Improving Public Hospital Performance

Sungjoo Choi

Kennesaw State University

Hal G. Rainey

University of Georgia

Managing Diversity in U.S. Federal Agencies: Effects of Diversity and Diversity Management on Employee Perceptions of Organizational Performance

Current Trends in Public Personnel Administration

Sungjoo Choi is an assistant professor in the Department of Political Science at Kennesaw State University. She received her doctorate from the University of Georgia. Her research interests include diversity management, organizational justice, perfor- mance management in public organizations, and comparative public administration. E-mail: [email protected]

Hal G. Rainey is Alumni Foundation Distinguished Professor in the School of Public and International Affairs at the University of Georgia. His book Understand- ing and Managing Public Organizations was published in 2009. This year, he received the Dwight Waldo Award from the American Society for Public Administration.

E-mail: [email protected]

Diversity in the workplace is a central issue for contemporary organizational management. Concomitantly, managing increased diversity deserves greater concern in public, private, and nonprofit organizations. The authors address the effects of diversity and diversity management on employee perceptions of organizational performance in U.S. federal agencies

by developing measures of three variables: diversity, diversity management, and perceived organizational performance. Drawing from the Central Personnel Data File and the 2004 Federal Human Capital Survey, their findings suggest that racial diversity relates negatively

to organizational performance. When moderated by diversity management policies and practices and team processes, however, racial diversity correlates positively with organizational performance. Gender and age diversity and their interactions with contextual variables produce mixed results, suggesting that gender and age diversity reflect more complicated relationships. This article provides evidence for several benefits derived from effectively managing diversity.

The American workforce has been increasingly diversified by greater access to jobs for women and minorities. Diversity and representation “politically integrate a diverse nation with a measure of legitimacy” (Brewer 2002, 1), but also enhance social justice (Kellough 1990; Krislov and Rosen- bloom 1981). Understanding the impacts of diversity on organizational outcomes, such as organizational performance, employee satisfaction, and turnover, has become essential (Milliken

pursuing affirmative action programs to taking advan- tage of differences to improve organizational effective- ness (Wise and Tschirhart 2000). Recent research has investigated diversity in relation to organizational effectiveness.

Public organizations, through equal employment opportunity (EEO) and affirmative action programs, have been more committed to workforce diversity than have private organizations, resulting in a higher level of diversity in public organizations (Cornwell and Kellough 1994; Foldy 2004; Riccucci 2002). Public organizations thus face the challenge of managing a diversified workforce, a challenge that calls for more research on the impacts of diversity

on organizational effectiveness (Wise and Tschirhart 2000). In spite of the greater diversity in government organizations, more research in the field of business management has examined the effects of diversity

on organizational outcomes. Inconsistent empirical evidence regarding the effects of diversity on organiza- tional outcomes suggests a more complex relationship between diversity and organizational effectiveness than is often realized (Jackson, Joshi, and Erhardt 2003; Milliken and Martins 1996; Williams and O’Reilly 1998; Wise and Tschirhart 2000). Recent research indicates that mediating or moderating effects of contextual factors, such as organizational culture, and demographic characteristics of group members and supervisors, explain the inconsistent research results.

and Martins 1996). Cox (1993) contends that managing diversity should be significant as a moral imperative, as a legal requirement, and as a factor in organizational performance. In recent years, diversity scholars have reached consensus that the perspective on diversity manage- ment has broadened from

... [This] study shows how managerial efforts and other contextual variables moderate the relationship between diversity and organizational outcomes.

The present study shows how managerial efforts and other contextual variables moder- ate the relationship between diversity and organizational outcomes. First, this article reviews the literature on diversity in work groups and its impacts on organizational outcomes. Then, it discusses theoretical arguments about the

Managing Diversity in U.S. Federal Agencies 109

impacts of diversity and diversity management. Next, it examines the effects of diversity in race, gender, and age, and of moderators such as diversity management, organizational culture, and team processes on individual- and organizational-level outcomes using a hierarchical regression analysis. The last section discusses findings and implications.

Literature on Diversity

Early studies of diversity focused on its effects on individual and organizational outcomes, indicating that it provides a great oppor- tunity for organizations as well as a challenge (Milliken and Martins 1996; Tsui, Egan, and Xin 1995). Some studies concluded that more heterogeneous work groups consider more perspectives and produce higher-quality solutions than homogeneous groups (Cox, Lobel, and McLeod 1991; McLeod and Lobel 1992; Milliken and Martins 1996; Watson, Kumar, and Michaelsen 1993). Others con- tended that heterogeneous groups show lower levels of integration and have higher levels of dissatisfaction and turnover than homoge- neous groups (Jackson, Joshi, and Erhardt 1993; O’Reilly, Caldwell, and Barnett 1989).

Two perspectives have shaped research on the impacts of diversity (Williams and O’Reilly 1998). One perspective, based on infor- mation and decision-making theories, argues that diversity can benefit organizations by providing a broad range of ideas, skills, and insights that can improve organizational capabilities to solve prob- lems and make better decisions (Cox 1993; Cox and Blake 1991; Ely 2004). The other perspective, based on social categorization and social identity theories (Turner 1987) and the similarity-attraction paradigm (Byrne 1971), contends that diversity may burden organi- zations with high costs of coordination and conflict resolution, thus compromising organizational effectiveness, given that people tend to distinguish between in-group and out-group

that contextual factors, such as task characteristics, organizational culture, team processes, and strategic context, significantly affected the relationship between diversity and organizational performance or behavior.

Nevertheless, the effects of other variables, such as diversity man- agement practices and leadership, need much more attention from researchers. Previous research has used a limited range of samples, such as private firms in some fields of business or laboratory studies, thereby raising questions about external validity. Although some recent studies examined diversity’s impacts on the organizational effectiveness of public organizations, they also used samples drawn from limited areas of government. The present study addresses these limitations of previous research, focusing on effective management of increased diversity in organizations rather than on the direct effects of diversity.

The study also draws on a very large sample of actual federal employees. The results show that the management of diversity significantly moderates diversity’s impacts on important outcomes at individual and organizational levels. Some diversity scholars, including Cox (1993), Ely (2004), and Foldy (2004), have argued for the importance of effective diversity management. Pitts (2009) examined the effects of diversity management on federal employ- ees’ perceptions of organizational performance. However, his study relied entirely on survey evidence, and reported no evidence about the levels of diversity in the federal agencies in which the federal employees were located. Therefore, he could provide no evidence about the relationship between levels of diversity and diversity management. He did not show, as does our present analysis, that diversity management can actually reduce the potential negative impacts of increased levels of diversity and enhance positive impacts

members, which may cause conflicts and mis- communication (Ely 2004). Although both of these perspectives are supported by research in laboratory settings (Ely 2004; Williams and O’Reilly 1998), the external validity of these findings is questionable, suggesting that researchers should take the effects of contex- tual factors into consideration, and seek more sources of evidence outside of the laboratory.

We measure the levels of diversity of 67 federal agencies and analyze how diversity management affects the relationship between levels of diversity and perceptions of organizational performance.

of diversity, which is the goal of diversity management policies and practices. Pitts’s study found that federal employees, especially minority employees, perceived their agen- cies to be more effective when they perceived effective diversity management. The study we report here provides a more advanced analysis of the effects of diversity management. We measure the levels of diversity of 67 federal agencies and analyze how diversity manage- ment affects the relationship between levels of diversity and perceptions of organizational

Inconsistent empirical evidence on the impacts of diversity on organizations suggests complex relation- ships between diversity and its consequences. Recently, diversity researchers have developed models to examine the potential impacts of contextual factors that may moderate the relationship between diversity and its outcomes, such as human resources practices, team processes, and organizational culture (Jackson, Joshi, and Erhardt 2003; Kochan et al. 2003; Milliken and Martins 1996; Pitts 2005; Williams and O’Reilly 1998). Williams and O’Reilly (1998), in a review of 40 years of diversity research, proposed that contextual variables should be more carefully examined. Milliken and Martins (1996), in their evaluation of six years (1989–94) of research, iden- tified four types of mediating variables, which are the short-term consequences of diversity that influence the long-term outcomes of diversity. More recently, Jackson, Joshi, and Erhardt (2003) reviewed 63 studies published between 1997 and 2002 and found

performance. We need more evidence about how diversity manage- ment affects diversity’s impacts on organizational effectiveness. This study provides such evidence.

