- Split View
-
Views
-
Cite
Cite
Andrew B. Whitford, Soo-Young Lee, Exit, Voice, and Loyalty with Multiple Exit Options: Evidence from the US Federal Workforce, Journal of Public Administration Research and Theory, Volume 25, Issue 2, April 2015, Pages 373–398, https://doi.org/10.1093/jopart/muu004
- Share Icon Share
Abstract
We assess the turnover intentions of federal employees using Hirschman’s theory of exit, voice, and loyalty. Specifically, we follow other studies that have tested the effects of loyalty and voice on the likelihood a person states their intention to leave. However, using large-scale survey evidence from the federal workforce, we are able to assess the impact of loyalty, voice, and other factors (including assessments of pay) on the likelihood that a respondent will retire, leave for another federal agency, or leave for another sector. Our statistical analysis provides evidence that perceptions about voice and loyalty limit exit. Yet, the effects of voice, loyalty, and pay vary with exit option.
INTRODUCTION
In 1970, Albert O. Hirschman argued in Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States that response to a decline in the quality of an organization causes people to choose from the actions of exit, voice, and loyalty. Loyalty keeps one from leaving an organization going in the “wrong direction”; exit and voice are alternatives to loyalty (Hirschman 1970, 78). Historically, “empirical studies have not been as extensive or as thorough as the number of citations to Hirschman (1970) would lead one to expect” (Dowding et al. 2000, 469), although researchers have compiled over 1,500 general studies on turnover in the case of private sector organizations (Muchinsky and Morrow 1980). Researchers often mix voluntary turnover (e.g., quits) and involuntary turnover (e.g., discharges) (Shaw et al. 1998). Yet, a person’s stated intention to leave helps predict actual turnover (Hom, Griffeth, and Sellaro 1984; Mobley 1977; Steel and Ovalle 1984; Steers and Mowday 1981; Van Breukelen, Vlist, and Steensma 2004).
Historically there were few public sector studies of intent to leave (Cotton and Tuttle 1986; Selden and Moynihan 2000; Tett and Meyer 1993), though this has changed recently. Core themes remain the organizational traits of agencies losing employees (Kellough and Osuna 1995), the characteristics of those who exit (Lewis 1991; Lewis and Park 1989), and the role of voluntary turnover (Kellough and Osuna 1995; Lewis 1991; Lewis and Park 1989; Selden and Moynihan 2000). One shortcoming of these previous studies is that the weight of the literature about turnover and turnover intention has been on exit options outside of the federal government (i.e., leaving the federal government). However, as discussed later, in the 2008 Federal Human Capital Survey (FHCS), the proportion of respondents who selected the response “to leave the current organization for another job within the federal government” when answering the question “Are you considering leaving your organization?” was 17.0%, whereas those who selected the response “to leave the federal government” was 2.3%. This result shows that it is necessary to pay attention to other exit options such as movements within the federal government.
In 2008, Lee and Whitford argued that the exit, voice, and loyalty framework was useful for understanding turnover in the case of the public sector. That article offered statistical evidence showing exit varies with perceptions about voice and loyalty. Yet, dissatisfaction with pay is also a substantial cause of intention to leave. Finally, the article showed evidence for “motivation crowding” when pay-based motivation is emphasized. This article fills a gap in the literature on public sector turnover by moving beyond the study of “intent to leave” in broad terms. We focus attention on the intent to leave when there are multiple exit options. Realistically, employees do not just leave organizations: they go somewhere else. This “else” is the focus of our article. Specifically, we test whether an individual’s intention to leave (exit) for a specific option depends on the organization’s commitment to responding to voice and organizational attempts to build loyalty. Generally speaking, “turnover intention” is a person’s deliberate and conscious resolve to leave the organization of which she is currently a member (Tett and Meyer 1993). However, because employees face multiple options, intention could be “to leave to retire,” “intention to leave to take another job within the federal government,” or “intention to leave to take another job outside the federal government,” instead of just the choice of leaving or not leaving.
The main research question of this study is as follows: what are the relationships among the four kinds of exit choices, voice, loyalty, and pay? To address this question, we use the “no intention to leave” response as a base group for comparison to test whether an individual’s exit choices depend on the organization’s commitment to responding to voice and organizational attempts to build loyalty. We also ask whether the exit intention depends on the individual’s hierarchical status as nonsupervisor or team leader, supervisor or manager, and/or executive. In contrast to Hirschman, we test whether exit depends on the use of extrinsic incentives like pay. Last, we ask whether the specific exit intention depends on the respondent’s time horizon—whether it depends on an individual’s place in the age strata.
We provide evidence for this position using a statistical model of data from the 2008 FHCS. Our dependent variable is the respondent’s stated intention to leave for a specific exit option, and so is an unordered polychotomous variable; using a multinomial logit model, we estimate effects across individuals working at different levels of the organization’s hierarchy, and accounting for structural aspects of the sampling frame. We show direct effects for both voice and loyalty on exit. We show that pay also motivates these individuals. Finally, we show that these effects are different for different exit options.
We first turn to an extended discussion of the multiple exit options federal employees face, and focus on why considering all these options is important for our study of an employees “intention to leave.” After that we briefly review the exit, voice, and loyalty approach to understanding organizations and specify a model that includes multiple exit options. We then estimate and assess the results of our model. In the conclusion, we discuss the importance of these results for helping us better understand why employees intend to leave the workforce.
WHERE EMPLOYEES GO
We start with a brief discussion of our dependent variable. Our data on employees’ perceived intention to quit come from the 2008 Federal Human Capital (FHC) Survey conducted by the US Office of Personnel Management (OPM). More than 210,000 Federal employees responded to the 2008 FHCS, for a governmentwide response rate of 51%. We concentrate on a stated intention to exit, which makes this study somewhat different from traditional work in this area. Most studies have examined actual turnover, although Bertelli (2007) shows how intent to leave depends on attempts to make employees more fiscally accountable. Many turnover studies center on organizational-level data. In any case, turnover intention helps predict actual turnover (Hom, Griffeth, and Sellaro 1984; Mobley 1977; Steel and Ovalle 1984; Steers and Mowday 1981; Van Breukelen, Vlist, and Steensma 2004).
Typically, the dependent variable Intention to leave is measured as a dichotomous response (1 for yes, 0 for no) to the survey item “Are you considering leaving your organization?” For example, in the 2004 FHC, the proportions of respondents responding “yes” are: 36.1% (nonsupervisors and team leaders), 33.5% (supervisors and managers), and 38.7% (executives); the overall turnover intention is 35.4%. Yet, the dependent variable in this research is federal employees’ exit choices (i.e., types of turnover), measured by an item asking “Are you considering leaving your organization within the next year, and if so, why?” The response categories of this question are “no,” “yes, to retire,” “yes, to take another job within the federal government” (i.e., “leave agency”), and “yes, to take another job outside the federal government” (i.e., “leave government”). In other words, this study focuses on leaving the current organization, not the current job. The original survey item also includes another response category (“yes, other”); we omit since we cannot interpret what this clearly means. In the 2008 FHC, the proportions of respondents saying “no” are 74.6%, “to retire” 6.1%, “for another agency” 17.0%, and “to leave federal government” 2.3%.
We believe it is important to stress the ways in which this alternative specification of the “intention to exit” changes the way in which we understand exit as a general process in organizations. First, and most obviously, this approach allows us to see the factors that influence the intent to retire. At a minimum, we can speculate that our approach here brings pause to the longstanding complaint that many federal employees intend to retire very, very soon (Thompson and Seidner 2009). For instance, the OPM (2006) expected that nearly 60% of all federal public servants would be eligible for retirement within the following 10 years and that almost 40% of them would actually retire. Then director of the OPM called this “federal retirement tsunami.” The marginals described above, specifically that 6.1% intended to retire in the next year, help us see that in contrast, when compared to the other options in the “intend to leave” basket, retirement is a smaller concern than intra-agency movements.
Second, although most of the literature about turnover among federal executives has centered on exit options to the private or nonprofit sectors, the approach we take here allows us to see what factors will affect intended movement within government—from one agency to another. Here we must refer specifically to the longstanding concern that federal employees may leave for options outside government, often in the private sector, as part of a “revolving door” process (e.g., Kelman 1993). For instance, according to the Government Accountability Office (2008, highlights), in 2006, 52 contractors hired 2,435 former Department of Defense (DOD) senior and acquisition officials who had previously worked as generals, admirals, senior executives, contracting officers, or in other acquisition positions; these positions made them subject to restrictions on their post-DOD employment. Although that sort of motivation is real, and likely varies for an individual over time based on personal characteristics, the nature of the external labor market, and secular changes in federal agencies that reflect those timetables of presidential administrations, the marginals above describe a situation in which movements from one agency to another are over seven times more likely. Below, we will describe a set of mechanisms that flow from Hirschman’s exit, voice, and loyalty framework that may have some power to limit the intent to exit to either outside the government or to another agency, or it may be that a given mechanism is less likely to affect only one exit route. For example, it may be that mobility within government may be limited by giving federal employees the perception that they have voice, a finding that speaks to the problem of having policy-motivated agents.
