Excerpt from:
Innovative Theory and Empirical Research on Employee Turnover

Rodger Griffeth, Georgia State University
Peter Hom, Arizona State University

Pages 3-32. Copyright 2004 by Information Age Publishing. All rights of reproduction in any form reserved.


James L. Price

This chapter's focus is on strategies for the construction of causal models. The paper has two parts. First, a description is offered of research conducted by Price/Mueller and their colleagues at the University of Iowa regarding the development of a causal model of voluntary turnover. Five research projects are described. Second, Price and Mueller's research is then used as basis for discussion of strategies for the construction of causal models. It is suggested that more papers regarding strategies should be written.

The purpose of this paper is to describe the development of a causal model of voluntary turnover. Voluntary and involuntary turnover are usually distinguished. Employees who leave an organization at their own discretion are examples of voluntary turnover (Price, 1977). "Quits" is a common designation for these employees. Dismissals, exits due to serious illness, and deaths are examples of involuntary turnover. Retirements may be either voluntary or involuntary turnover.

The model of turnover, whose development is described, is often referred to as the "Price-Mueller" model. This model is one of the three major explanations of turnover in the literature (Hom & Griffeth, 1995). The other two models are those proposed by Mowday and his colleagues (Mowday, Porter, & Steers, 1982) and Mobley (1982).(1)

The model's development will be described historically. Five phases of development will be indicated. After the development is described, reflections will be offered regarding the process. The focus of the reflections will be on strategies of causal model construction in the field of turnover research. Throughout the description "variables" and "determinants" will be used interchangeably.

In the Spring of 1972, Price and a small number of sociology graduate students at the University of Iowa reviewed literature which offered explanations of voluntary turnover. The purpose of this review was to develop a preliminary causal model of voluntary turnover. Models proposed by economists dominated the literature review (some examples are Burton & Parker, 1969; March & Simon, 1958; Pencavel, 1970; and Stoikov & Raimon, 1968). The work of psychologists, however, was also prominent (Farris, 1971 and Lyons, 1968 are illustrations).

The question arises as to why a preliminary model was sought. It would have been possible, for instance, to have estimated one of the economic models. However, based on his acquaintance with the turnover literature, Price was of the opinion that the economic models, which dominated the literature, focused on too narrow a range of determinants to explain turnover adequately. The monetary variables emphasized by the economists, for instance, seemed to be important, but needed to be supplemented by non-monetary determinants. Economic explanations also ignored the process whereby turnover was generated. The economists, for example, suggested the amount of monetary income as a determinant: the more money received, the less the likelihood of turnover. However, they did not indicate how monetary income impacted on turnover. Increased money received may, for instance, decrease turnover by increasing the employees' job satisfaction. It seemed necessary, therefore, to attempt to develop a more inclusive model of turnover than those proposed by economists-thus the 1972 review.

The review resulted in a series of summaries and critiques of the literature. There was a summary and critique for each piece of literature reviewed. These summaries and critiques posed a vexing set of problems. Different terminologies were used to describe the various determinants; some of the proposed determinants were overlapping; a causal order was not proposed for the determinants; a beginning for the causal sequence was not specified; and some of the determinants were not empirically well supported. A sizeable amount of work thus remained to be done to develop a preliminary model. The summaries and critiques, by themselves, were clearly not sufficient.

During the 1972-1973 academic year, Price was on leave from the University of Iowa as a research professor at Bradford University in Bradford, England. During this time he transformed the summaries and critiques into a preliminary model. The model will now be described (Price, 1975). Figure 1.1 constitutes a diagram of the model.

The model had four exogenous variables: pay, primary group, communication, and centralization. Pay was monetary income and is the type of variable commonly emphasized by economists. Two moderation conditions were proposed for pay.(2) Pay will not be a significant determinant unless it is important to the employees and is perceived to be high. The use of "moderating conditions" illustrates the contingency approach to the study of organizations (Woodward, 1965). The second determinant was not assigned a label but was simply referred to as "participation in a primary group." Kinship-type systems are examples of primary groups whose organizational importance, though not under this label, was emphasized by the Western Electric Research (Roethlisberger & Dickson, 1939). The primary group label came from scholars at Columbia University (Merton, 1968). Communication was the transmission of information among the members of an organization and centralization was the distribution of power within an organization. Turnover from an organization, it was hypothesized, was likely to be low if pay, primary group participation, and communication were high. Pay, however, must be important to the employees and be perceived as high. High centralization was hypothesized to increase turnover.

The model had two intervening, or process, variables, job satisfaction and opportunity. Job satisfaction was defined as a positive affective orientation toward the organization and opportunity as the number of jobs in the environment. The two intervening variables, however, impacted on turnover quite differently. Job satisfaction was believed to mediate the impact of the exogenous variables, whereas opportunity was a moderating variable. If, for instance, the exogenous variables produce more dissatisfaction than satisfaction, and if there are jobs in the environment, then turnover is likely to occur. This, of course, assumes that pay is important and is perceived to be low. The model stressed the necessity of examining the balance of satisfactions and dissatisfactions. This balance was referred to as "costs and benefits." More dissatisfactions than satisfactions, for instance, means a situation that is costly to the employees. It is assumed that employees will attempt to leave a costly situation.(3) Opportunity was accompanied by two conditions. If there were many jobs in the environment, the employees must know about these jobs and be free to leave before turnover will occur.
Two of the six variables in the model-pay and opportunity-are those commonly proposed by economists. However, the economists do not accompany pay and opportunity with any conditions and they do not specify the intervening process between their determinants and turnover.

Five variables were excluded from the model because of insufficient evidence: role clarity, programmed coordination, inequity, industrial concentration, and size. Role clarity was precise definition of the work that was to be done, programmed coordination was central management of work, inequality was the lack of fairness in the distribution of rewards, industrial concentration was the extent to which the output of an industry was produced by a small number of organizations, and size was scale of operation.(4) Turnover decreases with more role clarity, greater industrial concentration, and more programmed coordination. Inequity, however, increases turnover. No proposition was advanced for size.

The next phase of the research was to estimate the preliminary model. Price decided to estimate the model with a sample of registered nurses employed in a voluntary, short-term, general hospital. Registered nurses were a good sample because their turnover rate was comparatively high in the Fall of 1973, often reaching 50% a year. Hospitals were a good site because Price could perform the research under the auspices of the College of Medicine, where he had a joint academic appointment. Voluntary, short-term, general hospitals were a good site because they were the most common type of hospital in the United States. Price obtained permission, in the Fall of 1973, to study the Department of Nursing in a hospital in Iowa City, the location of the University of Iowa.

A pilot study was necessary because Price knew very little about nurses or hospitals. It was decided that most of the data to estimate the model would be collected by a questionnaire. Before the questionnaire could be constructed, Price had to observe and interview in the Nursing Department for several months. He spent so much time observing and interviewing that he was designed by the Nursing Director as "an honorary nurse." The pilot study also provided the opportunity to develop a questionnaire. The plan was to use the questionnaire in a later, large-scale study of nurses.

