During the 1980s and 1990s, many American firms adopted employee empowerment programs to help maintain their competitive edge in the face of rising global competition (Bowen and Lawler 1992, 1995; Conger and Kanungo 1988; Lawler, Mohrman, and Ledford 1995; Potterfield 1999; Spreitzer 1995, 1996; Thomas and Velthouse 1990). In the private sector, empowerment has been linked to higher levels of performance (Kirkman and Rosen 1999; Lawler, Mohrman, and Ledford 1992, 1995; Nielsen and Pedersen 2003; Spreitzer 1995), job satisfaction (Kirkman and Rosen 1999; Lawler, Mohrman, and Ledford 1995), and organizational commitment (Guthrie 2001; Kirkman and Rosen 1999; Lawler, Mohrman, and Ledford 1995). Empowerment gained currency among government reformers, having figured prominently in the new public management (NPM) reforms undertaken in the United States, United Kingdom, Australia, Canada, France, Sweden, and Norway (Kettl 2005; Matheson 2007; Peters 1996; Pollitt 1990; Wise 2002). A growing number of studies indicate employee empowerment is positively related to job satisfaction (Kim 2002; Lee, Cayer, and Lan 2006; Park and Rainey 2007; Wright and Kim 2004), organizational commitment (Park and Rainey 2007), and performance (Fernandez and Moldogaziev 2011; Lee, Cayer, and Lan 2006) in the public sector.
An important causal pathway by which employee empowerment influences performance is through innovative behavior on the part of frontline employees (Bowen and Lawler 1992, 1995; Gore 1993; Kanter 1983; Thomas and Velthouse 1990). Empowered employees improve performance by recovering quickly from errors in service delivery, learning from those recoveries, and generating innovative proposals for redesigning processes and products. Failure to encourage such behavior can seriously undermine the effectiveness of empowerment programs. The link between empowerment and encouragement to innovate is of particular significance in the public sector, where goal ambiguity, high levels of formalization, and restrictions on the ability to reward extrinsically might dampen or even neutralize the effects of empowerment efforts (see Rainey 2009). Based on Bowen and Lawler's (1992, 1995) conceptualization of employee empowerment, this study explores how different empowerment practices influence US federal government employees' level of encouragement to innovate. In this study, encouragement to innovate is viewed as an affective state or experience of feeling felt by public employees (see Locke and Latham 2004). The empirical results show that while employee empowerment as an overall approach can increase encouragement to innovate, empowerment practices have divergent effects, and some may even discourage innovation. Specifically, empowerment practices aimed at providing employees with access to job-related knowledge and skills and granting them discretion to change work processes increase encouragement to innovate. Offering employees rewards based on performance, however, reduces such encouragement. The next section offers a review of the literature on employee empowerment. The discussion then turns to the data and methods used in the analysis. The results of the statistical analysis are then presented. The study concludes with a discussion of its limitations and implications for theory and future research.
DEFINING EMPLOYEE EMPOWERMENT
Although scholars have made significant headway in developing the construct of employee empowerment, they have failed to reach consensus on what employee empowerment actually means (Conger and Kanungo 1988; Potterfield 1999; Thomas and Velthouse 1990). Two distinct theoretical perspectives have emerged in the literature, a managerial and a psychological one. From a managerial perspective, employee empowerment is a relational construct that describes how those with power in organizations (i.e., managers) share power and formal authority with those lacking it (i.e., employees) (Conger and Kanungo 1988). The intellectual origins of this construct date back to seminal contributions to the human relations movement in organization theory (e.g., Argyris 1957; Follett 1926; Likert 1967; McGregor 1960; see Potterfield 1999). Up until 1990, the tendency among scholars adopting the managerial perspective was to equate empowerment exclusively with delegating or sharing decision-making authority with frontline employees through various participative management techniques (Kanter 1983; Pettigrew 1972; Salancik and Pfeffer 1974). Dissatisfaction with this narrow characterization of employee empowerment led to two important developments: a refined view of empowerment as a multifaceted approach to management involving more than simply sharing authority with subordinates and the reconceptualization of empowerment as a psychological construct.
Bowen and Lawler (1992, 1995), who analyzed the growing use of empowerment practices among service firms, observed that sharing authority with frontline employees is necessary but insufficient for realizing the benefits of empowerment. As they noted, "many empowerment programs fail when they focus on 'power' without also redistributing information, knowledge and rewards" (1992, 32). Bowen and Lawler defined empowerment as an "approach to service delivery" entailing various management practices aimed at sharing four organizational "ingredients" with frontline employees: "(1) information about the organization's performance, (2) rewards based on the organization's performance, (3) knowledge that enables employees to understand and contribute to organizational performance, and (4) power to make decisions that influence organizational direction and performance" (1992, 32). Importantly, they argued that these four elements interact with each other to produce a multiplicative effect on performance.
Dissatisfied with the treatment of employee empowerment as a relational construct, another group of scholars worked to develop the psychological construct of empowerment. From this newer perspective, empowerment is an internal cognitive state characterized by enhanced feelings of self-efficacy (Conger and Kanungo 1988) or increased intrinsic task motivation (Thomas and Velthouse 1990). Based on Vroom's (1964) and Lawler's (1973) work on expectancy theory of motivation, Conger and Kanungo (1988) argued that one's motivation to increase effort is in part a function of two expectancies: the expectancy that one's effort will result in the desired level of performance (expectancy I, also called the self-efficacy expectation by Bandura, 1977, 1986) and the expectancy that performance will produce a desired outcome or reward (expectancy II). For Conger and Kanungo (1988), as employees become more empowered, their self-efficacy expectations will be enhanced, thereby increasing the amount of effort and time they dedicate to performing a task (476). Thomas and Velthouse (1990) defined empowerment as increased intrinsic task motivation that comes from making a task meaningful, identifying with it, and finding expressive value in it. Four personal assessments of a task are argued to positively influence intrinsic task motivation: impact, competence, meaningfulness, and choice. (1) To the extent that an employee makes positive assessments of these four aspects of the task, he or she will feel a heightened level of intrinsic task motivation and, therefore, be empowered. Spreitzer's (1995, 1996) research showed that elements of psychological empowerment resembling four task assessments by Conger and Kanungo are positively related to effectiveness and innovativeness.
Scholars clearly have divergent notions of what constitutes employee empowerment. One way out of this morass is to resist the temptation of taking sides in the debate over whether empowerment is a relational or motivational construct and instead to treat both as complementary pieces of the empowerment puzzle. These two constructs represent qualitatively different phenomena, the relational construct representing managerial behavior, and the motivational one representing employee cognition. Empowerment might best be understood as a process involving a set of management practices (sharing authority, resources, information, and rewards) that influence performance (effort, productivity) not only directly but also indirectly through their impact on employee cognition (self-efficacy, motivation, and job satisfaction) (see Bowen and Lawler 1992, 1995; Spreitzer 1995, 1996; Thomas and Velthouse 1990). The managerial perspective on empowerment has a long history and offers a set "levers" managers can pull to improve performance. The insight provided by the psychological perspective, however, should not be overlooked. As empowered employees have a higher expectancy in their ability to perform a task successfully, they exert greater effort and persist in those efforts when faced with adversity. A sense of autonomy at work, along with the feeling of having control over the outcome, also increases effort.
In addition to exerting greater effort or "working harder," empowered employees also seem to improve their performance by working "smarter" or by seeking out new and better ways of doing things. Thomas and Velthouse (1990) alluded to this when they explained that intrinsically motivated individuals "may demonstrate flexibility in controlling their own task accomplishment, [and] initiation of new tasks as problems or opportunities arise" (673; see also Kanter 1983). Bowen and Lawler noticed two forms of innovative behavior resulting in performance gains: when frontline employees take rapid and spontaneous steps to "recover" from poor service delivery and adapt services to meet the idiosyncratic needs of customers, and when they move beyond reactive recovery to proactively redesigning processes and systems and creating new products and services. The Clinton Administration's National Performance Review (NPR) identified employee empowerment as one of the keys to making government more efficient and effective. Frontline employees were argued to be the source of many innovative solutions to problems since they are closest to those problems and more knowledgeable about how to solve them. Reformers expected improved performance to come from "turning the entire management system upside down" by empowering frontline employees to exercise their judgment, giving them training and resources needed to get the job done, and holding them accountable for results (Gore 1993, 6). (2)
EMPOWERMENT PRACTICES AND ENCOURAGEMENT TO INNOVATE
In a dynamic external environment, adaptable organizations are more likely to survive and thrive than mechanistic ones (Burns and Stalker 1961). Adaptability often takes the form of innovation, the invention or adoption of processes, practices, products, and services by an organization (Becker and Whisler 1967; March and Simon 1993; Simon 1997). Product and technological innovations continue to be key sources of performance improvement and competitive advantage for private sector firms (Christensen 1997; Fagerberg, Mowery, and Nelson 2006; Porter 1985). NPM reforms in the United States, Australia, the United Kingdom, and elsewhere have stressed innovation as a way to improve public sector performance (Bartos 2003; Breul and Kamensky 2008; Gore 1993; Kamensky 1996; Kettl 2005; Pollitt and Bouckaert 2004). Even though the evidence to date is not abundant, a growing number of studies suggest innovation is positively linked to performance in government (Altshuler and Behn 1997; Borins 2008; Damanpour, Walker, and Avellaneda 2009; Light 1998; Walker and Damampour 2008). This positive relationship can be mediated by forms of organization capacity, such as a performance management orientation (Walker, Damampour, and Devece 2011). The benefits of innovation are not always realized, however, as many new ideas fail during their implementation (Hartley 2005). Moreover, organizational change can be disruptive, adversely affecting performance to the point of organizational decline and death (Amburgey, Kelly, and Barnett 1993). The impact of innovation on performance thus appears to be generally positive in the long-term but marginal or even negative in the short-term until new processes are learned and institutionalized.
Research on innovation in the public sector has shown that while elected officials and political appointees are the source of many innovations (Breaux et al. 2002; Chakerian and Mavima 2000; Kellough and Nigro 2002; Wallin 1997), so are frontline employees who generate novel ideas through experimentation, accidental occurrences, and other forms of experience (Altshuler and Zegans 1997; Borins 2000a, 2000b, 2012; Kamensky 1996; Light 1998; Thompson and Sanders 1997). Many of the innovations arising out of the NPR originated from the experiences of practitioners (Kamensky 1996). Reinvention laboratories were set up in many federal agencies to give employees dispensation to modify, streamline, and reinvent processes and structures in their agencies. Altshuler and Zegans (1997), in their review of award-winning innovations in American government, found that public servants who initiated the innovation were more likely to be street-level bureaucrats in direct contact with clients than senior managers. Similarly, Borins (2000a, 2000b, 2012) has found that career civil servants at the middle manager and frontline levels are frequent initiators of innovation in government.
If frontline employees are an important source of innovative ideas, how can empowerment be used to encourage them to innovate? It is important to keep in mind that according to Bowen and Lawler, employee empowerment is a multifaceted management approach involving four practices: providing information about goals and performance (e.g., goal setting, performance measurement, feedback); offering rewards based on performance (e.g., merit-based pay, profit sharing, employee recognition); providing access to job-related knowledge and skills (e.g., training and job-embedded learning to acquire technical, problem solving, interpersonal skills); and granting discretion to change work processes (e.g., creative rule breaking, employee participation, self-managed teams, job enrichment). It is essential, therefore, to understand how each of these practices influences feelings of encouragement to innovate.