Finally, we simply note that we are better off modeling choices that individuals face rather than settings that artificially constrain choices because of poor wording of survey questions. One aspect of working with data like those from the US Office of Personnel Management is that the survey designers do not have our academic interest in mechanism design as their first concern. As is widely noted, the purpose of such surveys is to further the management interests of a given administration, hence the interest in being able to obtain agency-level estimates of such variables as “organizational satisfaction,” rather than to understand the roles of higher-level concepts drawn from the study of work motivation. Unfortunately, historically OPM chose to assess exit intent by a crude instrument, and although such measures still provide a lens on the respondents’ work lives, much explanatory power was “left on the table.”
Our focus in this article is on these “multiple exit options,” on what drives exit, and how those factors’ influence may vary with options. The choice to go (or not go) “somewhere else” (with the “else” being explicit) is followed by a “because” statement. Next we turn to a synthetic theory of what drives the intent to exit an organization drawn from the path-breaking work of Albert O. Hirschman.
AN EVL THEORY OF EXITS
To exit means to leave an organization; expressing voice means bringing one’s dissatisfaction with the organization directly to management (or through a process that makes management subordinate) (Hirschman 1970, 4). Exit, voice, and loyalty are substitutes, though exit is a “reaction of last resort” (Hirschman 1970, 37). Loyalty is allegiance—in groups or organizations; and “loyalty makes exit less likely” (Hirschman 1970, 77). For instance, voice for workers may come through a trade union if unionization enhances negotiated relations with management (Freeman 1976, 1980; Freeman and Medoff 1984), but voice is usually more than just union membership (Dowding et al. 2000, 486). More broadly, studies added to this triumvirate the concept of neglect, which is similar to “shirking” in Brehm and Gates (1997) (see also Farrell 1983; Rusbult and Farrell 1983; Rusbult and Lowery 1985; Rusbult, Zembrodt, and Gunn 1982; Withey and Cooper 1989). Most studies offer the EVLN (exit, voice, loyalty, and neglect) components as dependent variables (e.g., Daley 1992). We focus this article on the argument that the likelihood a respondent says that they “intend to leave” depends on whether the organization enables them to exercise voice or encourages them to feel an obligation like loyalty. This is the reason why we do not include ‘neglect’ in our model. Unions are less relevant in the federal workforce because of limited variation, but in any case, managers can do other things to give people voice in organizations or to build organizational commitment generally.
As noted, numerous scholars in public administration and public management have studied employee turnover (i.e., exit) in the public sector because turnover increases the costs of hiring and training new employees and lowers productivity (Kim 2005; Lee and Whitford 2008). Table 1 summarizes previous studies on employee turnover in the public sector. These focused on actual turnover rates or public employees’ turnover intentions, usually to explain factors causing actual turnover or intentions.
Author(s) . | Independent Variables . | Dependent Variable(s) . | Data . | Unit of Analysis . |
---|---|---|---|---|
Lewis and Park (1989) | Sex (Are women more likely to quit than men?) | Exit (the employee either did or did not exit federal service) | 1% random sample of the Central Personnel Data File (United States) | Individual |
-Control variable: age, education, experience, salary | ||||
Lewis (1991) | General schedule (GS1-GS3, GS4-GS6, GS7-GS9, GS10-GS12, GS13, and above) | Exit rates | Central Personnel Data File (1973–1989, United States) | Individual |
-Control variable: gender, age, years of service, years of education | ||||
Kellough and Osuna (1995) | Percentage of the young workers, gender (% of female), race (% of minority), occupational characteristics (professional/administrative proportion, clerical proportion), agency size, union strength, percentage of the temporary employment | Quit rates | Pooled data from the US OPM (1986, 1988, 1990, and 1992) | Agency |
Selden and Moynihan (2000) | Unemployment rates, unionization, internal opportunity structure, pay, family-friendly policies (child care on site), training | Turnover rates | Data from the 1998 Government Performance Project, the National Association of State Personnel Executive (1999), the Book of State, the Bureau of Labor Statistics, and the US Bureau of the Census | State |
-Control variable: geographical region and size (median state annual income) | ||||
Kim (2005) | Job characteristics (work exhaustion, role clarity, role conflict), work environment (participatory management, resources), human resource management (advancement opportunities, training and development, pay and rewards satisfaction) | Turnover intentions (of IT employees) | Survey of central IT department employees (State governments of Nevada and Washington, 2003, United States) | Individual |
-Control variable: perceived job alternatives, gender, age, years of work education | ||||
Bertelli (2007) | Functional preferences (JIM: job involvement and intrinsic motivation), friendly workplace, quality of workgroup output, poor performance is addressed, quality of pay, quality of immediate supervisor, managerial review of goals and evaluation, supervisors receptive to change, hold leaders in high regard, reasonable workload, work-life balance supported, promotions are merit based, timely rewards for high performance, appraisal reflects performance, awards incentivize high performance, accountability for results, information regarding benefit changes | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: gender, minority | ||||
Bright (2008) | PSM(public service motivation) and P-O fit (person-organization fit) | Turnover intention and job satisfaction | Their own survey of public employees in Indiana, Kentucky, and Oregon (United States) | Individual |
-Control variable: age, minority status, gender, education, years of public sector experience | ||||
Lee and Whitford (2008) | Organizational satisfaction, voice, loyalty, pay | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: benefits, training, merit promotion, physical conditions, race, gender | ||||
Meier and Hicklin (2008) | Teachers’ turnover rates | Students’ performance (test score) | Texas school districts’ survey (1994–2002, United States) | School districts |
-Control Variable: resources (teacher pay, class size, state aid, percentage of noncertified teachers, teachers with advanced degrees, experienced teachers), constraints (poverty, race) | ||||
Moynihan and Landuyt (2008) | Individual characteristics (primary earner, household size, age, years in state of Texas, agency experience, education, gender, race), job characteristics (job satisfaction, workload, supervisor), human resource management practices (salary, perception of fair pay, benefits, merit promotion, family-friendly work practices, diverse workforce practices, employee development), work environment (loyalty, empowerment, voice) | Turnover intention | Survey of state employees in Texas (2004, United States) | Individual |
Moynihan and Pandey (2008) | Internal social network (obligation toward coworkers, coworker support), external social network, person-organization value fit | Turnover intention (short term and long term) | Their own survey of 12 organizations in the northeastern United States (2005) | Individual |
-Control variable: job satisfaction, age, years in position, nonprofit status | ||||
Choi (2009) | Race diversity, sex diversity, age diversity, diversity management, equal employment opportunity (EEO) complaints, job satisfaction | Turnover intention | Central Personnel Data File and 2004 FHCS (United States) | Individual |
-Control variable: size, tenure, gender, minority, supervisory status | ||||
Jung (2010) | Goal ambiguity, pay satisfaction, interpersonal relationship, merit promotion, diversity policy, workload satisfaction, benefit satisfaction | Turnover rates and turnover intention rates | FHCS (2006, United States) | Agency |
-Control variable: organizational size, minority rates, female rates | ||||
Lee and Jimenez (2011) | Performance-based rewards system and Performance-supporting supervision | Turnover intention | 2005 Merit Principles Survey (United States) | Individual |
-Control Variable: Work-related variables (job satisfaction, training, job position, year of employment, salary, union membership) and Socio demographic variables (gender, age, race) | ||||
Lee and Hong (2011) | Paid leave for family care, child care subsidy, telecommuting, alternative work schedule | Turnover rates | Federal Human Resources Data (OPM, 2005, 2007), FHCS (2004, 2006, United States) | Agency |
-Control Variable: Pay Satisfaction, Training Satisfaction, Physical Conditions Satisfaction | ||||
Cho and Lewis (2012) | Salary, federal experience, age, education, gender and race | Turnover intention and behavior | Central personnel data file, 2005 Merit Principles Survey | Individual |
Bertelli and Lewis (2013) | Agency-specific human capital, outside option, relative influence over policy | Turnover intention | 2007–2008 Survey on the Future of Government Service | Individual (federal executives) |
-Respondent ideal point, years employed, age of respondent, eligibility to retire, independent commission |
Author(s) . | Independent Variables . | Dependent Variable(s) . | Data . | Unit of Analysis . |
---|---|---|---|---|
Lewis and Park (1989) | Sex (Are women more likely to quit than men?) | Exit (the employee either did or did not exit federal service) | 1% random sample of the Central Personnel Data File (United States) | Individual |
-Control variable: age, education, experience, salary | ||||
Lewis (1991) | General schedule (GS1-GS3, GS4-GS6, GS7-GS9, GS10-GS12, GS13, and above) | Exit rates | Central Personnel Data File (1973–1989, United States) | Individual |
-Control variable: gender, age, years of service, years of education | ||||