Theoretical Background and Hypotheses

Direct Effects of Diversity

Some proponents of diversity have maintained that workforce diver- sity enhances organizational effectiveness and productivity (e.g., Ely 2004; Hambrick and Mason 1982; McLeod and Lobel 1992; Richard 2000; Wiersema and Bantel 1992). Ely’s (2004) informa- tion and decision-making theories suggest that diversity improves performance by contributing to higher-quality decisions and by taking advantage of a broader range of alternatives and new ideas (Cox 1994; Foldy 2004; Ospina 2001; Richard 2000). Some studies

110 Public Administration Review • January | February 2010

have found that work teams composed of people with different backgrounds tended to share more information, leading to better performance than homogeneous teams (Bunderson and Sutcliffe 2002). Greater heterogeneity may lead to low consensus in decision making. Some researchers contend, however, that this can improve problem solving (Dutton and Duncan 1987; Katz 1982).

On the other hand, some research has reported a negative or insig- nificant relationship between diversity and organizational perfor- mance (e.g., Chatman and Flynn 2001; Chatman et al. 1998; Foldy 2004; Gladstein 1984; Jehn 1995; Jehn, Northcraft, and Neale 1999). Heterogeneous groups may experience problems in integra- tion, coordination, motivation, and conflict management (Gladstein 1984; Jehn 1995; Jehn, Northcraft, and Neale 1999). Consequently, diverse work groups typically take more energy to accomplish tasks than homogeneous work groups (Greenhaus, Parasuraman, and Wormley 1990; Kanter 1979; Schneider and Northcraft 1999). Research has indicated that individuals from diverse groups can experience exclusion from the group’s internal information networks and from important decision-making processes (Cox 1994; Ibarra 1993; Pettigrew and Martin 1989).

Similarity-attraction theory (Byrne 1971) and social categorization and social identity theories (Turner 1987) support negative perspec- tives on diversity in work groups because people prefer similarity in their interactions (Schneider 1987; Tsui, Egan, and O’Reilly 1992). Individuals feel comfortable in a group they perceive as similar to themselves. Individuals in diverse groups feel less safe and trust each other less. Lower trust increases conflict within groups. Theories

of selection and socialization view similarity in values and demo- graphic characteristics as conducive to effective work environments (Jehn, Northcraft, and Neale 1999).

Social categorization and social identity theories propose that people classify themselves and others into social categories based on social status and membership of social groups (Ashforth and Mael 2001; Tajfel and Turner 1985; Turner 1987). People in a group sharing the same social identity distinguish members of their own social category from others (Ely 2004; Kramer 1991). Such distinctions decrease interaction within or between groups from different social categories and leads to discrimination and self-segregation. Thus, some research finds that diversity is associated with less social inte- gration, more conflict, and less cohesion in groups (Milliken and Martins 1996; Williams and O’Reilly 1998). Diversity in organiza- tions is therefore likely to decrease organizational effectiveness.

H1: Higher levels of workforce diversity will decrease organi- zational performance.

Effects of Diversity Management

Effective diversity management enhances the effects of diversity on organizational performance. Research has shown mixed findings about the impacts of diversity in tenure, race, and sex on organi- zational performance. When properly managed, however, diversity can enhance performance (Cox, Lobel, and McLeod 1991; O’Reilly, Williams, and Barsade 1997). Human resource management poli- cies that promote diversity can enhance the diversity–performance relationship (Jehn and Bezrukova 2004; Richard 2000; Richard and Johnson 2001). An organizational culture and managerial attitudes

supporting diversity can also improve performance (e.g., Jehn and Bezrukova 2004; O’Reilly, Williams, and Barsade 1997).

The integration and learning perspective proposes that organizations should incorporate employees’ diverse perspectives into organiza- tional tasks and provide opportunities to learn from differences (Thomas and Ely 1996). Diversity provides a source of growth, learning, and intuition, but only when it is properly managed. Foldy (2004) contended that the integration and learning perspective (Thomas and Ely 1996, 2001) should link to high learning frames and behaviors (Argyris and Schön 1996). High learning frames and behaviors make three important contributions: (1) they support

the perspective that diversity is a source of learning, (2) they call for cultural diversity to be discussed openly, and (3) they ask every member in a group to embrace diversity (Foldy 2004).

Numerous scholars have emphasized the important role of manag- ers in diversity management. Thomas and Ely (1996) suggested that effective leadership connects diversity to work effectiveness. Effective leadership encourages a diverse workforce to appreciate different perspectives and to value diverse opinions and ideas (Thomas and Ely 1996).

H2a: Effective diversity management will increase organiza- tional performance.

H2b: Diversity management will moderate the impact of diversity on organizational performance. In agencies that have more effective diversity management, higher levels of diversity will increase perceived organizational performance. In agencies that have less effective diversity management, higher levels of diversity will decrease perceived organizational performance.

Organizational Tenure

Pfeffer (1983, 323) asserted that organizations will be most pro- ductive when employees have organizational tenure “long enough to overcome some initial naiveté and learn the ropes and local practices.” As diverse teams work together longer, teamwork and collaboration increase. Enhanced intragroup contact will reduce the negative effects of social categorization (Chatman and Flynn 2000; Ely 2004; Harrison et al. 2002). The negative effects of surface-level diversity, such as demographic differences, become less important, reducing prejudices and stereotyping. Katz (1982) argued that longer organizational tenure stabilizes the organization, reducing goal conflict. Decreased negative effects of diversity imply that organizations can invest more in the improvement of organizational performance, while saving resources that would have been expended for coordination, conflict management, and control.

H3a: Longer organizational tenure will increase organizational performance.

H3b: Organizational tenure will moderate the impact of di- versity on organizational performance such that, for agencies that have longer average tenure of employees, higher levels of diversity will increase organizational performance. For agen- cies that have shorter average tenure, higher levels of diversity will decrease organizational performance.

Managing Diversity in U.S. Federal Agencies 111

Effect of Team Processes

Individual differences may lead to excessive conflict, so it is neces- sary to develop ways to counteract these effects. Relational conflicts, when not properly treated, decrease team members’ satisfaction (De Dreu and Weingart 2003). Team processes can moderate the relationship between diversity and conflict, so that cooperation

and communication among team members mitigate the negative effect of diversity (e.g., Ely 2004; Mohammed and Angell 2004). Mohammed and Angell (2004, 1021) argued that team processes enhance interactions between team members (Marks, Mathieu, and Zaccaro 2001) and the synergistic combination of individual efforts (Kozlowski and Bell 2003). Team processes should moderate the negative effects of diversity on organizational performance by reducing conflict.

H4a: Effective team processes will increase organizational performance.

H4b: Effective team processes will moderate the impact of di- versity on organizational performance, such that for agencies that have more effective team processes, diversity will increase organizational performance. For agencies that have less ef- fective team processes, diversity will decrease organizational performance.

Effect of Organizational Culture

Organizational culture is defined as “a common set of shared mean- ings or understandings about an organization (e.g., strongly held values, and norms about patterns of group members’ behaviors)” (Chatman and Jehn 1994; Jehn and Bezrukova 2004, 705–6; Reichers and Schneider 1990; Triandis and Suh 2002). Organi- zational culture may affect the relationship between diversity and performance by imbuing group members with the organization’s perspectives on diversity (Jehn and Bezrukova 2004).

A results-oriented organizational culture emphasizes individual accomplishment, usually leading to high competition among mem- bers in the organization. High competition may prevent individuals from working efficiently, leading to inefficiency (Jehn and Mannix 2001; Tsui, Egan, and O’Reilly 1992). Furthermore, the organiza- tion may lose opportunities to benefit from cooperation because of the competitive culture. Diversity in results-oriented organizational culture may decrease organizational performance.

H5a: Results-oriented organizational culture will increase organizational performance.

(CPDF) and the 2004 Federal Human Capital Survey (FHCS), both published by the U.S. Office of Personnel Management.1 The diversity measures for federal agencies were created using information from the CPDF in 2004. Index variables were developed using the questionnaire of the 2004 FHCS. A total of 67 subagencies under executive branch agencies and inde- pendent agencies participated in the 2004 FHCS, and approxi- mately 150,000 federal government employees responded to the survey.

Diversity Measure

Three kinds of diversity measures, including the coefficient of variation, the Blau index of heterogeneity, and the entropy index of diversity, are the most frequently used measures. Diversity researchers use different measures depending on the characteris- tics of variables. The Blau index of heterogeneity and the entropy index are appropriate for categorical variables such as gender, race/ethnicity, and functional and educational background, while the coefficient of variation is used for continuous variables such as age and tenure. The present study employs the entropy index of diversity to measure race, gender, and age diversity, which are categorical variables in this study. The entropy index of diversity, or the Shannon index (uncertainty index), is considered the appro- priate measure of diversity (Miller and Quigley 1990; Pielou 1977; Reardon 1998). The entropy index is computed with the following formula:

E = ∑ Qr logn (1/Qr) (r = 1, . . . , n),

where Qr denotes the proportion of the population of group r. The entropy index of diversity ranges from a minimum value of 0, which means that the population contains a single group, to a maximum value of 1, which means that all n groups are evenly represented in the population.

