Subordinate–manager gender combination and perceived leadership style influence on emotions, self-esteem and organizational commitment

Journal of Business Research 58 (2005) 115 – 125

Subordinate–manager gender combination and perceived leadership style influence on emotions, self-esteem and organizational commitment Janet R. McColl-Kennedya,*, Ronald D. Andersonb,1 a

UQ Business School, University of Queensland, Brisbane, Queensland 4072, Australia b Kelley School of Business, Indiana University, Indianapolis, IN 46202-25151, USA

Received 1 February 2002; received in revised form 1 October 2002; accepted 1 April 2003

Abstract A theoretical model was developed to investigate the relationships among subordinate – manager gender combinations, perceived leadership style, experienced frustration and optimism, organization-based self-esteem and organizational commitment. The model was tested within the context of a probabilistic structural model, a discrete Bayesian network, using cross-sectional data from a global pharmaceutical company. The Bayesian network allowed forward inference to assess the relative influence of gender combination and leadership style on the emotions, self-esteem and commitment consequence variables. Further, diagnostics from backward inference were used to assess the relative influence of variables antecedent to organizational commitment. The results showed that gender combination was independent of leadership style and had a direct impact on subordinates’ levels of frustration and optimism. Female manager – female subordinate had the largest probability of optimism, while male manager teamed with a male subordinate had the largest probability of frustration. Furthermore, having a female manager teamed up with a male subordinate resulted in the lowest possibility of frustration. However, the findings show that the gender issue is not simply female managers versus male managers, but is concerned with the interaction of the subordinate – manager gender combination and leadership style in a nonlinear manner. D 2003 Elsevier Inc. All rights reserved. Keywords: Leadership style; Gender combinations; Frustration; Optimism; Organization-based self-esteem; Organizational commitment; Bayesian networks

1. Introduction A recent study (Fletcher et al., 2000) claimed that women leaders especially place value on building and fostering relationships with their supervisees in order to realize performance outcomes. But this focus has not been appreciated as ‘‘real’’ work, being relegated to things ‘‘women do’’ or they are being ‘‘nice’’ and ‘‘helpful’’ (Fletcher et al., 2000) or worse being considered as a negative practice. However, no study to date has demonstrated that female managers have been able to significantly impact, either positively or negatively, organization-based self-esteem or organizational commitment. Furthermore, no study has investigated the impact * Corresponding author. Tel.: +61-7-3365-6673; fax: +61-7-33656988. E-mail addresses: [email protected] (J.R. McColl-Kennedy), [email protected] (R.D. Anderson). 1 Tel.: +1-317-274-2446; fax: +1-317-274-3312. 0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00112-7

of gender combinations of managers and subordinates on organizational outcomes. Also, the role of subordinate – manager gender combinations and perceived leadership style on subordinates’ emotions of optimism and frustration is an unstudied area. Knowledge of these influences has significant ramifications for both theory and practice. The structure of the paper is to first outline a theoretical framework while highlighting important gaps in the literature. Specifically, gender combination and interactions, leadership styles, emotions, organization-based self-esteem and organizational commitment are reviewed. The initial section concludes with a statement of the objectives of the study. Next, methodological issues are addressed, with an emphasis on the requirements of causal claims and the appropriateness of Bayesian networks. The Results section presents a causal model, in the form of a Bayesian network, and the probabilistic inference findings based on network interventions. Finally, a discussion of results, implications and limitations conclude the paper.

116

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

2. Conceptual framework and objectives 2.1. Gender combinations and interactions Differences in the way males and females manage have been widely acknowledged (Eagly and Johnson, 1990; Lewis, 2000). In particular, female superiors are regarded as less competent than male superiors when providing criticism to their subordinates (Sinclair and Kunda, 2000). Although the traditional stereotype of a leader is that of a male (Oakley, 2000), with men being perceived to be ‘‘better’’ leaders than women (Heilman et al., 1989; Nieva and Gutek, 1980; Gutek, 1985), a current contention is that female leaders particularly endeavor to develop mutually rewarding relationships with their subordinates, further highlighting the high value female managers appear to put on relational aspects (Fletcher et al., 2000). The key appears to lie in the way the women participate in ‘‘growth fostering relationships,’’ which implies mutual empathy and empowerment (Jordan et al., 1991). The female focus is on participating in ‘‘connection,’’ mutuality, interdependence and collectivity rather than the traditionally masculine focus on self-gratification, autonomy, competition and independence. Fletcher et al. (2000, pp. 251 –253) found in a detailed workplace study that women believed effective outcomes were best achieved through a ‘‘context of connection where the conditions of mutual psychological growth such as empathy, mutuality, authenticity and empowerment are met.’’ Effective managers do not work in isolation from their subordinates. Rather, they work with their subordinates in what is often referred to as dyadic relationships of manager – subordinate (Brower et al., 2000). The nature of the relationship between the manager and subordinate has been acknowledged as both complex and interactive. Furthermore, it is widely acknowledged that there is exchange and reciprocity in manager – subordinate relationships. Indeed, leader – member exchange theory claims that ‘‘both parties bring something of value to the exchange, and that the two individuals become interrelated’’ (Brower et al., 2000, p. 230). This clearly suggests that it is not sufficient to consider only one of the parties involved. Rather, we must consider the interaction of the subordinate –manager combination of the manager and the gender of each party. Therefore, given that (1) the findings by Oakley (2000) and Fletcher et al. (2000) suggest that female managers appear to interact with their subordinates differently to male managers by placing importance on developing and building mutually satisfying relationships and that this is likely to impact on outcomes for the individuals involved as well as the respective organizations; (2) the study of female leaders by Fletcher et al. (2000) found that female managers were able to achieve effective outcomes by connecting with their subordinates and developing their subordinate’s competencies including self-confidence; (3) that we cannot simply look at the gender of the manager in isolation from the gender of the subordinate; and (4) no study to date has