Kellough and Osuna (1995) | Percentage of the young workers, gender (% of female), race (% of minority), occupational characteristics (professional/administrative proportion, clerical proportion), agency size, union strength, percentage of the temporary employment | Quit rates | Pooled data from the US OPM (1986, 1988, 1990, and 1992) | Agency |
Selden and Moynihan (2000) | Unemployment rates, unionization, internal opportunity structure, pay, family-friendly policies (child care on site), training | Turnover rates | Data from the 1998 Government Performance Project, the National Association of State Personnel Executive (1999), the Book of State, the Bureau of Labor Statistics, and the US Bureau of the Census | State |
-Control variable: geographical region and size (median state annual income) | ||||
Kim (2005) | Job characteristics (work exhaustion, role clarity, role conflict), work environment (participatory management, resources), human resource management (advancement opportunities, training and development, pay and rewards satisfaction) | Turnover intentions (of IT employees) | Survey of central IT department employees (State governments of Nevada and Washington, 2003, United States) | Individual |
-Control variable: perceived job alternatives, gender, age, years of work education | ||||
Bertelli (2007) | Functional preferences (JIM: job involvement and intrinsic motivation), friendly workplace, quality of workgroup output, poor performance is addressed, quality of pay, quality of immediate supervisor, managerial review of goals and evaluation, supervisors receptive to change, hold leaders in high regard, reasonable workload, work-life balance supported, promotions are merit based, timely rewards for high performance, appraisal reflects performance, awards incentivize high performance, accountability for results, information regarding benefit changes | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: gender, minority | ||||
Bright (2008) | PSM(public service motivation) and P-O fit (person-organization fit) | Turnover intention and job satisfaction | Their own survey of public employees in Indiana, Kentucky, and Oregon (United States) | Individual |
-Control variable: age, minority status, gender, education, years of public sector experience | ||||
Lee and Whitford (2008) | Organizational satisfaction, voice, loyalty, pay | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: benefits, training, merit promotion, physical conditions, race, gender | ||||
Meier and Hicklin (2008) | Teachers’ turnover rates | Students’ performance (test score) | Texas school districts’ survey (1994–2002, United States) | School districts |
-Control Variable: resources (teacher pay, class size, state aid, percentage of noncertified teachers, teachers with advanced degrees, experienced teachers), constraints (poverty, race) | ||||
Moynihan and Landuyt (2008) | Individual characteristics (primary earner, household size, age, years in state of Texas, agency experience, education, gender, race), job characteristics (job satisfaction, workload, supervisor), human resource management practices (salary, perception of fair pay, benefits, merit promotion, family-friendly work practices, diverse workforce practices, employee development), work environment (loyalty, empowerment, voice) | Turnover intention | Survey of state employees in Texas (2004, United States) | Individual |
Moynihan and Pandey (2008) | Internal social network (obligation toward coworkers, coworker support), external social network, person-organization value fit | Turnover intention (short term and long term) | Their own survey of 12 organizations in the northeastern United States (2005) | Individual |
-Control variable: job satisfaction, age, years in position, nonprofit status | ||||
Choi (2009) | Race diversity, sex diversity, age diversity, diversity management, equal employment opportunity (EEO) complaints, job satisfaction | Turnover intention | Central Personnel Data File and 2004 FHCS (United States) | Individual |
-Control variable: size, tenure, gender, minority, supervisory status | ||||
Jung (2010) | Goal ambiguity, pay satisfaction, interpersonal relationship, merit promotion, diversity policy, workload satisfaction, benefit satisfaction | Turnover rates and turnover intention rates | FHCS (2006, United States) | Agency |
-Control variable: organizational size, minority rates, female rates | ||||
Lee and Jimenez (2011) | Performance-based rewards system and Performance-supporting supervision | Turnover intention | 2005 Merit Principles Survey (United States) | Individual |
-Control Variable: Work-related variables (job satisfaction, training, job position, year of employment, salary, union membership) and Socio demographic variables (gender, age, race) | ||||
Lee and Hong (2011) | Paid leave for family care, child care subsidy, telecommuting, alternative work schedule | Turnover rates | Federal Human Resources Data (OPM, 2005, 2007), FHCS (2004, 2006, United States) | Agency |
-Control Variable: Pay Satisfaction, Training Satisfaction, Physical Conditions Satisfaction | ||||
Cho and Lewis (2012) | Salary, federal experience, age, education, gender and race | Turnover intention and behavior | Central personnel data file, 2005 Merit Principles Survey | Individual |
Bertelli and Lewis (2013) | Agency-specific human capital, outside option, relative influence over policy | Turnover intention | 2007–2008 Survey on the Future of Government Service | Individual (federal executives) |
-Respondent ideal point, years employed, age of respondent, eligibility to retire, independent commission |
Author(s) . | Independent Variables . | Dependent Variable(s) . | Data . | Unit of Analysis . |
---|---|---|---|---|
Lewis and Park (1989) | Sex (Are women more likely to quit than men?) | Exit (the employee either did or did not exit federal service) | 1% random sample of the Central Personnel Data File (United States) | Individual |
-Control variable: age, education, experience, salary | ||||
Lewis (1991) | General schedule (GS1-GS3, GS4-GS6, GS7-GS9, GS10-GS12, GS13, and above) | Exit rates | Central Personnel Data File (1973–1989, United States) | Individual |
-Control variable: gender, age, years of service, years of education | ||||
Kellough and Osuna (1995) | Percentage of the young workers, gender (% of female), race (% of minority), occupational characteristics (professional/administrative proportion, clerical proportion), agency size, union strength, percentage of the temporary employment | Quit rates | Pooled data from the US OPM (1986, 1988, 1990, and 1992) | Agency |
Selden and Moynihan (2000) | Unemployment rates, unionization, internal opportunity structure, pay, family-friendly policies (child care on site), training | Turnover rates | Data from the 1998 Government Performance Project, the National Association of State Personnel Executive (1999), the Book of State, the Bureau of Labor Statistics, and the US Bureau of the Census | State |
-Control variable: geographical region and size (median state annual income) | ||||
Kim (2005) | Job characteristics (work exhaustion, role clarity, role conflict), work environment (participatory management, resources), human resource management (advancement opportunities, training and development, pay and rewards satisfaction) | Turnover intentions (of IT employees) | Survey of central IT department employees (State governments of Nevada and Washington, 2003, United States) | Individual |
-Control variable: perceived job alternatives, gender, age, years of work education | ||||
Bertelli (2007) | Functional preferences (JIM: job involvement and intrinsic motivation), friendly workplace, quality of workgroup output, poor performance is addressed, quality of pay, quality of immediate supervisor, managerial review of goals and evaluation, supervisors receptive to change, hold leaders in high regard, reasonable workload, work-life balance supported, promotions are merit based, timely rewards for high performance, appraisal reflects performance, awards incentivize high performance, accountability for results, information regarding benefit changes | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: gender, minority | ||||
Bright (2008) | PSM(public service motivation) and P-O fit (person-organization fit) | Turnover intention and job satisfaction | Their own survey of public employees in Indiana, Kentucky, and Oregon (United States) | Individual |
-Control variable: age, minority status, gender, education, years of public sector experience | ||||
Lee and Whitford (2008) | Organizational satisfaction, voice, loyalty, pay | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: benefits, training, merit promotion, physical conditions, race, gender | ||||
Meier and Hicklin (2008) | Teachers’ turnover rates | Students’ performance (test score) | Texas school districts’ survey (1994–2002, United States) | School districts |
-Control Variable: resources (teacher pay, class size, state aid, percentage of noncertified teachers, teachers with advanced degrees, experienced teachers), constraints (poverty, race) | ||||
Moynihan and Landuyt (2008) | Individual characteristics (primary earner, household size, age, years in state of Texas, agency experience, education, gender, race), job characteristics (job satisfaction, workload, supervisor), human resource management practices (salary, perception of fair pay, benefits, merit promotion, family-friendly work practices, diverse workforce practices, employee development), work environment (loyalty, empowerment, voice) | Turnover intention | Survey of state employees in Texas (2004, United States) | Individual |
Moynihan and Pandey (2008) | Internal social network (obligation toward coworkers, coworker support), external social network, person-organization value fit | Turnover intention (short term and long term) | Their own survey of 12 organizations in the northeastern United States (2005) | Individual |
-Control variable: job satisfaction, age, years in position, nonprofit status | ||||
Choi (2009) | Race diversity, sex diversity, age diversity, diversity management, equal employment opportunity (EEO) complaints, job satisfaction | Turnover intention | Central Personnel Data File and 2004 FHCS (United States) | Individual |
-Control variable: size, tenure, gender, minority, supervisory status | ||||
