Agency growth, salaries and the protected bureaucrat.

Johnson, Ronald N. ; Libecap, Gary D.

AGENCY GROWTH, SALARIES AND THE PROTECTED BUREAUCRAT

I. INTRODUCTION

In recent years, increased attention has been directed towards understanding bureaucratic behavior and its implications for economic performance. A key element of this literature is the application of utility maximization for deriving implications about bureaucratic behavior. Representative is Niskanen [1971, 38], who argues that bureaucratic decisions are made rationally to maximize utility, which is assumed to be a function of salary, perquisites, reputation, power, patronage, and output. Implicit in the literature is an activist bureaucracy which engages in purposeful action to advance its interests.

An important part of some of the bureaucracy literature is the notion that government employees are a source of the documented growth of government. It is commonly asserted that government employees, acting in their own self-interest, will foster the growth of their organization. Growth of the organization is seen as a means for increasing salaries, as well as the other arguments in the utility function. The positive link between salaries and growth is emphasized by Downs [1967, 11] in his classic study of bureaucracy: "Any organization experiencing rapid overall growth provides many more opportunities for promotion in any given time period than a static one." Additional references to the importance of agency growth for salaries include Niskanen [1971, 38-41; 1975, 619], Tullock [1974, 127], and Heclo [1977, 131]. Building on that theme, other researchers such as Bush and Denzau [1977], Borcherding, Bush, and Spann [1977], and Bennett and Orzechowski [1983] have argued that government employees will use their power to influence election outcomes to promote government growth and, hence, salaries.

Despite the often critical nature of the salary-growth relationship in recent discussions of bureaucratic behavior and the growth of government, there is no direct empirical evidence of the asserted linkage. Related evidence provided, for example, by Wolfinger and Rosenstone [1980, 97-101] that government employees have higher voter participation rates than do private employees is not conclusive support for the growth-salary assertion. Voting results, alone, do not permit one to distinguish the underlying reasons for the higher participation rates, and there are many. More importantly, for the federal government, at least, the assertion that salaries of government employees are significantly affected by agency growth seems inconsistent with the idea that the civil service system shields federal white-collar employees from the uncertain fluctuations of political outcomes.

Skepticism about the growth-salary linkage due to the constraints of the civil service system is noted by Wilson [1980, 375]: "Few officials need fear for their jobs, and their salaries are determined by government-wide laws and regulations, rather than by the size, rate of growth, or 'success' ... of the organization." Further, while senior agency officials would be in the best positions to promote the growth of their agencies, within the federal government the salaries of high GS (General Schedule) level and Senior Executive Service personnel are capped by law and may not exceed those paid to officials in Executive Level V. The restriction limits the salaries of senior managers regardless of the growth of the agency. One could, however, argue that managers promote growth to motivate their subordinates or that medium- and lower-level employees are the primary proponents of agency growth. But as Kaufman [1981, 79] notes, "most officers and employees, once they have attained career status, could do nearly as well materially without pushing themselves, as they could by working at capacity." The absence of incentives and resulting inertia can contribute to the often-cited tension between career bureaucrats and political appointees.

The above discussion points out a serious problem with the growth-salary linkage often asserted in the bureaucracy literature. Given the absence of direct supportive evidence, it is difficult to determine the existence, much less the magnitude, of the relationship. Accordingly, one cannot distinguish between two competing hypotheses regarding bureaucratic behavior. One hypothesis, which implicitly discounts the effect of civil service rules, argues that federal employees have a strong incentive to support policies that would expand the size of their agency as a means of increasing their salaries. The other hypothesis, which emphasizes the protection provided by the civil service system, argues that federal employees have little incentive to promote the expansion of their agency to increase their salaries.

Reconciling the growth-salary assertion with the protections provided government employees by the civil service system is important for understanding bureaucratic behavior. A strong salary-growth relationship would, of course, lend support to the argument that federal employees have an incentive to promote agency growth. But also, tying salaries to the performance of the agency, as judged by congressional budget reviews, likely would make rank-and-file workers more responsive to the demands of the agency's constituency. On the other hand, the absence of a significant salary-growth relationship suggests that it may be difficult to motivate federal workers with ensuing implications for the performance of government.