investigated the gender combination of manager and subordinate and the impact of this on outcomes; we argue that demonstrating that these gender combinations produce differential results in terms of organizational outcomes has major theoretical and practical implications. In sum, it may not be simply whether the manager is male or female but the gender combination of manager and subordinate that is important. 2.2. Leadership style The style of the leader is considered to be particularly important in achieving organizational goals (Dubinsky et al., 1995) and so it is not surprising that many studies endeavor to categorize leadership style. In this study, leadership style is categorized into four styles: laissez faire, management-byexception, contingency reward and transformational leadership styles, following Bass (1985a) and Bass and Avolio (1994). These style categories have been widely applied in training efforts and evaluation studies (Bass and Avolio, 1994), as well as a typology in academic research (e.g., Sosik and Dionne, 1997). Laissez faire is a passive style that is reflected by high levels of avoidance, indecisiveness and indifference. Typical behaviors of management-by-exception leadership style include standard setting, deviation monitoring, error searching, rule enforcement and a focus on mistakes. Transactional exchanges of rewards and recognition for accomplishments for desired outcomes characterize the contingency reward style. The concept of transformational leadership, named by Burns (1978) in his exploration of ‘‘world class leaders’’ and championed by Bass (1985a,b, 1990, 1997, 1998), has enjoyed wide theoretical and practical acceptance (Tichy and Devanna, 1986; Stewart, 1994; Avolio, 1998). Transformational leadership has been described as guidance through individualized consideration, intellectual stimulation, inspirational motivation and idealized influence (Bass, 1997). Individualized consideration emphasizes personal attention, while intellectual stimulation encourages use of reasoning, rationality and evidence. Further, inspirational motivation is assumed to raise levels of optimism and enthusiasm, while idealized influence provides a vision and sense of mission (Dubinsky et al., 1995). Studies have investigated relationships between transformational leadership style and a wide range of consequences, such as leadership trust (Podsakoff et al., 1996), self-efficacy beliefs (Kirkpatrick and Locke, 1996), leadership satisfaction (Yammarino and Bass, 1990), worker absenteeism and satisfaction (George and Jones, 1997; Staw et al., 1994; Weiss and Cropanzano, 1996) and performance outcomes (Barling et al., 1996; Howell and Avolio, 1993). Transformational leadership has consistently been demonstrated to offer benefits over the other three styles, particularly in terms of achieving organizational goals (Bass and Avolio, 1994; Dubinsky et al., 1995), and being able to evoke work effort (Barling et al., 1996).

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

2.3. Emotions Bagozzi et al. (1999) define emotions as mental states of readiness that arise from cognitive appraisals of events or from one’s thoughts. The past reluctance to even acknowledge the existence of emotions in a work setting is changing. It is clear that leaders and supervisees in their multiple interactions with fellow workers are exposed to situations that produce emotions, which can potentially influence their feelings, attitudes and behaviors. So much so that many have claimed that the workplace is ‘‘one of the most interpersonally frustrating contexts that people have to deal with’’ (Fitness, 2000, p. 148). Thus, it is expected that frustration will be frequently experienced. Indeed, Goleman (1998, p. 32) argues that ‘‘Interpersonal ineptitude in leaders lowers everyone’s performance: It wastes time, creates acrimony, corrodes motivation and commitment, builds hostility and apathy.’’ Leaders often express positive emotions such as enthusiasm and optimism in order to motivate their subordinates (Ashkanasy and Tse, 2000; Bass, 1990; Conger and Kanungo, 1998; Lewis, 2000). They also have been shown to express negative emotions such as anger and frustration in their interactions with their supervisees. Transformational leaders especially seem to use positive emotions and appear to be comfortable with their expression of such emotions (Dubinsky et al., 1995). Furthermore, they tend to be more optimistic and sensitive to what their supervisees are feeling (Spreitzer and Quinn, 1996). The mediating role of emotions on job attitudes and behaviors (Ashkanasy et al., 1998; Fisher, 1999) and on leadership style (Lewis, 2000) has also been investigated. Yet, notwithstanding these findings, much is still unknown about the influence of gender relationships and management style on the emotions of supervisees and how emotions in turn influence important outcomes, such as organization-based self-esteem and organizational commitment.

117

(Mowday et al., 1979). Organization-based self-esteem has been found to be positively related to organizational commitment in multiple studies (Pierce et al., 1989; Tang and Gilbert, 1994; Hui and Lee, 2000). 2.5. Objectives The first set of objectives is to develop a causal model that will test hypothesized relationships and provide a descriptive summary of the model variables. The model will allow evaluation of whether (1) subordinate – manager gender combinations have direct, indirect or no influence on perceived leadership style, the frequency of experienced frustration and optimism, organization-based self-esteem and organizational commitment; (2) leadership style has direct, indirect or no influence on frustration, optimism, self-esteem and commitment; and (3) frustration and optimism have direct, indirect or no influence on self-esteem and commitment. These objectives are accomplished within the context of a probabilistic structural model, a discrete Bayesian network (Pearl, 1988, 1998, 2000). The Bayesian network methodology provides a decision-oriented approach consistent with evaluations of actions necessary for potential change (Anderson and Lenz, 2001). Also, a Bayesian network allows prediction by forward inference and diagnostics from backward inference by application of Bayes’ theorem. The second set of objectives is concerned with assessment of the effects of assumed manipulations or interventions on system variables in the Bayesian network. A unique feature of Bayesian networks is such that interventions will allow forward inference for the assessment of the influence of antecedents on consequence variables and backward inference to evaluate changes in the antecedent variables.