Jung (2010) | Goal ambiguity, pay satisfaction, interpersonal relationship, merit promotion, diversity policy, workload satisfaction, benefit satisfaction | Turnover rates and turnover intention rates | FHCS (2006, United States) | Agency |
-Control variable: organizational size, minority rates, female rates | ||||
Lee and Jimenez (2011) | Performance-based rewards system and Performance-supporting supervision | Turnover intention | 2005 Merit Principles Survey (United States) | Individual |
-Control Variable: Work-related variables (job satisfaction, training, job position, year of employment, salary, union membership) and Socio demographic variables (gender, age, race) | ||||
Lee and Hong (2011) | Paid leave for family care, child care subsidy, telecommuting, alternative work schedule | Turnover rates | Federal Human Resources Data (OPM, 2005, 2007), FHCS (2004, 2006, United States) | Agency |
-Control Variable: Pay Satisfaction, Training Satisfaction, Physical Conditions Satisfaction | ||||
Cho and Lewis (2012) | Salary, federal experience, age, education, gender and race | Turnover intention and behavior | Central personnel data file, 2005 Merit Principles Survey | Individual |
Bertelli and Lewis (2013) | Agency-specific human capital, outside option, relative influence over policy | Turnover intention | 2007–2008 Survey on the Future of Government Service | Individual (federal executives) |
-Respondent ideal point, years employed, age of respondent, eligibility to retire, independent commission |
Author(s) . | Independent Variables . | Dependent Variable(s) . | Data . | Unit of Analysis . |
---|---|---|---|---|
Lewis and Park (1989) | Sex (Are women more likely to quit than men?) | Exit (the employee either did or did not exit federal service) | 1% random sample of the Central Personnel Data File (United States) | Individual |
-Control variable: age, education, experience, salary | ||||
Lewis (1991) | General schedule (GS1-GS3, GS4-GS6, GS7-GS9, GS10-GS12, GS13, and above) | Exit rates | Central Personnel Data File (1973–1989, United States) | Individual |
-Control variable: gender, age, years of service, years of education | ||||
Kellough and Osuna (1995) | Percentage of the young workers, gender (% of female), race (% of minority), occupational characteristics (professional/administrative proportion, clerical proportion), agency size, union strength, percentage of the temporary employment | Quit rates | Pooled data from the US OPM (1986, 1988, 1990, and 1992) | Agency |
Selden and Moynihan (2000) | Unemployment rates, unionization, internal opportunity structure, pay, family-friendly policies (child care on site), training | Turnover rates | Data from the 1998 Government Performance Project, the National Association of State Personnel Executive (1999), the Book of State, the Bureau of Labor Statistics, and the US Bureau of the Census | State |
-Control variable: geographical region and size (median state annual income) | ||||
Kim (2005) | Job characteristics (work exhaustion, role clarity, role conflict), work environment (participatory management, resources), human resource management (advancement opportunities, training and development, pay and rewards satisfaction) | Turnover intentions (of IT employees) | Survey of central IT department employees (State governments of Nevada and Washington, 2003, United States) | Individual |
-Control variable: perceived job alternatives, gender, age, years of work education | ||||
Bertelli (2007) | Functional preferences (JIM: job involvement and intrinsic motivation), friendly workplace, quality of workgroup output, poor performance is addressed, quality of pay, quality of immediate supervisor, managerial review of goals and evaluation, supervisors receptive to change, hold leaders in high regard, reasonable workload, work-life balance supported, promotions are merit based, timely rewards for high performance, appraisal reflects performance, awards incentivize high performance, accountability for results, information regarding benefit changes | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: gender, minority | ||||
Bright (2008) | PSM(public service motivation) and P-O fit (person-organization fit) | Turnover intention and job satisfaction | Their own survey of public employees in Indiana, Kentucky, and Oregon (United States) | Individual |
-Control variable: age, minority status, gender, education, years of public sector experience | ||||
Lee and Whitford (2008) | Organizational satisfaction, voice, loyalty, pay | Turnover intention | 2002 FHCS (United States) | Individual |
-Control variable: benefits, training, merit promotion, physical conditions, race, gender | ||||
Meier and Hicklin (2008) | Teachers’ turnover rates | Students’ performance (test score) | Texas school districts’ survey (1994–2002, United States) | School districts |
-Control Variable: resources (teacher pay, class size, state aid, percentage of noncertified teachers, teachers with advanced degrees, experienced teachers), constraints (poverty, race) | ||||
Moynihan and Landuyt (2008) | Individual characteristics (primary earner, household size, age, years in state of Texas, agency experience, education, gender, race), job characteristics (job satisfaction, workload, supervisor), human resource management practices (salary, perception of fair pay, benefits, merit promotion, family-friendly work practices, diverse workforce practices, employee development), work environment (loyalty, empowerment, voice) | Turnover intention | Survey of state employees in Texas (2004, United States) | Individual |
Moynihan and Pandey (2008) | Internal social network (obligation toward coworkers, coworker support), external social network, person-organization value fit | Turnover intention (short term and long term) | Their own survey of 12 organizations in the northeastern United States (2005) | Individual |
-Control variable: job satisfaction, age, years in position, nonprofit status | ||||
Choi (2009) | Race diversity, sex diversity, age diversity, diversity management, equal employment opportunity (EEO) complaints, job satisfaction | Turnover intention | Central Personnel Data File and 2004 FHCS (United States) | Individual |
-Control variable: size, tenure, gender, minority, supervisory status | ||||
Jung (2010) | Goal ambiguity, pay satisfaction, interpersonal relationship, merit promotion, diversity policy, workload satisfaction, benefit satisfaction | Turnover rates and turnover intention rates | FHCS (2006, United States) | Agency |
-Control variable: organizational size, minority rates, female rates | ||||
Lee and Jimenez (2011) | Performance-based rewards system and Performance-supporting supervision | Turnover intention | 2005 Merit Principles Survey (United States) | Individual |
-Control Variable: Work-related variables (job satisfaction, training, job position, year of employment, salary, union membership) and Socio demographic variables (gender, age, race) | ||||
Lee and Hong (2011) | Paid leave for family care, child care subsidy, telecommuting, alternative work schedule | Turnover rates | Federal Human Resources Data (OPM, 2005, 2007), FHCS (2004, 2006, United States) | Agency |
-Control Variable: Pay Satisfaction, Training Satisfaction, Physical Conditions Satisfaction | ||||
Cho and Lewis (2012) | Salary, federal experience, age, education, gender and race | Turnover intention and behavior | Central personnel data file, 2005 Merit Principles Survey | Individual |
Bertelli and Lewis (2013) | Agency-specific human capital, outside option, relative influence over policy | Turnover intention | 2007–2008 Survey on the Future of Government Service | Individual (federal executives) |
-Respondent ideal point, years employed, age of respondent, eligibility to retire, independent commission |
As table 1 shows, little previous research, especially with regard to the public sector, has focused on employees having several alternatives when they decide to leave their organizations. The broader management literature has considered this possibility. For instance, Carruthers and Pinder (1983) and Mueller and Price (1989) argued that employees consider and create alternatives (or job choices) across organizations as well as within the same organizations when they decide to leave their current job. Yet, “Most turnover models predicting job change decisions concentrate on employee’s leaving an organization and rarely take into account the employee’s destination choice” (Kirschenbaum and Weisberg 2002, 109). No large-scale statistical studies have paid attention to alternative forms of turnover destinations.
This is the case even though, as Miller (1996, 24) argued, turnover is “any voluntary movement out of a job” and most studies define it as “movement across the boundaries of an organization—or quitting.” He even includes in his definition movement within organizational boundaries such as quits, transfers, and promotions—aspects overlooked in a traditional definition of turnover as movement across organizational boundaries. In other words, turnover incorporates the mobility out of a job within the organization as well as any attempt to leave their organization. Therefore, researchers should realize that employees can have a set of turnover options such as quits, transfers and promotions within the organization, or planned retirement and “it is important to examine turnover by contrasting homogeneous groups of employees, based on their having experienced similar types of turnover” (Miller 1996, 24). For this reason, Miller (1996) identified four types of turnover: no turnover (staying), promotion, transfer, and resignation (quitting). Kirschenbaum and Weisberg (2002, 110–111) argued that “Employees’ movements from their current job or position could be within the same organization or crossing boundaries to another organization, or even withdrawing totally from the labor market,” which implies three types of turnover options such as quits, internal organizational destination options (e.g., another job in the same department, the same job in a different department, or a different job in a different department), and external destination choices involving interorganizational movements (e.g., the same job in a different organization or a different job in a different organization).