The indices of each agency’s attributes of diversity—race, gender, and age—are calculated using the entropy index of diversity.2 Our analysis classifies federal employees into five racial/ethnic groups— American Indian or Alaska Native, Asian or Pacific Islander, blacks, Hispanics, and whites—based on the CPDF’s categorization. Gender of federal employees is composed of two groups: male and female. The age variable is categorized into five groups: under 29, 30–39, 40–49, 50–59, and over 60.

Dependent Variables

The appendix presents the questionnaire items for the scales constructed for this study. The measure of perceived organiza-

H5b: Results-oriented organizational cul- ture will moderate the impact of diversity on organizational performance, such that for agencies that emphasize results-oriented culture, diversity will decrease organiza- tional performance.

Data and Methods

Data Sources and Sample

The present study uses data from two major sources: the Central Personnel Data File

The present study uses data from two major sources: the Central Personnel Data File (CPDF) and the 2004 Federal Human Capital Survey (FHCS), both published by the U.S. Office of Personnel Management.

tional performance was developed by using principal components factor analysis with varimax rotation. It includes survey questions about the quality of work, organizational performance, managerial capability, and job-relevant resources and skills. Respondents rated these four questions on a Likert scale from 1 (strongly disagree or very poor) to 5 (strongly agree or very good). Factor analysis of these items shows that factor loadings range between 0.715 and 0.794. The scale’s eigen- value is 2.280 and Cronbach’s alpha is 0.738.

112 Public Administration Review • January | February 2010

Main Independent Variable

We measure diversity management in two ways. One method mea- sures the diversity management of agencies according to employees’ perceptions. This study develops an index variable through factor analysis, using three survey questions from the 2004 FHCS, that ask about commitment to diversity and policies and practices to promote diversity (see the appendix). Principal components factor analysis and varimax rotation produce a single factor on which these three items loaded. The initial eigenvalue of the scale is 2.308 and the Cronbach’s alpha is 0.850. Factor loadings range between 0.868 and 0.886.

The other way of measuring the effectiveness of diversity manage- ment uses the number of EEO complaints in an agency as a proxy measure of effective diversity management. The Notification and Federal Employee Antidiscrimination and Retaliation (No FEAR) Act of 2002 requires every federal agency to report complaints about violations of antidiscrimination and whistle-blower protec- tion laws on a quarterly basis.4 Under section 301 of the No FEAR Act, each federal agency must post summary statistical data pertain- ing to complaints of employment discrimination filed against it

by employees, former employees, and applicants for employment under 29 C.F.R. Part 1614 (EEOC 2005). We assume that the effectiveness of an agency’s diversity management relates negatively to complaints reported, and that a lower number of EEO com- plaints represents more effective diversity management. The data come from the 2004 annual report posted on each agency’s public Web site. The measure divides the number of EEO complaints by organizational size.

Control Variables

Performance-oriented organizational culture. A results-oriented organizational culture emphasizes employees’ performance and productivity by rewarding high performance and penalizing poor performance. Such a culture emphasizes merit-based promotions, rewards, or penalties based on performance evaluations. The appen- dix shows the eight items that are integrated into a single factor, with factor loadings between 0.685 and 0.888, through principal factor analysis and varimax rotation. The initial eigenvalue of the scale is 5.736 and the Cronbach’s alpha is 0.942.

Team processes. This variable measures cooperation and communi- cation in work units. Two questions ask about employees’ coopera- tion to get the job done in work groups and about information and knowledge sharing among employees through communication. They were rated on a five-point Likert scale. These items are also integrated into one index variable through factor analysis. The initial eigenvalue is 1.462 and the Cronbach’s alpha is 0.626.

Average tenure. Average tenure represents the average tenure of full-time employees hired in an agency. This variable is measured for each agency based on the 2004 CPDF. The mean average tenure of federal agencies is 15.1 years, ranging between 7.9 and 22.6 years.

Demographic variables. Demographic variables including gender, minority, and supervisory status are recorded as dummy variables in order to control their effects on dependent variables. Previous research has shown that differences in gender, minority status, and supervisory status affect perceptions of organizational performance. We control for the impacts of these demographic characteristics

on dependent variables. The gender variable is coded 1 when a respondent is a female and 0 for a male. The minority variable is 0 when a respondent is a white male or a white female and is 1 when a respondent is neither a white male nor a white female. The CPDF defines American Indian or Alaska Native, Asian or Pacific Islander, black (not of Hispanic origin), and Hispanic as the minority race/ national origin groups for federal statistics and program adminis- trative reporting (OPM 2006). Supervisory status is classified into two types—supervisor or manager and executive—represented by two dummy variables. For a supervisor or a manager, the supervisor variable is recorded as 1, and when a respondent is an executive, the executive variable is recorded as 1. Otherwise, it is recorded as 0.

Methodology and Model Specification

We present hierarchical regressions to test a model of the effects of diversity. Moderated multiple regression involves hierarchical regression that first tests the relationship between the predictors (independent variables) of interest and the criterion variable (depen- dent variable), and then tests the relationship of a term that carries information about both predictors (the interaction term). The “hier- archical” form of regression indicates that predictors are not entered

Table 1 Descriptive Statistics Variables Mean Size 11.23

Std. Dev. 1.83 2.72 0.50 0.45 0.18 0.46 0.08 0.04 0.03 1 0.28 1 1

Min 6.22 7.9

0 0 0 0

0.21 0.81 0.73

–3.07 0.15 –3.96 –2.20

Max Unit 13.39 Natural log of the number of employees 22.6 Year

1 Female = 1, male = 0 1 Supervisor or manager = 1 1 Executive = 1 1 Minority = 1

0.77 Entropy index 1 Entropy index 0.96 Entropy index

1.44 Factor score 1.81 Proportion of the number of EEO complaints 1.28 Factor score 1.87 Factor score

Organizational tenure Female Supervisor Executive

Minority Race diversity Gender diversity Age diversity Diversity management EEO complaints Team processes Organizational culture

15.11 0.43 0.28 0.03 0.30 0.56 0.95 0.89 0 0.60 0 0

Managing Diversity in U.S. Federal Agencies 113

into the regression equation simultaneously, but in a logical order. Typically, the continuous predictor and the polychotomous predic- tor are entered in the first step, and the interaction term is entered in the second step (Aiken and West 1991). In addition, researchers decide how many predictors to enter and the order in which they are entered. Determining the order of entry is based on logical or theoretical considerations. F-tests are used to compute the signifi- cance of each added variable (or set of variables) to the explanation reflected in R2 (Garson 1998).

We test the main and moderating effects of diversity and diversity management on organizational performance using this method. Some researchers (e.g., Allison 1977; Friedrich 1982; Smith and Sasaki 1979; Wright 1976) have expressed the concern that because a multiplicative term is often highly correlated with its constitu- ent variables, the inclusion of the multiplicative term in a model with its constituent variables leads to multicollinearity, which yields unreliable and unstable coefficients. In fact, when we tested our model using the ordinary least squares method, we detected such

a problem in the interaction terms of our model. For this reason, many studies in organizational behavior and social psychology (e.g., Aiken and West 1991; Cohen 1978, 861; Cohen and Cohen 1983; Simons and Peterson 2000) have used a hierarchical regression procedure to resolve the problem, in which the effect of a multipli- cative term on a dependent variable should be assessed only after the effects of its constituent variables on the dependent variable have been partialed out (Friedrich 1982, 802).

Further, larger correlations between predictors can also increase multicollinearity. When the predictors are highly correlated, a statistically significant interaction term may result from a nonlinear multiplicative effect, and not because of a linear multiplicative effect (Friedrich 1982). Nevertheless, correlations between predictors that compose interaction terms are not substantially high, suggesting that multicollinearity related to interaction terms may not be a criti- cal problem in this model.