To address these competing hypotheses, a theoretical framework for the asserted growth-salary relationship is outlined first. Internal labor market arguments regarding the impact of organizational growth on employee salaries are examined. In addition, the important ramifications of the civil service system on job ladders and salaries for federal white-collar employees are considered. Next, the impact of agency growth in terms of staffing and budgets on federal white-collar salaries is estimated using two random samples, each with over 15,000 employees, from forty-five federal agencies.

Overall, the analysis reveals support for the second hypothesis, that most federal employees have little incentive to support the expansion of their agency in order to increase their salaries. The federal civil service system apparently protects white-collar employees so that their salaries are relatively insulated from changes in agency staffing and budgets.

II. THE LINK BETWEEN AGENCY GROWTH AND SALARIES

A basic tenet of much of the literature on bureaucratic behavior is that career advancement and salaries depend upon agency growth. For example, Arnold [1979, 22] states: "The income incentive is probably their most important personal goal...Ordinarily, promotions can be obtained only when others resign or retire; however, they are considerably easier to obtain if the organization grows and new positions are created. It is, therefore, in the bureaucrat's self-interest to promote organizational growth by working for budgetary increases." Similarly, Bennett and Orzechowski [1983, 272] state that "as the firm (bureau) grows and new jobs are created, there is an increase in the likelihood of an employee (bureaucrat) rising to another position within the firm (bureau) where wages and productivity are higher." Bennett and Orzechowski emphasize that "in the public sector, the effect is of greater importance because bureaucrats, as a voting block, can have an important influence on the outcome of elections." The growth-salary link provides an underlying motivation for government employees not only to vote for increases in the size of government, but also to support programs within the organization that foster its growth and to oppose attempts to reduce the size and budget of the agency.

While the growth-salary linkage is an important part of the literature, those who advocate the relationship have not provided a formal theory showing the conditions under which organizational expansion can be expected to increase promotion opportunities. Certainly, within a neoclassical framework there is no necessary link between organizational growth and employee salaries. Under the standard neoclassical model of the firm, factor prices can often be taken as exogenously determined. If a firm expands, wages need not be bid up. A similar argument could be made for a governmental bureau. Clearly, some managers may experience an increase in their remuneration, and some workers may be promoted to managerial positions. But the rank and file need not, nor expect to, experience an increase in their wages just because the organization grows.

On the other hand, with internal labor markets, as described by Doeinger and Piore [1971], growth may have a greater effect on earnings. The usual portrayal of internal labor markets is one wherein workers have, or develop, skills that are specialized to the organization. The ports of entry into the organization are usually thought of as being at the bottom of a hierarchical pyramid, and advancement is limited by a narrowing of available positions as individuals progress up the pyramid. Furthermore, wage rates are attached mainly to the position held rather than to the worker, a point emphasized by Williamson et al. [1975].

An important feature of internal labor markets is that the wages paid to employees are often inconsistent with the simple notion that workers receive their marginal product in each time period. High costs of monitoring to detect shirking provides an explanation as to why employers would choose an alternative to the standard wage contract. The rank-order tournament contract offers an example. Lazear and Rosen [1981] demonstrate that under that scheme, payment is based on an ordinal ranking of employees and provides performance incentives. Malcomson [1984] has further developed the argument and has shown that payment based on rankings fits with the main features observed in internal labor markets. In particular, rankings together with internal promotion offer an incentive structure even in the presence of asymmetric information.

Within the framework of the internal labor market model, organizational growth can improve promotion possibilities by creating more job openings. increase in the demand for the firm's output would likely expand the base of the hierarchical pyramid and increase the derived demand for workers at each level. Initially, with the firm relying on internal promotion to fill the higher-level positions, the proportion of workers promoted would increase. To facilitate that process, the standards for promotion will be lowered, yielding an inverse relationship between the standards for promotion and the proportion of workers promoted. Once the firm has reached its new optimal size, the proportion of workers being promoted will be reduced. Of course, in a fully dynamic model rational workers would hold expectations about future firm growth and opportunities for advancement. Nevertheless, promotional opportunities are an integral part of internal labor markets and are often tied to the performance of the firm. In the private sector, internal markets with their hierarchical pyramids and rank-order promotional schemes offer an incentive to employees to enhance the output of the firm.

The pyramidal job structure is assumed implicitly by those who assert a significant positive relationship between organizational growth and bureaucratic salaries. While Rosenbaum [1979] provides supportive evidence of the relationship in certain private sector firms, a key issue is whether or not a simple pyramidal job ladder applies to the federal civil service. The answer to that question requires a more detailed examination of the federal civil service system and how it affects the hiring and promotion of individuals.