Providing Information about Goals and Performance
Communicating goals and priorities to employees and offering feedback on performance are practices that have been found to encourage innovation. Specific and challenging goals in general serve to raise employee motivation and performance (Locke and Latham 1990). Top-down communication that conveys the leadership's priorities and goals can, therefore, encourage achievement-oriented employees to seek new strategies and tactics for attaining those goals. Negative feedback indicative of failure also signals the need to search for new ways of narrowing the performance gap (Cyert and March 1963; Fernandez and Wise 2010; Manns and March 1978; Salge 2011), thereby encouraging employees to innovate. While goal ambiguity in the public sector can undercut the effectiveness of goal setting as a motivational approach (Rainey 2009), at the level of the work team or the individual employee, goals are often sufficiently clear for this empowerment practice to have a positive impact on the extent to which an employee feels encouraged to innovate. This leads to the first hypothesis:
Hypothesis 1 The practice of providing employees with information about goals and performance will be positively related with encouragement to innovate.
Offering Rewards Based on Performance
Intrinsic job characteristics have been found to have a stronger impact on employee attitudes than extrinsic ones (Deci 1971; Mottaz 1985; O'Reilly and Caldwell 1980). A large body of research shows, however, that pay and other extrinsic rewards can still be used effectively to increase effort, performance, and job satisfaction (Greene and Haywood 2008; Lawler, Mohrman, and Ledford, 1992, 1995; Mottaz 1985; O'Reilly and Caldwell 1980). Even among public employees with higher levels of public service motivation, monetary rewards appear to be significantly valued (Alonso and Lewis 2001; Perry, Mesch, and Paarlberg 2006; Wittmer 1991; Wright 2007). Evidence suggests pay-for-performance in the public sector is often only marginally effective when it comes to improving performance (Kellough and Lu 1993; Perry, Engbers, and Jun 2009). This is due at least in part, however, to flawed implementation. (3)
Based on the research cited above, it is reasonable to conclude that empowerment practices aimed at offering financial and other rewards for performance will cause employees to feel more encouraged to innovate. There is reason to believe, however, that tying rewards to performance may discourage innovativeness in government agencies. Change can cause significant turbulence within organizations (Amburgey, Kelly, and Barnett 1993; Fernandez and Rainey 2006). This could incentivize employees to stick with routine ways of doing things and avoid disruptive innovations that might pay off only in the long term. In addition, self-determination theory indicates introducing extrinsic rewards for work that was intrinsically motivating may actually reduce one's motivation and effort to complete a task (Deci 1971; Ryan and Deci 2000). In light of the divergent views expressed above, the second hypothesis is proposed:
Hypothesis 2 The practice of offering employees rewards based on performance will be related with encouragement to innovate, but the direction of the relationship could be either positive or negative.
Providing Access to Job-Related Knowledge and Skills
Efforts to enhance employees' access to job-related knowledge and skills through training and job-embedded learning have been linked to receptivity to new ideas and creativity. Training and professional development activities help to diffuse innovations, as employees learn about and introduce ideas applied successfully in other organizations. They also expose employees to a broader palette of ideas that can be brought to bear on new problems (Damanpour 1991; Katz and Tushman 1981; Thompson 1965). Because training and development improves an employee's ability to diagnose and solve technical problems, the odds are increased that innovative proposals will become effective practices (Dewar and Dutton 1986; McGinnis and Ackelsberg 1983). Enhanced knowledge of alternatives for improving performance and greater confidence in the efficacy of those alternatives should cause employees to feel more encouraged to innovate. Thus, the third hypothesis:
Hypothesis 3 The practice of providing employees with access to job related knowledge and skills will be positively related with encouragement to innovate.
Granting Discretion to Change Work Processes
The relationship between practices aimed at sharing power with employees and innovation is one that is well-established in the innovation literature. There are various ways in which granting discretion to employees can increase innovativeness. By loosening controls, managers give entrepreneurial employees freedom to tinker with existing elements and practices and reconfigure them in new ways (Kanter 1983; Levin and Senger 1994). Pushing authority downward can also encourage employees to innovate by imparting a sense of control and responsibility for the quality of their work (Hackman and Oldham 1976). Finally, being granted the authority to modify work processes may increase encouragement to innovate by raising one's level of confidence that he or she will not be called out or punished for failed innovations (Edmonson 1999; Light 1998). Importantly, public agencies have been found to have higher levels of formalization than private firms (Rainey and Bozeman 2000). The profusion of rules and regulations in government might prevent public managers from granting enough discretion to achieve more than just trivial changes in the way work is structured and performed. A highly formalized work setting can also undermine psychological safety and diminish the extent to which employees feel encouraged to innovate. Notwithstanding these constraints typical of the public sector, the fourth hypothesis is proposed:
Hypothesis 4 The practice of granting employees discretion to change work processes will be positively related with encouragement to innovate.
Finally, since employee empowerment is described by Bowen and Lawler as a multifaceted approach to service delivery that comprises the four practices describe above, it is important to also analyze an employee empowerment approach's overall relationship to encouragement to innovate. This leads to the final hypothesis:
Hypothesis 5 An employee empowerment approach will be positively related with encouragement to innovate.
DATA AND METHODS
This section provides a description of the data and research setting, the variables used in the analysis, and the regression model selection and estimation process.
Data and Research Setting
The data for the analysis are from the 2006 Federal Human Capital Survey (FHCS) conducted by the US Office of Personnel Management (OPM). The 2006 FHCS was administered electronically via the Internet (with limited distribution of paper surveys to those without Internet access) to 390,657 federal government employees at five levels ranging from nonsupervisory to Senior Executive Service employees. The government-wide response rate was 57% (N = 221,479). (4) Respondents worked for 82 cabinet-level and smaller independent agencies representing 97% of the executive branch workforce. OPM used a stratified sampling approach to produce generalizable results for each individual agency as well as the entire federal government; in some of the smaller agencies, all employees were surveyed. The stratified sampling approach also helped to ensure that respondents from various hierarchical levels were represented adequately (see Enticott, Boyne, and Walker 2009).
Out of the 221,479 respondents to the 2006 FHCS, 66,686 have missing data on the outcome, explanatory, and/or control variables. No meaningful differences between observations with and without missing data were found. The concept of employee empowerment applies particularly to frontline and lower-level employees. The focus of the analysis, therefore, is on those employees at the three lowest levels of the federal bureaucracy: nonsupervisory employees, team leaders, and supervisors; managers and senior executives are excluded from the analysis. (5)
The dependent variable, encouragement to innovate, is defined as an affective state or experience of feeling felt by public employees that makes them inclined to innovate. It represents only one component--the emotion or affect component--of the complex motivational process involving needs, values, motives, emotional appraisals, and behavioral responses to them (see Locke and Latham 2004). As a result, it should not be confused with motivation to innovate or actual innovative behavior. The dependent variable is measured using responses to the following ordinal survey item: "I feel encouraged to come up with new and better ways of doing things." The response categories ranged from 1 = "strongly disagree" to 5 = "strongly agree." Approximately 5% of respondents answered "strongly disagree"; 13% answered "disagree;" 19% answered "neither agree nor disagree"; 40% answered "agree"; and 22% answered "strongly agree." This distribution indicates sufficient variance and representation of each of the five response categories to allow reasonable estimations and tests of the hypotheses offered above.
Measuring the dependent variable using a single survey item is one of this study's limitations. Single-item measures of a construct can be just as valid as multi-items ones, particularly in terms of predictive validity, when the construct consists of a concrete singular object and a concrete attribute of that object (e.g., consumer appeal of a brand or product) (Bergkvist and Rossiter 2007). Encouragement to innovate appears to be a more complex psychological construct warranting the use of a multiple-item measure. Single-item measures of psychological constructs like job satisfaction, however, have been found to correlate at about 0.70 with multi-item measures (Wanous, Reichers, and Hudy 1997). Thus, whereas a multi-item measure would be preferable, the single-item measure of encouragement to innovate is suitable for the purpose of this study.
Survey indicators were used to construct standardized summated rating scales for practice 1, providing information about goals and performance; practice 2, offering rewards based on performance; practice 3, providing access to job related knowledge and skills; and practice 4, granting discretion to change work processes. As can be seen in Appendix 1, the survey items selected for each of the scales exhibit face validity and appear to be measuring the type of managerial behavior described by Bowen and Lawler. Cronbach's alpha tests show moderate to high levels of internal consistency, with scale reliability statistics ranging from 0.74 for practice 3 to 0.88 for practice 2.
A major issue that comes forward is whether these measures of empowerment stand the test of discriminant and convergent validity. The four empowerment variables were subjected to a higher order confirmatory factor analysis (CFA) performed using the R software package (see Appendix 4). A variety of goodness-of-fit indices (GFIs) from the CFA support a four-factor model of employee empowerment, whereas rejecting a single-factor model. (6) Tests using the average variance explained (AVE) statistic offer evidence of both convergent and discriminant validity in the higher order CFA. (7) A second CFA was performed without the higher order factor (i.e., the empowerment construct) using the R software package (see Appendix 5). These additional results are very similar to those from the higher order CFA and support a four-factor model over the single-factor model. (8) Tests using the AVE statistic for this second CFA continue to offer evidence of discriminant validity. (9)
The regression model used to test the hypotheses includes control variables for factors that influence innovativeness. The perception that innovative behavior is rewarded should be a strong motivator of such behavior. The model thus includes the variable rewards for innovation, which is measured using an indicator of the extent to which the respondent feels creativity and innovation are rewarded. The actual survey items used to measure this and other control variables are shown in Appendix 2. This control variable and the independent variable practice 2 are correlated at the r = .80 level (see Appendix 3). Despite the high correlation and overlap between the two variables (both represent efforts to reward employees), there are conceptual and empirical reasons for treating them as distinct variables. First, these practices differ in the performance criteria used as a basis for rewards, with the former focusing on the production of outputs and outcomes and the latter on proposed or actual changes in processes. (10) They also differ in terms of temporal perspective, with the former based on annual performance as determined by an employee performance appraisal and the latter on projections of impact or historical data going back much further than the previous year. These differences help to explain why many public organizations have developed systems for rewarding innovation that operate independently of the annual performance appraisal and reward system, with different staff typically assigned to them. (11) Moreover, data from the 2006 FHCS also offer evidence in support of treating practice 2 and rewards for innovation as related and overlapping but distinct variables. Adding the indicator used to measure rewards for innovation as a fifth observable indicator for practice 2 produces worse CFA model fit statistics compared with the original CFA model. (12)
Satisfied employees are generally more committed to the organization and thus more likely to look for ways to improve performance (Hage and Aiken 1967; Thompson 1965). As a result, the variable job satisfaction, a global measure of employee job satisfaction, is included in the model. Poor or substandard performance has been found to induce search behavior among organizational members (Cyert and March 1963; March and Simon 1993; Salge 2011). The effect of perceptions of performance on innovativeness is controlled for by including the variable perceived performance in the model. Successfully cultivating innovative ideas often requires a commitment of financial and material resources to launch and sustain an innovation (Berry 1994; Cyert and March 1963; Fernandez and Wise 2010). When resources are scarce, managers are less likely to commit them toward anything other than ongoing operations, thereby discouraging frontline employees from innovating. The variable sufficient resources is included in the model, therefore, to control for the effects of perceived adequacy of resources on encouragement to innovate.