3. Method 3.1. Causal modeling theory

2.4. Organization-based self-esteem and organizational commitment Self-esteem is a concept derived from the multidimensional concept of ‘‘self’’ (Coopersmith, 1967). Self-esteem expresses an ‘‘attitude of approval or disapproval of self; it is a personal evaluation reflecting what people think of themselves as individuals; it is the extent to which individuals believe themselves to be capable, reflecting a personal judgment of worthiness’’ (Pierce et al., 1989, p. 625). Thus, organization-based self-esteem is taken as an attitude that represents self-esteem that is specific to contexts and perceptions of an organization as constructed from past experiences (Hui and Lee, 2000). Organizational commitment has been defined as the relative strength of an employee’s identification with the organization and is measured by expressed pride in the organization and caring about the organization’s future

Our objective is to construct a relationship model that approximates reality and may be given a causal interpretation. The theory of probabilistic causation provides the conceptual basis for causal modeling (Reichenbach, 1956; Suppes, 1970; Eells, 1991). The key assumptions underlying the probabilistic causation perspective are (1) incomplete knowledge of causes results in probabilistic cause – effect relationships and (2) the occurrence of an event C increases the probability of E, p(EjC)>p(E). The first conceptualization, labeled as pseudoindeterminism (Spirtes et al., 1993), assumes that C does not itself determine E but does so in conjunction with an unobserved variable U that represents unspecified causes: C ! E p U. This notion provides a basis for separating causal relations from spurious associations. Recent work in causal modeling has employed elements of graph theory and probability theory to translate the

118

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

metaphysical concept of pseudoindeterminism into a useful methodology. The utilized graphical structure is a directed acyclic graph (DAG) that represents the set of variables as a path diagram structure. The structure is characterized as directed, since two-headed arrows depicting noncausal association are not allowed and as acyclic, feedback loops (e.g., X ! Y ! X) are not allowed. Any graph that conforms to the DAG requirements characterizes the causal Markov condition, which states that every consequence variable is independent of all variables that are not its effects, conditional on its direct antecedents in the structure (Spirtes et al., 1993). Thus, a DAG portrays recursive relationships where each arrow depicts causal dependence, and the absence of a connecting arrow indicates causal independence. The causal Markov condition allows the joint probability distribution for a DAG (e.g., W ! X ! Z p Y) to be decomposed into a product of conditional probabilities [e.g., p(w) * p(xjw) * p( y) * p(zjx,y)]. This decomposition property provides the capability to assess how the system of variables under study would respond to the manipulation or intervention of selected variables (Scheines, 1997). Since a DAG provides a portrayal of the conditional independence assertions consistent with the pseudoindeterminism concept, specified causes ! effect p unspecified causes, the graph can always be translated into a set of recursive structural equations with independent errors that satisfy the causal Markov condition (Kiiveri and Speed, 1982). The set of recursive structural equations that describe the data-generating process of a DAG is specified by vi = fi(directi,ui), where vi is a value of a consequence variable Vi linked by a function fi to a configuration of direct antecedent causes, directi and ui, a variable representing unspecified disturbances (Pearl, 1996, 1998, 2000). The disturbances are assumed to represent mutually independent unobserved variables, each with a probability distribution function p(ui) that reflects the uncertainty of Vi. The equation, vi = fi(directi,ui), reflects the view that nature possesses stable causal mechanisms that are deterministic functional relationships between variables, represented by fi, while p(ui) reflects our incomplete knowledge of causes of Vi. Mathematically, the equation vi = fi(directi,ui) can accommodate relationships expressed in terms of the estimated probability of an effect as a nonlinear function of direct antecedents.

Markov condition. Therefore, each variable is independent of all network variables that are not its effects, conditional on its direct antecedents. Further, the empirical regularities of conditional independence relations in a population are assumed due to a causal structure not coincidence or specific parameter values. This assumption, labeled as the stability condition (Pearl, 2000) or the faithfulness condition (Spirtes et al., 1993), allows the joint population probability distribution over the network variables to be considered faithful to the independence relations of the underlying causal structure as specified by the network structure. The causal Markov and faithfulness conditions jointly provide the basis for separating causal relations from spurious associations if the set of network variables includes all relevant common causes. The variable set is said to be causally sufficient when no common causes are omitted. Bayesian networks combine the pseudoindeterminism perspective with specific causal assertions and the associated conditional probability distributions to provide an operational tool for causal modeling. Assertions described in a Bayesian network are derived from background knowledge of temporal order, systematic conjecture and prior empirical results. Such assertions are assumed to be causal if they satisfy causal sufficiency, the causal Markov and faithfulness conditions, independent of specified and unspecified causes and association evidence. Conceptually, the framework for Bayesian networks is applicable to continuous or discrete variables or both. However, applications have been concerned with discrete Bayesian networks where the vast majority of computational work has been focused. Cooper (1999), Cowell (1999) and Heckerman (1999) provide technical reviews of Bayesian networks. The Bayesian network methodology offers an explanatory description of causal relationships plus manipulation capabilities for diagnosing the key changes necessary for system improvement and for predicting the impacts of potential change actions. The intervention capabilities are derived from the joint probability decomposition property of the network. As such, this methodology is particularly appealing in that it provides an explanatory description of causal relationships between gender combinations, leadership style, emotions, organization-based self-esteem and organizational commitment. 3.3. Sample

3.2. Bayesian networks A Bayesian network is a DAG with nodes representing random variables and arrows portraying direct causal relationships between variables. Each node in a Bayesian network has a conditional probability distribution that assigns probabilities to the values of the variable represented by the node. Thus, the probability of a variable value is conditional upon the set of values of its direct causes. A network encapsulates a set of independence causal assertions between represented variables according to the causal