III. THE CIVIL SERVICE SYSTEM

Advancement and Promotion

At first glance, the federal civil service system appears to be a classic internal labor market with salaries a function of position rather than of the individual. Salaries increase as individuals move to higher positions within the hierarchy. Nevertheless, as Stewman and Konda [1983] stress, a pyramidal position structure alone does not imply pyramidal promotion opportunities for individual employees. As described below, seniority in the federal government is so weighted in promotion determination that individuals who meet the job qualification standards advance regardless of changes in the budget or staffing of their agency. Moreover, qualification standards can impose an important barrier to individual advancement, even in an expanding agency environment. As individuals advance in their job ladder, further progression to higher-level positions requires substantial increases in formal education and training, which the individual may not have. There also is competition for higher-level positions by contenders outside the federal government. Ports of entry into the federal white-collar labor force exist throughout the range of GS grades, not just at the lower levels. Thus, new entrants may mitigate any positive effects agency growth might otherwise have for the promotion of existing workers.

Since civil service rules describe a more complex promotion arrangement than a simple pyramid, it is necessary to examine them in some detail to see how they might moderate any agency growth effects on salaries. The general schedule of the civil service system, which applies to 72 percent of all civilian federal employees, covers most white-collar positions in the federal government. Within the general schedule there are eighteen pay grades, and within grades GS1 through GS15 there are ten steps with pay increases of approximately 3 percent each.

Individual salaries increase with advancement through grade steps that reflect time in service. The prescribed time periods for step advances are usually 52 weeks for steps 2 through 4, 104 weeks for steps 5 through 7, and 156 weeks for steps 8 through 10. Grade step increases are a major contributor to the earnings of most federal workers and are virtually automatic. The General Accounting Office (GAO) [1975, 34] reported that 99 percent of those eligible received their step pay raises. It is important to note that the associated pay increases occur regardless of whether or not the agency is growing.

Under civil service rules, GS grades are linked to specific federal white-collar positions through the position classification process. The general schedule rules are uniform across the federal government. Each agency has the responsibility to define work positions and the corresponding GS levels following the classification framework provided by the Office of Personnel Management (OPM). Through the agency-based classification process, each white-collar position is assigned a title, occupation code, GS grade, qualification requirements, and performance standards. The range of possible GS grades within an occupation define the job ladder. As with grade step increases, the ladder is so structured that agency growth or decline may have little impact on promotion possibilities.

New white-collar employees enter the federal work force with a probationary period of one to two years until being tenured. Individuals advance across grades, within the job ladder, so long as they meet the higher requirements of the position. As noted by the Grace Commission [1984, 86], advancement can be relatively routine and rapid within the job ladder.

To illustrate the nature of civil service job ladders and promotion opportunities, consider the Safety and Occupational Health occupation. Individuals can enter this occupational job ladder as a Safety and Occupational Health Specialist with a range of possible GS levels from GS5 through GS12, or as a Safety and Occupational Health Manager with a range of possible GS levels from GS11 through GS15. Higher GS levels involve greater responsibilities and qualification requirements. Entry at the GS5 level requires four years undergraduate college education. Entry at or advancement to the GS9 requires a master's degree; and a GS11 or higher requires a Ph.D. degree. In addition, there are increased experience requirements, though, to a limited degree education and experience are substitutable. Normally, as with any white-collar position, a Safety and Occupational Health Specialist can expect to advance through a GS12, so long as the individual meets the experience and education requirements. Movement through GS15, however, requires competitive promotion to the different job ladder of a Safety and Occupational Health Manager. Those who do not meet the requirements for promotion within or across job ladders remain in their position, and their salaries increase with step advancement. Unlike the military, there is no up-or-out requirement within the federal civilian labor force.

Given the education and experience requirements, not all employees can advance to the top of their particular occupational job ladder. Moreover, outside competition for open positions can occur at relatively high levels. for example, all federal white-collar occupations are classified as to whether they are professional, administrative, technical, clerical, or other (PATCO). Professional and administrative occupations have the highest educational and experience requirements for entry. In 1986, the median GS grade for new hires for professional and administrative positions was GS9. The average GS grade for all positions was 8.4.