There were continuities and changes in the pilot study from the preliminary model. Figure 1.2 provides a diagram of the model estimated (Price & Bluedorn, 1980).

Six variables from the preliminary model were included in the estimation: pay, primary group, communication, centralization, satisfaction, and opportunity. Pay was again defined as monetary income and its importance was again a condition. However, no mention was made about the perception of pay's level. Based on the work of Blau (1959-1960), primary group was termed "integration." This variable previously had no label. Communication was narrowed to the transmission of job-related information and was designated as "instrumental communication." Pay, integration, instrumental communication, and centralization continued as exogenous variables; satisfaction and opportunity were again viewed as intervening variables. Satisfaction and opportunity continued as mediating and moderating variables, respectively.

Routinization and distributive justice were added to the model as the result of another review of the literature conducted for a book on turnover (Price, 1977). Repetitive work was defined as routinization and distributive justice was conformity to organizational norms followed by positive sanctions. Distributive justice was previously excluded from this preliminary model developed in England due to insufficient data. "Inequity"-the earlier term-was not widely used as a label in the literature as distributive justice. The idea was to standardize labels as much as possible. Widely-used labels, it seemed, would result in greater standardization than narrowly used labels.

Changes were also made in the pilot study regarding the four variables excluded from the preliminary model: role clarity, programmed coordination, industrial concentration, and size. The conceptual content of role clarity seemed to be captured by instrumental communication; programmed coordination's conceptual content appeared to be caught up by centralization; industrial concentration looked like a demographic variable, a term to be discussed shortly; and size still lacked sufficient supporting data. Role clarity and programmed coordination were thus included in the pilot study but under different labels.(5)

The nurses were professional employees, a fact which prompted an examination of the sociological literature about professions. This literature seemed to suggest that highly professionalized employees generally experienced greater turnover than less professionalized employees. It was more difficult, for instance, for the employer to meet the high standards required by highly professionalized employees. Professionalism was defined as the extent to which the members of a profession conformed to its norms. The nurses in the pilot study appeared likely to differ in professionalism, since their training was quite varied. Some nurses were university graduates (12%), a few were graduates of community colleges (10%), and most were trained in hospitals (78%).

During the pilot study it became apparent to Price that there were two distinct categories of nurses employed by the hospital. The first category consisted of nurses who had been born in the community, were members of a local Roman Catholic church, were trained in the nursing school run by the hospital, and were married to local men, when they were married. The hospital studied was affiliated with the Roman Catholic church and all the nurses were women. This first category of nurses might be termed "locals." The second category consisted of nurses whose husbands were obtaining education or training at the University of Iowa, were also usually born outside the community, less likely to be Roman Catholics, were not trained in the hospital's nursing school, and were not married to local men, when they were married. This second category of nurses might be termed "non-locals." The last three variables of the model-community participation, social class, and work commitment-were an attempt to capture conceptually the key variables which seemed to distinguish the local and non-local nurses.

Community participation was involvement in the life of the non-hospital environment, social class was the community's prestige structure, and work commitment was the extent to which work was the central life interest of the employee. Local nurses were expected to score higher than non-local nurses on community participation, social class, and work commitment. Work commitment was expected to be lower for the non-local nurses, since they had the reputation of quitting their jobs to go with their husbands when their spouses had completed their university education or training, thus indicating greater commitment to a kinship role (wife) than an occupational role (nurse).

Demographic variables, such as age and seniority, were distinguished and excluded from the model. "Correlates" was the label for the demographic variables in the reported research. Demographic variable was a more widely used label than correlates. The demographic variables were included in the analysis to check the explanatory power of the model (Price, 1995; Price & Kim, 1993).

As previously indicated, this site for the pilot study was a voluntary, shortterm, general hospital. The hospital was of medium size (about 250 beds). No attempt was made to generalize the results of the research to government or specialty hospitals. Only non-supervisory, registered nurses were included in the sample. All administrative employees in the Nursing Department-managers and clerks-were excluded from the sample. The N for the nurses was 130. Systematic data to estimate the model were collected by a mailed questionnaire administered to the nurses by Price in February 1974 and returned to the University of Iowa. Turnover data were collected from the hospital in February 1975. The sample was divided into "leavers" and "stayers." Since none of the nurses left because of death, retirement, or dismissal, it was likely that most of the nurses voluntarily left the hospital. The sample consisted of 98 stayers and 32 leavers, a 75/25 split. Data were analyzed by regression and path analytic techniques.

The most important determinants, in terms of total effects, were the following: satisfaction (.30), professionalism (.26), integration (-.24), pay (-.29), distributive justice (-.23), and routinization (.15). Turnover was increased by professionalism and decreased by integration, pay, distributive justice, and routinization. All of these results were expected. It was not anticipated, however, that satisfaction would increase turnover. This unanticipated result may have been due to the local and non-local types of nurses in the sample. Many of the non-local nurses would ordinarily have been stayers. However, when their husbands finished their university training or education, the non-local nurses quit their jobs to go with their spouses to a new location. The adjusted explained variance, with the demographic variables included, was 33%.

As previously indicated, demographic variables were included in the analysis to check the explanatory power of the model. With a sophisticated model and psychometrically sound measures, the demographic variables should not be important. The conceptual content indicated by demographic variables should be captured by the theoretical variables of the model. However, two demographic variables-length of service (-.37) and age (-.24)-had substantial total effects. The effects for length of service (seniority) and age were in agreement with the literature. Time worked, whether full time or part time, had a relatively small total effect (-.12). The adjusted explained variance, without the demographic variables, was not specified but it was probably fairly high, to judge from the total effects of length of service and age. The size of the demographic variables in the pilot study did not auger well for the model's power.

The plan was to expand the pilot study after it was completed. When the pilot study was finished, nursing turnover was still high and voluntary, short-term, general hospitals remained the dominant type of patient-care facility in the United States. Price continued to hold his joint appointment with the Medical College, so it was possible to perform another study with official health sponsorship. The completion of the pilot study also provided a track record which facilitated obtaining research grants from the Division of Nursing of the Bureau of Health Manpower and the American Nurses' Foundation. The assistance of Dr. Myrtle Aydelotte, who agreed to serve as the Principal Investigator for the Iowa-Illinois study, was the major factor in obtaining the research grants. Dr. Aydelotte had previously been Dean of the College of Nursing at the University of Iowa and Director of the Nursing Department of the University of Iowa Hospitals and Clinics. Before the research could begin, Dr. Aydelotte relinquished her involvement with the research to become Executive Director of the American Nurses' Association. Price then became the Principal Investigator of the project. Dr. Charles Mueller joined the research project. Dr. Mueller was in the Sociology Department at the University of Iowa and specialized in statistical analysis. Early in 1976, Price obtained permission to study nonsupervisory registered nurses in seven hospitals located in Iowa and Illinois. Five of the hospitals were located in Iowa and two were in Illinois. The hospitals were all voluntary, short-term, general units.