A survey of 139 sales representatives of a global pharmaceutical company in Australia was undertaken. Only two of the representatives did not complete the self-administered questionnaires, yielding a response rate of 98.6%. Of the 137 respondents, complete measurements on study variables were available for 127 representatives. The sample was comprised of 39 male representatives reporting to a male manager, 20 male representatives with a female manager, 24 female representatives reporting to a female manager and 44 female representatives reporting to a male manager. The

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

average age for females was 33 years and 39 years for males. Seventy-two percent of the males and 75% of the females had completed a university degree. 3.4. Measurements The variables to be modeled are subordinate– manager gender combinations (gender), leadership style (leadership), experienced frequency of optimism (optimism) and frustration (frustration), organization-based self-esteem (esteem) and organizational commitment (commitment). The four subordinate – manager gender combinations were obtained from company records. The remaining measurements for leadership style, optimism, frustration, organization-based self-esteem and organizational commitment were based on self-reported responses to questionnaire items. Leadership is often viewed as a multivariate concept and the measurement is assumed to reflect a continuous variable that encompasses multiple facets. However, this study is concerned with the type of leadership style of the manager as perceived by the subordinate. So although each manager will possess certain aspects of laissez faire, management-byexception, contingency reward and transformational leadership qualities, respondents were asked to ‘‘typecast’’ their manager into the most descriptive category of their management style. This use of discrete categories has been used frequently in describing and comparing leadership styles (Bass, 1985a, 1990; Bass and Avolio, 1994; Sosik and Dionne, 1997). We consider this form of scenario categorization to be consistent with semantics used daily in the workplace for evaluation of people and objects. The style selection was implemented by asking each subordinate to select one of four available unlabeled descriptions that best characterized the leadership style of their manager. The experienced frequencies of frustration and optimism are conceptually viewed as variables with a mixture of discrete and continuous aspects. The discrete part has two states, no frustration or frustrated and no optimism or optimistic. The continuous part is the magnitude of frustration or optimism. The questionnaire measurement of frustration involved three variables: the self-reported frequency of experienced irritation, tenseness and frustration within the organization. The measures for optimism were also selfreported frequencies of experienced enthusiasm, excitement and optimism. The frequency of each emotion variable was measured on a five-point ordinal scale (‘‘never,’’ ‘‘very seldom,’’ ‘‘sometimes,’’ ‘‘often,’’ and ‘‘very frequent’’). Variables that represent the degree of organization-based self-esteem and organizational commitment are viewed conceptually as continuous, with high positively skewed distributions. Four questionnaire items were used to measure organization-based self-esteem: ‘‘I count around here,’’ ‘‘I am trusted in this organization,’’ ‘‘This organization has confidence in me,’’ and ‘‘I can make a difference to this organization’’ (Pierce et al., 1989). Four items were also used to measure organizational commitment: ‘‘I put in effort

119

beyond what is expected,’’ ‘‘My values and the organization values are similar,’’ ‘‘I am proud to tell others that I am part of this organization,’’ and ‘‘I really care about the future of this organization’’ (Mowday et al., 1979). Each questionnaire item for organization-based self-esteem and organizational commitment was measured on a six-point agreement scale (‘‘strongly disagree’’ to ‘‘strongly agree’’). The variables measured in the questionnaire were used for the measurements of the frustration, optimism, organization-based self-esteem and organizational commitment concepts. The categorical measurements for each concept were obtained from applications of k-means cluster analysis using the questionnaire item responses as inputs. Discriminant analyses verified that little information was lost due to the two-state categorization. The analyses showed 100% correct classification for the two-state concept representation of frustration and optimism based on the questionnaire measurements. The two-state concept representation of selfesteem and commitment yielded 98.43% and 96.06% correct classifications. Of course, the results of these discrimination analyses are to be interpreted as only descriptive, not inferential, but they are supportive of a meaningful translation of the questionnaire measures into concept variables. An issue as to whether the subordinate– manager combination is more appropriate than simply the gender of the manager is addressed before moving on to building the model structure. The ratio of dispersion explained by the gender combination to the dispersion explained by manager gender for other model variables provided a quantitative assessment. The ratios of explained dispersion from logit analyses were all in favor of the gender combination: 1.70 for leadership, 1.25 for frustration, 1.26 for optimism, 14.38 for esteem and 3.67 for commitment. Thus, the subordinate –manager combination was found to have a substantive prediction advantage over just the manager’s gender. 3.5. Building the causal structure The set of model variables for the Bayesian network application are V={gender, leadership, frustration, optimism, esteem, and commitment}, and the set of unspecified error variables, U={u1, u2, u3, u4, u5, and u6}. Each probability distribution of ui, p(ui), is assumed to be a collection of multinomial distributions, one for each configuration of direct causes. The process of constructing a graphical structure starts with a temporal ordering of the study variables V. The following assumptions are made based on the findings of the reviewed studies and conceptual developments outlined in the first section of the paper: 1. Organization-based self-esteem is prior to and is a direct contributor to the formation of organizational commitment (i.e., esteem ! commitment). 2. The frequencies of experience of frustration and optimism are taken to be prior to organization-based self-

120

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

4. Results 4.1. Tests of the causal claims

Fig. 1. Hypothesized relationships.

