, Sect. 2.3.3 correspondingly extends this potential explanation of the contradictory evidence of the discretion puzzle.

As explained in Box 2.1, it is possible to assume that the principal is less performance-maximising than the agent, in which case managerial discretion may positively affect performance (Chang and Wong 2003, pp. 1–7). However, this assumption is rarely made in principal-agent theory (Thomsen and Pedersen 2000, p. 690). Nevertheless, it should be noted that the principal-agent literature recognises that maintaining a certain level of performance might be in the manager’s own best interest due to ‘both the discipline and opportunities provided by the markets for their services, both within and outside of the firm’ (Fama 1980, p. 289).

As noted above, Sect. 2.4.3 explains that this prediction of principal-agent theory (\( d<0 \)) is recorded as the principal-agent hypothesis for Hypothesis 4, which is indicated on the downwards-sloping line in Fig. 2.4.

As noted above, if one instead assumes that the agent’s (i.e. plant manager’s) interests are more aligned with maximising performance than those of the principal (i.e. top management at corporate headquarters in China), then granting discretion might increase performance (e.g. Chang and Wong 2003). The plant manager would still use discretion opportunistically to produce the outcomes that he/she desired, but when assuming that these agent-desired outcomes are more performance-maximising than those of the principal, then this self-interest seeking behaviour of the agent would increase performance (albeit to the discontent of the principal).

Moderating effects were formally introduced in Sect. 2.2.1 and are discussed at length in Sect.​ 4.​3. It was shown that the relationship between discretion and performance is often modelled by an equation similar to \( P=d\cdot D+c\cdot C+m\cdot \left( {D\cdot M} \right)+\varepsilon \) (see Equation (2.1) in Sect. 2.2.1), which is depicted by the downwards-sloping line in Fig. 2.4 (when the control variable \( C \) and the moderator variable \( M \) are normalised to zero). By partial differentiation, the total impact of discretion on performance was derived as \( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \) (see Equation (2.2) in Sect. 2.2.1), i.e. the sum of the direct effect of discretion on performance (\( d \)) and the moderating effect (\( m \)) multiplied by the moderator variable’s value (\( M \)) (see Sect.​ 4.​3.​2 on comparative statics; e.g. Aiken and West 1991, p. 37; Dowling 2000, pp. 284–291; Finney et al. 1984; Henseler and Fassott 2010, p. 728; Hirschey 2009, p. 99). Hence, the slope of the line in Fig. 2.4 (i.e. \( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \)) is equal to the direct effect of discretion on performance (\( d \)) when the moderator variable is zero (\( M=0 \)) and is adjusted upwards or downwards to the extent that the moderator diverges from zero. Moderating effects (\( m \)) therefore adjust the impact of discretion on performance upwards or downwards beyond the direct effect (\( d \)) and thus pivot the line in Fig. 2.4, as indicated by the curved arrow therein.

For example, when the plant manager’s relative interest in performance-maximisation (\( M \)) increased (whether due to natural predisposition or compensation control mechanisms), the impact of discretion on performance (\( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \)) could become less negative, because the plant manager would then use discretion to produce outcomes that were to a lesser extent harmful to organisational performance.

Hypothesis 2 and Hypothesis 3 regarding the respective moderating effects of firm type and firm size on the impact of discretion on performance (see Sect. 2.4.2) are methodologically motivated by the discussion of moderating effects in Box 2.2. In particular, principal-agent theory was shown to imply that one moderator variable (i.e. the agent’s interests) can moderate both the direct effect of discretion on performance (i.e. \( d>0 \) versus \( d<0 \)) and the moderating effects of other moderator variables (i.e. \( m>0 \) versus \( m<0 \)). In parallel, the present study specifies that one moderator variable (i.e. firm type) can moderate both the direct effect of discretion on performance and the moderating effect of firm size (see Sect. 2.4.2).