The model of diversity and organizational performance categorizes the independent variables into eight sets and enters them in the following order: demographic control variables (step 1), the main effects of diversity (step 2), the main effects of diversity manage- ment (step 3), the main effects of team processes and organizational culture (step 4), the moderating effects of diversity management (step 5), the moderating effects of tenure (step 6), the moderat-

ing effects of team processes (step 7), and the moderating effects of organizational culture (step 8). The model of organizational performance is specified as follows:

Step 1: Pi = + Ci + ei 01

Step 2: Pi = 0 + 1Ci + 2Di + ei Step 3: Pi = 0 + 1Ci + 2Di + 3Mi + 4Ei + ei

Step 4: Pi = 0 +1Ci+2Di+3Mi+4Ei+5Ci+6Ti+7Ri + ei

Step 5: Pi = 0 +1Ci+2Di+3Mi+4Ei+5Ci+6Ti+7Ri + 8Di * Mi + 9Di * Ei + ei

Step 6: Pi = 0 +1Ci+2Di+3Mi+4Ei+5Ci+6Ti+7Ri + 8Di * Mi + 9Di * Ei +

10 * Yi + ei 114 Public Administration Review • January | February 2010

Step 7: Pi = 0 + 1Ci + 2Di + 3Mi + 4Ei + 5Ci + 6Ti + 7Ri +8Di*Mi+9Di*Ei+

10 *Yi+11Di*Ti+ei Step 8: Pi = 0 + 1Ci + 2Di + 3Mi + 4Ei + 5Ci + 6Ti + 7Ri

+ 8Di * Mi + 9Di * Ei + 10 *Yi+11Di*Ti+8Di*Ri+ei

where Pi is perceived organizational performance; Ci is the vector of organizational contextual variables (organizational size, average tenure of employees) and demographic characteristics of employees (female, minority status, and supervisory status); Di is the vector of attributes of diversity (race diversity, gender diversity, and age diversity); Yi is organizational tenure; Ti represents team processes; Ri denotes results-oriented organizational culture; Mi is perceived diversity management;

Ei is the percentage of reported EEO complaints; and ei is an error term.

Results

The main effects of diversity, diversity management, and other contextual variables and moderating effects on organizational per- formance were examined through the eight steps of the hierarchical regression analysis. The model fit of each of the eight blocks is good and the complete model explains 57 percent of the variation in the dependent variable.

Diagnostic tests for normality, linearity, multicollinearity, homosce- dasticity, and the existence of outliers did not indicate serious violations of the assumptions for the ordinary least squares regres- sion, except for one minor problem. Residual analyses detected one outlier, but it does not appear to influence the relationship signifi- cantly. Table 3 describes the correlations of variables. The largest correlation is 0.66, and the average and median correlations are 0.03 and 0.01, respectively. The variance inflation factor (VIF) test for multicollinearity revealed that the largest VIF value and the average were 1.95 and 1.40 respectively, much lower than the typical cutoff point of 3.5.

Direct Effects of Diversity (H 1 ) Only the effect of racial diversity on perceived organizational performance (hereafter, organizational performance) was supported. Racial diversity (–.42, p < .01) is negatively related to organizational performance, in support of hypothesis 1, suggesting that in agencies that have higher levels of racial diversity, organizational performance

Table 2 Summary of Model Fitç Block F R2

Change in R2

.0015 .3132 .2373 .0003 .0001 .0001 .0001

1 224.09*** 2 39.01*** 3 18090.81*** 4 21247.24*** 5 8.67*** 6 7.00*** 7 3.85** 8 5.15***

.0171 .0186 .3319 .5691 .5694 .5695 .5696 .5697

** Significant at .05 level; *** significant at .01 level.

Table 3 Correlations of Variables

1 2 3 4 5 6 7 8 9 10 11 12 13

1.Organizational performance 2. Race diversity –.05 3. Gender diversity .01 .14

4. Age diversity –.01 5. Size .02 6. Tenure .02 7. Gender –.00a 8. Minority –.02 9. Supervisor .08

10. Executive .08 11. Diversity management .57 12. EEO complaints –.05 13. Team processes .63 14. Organizational culture .66

Note: all of the coefficients, except those

.09 .27 –.16 –.35 –.13 .01

.09 .12

–.09 –.45 –.19

.01 –.02 .05 .01 –.01 –.09 –.01

–.14 .04

.03 .19 –.14 –.14 –.07

.02 –.06 –.06 –.12 .03 –.09 –.19 .15 .11 .11 .10 .11 –.04 –.05 –.06 .01 –.06 –.07 .11 .07 .49 –.04 .04 –.06 –.04 .21 .15 .64 –.07 .52

was perceived lower than that in other agencies. On the other hand, gender diversity (.31, p < .01), which indicates a better balance of the two sexes, is positively related to organizational performance. In agencies where the proportions of male and female employees are similar, organizational performance was perceived higher than it was in others. The effect of age diversity on organizational performance was not significant.

Findings supported our hypotheses of the effects of diversity man- agement on organizational performance. We developed the two measures of diversity management—perceived diversity manage- ment and EEO complaints, which are negatively related to each other. Higher levels of perceived diversity management indicate that the agency manages workforce diversity more effectively, while higher levels of EEO complaints reported to the agency’s EEO office mean that the agency manages diversity less effec- tively. Supporting hypothesis 3a, perceived diversity management (.58, p < .01) is positively related to organizational performance, while EEO complaints (–.031, p < .01) are negatively related to organizational performance. In other words, when agencies have more effective diversity management policies and practices, they also perform better than others that are less effective at managing diversity.

The results of the moderating effects of diversity management are complicated to interpret. Perceived diversity management (.13, p < .01) positively moderated the relationship between racial diver- sity and organizational performance, suggesting that in agencies that effectively manage diversity, racial diversity is associated with increased perceived organizational performance. This indicates that although higher levels of racial diversity will decrease organizational performance, effective management will turn negative effects on organizational performance into positive effects. However, the mod- erating effects of perceived diversity management on the other two types of diversity are found to be insignificant, in partial support of hypothesis 2b.

Hypothesis 2b also proposed that the effects of EEO complaints moderate the relationships between the attributes of diversity and organizational performance such that less effective diversity management will enhance the negative effects of diversity on orga- nizational performance. Interestingly, in the agencies with higher levels of EEO complaints, which indicates poor management for diversity, higher levels of gender diversity (.85, p < .01) had a posi- tive association with organizational performance, while racial (–.52, p < .01) and age diversity (–1.34, p < .01) related negatively to performance.

Hypothesis 3a, which examined the effect of employees’ organiza- tional tenure on organizational performance, was not supported. The findings of the moderating effect of tenure were also not consistent with our hypothesis 3b. We predicted that organizational tenure would positively moderate the association between diver- sity and organizational performance. However, the results indicate that higher levels of racial diversity are negatively associated with perceived organizational performance when employees of the agency have longer organizational tenure. In contrast, higher levels of gen- der diversity tend to increase organizational performance when the employees worked together longer. The moderating effect of tenure on age diversity was not significant.

The effects of team processes and results-oriented organizational culture on organizational performance supported hypotheses 4a 5a. Both of these two variables were positively associated with organiza- tional performance (.37, p < .01 and .38, p < .01, respectively).

Hypothesis 4b predicted that team processes would positively moderate the associations of the three attributes of diversity and organizational performance. Team processes showed a pattern of moderating effects similar to that of perceived diversity manage- ment. The moderating effects of team processes (.081, p < .05) were significant only in the relationship between racial diversity and organizational performance, suggesting that in agencies effectively

marked by

“a,” are statistically significant at the .01 or .001 levels.

Managing Diversity in U.S. Federal Agencies 115

practicing teamwork and cooperation among members in work- groups, higher levels of racial diversity increased organizational performance.

Hypothesis 5b proposed that organizational culture should negatively moderate the relationship between diversity and orga- nizational performance. However, the moderating effects of organizational culture on the relationship between diversity and organizational performance were mixed. The results showed that in agencies with high results-oriented culture, higher age diversity increased organizational performance, while the moderating effects of organizational culture on racial and gender diversity were insignificant.

cohesion in groups, consequently decreasing organizational perfor- mance (Milliken and Martins 1996; Williams and O’Reilly 1998). In contrast, higher levels of racial diversity decreased perceived organizational performance, while higher levels of gender diver- sity increased perceived organizational performance. One possible interpretation is that gender may lead to fewer conflicts than do other types of diversity, thus boosting the positive effects of diversity on organizational performance (Pelled, Eisenhardt, and Xin 1999). Gender diversity may be a less sensitive issue in work groups, as compared to racial diversity, and may require lower costs for conflict resolution or coordination. On the other hand, work groups with

higher levels of racial diversity may experience more conflicts because of complicated dif- ferences arising from residual effects of racial injustice and related factors.

The moderating effects of diversity manage- ment on the relationship between diversity and organizational performance contribute the most interesting findings of this study. Although policy makers and public managers may understand the importance of workforce diversity very well, they may not realize the crucial need for effective management prac- tices to enhance the positive impacts of diver- sity on organizational performance. This study developed two measures of diversity manage- ment—a subjective measure and an objective measure. Even though the results of these two measures did not exactly match, the pattern of the impacts is consistent. The findings suggest that the effects of diversity management can enhance or even reverse the main impacts of diversity on organizational outcomes. When an agency’s employees perceived that leaders managed racial diversity effectively, higher levels of racial diversity improved perceived organizational performance. In contrast, higher levels of racial diversity in agencies that

were not successful in managing diversity significantly decreased perceptions of the agency’s performance. These results support the integration and learning perspective on diversity, which proposes that if organizations properly manage employees’ varied perspec- tives, diversity can be a source of growth, learning, and intuition, thus enhancing organizational performance (Foldy 2004; Thomas and Ely 1996, 2001). The results are also consistent with Cox

and Blake’s (1991) conclusion that managing diversity can bring a number of benefits to organizations by improving creativity and problem-solving capacity and reducing turnover.