In addition to vertical or downward communication captured in part by practice 1, horizontal communication and exchange of information among employees has been found to be a predictor of innovativeness (Kanter 1982; Monge, Cozzens, and Contractor 1992; Tjosvold and McNeely 1988). To control for this effect, the variable knowledge sharing is included in the model. High exchange relationships between superiors and subordinates characterized by high levels of trust have been linked to higher subordinate satisfaction, stronger organizational commitment by the subordinate, and higher subordinate performance (see Bass 1990). Since the extent to which the respondent has trust and confidence in his/her supervisor could encourage innovative behavior, the variable trust in leader is included in the model. Finally, the model includes a set of controls for demographic characteristics of the survey respondent, including a dummy variable for whether or not the respondent works in a field office (location), a dummy variable for whether or not the respondent is nonwhite (minority), and an ordinal variable for the respondent's age (age).
Regression Model Selection and Estimation
The variable encouragement to innovate is a limited dependent variable that is measured using an ordinal survey indicator. Since using ordinary least squares regression to estimate this type of dependent variable can result in biased coefficients and misleading results (Long 1997; Long and Freese 2005; McKelvey and Zavoina 1975; Winship and Mare 1984), the preferred estimation approach is ordered logit model (OLM) or ordered probit model (OPM) regression. OLM and OPM regression are based on the rationale of proportional odds or parallel regression equations. If the ordered models violate this assumption, then a higher order specification such as the multinomial logit model (MNLM) is preferred. A variety of tests show that the OLM estimation violates the fundamental assumption of parallel regression. The Brant-test of parallel regression assumption suggests that the four empowerment practices taken together violate the assumption of parallel regression. When taken individually, none of the empowerment practices passes the test, failing the Brant-test quite significantly (p < .001). None of the control variables in the OLM model pass the parallel regression test either. The overall model fails the test as well, indicating a multinomial specification should be used. Further, in order to relax the parallel regression assumption employed by a traditional ordered logit model, an omnibus likelihood ratio test for generalized ordered logit constrained and unconstrained models is run. The likelihood ratio test for proportional odds shows that the model fails the omnibus test, too [[chi].sub.LR;df=392]=3637.83; p < .001). These results point to the MNLM as the preferred model specification.
In the MNLM, the nonlinear probability of an outcome to occur, that is, y = m given x is:
Pr(y=m[x.sub.i]=exp([x.sub.i][[beta].sub.m]/[[summation].sup.j.sub.j=1]exp([x.sub.i][[beta].sub.j]) where [[beta].sub.A]=0,
and where, x[beta] represents the model equation. The dependent variable is encouragement to innovate.
The results of the empirical analysis are presented in this section. Descriptive statistics appear in table 1. The results of the hypotheses tests are discussed first, followed by the results of various post hoc analyses.
The results of the MNLM regressions used to test the five hypotheses are reported in tables 2 and 3. They show that three of the four empowerment practices-practice 2, practice 3, and practice 4--are statistically correlated with encouragement to innovate (p < .001), whereas practice 1 fails to achieve statistical significance (p = .478). (13) For the purpose of interpreting the MNLM results, it should be noted that the comparison response category is "agree." (14) Importantly, for practice 2, practice 3, and practice 4, the substantive magnitude of the relationships are sizeable. This is illustrated graphically in figures 14, which plot the predicted levels of the dependent variable across the range of the empowerment practices, with all other variables held constant at their mean values (also see table 3).
Figure 1 plots the levels of encouragement to innovate across practice 1, providing information about goals and performance. According to Hypothesis 1, practice 1 should be positively correlated with the dependent variable. The MNLM regression results show no apparent relationship between this empowerment practice and the dependent variable, thus rejecting the first hypothesis. All the lines representing the predicted probabilities for the five response categories of the dependent variable are generally flat across the range of practice 1. Even when a slight slope is observed (e.g., strongly agree), the magnitude of the effect is close to nil.
Hypothesis 2 states that practice 2, offering rewards based on performance, will be correlated with encouragement to innovate, but the direction of the correlation could be either positive or negative. The MNLM regression results support this hypothesis, as they indicate practice 2 is negatively correlated with the dependent variable. Figure 2 plots the predicted probabilities of the responses categories for the dependent variable across this empowerment practice. The predicted probabilities of categories "disagree" and "neither agree nor disagree" increase significantly by about 10% (from 3% to 13%) and 15% (from 19% to 34%), respectively, when moving across the range of practice 2. Conversely, the predicted probabilities of categories "agree" and "strongly agree" both decrease by about 13% across the range of practice 2. The predicted probabilities of category "strongly disagree" remain fairly steady.
The MNLM regression results support Hypothesis 3, which states that practice 3, providing access to job-related knowledge and skills, will be positively correlated with encouragement to innovate. From among the four empowerment practices, the relationship between practice 3 and the dependent variable is the largest in substantive magnitude. Figure 3 plots the predicted probabilities of the response categories for the dependent variable across practice 3. The predicted probabilities of categories "strongly agree" and "agree" increase sharply, but not linearly, by 0.44 and 0.25, respectively, when moving across the range of this empowerment practice. On the other hand, the predicted probabilities of categories "neither agree nor disagree" and "disagree" decrease sharply by more than 0.30 when moving across the range of practice 3. The line representing the predicted probabilities of response category "strongly disagree" shows only a slight negative slope.
The MNLM regression results show that practice 4, granting discretion to change work processes, is positively correlated with encouragement to innovate, thus supporting Hypothesis 4. The predicted probabilities of the dependent variable across this empowerment practice are depicted in figure 4, with the results resembling those for practice 3. The predicted probabilities of categories "strongly agree" and "agree" increase sharply by 0.20 and 0.35, respectively, when moving across the range of this empowerment practice. Conversely, the figure shows the predicted probabilities of categories "neither agree nor disagree" and "disagree" decline by about 0.25 across the range of practice 4. Predicted probabilities for category "strongly disagree" remain steady.
[FIGURE 1 OMITTED]
The substantive magnitude of the relationships has so far been interpreted in terms of changes in the predicted probabilities of the response categories. Another useful approach involves interpreting the odds ratio coefficients of the empowerment variables (see table 4). This odds ratios method gives a better sense of the magnitude of the relationship regardless of the location on a scale across the range of an independent variable. The relationship between practice 2 and encouragement to innovate is negative and of a substantively significant size. A unit increase in this empowerment practice increases the odds of response category "disagree" versus category "agree" occurring by a factor of 1.53, with all other variables held constant (p < .001). In a similar manner, a one-unit increase in practice 2 increases the odds of category "neither agree nor disagree" versus" agree" occurring by a factor of 1.26, all else held equal (p < .001). Conversely, for every one-unit increase in this empowerment practice, the odds of category "strongly agree" versus category "agree" occurring decrease by a factor of 0.70 (a difference of roughly 43%), all else equal (p < .001).
Both practice 3 and practice 4 are positively related to encouragement to innovate, with the magnitude of the relationships being substantively significant. A one-unit increase in practice 3 increases the odds of category "strongly agree" versus "agree" occurring by a factor of 4.52, all else equal (p < .001). Conversely, a unit increase in this empowerment practice decreases the odds of categories "neither agree nor disagree," "disagree," and "strongly disagree" versus category "agree" occurring by factors of 0.55, 0.36, and 0.19 (decreases of about 82%, 178%, and 438%), respectively, all other variables held constant (p < .001). A similar pattern in found with practice 4. A one-unit increase in this last empowerment practice increases the odds of category "strongly agree" versus category "agree" occurring by a factor of 1.78, all else equal (p < .001). Conversely, for every unit increase in practice 4, the odds of categories "neither agree nor disagree" and "disagree" versus category "agree" occurring decrease by factors of 0.57 and 0.32 (decreases of 75% and 213%), respectively, all else held constant (p < .001).
[FIGURE 2 OMITTED]
Although the focus of the analysis has so far been on the relationship between individual empowerment practices and encouragement to innovate, the relationship between empowerment as an overall approach and the dependent variable is also explored. Hypothesis 5 states that an employee empowerment approach will be positively related with encouragement to innovate. Model 2 in table 2 shows the results of a MNLM regression in which the primary independent variable is the overall factor score developed from all the indicators used to measure the four empowerment practices. The results indicate a sizeable, positive relationship between this overall empowerment measure and encouragement to innovate (-2.24, p < .001), thus providing support for the fifth and last hypothesis. This new model and the one with variables for the four empowerment practices are nearly identical in terms of model fit statistics, coefficients, and levels of statistical significance.
Shifting the focus now to the control variables, table 2 indicates that all but one control variable, sufficient resources, is statistically correlated with encouragement to innovate. However, only rewards for innovation, trust in leader, and job satisfaction seem to have a substantively significant relationship with the dependent variable (see also tables 3 and 4). (15) The variable rewards for innovation is positively correlated with the dependent variable, with the magnitude of the relationship rivaling that of practice 3 and practice 4. For every unit increase in this variable, the odds of "strongly disagree," "disagree," and "neither agree nor disagree" occurring compared with "agree" decrease by factors of0.19, 0.40, and 0.66 (differences of 438%, 150%, and 52%), respectively, all else equal (p < .001). Conversely, for every one-unit increase in rewards for innovation, the odds of category "strongly agree" versus "agree" occurring increase by a factor of 1.49, all else equal (p < .001).
[FIGURE 3 OMITTED]
The variable trust in leader is positively correlated with the dependent variable. A unit increase in this control variable decreases the odds of "disagree" and "neither agree nor disagree" occurring compared with "agree" by factors of 0.70 and 0.84 (differences of 43% and 19%), respectively, all else constant (p < .001). Conversely, a unit increase in trust in leader increases the odds of "strongly agree" versus "agree" occurring by a factor of 1.49, all else equal (p < .001). The control variable job satisfaction is also positively correlated with motivation to innovate. For every one-unit increase in this variable, the odds of "disagree" and "neither agree nor disagree" occurring compared with "agree" decrease by 0.87 and 0.90 (differences of roughly 15% and 11%), respectively, all else constant (p < .001). Alternatively, for every one-unit increase in job satisfaction, the odds of "strongly agree" versus "agree" occurring increase by a factor of 1.29, all else equal (p < .001).
[FIGURE 4 OMITTED]
Post hoc Analyses
The analysis above is based on a sample of federal employees at three levels: non-supervisory employees, team leaders, and supervisors. Separate MNLM regression analyses were performed for each of these groups of employees to determine if the results varied by level. The additional results were remarkably similar to those reported above, with one minor exception. The variable practice 1, providing information about goals and performance, which fails to achieve statistical significance in the multinomial regression using the larger sample, achieves statistical significance in the sample of just those employees with supervisory positions. The relationship, however, is close to zero in terms of substantive significance.
Conceptual as well as empirical reasons were offered for treating practice 2, offering rewards based on performance, and rewards for innovation as separate variables capturing overlapping but distinct practices. The high correlation between these two variables, however, raises the question of what occurs when organizations combine the practice of rewarding outputs and outcomes with rewarding innovative changes in processes. To answer this question, a new variable was created by combining the four survey indicators used to measure practice 2 with the single indicator used to measure rewards for innovation into a standardized summated rating scale and including this new variable in the MNLM. The new results (not shown) indicate this new variable is strongly, positively correlated with encouragement to innovate, with the other coefficients virtually unchanged compared with those reported in tables 2 and 3. (16) This suggests innovation can be encouraged further by integrating systems for rewarding innovation with annual performance appraisal and reward systems. How to accomplish this remains an ongoing challenge, however, given the difficulty of establishing goals for innovative behavior and of providing employees with sufficient slack to innovate without adversely affecting performance.