It was explained above that in the extreme case where the agent’s interest in performance surpasses the principal’s interest in performance (i.e. \( M \) becomes large), the total impact of discretion on performance might become positive. This is consistent with the described moderating effect, since as \( M \) becomes large for a firm in a sample with a highly performance-aligned agent, the positive moderating effect (\( m \)) in this firm becomes so large that it outweighs the overall negative direct effect of discretion on performance (\( d<0 \)) estimated for the overall sample, i.e. \( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M>0 \) becomes positive for the performance-aligned agent despite \( d<0 \).

Algebraically, the impact of discretion on performance is \( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M=d+m\cdot 0=d \) with weaker monitoring (\( M=0 \)) and \( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M=d+m\cdot 1=d+m \) with better monitoring (\( M=1 \)). The moderating effect on the impact of discretion on performance from imposing better monitoring (i.e. from increasing \( M \) from \( M=0 \) to \( M=1 \)) is thus equal to \( m \) (i.e. the difference between these two equations). Hence, the impact of discretion will be less negative with better monitoring to the extent that the moderating effect is positive (\( m>0 \)).

This three-way interaction (Henseler and Fassott 2010, p. 722) in the sense that the agent’s interests influence whether ability, knowledge, and information improve or dampen the impact of discretion on performance is consistent with seemingly opposing views in the literature. Hayek (1945, pp. 524–526) advocates extending discretion downwards along the line of control (i.e. decentralisation) on the premise that this may facilitate the use of everyone’s specialised knowledge. On a similar note, Li (2007) argues that the board of directors (i.e. principal) may hire a general manager (i.e. agent) as the manager possesses managerial expertise. Caza (2007, p. 13) also contends that extending discretion may improve performance only if the manager’s ability (e.g. education and experience) is sufficient so as to use discretion effectively for improving performance. However, Chang and Wong (2003, p. 24) explain that managers ‘would not make productive use of their expertise to improve firm performance if their self-interests were not somehow tied to firm performance.’ In other words, only if a manager’s interests are sufficiently aligned with improving performance may the manager’s ability, knowledge, and information positively moderate the way managers use discretion to affect performance. If instead the manager’s interests are poorly aligned with performance, then greater discretion that allows managers to ‘respond more quickly to changing circumstances’ may be abused by the manager so as to fulfil his/her private interests at the expense of those of the principal (Cheng et al. 2006, pp. 341–342).

It should be noted that the moderating effect of the agent’s interests could also be modelled as a continuous (rather than a dichotomous) moderator variable. However, the dichotomy of the multi-group comparison approach is chosen instead in order to ensure continuity with the present study’s new model in Chap.​ 4.

Assumption 2 and assumption 3 from principal-agent theory are similar in stewardship theory, namely that the principal cannot fully monitor the steward’s actions (i.e. asymmetric information) and that the steward’s managerial action is rational in terms of tending to produce the outcomes that the steward intends (e.g. Davis et al. 1997a, p. 612; Davis et al. 1997b, pp. 23–24; Fox and Hamilton 1994, p. 78).

It was shown in Box 2.1 in Sect. 2.3.2.2 that the nature of the impact of discretion on performance may depend on whose interests are more performance-maximising (assumption 4), which is further discussed below.

For example, consider granting a plant manager additional discretion for making capital investments, i.e. increasing the maximum capital investment that the plant manager (i.e. agent or steward) can undertake without prior authorisation from top management at corporate headquarters in China (i.e. principal) from 10,000 RMB to 20,000 RMB. With the self-serving utility function assumed in principal-agent theory (see the first assumption above), if the plant manager’s and top management’s interests were equally performance-maximising (see the fourth assumption above), then capital investments with the same performance consequences would be predicted with and without the higher discretion (i.e. whether or not the plant manager needed to seek prior authorisation from top management for the investments between 10,000 RMB to 20,000 RMB). With the pro-organisational utility function assumed in stewardship theory, however, even if interests were equally performance-maximising, then the higher discretion would strengthen the plant manager’s motivation (e.g. to make greater efforts to choose investments wisely) and thus would be predicted to boost performance.

This is the additional theoretical mechanism of stewardship theory which was mentioned in Sect. 2.3.2.2 to help explain a positive impact of discretion on performance.

Box 2.3 in Sect. 2.3.3.2 creates transparency on how two theoretical mechanisms translate the assumptions of stewardship theory into this predicted positive impact of discretion on performance.