The discussion of civil service rules points out that promotional opportunities outside the job ladder are not simple sequential advancements up a pyramid. Because of outside competition for higher-level positions and the existence of explicit qualification requirements for promotions, agency growth is unlikely to be of important assistance in the advancement of many individuals.

It could be argued, however, that in order to attract new employees as an agency grows, the general level of salaries within the agency must be increased. But studies by Smith [1977], Quinn [1979], and Venti [1985] show that federal workers are generally paid more than their private sector counterparts with the same socio-economic attributes. Additionally, Venti [1985] and Kruger [1986] indicate that there are extensive job queues for most federal positions. The existence of queues suggest that an expanding agency need not raise salaries to attract qualified personnel, even if the supply schedule were upward sloping. Further, general salary increases are not agency specific, since the federal salary structure is largely exogenous to one agency. When raising salaries, Congress has historically used across-the-board percent increases for all GS employees. With a standardized classification and pay system for the entire federal civil service system, the incentive of white-collar employees within a particular agency to foster agency growth is reduced.

Of course, government employees as a group can still be expected to lobby Congress for overall pay increases. But their incentive to push for a general increase may have little or nothing to do with the growth of their particular agencies. Moroever, there is little reason for federal employees to lobby for increases in total federal employment with the hope that salaries would be increased in the process. Given the existence of queues for federal positions, Congress may simply increase employment without raising the general schedule of wages. Although Tullock [1974] emphasized that growth is a means for increasing salaries, he also noted that governmental employees may be willing to lobby for a reduced size of government employment in exchange for higher wages.

While federal civil service rules can limit the effect of agency growth on promotion and salaries, those rules also have important affects on the impact of declines in agency employment on promotion and salaries. In the following subsection, reductions-in-force, or RIFs, are examined.

Reduction-in-Force (RIF) Procedures

Federal agencies are generally allocated a limited number of full-time equivalent positions along with a budgeted amount for salaries and expenses. The latter includes pay and fringe benefits, travel, office equipment and other related expenditures. According to Hartman [1983, 108], the slot constraint dominates as "the agency typically runs up against its slot limit before it runs out of salary and expense money." That is, the budget process is sufficiently fungible that salaries are not a binding constraint when it comes to staffing decisions. While that suggests that government managers may hire an excessive number of higher-grade personnel, it also indicates that if an agency were required by Congress to reduce staffing slots, there is ample salary and fringe monies available to allow the reduction to occur at the lower-grade levels. Proportional declines in personnel across all grade levels, essentially reducing both the height and width of the hierarchical structure, are not required. That flexibility allows for the protection of tenured, more senior personnel.

The protection offered by the civil service system to tenured white-collar employees is demostrated by the application of federal reduction-in-force procedures. RIFs are the most drastic form of position reduction in the federal government. They can result in involuntary separations and the downgrading of employees to lower ranked positions, and, therefore, could seriously affect salaries. Nevertheless, the procedural requirements are so designed to protect seniority that the permanent impact of RIFs on the salaries of all but nontenured federal employees is limited.

RIF regulations, as defined in the Federal Personnel Manual, Chapter 351, call for employees in the targeted positions to compete for remaining jobs. Retention is based on tenure, veterans's status, seniority, and performance ratings, with seniority the most important factor. High performance ratings and veteran's status can add to seniority calculations. Under these criteria, temporary employees must be released before probationary and permanent employees are affected. Once an RIF extends to permanent employees, those with the lowest seniority and veteran's status ranking are the most vulnerable. They can receive lateral transfers to positions in the agency not subject to RIFs, or they can bump lower-ranked employees. Any employee who is downgraded, however, retains his GS rank and salary, tenure, and grade step increases appropriate for his pre-RIF position for up to two years. Beyond that time, salaries will be adjusted gradually to reflect the new position.

The limited impact on the salaries of permanent employees is indicated by Reagan administration RIFs. Between 1981 and 1983, over forty-two agencies were involved in reductions-in-force. Despite this, the number of workers who were affected involuntarily was small. Ninety percent of the work force reduction was through voluntary attrition and early retirement, which was costly to the government. In 1981, only 2,629 personnel lost their jobs through RIFs, which as 0.4 percent of total 1981 separations and represented only 0.09 percent of the average total civilian work force for the year. Furthermore, a GAO [1985, 35-37] study of RIFs in eight agencies from 1981 through 1983 showed that of the employees who were downgraded, many were quickly repromoted. Of the 2,055 employees affected by the RIFs, 744 were downgraded to lower GS level positions. In less than two years, however, nearly half were repromoted to pre-RIF positions.