esteem. Since self-esteem is viewed as an attitude constructed from past organizational experiences (Pierce et al., 1989, 1993; Hui and Lee, 2000) and emotions are a part of those experiences, the priority assumption seems justified. We postulate that each emotion is a potential direct cause of self-esteem (frustration ! esteem and optimism ! esteem). Further, we see no rationale for assuming a temporal priority for frustration versus optimism, nor for assuming a causal relationship between frustration and optimism, although a negative noncausal association may be manifested in the sample data. 3. The frequency of experienced frustration and optimism and formation of organization-based self-esteem and organizational commitment attitudes should result, in part, from subordinate interactions with the manager. Thus, we take the leadership style of the manager to be prior to these variables. Leadership style is postulated to be a direct cause in the formation of organizationbased self-esteem (leadership ! esteem), and leadership style is hypothesized to have an indirect effect on organizational commitment, mediated by self-esteem (leadership !esteem ! commitment). Further, we hypothesize that the frequency of frustration and optimism is, in part, a direct result of the leadership style of a manager (leadership ! frustration and leadership ! optimism). 4. A subordinate –manager gender combination is taken to be prior to all other model variables. Thus, gender combination is viewed as an exogenous variable. An extended discussion of exogeneity is provided by Pearl (2000, pp. 165 –170). We postulate gender combination to be a direct cause of each emotion (gender ! frustrafrustration and gender ! optimism). The relationship between gender combination and the other study variables is an open question and is investigated in the analysis. The above temporal assumptions and hypothesized causal relationships are the basis for the graphical representation of Fig. 1 that satisfies the requirements of a DAG.

The theory of probabilistic causation provides the basis for supporting or rejecting the validity of causal claims. The statistical evaluation of the claims of Fig. 1 was investigated using the sequential testing procedure, the PC algorithm of TETRAD II (Scheines et al., 1994). The procedure provided values of the G2 test statistic and P values associated with tests of independence, analogous to those described above, guided by the specified temporal order of the variables. Scheines et al. (1994), based on a series of simulation experiments, recommend a nominal significance level of .10 for sample sizes of 100– 300 to address the issue of overall error rate for the sequential testing. However, the traditional significance level of .05 was taken as the basis for rejection of independence and as evidence to support the decision to retain a dependency relationship A ! B. Conversely, when the hypothesis of statistical independence is retained, the dependency relationship is rejected. The sequential statistical testing began with a fully connected graph with temporal order specifications and the restriction of causal independence of frustration and optimism. Fourteen tests of independence were conducted, and the obtained values of G2 and the P values are summarized in Table 1. The hypothesized relationships of Fig. 1 were supported with two exceptions. The causal claim optimism ! esteem is rejected based on retaining the hypothesis of independence of optimism and esteem, given leadership. Thus, the evidence is that leadership as a common cause of both optimism and esteem screens off the relationship between these effects. Further, the hypothesis of conditional independence of optimism and commitment, given esteem, was rejected. Thus, the evidence is that esteem does not screen off optimism from commitment, and the dependency link optimism ! commitment should be added to Fig. 1.

Table 1 Summary of independence testing Relationship

G2

df

p value

Relationship decision

Gender ! leadership Gender ! optimism Gender ! frustration Gender ! esteem Gender ! commitment Leadership ! optimism Leadership ! frustration Leadership ! esteem Leadership ! commitment Esteem ! commitment Optimism ! esteem Optimism ! commitment Frustration ! esteem Frustration ! commitment

13.82 7.84 13.42 2.60 6.64 16.90 17.60 15.80 7.03 19.40 1.71 9.36 10.29 5.49

9 3 3 3 3 3 3 3 3 1 4 2 4 2

.135 .049 .004 .464 .088 < .001 < .001 .002 .075 < .001 .789 .009 .038 .068

Reject Retain Retain Reject Reject Retain Retain Retain Reject Retain Reject Retain Retain Reject

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

The structure of Fig. 1 was modified to conform to the causal claim findings, and conditional probability estimates were estimated to conform to the supported dependency linkages. The conditional probability tables were calculated using the Bayesian Knowledge Discoverer program (Ramoni and Sebastiani, 1998). Fig. 2 displays the resulting Bayesian network. 4.2. Model fit The statistical analysis of the causal claims and a descriptive evaluation of the internal consistency of the network provide support for the fit of the Bayesian network of Fig. 2. A leave-one-out cross validation was used to assess the internal consistency of the Bayesian network (Stone, 1974, 1977). The method selected one case as the response and used the remaining cases in a model to predict the omitted case. The procedure was repeated for every case in the sample. Frustration had the lowest accuracy at 73.8% and esteem had the highest accuracy at 85.7%. The overall predictive accuracy for the consequence variables was 78.6%.

121

4.3. Conditional probabilities The conditional probability tables, displayed in Fig. 2, show the transformational leadership style had the largest probability of optimism within each gender combination. The subordinates of female managers exhibited larger probabilities of optimism than the subordinates of male managers for each leadership style, except for the management-byexception style. This result is consistent with the connection hypothesis discussed above since the management-by-exception style places emphasis on negative outcomes. The probabilities of frustration were lower in the male subordinate –female manager category than in each leadership style, except for the transformational leadership style. Male subordinates with male managers exhibited higher levels of frustration across all leadership categories. The laissez-faire style had the larger probabilities of frustration across gender combinations, followed by management-by-exception style. Transformational leadership retained a high probability of self-esteem in the state of frustration, whereas other leadership styles displayed a sharp decrease.

Fig. 2. The Bayesian network.

122

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

Table 2 Influence indexes for gender combination and leadership style interventions Intervention

High frustration

High optimism

High self-esteem

High commitment

114 100 97 93

98 109 96 101

103 104 97 99

66 86

126 96

120 96

118 97

127

52

85

81

175

99

81

90

Subordinate – manager gender Female – female 95 Male – female 52 Male – male 136 Female – male 95 Leadership style Transformational Contingency reward Management by exception Laissez faire