Along similar lines, when an agency manages diversity poorly, higher levels of age diversity decrease organizational performance. On the other hand, the moderating effects of diversity management on the relationship between gender diversity and organizational performance appear to be more complicated to interpret. In agencies that manage gender diversity poorly, higher levels of gender diversity increased organizational performance. As mentioned previously, gender diver- sity may not cause high costs for conflict resolution and coordination, thereby not requiring managerial capacity or efforts for diversity.

Results of Control Variables

Female employees, supervisors, and Senior Executive Service members perceive higher organizational performance, while racial minority employees perceive lower organiza- tional performance. However, organizational tenure of employees and organizational size were not significantly associated with organi- zational performance.

Discussion and Conclusion

Previous empirical research on the effects of diversity on organizational performance has found mixed results. Recently, researchers have reported that the actual effects of diversity under complex organizational conditions are much more complicated than they had pre- dicted, suggesting the need for a more nuanced interpretation of diversity and for research that analyzes the impacts of diversity taking into account the complex reality of organizations (Bell and Berry 2007; Klein and Harrison 2007). This study responds to these needs, ana- lyzing not only the main impacts of diversity but also moderating effects of contextual fac- tors on organizational performance. The results of this study add to recent research that indicates the importance of contextual variables for understanding impacts of diversity on organizational outcomes (Jackson, Joshi, and Erhardt 2003; Kochan et al. 2003; Milliken and Martins 1996; Williams and O’Reilly 1998). The results suggest that contextual variables significantly affect the relationship between diversity and organizational performance. The most important evidence of such moderating effects comes from the finding that this very large sample of federal employees perceives lower organizational effectiveness where racial diversity is higher, but strikingly, where they perceive that diversity is effectively managed, perceived organizational effectiveness is higher when racial diversity is greater.

The findings suggest that the main effects of diversity may vary according to the types of diversity. This is consistent with the findings of previous research (Sackett and DuBois 1991; Watson, Kumar, and Michaelsen 1993; Wise and Tschirhart 2000). The main effect of racial diversity was consistent with social categori- zation and social identity theories that propose greater diversity is associated with less social integration, more conflict, and less

Recently, researchers have reported that the actual effects of diversity under complex organizational conditions are much more complicated than they had predicted, suggesting the need for a more nuanced interpretation of diversity and for research that analyzes the impacts of diversity taking into account the complex reality of organizations.

The moderating effects of diversity management on the relationship between diversity and organizational performance contribute the most interesting findings of this study.

Donald P. Moynihan

University of Wisconsin

Noel Landuyt

University of Texas

How Do Public Organizations Learn? Bridging Cultural and Structural Perspectives

New Ideas for Improving Public Administration

Donald P. Moynihan is an associate professor and Romnes Fellow in the La Fol- lette School of Public Affairs, University of Wisconsin–Madison. His research examines the application of organization theory to public management issues such as admin- istrative reform, performance management, and employee behavior. His book The Dynamics of Performance Management won the 2009 Academy of Management award for best public and nonprofit book. E-mail: [email protected]

Noel Landuyt is a research associate and lecturer at the University of Texas at Austin in the Center for Social Work Research. He is also the director of the Institute for Organizational Excellence. E-mail: [email protected]

How do public organizations learn? The organizational learning literature suggests distinct cultural and structural routes to learning. However, such categorizations oversimplify. Leaders seeking to foster learning should recognize that most relevant organizational variables combine structural and cultural aspects, which are mutually dependent on one another. The strongest influences are the existence of work groups that are purpose driven and incorporate the views of all members, including dissenting views. Such learning forums can

be fostered through formal requirements, but they need appropriate cultural characteristics to succeed. Mission orientation, decision authority, information systems, and resource adequacy are also positively related to improved organizational learning.

Many of the criticisms leveled at public organ- izations imply a failure to use information and experience to make better decisions— in short, a failure of organizational learning. There is relatively little research done on organizational learn- ing in the public sector. This is surprising, because the concept of organizational learning—that organizations can improve if organizational actors identify and use information to improve actions—is the key assump- tion that underpins much of contemporary public management reform, such as total quality manage- ment, reengineering, benchmarking, performance management, and performance budgeting.

This article attempts to go beyond the question, can public sector organizations learn? (Barrados and Mayne 2003). Generally, those who

pose this question accept that public organizations can and should learn. Instead, we ask, how do public organiza- tions learn? Existing empirical research relies largely on case studies, which have been invalu- able in identifying examples of learning and in suggesting

the characteristics of learning organizations. But the question of how learning can be engineered remains in dispute. The debate falls essentially along two lines, explored in the next section of the article. One approach argues that learning is something that emerges from the culture of the organization. By contrast, structural proponents argue that the cultural viewpoint underestimates the extent to which formal procedures can be used to foster learning.

This article uses a sample of Texas state employees to test which variables foster organizational learn- ing. There have been few tests of the antecedents of learning using large-scale quantitative data sets (Vince, Sutcliffe, and Olivera 2002), particularly in public settings. This may be because a survey approach is

not adept at identifying processes by which organiza- tions store knowledge. However, it does allow us to understand what factors are associated with individual perceptions of whether learning is occurring in the workplace.

The findings suggest that both structural and cultural approaches are important, and are intertwined with one another to a degree that undercuts the claim that they are distinct approaches to learning. Many of the variables that explain learning clearly involve both structure and culture. While the dichotomy

of structure versus culture is heuristically appealing, it obfuscates a messier reality. We argue that bridg- ing the cultural and structural perspectives requires

treating them as connected and interdependent factors. To make this argument clearer, we frame our findings using assump- tions of structuration theory (Giddens 1984), which treats structure and culture as part of the broader social forces that en- able and constrain social action, but are, in turn, reshaped by human agency. This suggests the possibilities for leaders to foster

This article attempts to go beyond the question, can public sector organizations learn?. . . . Generally, those who pose this question accept that public organizations can and should learn. Instead, we ask, how do public organizations learn?

How Do Public Organizations Learn? 1097

learning, and we consider the practical implications of our findings before we conclude.

Structural and Cultural Approaches to Learning

Organizational learning comes from the ability of organizational actors to relate experience and information to routines and problems (Argyris and Schön 1996, 16; Mahler 1997, 519). We follow Ber- ends, Boersma, and Weggeman’s definition of organizational learning as “the development of knowledge held by organizational members, that is being accepted as knowledge and is applicable in organization- al activities, therewith implying a (potential) change in those activi- ties” (2003, 1042). Therefore, we treat learning in terms of individual perceptions of the deployment of knowledge in a group setting.

A tension in the organizational learning literature is whether learn- ing is fostered by culture or structure. Cook and Yanow (1993) propose what they describe as a cultural perspective on learning, where learning derives from the intersubjective meaning experienced by organizational actors and is reflected largely in tacit knowledge rather than explicitly considered decisions. While Cook and Yanow’s approach seeks to distance itself from mainstream organizational learning, the emphasis on culture as critical

The structural approach is sharper in its criticism of the explicitly cultural perspective on learning and argues that authors such as Cook and Yanow exclude the role of individual cognition. Struc- turalists agree with Simon, who claims that “all learning takes place inside individual human heads; an organization learns in only

two ways: a) the learning of its members, or b) by ingesting new members who have knowledge the organization previously didn’t have” (1991, 125). Simon leaves open the question of how such an individual’s knowledge becomes organizational. For structural- ists, therefore, the key challenge for organizational learning is not to study the intersubjective meanings of individuals, but to study the structural procedures by which individual learning is acquired and utilized for organizational purposes. Lipshitz, Popper, and Oz (1996) describe such procedures as organizational learning mecha- nisms (OLMs). From the structuralist standpoint, OLMs provide a necessary (though not sufficient) basis for organizational learning (Lipshitz, Popper, and Friedman 2002).

A central point of this article is that an unquestioning acceptance of the cultural/structural dichotomy can lead to the exclusion of reality in the name of parsimony. Many of the features of organizational

to learning is not a dramatic break. The most influential writings on organizational learning also assume that learning is facilitated through shared norms, and that some cultures will be more conducive to learning than others. For instance, Senge (1990) argues that learning is based on shared experiences, norms, and un- derstandings that foster intelligent behavior.