For instance, Donaldson and Davis (1991) recommend having a chief executive officer (CEO) who is a steward as the chair of the board of directors, since this empowering governance mechanism grants the CEO greater discretion to shape strategy in the company’s best interest without fear of interference by an outside chair.

The slopes of the lines measure the impact of discretion on performance (i.e. \( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \)) and are therefore equal to the direct effect of discretion on performance (\( d \)) when the moderator variable is zero (\( M=0 \)), and are adjusted upwards or downwards to the extent that the moderator diverges from zero.

As explained in Sect. 2.3.1, managerial discretion theory does not make specific predictions regarding the nature of the impact of managerial discretion on performance (i.e. whether it is positive, neutral or negative; e.g. Adams et al. 2005; Crossland and Hambrick 2007; Finkelstein and Hambrick 1990; Hambrick and Finkelstein 1987; Misangyi 2002; Quigley and Hambrick 2009; Tang 2008; Zhao et al. 2010).

This extends the attempted explanation derived from principal-agent theory in Box 2.1 in Sect. 2.3.2.2.

This reasoning resonates with Aghion and Tirole’s (1997) view shared by e.g. Xiaoyang Li (2007) and Xu et al. (2005) that extending discretion downwards along the line of control (decentralisation) can raise the agent’s initiative to acquire productive information and facilitate his/her participation in the contractual relationship.

As stated earlier, moderator variables (\( M \)) with moderating effects (\( m \)) may influence the extent to which the impact of discretion on performance (\( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \)) is positive or negative, or equivalently, the slopes of the lines in Fig. 2.5 (\( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \)). Strong moderating effects may therefore help explain whether discretion increases (\( {{{\partial P}} \left/ {{\partial D}} \right.}>0 \)) or decreases performance (\( {{{\partial P}} \left/ {{\partial D}} \right.}<0 \)), or equivalently, whether the line in Fig. 2.5 is upwards-sloping or downwards-sloping. Moderating effects might thus potentially explain in which cases (i.e. for which values of \( M \)) the predictions of stewardship theory versus principal-agent theory hold true, thus paving a way for reconciling the theories and resolving the discretion puzzle.

Initial research on stewardship theory contrasted principal-agent theory and stewardship theory (e.g. Donaldson and Davis 1989, 1991, 1994; Fox and Hamilton 1994) and tended to assume that stewardship theory is correct and principal-agent theory is incorrect (see Davis et al. 1997b, p. 21). The seminal research by Davis et al. (1997b) from which moderating effects are derived herein no longer assumes that only stewardship theory is correct but rather attempts to reconcile the two theories’ differences by delineating the conditions (i.e. the moderator variables) under which each of the theories is necessary.

As explained above in the discussion on antecedents, stewardship theorists refer to these variables as ‘antecedents in the prediction of stewardship versus agency relationships’ (Davis et al. 1997b, p. 37).

If the choice between stewardship and agency behaviour (which is argued to depend on these variables) is viewed as a binary choice, then these variables represent moderator variables that switch between a positive (stewardship) and a negative (agency) impact of discretion on performance. If instead the choice is viewed as lying on a continuum between stewardship and agency behaviour (e.g. Albanese et al. 1997, p. 610), then these variables can have continuous moderating effects. The more the variables match the assumptions of stewardship rather than agency theory, the more positive the impact of discretion on performance is predicted to be. As these moderators work via altering the manager’s natural predisposition to performance maximisation, they may moderate the impact of other moderators (e.g. monitoring control mechanisms)—which was discussed in Sect. 2.3.2 in terms of three-way interaction (e.g. Henseler and Fassott 2010, p. 722) and enter into the study’s new discretion model in Chap.​ 4.

For example, suppose the manager acted as a steward but was controlled by the principal as if he/she were an agent, e.g. by harshly constraining the steward’s discretion through control mechanisms. The steward might then feel demotivated due to the theoretical mechanism related to motivation described earlier and potentially engage in antiorganisational behaviour (Argyris 1964; Fleishman and Harris 1962; Herzberg et al. 1959). Likewise, in the other type of mixed-motive relationship, the ‘principal is likely to feel betrayed and angry and may increase controls, withdraw from the situation, or attempt to remove the manager’ (Davis et al. 1997b, p. 40).