According to Hom et al. (1992), these turnover options depend on the nature of the labor market and employees’ perception of their organization. In other words, employees’ perceptions of specific factors in their work environment affect their turnover choices. Kirschenbaum and Weisberg (2002, 110) claimed “Those employees, whose career aspirations are best met by moving across or within their work organization . . . will be more sensitive to signs at their work place that will either reinforce or dampen their destination options.” Doeringer and Piore (1970) also pointed out that these turnover options come from the nature of the labor market in which work conditions and work-related behaviors are different according to career pathways. That is, the innate dynamics of labor markets provide different opportunity structures and working conditions for different career options (Spilerman 1977) and employees will respond to contrasting conditions between their present status and available opportunities (Kirschenbaum and Weisberg 2002). From this perspective, Kirschenbaum and Weisberg (2002) stated that intentions to improve employment conditions (including working conditions) and perceptions of other organization-related factors may decide a choice of one turnover option over another. In their view, it is natural for employees to try to move up to a better position and move out of a current organization for a better job in order to match their improved skills and knowledge with alternative job opportunities involving career and residential changes.
To start, an employee’s overall satisfaction with the organization gives context to Hirschman’s choice of exit, voice, or loyalty. The employee’s perception of organizational satisfaction forms a baseline for the exit calculus. We include the employee’s satisfaction with the organization in our models in order to check the employee’s overall psychological state with the organization:
H1: An employee is less likely to state her intention to leave for any given exit option when her satisfaction with the organization is high.
Researchers usually define job satisfaction as an affective reaction that depends on how a person compares desired outcomes and actual outcomes (e.g., Cranny, Smith, and Stone 1992). Employees are less likely to state intent to leave a firm when they are satisfied with their jobs (Gray-Toft and Anderson 1981; Mobley 1977; Murphy and Gorchels 1996; Ostroff 1992). For instance, lack of satisfaction with salary or career opportunities increases intent to leave (Rosse and Miller 1984). We note that Hirschman concentrated on organizational satisfaction, not job satisfaction (Driscoll 1978).1
What mechanisms other than unionization are available to increase voice? For Hirschman, voice is “any attempt at all to change, rather than to escape from, an objectionable state of affairs . . . to make an attempt at changing the practices, policies, and outputs of the firm from which one buys or of the organization to which one belongs” (1970, 30)—to achieve change “from within” (1970, 38). We can connect this to empowerment: the “process of enhancing feelings of self-efficacy among organizational members through the identification of conditions that foster powerlessness and through their removal by both formal organizational practices and informal techniques of providing efficacy information” (Conger and Kanungo 1988, 474). Usually, empowerment depends “on the assumption that individuals can have a high level of ‘voice’ in shaping and influencing organizational activities” (Spreitzer 1996, 484); it is an “active, rather than a passive, orientation to a work role” (Spreitzer 1995, 1444). For other views on this, see Eby et al. (1999), Liden, Wayne, and Sparrowe (2000), and Thomas and Velthouse (1990). We offer this hypothesis:
H2: An employee is less likely to state her intention to leave for any given exit option when she perceives increased opportunities to express voice.
Does loyalty make people less likely to exit? Hirschman notes that loyalists “suffer in silence, confident that things will soon get better” (1970, 38). Of course, loyalists may try to protect their organization from critics (Daley 1992). Loyalists are committed, exert effort, and maintain membership (Angle and Perry 1981; Mowday, Steers, and Porter 1979). Commitment resonates in this view: “organizational commitment refers to an individual’s loyalty or bond to his or her employing organization” (Bozeman and Perrewé 2001); a committed employee “should be loyal to his organization, should make sacrifices on its behalf, and should not criticize it” (Wiener and Vardi 1980, 86). Loyalty or commitment helps explain turnover intention (Mathieu and Zajac 1990), to the point that managers may try to enhance it (Menon 2001). We offer this hypothesis:
H3: An employee is less likely to state her intention to leave for any given exit option when she perceives the organization as attempting to foster loyalty among members of the workforce.
Our next hypothesis asks about monetary incentives, with the point that financial incentives can substitute for social motivations in hierarchies (Miller and Whitford 2002). Social psychologists have shown that intrinsic motives like voice and loyalty are important. Do monetary incentives reduce their power? For instance, Bertelli (2007) shows that supervisors under paybanding are less likely to leave. It appears that turnover does not move with pay at the aggregate level (Lewis 1991), although others disagree (Blau and Kahn 1981; Kim 1999; Leonard 1987; Utgoff 1983), potentially influencing quit rates (Park, Ofori-Dankwa, and Bishop 1994; Powell, Montgomery, and Cosgrove 1994; Shaw et al. 1998).
H4: An employee is less likely to state her intention to leave for any given exit option when her pay is satisfactory.
To build a full model of turnover intention, we also assess the impact of other possible causes, including satisfactions with benefits, training, promotion, and physical conditions, minority, and gender. Cotton and Tuttle (1986) note that turnover depends on external effects (union presence, unemployment rate), work-related effects (pay, overall job satisfaction, role clarity), and personal effects (age, education, gender). One important study in public administration shows that voluntary turnover depends on organizational factors (size, unionization), work force characteristics (professional/administrative proportion), and individual factors (age, gender, race) (Kellough and Osuna 1995). At the state level, environmental factors (unemployment, region), organizational factors (size, unionization), and human resource management practices (pay, training) also affect voluntary separation (Selden and Moynihan 2000).
Among these possible causes, minority and gender have been considered as standard controls in most studies of turnover and turnover intention (Moynihan and Landuyt 2008, 4). Yet, the analysis results have not reached an agreement on the impacts of minority and gender on turnover and turnover intention. With regard to minority, some scholars have found that minority employees are more likely to leave their current organizations than whites due to at-work racial discrimination, local unemployment, and commuting time (Shields and Price 2002; Zax 1989), whereas others have argued that quit rates are lower for minority employees than for whites (Blau and Khan 1981; Weiss 1984). Of course, a few recent research findings show no relationship between minority and turnover intention (Bertelli 2007; Lee and Whitford 2008). Regarding gender, some studies argued that women are more likely to leave their organizations due to the different job-matching processes for men and women (Cotton and Tuttle 1986; Kellough and Osuna 1995; Meitzen 1986). However, Moynihan and Landuyt (2008) pointed out and found that the traditional hypothesis that females are more likely to quit is no longer valid because of the changing patterns of workforce participation and the particular attraction of the public sector.2
Rather than take the position of other studies stating that intent to leave is a single exit option, we argue that the intent to leave varies across exit options: for example, retirement, for another agency, or outside the federal workforce. Accordingly, we move beyond previous studies that use federal workforce data to account for a peculiar potential cause of exiting through retirement: the respondent’s age. In this case we account for the respondent’s location in five age strata. We discuss these strata below. We expect that the likelihood of selecting retirement increases with movement through these age strata because older respondents will be closer to retirement and less likely to think about a change of career (Moynihan and Pandey 2008). Moynihan and Pandey (2008, 220) found that age is not a statistically significant cause of short-term turnover intention (i.e., active consideration of the movement outside the current organization) and age has a significant negative impact on long-term turnover intention (i.e., lifetime employment which intends to spend the rest of their career at the current organization). At a minimum, not accounting for age—even in simple form (as are strata)—is a form of omitted variable bias.
We also account for the respondent’s work status: for example, nonsupervisor/team leader, supervisor/manager, and executive. Knowledge of a person’s hierarchical level helps us understand perceived exit options. We expect that those inhabiting the lowest of the three hierarchical tiers—nonsupervisors and team leaders—are least likely to state intent to leave. Those in the higher tiers are more likely to respond as intending to leave because of access to alternative positions (Cotton and Tuttle 1986). Such intent would suggest position-specific retention methods. Top-level workers are more affected by long-term strategic considerations, which give them different time horizons than those of lower-level workers (Brinkerhoff 1972).
MODEL SPECIFICATION
Our variable Organizational Satisfaction is a single item “Considering everything, how satisfied are you with your organization?,” a five-point scale from “very dissatisfied” (1) to “very satisfied” (5), to measure the respondent’s perception of the ascent or decline of the organization. Voice is constructed from four items by dividing the sum of the responses to the items by the number of the items. Hirschman’s voice connects directly to the empowerment of employees in organizations (Lee and Whitford 2008); empowerment is the ‘‘process of enhancing feelings of self-efficacy among organizational members” (Conger and Kanungo 1988, 474) (Cronbach’s α is 0.87). The items are: “Employees have a feeling of personal empowerment and ownership of work processes,” “I feel encouraged to come up with new and better ways of doing things,” “My talents are used well in the workplace,” and “Supervisors/team leaders in my work unit provide employee(s) with the opportunities to demonstrate their leadership skills.” Loyalty is built from two items by dividing the sum of the responses to the items by the number of the items (Cronbach’s α is 0.77).3 The items are: “In my organization, leaders generate high levels of motivation and commitment in the workforce,” and “I recommend my organization as a good place to work.” The perception of pay (Pay) is the response to the question: “Considering everything, how satisfied are you with your pay?,” a five-point scale from “strongly disagree” (1) to “strongly agree” (5).