There were continuities and changes in the new model to be estimated (Price & Mueller, 1981). Figure 1.3 contains a diagram of the model.

The following seven exogenous variables were the same in the pilot study and the Iowa-Illinois Study: routinization, centralization, instrumental communication, integration, pay, distributive justice, and professionalism. "Participation" was substituted for centralization to make the label consistent with the definition of centralization. Centralization was defined as the power exercised by an employee in his or her job-an individual focus-and the new label had an individual rather than an organizational reference. Art Brief, a faculty member in the College of Business at the University of Iowa, suggested the use of the participation label. Brief (1976) had conducted research on nursing turnover. The possibility that pay's importance may vary was discussed but not estimated. No mention was made of the perception of pay's magnitude.

There were four substantial changes in the Iowa-Illinois Study, compared to the pilot study. First, opportunity was changed from a moderating variable to an exogenous variable. This change was suggested by Allen Bluedorn (1976) as the result of his turnover research. Bluedorn had also worked with Price in the analysis of the data from the pilot study and had participated in the Spring 1972 review of turnover models. No mediating variables were suggested between opportunity and turnover. Second, promotional opportunity-the chance to advance in the organization-was added as an exogenous variable. The literature on "internal labor markets" suggested promotional opportunity as a determinant (Osterman, 1984). Satisfaction was hypothesized to mediate the relationship between promotional opportunity and turnover. Promotions bring increased pay, power, and prestige and this should result in more satisfaction. The third change in the exogenous variables was the addition of general training, that is, the ability of increase productivity in different organizations. General training comes from the human capital tradition of research in economics (Becker, 1964). The impact of general training on turnover was hypothesized to be through intent to stay, a concept to be discussed shortly. Kinship responsibility was the fourth new exogenous variable. The pilot study suggested that local and non-local nurses appeared to differ by community participation, social class, and work commitment. However, in the empirical data, none of these variables turned out to be a statistically significant determinant of turnover. Kinship responsibility was proposed as another variable to capture an apparent conceptual difference between the local and nonlocal nurses. Price's fieldwork had already indicated that the local and nonlocal nurses appeared to be quite different and he wanted conceptually to capture this difference. Kinship responsibility was defined as obligations to relatives living in the community. The local nurses, it was hypothesized, should have more kinship responsibility than the non-local nurses. Kinship variables were not commonly studied by the economists and psychologists who did most of the research on turnover. However, scholars interested in migration, who were often sociologists, stressed the importance of kinship variables in migration. Since migration is conceptually similar to turnover (both concepts involve movement out of social systems), kinship responsibility seemed to be an interesting determinant to explore. Kinship responsibility was hypothesized to impact on turnover through intent to stay. No specification was made as to how kinship responsibility would increase intent to stay.

Satisfaction continued as mediating variables. Intent to stay, however, was added as a variable that mediated the relationship between satisfaction and turnover. Intent to stay has an interesting history in the development of the model.

Graduate sociology students who planned to do their dissertations on turnover at the University of Iowa did not relish the idea of waiting for turnover to occur in the organizations they were studying. The practice was first to ask the employees to describe, by questionnaire, their work situations and then wait for the employees to quit. The content of these descriptions was indicated by the variables of the model estimated. The period of waiting was usually about a year. A speeding up of the process was needed by the students. Price discovered a major study of turnover of priests by Greeley (1972) who measured not turnover but intent to turnover. Greeley provided no data to support a link between intent and turnover. Intent to turnover became intent to stay in the Iowa-Illinois Study. Intent to stay was given scholarly legitimacy when Mobley (1982) used it as intervening variable in his model. Mobley hypothesized that intent was the variable immediately preceding turnover.

During the course of the Iowa-Illinois Study, Price became aware of substantial data supporting commitment as a possible determinant of turnover. Mowday and his colleagues (Mowday et al., 1982) proposed commitment as an intervening process and suggested that it was more important than satisfaction. Intent to stay seemed, to Price, to be a dimension of commitment and thus provided a means to link the Iowa-Illinois Study to the research conducted by Mowday and his colleagues. Mowday et al.'s research was part of the expectancy tradition which had its contemporary origins in the work of Vroom (1964).

As with the pilot study, demographic variables were excluded from the model. The demographic variables, termed "correlates" in the Iowa-Illinois Study, were used as controls in the analysis. Three demographic variables were used as controls: length of service, age, and amount of time worked (whether full time or part time).

The sample and site of the Iowa-Illinois Study were the same as the pilot study. As intended, however, the Iowa-Illinois Study was larger than the pilot study. Eleven hundred and one nurses returned questionnaires in the Iowa-Illinois Study; these nurses, as previously indicated, were distributed among seven hospitals. As before, most of the data were collected by questionnaires mailed to the homes of the nurses and returned to the University of Iowa. (6) Questionnaires were mailed to the nurses' homes in August 1976 and turnover data were collected from the hospitals in October 1977. Interviews with hospital personnel were used to eliminate from the sample all nurses who had left involuntarily. The final sample consisted of 880 stayers and 221 voluntary leavers, an 80-20 split. Factor analysis was used to construct the measures and to assess validity. Reliability was evaluated with coefficient alpha.(7) The resulting measures possessed adequate psychometric properties. As with the pilot study, analysis was conducted with regression and path analytic techniques.

The five most important determinants, in terms of total effects for the entire sample, were intent to stay (-.37), opportunity (.16), general training (.13), satisfaction (-.10), and kinship responsibility (-.07). Turnover was increased by opportunity and general training but decreased by intent to stay, satisfaction, and kinship responsibility. All of these results were consistent with the model. The results for satisfaction were especially encouraging, since its outcome had been inconsistent with the model in the pilot study. The adjusted explained variance for turnover, including the demographic variables, was 18%.(8)

Total effects were given for amount of time worked but not for age and length of service. The total effects for amount of time worked was not comparatively large (.01). The three demographic variables only added 1% to the adjusted explained variance, the type of result that augers well for the explanatory power of this model.

Reflections suggested three features of the Iowa-Illinois Study that should be modified. First, too much time has elapsed-fourteen months-between the collection of data about the nurses' situation at time one and the collection of the turnover data. Since the data at time one were used to explain turnover, the situations of the nurses might have changed substantially by the time the turnover data were collected. There was a problem here, however. If the time elapsed was to be substantially shortened, a situation must be found with very high turnover if one was going to obtain an adequate split between the leavers and stayers. The Iowa-Illinois Study had an 80-20 split between the leavers and stayers and future research, it was thought, should not go much below this, to a 90-10 split, for instance. A 50-50 split is, of course, the preferred ratio. It took fourteen months to get the 80-20 split in the Iowa-Illinois Study.(9) Second, the focus on non-supervisory nurses was helpful in that occupation and supervisory position were controlled. Since these two variables commonly produce many differences in the employees, it was helpful that they were controlled. However, the focus on non-supervisory registered nurses probably created too homogenous a sample to detect variations in the determinants of turnover. Pay, for example, despite its massive support in the literature, was not a significant determinant in the Iowa-Illinois Study. The lack of significance for pay may have been due to the fact that there was little variance in pay in this sample. Third, since all the nurses in the Iowa-Illinois Study were female, this raised questions about kinship responsibility as a determinant. Females were the traditional providers of kinship services, so perhaps kinship responsibility was a determinant due to the female composition of the sample. More males were needed in the next study. Price and Mueller arranged to study five hospitals in the area of Denver, Colorado to make the changes which needed to be made.