The high correlation between practice 2 and rewards for innovation as well as between some of the empowerment practices themselves raises concern about a multicollinearity problem with the MNLM regression analysis. A variety of diagnostic tests were performed to assess the presence of such a problem. The variance inflation factor (VIF), tolerance score, and conditioning number were estimated for each of the variables in the MNLM. VIFs greater than 10 (or more conservatively greater than 5), tolerance scores below 0.10, and conditioning numbers greater than 20 are indicative of a multicollinearity problem (Baum 2006; Gujarati and Porter 2010). The results of these diagnostic tests fail to detect a multicollinearity problem, as none of the VIFs are greater than 3.5, all tolerance scores are at or above 0.30, and the largest conditioning number is 5.86. To explore the issue of multicollinearity further, the sample used in the analysis was split into three subsamples of about 50,000. The MNLM regression results are consistent across the three subsamples and the full sample, thus failing to provide evidence of a multicollinearity problem. As a final check on multicollinearity, a series of MNLM regressions were estimated in which one or more of the empowerment practices were withdrawn. These additional MNLM regression results show no meaningful differences in statistical or substantive significance for the coefficients. In short, none of the diagnostic tests offered evidence indicative of a multicollinearity problem causing unstable and biased regression estimates.
Finally, the analysis included a test for interaction effects among the empowerment variables. In a full-order interaction MNLM regression, the coefficients for the four-way interaction involving all four empowerment practices and for all possible three-way interaction combinations failed to achieve statistical significance at the p < .05 level (results not shown). A MNLM regression was also tested that included all possible two-way interaction combinations. Adding those interaction terms had only a marginal effect on the fit statistics (the [R.sup.2] increased by about .02), but the model's coefficients reveal some interesting results (see table 5). The interactions between practice 1, providing information about goals and performance, and practice 2, offering rewards based on performance, as well between practice 2 and practice 4, granting discretion to change work processes, produce very small changes in the response categories of the dependent variable and appear to be trivial in their substantive significance.
Evidence is found that providing employees with greater opportunities to learn and develop enhances the effectiveness of all the other empowerment practices when it comes to encouraging innovation. For the three two-way interactions involving practice 3, providing access to job-related knowledge and skills, an increase in the value of practice 3 increases the probability of response category "strongly agree" occurring compared with most other response categories as the values of practice 1, practice 2, and practice 4 increase. This suggests employees are more encouraged to innovate when they have considerable knowledge and skills and ample discretion compared with when they have just one or the other. Setting goals and providing performance feedback seem to encourage innovation only when employees enjoy significant opportunities to learn and develop, perhaps because they are better equipped to achieve those goals and respond to feedback. It even appears that the practice of offering rewards based on performance, practice 2, can have a marginally positive effect on the dependent variable, but only when used together with practice 3.
Evidence is also found of a two-way interaction between practice 1, providing information about goals and performance, and practice 4, granting discretion to change work processes. Increasing the value of practice 1 increases the probability of response categories "strongly agree" and "agree" occurring compared to "neither" as the value of practice 4 increases. This suggests that setting goals and providing feedback may be used to encourage innovative behavior only when employees feel they are involved in decisions affecting how they structure and perform their work. By itself, however, this empowerment practice appears to be ineffective.
DISCUSSION AND CONCLUSION
Employee empowerment has gained widespread popularity as a high performance management approach that enables organizations to remain competitive and innovative. Although it was first widely adopted in the private sector, empowerment has become part and parcel of major government reforms around the world. This study explored the effectiveness of various empowerment practices at encouraging innovative behavior among frontline employees in the US federal government. The empirical results show that not all empowerment practices encourage innovative behavior and that some even appear to discourage innovation. They also show that as a multifaceted managerial approach, empowerment increases encouragement to innovate.
Empowerment practices aimed at granting employees discretion to change work processes and at providing them with opportunities to acquire job-related knowledge and skills are strongly, positively correlated with employee encouragement to innovate. These results confirm a consistent pattern reported in the innovation literature: organizations that grant employees ample opportunities to exercise discretion and to learn and grow tend to be more innovative than others. It is also found that attempting to empower employees by offering rewards based on performance, when performance is defined in terms of outputs and outcomes, inhibits innovativeness. Rewarding short-term performance, as pay-for-performance schemes often do, seems to foster a myopic mindset among employees, causing them to settle for proven ways of doing things while eschewing disruptive changes that might produce only long-term gains. Interestingly, rewarding innovative changes in processes rather than outputs and outcomes seems to encourage innovation. The data used in the analysis prevent one from identifying the different intrinsic and extrinsic rewards conferred on innovative employees, thus creating the need for further research.
The practice of empowering frontline employees by providing information about goals and performance is ineffective when used alone, but in combination with other empowerment practices may produce small gains in innovativeness. Goal setting and performance feedback are often used in place of rules to coordinate and control behavior in organizations (Mintzberg 1979). Efforts to control employees this way appear to discourage them from seeking out and trying innovative solutions to problems. They may even cause resentment among employees who feel overburdened by reporting requirements and doubt the quality and usefulness of feedback (Tosi and Carroll 1968). In addition, the effect of goal setting on learning appears to vary according to one's level of cognitive ability, with those with low cognitive ability benefiting more from goal setting than others (Kanfer and Ackerman 1989). Research also suggests that setting clear and challenging goals is more effective at motivating employees facing simple, programmable tasks than those performing more complex tasks where learning and trial and error are at a premium (Winters and Latham 1996). In short, when it comes to encouraging innovative behavior, the benefits of providing information about goals and performance are contingent on the situation and may be offset by the costs.
The relationships between different empowerment practices and encouragement to innovate appear to be largely independent of each other. Some evidence was found of interactions between two-way combinations of empowerment practices. For example, providing employees with greater opportunities to learn and develop can slightly enhance the effectiveness of each of the other empowerment practices. The magnitude of these interaction effects, however, is quite small.
One limitation of the study is that the dependent variable only captures inclination to innovate. It is not possible to discern with any degree of certainty whether or not such inclination will translate into actual innovative proposals, whether or not those proposals are accepted, or the possible impact of those innovations on performance. Additional longitudinal research is needed to analyze the effects of encouragement to innovate on the frequency of innovations and on the consequences of those innovations (e.g., see Walker, Damanpour, and Devece 2011).
The use of self-reported data from a single survey raises the specter of common method bias. Common method bias is generally believed to produce artificially inflated correlations (Crampton and Wagner 1994), although in some cases the bias can also deflate correlations (Cote and Buckley 1988; Podsakoff et al. 2003). Two approaches were taken to detecting common method bias. First, a Harman single factor test of all the survey items in the 2006 FHCS produced a multiple factor solution. Second, it should be recalled that the CFA supported a four-factor model, and importantly, that a one-factor model failed every single test of goodness of fit. These results, while not refuting the presence of common method bias, fail to produce convincing evidence of its presence. If indeed present, common method bias could conceivably inflate the fairly strong correlations between the empowerment practices and the dependent variable. Some care should be taken, therefore, in interpreting the results of this analysis.
The CFA results, while providing evidence of discriminant and convergent validity, suggest the need for further refinement in measurement. Nearly all of the fit statistics for the four-dimensional model of empowerment were found to be at acceptable levels, but many hovered around conventional cutoff points. In particular, the measurement of practice 4, granting discretion to change work processes, could be improved with additional indicators measuring specific ways in which discretion is used. The four-dimensional definition of empowerment used in this study needs further validation across other large samples of public sector employees. Moreover, our understanding of these empowerment practices and how they encourage innovation could be explored further using in-depth case studies based on multiple data sources and methods of data collection.
The link between empowerment practices and encouragement to innovate is an important causal path by which empowerment can improve performance but certainly not the only possible one. Bowen and Lawler's empowerment practices could very well influence performance through increased job satisfaction, another employee attitude associated with increased effort and productivity (Judge and Church 2000). Although previous studies have shown empowerment can increase job satisfaction in the public sector, they have not ventured beyond to explore the relationship between job satisfaction and performance. The use of structural equation modeling techniques could prove fruitful in describing the complex pattern of causal relationships among the various managerial practices, cognitions, and behavioral outcomes involved in the empowerment process. Unfortunately, such analyses remain beyond the scope of this study.
The large sample size used in the study and its widespread coverage of federal agencies bode well for the generalizability of the results across large portions of the federal bureaucracy, particularly at the frontline and nonsupervisory echelons. There is reason to believe, however, that a somewhat different pattern of results might emerge from analyses conducted at the state or local level of government. Previous research on empowerment suggests empowerment practices work best in service delivery organizations where frontline employees are in direct contact with clients (Bowen and Lawler 1992, 1995; Potterfield 1999). The fact that much larger proportions of state and local government employees are involved in direct service delivery could mean that empowerment practices are even more effective at encouraging innovative behavior at those levels of government than what is found in this study.
Finally, normative and empirical questions remain about the consequences of empowering frontline public employees in a democratic system of government. How will that discretion be used? One possibility is that extensive use of empowerment breeds reckless rule breaking among unelected public employees. Although previous studies have shown that public managers behave responsibly when engaging in innovative behavior (Berman and West 1998; Borins 2000a) and seek political approval from elected officials when launching new initiatives (Golden 1990), this issue is worthy of further investigation.
The authors received no external funding to conduct this research.
APPENDIX 1 Measures, Employee Empowerment Practices
Practice 1 (providing information about goals and performance)
I1. Managers review and evaluate the organization's progress toward meeting its goals and objectives (1 = strongly disagree through 5 = strongly agree).
I2. Supervisors/team leaders provide employees with constructive suggestions to improve their job performance (1 = strongly disagree through 5 = strongly agree)
I3. How satisfied are you with the information you receive from management on what's going on in your organization? (1 = very dissatisfied through 5 = very satisfied)
Cronbach's alpha test, mean interval covariance = 0.62
Cronbach's alpha test, scale reliability coefficient = 0.80
Practice 2 (offering rewards based on performance)
I4. Promotions in my work unit are based on merit. (1 = strongly disagree through 5 = strongly agree)
I5. Employees are rewarded for providing high-quality products and services to customers (1 = strongly disagree through 5 = strongly agree)
I6. Pay raises depend on how well employees perform their jobs (1 = strongly disagree through 5 = strongly agree)
17. Awards in my work unit depend on how well employees perform their jobs (1 = strongly disagree through 5 = strongly agree)
Cronbach's alpha test, mean interval covariance = 0.87
Cronbach's alpha test, scale reliability coefficient = 0.88
Practice 3 (providing access to job-related knowledge and skills)
I8. I am given a real opportunity to improve my skills in my organization (1 = strongly disagree through 5 = strongly agree)
I9. The workforce has the job-relevant knowledge and skills necessary to accomplish organizational goals (1 = strongly disagree through 5 = strongly agree)
I10. Supervisors/team leaders in my work unit support employee development (1 = strongly disagree through 5 = strongly agree)
Cronbach's alpha test, mean interval covariance = 0.49
Cronbach's alpha test, scale reliability coefficient = 0.74
Practice 4 (granting discretion to change work processes)
I11. Employees have a feeling of personal empowerment with respect to work processes (1 = strongly disagree through 5 = strongly agree).
I12. How satisfied are you with your involvement in decisions that affect your work? (1 = very dissatisfied through 5 = very satisfied).