This study includes six control variables. Benefits is built from two items by dividing the sum of the responses to the items by the number of the items (Cronbach’s α is 0.73); the four items are: “How satisfied are you with child care subsidies?” and “How satisfied are you with work/life programs (for example, health and wellness, employee assistance, eldercare, and support groups)?” Training is constructed from two items (Cronbach’s α is 0.80): “How satisfied are you with the training you receive for your present job?” and “My training needs are assessed.” Merit Promotion is measured by the response to “Promotions in my work unit are based on merit.” Physical Conditions is measured by the response to “Physical conditions (for example, noise level, temperature, lighting, cleanliness in the workplace) allow employees to perform their jobs well.” Race is recorded “0” for white and non-Hispanic, non-Latino, and non-Spanish respondents, and “1” for others. Women are coded “1” for the variable Gender.
We divide respondents into nonsupervisor and/or team leader, supervisor and/or manager, and executive. A nonsupervisor does not supervise other employees. Team leaders provide employees with day-to-day guidance in conducting work projects, but do not have supervisory responsibilities and are not official supervisors. Supervisors oversee employees but not other supervisors; managers oversee one or more supervisors. An executive is a member of the Senior Executive Service or its equivalent. Our age strata come from responses to the following question: “What is your age group?” Age stratum 1 is composed of those responding “29 and under”; 2 is “30–39”; 3 is “40–49”; 4 is “50–59”; and 5 is “60 or older.” Table 2 shows the descriptive statistics.
. | Mean . | SD . |
---|---|---|
Organizational Satisfaction | 3.510 | 1.074 |
Voice | 3.502 | 0.902 |
Loyalty | 3.436 | 1.000 |
Pay | 3.564 | 1.070 |
Merit promotion | 2.969 | 1.183 |
Physical conditions | 3.679 | 1.076 |
Benefits | 3.223 | 0.725 |
Training | 3.480 | 0.910 |
Race | 0.138 | |
Gender | 0.337 | |
Age stratum 1 | 0.042 | |
Age stratum 2 | 0.177 | |
Age stratum 3 | 0.303 | |
Age stratum 4 | 0.397 | |
Age stratum 5 | 0.081 | |
Status stratum 1 | 0.759 | |
Status stratum 2 | 0.236 | |
Status stratum 3 | 0.005 | |
N | 120,547 |
. | Mean . | SD . |
---|---|---|
Organizational Satisfaction | 3.510 | 1.074 |
Voice | 3.502 | 0.902 |
Loyalty | 3.436 | 1.000 |
Pay | 3.564 | 1.070 |
Merit promotion | 2.969 | 1.183 |
Physical conditions | 3.679 | 1.076 |
Benefits | 3.223 | 0.725 |
Training | 3.480 | 0.910 |
Race | 0.138 | |
Gender | 0.337 | |
Age stratum 1 | 0.042 | |
Age stratum 2 | 0.177 | |
Age stratum 3 | 0.303 | |
Age stratum 4 | 0.397 | |
Age stratum 5 | 0.081 | |
Status stratum 1 | 0.759 | |
Status stratum 2 | 0.236 | |
Status stratum 3 | 0.005 | |
N | 120,547 |
. | Mean . | SD . |
---|---|---|
Organizational Satisfaction | 3.510 | 1.074 |
Voice | 3.502 | 0.902 |
Loyalty | 3.436 | 1.000 |
Pay | 3.564 | 1.070 |
Merit promotion | 2.969 | 1.183 |
Physical conditions | 3.679 | 1.076 |
Benefits | 3.223 | 0.725 |
Training | 3.480 | 0.910 |
Race | 0.138 | |
Gender | 0.337 | |
Age stratum 1 | 0.042 | |
Age stratum 2 | 0.177 | |
Age stratum 3 | 0.303 | |
Age stratum 4 | 0.397 | |
Age stratum 5 | 0.081 | |
Status stratum 1 | 0.759 | |
Status stratum 2 | 0.236 | |
Status stratum 3 | 0.005 | |
N | 120,547 |
. | Mean . | SD . |
---|---|---|
Organizational Satisfaction | 3.510 | 1.074 |
Voice | 3.502 | 0.902 |
Loyalty | 3.436 | 1.000 |
Pay | 3.564 | 1.070 |
Merit promotion | 2.969 | 1.183 |
Physical conditions | 3.679 | 1.076 |
Benefits | 3.223 | 0.725 |
Training | 3.480 | 0.910 |
Race | 0.138 | |
Gender | 0.337 | |
Age stratum 1 | 0.042 | |
Age stratum 2 | 0.177 | |
Age stratum 3 | 0.303 | |
Age stratum 4 | 0.397 | |
Age stratum 5 | 0.081 | |
Status stratum 1 | 0.759 | |
Status stratum 2 | 0.236 | |
Status stratum 3 | 0.005 | |
N | 120,547 |
MODEL ESTIMATION AND RESULTS
For the analysis, we will use multinomial logit model because our dependent variable (i.e., exit destination choices) is an unordered polychotomous variable with four response categories (Long and Freese 2006, 223). We are interested in the effects for the variables Organizational Satisfaction, Voice, Loyalty, Pay, Benefits, Training, Merit Promotion, Physical Conditions, Race, Gender, Age, and Status. Table 3 shows the model for the 2008 FHC data in three columns. The baseline effect is for the option of “do not leave.” The entry for each variable in the column “retirement” indicates the relative risk ratio (RRR) for that variable for the exit option of retiring, and so on; below we interpret these option-specific effects in relation to the option of not leaving. We also account for the fact that some groups were undersampled by including survey weights in our estimation.
. | Retirement . | Inside Federal Government . | Outside Federal Government . | |||
---|---|---|---|---|---|---|
RRR . | SE . | RRR . | SE . | RRR . | SE . | |
Organizational satisfaction | 0.738 | 0.014*** | 0.690 | 0.011*** | 0.623 | 0.019*** |
Voice | 0.966 | 0.022 | 0.759 | 0.014*** | 0.639 | 0.022*** |
Loyalty | 0.895 | 0.020*** | 0.888 | 0.017*** | 0.954 | 0.034 |
Pay | 1.088 | 0.014*** | 0.828 | 0.008*** | 0.687 | 0.012*** |
Merit promotion | 0.987 | 0.014 | 0.860 | 0.010*** | 0.881 | 0.021*** |
Physical conditions | 0.948 | 0.011*** | 1.071 | 0.010*** | 1.046 | 0.019* |
Benefits | 1.061 | 0.021** | 0.960 | 0.014** | 0.786 | 0.021*** |
Training | 1.137 | 0.020*** | 0.906 | 0.013*** | 1.013 | 0.026 |
Race | 1.004 | 0.041 | 1.466 | 0.047*** | 0.417 | 0.035*** |
Gender | 0.692 | 0.019*** | 1.031 | 0.026 | 0.610 | 0.031*** |
Age stratum 2 | 1.41×107 | 1.09×1010 | 0.849 | 0.038 | 0.882 | 0.086*** |
Age stratum 3 | 2.56×107 | 1.98×1010 | 0.520 | 0.023*** | 0.417 | 0.041 |
Age stratum 4 | 3.46×108 | 2.68×1011 | 0.264 | 0.012*** | 0.365 | 0.036*** |
Age stratum 5 | 1.04×109 | 8.05×1011 | 0.130 | 0.010*** | 0.232 | 0.031*** |
Status stratum 2 | 1.262 | 0.033*** | 1.249 | 0.032*** | 1.426 | 0.071*** |
Status stratum 3 | 1.186 | 0.162 | 0.774 | 0.226 | 9.172 | 1.567*** |
Agency-specific effects | Included | |||||
N | 120,547 | |||||
LR χ2 | 36552.26 | *** | ||||
Log-likelihood | −73338.0 | |||||
Pseudo R2 | 0.1995 |
. | Retirement . | Inside Federal Government . | Outside Federal Government . | |||
---|---|---|---|---|---|---|
RRR . | SE . | RRR . | SE . | RRR . | SE . | |
Organizational satisfaction | 0.738 | 0.014*** | 0.690 | 0.011*** | 0.623 | 0.019*** |
Voice | 0.966 | 0.022 | 0.759 | 0.014*** | 0.639 | 0.022*** |
Loyalty | 0.895 | 0.020*** | 0.888 | 0.017*** | 0.954 | 0.034 |
Pay | 1.088 | 0.014*** | 0.828 | 0.008*** | 0.687 | 0.012*** |
Merit promotion | 0.987 | 0.014 | 0.860 | 0.010*** | 0.881 | 0.021*** |
Physical conditions | 0.948 | 0.011*** | 1.071 | 0.010*** | 1.046 | 0.019* |
Benefits | 1.061 | 0.021** | 0.960 | 0.014** | 0.786 | 0.021*** |
Training | 1.137 | 0.020*** | 0.906 | 0.013*** | 1.013 | 0.026 |
Race | 1.004 | 0.041 | 1.466 | 0.047*** | 0.417 | 0.035*** |
Gender | 0.692 | 0.019*** | 1.031 | 0.026 | 0.610 | 0.031*** |
Age stratum 2 | 1.41×107 | 1.09×1010 | 0.849 | 0.038 | 0.882 | 0.086*** |
Age stratum 3 | 2.56×107 | 1.98×1010 | 0.520 | 0.023*** | 0.417 | 0.041 |
Age stratum 4 | 3.46×108 | 2.68×1011 | 0.264 | 0.012*** | 0.365 | 0.036*** |
Age stratum 5 | 1.04×109 | 8.05×1011 | 0.130 | 0.010*** | 0.232 | 0.031*** |
Status stratum 2 | 1.262 | 0.033*** | 1.249 | 0.032*** | 1.426 | 0.071*** |
Status stratum 3 | 1.186 | 0.162 | 0.774 | 0.226 | 9.172 | 1.567*** |
Agency-specific effects | Included | |||||
N | 120,547 | |||||
LR χ2 | 36552.26 | *** | ||||
Log-likelihood | −73338.0 | |||||
Pseudo R2 | 0.1995 |
***Indicates significant at better than .001 (two-tailed test).
**Indicates significant at better than .01 (two-tailed test).
*Indicates significant at better than .05 (two-tailed test).