Professor Sam Levey, Chair of the Graduate Program in Hospital and Health Administration at the University of Iowa, joined the Denver project. Professor Levey was especially interested in hospital administration and was a major reason why Federal funds were obtained from the National Center for Health Services Research. Professor Levey also provided a site for Price and Mueller to do their research at the Medical School.

The expectations for the Denver Study were not fully realized. A recession was experienced shortly after the Denver Study was begun in the Summer of 1980. What had been a booming economy in the Denver area was transformed into a recession and turnover rates were depressed throughout the region. More variance, however, was obtained in the determinants, since all employees in the hospitals were sampled and the sample became somewhat less dominated by females. The new sample, however, was still predominantly female, since 88% of the employees in the five Denver hospitals were female. A plus for the Denver study was that three of the hospitals were small (under 100 beds) and two were of medium size (between 100 and 500 beds), thereby making it easy to include size as a possible determinant. Size had previously been excluded because it supporting evidence was mixed.

Most of the variables and propositions used in the Denver Study (Price & Mueller, 1986) were the same as those used in the Iowa-Illinois Study: opportunity, routinization, centralization, instrumental communication, integration, pay, distributive justice, promotional opportunity, professionalism, general training, satisfaction, and intent to leave. "Participation" was changed to "centralization," integration became close friends in the immediate work unit rather than close friends somewhere in the hospital, and "intent to leave" replaced previous "intent to stay." Figure 1.4 is a diagram of the model used in the Denver Study.

Three changes were made in the model used in the Denver Study. First, role overload-excessive work demands-was added as an exogenous variable. Role overload was included because hospital personnel emphasized its importance during the field work Price conducted prior to the administration of the first questionnaire. There was also data in the literature supporting role overload, conceptualized as a dimension of job stress, as a determinant of turnover (House, 1981). Second, commitment was added as a variable that intervened between satisfaction and intent to leave. Following the work of Mowday and his colleagues (Mowday et al., 1982), commitment was viewed as employee loyalty to the organization. Increased commitment was hypothesized to decrease turnover. Third, as previously indicated, size was also added as an exogenous variable. Hospital and work unit size were examined, but no propositions were proposed.(10)

The number of checks for moderating variables were increased in the Denver Study.(11) Moderating checks were made for all the exogenous variables. It is possible, it was hypothesized, that the importance of all the exogenous variables could vary for the employees. The previous focus had only been on the importance of pay for the employees.

Like the Iowa-Illinois Study, the Denver Study used the demographic variables as controls and not as components of the model. Fifteen demographic variables are used as controls-the most ever.

All of the employees in the five hospitals were included in the sample. Physicians working in the hospitals were not considered employees. The sample consisted of 2,152 employees. One hundred and twenty-nine work units were also analyzed, because work unit results can differ from individual results. The five hospitals were all voluntary, short-term, general units. The Denver Study included two questionnaire administrations, in November 1980 and June 1981. Turnover data were collected by mail from each hospital. During the field work, discussions were held with each hospital to help them distinguish voluntary and involuntary leavers. The sample analyzed had the following distribution: 1,748 stayers (80%), 404 voluntary leavers (18%), and 40 voluntary leavers (2%). An unsuccessful effort was also made to collect data about each of the employees who left. Preceding the first questionnaire administration, Price conducted field work in the five hospitals from June 1980 to November 1980. Factor analysis was again used to construct the measurements. As before, the measures used had acceptable psychometric properties. Regression and path analytic techniques were again used to analyze the data.

Seven determinants had substantial total effects on turnover: intent to leave (.32), satisfaction (-.11), pay (-.09), kinship responsibility (-.08), opportunity (.07), and integration (-.07). 12 All of the effects were in agreement with the model. Turnover was increased by intent to leave and more opportunity and decreased by high satisfaction, pay, kinship responsibility, commitment, and integration. Pay referred to individual and not family income. The adjusted explained variance for turnover, without the demographic variables, was 12%. None of the moderating variables were statistically significant.

Total effects were not computed for the demographic variables. However, the fifteen demographic variables increased the adjusted explained variance by 1%, the type of result which indicates that the model is capturing most of the conceptual content indicated by the demographic variables, the type of situation anticipated by Price and Mueller.

The finding that kinship responsibility again decreased turnover emphasized the need for a sample with more males, especially males delivering healthcare services. An opportunity to obtain such a sample came when Price obtained a U.S. Air Force Summer Fellowship in 1990. Price worked at Armstrong Laboratory, which was near Wilford Hall, an important military hospital located at Lackland Air Force Base. Price obtained permission from the hospital to study its military medical personnel. In addition to its large core of males employed as nurses, doctors, and dentists, the Air Force also made available demographic data about all the military medical personnel who completed the questionnaires. This demographic data eased the problem of data collection. All the military medical personnel in the hospital worked full time. The three pervious samples had consisted of both full time and part time employees.

There were continuities and changes in the Texas Study. Figure 1.5 is a diagram for the model estimated in the Texas Study (Kim, Price, Mueller, & Watson, 1996).

Eleven determinants and propositions were basically the same from the Denver Study: kinship responsibility, opportunity, general training, autonomy, distributive justice, pay, promotional chances, routinization, satisfaction, commitment, and intent to stay. There were two minor changes in labels: "centralization" became "autonomy" and "intent to leave" became "intent to stay." Autonomy was a widely used label and more accurately described the variable previously investigated as centralization in the Denver Study. All subsequent research used the label of autonomy for this concept. It is not clear why intent to stay-rather than intent to leave-became the label for the dependent variable. Since the Denver Study had started, new data indicated that intent to stay was moderately correlated (about .50) with turnover (Steel & Ovalle, 1984). Another reason for using intent to stay rather than turnover was the fact that the military medical personnel in the sample had signed contracts to remain in the Air Force to repay the government for their medical education. It was not unusual for these contracts to last for six years, a long time to wait for turnover data.