What cultural attributes foster or discourage learning? Characteristics of a learning culture include high employee empowerment, partici- pation, and organizational openness (Argyris and Schön 1996; Fiol and Lyles 1985; Hult

Scholarly efforts that attempt to divide all of the antecedents of learning into structural or cultural variables will misdiagnose the causal mechanisms of learning by underestimating the importance of culture to what are classified as structural variables, and the importance of structure to variables deemed to be cultural.

life—including variables that affect learning— simultaneously feature both cultural and structural components. Scholarly efforts that attempt to divide all of the antecedents of learning into structural or cultural variables will misdiagnose the causal mechanisms of learning by underestimating the importance of culture to what are classified as structural variables, and the importance of structure to variables deemed to be cultural.

Our model is presented in figure 1. Some variables, such as information systems and resources, reflect largely structural influ- ences on learning. However, other variables demonstrate the lack of a clear distinction

et al. 2000). On the other hand, defensive norms and the existence of taboo subjects weaken the capacity to learn. Overcoming defensiveness depends on a high measure of trust among employees and an understanding that an acknowledgment of error will not be used for punitive purposes. Once a nonpunitive environment of inquiry is established, employees are expected to consistently seek new approaches to provide better services. Cook and Yanow (1993) do not disagree with such assertions, but go fur- ther—culture does not just support learning, it is the primary means through which learning is transferred. A true study of organizational learning, therefore, focuses on collective meaning rather than indi- vidual learning (Yanow 2000).

A structural viewpoint, represented most clearly by Lipshitz, Popper, and Oz, complains that the organizational learning literature “offers relatively few treatments of the problem of how to build learning organizations” (1996, 301). The attributes that make culture an explanatory variable—its embeddedness in all aspects of organiza- tional life—also make culture a difficult constraint to change. The structural criticism of mainstream organizational learning is not that culture is irrelevant, but that too little attention has been given to more immediate formal mechanisms that can foster organizational learning.

between structure and culture. Formal rules can try to establish a clear understanding of purpose and empower managers, but mission orientation and patterns of decision authority also rely on comple- mentary cultural norms. Learning forums can be created by formal rules, but only the appropriate cultural traits, such as a willingness to acknowledge error and entertain the views of others, can ensure the success of such forums.

Resources

Resources are a constraint formally created by actors outside the public organization, and so they can be considered a structural variable. We hypothesize that when an organization has adequate

Information Systems

Mission Orientation

Organizational Learning

Adequacy of Resources

Decision Flexibility

1098 Public Administration Review • November | December 2009

Figure 1 Learning

A Structural-Cultural Model of Organizational

Learning Forums

resources, it is more likely to learn. This runs contrary to the “neces- sity is the mother of invention” argument, which suggests that tight budgets will force organizations to come up with more innovative approaches. But agencies that are consistently underfunded are likely to have already wrung out any inefficiencies. Such organiza- tions are in a reactive situation and focus on coping with the prob- lems created by low resources and finding ways to increase the flow of resources. When organizations have some measure of organiza- tional slack, they are more likely to be able to think proactively, and devote specialized resources and time to learning. Both quantitative (Askim, Johnsen, and Christophersen 2008; De Lancer Julnes and Holzer 2001) and qualitative (Berends, Boersma, and Weggeman 2003; Moynihan 2008) research that examines the link between resources and learning supports this claim.

H1: Adequacy of resources is positively related to organiza- tional learning.

Information Systems

What do OLMs as a structural approach to learning actually look like? Lipshitz, Popper, and Oz describe OLMs as “institutionalized structural and procedural arrangements that allow organizations to systematically collect, analyze, store, disseminate, and use informa- tion that is relevant to the effectiveness of the organization” (1996, 293). In applying the learning model to the public setting, Moyni- han (2005) argues that performance information systems that col- lect, store, and disseminate information represent a common OLM in the public sector.

Performance information systems often fail to generate valid, legitimate, and functional performance information (Bouckaert 1993), or fail to distribute this information in a timely fashion to the right audience. These are not modest problems. A formal requirement for a performance information system is a poor guarantee of learning. For example, Argyris and Kaplan (1994) describe how the introduction of activity-based costing systems had a weak impact on learning. This makes testing the impact of information systems difficult. If one were to test the relationship between the existence of a performance information system and learning, and find no correlation, the interpretation could be either that (1) the information system does not matter to learning, or (2) the information system was not effective. To eliminate the latter interpretation from our model, we measure effective information systems, that is, information systems that appear to meet the basic criteria of collecting and diffusing relevant information in a timely fashion to the target users.

H2: Effective information systems are positively related to organizational learning.

Mission Orientation

Relative to performance information systems, the employee’s mis- sion orientation—their understanding of the mission, vision, goals of the organization—is a more cultural aspect of performance man- agement. A mission orientation might be the product of structures of strategic planning and communication, but it also reflects the compatibility of an organizational culture with mission and goals. When a mission-based culture exists, employee behavior is guided by shared norms and assumptions about the organizational purpose.

Mission orientation overlaps with two aspects of Senge’s (1990) model of learning organizations. The first is building a shared vision, where employees become committed and align their actions to the organizational vision. The second is systems thinking. Senge argues that if employees understand the broader system of which they are a part, they are in a stronger position to place their actions in this broader context, and to learn in a way that contributes to the whole. If individuals perceive a high measure of ambiguity in their environment, this will frustrate systems thinking and make learning more difficult. Mahler (1997) suggests that when organiza- tions deal with a high measure of uncertainty, learning is limited. When employees understand the mission and do not face goal ambiguity, they are more likely to engage in learning (Kaplan and Norton 1996).

.

Decision Flexibility

Recent public management reform has called for empowering employees to make management decisions, on the assumption that the employees closest to the work best understand how to foster improvement (Moynihan 2008). But a mismatch between knowl- edge and authority can affect learning. Operational staff with the time, interest, and expertise to make informed judgments may lack the authority to make the appropriate changes, while senior managers or elected officials with authority are likely to lack the motivation and operational expertise to consider specific processes. As a result, information is likely to remain unused, potential learn- ing opportunities untaken, and ineffective organizational processes unchanged.

Decision flexibility allows operators to participate in decision mak- ing and a chance to link learning with decisions. One of the major barriers to learning is when teams “lack the power to act in the domains about which they are learning” (Senge 199, xvii). Popper and Lipshitz (1998) recommend providing employees with “elbow room” to consider alternatives and experiment. Schulz (2001) has found that work units with more autonomy report higher levels of learning.

Decision flexibility has both structural and cultural components. An excellent illustration comes from reform efforts at the federal level during the 1990s. At a time when the federal government was trying to provide agencies with greater flexibility through formal grants of authority and eliminating rules, Ban (1995) found that organizational culture shaped agencies’ willingness to exert flexibility and work around formal constraints. Subsequent empirical work has provided additional evidence on the cultural aspects of flexibility, showing that organizational culture interacts with perception of rule constraints to affect the performance of agencies (Pandey, Coursey, and Moynihan 2007

.

Learning Forums

While performance management reforms created OLMs to collect, store, and disseminate data, they have generally neglected learning

How Do Public Organizations Learn? 1099

forums, that is, organizational routines in which employees seek to examine and discuss information and consider what it implies for subsequent action (Moynihan 2005). Learning forums ideally occur on a regular basis, are based on a dialogue between key actors, include experiential and hard data, and are focused on improving the organization.

Learning forums represent a marriage of the cultural and structural approaches. Such routines are likely to be formally established. Pisano, Bohmer, and Edmondson (2001) find that firms with formal procedures for learning are more effective learners. However, the nature and efficacy of the dialogue in such forums will depend greatly on the cultural attributes of the organization. Learning fo- rums work best if they occur within a culture that is purpose driven, encourages the open sharing of information, supports the presenta- tion of different perspectives, and examines errors to solve problems rather than to allocate blame (Moynihan 2005). The literature on organizational learning suggests that confrontational uses of data lead to defensive reactions rather than learning, and that learning forums that establish collegiality and an equal footing for members are likely to overcome defensiveness and foster information sharing.