Section 1.​1 on the research gap mentions this combination of principal-agent theory and stewardship theory as one of the existing attempts to reconcile the relevant theories. The other attempts mentioned in Sect.​ 1.​1 relate to combinations of managerial discretion theory and principal-agent theory (e.g. Caza 2007; 2011; Finkelstein and Peteraf 2007), which have been discussed in Sect. 2.3.2.

A subset of these variables has been empirically investigated (e.g. Caza 2007; 2011; Mills and Keast 2009; Van Slyke 2007; Vargas Sánchez 2001, 2004). For agrarian cooperative societies in Spain (Vargas Sánchez 2001) and Wales (Vargas Sánchez 2004), for example, a subset of the psychological characteristics and situational characteristics are empirically confirmed as determining agency versus stewardship relationships of chairmen and managers. For Spain, individuals motivated by higher-order needs as well as intrinsic needs were found to be more likely to develop steward relationships, as were individuals who tended to use personal power rather than institutional power (Vargas Sánchez 2001). For Wales, by contrast, stewardship relationships were predicted by greater identification with the organisation and to a lesser extent by a lower power distance (Vargas Sánchez 2004). The other tested factors, such as value commitment, involvement-orientation, and collectivist cultures received no empirical support in either study. In addition, Caza (2007) tests for moderating effects of managerial commitment (i.e. identification; see the psychological characteristics above) on the impact of discretion on performance, but finds no significant evidence. By contrast, Van Slyke (2007) finds a positive association between the risk that either of the parties perceives and the extent to which agency behaviour is adopted (see the expectations above).

A thorough discussion of the research gap and research objective can be found in Sects.​ 1.​1 and 1.​2.

As mentioned above, the present study’s discretion puzzle relates to the managerial discretion of managers in organisations and is therefore distinct from the puzzle of discretion (Pratt and Sossin 2009) that concerns judicial discretion in law.

Managerial discretion theory (e.g. Abrahamson and Hambrick 1997; Berman et al. 2005; Carpenter and Golden 1997; Caza 2007; 2011; Huiyuan Chen 2006; Crossland 2007; Crossland and Hambrick 2007; Datta et al. 2003; Finkelstein and Boyd 1998; Finkelstein and Hambrick 1990; Finkelstein and Peteraf 2007; Haleblian and Finkelstein 1993; Hambrick and Abrahamson 1995; Hambrick and Finkelstein 1987; Hambrick et al. 1993; Hutzschenreuter and Kleindienst 2007; Keegan 2006; Keegan and Kabanoff 2008; Key 2002; Yougen Li and Zhao 2004; Magnan and St-Onge 1997; Quigley and Hambrick 2009; Rajagopalan and Finkelstein 1992; Thomas and Peyrefitte 1996; Zhang and Li 2008b; Zhang et al. 2006a, b).

Principal-agent theory (e.g. Agrawal and Knoeber 1996; Baysinger and Butler 1985; Berger et al. 1997; Brush et al. 2000; Chang and Wong 2003; Childs and Mauer 2008; Denis et al. 1997; Eisenhardt 1989; Fama 1980; Fama and Jensen 1983a, b; He et al. 2009; Jensen 1986; Jensen and Meckling 1976; Jensen and Murphy 1990; Jensen and Ruback 1983; Laffont and Martimort 2002; Lang et al. 1995; Levinthal 1988; Ongore 2011; Shleifer and Vishny 1997; Spremann 1987; Thépot 2007; Thomsen and Pedersen 2000; Walters 1995; Wang et al. 2008; Weidenbaum and Jensen 1993; Werner and Tosi 1995, p. 1673; Xu et al. 2005; Zou 1989).