There were nine major changes in the model to be estimated. First, job motivation became a determinant. Price became convinced that the concept that had been investigated as "professionalism" was more accurately designated as "job motivation." Work commitment had been investigated in the pilot study but it was not statistically significant. However, a new measure was available (Kanungo, 1982), so work commitment was reintroduced as job motivation. The label of professionalism was dropped. Motivation was also a label widely used in the psychological literature. Job motivation was viewed as willingness to exert effort at work and was hypothesized to impact on intent to stay through satisfaction and commitment. No links were specified between job motivation and satisfaction/commitment. Second, met expectations was added from the work of Mowday and his colleagues (Mowday et al., 1982). For a long time, Price believed that expectations, a critical concept in the Parsonian-Mertonian tradition in which Price was trained, was identical to norms. One of the Iowa students (Jean Wallace), however, argued that two different concepts were involved and her arguments were accepted for the Texas Study.

Third, positive and negative affectivity were added as exogenous variables. These were dispositional-that is, psychological-concepts that referred, respectively, to the tendency to experience pleasant/unpleasant emotional states. The affectivity variables were hypothesized to impact on intent to stay through satisfaction and commitment. The hypothesized impact on satisfaction was suggested by Brief and his colleagues (Brief, Burke, George, Robinson, & Webster, 1988). No links were specified between the affectivity variables and satisfaction/commitment. Brief and his colleagues (1988) also argued that the affectivity variables must be controlled in the investigation of satisfaction, because they might contaminate the exogenous variables. Watson and Clark (1987) were the major scholars working with the affectivity variables.

Fourth, job hazards were added as an exogenous variable. Job hazards came from the research by Viscusi (1979) which indicated that turnover was increased by physically dangerous work. Factory work had historically been viewed as often hazardous but current research on hospitals had indicated that its work was often quite hazardous. An interesting feature of Viscusi's research was that its controls were almost exclusively demographic variables. Price and Mueller's research mostly used theoretical variables as controls and the Texas Study provided an opportunity to check Viscusi's analysis. With many theoretical controls, job hazards may not be a significant determinant. Fifth, role overload from the Denver Study was expanded to "job stress" and included three variables in addition to role overload: role inadequacy, role ambiguity, and role conflict. Job stress was viewed as the extent to which job duties cannot be fulfilled and was hypothesized to impact negativity on intent to stay through satisfaction and commitment. The previous emphasis on instrumental communication seemed to be conceptually captured by role ambiguity. Instrumental communication was thus no longer used as a label. The job stress label came from research done at the Survey Research Center of the University of Michigan (House, 1981).

Sixth, Mangelsdorff's (1989) work emphasized the importance of professional growth in a military setting. Since most of the Texas sample were professional employees, and since research on professionals stressed the importance of knowing the literature, this seemed to be an important determinant to explore. Mangelsdorff emphasized the importance of this variable in a discussion with Price in Texas. Professional growth was the degree to which the Air Force provided the chance to increase job-related knowledge and skills. With a good chance for professional growth should come increased intent to stay through satisfaction and commitment. The empirical data supporting professional growth, however, was not as extensive as the evidence for job hazards. Seventh, integration became a dimension of social support. Social support was viewed as assistance with jobrelated problems and could come from the family, supervisor, and peers. Peer support was previously termed "integration." Increased social support was hypothesized to increase intent to stay by its positive impact on satisfaction and commitment. The social support label also came from research performed at the Survey Research Center (House, 1981) and was a widely used label in the literature.

Eighth, search behavior was the extent to which an employee was looking for another job. The proposition was that the greater the search behavior, the greater the likelihood of decreased intent to stay. Search behavior was a variable emphasized by economists and, when combined with intent to stay, partially captures the conceptual content of Hom and Griffeth's (1995) "thoughts of quitting." Ninth, the Denver Study had intensively investigated the moderating variables of values without formally incorporating them into the model. The early research, for example, did not support values' hypothesized moderating effects. However, most of the turnover literature emphasized the importance of values' moderating effects, especially the research of Mowday and his colleagues, so they were finally incorporated into the model in the Texas Study, as Figure 1.5 indicates. It was also hypothesized that autonomy and social support would moderate the impact of job stress on satisfaction (Karasek & Thorell, 1990). Job stress, for example, it was hypothesized, will not decrease satisfaction if autonomy and social support are high. The moderating effect of autonomy and support was not formally incorporated into this model, however. There was more evidence for the moderating effect of values, it seemed, than for the moderating effect of autonomy and social support.

The Texas Study also divided the determinants into four classes: endogenous, environmental, individual, and structural. Satisfaction, commitment, search, and intent were endogenous variables, whereas kinship responsibility and opportunity were environmental variables. General training, job motivation, met expectations, and the affectivity determinants were individual variables. The remaining thirteen variables were classified as structural.

As in previous research, demographic variables were included in the analysis as controls but were not incorporated into the model. Four demographic variables were included in the analysis: education, rank, age, and length of military service obligations.

The model was estimated for a sample of male physicians to see how well it would explain intent to stay for a non-female sample (Kim et al., 1996). As previously indicated, all the previous estimates of the model had been made for females. The results for kinship responsibility were of special concern. There were 244 male physicians in the sample. The hospital at Lackland Air Force Base-Wilford Hall-was a tertiary care center with 806 beds, a large hospital. Data were collected by questionnaires and records. The questionnaires were mailed in July 1990 and returned in August and September to Armstrong Laboratory, where Price worked. Demographic data about the sample were obtained from military records. Price conducted intensive field work in the hospital for two months before the questionnaires were administered. Factor analysis was used to construct the measures, which turned out to have adequate psychometric properties. The data were analyzed with regression analysis and path analytical techniques.

Six determinants had substantial total effects: commitment (.42), satisfaction (.23), search behavior (-.23), opportunity (-.19), met expectations (.16), and positive affectivity (.12). There were no total effects for kinship responsibility. So with a male sample, kinship responsibility was not a significant determinant. Nor was job hazards a significant determinant. All of the significant effects were in agreement with the model. None of the moderating variables was statistically significant. The adjusted explained variance, with the demographic variables included, was 41%.

Total effects were not computed for the four demographic variables used in the analysis. Data also were not given regarding the amount of explained variances added by the demographic variables. The demographic variables were important early in the development of the model, but by this phase of the research, they did not yield much information, a result that was anticipated. Significant results for the demographic variables would indicate that the model was not working well.

Strengths and weaknesses of the developmental process will now be presented. The focus will be on strategies of model construction in the context of turnover research. Three strengths and seven weaknesses are apparent. Consider first the strengths.

First, reviews of the literature were helpful. A review of models, as was previously indicated, was first conducted in the Spring of 1972. When Price returned to Iowa in 1973, after his research professorship in England, he began a review of the entire field of turnover research-definitions, measures, models, empirical generalization, and extent of turnover (Price, 1977). Reviews are, of course, traditional in research, so there is nothing unusual here. What was a bit different about these reviews, from the perspective of turnover research, is that they were not bound by the confines of any single discipline or applied area. Everything pertinent to turnover was examined. Many turnover researchers, especially economists, stay very close to their disciplines or applied areas. As the descriptions indicated, smaller reviews continued during the entire process of model development.