4.4. Probabilistic inference A Bayesian network allows manipulation from an intervention in the context of a thought or hypothetical experiment. That is, a selected variable or set of variables is fixed in a given state, and the application of Bayes’ theorem provides revised probabilities for all other system variables. This process, termed as probabilistic inference, is concerned with the revised probabilities for a set of variables, called the query, when an intervention fixes the values of another set of variables, called the evidence. Any set of variables in the Bayesian network can serve as a query or as evidence. Thus, inference can be focused on prediction of consequence variables or on diagnostics of antecedent variables. The preintervention probabilities provided a baseline for evaluation of the magnitude of change created by an intervention. The magnitude of change for a query from intervention evidence was measured by an influence index, computed as the ratio of post- to preintervention probability and

expressed as a percentage. Thus, an influence index of 100 indicates no change from the intervention, an index less than 100 indicates a probability decrease and an index greater than 100 indicates a probability increase. Table 2 provides a summary of the influence indexes for frustration, optimism, self-esteem and commitment resulting from interventions on gender combination, leadership style and commitment. Female managers produce the highest levels of optimism in their subordinates irrespective of whether the subordinate is male or female, and the highest levels of frustration were experienced when male subordinates were teamed up with male managers. Similarly, the highest levels of self-esteem among subordinates were experienced when the manager was female and the subordinate male. But interestingly, the next highest level of self-esteem was experienced among male manager and female subordinate combinations. Highest levels of commitment were experienced when female and male subordinates were teamed up with female managers. Large changes in the probability of frustration for male subordinates from gender combination interventions are indicated by the influence indexes of Table 2. The male subordinate– male manager intervention showed an increase in frustration of 36% from preintervention probability. However, the male subordinate –female manager intervention yielded a 48% decrease in frustration with reference to the preintervention probability. Changes from interventions of female subordinates were very small. Relative changes in the probability of optimism were much smaller than those of frustration gender interventions. The female subordinate – female manager combination showed the greatest positive change of 14%, and the female subordinate– male manager had the largest negative change of 7%. The indirect effects of gender combination on organization-based self-esteem and organizational commitment were small as indicated by the influence indexes.

Table 3 Influence indexes for interventions on both gender combination and leadership style Intervention Gender combination

Leadership style

Female – female Female – female Female – female Female – female Male – female Male – female Male – female Male – female Female – male Female – male Female – male Female – male Male – male Male – male Male – male Male – male

Transformational Contingency reward Management by exception Laissez faire Transformational Contingency reward Management by exception Laissez faire Transformational Contingency reward Management by exception Laissez faire Transformational Contingency reward Management by exception Laissez faire

High frustration

High optimism

High self-esteem

High commitment

18 68 195 209 77 30 32 61 59 89 105 168 95 123 159 220

143 122 20 133 141 96 20 106 107 88 91 72 130 88 43 101

120 99 73 74 120 103 103 104 120 96 89 83 120 93 80 71

122 103 71 93 121 99 81 103 112 94 91 84 118 93 78 85

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

The influence indexes resulting from the four leadership style interventions are also displayed in Table 2. Transformational leadership had the lowest probability of frustration and the largest probability of optimism. The laissez-faire style had the largest probability of frustration, and the management-by-exception style had the lowest probability of optimism. Transformational leadership also had the largest probability of high self-esteem and commitment. The laissez-faire leadership style had the lowest probability of high self-esteem, and the management-byexception style had the smallest probability of high commitment. Table 3 displays the influence indexes from interventions on both gender combination and leadership style. These results show that over all gender combinations, the probability of frustration decreases (influence index < 100) with the transformational leadership style; the probabilities of optimism, self-esteem and commitment are at their maximum levels with the transformational leadership style, and the probabilities of optimism and commitment decrease with the management-by-exception style. Also, the patterns of the influence indexes are similar across the leadership styles for the female – female, female – male and male – male subordinate – manager combinations. The male subordinate – female manager combination exhibits relatively small probabilities of frustration over all leadership styles. Organizational commitment interventions provide backward inference for the other variables in the Bayesian network. The changes from pre- to postintervention probabilities for gender differences were minor. However, the influences of leadership style, optimism, frustration and esteem were substantial. The largest change in probability from pre- to postintervention for low versus high commitment was self-esteem, followed by optimism. A summary of the relative changes in self-esteem, frustration and optimism as antecedents of commitment under low state and high state interventions is given in Fig. 3. The postintervention probability of transformational leadership style, under the condition of low commitment, de-

Fig. 3. Relative change in optimism, frustration and esteem: low versus high commitment.

123

Fig. 4. Relative change in leadership style: low versus high commitment.

creased 35% from the baseline preintervention probability. The probabilities of management-by-exception and laissezfaire styles increased 38% and 24% from the baseline level. The high commitment intervention showed an 18% increase in the influence index for transformational leadership style and exhibited negative changes for all other styles. Fig. 4 summarizes the influence indexes for the leadership styles under low and high commitment interventions.

5. Discussion and conclusions The lack of a direct relationship between gender and leadership style provides partial support for the contention that management guidelines are not related to gender alone (Moncrief et al., 2000). That is, it is not simply whether the leader is male or female that is important. What is important is the gender combination of manager and subordinate, as the interaction between manager and subordinate impacts on other variables. Our study found an interacting set of relationships involving subordinate – manager combinations and leadership styles. For example, we found that the female subordinate – female manager combination with transformational leadership style had the most favorable set of probabilities for positive emotions, self-esteem and commitment (see the influence indexes in Table 3). But the female subordinate –female manager gender combination with the management-by-exception style had the least favorable profile of probabilities for the consequence variables. Further, the male subordinate – female manager combination did not show similar results with the female subordinate – female manager combination. Furthermore, our study showed gender combination to have a minor impact under the high commitment intervention (influence indexes ranged from 100 to 106) and under the low commitment intervention (influence indexes ranged from 94 to 106). In contrast, leadership style, frustration, optimism and self-esteem had much greater influences on the commitment state (see Figs. 3 and 4). In general, the findings demonstrate that it is not simply female versus male managers, but a more complex nonlinear interaction of the subordinate –manager gender combination and leadership style.