Stewardship theory (e.g. Albanese et al. 1997; Arthurs and Busenitz 2003; Corbetta and Salvato 2004; Davis et al. 1997a, b; Dicke and Ott 2002; Donaldson 1990; Donaldson and Davis 1989, 1991, 1993, 1994; Eddleston and Kellermanns 2007; Fox and Hamilton 1994; Lane et al. 1999; Liu and Cai 2004; Miller and Le Breton-Miller 2006; Mills and Keast 2009; Muth and Donaldson 1998; Salvato 2002; Tian and Lau 2001; Tosi et al. 2003; Van Slyke 2007; Vargas Sánchez 2001, 2004, 2005; Zahra 2003).

More specifically, the research objective is to establish that the failure of the extant literature to account for granularity in the way that managers use discretion is a potential cause of the discretion puzzle—and that theories and empirical studies must therefore differentiate discretion’s impact by this granularity (i.e. by dimensions of discretion and influences on managers) to resolve the discretion puzzle (see Sect.​ 1.​2).

For example, a given manager might use additional discretion for making capital investments in a way that improves performance but use additional discretion for hiring workers in a way that reduces performance. According to this postulate, discretion could have positive, neutral, and also negative impacts on performance depending on the dimensions of discretion (e.g. capital investment discretion versus hiring discretion), firm type (e.g. Chinese firms versus multinationals), and firm size (e.g. 150 versus 5,000 employees).

For example, if two different studies adopt two different measures of discretion, each of which taps into a distinct dimension of discretion, then one study might find a positive and the other a negative impact of discretion on performance, provided the distinct discretion dimensions have different performance impacts.

It should be noted that as explained in Box 1.1 in Sect.​ 1.​2 on the delimitations of the research objective, the present study’s approach is designed as a proof-by-counter-example and is subject to the caveat of observational cross-sectional studies in terms of demonstrating causality (e.g. Caza 2007, p. 46; Finkelstein and Hambrick 1990, p. 500; Granger 1969; Sánchez 2008, p. 5; Simon 1954, pp. 477–478; Wagner 2002, pp. 287–292; see Sect.​ 5.​4.​1). In simple terms, if in the instance of the present study it is found that discretion is multidimensional, this would constitute a proof-by-counter-example that discretion is not unidimensional in all cases. While this does not prove that unidimensional measures of discretion are flawed in the existing studies and responsible for the discretion puzzle, it would offer evidence that this is a possibility and warrants further investigation.

Among the many empirical studies resorting to proxy measures for gauging managerial discretion, discretion has prevalently been modelled as unidimensional as well. Empirical studies have modelled unidimensional discretion constructs by measuring one or several proxies related to e.g. ratings of managerial power, internal representation on the board of directors, managerial stock ownership, and financial ratios (e.g. Huiyuan Chen 2006; Khanchel 2009; Yougen Li and Zhao 2004; Zhang and Li 2008b; Zhang et al. 2006a, b) as well as multiple antecedents mainly drawn from the task environment (e.g. Agarwal et al. 2009; Berman et al. 2005; Cameron 2000; Finkelstein and Boyd 1998; Finkelstein and Hambrick 1990; He et al. 2009; Magnan and St-Onge 1997; Rajagopalan and Finkelstein 1992; Williamson 1963). In addition, industry-level discretion has been frequently proxied in existing studies (e.g. Abrahamson and Hambrick 1997; Datta et al. 2003; Finkelstein and Hambrick 1990; Hambrick and Abrahamson 1995; Hambrick et al. 1993; Keegan 2006; Keegan and Kabanoff 2008; Thomas and Peyrefitte 1996).

As is explained in Chap.​ 7, erroneously treating discretion as unidimensional when it is in fact multidimensional might produce misleading estimates of the impact of discretion on performance that could potentially be a cause of the contradictory empirical evidence that gives rise to the discretion puzzle.

Section 2.3 has derived these influences (i.e. moderators) from the theories’ assumptions, as these determine the theories’ predicted impact of discretion on performance. For example, the principal’s and manager’s psychological characteristics, perceptions of the organisation’s situational characteristics, and expectations are all postulated to influence whether the manager’s natural predisposition to performance maximisation is as assumed in principal-agent theory (i.e. discretion is abused to pursue self-serving interests at the expense of performance) or as assumed in stewardship theory (i.e. discretion is used to diligently improve performance). Empirical studies focusing on managers as units of analysis where the assumptions of stewardship theory apply (i.e. stewards) might then find a positive impact of discretion on performance, whereas studies focusing on agents might find a negative impact, and studies mixing stewards and agents in their samples might find an insignificant (neutral) impact. It follows that taking a more granular approach that differentiates the unit of analysis according to moderators related to the theories’ assumptions (i.e. firm type and firm size in this study) might reveal that certain units of analysis lead to a positive and others to a negative impact of discretion. This greater granularity could then potentially help explain the contradictory evidence of the discretion puzzle.

