The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Economic m i s i a i f l R E V I E W 1988 Quarter 1 Vol. 24, Ho A Can C om petition Am ong 2 Lo cal Governm ents Con strain Governm ent Spending? Economic Review is published quarterly by the Research D e p a rt m ent of the Federal Reserve Bank by R andall W. E berts of C leveland. C opies of the and T im o th y J. G ronb erg are available through our Public Review Inform ation D epa rtm en t, T h e decentralized U .S . governm ental structure has been both praised for 216 /5 79 -215 7. prom otin g efficiency and blam ed for stim ulating excessive local g o ve rn m ent spen ding. B y exam ining the relationship between the num ber of local g ove rnm e nts within local labor m arkets and their expenditures, the authors find that the existing structure of gove rnm e nt creates two o p p o s ing forc es. C om petition am on g general-purpose units constrains local C oordinating Eco n o m is t: Randall W . Eb erts Editor: W illiam G . M urm a nn gove rnm e nt spen d in g , while the overlapping labyrinth of single-purpose gove rnm e nts stim ulates local gove rnm e nt spending. A s sista n t Editor: Robin Ratliff Design: M ichael G alka Typesetting: L iz H an n a E x it Barriers in the Steel Industry 10 Opinions stated in Review Economic are those of the authors by Mary E. Deily and not necessarily those of the U n derstan ding how resources exit from a declining industry is im portant Cleveland or of the Board of for evaluating the perform an ce of those industries. This paper exam ines Governors of the Federal Reserve the nature and size of exit barriers in the steel industry and argues that S y s te m . Federal R eserve B an k of these barriers have greatly prolonged the in d u s try’s contraction. The effects of current trade protection and pension policies on the size of the exit barriers are also discussed. Material m ay be reprinted pro vided that the source is credited. Please send copies of reprinted W h y Do W ages V ary 19 material to the editor. Am on g Em ployers? b y E rica L. G roshen This paper review s the burgeoning literature on w age differences am on g industries and establishm ents. First, the empirical evidence on intra- and inter-industry w age differentials is s urveyed . Then the author discusses five alternative explanations for large, persistent, system atic w age differentials for observationally equivalent w orkers across em plo yers. The paper co ncludes with a short discussion of the policy im plications of alternative explanations for em ployer w age differentials. IS S N 0013-0281 Can C om petition Am ong Local Governm ents C onstrain Governm ent Spending? by R andall W. E be rts and T im o th y J. G ronb erg Randall W . Eb erts is an assistant vice president and ec onom ist at the Federal Reserve B ank of C levelan d. T im o th y J . G ron berg is an associate professor of econom ics at Texas A & M University. T h e authors wish to thank Ralph D ay and D avid D ’Alessandris for excellent co m p uter assistance and Erica G rosh en and Daniel M cM illen for helpful co m m en ts and suggestions. Introduction The United States contains more than 80,000 separate governmental units. If none of these units overlapped, each government would serve fewer than 2,000 individuals. Governmental units do overlap, however, resulting in several layers of jurisdictions. Residents within a metro politan area typically receive public services from a municipality, a township, a county, and a host of special districts. In addition, at each level of government, several similar governmental units may provide services within the same geographical area. For example, the Chicago metropolitan area alone contains more than 250 municipalities, each responsible for the same array of governmental functions. Overlapping these governments are 835 special districts, which usually perform only a single function, such as providing regional transportation or enforcing environmental pro tection regulations. The impact of this structure on government behavior is varied, and the net effects are not yet fully understood. Critics of the decentralized structure of local governments blame the pro liferation of local governments for what they see to be “runaway” spending. They argue that duplication of efforts by similar but independent jurisdictions within the same geographical area is an inefficient way to provide public services and that the resulting fragmentation could ne gate any benefits derived from economies of scale. Proponents of a decentralized public sector counter with the argument that it fosters increased efficiency in the production of public goods. They maintain that competitive pres sures induce local governments to adopt the most efficient provision techniques and to tailor the levels of provision of public goods to the preferences of societal subgroups (Oates [1972]). The phenomenal expansion of the local pub lic sector adds fuel to this controversy. Since 1950, state and local government expenditures have increased at a faster rate than either the gross national product, federal expenditures, or expenditures on private-sector services. State and local governments currently claim 17 per cent of total personal income, in contrast to 10 percent in 1950. Currently, they spend two and one-half times more than the federal govern ment spends on civilian services such as educa tion, roads, welfare, public health, hospitals, police, and sanitation. How much of this growth is due to govern ment structure and how much is due to other factors, such as demand for local services, is an empirical question. Even the effect of govern mental structure can work in opposite direc tions. For instance, a decentralized public sector may increase local public spending due to duplication of efforts, but at the same time, competition among these units may constrain spending. The net effect of our present govern mental structure on government spending depends on which of these various factors is more important. To further complicate matters, there are two distinct types of local governments. One type provides a variety of services to a subgroup of the county or metropolitan population, while the second type typically provides a single