Second, contact with other researchers working on turnover was basic to the process. The results of this contact are apparent throughout the descriptions of the developmental process. The importance of the research community-the famous "invisible college"-is, of course, traditional so there is nothing extraordinary here. What is somewhat different, again in the context of turnover research, is the range of contacts. Price sought to discuss turnover with everyone he could find who did research on the topic. He described the research at meetings, especially professional gatherings; he eagerly spoke to classes of students about the research; he commented, by letter and telephone, on publications and asked for additional material cited in the publications; he sent his publications to others to stimulate discussion; he sent large amounts of material about the Iowa research to individuals just beginning their study; to open discussions with managers, he spoke at meetings devoted to the reduction of absenteeism and turnover; and he served as a paid consultant on numerous turnover projects, especially projects studying nursing turnover. Foreign and domestic meetings, where material about turnover were presented, were regularly attended. These contacts began in the Spring of 1972 and continued to the present time.

Third, repeated estimations of the model were integral to the developmental process. Models must, of course, be estimated to be pronounced as empirically sound. What was somewhat different in this process were the repeated estimations. And the description in this paper has not examined the thirty-three theses and dissertations which Price and Mueller directed. The theses and dissertations continually improved the measures and confirmed the different components of the model. Every estimation was a learning process and sought to improve on preceding studies.

Mistakes were, of course, made in the developmental process. Seven mistakes are noteworthy.

First, the process was too often ignored in the development of the model. The intervening process between the determinants and turnover has been a component of the model from the beginning. Satisfaction and opportunity, for instance, were in the preliminary model developed when Price was on leave in England in 1972-1973. Satisfaction and opportunity were, respectively, mediating and moderating variables in the preliminary model. The focus on process was an explicit attempt to remedy the lack of a process emphasis in the economic research on models examined in the Spring of 1972. However, in many instances determinants were proposed without specification of how the variable impacted on turnover. When opportunity, for example, was made an exogenous variable in the Denver Study, no mediating variables were proposed. Again, when kinship responsibility was included in the Iowa-Illinois Study, in an attempt to conceptualize the apparent differences between local and non-local nurses observed in the pilot study, no mediating variables were proposed. Still again, when positive and negative affectivity were used as exogenous variables in the Texas Study, a process was not hypothesized between these variables and satisfaction/commitment. However, when the Texas Study, focusing on physicians, was published in 1996, some of the neglected processes were suggested. The most recent statement of the model (Price, 2000) attempts to specify all of the intervening processes.(13) However, for far too long, many intervening processes were ignored.

Journal reviewers forcefully noted the lack of a process for many determinants. Price then became aware of the fact that those processes were often in his mind but had not been specified. Mueller was especially discouraged by the journal reviews and the fact that sociologists were not much interested in turnover.(14) An effort was then made to specify the processes for the propositions, such as the 1996 article on physicians. Of major importance is the fact that Prices's training never stressed the need for process to make propositions complete. Price was aware of the need for some processes-his criticism of economic research illustrates this awareness-but his conception of theory never included an explicit directive about specifying processes for propositions, when such processes were needed.

Second, scope conditions for the models were never specified. Consider the situation of full-time and part-time employees. The hospitals in the first three studies employed both full-time and part-time employees. This employment pattern was viewed as an asset, since the research was able to examine a growing segment of the labor force, the part-time employee. Controls were, of course, always used for amount of time worked, whether full time or part time. In the Iowa-Illinois Study, amount of time worked and age were even used to classify the nurses for subgroup analysis. Never did it occur to Price and his colleagues to consider the possibility that the model might apply most readily to the full-time employees. During most of the time that research has been done on turnover-from the early 1900s-nearly all of the employees had worked full time, so the models developed would naturally be constructed to explain the turnover of full-time employees.

Price never thought of the possibility of specifying the universe to which the model would apply until he was introduced to the idea of "scope conditions" by the Stanford-trained researchers in the Sociology Department at Iowa (Cohen, 1991), especially by Barry Markovsky.(15) Price then realized that the model seemed to work best for full-time employees and he began to use amount of time worked as a scope condition.(16)

Third, some of the samples and sites were designed too narrowly. The pilot study and the Iowa-Illinois Study were probably all right, since the research was just beginning and it is difficult to avoid narrowness at the beginning. However, the Denver Study was probably a mistake, since it was not very different from the first two studies. Due to the recession in the early 1980s, the rate of turnover in the Denver area was also greatly depressed, thereby nullifying the major reason for this sample and site. Too much attention was devoted to obtaining a high rate of turnover in a brief period of time. The sample and site for the Texas Study was good, since it was very different from the first three studies. The problem with the Texas Study was that during this time Mueller decided to end most of his participation in the developmental process. Price continued the Texas Study with a graduate student (Sang-Wook Kim) but the range of the research was considerably reduced.

The narrowness of the sample and site raises questions about the generality of the model. Kinship responsibility, for instance, was significant as long as females were studied, but when male physicians were examined, it was no longer significant. The narrowness of the sample may also account for the fact that none of the moderating variables were statistically significant. All of the samples were mostly composed of middle-class employees, for example, and these individuals may be too homogenous in what they value. Rather than the Denver sample and site, a large, public hospital in a major metropolitan area, such as Chicago, would have obtained more lower-class employees. It was probably a good idea to keep doing research in the healthcare area, despite the problem of generality, because of official health sponsorship-the Medical School at the University of Iowa-and the opportunity to develop a track record in the area, which facilitated funding. There is a great variety in the American healthcare delivery system, but this variety was not explored.

Fourth, analysis was somewhat slighted in the studies. This neglect characterizes all the studies. The pilot study should have, but never did, conduct a systematic analysis of the local and non-local nurses. The Iowa-Illinois Study found that the classification of the nurses by age and amount of time worked did not very well explain the differences observed, but other subdivisions-by theoretical variables, for instance-was not attempted. The Denver Study was very concerned to get more men into the sample, and succeeded in doing so, but never did a matched subgroup analysis of males and females. Nor did the Denver Study, with its longitudinal data, wrestle with the causal orders of the exogenous/intervening variables and within the endogenous variables. Two points in time may not be ideal for longitudinal analysis, but two points are better than the one point of a cross-sectional analysis. When the Texas Study was written up, there was an awareness of the controversy concerning the causal order of search behavior and intent to stay (Sager, Griffeth, & Hom, 1998), but this topic was not even considered, and the Texas Study had longitudinal data-though for two points in time-available for the analysis. An immense amount of analysis was conducted in the development of the model. However, at critical points, analysis was not thoroughly explored. There was too much of a rush through analysis in the development process.