By testing whether the impact of discretion on performance can differ by these influences (i.e. firm type and firm size), the present study establishes whether managers in different firm types and firm sizes tend to use their discretion in different ways (i.e. more like stewards or more like agents, leading to positive impacts in some situations and negative impacts in others). This therefore tests for whether the influences on managers help explain a positive versus negative impact of discretion on performance and thereby potentially help explain the contradictory empirical results that give rise to the discretion puzzle.

Similarly, with different types of firms attracting different people, the psychological characteristics and expectations that stewardship theorists have argued to influence how managers use discretion might differ between Chinese firms and multinationals (see Sect. 2.3.3.2; e.g. Argyris 1973a, b; Brown 1969; Caza 2007; Davis et al. 1997b, pp. 38–43; French and Raven 1959; Gibson et al. 1991; Katz and Kahn 1978; Khanchel 2009, p. 98; Maslow 1970; McGregor 1960; Mills and Keast 2009; Simon 1957a, b; Turner 1981; Van Slyke 2007; Vargas Sánchez 2004, pp. 4–5, 2005, pp. 24–25). In addition to these influences derived from stewardship theory that can affect the manager’s natural predisposition to performance maximisation, Chinese firms and multinationals might also differ in terms of their compensation control mechanisms, which can moderate the impact of discretion on performance in principal-agent theory (see Sect. 2.3.2; e.g. Agrawal and Knoeber 1996, p. 378; Berger et al. 1997, p. 1411; Burkart et al. 1997, p. 705; Chang and Wong 2003, p. 6; Cheng et al. 2006; Eisenhardt 1989, p. 60; Fama and Jensen 1983a, p. 345; Jensen and Meckling 1976; Jensen and Murphy 1990, p. 226; Levinthal 1988, p. 153; Spremann 1987, p. 10; Wang et al. 2008; Werner and Tosi 1995, p. 1673; Zhang and Li 2008b).

For the unit of analysis of the present study (see Chap.​ 3), this would predict that top management at corporate headquarters in China could, ceteris paribus, better monitor the plant manager in a larger firm with multiple plants and plant managers than in a smaller firm with only a single plant and a single plant manager.

For example, if plant managers in Chinese firms tended to be more performance-maximising than top management but plant managers in multinationals tended to be less performance-maximising than top management, then better monitoring control mechanisms in e.g. larger firms that aligned the plant manager’s actions more closely with those desired by top management would reduce performance in Chinese firms but improve performance in multinationals. Firm type would then influence (i.e. moderate) whether firm size has a positive or negative moderating effect on the impact of discretion on performance. This study’s model allows for such three-way interaction by estimating separate moderating effects of firm size by firm type.

Empirical evidence exists for both approaches. On the one hand, scholars have tested moderators that derive directly from the assumptions of principal-agent theory and stewardship theory, such as managerial incentives (Cheng et al. 2006), managerial commitment, the number of similar units, managerial education, and managerial experience (Caza 2007; 2011), top management pay gap (Zhang and Li 2008b), and corporate control via performance monitoring, incentive systems, and social integration (Wang et al. 2008). On the other hand, scholars have tested moderators that aggregate individual influences to an organisational context, such as ownership concentration (Yougen Li and Zhao 2004), export joint venture group composition (López-Navarro and Camisón-Zornoza 2003), the organisational type of state-owned enterprises (Xu et al. 2005), and market competition (Zhao et al. 2010). The organisational context measured by these moderators may tap into individual influences from the theories, e.g. competition might tap into monitoring control mechanisms: ‘The firm is disciplined by competition from other firms, which forces the evolution of devices for efficiently monitoring the performance of the entire team and of its individual members’ (Fama 1980, p. 289).