124

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125

6. Management implications Our study clearly demonstrates that (1) female managers produce the highest levels of optimism in their subordinates irrespective of whether the subordinate is male or female (and that the highest levels of frustration were experienced when male subordinates were teamed up with male managers); (2) the highest levels of self-esteem among subordinates were experienced when the manager was female and the subordinate male; and (3) the highest levels of high commitment were experienced when the manager was female. Therefore, we suggest that organizations could particularly benefit from developing management training programs that focus on developing ‘‘female’’ management qualities, such as nurturing, developing relational aspects and engaging in ‘‘connection’’ and being sensitive to the emotional context of their subordinates rather than focusing on the traditional (and essentially ‘‘male’’) self-gratification. Traditionally, male managers have not been encouraged to demonstrate these ‘‘female’’ qualities of involving and nurturing subordinates, showing empathy and developing the subordinate’s competence and confidence. However, the gender of the manager alone should not solely be considered. Table 2 in particular showed that having a female manager did not always result in ‘‘better’’ outcomes. Teaming up a female manager with a male subordinate resulted in an increase in high self-esteem, but teaming up a female manager with a female subordinate did not improve subordinate self-esteem. Certainly, the female subordinate – female manager combination with transformational leadership style had the most favorable set of probabilities for emotions, self-esteem and commitment. But the female subordinate – female manager gender combination with the management-by-exception style had the least favorable profile of probabilities for the consequence variables. Further, the male subordinate –female manager combination did not show similar results with the female subordinate – female manager combination. What is critical is that the management recognizes that gender combinations coupled with leadership style results in differential outcomes for the organization.

7. Limitations and research directions The results of any causal modeling effort are limited by variable selection, the available database and the quantitative implementation. We have attempted to make a strong case, theoretically and operationally, for the employed modeling methodology. However, the study results must be evaluated with reference to the coarse-grain categorical measurement and the Bayesian network method. Further, the potential exclusion of variables that could have causal influences on the consequence variables is a possible threat to the findings. Finally, the study is conducted with the data of a single firm, rather than a sample of organizations, so it

is not appropriate to generalize the results to all organizations. Future research should explore whether the model of nonlinear interaction of manager –subordinate gender combinations and leadership style model of this study holds true for other organizations.

References Anderson RD, Lenz RT. Modeling the impact of organizational change: a Bayesian network approach. Organ Res Methods 2001;4(2):112 – 30. Ashkanasy NM, Tse B. Transformational leadership as management of emotion: a conceptual review. In: Ashkanasy N, Ha¨rtel C, Zerbe W, editors. Emotions in the workplace: developments in the study of the managed heart. Westport (CT): Quorum Books; 2000. p. 221 – 35. Ashkanasy NM, Fisher CD, Ha¨rtel CEJ, Ashforth BE. Investigating the causes and consequences of emotional experience at work. In: Ha¨rtel CEJ, Ashkanasy NM, chairs. Emotion at work: new research directions. Symposium presented at the Annual Meeting of the Society of Australian Social Psychologists Christchurch (New Zealand); 1998. April. Avolio BJ. Full leadership development: building the vital forces in organizations. Thousand Oaks (CA): Sage Publications; 1998. Bagozzi RP, Gopinath M, Nyer PU. The role of emotions in marketing. J Acad Mark Sci 1999;27(2):184 – 206. Barling J, Weber T, Kelloway EK. Effects of transformational leadership training and attitudinal and fiscal outcomes: a field experiment. J Appl Psychol 1996;81:827 – 32. Bass BM. Leadership and performance beyond expectations. New York: Free Press; 1985a. Bass BM. Leadership: good, better, best. Organ Dyn 1985b;13:26 – 40. Bass BM. Bass and Stogdill’s handbook of leadership. New York: Free Press; 1990. Bass BM. Personal selling and transactional/transformational leadership. J Pers Sell Sales Manage 1997;27(3):19 – 28. Bass BM. Transformational leadership. Mahwah (NJ): Erlbaum; 1998. Bass BM, Avolio BJ. Improving organizational effectiveness through transformational leadership. Thousand Oaks (CA): Sage; 1994. Brower HH, Schoorman FD, Tan HH. A model of relational leadership: the integration of trust and leader – member exchange. Leadersh Q 2000; 11(2):227 – 50. Burns JM. Leadership. New York: Harper and Row; 1978. Conger J, Kanungo R. Charismatic leadership in organizations. Thousand Oaks (CA): Sage; 1998. Cooper G. An overview of the representation and discovery of causal relationships using Bayesian networks. In: Glymour C, Cooper G, editors. Computation, causation and discovery. Menlo Park (CA): MIT Press; 1999. p. 3 – 62. Coopersmith J. The antecedents of self-esteem. CA: Freeman; 1967. Cowell R. Introduction to inference for Bayesian networks. In: Jordan MI, editor. Learning in graphical models. Cambridge (MA): MIT Press; 1999. p. 9 – 49. Dubinsky AJ, Yammarino FJ, Jolson MA, Spangler WD. Transformational leadership: an initial investigation in sales management. J Pers Sell Sales Manage 1995;15:17 – 29. Eagly AH, Johnson BT. Gender and leadership style: a meta analysis. Psychol Bull 1990;108(2):233 – 56. Eells E. Probabilistic causality. Cambridge: Cambridge Univ Press; 1991. Fisher CD. Affective events theory and organizational citizenship behavior. In: Daus C, chair. Affective events theory: empirical evidence and research implications. Symposium presented at the 14th Annual Conference of the Society for Industrial and Organizational Psychology, Atlanta (GA); 1999. May. Fitness J. Anger in the workplace: an emotion script approach to anger episodes between workers and their superiors, co-workers and subordinates. J Organ Behav 2000;21:147 – 62.