Fifth, a systematic procedure to access the empirical validity of the different components of the model was never developed. The procedure was to evaluate the studies on a case-by-case basis. The rule was that each study should improve on the preceding study, theoretically and methodically. When a study was done, if a determinant was statistically significant, the determinant was retained in the next study. If, however, a better measurement instrument was found, the next study would incorporate the new measurement. If a determinant was not statistically significant, an effort was made to see what went wrong. A variable would not be estimated unless it was supported by some empirical data, so if the variable was not statistically significant, something should be wrong. Usually what was wrong was an inferior measure. Many variables were estimated which had not been subjected to the extensive controls that were used in the research, so it was anticipated that quite a few of these variables would not be statistically significant. The bias of the research was to retain determinants for the next study.

A major result of this case-by-case approach was that knowledge was not available regarding the explanatory power of the different variables. An example illustrates this lack of knowledge. In the Spring Semester of the 1987-1988 academic year, Price was on leave at Birkbeck College of the University of London (England). While at Birkbeck, he was asked to describe the turnover research to the faculty and students. The description focused mostly on the model developed and the studies performed to estimate the model. When his oral presentation was finished, Price was asked to describe the major results of the research. Price could cite no major results: there was only a series of different studies, each one, supposedly, better than the preceding one. The studies had not been "added up," had not been cumulated.

Meta-analysis (Hunter, Schmidt, & Jackson, 1982) is a systematic way to assess, quantitatively, the results of a series of studies. One of the graduate students in Sociology (Roderick Iverson) started a meta-analysis of the Iowa Research in the early 1990s, but returned to Australia before the research was finished. The meta-analysis was dropped because Iverson began a series of new empirical studies of turnover in Australia which fully occupied his time. A graduate student in the College of Business (Chris Quinn-Trank) worked several years in the mid-1990s on a meta-analysis of the Iowa research, but was unable to use the results of her work for a Ph.D., so she did not complete the project. Frank Schmidt, a major scholar of meta-analysis in the College of Business, would not approve Quinn-Trank's dissertation because the Iowa research did not use a standard set of determinants. Finally, one of Rodger Griffeth's graduate students in the College of Business at Georgia State University (Stefan Gaertner) did a meta-analysis, with a standard set of determinants, on part of the Iowa research (Gaertner, 1990). Meta-analysis should have been done throughout the development of the model. Too much time had elapsed before such an analysis was done.

Sixth, longitudinal analysis was not fully used in the research. All of the studies which focused on turnover made sure that it was measured after the exogenous and intervening variables were assessed. There was, therefore, a longitudinal element in all these studies. The Denver Study also collected data at two points in time and had a longitudinal dimension in its analysis. Data were collected for a longitudinal analysis in the Texas Study, but the data were never analyzed. None of the studies, however, collected data at three points in time and built the entire analysis around the longitudinal data. In the Denver data, the longitudinal analysis is quickly inserted after a long series of cross-sectional analyses. Three data-collection points seem to be necessary for a complete longitudinal analysis.

Seventh, none of the studies used event history analysis as a statistical technique (Allison, 1984). The samples were always divided into leavers and stayers, thereby ignoring the different lengths of time that the sample's member had been employed. Important data were thus lost with the leaver/stayer dichotomy. Two of the graduate students working with Price and Mueller designed a study of the Chicago Public Schools to make use of event history analysis. The students obtained their dissertations (Currivan in 1998 and Iverson in 1992) from the Chicago research, but never used event history analysis in a published study. Event history analysis requires a large entering cohort-thus the selection of an entering cohort of teachers from the Chicago Public Schools-but it is clear that there are genuine advantages in this technique.

Three final comments should be made before concluding this paper. The comments deal with largely misplaced criticisms of the model.

First, the model is often criticized for having too many variables. The Texas Study had, for instance, twenty-four variables. This is certainly too many variables. Simpler models are, of course, favored over complex ones. However, there are many explanations of turnover in the literature and these explanations, at least the major ones, must be estimated at some time. The explanations must also be estimated against each other, which means including a substantial number of the variables in the same research. An effort can be made to eliminate variables with little support, but this will still, if past experience is a guide, leave too many variables. The models will eventually be simplified, but this will take time. Meta-analysis should facilitate the simplification.

Second, the model is often criticized for a large percentage of insignificant variables. Twenty-four variables may be proposed, for instance, but only 50% may be statistically significant. Three comments are pertinent to this criticism. (1) Some of the determinants have been proposed without many theoretical controls. Viscusi's (1979) study is an example of such research. With extensive theoretical controls, these determinants often do not become significant. (2) Statistical significance is strongly influenced by sample size. Statistical significance is easier to obtain with a large sample. And few studies of turnover have samples of a thousand or more. The samples must, of course, be even larger with many variables in the model. (3) Explained variance is an important criterion by which to evaluate the explanatory power of a model. But so also is the extent to which the results conform to predictions of the model. Price is partial to the use of confirmed predictions, but both criteria should be used for evaluation.

Third, the model contains two paths between the exogenous variables and turnover (see Figure 1.5). One path goes from kinship responsibility and opportunity to turnover, the second path goes from the remaining exogenous variables to turnover through satisfaction and commitment. All of the exogenous variables are not hypothesized to impact on turnover through satisfaction and commitment.

An effort has been made in this paper to describe the development of the Price/Mueller causal model of voluntary turnover by analyzing select studies conducted by these researchers. An effort has been made to select studies which illustrate critical features of the model's development. A sizeable amount of research has, of course, been excluded, since the Iowa research has been conducted for more than twenty-five years. The most serious exclusion has been the thirty-three dissertations and theses directed by Price and Mueller. These theses and dissertations have contributed substantially to the research that Price and Mueller have conducted. It has also been a distinct pleasure working with the students who did the research; we found these students to be marvelous colleagues. What has been outstanding is the fact that five of the dissertations have been performed in South Korea. These South Korean dissertations are an attempt to extend the generality of a model developed in the west. However, the four studies described seem to represent the development process quite accurately. Critical features of the model's development are covered by the four studies. Descriptions, such as the one presented in this paper, are rare and it is hoped that other turnover researchers will benefit from this description of model development and will provide descriptions of the development of other models.