Cronbach's alpha test, mean interval covariance = 0.74
Cronbach's alpha test, scale reliability coefficient = 0.77
Note: Mean internal covariance, also known as average inter-item correlation, is a statistic used to assess the reliability of a scale. Conventionally, if this statistic is greater than .6, then item standardization and index construction is justified (Nagel and Garson 1996), although when theory justifies it, lower scores of this measure may be selected if the scale reliability coefficient is greater than. 7 (Nunnally and Bernstein 1994). It is also a common practice to view a scale reliability coefficient greater than .7 as an indicator that the scale index is reliable.
Measures, Control Variables
Rewards for innovation
"Creativity and innovations are rewarded" (1 = strongly disagree through 5 = strongly agree).
"Considering everything how satisfied are you with your job?" (1 = very satisfied through 5 = very dissatisfied).
"How would you rate the overall quality of work done by your work group?" (1 = very poor through 5 = very good).
"Employees share job knowledge with each other." (1 = strongly disagree through 5 = strongly agree)
Trust in leader
"I have trust and confidence in my supervisor" (1 = strongly disagree through 5 = strongly agree).
"I have sufficient resources to get my job done" (1 = strongly disagree through 5 = strongly agree).
Respondent's work location (1 = field office, 0 = headquarters).
Respondent's race (1 = nonwhite, 0 = white).
Respondent's age (1 = 26-29; 2 = 30-39; 3 = 40-49; 4 = 50-59; 5 = 60-69).
APPENDIX 3 Correlation Matrix 1 2 3 4 5 1. Encouragement to innovate 1.00 2. Practice 1 0.61 1.00 3. Practice 2 0.57 0.69 1.00 4. Practice 3 0.70 0.74 0.67 1.00 5. Practice 4 0.68 0.75 0.70 0.73 1.00 6. Rewards for innovation 0.64 0.66 0.80 0.66 0.70 7. Job satisfaction 0.58 0.63 0.57 0.66 0.70 8. Overall performance 0.44 0.47 0.43 0.53 0.48 9. Knowledge sharing 0.38 0.48 0.43 0.50 0.46 10. Trust in leader 0.59 0.63 0.57 0.65 0.62 11. Sufficient resources 0.36 0.43 0.40 0.44 0.44 12. Location -0.04 0.01 -0.03 -0.02 -0.01 14. Age -0.01 -0.02 -0.01 -0.02 -0.01 6 7 8 9 10 1. Encouragement to innovate 2. Practice 1 3. Practice 2 4. Practice 3 5. Practice 4 6. Rewards for innovation 1.00 7. Job satisfaction 0.56 1.00 8. Overall performance 0.41 0.46 1.00 9. Knowledge sharing 0.40 0.41 0.46 1.00 10. Trust in leader 0.54 0.56 0.46 0.38 1.00 11. Sufficient resources 0.38 0.40 0.29 0.24 0.33 12. Location -0.03 0.02 -0.02 0.03 -0.02 14. Age 0.01 0.01 0.04 -0.01 -0.01 11 12 13 14 1. Encouragement to innovate 2. Practice 1 3. Practice 2 4. Practice 3 5. Practice 4 6. Rewards for innovation 7. Job satisfaction 8. Overall performance 9. Knowledge sharing 10. Trust in leader 11. Sufficient resources 1.00 12. Location -0.04 1.00 14. Age -0.02 0.01 -0.06 1.00
Higher Order CFA, Employee Empowerment
CFA, Employee Empowerment Practices.
Alonso, Pablo, and Gregory B. Lewis. 2001. Public service motivation and job performance: Evidence from the federal sector. American Review of Public Administration 31:363-80.
Altshuler, Alan A., and Robert D. Behn. 1997. Innovation in American Government: Challenges, opportunities, and dilemmas. Washington, DC: Brookings Institution.
Altshuler, Alan A., and Marc D. Zegans. 1997. Innovation and public management: Notes from the State House and City Hall. Innovation in American Government: Challenges, opportunities, and dilemmas, 68-80. Washington, DC: Brookings Institution.
Amburgey, Terry L., Dawn Kelly, and William P. Barnett. 1993. Resetting the clock: The dynamics of organizational change and failure. Administrative Science Quarterly 38:51-73.
Argyris, Chris. 1957. Personality and organization: The conflict between system and individual. New York: Harper.
Bandura, Albert. 1977. Social learning theory. Englewood Cliffs, NJ: Prentice Hall.
--. 1986. Social foundations of thought and action: A social cognitive theory. Englewood Cliffs, NJ: Prentice-Hall.
Bartos, Stephen. 2003. Creating and sustaining innovation. Australian Journal of Public Administration 62:9-14.
Barzelay, Michael. 2001. The new public management: Improving research and policy dialogue. Berkeley: Univ. of California Press.
Bass, Bernard M. 1990. Bass and Stogdill's handbook of leadership: Theory, research, and managerial applications, 3rd ed. New York: Free Press.
Baum, Christopher F. 2006. An introduction to modern econometrics using Stata. College Station, TX: Stata Press.
Becker, Selwyn W., and Frank Stafford. 1967. Some determinants of organizational success. Journal of Business 40:511-18.
Bergkvist, Lars, and John R. Rossiter. 2007. The predictive validity of multiple-item versus single-item measures of the same constructs. Journal of Marketing Research 44:175-84.
Berman, Evan K., and Jonathan P. West. 1998. Responsible risk-taking. Public Administration Review 58:346-52.
Berry, Frances S. 1994. Innovation in public management: The adoption of strategic planning. Public Administration Review 54:322-30.
Borins, Sanford. 2000a. Loose cannons and rule breakers, or enterprising leaders? Some evidence about innovative public managers. Public Administration Review 60:498-507.
--. 2000b. What border? Public management innovation in the United States and Canada. Journal of Policy Analysis and Management 19:46-74.
--. 2001. Encouraging innovation in the public sector. Journal of Intellectual Capital 2:310-19.
--. 2008. Innovations in government: Research, recognition, and replication. Washington, DC: Brookings Institution.
--. 2012. Making narrative count: A narratological approach to public management innovation. Journal of Public Administration Research and Theory 21:165-89.
Bowen, David E., and Edward E. Lawler. 1992. The empowerment of service workers: What, why, how, and when. Sloan Management Review 33:31-39.
--. 1995. empowering service employees. Sloan Management Review 36:73-84.
Breaux, David A., Christopher M. Duncan, C. Denise Keller, and John C. Morris. 2002. Welfare reform, Mississippi style: Temporary assistance for needy families and the search for accountability. Public Administration Review 62:92-103.
Breul, Jonathan D., and John M. Kamensky. 2008. Federal government reform: Lessons from Clinton's "Reinventing Government" and Bush's "Management Agenda" initiatives. Public Administration Review 68:1009-26.
Burnham, Kenneth P., and David R. Anderson. 2004. Understanding AIC and BIC in model selection. Sociological Methods 33:261-304.
Burns, Tom, and George M. Stalker. 1961. The Management of Innovation. London: Tavistock.
Chakerian, Richard, and Paul Mavima. 2000. Comprehensive administrative reform implementation: Moving beyond single issue implementation research. Journal of Public Administration Research and Theory 11:353-77.
Christensen, Clayton M. 1997. The innovator "s dilemma. When new technologies cause great firms to fail. Cambridge, MA: Harvard Business.
Cohn, Daniel. 1997. Creating crises and avoiding blame: The politics of public service reform and the new public management in Great Britain and the United States. Administration and Society 29:584-616.
Conger, Jay A., and Rabindra N. Kanungo. 1988. The empowerment process: Integrating theory and practice. Academy of Management Review 13:471-82.
Cote, Joseph A., and Ronald Buckley. 1988. Measurement error and theory testing in consumer research: An illustration of the importance of construct validation. Journal of Consumer Research 14:579-82.
Crampton, Suzanne, and John Wagner. 1994. Percept-percept inflation in micro-organizational research: An investigation of prevalence and effect. Journal of Applied Psychology 79:67-76.
Cyert, Richard M., and James G. March. 1963. A behavioral theory of the firm. Upper Saddle River, NJ: Prentice Hall.
Daley, Dennis M. 1992. Performance appraisal in the public sector. Wesport, CT: Quorum Books.
Damanpour, Fariborz. 1991. Organizational innovation: A meta-analysis of effects of determinants and moderators. Academy of Management Journal 34:555-90.
Damanpour, Fariborz, Richard M. Walker, and Claudia N. Avellaneda. 2009. Combinative effects of innovation types on organizational performance: A longitudinal study of public services. Journal of Management Studies 46:650-75.
Deci, Edward. 1971. Effects of externally mediated rewards on intrinsic motivation. Journal of Personality and Social Psychology 18:105-15.
Dewar, Robert, and Jane Dutton. 1986. The adoption of radical and incremental innovation: An empirical analysis. Management Science 32:1422-1433.
Edmonson, Amy. 1999. Psychological safety and learning behavior in work teams. Administrative Science Quarterly 44:350-83.
Enticott, Gareth, George A. Boyne, and Richard M. Walker. 2009. The use of multiple informants in public administration research: Data aggregation using organizational echelons. Journal of Public Administration Research and Theory 19:229-53.
Fagerberg, Jan, David C. Mowery, and Richard R. Nelson, eds. 2006. The Oxford handbook of innovation. Oxford: Oxford Univ. Press.
Fan, Xitao, Bruce Thompson, and Lin Wang. 1999. Effects of sample size, estimation method, and model specification on structural equation modeling fit indexes. Structural Equation Modeling 6:56-83.
Fernandez, Sergio, and Tima Moldogaziev. 2011. Empowering public sector employees to improve performance: Does it work? American Review of Public Administration 41:23-47.
Fernandez, Sergio, and Hal G. Rainey. 2006. Managing successful organizational change in the public sector: An agenda for research and practice. Public Administration Review 66:168-76.
Fernandez, Sergio, and Lois R. Wise. 2010. An exploration of why public organizations "ingest" innovations. Public Administration 88:979-98.
Follet, Mary P. 1926. The psychological foundations: Constructive conflict. In Scientific foundations of business, ed. Henry C. Metcalf. Baltimore, MD: Williams and Wilkins.
Fornell, Claes, and David Larcker. 1981. Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research 18:39-50.
Golden, Olivia. 1990. Innovation in public sector human services programs: The implications of innovation by "groping along." Journal of Policy Analysis and Management 9:219-48.
Gore, Albert. 1993. From red tape to results: Creating a government that works better and costs less. Report of the National Performance Review, Washington, DC.
Greene, Colin, and John S. Haywood. 2008. Does performance pay increase job satisfaction? Economica 75:710-28.
Gujarati, Damodar, and Dawn Porter. 2010. Essentials of econometrics. New York: McGraw-Hill.
Guthrie, James. 2001. High-involvement work practices, turnover, and productivity: Evidence from New Zealand. Academy of Management Journal 44:180-92.
Hackman, J. Richard, and Greg R. Oldham. 1976. Motivation through the design of work: Test of a theory. Organizational Behavior and Human Performance 16:250-79.
Hage, Jerald T., and Michael Aiken. 1967. Program change and organizational properties: A comparative analysis. American Journal of Sociology 72:503-19.
Hartley, Jean. 2005. Innovation in governance and public services: Past and present. Public Money 25:27-34.
Heinrich, Carolyn J. 2007. False or fitting recognition? The use of high performance bonuses in motivating organizational achievements. Journal of Policy Analysis and Management 26:281-304.