. | Retirement . | Inside Federal Government . | Outside Federal Government . | |||
---|---|---|---|---|---|---|
RRR . | SE . | RRR . | SE . | RRR . | SE . | |
Organizational satisfaction | 0.738 | 0.014*** | 0.690 | 0.011*** | 0.623 | 0.019*** |
Voice | 0.966 | 0.022 | 0.759 | 0.014*** | 0.639 | 0.022*** |
Loyalty | 0.895 | 0.020*** | 0.888 | 0.017*** | 0.954 | 0.034 |
Pay | 1.088 | 0.014*** | 0.828 | 0.008*** | 0.687 | 0.012*** |
Merit promotion | 0.987 | 0.014 | 0.860 | 0.010*** | 0.881 | 0.021*** |
Physical conditions | 0.948 | 0.011*** | 1.071 | 0.010*** | 1.046 | 0.019* |
Benefits | 1.061 | 0.021** | 0.960 | 0.014** | 0.786 | 0.021*** |
Training | 1.137 | 0.020*** | 0.906 | 0.013*** | 1.013 | 0.026 |
Race | 1.004 | 0.041 | 1.466 | 0.047*** | 0.417 | 0.035*** |
Gender | 0.692 | 0.019*** | 1.031 | 0.026 | 0.610 | 0.031*** |
Age stratum 2 | 1.41×107 | 1.09×1010 | 0.849 | 0.038 | 0.882 | 0.086*** |
Age stratum 3 | 2.56×107 | 1.98×1010 | 0.520 | 0.023*** | 0.417 | 0.041 |
Age stratum 4 | 3.46×108 | 2.68×1011 | 0.264 | 0.012*** | 0.365 | 0.036*** |
Age stratum 5 | 1.04×109 | 8.05×1011 | 0.130 | 0.010*** | 0.232 | 0.031*** |
Status stratum 2 | 1.262 | 0.033*** | 1.249 | 0.032*** | 1.426 | 0.071*** |
Status stratum 3 | 1.186 | 0.162 | 0.774 | 0.226 | 9.172 | 1.567*** |
Agency-specific effects | Included | |||||
N | 120,547 | |||||
LR χ2 | 36552.26 | *** | ||||
Log-likelihood | −73338.0 | |||||
Pseudo R2 | 0.1995 |
. | Retirement . | Inside Federal Government . | Outside Federal Government . | |||
---|---|---|---|---|---|---|
RRR . | SE . | RRR . | SE . | RRR . | SE . | |
Organizational satisfaction | 0.738 | 0.014*** | 0.690 | 0.011*** | 0.623 | 0.019*** |
Voice | 0.966 | 0.022 | 0.759 | 0.014*** | 0.639 | 0.022*** |
Loyalty | 0.895 | 0.020*** | 0.888 | 0.017*** | 0.954 | 0.034 |
Pay | 1.088 | 0.014*** | 0.828 | 0.008*** | 0.687 | 0.012*** |
Merit promotion | 0.987 | 0.014 | 0.860 | 0.010*** | 0.881 | 0.021*** |
Physical conditions | 0.948 | 0.011*** | 1.071 | 0.010*** | 1.046 | 0.019* |
Benefits | 1.061 | 0.021** | 0.960 | 0.014** | 0.786 | 0.021*** |
Training | 1.137 | 0.020*** | 0.906 | 0.013*** | 1.013 | 0.026 |
Race | 1.004 | 0.041 | 1.466 | 0.047*** | 0.417 | 0.035*** |
Gender | 0.692 | 0.019*** | 1.031 | 0.026 | 0.610 | 0.031*** |
Age stratum 2 | 1.41×107 | 1.09×1010 | 0.849 | 0.038 | 0.882 | 0.086*** |
Age stratum 3 | 2.56×107 | 1.98×1010 | 0.520 | 0.023*** | 0.417 | 0.041 |
Age stratum 4 | 3.46×108 | 2.68×1011 | 0.264 | 0.012*** | 0.365 | 0.036*** |
Age stratum 5 | 1.04×109 | 8.05×1011 | 0.130 | 0.010*** | 0.232 | 0.031*** |
Status stratum 2 | 1.262 | 0.033*** | 1.249 | 0.032*** | 1.426 | 0.071*** |
Status stratum 3 | 1.186 | 0.162 | 0.774 | 0.226 | 9.172 | 1.567*** |
Agency-specific effects | Included | |||||
N | 120,547 | |||||
LR χ2 | 36552.26 | *** | ||||
Log-likelihood | −73338.0 | |||||
Pseudo R2 | 0.1995 |
***Indicates significant at better than .001 (two-tailed test).
**Indicates significant at better than .01 (two-tailed test).
*Indicates significant at better than .05 (two-tailed test).
All of the effects are represented as relative risk ratios. Multinomial logit is a nonlinear estimation routine, so the effect of an independent variable on the probability a person expresses a specific intent to leave varies across the range of that variable. We estimate the effect of a variable on the probability a dependent variable takes a specific, unordered value. We interpret effects as the change in probability of stating an exit-specific intent to leave. We also account for heterogeneity that depends on unobservable agency-level characteristics. The model fits well.
First, we find that satisfaction with the organization reduces the likelihood of stating intention to leave for all three exit options. Indeed, high organizational satisfaction translates into a much lower risk of leaving for a position outside government (i.e., leave federal government) than in leaving for a position in another agency (i.e., leave agency). A Wald test of the equality of the coefficients for the satisfaction variable in the “leave agency” and “leave government” equations shows that the latter is more negative (lower in terms of relative risk ratios) (Wald χ2 = 10.15). Similarly, a Wald test of the equality of the coefficients for the satisfaction variable in the “retire” and “leave agency” equations shows that the latter is more negative (lower in terms of relative risk ratios) (Wald χ2 = 8.55). This is robust evidence for the dependence of exit on perceptions of the ascent or decline of the organization. Figure 1 shows the estimated effects of satisfaction with the organization across its entire range on the likelihood of stating intention to leave for the three exit options. The curves show how the probabilities of the average respondent answering “retire,” “leave agency,” or “leave government” fall as satisfaction with the organization goes from its minimum (1) to its maximum (5), given that all other variables are at their means. Note that the probability of stating intent to leave to move to another agency falls from around 0.17 to 0.05, and that this change is larger than those for the other two response categories.
More importantly, we find that respondents are less likely to reply that they intend to exit when they perceive that management has sought to expand voice and encourage employees to feel loyalty. With regard to Hypothesis 2 above, table 3 also shows that the effects of voice are the same for both the “leave agency” and “leave government” equations, although the effect is not significant for the “retirement” equation. A Wald test of the equality of the coefficients for the voice variable in the “leave agency” and “leave government” equations shows that the latter is more negative (lower in terms of relative risk ratios) (Wald χ2 = 21.26). Similarly, a Wald test of the equality of the coefficients for the voice variable in the “retire” and “leave agency” equations shows that the latter is more negative (lower in terms of relative risk ratios) (Wald χ2 = 73.39). Figure 2 presents the estimated effects of voice across its entire range on the likelihood of stating intention to leave for the three exit options. The curves show that the probabilities of the average respondent answering “retire,” “leave agency,” or “leave government” fall as satisfaction with enhancing voice goes from its minimum (1) to its maximum (5), given that all other variables are at their means. Like the case of satisfaction with the organization, the probability of stating intent to leave to move to another agency falls from .14 to around .05, and this change is larger than those for the other two exit options. The probability of retirement also falls slightly.