  1. An excellent review of the field of turnover research is Hom and Griffeth (1995). The Hom and Griffeth review comments on the major models of turnover.
  2. The two moderating conditions are not in Figure 1.1, because they were not in the original text.
  3. The motivational approach of Homans (1961) is basically accepted, though often with different terminology, by all of the research by Price and Mueller.
  4. Scale of operations was not the definition in the original text.
  5. It was a mistake to equate programmed coordination with centralization. Programmed coordination, for example, decreases whereas centralization increases turnover.
  6. The questionnaires for one of the hospitals were distributed to the nurses when they picked up their pay checks.
  7. Factor analysis and alpha were used in all of the studies described in this historical review.
  8. Results for the work units and subgroups are not presented mostly due to space considerations.
  9. Extreme splits can, of course, but analyzed with logistic regression. There was too much emphasis on avoiding extreme splits in the Denver Study.
  10. Absenteeism was also added as a dependent variable in the Denver Study. However, absenteeism is excluded from consideration in this paper to simplify the presentation.
  11. The focus on more moderators was made, because of the work of Mowday and his colleagues (Mowday et al., 1982).
  12. The emphasis is on individual and cross-sectional data, because these results were confirmed by the work unit and time-series data.
  13. The most recent statement of the model also indicates the model's connection to the exchange approach (Blau, 1964). An exchange of benefits between the employers and the employees is assumed by this model. The employers distribute benefits-the structural variables can be viewed as benefits-to the employees in return for the employees' service to the organization.
  14. Sociologists may also have not been interested in the Price/Mueller model. This lack of interest may be because turnover and absenteeism have a connection with cost consideration in business, a topic in which sociologists are not too interested.
  15. The Stanford sociologists also did not favor the use of demographic variables, like amount of time worked, as scope conditions. They preferred theoretical variables as scope conditions. Price is not aware, at this time, of the theoretical variable(s) indicated by amount of time worked.
  16. It was previously noted that the Iowa-Illinois Study used age and amount of time worked to conduct a subgroup analysis of the nurses. This analysis demonstrated to Price that demographic variables should not be used to construct subgroups for analysis. The differences which were observed for the subgroups of nurses could not be accounted for by age and amount of time worked. Demographic variables, by their very nature, do not indicate what are producing effects. Subgroups should not be created with theoretical variables (Price, 1995; Price & Kim, 1993).

Allison, P.D. (1984). Event history analysis. Beverly Hills, CA: Sage.

Becker, G.S. (1964). Human capital. New York: Columbia University Press.

Blau, P.M. (1959-1960). Social integration, social rank, and process of interaction. Human Organization, 18, 152-157.

Blau, P.M. (1964). Exchange and power in social life. New York: Wiley.

Bluedorn, A.C. (1976). A causal model of turnover in organizations. Unpublished Ph.D. dissertation, University of Iowa.

Brief, A.P. (1976). Turnover among hospital nurses: A suggested model. Journal of Nursing Administration, 6, 55-58.

Brief, A.P., Burke, M.J., George, J.M., Robinson, B.S., & Webster, J. (1988). Should negative affectivity remain an unmeasured variable in the study of job stress? Journal of Applied Psychology, 73, 193-198.

Burton, J.F., & Parker, J.E. (1969). Interindustry variations in voluntary labor mobility. Industrial and Labor Relations, 22, 199-216.

Cohen, B.P. (1991). Developing sociological knowledge. Chicago: Nelson-Hall.

Currivan, D.B. (1998). An analysis of causal relationships in a model of organizational commitment. Unpublished dissertation, University of Iowa.

Farris, G.F. (1971). A predictive study of turnover. Personnel Psychology, 24, 311-328.

Gaertner, S. (1999). Structural determinants of job satisfaction and organizational commitment in turnover models. Human Resource Management Review, 9, 479-493.

Greeley, A.M. (1972). Priests in the United States. Garden City, NY: Doubleday.

Hom, P.W., & Griffeth, R. (1995). Employee turnover. Cincinnati, OH: Southwestern College Publishing.

Homans, G.C. (1961). Social behavior. New York: Harcourt, Brace.

House, J.S. (1981). Work stress and social support. Reading, MA: Addison-Wesley.

Hunter, J.E., Schmidt, F.L., & Jackson, G.B. (1982). Meta-analysis: Cumulating research findings across studies. Beverly Hills, CA: Sage.

Iverson, R.D. (1992). Employee intent to stay: An empirical test of a revision of the Price and Mueller model. Unpublished dissertation, University of Iowa.

Kanungo, R.N. (1982). Work alienation. New York: Praeger.

Karasek, R., & Thorell, T. (1990). Healthy work. New York: Basic Books.

Kim, S.W., Price, J.L., Mueller, C.W., & Watson, T.W. (1996). The determinants of career intent among physicians at a U.S. Air Force hospital. Human Relations, 49, 947-975.

Lyons, T.F. (1968). Nursing attitudes of turnover. Ames: Iowa State University, Industrial Relations Center.

Mangelsdorff, A.D. (1989). A cross-validation study of factors affecting military psychologists' decision to remain in the service: the 1984 active duty psychologists survey. Military Psychology, 4, 241-251.

March, J.G., & Simon, H.A. (1958). Organizations. New York: Wiley.

Merton, R.K. (1968). Social theory and social structure. Glencoe, IL: The Free Press.

Mobley, W.H. (1982). Employee turnover. Reading, MA: Addison-Wesley.

Mowday, R.T., Porter, L.W., & Steers, R.M. (1982). Employee-organization linkages. New York: Academic Press.

Osteman, P. (Ed.). (1984). Internal labor markets. Cambridge, MA: MIT Press.

Pencavel, J.H. (1970). An analysis of the quit rate in American manufacturing industry. Princeton, NJ: Princeton University Industrial Relations Section.

Price, J.L. (1975). A theory of turnover. In B.O. Pettman (Ed.),Labor turnover and retention (pp. 51-75). Epping: Gower Press.

Price, J.L. (1977). The study of turnover. Ames: Iowa State University Press.

Price, J.L. (1995). A role for demographic variables in the study of absenteeism and turnover. The International Journal of Career Management, 7, 26-32.

Price, J.L. (2000). Reflections on the determinants of turnover. International Journal of Manpower, 22, 600-624.

Price, J.L., & Bluedorn, A.C. (1980). In D. Dunkerley & G. Salaman (Eds.), The international yearbook of organizational studies 1979 (pp. 217-236). London: Routledge & Kegan Paul Ltd.

Price, J.L., & Mueller, C.W. (1981). Professional turnover: The case of nurses. Bridgeport, CT: Luce.

Price, J.L., & Mueller, C.W. (1986). Absenteeism and turnover of hospital employees. Greenwich, CT: JAI Press.

Price, J.L., & Kim, S.W. (1993). The relationship between demographic variables and intent to stay in the military: Medical personnel in a U.S. Air Force hospital. Armed Forces and Society, 20, 125-144.

Roethlisberger, F., & Dickson, W. (1939). Management and the worker. Cambridge, MA: Harvard University Press.

Sager, J.K., Griffeth, R.W., & Hom, P.W. (1998). A comparison of structural models representing turnover cognitions. Journal of Vocational Behavior, 53, 254-273.

Steel, R.P., & Ovalle, N.K. (1984). A review and meta-analysis of research on the relationship between behavioral intentions and employee turnover. Journal of Applied Psychology, 69, 673-686.

Stoikov, V., & Raimon, R.L. (1968). Determinants of differences in the quit rate among industries. American Economic Review, 58, 1283-1298.

Viscusi, W.K. (1979). Employment hazards. Cambiridge, MA: Harvard University Press.

Vroom, V.H. (1964). Work and motivation. New York: Wiley.

Watson, D., & Clark, L.A. (1984). Negative affectivity: The disposition to experience aversive emotional states. Psychological Bulletin, 96, 465-490.

Woodward, J. (1965). Industrial organization. London: Oxford University Press.