J.R. McColl-Kennedy, R.D. Anderson / Journal of Business Research 58 (2005) 115–125 Fletcher JK, Jordan JV, Baker Miller J. Women and the workforce: applications of a psychodynamic theory. Am J Psychoanal 2000;60(3):243 – 61. George JM, Jones GR. Experiencing work: values, attitudes, and moods. Hum Relat 1997;50:393 – 416. Goleman D. Working with emotional intelligence. New York: Bantam Books; 1998. Gutek B. Sex and the workforce. San Francisco: Jossey-Bass; 1985. Heckerman D. A tutorial on learning with Bayesian networks. In: Jordan MI, editor. Learning in graphical models. Cambridge (MA): MIT Press; 1999. p. 301 – 54. Heilman ME, Block CJ, Martell RF, Simon MC. Has anything changed? Cultural characteristics of men, women and managers. J Appl Psychol 1989;74(6):935 – 42. Howell JM, Avolio BJ. Transformational leadership, transactional leadership, locus of control and support for innovation: key predictors of consolidated-business-unit performance. J Appl Psychol 1993;78: 891 – 902. Hui C, Lee C. Moderating effects of organization-based self-esteem on organizational uncertainty: employee response relationships. J Manage 2000;26:215 – 32. Jordan J, Kaplan A, Miller JB, Stiver I, Surrey J. Women’s growth in connection. New York: Guilford Press; 1991. Kiiveri HT, Speed TP. Structural analysis of multivariate data: a review. In: Leinhardt S, editor. Sociological methodology. CA: Jossey-Bass; 1982. p. 209 – 89. Kirkpatrick SA, Locke EA. Direct and indirect effects of three core charismatic leadership components on performance and attitudes. J Appl Psychol 1996;81:36 – 51. Lewis KM. When leaders display emotion: how followers respond to negative emotional expression of male and female leaders. J Organ Behav 2000;21:221 – 34. Moncrief WC, Babakus E, Cravens DW, Johnston MW. Examining gender differences in field sales organizations. J Bus Res 2000;49:245 – 57. Mowday RT, Strees RM, Porter LW. The measurement of organizational commitment. J Vocat Behav 1979;14:224 – 47. Nieva VF, Gutek B. Sex effects on evaluations. Acad Manage Rev 1980;5: 267 – 76. Oakley JG. Gender-based barriers to senior management positions: understanding the scarcity of female CEOs. J Bus Ethics 2000;27(4):321 – 34. Pearl J. Probabilistic reasoning in intelligent systems. CA: Morgan Kaufmann; 1988. Pearl J. Structural and probabilistic causality. In: Shanks DR, Holyoak KJ, Medin DL, editors. The psychology of learning and motivation, vol 3. CA: Academic Press; 1996. p. 393 – 435. Pearl J. Graphs, causality, and structural equation models. Sociol Methods Res 1998;27:226 – 84. Pearl J. Causality: models, reasoning and inference. New York: Cambridge Press; 2000.

125

Pierce JL, Gardner DG, Cummings LL, Dunham RB. Organization-based self-esteem: construct definition, measurement, and validation. Acad Manage J 1989;32:622 – 48. Pierce JL, Gardner DG, Dunham RB, Cummings LL. Moderation by organization-based self-esteem of role condition-employee response relationships. Acad Manage J 1993;36:271 – 88. Podsakoff PM, MacKenzie SB, Bommer WH. Transformational leader behaviors and substitutes for leadership as determinants of employee satisfaction, commitment, trust and organizational citizenship behaviors. J Manage 1996;22:259 – 98. Ramoni M, Sebastiani P. Bayesian knowledge discoverer. UK: Knowledge Media Institute, Open University; 1998. Reichenbach H. The direction of time. CA: University of California Press; 1956. Scheines R. A introduction to causal inference. In: McKim VR, Turner SP, editors. Causality in crisis? IN: University of Notre Dame Press; 1997. p. 185 – 99. Scheines R, Spirtes P, Glymour C, Meek C. TETRAD II: tools for causal modeling. NJ: Erlbaum; 1994. Sinclair L, Kunda Z. Motivated stereotyping of women: she’s fine if she praised me but incompetent if she criticized me. Pers Soc Psychol Bull 2000;26(11):1329 – 42. Sosik JJ, Dionne SD. Leadership styles and Deming’s behavior factors. J Bus Psychol 1997;11:447 – 62. Spirtes P, Glymour C, Scheines R. Causation, prediction, and search. New York: Springer-Verlag; 1993. Spreitzer GM, Quinn RE. Empowering middle managers to be transformational leaders. J Appl Behav Sci 1996;32(3):237 – 61. Staw B, Sutton R, Pelled L. Employee positive emotion and favorable outcomes at the workplace. Organ Sci 1994;5:51 – 71. Stewart TA. How to lead a revolution. Fortune 1994;48 – 50 [November 28]. Stone M. Cross-validatory choice and assessment of statistical predictions. Royal Stat Soc B 1974;36:111 – 47. Stone M. Asymptotics for and against cross-validation. Biometrika 1977;64: 29 – 35. Suppes P. A probabilistic theory of causality. Amsterdam: North-Holland; 1970. Tang TL, Gilbert PR. Organizational-based self-esteem among health workers: a replication and extension. Public Pers Manage 1994;23:127 – 35. Tichy N, Devanna M. The transformational leader. New York: Wiley; 1986. Weiss HM, Cropanzano R. Affective events theory: a theoretical discussion of the structure, causes and consequences of affective experiences at work. In: Sutton RI, Staw BM, editors. Research in organizational behavior, vol 18. Greenwich (CT): JAI Press; 1996. p. 1 – 74. Yammarino FJ, Bass BM. Long-term forecasting of transformational leadership and its effects among naval officers: some preliminary findings. In: Clark KE, Clark MB, editors. Measures of leadership. Greensboro (NC): Center for Creative Leadership; 1990. p. 151 – 69.