Learning forums can take a variety of forms, such as strategic plan- ning routines, after-action reviews, benchmarking processes, or other routines in which data is examined. In this article, we examine work groups as learning forums. This approach is consistent with Kaplan and Norton’s (1996, 252) argument that learning is fostered by goal-based problem solving among teams who use information to intelligently respond to organizational conditions, and Yanow’s (2000) recommendation to study communities of practice. We looked for team environments with the ideal characteristics of learn- ing forums, specifically those in which workers were likely to take into account the opinions of others, work groups were given feed- back on performance, and work groups focused on organizational improvement. However, the potential cohesiveness arising from these qualities can lead to groupthink in teams “when the members’ strivings for unanimity override their motivation to realistically appraise alternative courses of action” (Janis 1982, 9). Therefore, we also looked for groups that demonstrated a cultural value of inquiry, implying “a willingness to accept a degree of uncertainty and to suspend judgment until a satisfactory understanding is achieved and is similar to the value of intellectual curiosity (questioning the status quo)” (Lipshitz, Popper, and Friedman 2002, 85). learning.

demographic information, can be found in Moynihan and Landuyt (2008, 127–29). The appendix lists the items used to construct the variables, as well as descriptive data for the variables and Cronbach’s alphas for indexes.

We include controls for gender, age, minority status, length of service with the organization, supervisory status, and education. We employed a fixed-effects approach, running an ordinary be significant or altered the nature of the relationship between the independent and dependent variables.

Measuring Learning

Learning is a variable plagued by measurement difficulties. There- fore, we discuss our measurement of organizational learning in particular detail. A strict cultural approach argues that learning largely involves context-specific tacit knowledge that is impossible to measure, but is best studied from an interpretative standpoint (Yanow 2000). Such research uses case studies to study the collective rather than the individual, by examining the cultural artifacts, social practices, and language through which norms are communicated. Case research also dominates more mainstream learning research, and has been used to impute learning across time by developing limited historical narratives. But the case approach raises issues of generalizability and imprecision in the measurement of variables— one researcher’s perception of learning may vary from that of another. Qualitative research may also have led to an overemphasis on cultural attributes. As researchers see learning accompanied by certain cultural traits, they assume a causal relationship, but such relationships have been largely untested using quantitative methods. Quantitative research on learning has increased, largely in a private sector context, often focusing on strategic issues related to joint ven- tures and alliances (Bapuji and Crossan 2004). Efforts to measure learning have generally relied on perceptual measures from surveys (e.g., Askim, Johnsen, and Christophersen 2008; Schulz 2001).

Our approach assumes that individual learning is captured and used for organizational purposes. The role of the individual is central to most treatments of learning (Fiol and Lyles 1985), but the struggle has been in accounting for how the organization captures individual knowledge. We rely on close-ended individual responses, and there- fore cannot capture examples of collective learning in the way that an interpretative approach would. But while we survey individuals, our questions are not primarily related to individual incentives and individual cognitive processes, but seek to probe the individual’s perception of group action. The survey instructions ask the respond-

Data and Methods

The data were obtained from a 2004 survey of Texas State agencies, the Survey of Or- ganizational Excellence. A total of 62,628 employees were surveyed in 53 different state agencies, resulting in 34,668 usable responses, a response rate of more than 55 percent.

This confidential survey was administered primarily by e-mail but, when necessary, employees were provided a paper version of the survey. Additional information about the survey, including the agencies surveyed and

The data were obtained from a 2004 survey of Texas State agencies, the Survey of Organizational Excellence. A total of 62,628 employees were surveyed in 53 different state agencies, resulting in 34,668 usable responses, a response rate of more than 55 percent.

ent to answer “from the perspective of your immediate workplace.” The question structure reinforces the emphasis on group, and only the mission orientation variable and demo- graphic measures ask the respondent to refer to his or her individual situation.

Our dependent variable is a scale of differ- ent learning items that share the assumption that groups of employees use knowledge to make decisions that benefit the organization. Consistent with Barrados and Mayne (2003), we include a measure of experiential learning

1100 Public Administration Review • November | December 2009

from errors (“We are encouraged to learn from our mistakes”). We also incorporate a customer-service perspective on learning that is consistent with Tippins and Sohi (2003) (“We use feedback from those we serve to improve performance”). Our third item is a general measure of learning based on information (“We integrate information and act intelligently on that information”). The data come from the Organizational Excellence Group, which collected the survey and provided aggregate summaries of the data to the organization. The use of these external evaluations by agencies is an indicator of whether the organization will make use of relevant information. Therefore, the final part of our index asks respondents to agree or disagree with the statement, “I believe we will use the information from this survey to improve our performance.”

Results and Discussion

The results of the regression are presented in table 1. The results suggest strong support for the model. All of the independent vari- ables have statistically significant relationships with the dependent variable and in the direction hypothesized.

The large sample size tends to inflate even relatively trivial (though real) relationships between the independent and dependent variables into statistically significant relationships. Therefore, in interpreting the individual independent variables, it is helpful to also consider the size of the relative impact of the regression coefficients. All of the theoretical variables in the survey are measured on a 1–5 scale and index items have been converted to a 1–5 scale, so examining the coefficients provides us with a transparent basis for comparing the relative strength of the theoretical variables (the individual control variables have not been adjusted).

The most influential independent variable is the work groups as learning forums. A one-unit increase in this variable is associ- ated with approximately a .42 increase in the dependent variable. The more that work groups take on the characteristics of learning forums—inclusiveness and dialogue, consideration of performance

Table 1 OLS Regression on Organizational Learning Controlling for Agency-Level Effects

information, a desire to foster improvement—the greater the sense of organizational learning among respondents. The information systems variable is the next most influential variable, with a one-unit increase in this measure leading to a .15 increase in the dependent variable. The difference in the size of the effects between learning forums and information systems variables suggests that while both are important, public organizations would see a greater payoff if they devoted as much energy to creating learning forums as they have done to creating information systems.

The mission orientation variable was the next most influential. A one-unit increase in understanding of the mission, values, and stra- tegic plan of the organization fostered a .14 increase in our measure of organizational learning. Are the findings on mission orientation and information systems a vindication for the performance man- agement movement? Both variables have a statistically significant relationship with learning. However, the relationship must be taken in the context of how the independent variables were measured. We measured with the assumption that information systems were working as they should be, with the right people receiving the right information at the right time. This is not the same thing as simply having an information system. We measured whether employees clearly understood the organization’s mission, vision, and strate- gies. This is not the same thing as producing a strategic plan. In both instances, there are considerable implementation barriers. But if public organizations can overcome those barriers, information systems and a mission orientation can foster organizational learning.

Decision flexibility also has a significant relationship with learning, but the variable has lower explanatory power. A one-unit increase in the perception that managers have decision authority increases the measure of organizational learning by .07. Along with resources, this is the weakest impact of any of the variables in the model. Why does flexibility not have a stronger influence on learning? A possible explanation comes from case evidence on the use of performance information in state government (Moynihan 2008). Existing flex- ibility often allows for many of the changes that managers wish to pursue. So although a sense of greater authority is helpful, learning can occur under existing authority. In addition, employees who are not used to additional authority may not have a clear idea of how to use it.

The impact of resources is also significant, but more modest. A one- unit increase in perceived adequacy of resources is predicted to lead to a .08 increase in the measure of organizational learning. Clearly, resources matter. The findings suggest that organizational slack fos- ters learning. In part, this may be because slack allows organizations to be less reactive and to pursue deliberate change. In part, it may be because organizational slack enables the provision of financial resources and staff time to support learning forums.

Finally, we briefly note the relevance of the control variables. While most of the variables are statistically significant, none approaches the explanatory power of any of any of the theoretical variables tested. We find that age is positively related to perceptions of learn- ing, but length of state service is not. This finding is consistent with other results showing that older public sector employees tend to have positive job attitudes, but that those who have been in the same organization or position for a long time (controlling for age)

Variables Coefficient

Standard error .135 .003 .004 .003 .005 .003

.003 .002 .0071 .002 .006 .006

Constant Information systems Mission orientation Decision flexibility Learning forums Adequacy of resources Control Variables Age Years with organization Supervisor Education Female Minority

.843*** .147*** .137*** .070*** .419*** .078***

.014*** –.017*** .020** –.015***

.049***

.017**

Note: Agency-specific effects are included but not reported. N = 24,614. *** p < .001; **p < .01 (two tailed).

How Do Public Organizations Learn? 1101

tend to be less engaged (Moynihan and Pandey 2008; Traut, Larsen, and Feimer 2000). The impact of tenure may cause employees to negatively modify their expectations toward organizational activity, so that they are more likely to be critical of and less involved in or- ganizational learning efforts. We find that supervisors, females, and minorities are more likely to perceive evidence of learning, although the impact of supervisory and minority status is relatively small and significant only at the .01 level. Interestingly, level of education is negatively related to perceptions of organizational learning. This may be because those with more educational experience have higher standards for what constitutes learning. A complementary explana- tion is that those with greater education begin with a higher level of knowledge, and so their organizational experiences are less likely to engender new knowledge.