Table 3 shows that the effects of loyalty are also in the predicted direction and significantly different from zero, and that the effects vary across exit option. The magnitudes of these effects are smaller than for satisfaction with the organization, but are important when one considers which factors are more available to managers concerned about exit from the organization. Interestingly, a Wald test of the equality of the coefficients for the loyalty variable in the “leave agency” and “leave government” equations shows that we can only distinguish between the two at the 0.10 significance level (that they are not so different in terms of relative risk ratios) (Wald χ2 = 3.52). Similarly, a Wald test of the equality of the coefficients for the loyalty variable in the “retire” and “leave agency” equations shows that the two are statistically equivalent (the same in terms of relative risk ratios) (Wald χ2 = 0.07). Figure 3 presents the estimated effects of loyalty across its entire range on the likelihood of stating intention to leave for the three exit options. The curves show that, as in satisfaction with the organization or voice, the probabilities of the average respondent answering “retire,” “leave agency,” or “leave government” fall as loyalty (or commitment) to the organization goes from its minimum (1) to its maximum (5), given that all other variables are at their means. The probability of stating intent to leave to move to another agency falls from .11 to around .07, and this change is bigger than those of other two exit options. The probability of stating to intent to leave federal government also falls slightly.
Table 3 shows unusual variations in the effects of pay—that pay satisfaction reduces the likelihood a respondent reports an intention to leave, except in the case of retirement. Pay satisfaction increases the likelihood of planning retirement. In the retirement model, however, the effects of pay may not be strange in that high satisfaction with pay likely picks up a sense of a healthy retirement account and could lead to an increase in retirement plans. Table 3 shows that the effect is significant at the 0.001 level (two-tailed test). In contrast, pay satisfaction reduces the odds of leaving the agency or leaving the government for another position. A Wald test of the equality of the coefficients for the pay variable in the “leave agency” and “leave government” equations shows that the latter is more negative (lower in terms of relative risk ratios) (Wald χ2 = 93.12). Not surprisingly, a Wald test of the equality of the coefficients for the pay variable in the “retire” and “leave agency” equations shows that the latter is more negative (Wald χ2 = 318.82). This is robust evidence for the dependence of exit outside government on pay satisfaction. No study has ever modeled all three exit options together and assessed the role of pay on intent to leave using this kind of nonlinear model. It is possible that other articles fail to find similar effects due to differences in modeling strategies. Figure 4 shows the estimated effects of pay satisfaction across its entire range on the likelihood of stating intention to leave for the three exit options. The curves show that the probabilities of the average respondent answering “leave agency” and “leave federal government” fall as satisfaction with pay goes from its minimum (1) to its maximum (5), given that all other variables are at their means. The probability of stating intent to leave to move to another agency falls from .12 to around .07, and this change is bigger than those for the other two exit options. As noted earlier, pay satisfaction appears to increase the likelihood of stating “retirement.”
We also want to point out that the control variables reveal other effects. Increases in the benefits index reduce the odds of exit to another agency or outside government (but curiously increase the probability of retirement). Increases in the training index increase the odds of retirement or exit outside of government (but reduce it for exit to another agency). Merit-based promotion reduces exit (except for retirement, in which case there is no effect); better physical conditions reduce the odds of exit to another agency or outside government (but curiously increase the odds of retirement). Minorities have a greater odds of exit to other agencies (but lower risk to leaving the government) than do white employees. Women’s risks of exit are lower in the case of retirement and leaving government employment.
We also note that significant differences remain even after we account for status-level motivations for exit, with those in the second stratum (i.e., supervisor/manager) more likely to retire, to leave for another agency, and to leave government employment. Age strata effects are also present in the case of leaving for other agencies and leaving government (but not for retirement; in that case, although the estimated effects are huge, we have great uncertainty about their magnitudes).
Is the coefficient for loyalty larger than that for voice? The Wald test rejects the null hypothesis of no difference in the effects of voice and loyalty in the exit option of leaving government (Wald χ2(1) = 47.16) and in the exit option of leaving for another agency (Wald χ2(1) = 26.02). Is the coefficient for voice larger than that for pay satisfaction? For the exit option of leaving government, the coefficient for voice is more negative than that for pay at the 0.001 level (Wald χ2(1) = 17.75); for the case of the leaving government employment, the coefficient is more negative than that for pay only at the 0.10 level (Wald χ2(1) = 3.45). Is the coefficient for loyalty larger than that for pay satisfaction? For the case of the leaving government employment, we can reject the null that the coefficients are the same at a conventional level (Wald χ2(1) = 65.90); the same is in the case of leaving for another agency (Wald χ2(1) = 11.22). Together these results suggest that in aggregate, voice and loyalty are more important than pay.
DISCUSSION AND CONCLUSION
We assess the context of voluntary exit from organizations in the public workforce when respondents have multiple exit options such as “retirement,” “leaving agency,” and “leaving federal government,” by focusing on exit, voice, and loyalty. Our approach follows other studies to test how loyalty and voice affect a person’s stated intention to leave in the context of the 2008 FHCS. We find that satisfaction with the organization reduces the likelihood of stating intention to leave for all three exit options. We also find compelling evidence for the reduction in the likelihood of exit based on individual-level perceptions of management’s dedication to both expanding the use of voice and enhancing employees’ sense of loyalty to the organization. The effects of voice are the same for both the “leave agency” and “leave government” equations, although the effect is not significant for the “retirement” equation. This result can be connected to formal theories of delegation such as Gailmard and Patty (2007) and Gailmard (2010) which are emphasizing policy-motivated agents.4 Empowering public employees through increasing voice makes them perceive that they have greater discretion and decide to opt into or stay in employment in the public sector. That is, giving federal employees the perception that they have voice can restrain their mobility. The effects of loyalty are also in the predicted direction and significantly different from zero, except for the “leave federal government” category. In addition, pay satisfaction reduces the likelihood a respondent reports an intention to leave, except in the case of retirement. We also note that significant differences remain even after we account for status-level and age strata motivations for exit.
Hirschman’s theory narrows the gap between study environment and causal mechanism. A distinct contribution of this study may be in the realization that the effects of voice, loyalty, and pay vary across exit options. What is interesting is that pay and the EVL components have similar effects for the likelihood of leaving government and moving to other agencies, although they have different effects for the likelihood of planning retirement. Especially, empowering employees for increasing voice, enhancing commitment for increasing loyalty, and pay satisfaction reduce much more the likelihood of the intent to move to another agency, compared to the other two exit options.
We know that our measures of Voice and Loyalty are neither perfect nor direct—that they probably include some “measurement error,” which is a common situation of empirical research in the world of public management and may carry substantive consequences. Here, we want to discuss the methodological implications of this problem for our study and those of other researchers. Specifically, one option is to bootstrap a model like this one, but there are a number of caveats. For one thing, model size matters because of the computational intensity of the process. A second issue is that conditional on model size, such methods are especially difficult in nonsmooth distributions because of the possibility that a bootstrap draw may have no presence of a given choice from the unordered polychotomous set. Other methods are interesting options and should be encouraged in the public management literature. For instance, errors-in-variables approaches add a priori information into the estimation process vis-à-vis measurement uncertainty, or principled approaches exist like simulation-extrapolation (SIMEX). Unfortunately, such approaches are very difficult for models like ours. The main reason is that, unlike a number of other articles that have studied data coming from OPM, we include sampling weights. Our sampling weights (really, a type of “replicate weight”) are becoming typical when one worries about privacy concerns of respondents (as is true in the case of these large federal surveys where agency information is included for small agencies). Such factors make it difficult to account for measurement error because bootstraps cannot pull from the typical target distribution but instead from the distribution as sampled after accounting for these weights.
However, the central contribution of this article is to broaden the focus of our attention from just “leaving the organization” to “going to a specific place.” In the end, people leave organizations largely because they are going to other, expected destinations. We care not that people are traveling but that they are going somewhere. Our approach centers on the destination, although it might be pointed out that some people leave for reasons outside their own control. These people do not “intend” to leave, but are “intended” by another party, perhaps a manager. We believe that sort of research would be appropriate for future research projects.
FUNDING
National Research Foundation of Korea Grant funded by the Korean Government (NRF-2012-047320130014 S-Y-L).
REFERENCES
———.
———.
Andrew Whitford: This is a d3lab product. Andrew Whitford thanks Tony Bertelli for his many comments on this article.
We do not concentrate on “leadership” because it often depends on enhancing voice and loyalty (Ashour 1982; Larson 1986; Locke and Latham 1990; Mumford 1986; O’Driscoll and Beehr 1994; Tjosvold 1984; Williams and Hazer 1986).
We include all these control variables to avoid excluded variable bias. In this article, we are unable to address dynamic effects since our data are not coded to the individual respondent.
We want to be clear that our measures of Voice and Loyalty are not perfect or direct—that they are coarse representations of the range of considerations that Hirschman packaged into theoretical framework. Of course, this is not unusual in empirical research that bridges past theoretical work with specific cases drawn from the world of public management. However, we note that this problem is frequently encountered when using existing data like the Merit Principles Survey or FHCS, which was specifically constructed for purposes other than the illumination of academic concerns about human motivation.
Policy-motivated agents are those who “attach intrinsic concern to public policy in the sense that they will sacrifice other important goals to improve it” (Gailmard 2010, 37).