The protection offered by the civil service system when agency staffing and budgets decline can be seen further in the experience of white-collar employees of the Civil Aeronautics Board (CAB), which was dismantled by Congress on January 1, 1985. Beginning in 1980, budgets and staffing levels were gradually reduced, with employment falling from 734 in 1980 to 367 in 1984 through attrition without an RIF. At the end of 1984, most of the remaining CAB functions and personnel, including Division Chiefs and staffs, were transferred to the Department of Transportation, with many offices remaining essentially unchanged.

The protection provided employees and their salaries from changes in congressional mandates calling for declines in agency staffing and budgets is reflective of the political power held by federal unions and employee organizations. The protection from declines in agency staffing when coupled with the relatively automatic step advances and within-job-ladder promotions describes an institutional environment wherein agency growth may have little effect on an individual's salary. The next section includes evidence on the magnitude of the salary/agency growth relationship using individual personnel records from OPM's Central Personnel Data File (CPDF) for March 1980 and March 1985. The selection of the two time periods was based on the premise that changes in administrations and corresponding policies after the November 1980 elections would allow for significant alterations in employment of personnel at various agencies.

IV. AGENCY GROWTH AND BUREAUCRATIC SALARIES

The CPDF contains data on individual annual earnings, agency affiliation and a large array of socioeconomic variables. The latter includes variables pertaining to an individual's educational achievement, experience, sex, minority status, veteran's preference, region of employment, and handicap status. For the larger agencies a 1 percent random sample of full-time employees was obtained. A 10 percent sample, excuding those agencies with less than 200 employees, was utilized for the smaller agencies. The forty-five federal agencies included in the sample accounted for 97 percent of all GS employees.

The first step in developing a test for a salary/agency growth relationship was the use of a standard wage regression to obtain estimates of the determinants of individual salaries. The basic wage equation was of the form 1n W.sub.ij = f (X.sub.j., A.sub.i.), where 1n W.sub.ij denotes the log of the annual salary of individual j in agency i, X.sub.j is a vector of socio-economic variables for individual j and A.sub.i is a vector of agency dummy variables. For both the 1980 and 1985 data sets the estimated coefficients on the individual characteristics variables had the expected signs and were similar to those obtained by Borjas [1980].

Despite the use of numerous controlling variables, the wage regression results revealed some large differences in earnings across agencies. For example, the coefficients on the agency dummies obtained from the 1980 data set varied from -.209 for the U.S. Soldiers' and Airmen's Home to .306 for the Federal Mediation and Conciliation Service. The F-statistic on restricting all agency coefficients to be equal was 39.04. Similar results were obtained using the 1985 data set. The persistence of the agency earnings differentials is consistent with the findings of Scism [1974] and Heclo [1977, 116-20] that interagency mobility of white-collar workers is limited.

Borjas [1980] argues that much of the observed difference in earnings across agencies can be attributed to agency constituency relationships. Hartman [1983, 44], however, suggests that the main cause for the variation is the occupational classification procedures used by the federal government. Including a broad classification of occupation codes into the regression runs along with variables denoting whether the individual was in a professional, administrative, technical, clerical or other type of position did have a significant effect on the agency variables. With the inclusion of occupation-related variables the F-statistic on the agency coefficients fell to 20.26. Nevertheless, substantial agency variation remains and provides the conditions for testing the salary-growth hypothesis. The results, presented below, show that our general conclusions are largely unaltered by inclusion of the occupational variables.

As already noted, an employment growth-salary relationship is central to much of the literature on bureaucracy. The internal labor market model provides the framework for the relationship. Increased staffing is seen as expanding the hierarchical pyramid resulting in promotions to higher GS levels. To test for the impact of staffing growth on salaries, the approach here is to see how the agency earnings differentials changed between 1980 and 1985 and to relate these changes to agency growth. The testing equations are designed to detect whether the salaries of individuals with similar attributes are advanced more rapidly in faster-growing agencies. Importantly, the GS salary structure pertaining to pay for each grade and step is exogenously determined and given to each agency. Thus, relatively higher salary growth can be attributed to more rapid promotions within a particular agency. For the period 1980 to 1985, total GS

employment increased by only 3.4 percent. As will be shown, however, employment growth varied substantially across agencies.