From Dualism to Duality: Bridging Structure and Culture

One conclusion that emerges from our findings is the difficulty of separating structural and cultural approaches to learning. The theoretical variables, with the exception of information systems and resources, have both cultural and structural attributes. Therefore, culture and structure intertwine to form expectations and shape behavior. This suggests that those seeking to foster organizational learning can pursue different avenues, but ideally should seek to en- sure that structural approaches mesh with cultural approaches. While culture and structure are useful inasmuch as they categorize different approaches to organizational life, such categorizations can become too constraining if viewed as alternate approaches to learning.

The question that motivates this article is, how do public organiza- tions learn? The answer moves us beyond the cultural-structural de- bate, and instead focuses on how human agency can reshape broader social norms that foster learning. Learning practices are recursive, shaped and reshaped by norms, rules, and resources (Berends, Bo- ersma, and Weggeman 2003). For example, learning forums can be seen as a fluid social practice through which organizational learning can be created and restructured (Berends, Boersma, and Weggeman 2003, 1053; Nonaka and Toyama 2003). Individuals can redefine relevant structural properties through social action, thereby re-creating the relevant context for learning. The following section reviews the potential for practitioners to affect the learning variables tested in our model.

Practical Implications for Leadership

What are the practical implications of our model for organizational learning? We look in particular at the role of leadership, given that leadership can explain variation in learning across similar organiza- tions (Moynihan 2005; Lipshitz and Popper 2000), and in the same organization across time (Berends, Boersma, and Weggeman 2003; Popper and Lipshitz 2000). A structuration perspective suggests that it is not simply the beliefs of leaders that matter, but how these beliefs motivate social actions to reshape learning, while treating structure and culture as mutually dependent forces rather than as alternatives.

How do leaders create norms that support learning? Leaders can claim a wide variety of priorities, but employees look to a leader’s

What is required, therefore, is an ability to rethink culture and structure, recognizing them as broad and connected norms that shape behavior. Structuration theory offers a suitable theoretical framing device. Struc- turation theory takes seemingly opposed dualisms in social theory and reconceptual- izes them as mutually reinforcing dualities (Giddens 1984). Structuration theory uses the term “structure” to capture what tradi- tional organizational theory would treat as both structure and culture, including norms, interpretations, rules, and resources. Consist- ent with the cultural approach to learning, structuration theory argues that norms and interpretations shape behavior. Consistent with a structural approach to learning, struc- turation theory points to the importance of rules and resources to social action.

What are the implications of structuration theory in helping to bridge cultural and structural perspectives? Not only does structura- tion theory offer a way to reconceptualize the dualism of structure and culture into one overarching understanding of social institu- tions, it also proposes a role for human agency and, therefore, change. A central tenet of structuration is the recursive nature of social action. Social action both reflects and reconstitutes struc- ture: structure influences agents, agents influence structure, and so on. Through its emphasis on constant reproduction, structura- tion theory views individuals as more than the passive recipients of organizational influence, making them agents capable of reconstitut- ing the broader social norms.

actions—specifically, how leaders spend their time, attention, and resources—to infer where the real priorities lie (Popper and Lips- hitz 2000). For example, Askim, Johnsen, and Christophersen (2008) find that when leaders spend time participating in benchmarking processes, these learning forums are more likely to influence decisions. Offering actual or symbolic rewards for behavior consist-

ent with desired cultural values positively reinforces such values. Leaders can also direct organizational resources to OLMs, includ- ing financial resources, specialized staff, and general staff time. These actions use cultural

and structural organizational levers to help establish norms consist- ent with learning.

Even with structural OLMs such as information systems, such norms are important. For most organizations, information systems are already in place, and so the leadership challenge is to ensure that such systems are truly useful, providing relevant information to the appropriate decision makers in a timely fashion. The more leaders devote time, attention, and resources to make clear that informa- tion systems are central to important decisions, the more likely it becomes that employees will use them (Moynihan 2005; Moynihan and Ingraham 2004).

Culture matters to factors such as decision flexibility and mission orientation, and leaders can reshape these factors by infusing the organization with supportive values (Schein 1992). Leaders can encourage learning by formally decentralizing decision authority to

How do leaders create norms that support learning? Leaders can claim a wide variety of priorities, but employees look to a leader’s actions—specifically, how leaders spend their time, attention, and resources—to infer where the real priorities lie.

1102 Public Administration Review • November | December 2009

those closest to the decisions. But such patterns of authority diffu- sion are also determined by cultural values that support empowered decision making over rule adherence (Ban 1995). Communications systems and strategic planning routines are structural mechanisms that can better explain the purpose of the organization, thereby fostering a mission orientation. But communication occurs beyond formal communication systems, and leaders can seek to shape the informal talk, symbols, and actions that form part of the organiza- tional culture.

The duality of structure and culture is perhaps best illustrated by the example of learning forums. Learning forums are a form of OLM consistent with a structural approach, and a relatively straightforward piece of advice for leaders is to establish routines of information use. But what makes such routines effective is the cultural attributes of the group and the wider organization. In this article, we tested not the existence of learning forums, but the degree to which positive learning forum characteristics were present among work groups. Purpose-driven work groups that used information but also allowed rigorous debate were more likely to be associated with organizational learning. Such characteristics depend on a wider organizational culture that values learning, acknowl- edges error without provoking defensiveness, welcomes multiple perspectives, and focuses on the assumptions challenged and infor- mation assessed rather than the status of the individuals involved. Simply declaring that learning forums should take place without seeking to ensure that such attributes are in place would reflect a failure by leaders to recognize just how much structure depends on culture.

Conclusion

The research presented in this article advances our understanding of learning in public organizations in a number of ways. First, a methodological contribution: large-N quantitative methods are infrequently

A second contribution of the article is to offer quantitative evi- dence that supports a range of variables. There are sound theoretical reasons, and in some cases qualitative evidence, why information systems, mission orientation, decision flexibility, and resource ad- equacy should support learning, but the findings offer another type of evidentiary support that they matter.

A third contribution is the strong support for the role of learning forums. The concept of learning forums is not as well established as the other variables tested here, and the relative influence of this variable marks it as potentially important for future investigation. The essential idea of learning forums—employees are given time and space to create a dialogue about what information means—is not limited to the work groups we investigate. Additional research could usefully examine how different types of learning forums mat- ter in different decision settings. Such research might also consider trade-offs between learning forum values. For example, a strong emphasis on error tolerance may be at odds with a pursuit of public accountability.

The fourth major contribution of the article is the effort to bridge the cultural and structural divide. Different camps within organi- zational learning have argued for each approach, while sometimes acknowledging that the other is important to learning. We move beyond this division by pointing out that, in practice, key variables often incorporate both structural and cultural aspects. Structuration theory offers a useful frame for reconsidering how structure and culture matter, emphasizing overarching norms that shape behavior, and the role of human agency in reconstituting these norms. This implies that public actors looking to foster learning should under- stand the extent to which they are limited by past norms while find- ing ways to reshape these norms by leveraging both the structural

and cultural aspects of their organization. There is much additional research that can be

done to answer the question of how pub- lic organizations learn. A complementary research question is whether such learning matters to performance. The organizational learning literature could be broadly but ac- curately categorized as instrumental in its approach, in that it advocates the pursuit of learning to improve organizational capacity and performance. In case studies, it is pos- sible to construct convincing narratives that

link learning to organizational improvements, but such claims are difficult to generalize. Having presented a quantitative approach to measuring learning, a next logical step is to link learning to an indicator of public sector performance.

Notes

1. To determine whether ordinary least squares was the appropriate estimation technique, we examined our data for heteroscedasticity, influential data and multicollinearity. A histogram of the standard- ized residuals shows that they are normally distributed. A scatter plot illustrates that the errors are relatively constant (homoscedas- tic) and independent of one another. For the basic model that includes the theoretical variables and individual-level controls, but not agency level controls, multicollinearity does not appear to be a

used to study organizational learning, espe- cially in public settings. In part, this is because of the difficulty of measuring learning in a satisfactory way. It is worth noting that our methodological strategy results in a number of limitations. We have the general limitations of any survey-based research, including the potential for common source bias to affect the results. The data are cross-sectional, which makes it difficult to attribute causality to the results. There are also more particular concerns. We rely on data from a particular state government, which may have significant differences with other states and other levels of government. By relying on individual responses about group actions, we assume that the respondent is capable of relating their social environment. And any survey or quantitative analysis of learning will necessarily miss aspects of organizational learning that can be learned only through an interpretative approach and qualitative methodology (Yanow 2000). However, case research brings its own weaknesses, including imprecision in variable definition and difficulty in replication that limits the ability of researchers to develop cumulative contingent knowledge. The relative dominance of qualitative research on learn- ing in the public sector setting provides additional relevance to our findings.

There is much additional research that can be done to answer the question of how public organizations learn. A complementary research question is whether such learning matters to performance.