Judge, Timothy A., and Allan H. Church. 2000. Job satisfaction: Research and practice. In Industrial and Organizational Psychology, ed. Cary L. Cooper and Edwin A. Locke. Oxford: Blackwell, 166-98.
Kamensky, John. 1996. Role of "reinventing government movement" in federal management reform. Public Administration Review 56:247-55.
Kanfer, Ruth, and Phillip L. Ackerman. 1989. Motivation and cognitive abilities: An integrative/aptitude-treatment interaction approach to skill acquisition. Journal of Applied Psychology 74:657-90. Kanter, Rosabeth M. 1982. The middle manager as innovator. Harvard Business Review 60:95-105.
--. 1983. The change masters: Innovations for productivity in the American corporation. New York: Simon and Schuster.
Katz, Ralph, and Michael Tushman. 1981. An investigation into the managerial roles and career paths of gate keepers and project supervisors in a major R&D facility. R&D Management 11:103-10.
Kellough, J. Edward. 2002. Employee performance appraisal and pay for performance in the public sector: A critical examination. New York: Addison Wesley Longman.
Kellough, J. Edward, and Haoran Lu. 1993. The paradox of merit pay in the public sector. Review of Public Personnel Administration 13:45-64.
Kellough, J. Edward, and Lloyd G. Nigro. 2002. Pay for performance in Georgia state government: Employee perspectives on Georgia gain after 5 years. Review of Public Personnel Administration 22:146-66.
Kettl, Donald F. 2005. The global public management revolution: A report on the transformation of governance, 2nd ed. Washington, DC: Brookings Institution.
Kim, Soonhee. 2002. Participative management and job satisfaction: Lessons for management leadership. Public Administration Review 62:231-41.
Kirkman, Bradley L., and Benson Rosen. 1999. Beyond self-management: Antecedents and consequences of team empowerment. Academy of Management Journal 42:58-74.
Lawler, Edward E., III. 1973. Motivation in work organizations. Monterey, CA: Goodyear.
Lawler, Edward. E, III., Susan. A. Mohrman, and Gerald. E. Ledford 1992. The fortune 1000 and total quality. Quality and Participation 15:6-10.
--. 1995. Creating high performance organizations: Impact of employee involvement and total quality management. San Francisco, CA: Jossey-Bas.
Lee, Haksoo, N. Joseph Cayer, and G. Zhiyong Lan. 2006. Changing federal government employee attitudes since the Civil Service Reform Act of 1978. Review of Public Personnel Administration 26:21-51.
Levin, Martin A., and Mary B. Sanger. 1994. Making government work: How entrepreneurial executives turn bright ideas into real results. San Francisco, CA: Jossey-Bass.
Light, Paul. 1998. Sustaining innovation: Creating nonprofit and government organizations that innovate naturally. San Francisco, CA: Jossey-Bass.
Likert, Rensis. 1967. The human organization. New York: McGraw-Hill.
Locke, Edwin A., and Gary P. Latham. 1990. A theory of goal setting and task performance. Upper Saddle River, NJ: Prentice Hall.
--. 2004. What should we do about motivation theory? Six recommendations for the twenty-first century. Academy of Management 29:388-403.
Long, Scott. 1997. Regression methods for categorical and limited dependent variables. Thousand Oaks, CA: Sage.
Long, Scott, and Jeremy Freese. 2005. Regression models for categorical dependent variables using Stata, 2nd ed. College Station, TX: Stata Press.
Mamas, Curtis L., and James G. March. 1978. Financial adversity, internal competition, and curriculum change in a university. Administrative Science Quarterly 23:541-52.
March, James, and Herbert Simon. 1993. Organizations, 2nd ed. Cambridge, UK: Blackwell.
Matheson, Craig. 2007. In praise of bureaucracy? A dissent from Australia. Administration and Society 39:233-61.
McGinnis, Michael A., and M. Robert Ackelsberg. 1983. Effective innovation management: The missing link in strategic planning? Journal of Business Strategy 4:59-66.
McGregor, Douglas. 1960. The human side of enterprise. New York: McGraw-Hill.
McKelvey, Richard D., and William Zavoina. 1975. A statistical model for the analysis of ordinal level dependent variables. Journal of Mathematical Sociology 4:103-20.
Mintzberg, Henry. 1979. The structuring of organizations. Upper Saddle River, NJ: Prentice Hall.
Monge, Peter R., Michael D. Cozzens, and Noshir S. Contractor. 1992. Communication and motivational predictors of the dynamics of organizational innovation. Organization Science 3:250-74.
Mottaz, Clifford J. 1985. The relative importance of intrinsic and extrinsic rewards as determinants of work satisfaction. Sociological Quarterly 26:365-85.
Nagel, Stuart S., and G. David Garson. 1996. Advances in social science and computers. Greenwich, CT: JAI Press.
Nielsen, Join F., and Christian P. Pedersen. 2003. The consequences and limits of empowerment in financial services. Scandinavian Journal of Management 19:63-83.
Nunnally, Jum, and Ira Bernstein. 1994. Psychometric theory. New York: McGraw-Hill.
O'Reilly, Charles A., and David F. Caldwell. 1980. Job choice: The impact of intrinsic and extrinsic factors on subsequent satisfaction and commitment. Journal of Applied Psychology 65:559-65.
Park, Sung Min., and Hal G. Rainey. 2007. Antecedents, mediators, and consequences of affective, normative, and continuance commitment: Empirical tests of commitment effects in federal agencies. Review of Public Personnel Administration 27:197-226.
Perry, James L., Trent A. Engbers, and So Yun Jun. 2009. Back to the future? Performance-related pay, empirical research, and the perils of performance. Public Administration Review 69:39-51.
Perry, James L., Debra Mesch, and Laurie Paarlberg. 2006. Motivating employees in a new governance era: The performance paradigm revisited. Public Administration Review 66:505-14.
Peters, B. Guy. 1996. The future of governing: Four emerging models. Lawrence: Univ. Press of Kansas.
Peters, B. Guy, and Jon Pierre. 2000. Citizens versus the new public manager: The problem of mutual empowerment. Administration and Society 32:9-28.
Pettigrew, Andrew M. 1972. Information control as a power source. Sociology 6:187-204.
Podsakoff, Phillip M., Scott B. MacKenzie, Jeong-Yeon Lee, and Nathan Podsakoff. 2003. Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology 88:879-903.
Pollitt, Christopher. 1990. Managerialism and the public services. Oxford: Blackwell.
Pollitt, Christopher, and Geert Bouckaert. 2004. Public management reform, 2nd ed. Oxford: Oxford Univ. Press.
Porter, Michael E. 1985. Competitive advantage. New York: Free Press.
Potterfield, Thomas A. 1999. The business of employee empowerment: Democracy and ideology in the workplace. Westport, CT: Quorum Books.
Rainey, Hal G. 2009. Understanding and managing public organizations, 4th ed. San Francisco, CA: Jossey-Bass.
Rainey, Hal G., and Barry Bozeman. 2000. Comparing public and private organizations: Empirical research and the power of the a priori. Journal of Public Administration Research and Theory 10:447-69.
Ryan, Richard, and Edward Deci. 2000. Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist 55:68-78.
Salancik, Gerald R., and Jeffrey Pfeffer. 1974. The bases and use of power in organizational decision making: The case of a university. Administrative Science Quarterly 19:453-73.
Salge, Torsten.. 2011. A behavioral model of innovative search: Evidence from public hospital services. Journal of Public Administration Research and Theory 21:181-210.
Schumacker, Randall, and Richard Lomax. 2004. A beginner's guide to structural equations modeling. Mahwah, NJ: Lawrence Erlbaum.
Selden, Sally C. 2009. Human capital: Tools and strategies for the public sector. Washington, DC: CQ Press.
Simon, Herbert A. 1997. Administrative behavior, 4th ed. New York: Free Press.
Spreitzer, Gretchen M. 1995. Psychological empowerment in the workplace: Dimensions, measurement, and validation. Academy of Management Journal 38:144-245.
--. 1996. Social structural characteristics of psychological empowerment. Academy of Management Journal 39:483-504.
Thomas, Kenneth W., and Betty A. Velthouse. 1990. Cognitive elements of empowerment: An "interpretive" model of intrinsic task motivation. Academy of Management Review 15:666-81.
Thompson, James R., and Ronald P. Sanders. 1997. Strategies for reinventing federal agencies. Public Productivity and Management Review 21:137-55.
Thompson, Victor A. 1965. Bureaucracy and innovation. Administrative Science Quarterly 10:1-20.
Tjosvold, Dean, and Leonard Y. McNeely. 1988. Innovation through communication in an educational bureaucracy. Communication Research 15:568-81.
Tosi, Henry L., and Stephen J. Carroll. 1968. Managerial reaction to management by objectives. Academy of Management Journal 11:415-26.
U.S. Office of Personnel Management (OPM). 2001. A handbook for measuring employee performance: Aligning employee performance plans with organizational goals. Washington, DC: U.S. OPM.
--. 2011. Performance appraisal rating tool. Washington, DC: U.S. OPM.
Vroom, Victor. 1964. Work and motivation. New York: John Wiley and Sons.
Walker, Richard M., and Fariborz Damanpour. 2008. Innovation type and organizational performance: An empirical exploration. In Managing improvement in public service delivery." Progress and prospects, ed. C. Donaldson, J. Hartley, C. Skelcher, and M. Wallace. Cambridge: Cambridge Univ. Press, 217-35.
Walker, Richard M., Fariborz Damanpour, and Carlos A. Devece. 2011. Management innovation and organizational performance: The mediating effect of performance management. Journal of Public Administration Research and Theory 21:36786.
Wallin, Bruce A. 1997. The need for a privatization process: Lessons from development and implementation. Public Administration Review 57:11-20.
Wanous, John P., Arnon E. Reichers, and Michael J. Hudy. 1997. Overall job satisfaction: How good are single-item measures? Journal of Applied Psychology 82:247-52.
Winship, Christopher, and Robert D. Mare. 1984. Regression models with ordinal variables. American Sociological Review 49:512-25.
Winters, Dawn, and Gary P. Latham. 1996. The effect of learning versus outcome goals on a simple versus a complex task. Group and Organization Management 21:236-50.
Wise, Louise R. 2002. Public management reform: Competing drivers of change. Public Administration Review 62:543-54.
Wittmer, Dennis. 1991. Serving the people or serving for pay: Reward preferences among government, hybrid sector, and business managers. Public Productivity and Management Review 14:369-83.
Wright, Bradley E. 2007. Public service and motivation: Does mission matter? Public Administration Review 67:54-64.
Wright, Bradley E., and Soonhee Kim. 2004. Participation's influence on job satisfaction: The importance of job characteristics. Review of Public Personnel Administration 24:18-40.
(1) "Impact" refers to the extent to which behavior is seen to make a difference as to whether or not a task will be accomplished. Competence refers to self-efficacy or "the degree to which a person can perform task activities skillfully when he or she tries" (Thomas and Velthouse 1990, 672). Meaningfulness is conceived in terms of how much value an employee places on accomplishing the task. Finally, choice is defined as locus of causality, referring to "the issue of whether a person's behavior is perceived as self-determined" (Thomas and Velthouse 1990, 673).