The estimating procedure followed is one of testing across the two data sets, 1980 and 1985. The initial regression run entails pooling both data sets, introducing a set of dummy variables for each agency, and dropping the constant term. The testing equation is where D.sub.i equals one if agency i, zero otherwise, and D.sub.i equals one if agency i and year equals 1985, zero otherwise. The X's are the socio-economic variables, and we allow for different slope coefficients as D equals one if year equals 1985, zero otherwise. The agency salary-growth coefficients, [alpha].sub.i., reflect changes in the agency earnings differentials over the period 1980-85.

Table I reports the agency/salary growth coefficients and relative employment growth for each agency from 1980 to 1985. The salary growth coefficents reveal relative differences in the growth of earnings across agencies over the period ranging from a high of .503 for the Farm Credit Administration to a low of .319 for the Department of Energy. The employment growth figures show substantially more variation, and as demostrated below the salary growth coefficients are not systematically related to agency growth.

The following linear relations are used to test formally for the effect of employment growth on wages. [alpha].sub.i = C + b.GROWTH(85-80)+e.sub.i, (3a) and [alpha].sub.i=C'+b.sub.1.GROWTH(85-84) + b.sub.2.GROWTH(84-83) (3b) +b.sub.3.GROWTH(83-82) + b.sub.4.GROWTH(82-81) +b.sub.5.GROWTH(81-80)+eH.sub.i.

Here, GROWTH(85-80) is the difference, in logs, between total GS employment in agency i in 1985 and 1980. The error terms are assumed to be normally distributed with constant variance. The second equation (3b) allows for a free-form distributed effect by using differences, in logs, in annual GS employment. The use of logs, as opposed to relative changes, facilitates testing for possible asymmetry of the growth effects and the equality of the b's in equation (3b). The rationale for a distributed effects specification is that promotions and, hence, higher salaries may lag behind employment growth. To estimate the coefficients in equations (3a) or (3b), a pooled sample approach is considered first, whereby the right-hand side of equation (3) is substituted into equation (2) and re-estimated using the full sample of individual observations. While the procedure introduces heteroskedasticity, maximum-likelihood estimation and ordinary least squares produced almost identical results. Table II presents some selected results.

The first two columns in Table II pertain to a sample containing all GS employees. The second set of regression results were obtained by using observations on employees classified as being in professional or administrative positions only. Focusing on these two groups separately allows testing as to whether individuals in higher managerial and professional positions are more affected by agency growth than are those in the remaining classifications of clerical, technical or other. The third set of regressions (in columns 5 and 6) shows the effect of accounting for the individual's general occupational classiffication and designation of position as either professional, administrative, technical, clerical or other. The number of observations in the first and third groups are the same. The fourth (and last) group is the same as the first with the exception that all observations on individuals working for the Department of Defense were dropped. The Department of Defense represented 34 percent of the full sample.

Based on the results shown in Table II the effect of employment growth on salaries seems minimal and sensitive to the data set used. The only coefficient on the agency/employment growth variable, GROWTH(85-80), that is statistically different from zero at the 1 percent level is that for the full sample where occupational classification is not accounted for (column 1). The coefficient is .044, suggesting that an agency must more than double its size in order for salaries to increase by 4.4 percent relative to an agency that did not grow.

The coefficients on the distributed effects specification appear to indicate a larger initial response. However, the general pattern of the effects is contrary to our prior expectations, and it is difficult to explain the negative growth effects such as that indicated by the coefficient on GROWTH (83-82) in column 2. An F-test, however, showed that the distributed effects specification did not predict better than simply using GROWTH(85-80).

Restricting the sample to higher administrative and professional personnel does not provide a stronger growth effect, and the results in column 5 show that after accounting for occupation, the agency/salary growth effect is not significantly different from zero. Moreover, excluding the Department of Defense from the regression runs (as shown in column 7) yields growth coefficients that are smaller and generally not significant. The latter suggests that the pooled regression results showing a significant positive effect of growth on salaries for the entire federal white-collar workforce is largely dependent on the inclusion of the numerous observations on the Department of Defense.