As noted in Sect. 2.3.3.2, even if it were empirically proven that certain psychological characteristics, perceptions, and expectations predicted whether a given manager acted as a steward versus an agent, it would be intricate for a principal in practice to observe these factors and thus evaluate ex ante whether or not the manager would act as a steward or an agent (Davis et al. 1997b, p. 22; Williamson 1985)—making it difficult to recommend whether to increase or decrease managerial discretion in practice.

In terms of the algebraic notation utilised throughout this chapter, Fig. 2.6 depicts performance as a linear function of discretion (i.e. \( P=d\cdot D+c\cdot C+m\cdot \left( {D\cdot M} \right) \)), with the impact of discretion on performance (i.e. \( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \)) represented by the slopes of the lines. When the impact of discretion on performance is positive (\( {{{\partial P}} \left/ {{\partial D}} \right.}>0 \)), the line is upwards-sloping and when the impact is negative, it is downwards-sloping.

As discussed in Sects.​ 1.​1 and 2.3, existing combinations of managerial discretion theory, principal-agent theory, and stewardship theory also do not fully explain the impact of discretion on performance (e.g. Caza 2007; 2011; Davis et al. 1997b, pp. 27–43; Finkelstein and Peteraf 2007, pp. 237–243; Lane et al. 1999, p. 1079; Vargas Sánchez 2001, 2004, 2005).

As explained in Sect. 2.3.1, managerial discretion theory does not clearly specify the nature of the impact of discretion on performance and therefore makes no predictions on e.g. positive versus negative impacts (e.g. Adams et al. 2005; Crossland and Hambrick 2007; Finkelstein and Hambrick 1990; Hambrick and Finkelstein 1987; Misangyi 2002; Quigley and Hambrick 2009; Tang 2008; Zhao et al. 2010). Although principal-agent theory can be consistent with the null hypothesis (\( H_4^0 \)) that discretion does not alter performance in exceptional cases—i.e. when control mechanisms are so effective that they unhinge the theory’s assumptions (see Box 2.1 and Box 2.2 in Sect. 2.3.2.2; e.g. Agrawal and Knoeber 1996, p. 377; Chang and Wong 2003, p. 7)—the main prediction of principal-agent theory is the alternative hypothesis (\( H_4^1 \)) that discretion decreases performance (see Sect. 2.3.2; e.g. Caza 2007, p. 10; Caza 2011; Chang and Wong 2003, p. 7; Davis et al. 1997b, p. 38; Hutzschenreuter and Kleindienst 2007, p. 4; Jensen and Murphy 1990; Zhao et al. 2010).

In terms of the algebraic notation utilised throughout this chapter, the parameter \( d \) measures the direct effect of managerial discretion (\( D \)) on performance (\( P \)), which is equal to the total impact of discretion on performance (\( {{{\partial P}} \left/ {{\partial D}} \right.}=d+m\cdot M \)) when moderators are zero (\( M=0 \)), i.e. when they take their average values (see Sect.​ 4.​3.​2 on comparative statics; e.g. Aiken and West 1991, p. 37; Dowling 2000, pp. 284–291; Finney et al. 1984; Henseler and Fassott 2010, p. 728; Hirschey 2009, p. 99).

Universally rejected refers to rejecting the null hypothesis after accounting for the greater granularity, e.g. rejecting the null hypothesis when differentiating discretion’s impact by multiple dimensions of discretion.

If the empirical results are not consistent with any single existing theory for the present study’s particular instance of plant managers in China, this would serve as a proof-by-counter-example for the aforementioned postulate that neither existing theory can fully explain the empirical impact of discretion on performance after applying the greater granularity, since then neither theory could fully explain the impact of discretion on performance at least in the particular instance of the present study and therefore not universally in all studies.

As explained in Chap.​ 3, there are tens of thousands of plant managers in China alone (Guojia tongji ju [National Bureau of Statistics] 2007, 14–1, 14–2, 14–18). With plant managers being but one example of middle managers, this translates into an even larger number of middle managers in organisations worldwide.