(2) Ironically, the NPR and other NPM reform initiatives involved adopting empowerment programs in concert with reforms such as downsizing and privatization that entailed greater managerial control and signaled dissatisfaction and mistrust of public employees (Barzelay 2001; Cobn 1997; Peters and Pierre 2000). The extent to which these reforms have created an empowered state of mind among public employees thus remains an open question.
(3) Budget constraints and inadequate funding of pay-for-performance schemes in the public sector have resulted in reduced employee expectancy of receiving financial rewards and the valence attached to such rewards (Heinrich 2007; Kellough and Lu 1993; Kellough and Nigro 2002).
(4) OPM appears not to have performed an analysis of nonresponse bias.
(5) An MNLM regression using the full sample of respondents produces remarkably similar results to those reported above from a sample of only nonsupervisory employees, team leaders, and supervisors.
(6) A higher order CFA was performed to assess the measurement of Bowen and Lawler's four-dimensional empowerment construct. Multiple survey items shown in Appendix 1 were used to measure the four empowerment practices. Given the categorical nature of these items, a categorical polychoric maximum likelihood (CPML) method with bootstrapped standard errors was used to estimate the model in R. In the four-dimensional model shown in Appendix 4, each of the survey items loaded strongly and in the anticipated direction with the corresponding factor (i.e., empowerment practice) (p < .001). Those four factors, in turn, have very strong positive correlations with a second-order factor representing the underlying construct of employee empowerment (p < .001). The statistics for several GFIs support the four-factor model of empowerment. The statistic for the comparative fit index (CFI), which is minimally affected by sample size, is 0.92, indicating a good fit for the four-factor model (Fan, Thompson, and Wang 1999). The Joreskog and Sorbom GFI of 0.90 also suggests a good model fit. The normed fit index (NFI) statistic of 0.90 also points to an acceptable fit for the four-factor model (Schumacker and Lomax 2004). Complex models are more likely to generate better-fit statistics than parsimonious ones. It is recommended, therefore, that models be subjected to goodness-of-fit measures that penalize for lack of parsimony. The model with a four-factor structure has parsimony ratio and parsimony NFI statistics of 0.78 and 0.71, respectively, both of which are indicative of a reasonably parsimonious fit. It should be noted that the chi-square test results reject the four-factor model (109,365, N = 154,793, 50 degrees of freedom) at thep < .01 level. Large sample sizes like the one used in this CFA are much more likely to result in type II errors. Schumacker and Lomax (2004) suggest, therefore, discounting the chi-square results if other fit statistics support a model with such a large sample size. In contrast to the evidence favoring a fourfactor model of employee empowerment, the higher order CFA results reject a model with a one-factor structure. The CFI and NFI statistics for a one-factor model fail to reach the 0.90 cutoffpoint. In addition, a comparison of the fourfactor and one-factor models, in terms of their Bayesian information criterion (BIC) statistics, favors the former over the latter. The lower BIC statistic for the four-factor model (108,768) is considerably lower than the BIC statistic for the one-factor model (126,400), indicating a significantly better model fit (Burnham and Anderson 2004; Long 1997). Finally, the absolute value of the difference in chi squares between the four-factor and one-factor model is sufficiently large and indicative of a statistically significant difference in support of the four-factor model (p < .00l).
(7) According to Fornell and Larcker (1981), an AVE greater than 0.50 is indicative of convergent validity. The four empowerment practices have AVEs ranging from 0.58 (practice 3) to 0.75 (practice 2). Discriminant validity is assessed by comparing the square root of the AVE of an empowerment practice to the correlations between that practice and the remaining practices. A square root of an AVE greater than the correlations between an empowerment practice and the remaining practices is indicative of discriminant validity. The results show that the square root of AVE is greater than all the relevant correlations for all four empowerment practices, with differences ranging from 0.22 (practice 1) to 0.06 (practice 3).
(8) The four-factor model without a higher order factor was estimated in R using the CPML method with bootstrapped standard errors. The results appearing in Appendix 5 show that each of the survey items loaded strongly and in the anticipated direction with the corresponding factor (i.e., empowerment practice) (p < .001). The fit statistics are nearly identical to those from the higher-order four-factor model, and in the case of the CFI and NFI indicate slightly better fit (0.93 versus 0.92). Due once again to large sample size, the chi square test rejects the four-factor model (107,773, n = 154,793, 50 degrees of freedom) at the p < .01 level. The results support this four-factor model over the one-factor model, as the former has a considerably lower Bayesian information criterion (BIC) statistic (I 07,200 versus 126,400), and the difference in chi-squares between the four-factor and one-factor model is sufficiently large and indicative of a statistically significant difference in support of four-factor model (p < .001).
(9) The results appearing in Appendix 5 show that the correlations between the four empowerment practices are all between 0.60 (practice 3 and practice 4) and 0.68 (practice 2 and practice 4). Discriminant validity was again assessed by comparing the square root of the AVE of an empowerment practice to the correlations between that practice and the remaining practices. A square root of an AVE greater than the correlations between an empowerment practice and the remaining practices is indicative of divergent validity (Fornell and Larcker 1981). The results show that the square root of AVE is greater than all the relevant correlations for all four empowerment practices, with differences ranging from 0.25 (practice 1) to 0.13 (practice 3).
(10) The independent variable practice 2, as characterized by Bowen and Lawler (1992, 1995; see also Lawler, Mohrman, and Ledford 1995), refers to practices like merit pay and employee recognition systems that reward employees based on results-oriented performance criteria, such as outputs, quality, profitability, and customer satisfaction. Their work makes no mention of offering rewards for producing innovations, which typically involve proposed or actual changes in process. Similarly, in the public sector, performance appraisal and reward systems have a tendency to focus on outcomes, outputs, and task-oriented behaviors as performance criteria, with very little consideration for using innovative behavior or process changes as criteria on which to judge performance (see Daley 1992; Kellough 2002; OPM 2001, 2011; Selden 2009).
(11) A number of federal agencies (e.g., US Department of Agriculture, National Institutes of Health, National Aeronautics and Space Administration, and US Agency for International Development) have employee suggestion programs in which an ad hoe committee evaluates innovative proposals submitted by individuals or groups as they come forth and determines whether or not to grant a reward based on the projected gains in productivity that would be achieved from the adoption of the proposal. Other agencies (e.g., the US Department of Health and Human Services and General Services Administration) have innovation programs that recognize employees who have been able to demonstrate the success of their innovations, in some cases over the course of many years.
(12) Compared with the original CFA model in which practice 2 is measured using four observable indicators (see Appendix 4), the new CFA model with five observable indicators for practice 2 has lower Joreskog and Sorbom GFI and NFI statistics (0.87 and 0.89, respectively) that no longer reach levels indicative of good model fit. The latter model also has a considerably higher chi-square statistic (140,766 compared with 109,365, p < .001), suggestive of a worse fit.
(13) Since the survey respondents are clustered by agency, the same MNLM was estimated using clustered standard errors and using a series of agency dummy variables. Doing this failed to have a meaningful effect on the MNLM results in terms of statistical and substantive significance and model fit.
(14) "Agree" was chosen because it is the category with the highest frequency in the dependent variable's distribution, a common practice when estimating MNLM regression. Almost all statistical packages use the category with the highest frequency as a comparison category by default. The multinomial results in table 2 (third column) provide important but limited information; the sign of the reported coefficient provides the direction of the effect for a binary comparison of the lowest outcome choice "strongly disagree" to the comparison category "agree." That is why the results in tables 3 and 4 are very important. Table 3 provides the coefficients, log-odds, and levels of significance for the effects of the four empowerment practices on levels of the dependent variable for all binary comparisons: "strongly disagree," "disagree," "neither disagree or agree," and "strongly agree" versus "agree," the comparison choice.
(15) The control variable knowledge sharing is negatively correlated with the dependent variable, the opposite of what was expected. In terms of substantive significance (as indicated by the odds ratio coefficients), however, the magnitude of the relationship is close to zero.
(16) The full set of results are available from the authors upon request.
Sergio Fernandez *, Tima Moldogaziev ([dagger])
* Indiana University; ([dagger]) University of South Carolina
Address correspondence to the author at email@example.com.
Advance Access publication May 23, 2012