The sensitivity of the results to the sample's composition suggests that an alternative approach to estimating the growth coefficients in equations (3a) and (3b) may be more appropriate. The alternative is to utilize a two-stage procedure and investigate variations in regression coefficients. In the first stage the agency/salary growth coefficients are obtained as reported in Table I. The second stage involves regressing those agency coefficients on GROWTH(85-80) or the distributed effects specification. Here, the sample size is forty-five.

Amemiya [1978] and Borjas [1982] have shown the conditions under which the two-stage procedure and the pooled procedure reported in Table II can yield identical results. In an operational context, however, the statistical inference will vary. Using the coefficients from Table I and following Saxonhouse's [1977] suggested corrections for heteroskedasticity, equations (3a) and (3b) were re-estimated. For equation (3a), and using the full sample, the estimated coefficient on GROWTH(85-80) was--.0001 (t-statistic--.006). The results using equation (3b) revealed that none of the distributed growth coefficients were significant even at the 20 percent level. Additional two-stage runs were made for the same categories as reported in Table II for the pooled regressions. Again, the coefficients on the agency growth measures were not significant. Thus, none of the two-stage regressions indicate that changes in agency earnings differentials (Table I) are positively related to agency growth.

The empirical results reported are only a subset of numerous runs undertaken. Tests on the symmetry of positive and negative growth effects were also conducted. In none of the runs could the null hypothesis of symmetry be rejected. It could be argued, however, that the estimates of the effects of agency growth on salary understate the true effects because of new hires. Since there are often queues for federal positions, the salary of new hires may not be affected by agency growth. But restricting the CPDF sample to only those individuals with five or more years of federal experience in 1985 produced results similar to those shown in Table II.

Consideration was also given to the possibility that changes in an agency's budget could have a significant impact on salaries. The previous discussion, however, arguing that agencies are more constrained by personnel ceilings than by budgetary constraints, suggests that inclusion of a budget growth measure into the regression equations should yield an insignificant t-statistic on the budget variable. The results support this prediction. The budget variable did not enter significantly, nor were the coefficients on the staffing growth variables affected.

Finally, an indirect test of the salary/agency growth relationship was attempted. Although Ippolito [1987] reports that quit rates are relatively low in the federal government, it is possible that slow growth may encourage individuals to leave an agency. To test for that possible effect annual relative changes in agency quit rates, computed over the period 1984-86, were regressed on concurrent and past agency employment growth. None of the runs showed a significant negative effect of employment growth on changes in agency quit rates.

To summarize, it would appear very difficult to conclude from the results presented here that a strong relationship between agency growth and salaries exists. At best, agency size would have to double for salaries to increase by 4.4 percent over a five-year period. Moreover, the sensitivity of the pooled regression results to the exclusion of the Department of Defense was demonstrated. A two-stage procedure yielded growth coefficients that were negative, although not significant. It is also worth nothing that simple inspection of Table I shows that the salary growth of the two agencies in the sample (the Selective Service System and Federal Emergency Management Agency) that more than doubled in size was actually below the norm.

V. CONCLUDING REMARKS

The notion that the salaries of bureaucrats are closely tied to the fortunes of the agency has been an important part of the literature dealing with bureaucratic behavior and the growth of government. Yet, the results presented in this paper do not reveal a strong significant linkage between changes in staffing and salaries within an agency. For the typical federal white-collar employee, promoting growth does not appear to be the way to increase salaries.

Of course, agency growth may matter for other reasons. Staffing changes can result in lateral reassignments to less-preferred locations or can reduce the set of opportunities for job changes. Certainly for nontenured employees, growth would seem to matter. But those individuals are not in positions that would allow them much influence over agency policy. For the top-ranked personnel, those who may be in a position to influence policy, salary does not appear to be the major motivating force, as their salaries are capped. However, agency growth may still be important to them because they associate with the mission of the agency and/or because growth may lead to greater opportunities to consult or contract with the agency after retirement. Greater prestige and additional prequisites may also be associated with a growing agency, but these are all difficult to quantify.

While agency growth may matter for other reasons, a major conclusion of this paper is that the salaries of federal white-collar workers are highly insulated from changes in staffing and budget appropriations. Federal employees also appear to be highly paid. The civil service system, often lauded because of the protection it gives to civil servants, provides a relatively uncommon set of benefits. Whether having such an insulated group is in the interest of the general voter/taxpayer is a question beyond the scope of this paper; but understanding the protected environment in which federal workers reside is a first step in answering these broader questions.