Table 1 Descriptive Statistics for Independent and Control Variables Variable Mean SD Min. Max. Practice 1 (providing information about -0.04 0.85 -2.37 1.50 goals and performance) Practice 2 (offering rewards based on -0.06 0.92 -1.93 1.74 performance) Practice 3 (providing access to job-related -0.04 0.84 -2.66 1.27 knowledge and skills) Practice 4 (granting discretion to change -0.04 0.79 -1.90 1.34 work processes) Rewards for innovation 3.14 1.13 1 5 Job satisfaction 3.70 1.03 1 5 Overall performance 4.20 0.78 1 5 Knowledge sharing 3.81 1.00 1 5 Trust in leader 3.70 1.19 1 5 Sufficient resources 3.14 1.21 1 5 Location 0.61 0.49 0 1 Minority 0.26 0.44 0 1 Age 3.34 0.98 1 5 Source: FHCS 2006 dataset sample size = 189,856. Note: Sample restricted to 154,793 observations in the multivariate analysis due to missing data. Table 2 Multinomial Logit Regression Estimates and Tests of Fit, Dependent Variable = Encouragement to Innovate, Comparison Response Category = "Agree," N = 154,793 Variable Description Model I Employee empowerment approach Practice 1 (providing information -0.020 (-0.71) about goals and performance) Practice 2 (offering rewards 0.617 *** (19.83) based on performance) Practice 3 (providing access to -1.680 *** (-57.32) job-related knowledge and skills) Practice 4 (granting discretion to -0.804 *** (-49.61) change work processes) Rewards for innovation -1.683 *** (-58.05) Job satisfaction -0.269 *** (-14.59) Overall performance -0.166 *** (-8.12) Knowledge sharing 0.124 *** (8.08) Trust in leader -0.507 *** (-32.27) Sufficient resources -0.023 (-1.58) Location (1 = field office) 0.327 *** (9.73) Minority (1 = nonwhite) -0.233 *** (-6.28) Age category -0.063 ** (-3.67) AIC 312,262.4 BIC 312,819.6 BIC prime -133,207.5 Likelihood ratio [chi square] 133,828.9 McFadden's [R.sup.2] .300 ML (Cox-Snell) [R.sup.2]] .579 Cragg-Uhler's (Nagelkerke) [R.sup.2]] .613 Variable Description Model II Employee empowerment approach -2.244 *** (-63.09) Practice 1 (providing information about goals and performance) Practice 2 (offering rewards based on performance) Practice 3 (providing access to job-related knowledge and skills) Practice 4 (granting discretion to change work processes) Rewards for innovation -1.294 *** (-50.68) Job satisfaction -0.425 *** (-24.09) Overall performance -0.225 *** (-11.35) Knowledge sharing 0.124 *** (8.08) Trust in leader -0.485 *** (-31.83) Sufficient resources -0.062 *** (-4.32) Location (1 = field office) 0.299 *** (9.12) Minority (1 = nonwhite) -0.211 *** (-5.84) Age category -0.047 ** (-2.82) AIC 332,906.3 BIC 323,344.1 BIC prime -122,683.1 Likelihood ratio [chi square] 123,161.1 McFadden's [R.sup.2] .276 ML (Cox-Snell) [R.sup.2]] .549 Cragg-Uhler's (Nagelkerke) [R.sup.2]] .581 * p <.05; ** p <.01; *** p <.001. Table 3 Mulinomial Logit Regression Coefficients, Dependent Variable = Encouragement to Innovate, Comparison Response Category = "Agree", N = 154,793 Coefficients for Empowerment Practices Practice 1 (Providing Information about Goals and Comparison Statistic Performance) SD|A [[beta].sub.SD|A] -0.020 exp([[beta].sub.SD|A]) 0.980 z -0.710 p .478 D|A [[beta].sub.D|A] -0.009 exp([[beta].sub.D|A]) .991 z -0.461 p .645 N|A [[beta].sub.N|A] 0.003 exp([[beta].sub.N|A]) 1.003 z 0.205 p .838 SA|A [[beta].sub.SA|A] -0.125 exp([[beta].sub.SD|A]) 0.882 z -6.912 p .001 Coefficients for Empowerment Practices Practice 2 (Offering Rewards Based on Comparison Statistic Performance) SD|A [[beta].sub.SD|A] 0.617 exp([[beta].sub.SD|A]) 1.853 z 19.825 p .001 D|A [[beta].sub.D|A] 0.427 exp([[beta].sub.D|A]) 1.533 z 22.494 p .001 N|A [[beta].sub.N|A] 0.233 exp([[beta].sub.N|A]) 1.262 z 15.563 p .001 SA|A [[beta].sub.SA|A] -0.359 exp([[beta].sub.SD|A]) 0.698 z -21.988 p .001 Coefficients for Empowerment Practices Practice 3 (Providing Access to Job-Related Comparison Statistic Knowledge and Skills) SD|A [[beta].sub.SD|A] -1.680 exp([[beta].sub.SD|A]) 0.186 z -57.317 p .001 D|A [[beta].sub.D|A] -1.021 exp([[beta].sub.D|A]) 0.360 z -51.892 p .001 N|A [[beta].sub.N|A] -0.597 exp([[beta].sub.N|A]) 0.551 z -36.135 p .001 SA|A [[beta].sub.SA|A] 1.508 exp([[beta].sub.SD|A]) 4.519 z 67.057 p .001 Coefficients for Empowerment Practices Practice 4 (Granting Discretion to Change Comparison Statistic Work Processes) SD|A [[beta].sub.SD|A] -1.804 exp([[beta].sub.SD|A]) 0.165 z -49.611 p .001 D|A [[beta].sub.D|A] -1.146 exp([[beta].sub.D|A]) 0.318 z -53.125 p .001 N|A [[beta].sub.N|A] -0.569 exp([[beta].sub.N|A]) 0.566 z -33.006 p .001 SA|A [[beta].sub.SA|A] 0.579 exp([[beta].sub.SD|A]) 1.784 z 27.235 p .001 Coefficient for Select Control Variables Comparison Statistic Rewards for Job Satisfaction Innovation SD|A [[beta].sub.SD|A] -1.683 -0.269 exp([[beta].sub.SD|A]) 0.186 0.764 z -58.054 -14.595 p .001 .001 D|A [[beta].sub.D|A] -0.927 -0.144 exp([[beta].sub.D|A]) 0.396 0.866 z -60.565 -11.340 p .001 .001 N|A [[beta].sub.N|A] -0.416 -0.107 exp([[beta].sub.N|A]) 0.659 0.899 z -35.594 -9.946 p .001 .001 SA|A [[beta].sub.SA|A] 0.340 0.353 exp([[beta].sub.SA|A]) 1.491 1.287 z 30.466 17.829 p .001 .001 Coefficient for Select Control Variables Comparison Statistic Trust in Leader SD|A [[beta].sub.SD|A] -0.507 exp([[beta].sub.SD|A]) 0.602 z -32.265 p .001 D|A [[beta].sub.D|A] -0.351 exp([[beta].sub.D|A]) 0.704 z -33.620 p .001 N|A [[beta].sub.N|A] -0.175 exp([[beta].sub.N|A]) 0.840 z -20.032 p .001 SA|A [[beta].sub.SA|A] 0.395 exp([[beta].sub.SA|A]) 1.484 z 32.625 p .001 Note: (a) [beta] is a logit coefficient, (b) exp ([beta]) is a factor change, (c) z is a z-statistic, and (d) p is a significance level. Encouragement to innovate (five-choice outcome): SD, "strongly disagree"; D, "disagree"; N, "neither agree nor disagree"; A, "agree"; and SA, "strongly agree." Table 4 Discrete Change in the Probability of Encouragement to Innovate for Multinomial Logit Regression (All Other Variables Held at Mean Values) Variable Change [bar.[DELTA]] Mean Overall P([??] at -- mean Practice 1 (providing [DELTA] Range 0.01 information aboutgoals and performance) [DELTA] 1 0.00 SD[DELTA] 0.00 Practice 2 (offering rewards [DELTA] Range 0.10 based on performance) [DELTA] 1 0.03 SDA 0.03 Practice 3 (providing access [DELTA] Range 0.28 to job-related knowledge and skills) [DELTA] 1 0.08 SD[DELTA] 0.07 Practice 4 (granting discretion [DELTA] Range 0.22 to change work processes) [DELTA] 1 0.07 SD[DELTA] 0.05 Rewards for innovation [DELTA] Range 0.21 [DELTA] 1 0.05 SD[DELTA] 0.06 Job satisfaction [DELTA] Range 0.05 [DELTA] 1 0.01 SD[DELTA] 0.01 Trust in leader [DELTA] Range 0.10 [DELTA] 1 0.02 SD[DELTA] 0.03 Variable Change SD D Overall P([??] at 0.002 0.074 mean Practice 1 (providing [DELTA] Range -0.00 0.00 information aboutgoals and performance) [DELTA] 1 -0.00 0.00 SD[DELTA] -0.00 0.00 Practice 2 (offering rewards [DELTA] Range 0.00 0.10 based on performance) [DELTA] 1 0.00 0.03 SDA 0.00 0.02 Practice 3 (providing access [DELTA] Range -0.05 -0.33 to job-related knowledge and skills) [DELTA] 1 -0.00 -0.07 SD[DELTA] -0.00 -0.06 Practice 4 (granting discretion [DELTA] Range -0.03 -0.29 to change work processes) [DELTA] 1 -0.00 -0.07 SD[DELTA] -0.00 -0.06 Rewards for innovation [DELTA] Range -0.04 -0.27 [DELTA] 1 -0.00 -0.06 SD[DELTA] -0.00 -0.06 Job satisfaction [DELTA] Range -0.00 -0.04 [DELTA] 1 -0.00 -0.01 SD[DELTA] -0.00 -0.01 Trust in leader [DELTA] Range -0.01 -0.11 [DELTA] 1 -0.00 -0.02 SD[DELTA] -0.00 -0.03 Variable Change N SA Overall P([??] at 0.261 0.079 mean Practice 1 (providing [DELTA] Range 0.01 -0.04 information aboutgoals and performance) [DELTA] 1 0.00 -0.01 SD[DELTA] 0.00 -0.01 Practice 2 (offering rewards [DELTA] Range 0.15 -0.13 based on performance) [DELTA] 1 0.04 -0.03 SDA 0.04 -0.03 Practice 3 (providing access [DELTA] Range -0.31 0.44 to job-related knowledge and skills) [DELTA] 1 -0.12 0.14 SD[DELTA] -0.10 0.11 Practice 4 (granting discretion [DELTA] Range -0.24 0.18 to change work processes) [DELTA] 1 -0.10 0.06 SD[DELTA] -0.08 0.05 Rewards for innovation [DELTA] Range -0.21 0.17 [DELTA] 1 -0.07 0.04 SD[DELTA] -0.08 0.04 Job satisfaction [DELTA] Range -0.09 0.07 [DELTA] 1 -0.02 0.02 SD[DELTA] -0.02 0.02 Trust in leader [DELTA] Range -0.13 0.11 [DELTA] 1 -0.03 0.03 SD[DELTA] -0.04 0.04 Variable Change A Overall P([??] at 0.584 mean Practice 1 (providing [DELTA] Range 0.02 information aboutgoals and performance) [DELTA] 1 0.01 SD[DELTA] 0.00 Practice 2 (offering rewards [DELTA] Range -0.13 based on performance) [DELTA] 1 -0.04 SDA -0.03 Practice 3 (providing access [DELTA] Range 0.26 to job-related knowledge and skills) [DELTA] 1 0.06 SD[DELTA] 0.05 Practice 4 (granting discretion [DELTA] Range 0.36 to change work processes) [DELTA] 1 0.11 SD[DELTA] 0.09 Rewards for innovation [DELTA] Range 0.35 [DELTA] 1 0.09 SD[DELTA] 0.10 Job satisfaction [DELTA] Range 0.06 [DELTA] 1 0.01 SD[DELTA] 0.01 Trust in leader [DELTA] Range 0.13 [DELTA] 1 0.02 SD[DELTA] 0.03 Note: (a) [bar.[DELTA]] Mean is the average absolute discrete change, (b) [DELTA] Range is change from the minimum to the maximum, (c) [DELTA]l is centered change of 1 around the mean, and (d) SD[DELTA] is centered marginal change around the mean. Encouragement to innovate (five-choice outcome): SD, "strongly disagree"; D, "disagree"; N, "neither agree nor disagree"; A, "agree"; and SA, "strongly agree." Table 5 Discrete Change in the Probability of Encouragement to Innovate for Multinomial Logit Regression, Six Two-Way Interactions (All Other Variables Held at Mean Values) Variable Change [bar.[DELTA]] Mean SD Practice 1 x practice 2 [DELTA]1 0.00 0.00 SD[DELTA] 0.00 0.00 Practice 1 x practice 3 [DELTA]1 0.02 0.00 SD[DELTA] 0.02 0.00 Practice 1 x practice 4 [DELTA]1 0.02 0.00 SD[DELTA] 0.01 0.00 Practice 2 x practice 3 [DELTA]1 0.02 0.00 SD[DELTA] 0.02 0.00 Practice 2 x practice 4 [DELTA]1 0.01 0.00 SD[DELTA] 0.01 0.00 Practice 3 x practice 4 [DELTA]1 0.02 0.00 SD[DELTA] 0.02 0.00 Variable Change D N SA A Practice 1 x practice 2 [DELTA]1 0.01 -0.01 0.00 0.00 SD[DELTA] 0.00 -0.01 0.00 0.00 Practice 1 x practice 3 [DELTA]1 0.00 -0.01 0.05 -0.04 SD[DELTA] 0.00 -0.01 0.05 -0.04 Practice 1 x practice 4 [DELTA]1 0.00 -0.04 0.03 0.01 SD[DELTA] 0.00 -0.03 0.02 0.01 Practice 2 x practice 3 [DELTA]1 -0.01 -0.02 0.04 -0.01 SD[DELTA] -0.01 -0.02 0.04 -0.01 Practice 2 x practice 4 [DELTA]1 -0.01 -0.02 0.01 0.01 SD[DELTA] 0.00 -0.02 0.01 0.01 Practice 3 x practice 4 [DELTA]1 -0.01 -0.04 0.03 0.02 SD[DELTA] -0.01 -0.03 0.03 0.02 Note: (a) [bar.[DELTA]] Mean is the average absolute discrete change, (b) [DELTA] Range is change from the minimum to the maximum, (c) [DELTA]1 is centered change of 1 around the mean, and (d) SD[DELTA] is centered marginal change around the mean. Encouragement to innovate (five-choice outcome): SD, "strongly disagree"; D, "disagree"; N, "neither agree nor disagree"; A, "agree"; and SA, "strongly agree."…