serv ice to the entire local area. Possible differences in behavior of these two government types must be taken into account. Two previous studies, one by Oates (1985) and a follow-up by Nelson (1987), have estimated the relationship between decentralization and government spending, but without conclusive results.1 The purpose of this paper is to continue the inquiry into the relationship between decentral ization and the size of the local public sector. We test the decentralization hypothesis proposed by Oates, in which an increase in the number of governmental units reduces local government spending as a percentage of personal income. However, unlike Oates (and Nelson), we contend that the hypothesized effects will most likely be observed at the metropolitan and county levels (referred to as the local level), not at the state or national levels. We believe that most of the “disci pline” derived from competition for households and firms would be observed at these levels of disaggregation, because these levels more closely approximate local labor markets within which firms and labor are most mobile. Oates (1985), in fact, argues that the “discipline” resulting from fiscal competition should increase as the geo graphical size of the unit of analysis decreases. However, neither Oates nor Nelson uses a unit of analysis less aggregated than the state. To test our point, we use various levels of aggregation from the county to the state level. We find solid statistical support for the decentralization hypothesis at the metropolitan and county levels. Increases in the number of ■ 1 A n unpublished paper by Z a x (19 8 7), recently brought to our attention, also takes exception to the use of state-level data by Oa tes and N e ls o n . Us in g county-level data, he finds a negative and statistically significant relationship between the n u m b er of gove rnm e nts and the size of the local public sector. H is study differs from ours in at least three w ays. Firs t, he uses ow n-source revenue as a dependent variable, w hereas we use local expenditures on selected fun ction s. S e c o n d , we explore these effects at various levels of agg regation , not just at the c o u n ty level. T h ird , he finds that an increase in the n um ber of special districts also reduces the size of the local public sector. W e find the opposite effect at each level of disaggregation. competing general-purpose government units are associated with a statistically significant decrease in the relative income share of local public expenditures. At the same time, we find a distinct difference in behavior between the two types of government. An increase in the number of single-purpose districts increases the share of personal income going to local government expenditures. To further support our point, we find that these relationships are not significant at the state level, which is consistent with the results of Oates and Nelson. I. Com petition Am ong Local Governm ent Jurisdictions The potential benefits of competition among local government jurisdictions are similar to the benefits associated with competition in private markets. In the private sector, competition induces profit-maximizing firms to provide goods or services preferred by consumers at the lowest resource cost. The motivating force behind this behavior is the choice of suppliers available to consumers. If a firm raises its price, consumers will switch to the supplier with the lowest price, assuming that all firms are identical and that consumers incur no additional cost in searching for another supplier. Given enough competing firms (that is, choices to the con sumer), no firm can set prices above the per-unit cost of production. The same competitive forces exist among local government jurisdictions. By law, local governments cannot earn profits. However, according to Niskanen (1971), public adm in istrators may be motivated to maximize revenue, and thus expenditures, in order to expand desir able aspects of their working environment. Pub lic administrators thereby “consume” profits on the job instead of taking them home. The capacity of governments to increase revenues depends upon the customer base— taxpayers w ho live within their jurisdictions. If local governments attempt to raise taxes or to reduce the level and quality of services, then taxpayers will have an incentive to locate in neighboring jurisdictions that provide a service/ tax package more in line with the taxpayers’ preferences. The loss of households and firms reduces a government’s tax base and, in turn, reduces its ability to raise revenue. Thus, the basis for the constraining effect of decentralization is founded upon the interjurisdictional competition for mobile resources, both labor and firms. The line of argument follows the old industrial-organization paradigm of structure, conduct, and performance. Applied to the public sector, the argument runs from an increase in the number of independent public jurisdictions (suppliers), to an increase in the degree of competition, to a decrease in the relative size of the public sector. However, the efficacy of governmental fragmentation depends on the mobility of households and firms. The net benefit of the move determines the extent to which mobility occurs or is likely to occur. This benefit comes from either the sav ings derived from locating in a lower-cost juris diction or the advantages gained from residing within a jurisdiction that provides more or better services, everything else being equal. The costs associated with choosing between local governments are generally greater than the costs incurred in searching for alternative sup pliers of private goods and services. To change local governments, a household must change residence and incur the costs of purchasing a new home and finding a new job, and must bear the emotional costs of moving to a new area. However, these costs are in direct proportion to the distance one must move in order to find a more preferable governmental unit. For exam ple, if enough choices of local governments are available within the same metropolitan area, then the discontented taxpayer may not need to change jobs in order to change jurisdictions. Consequently, the mobility of households and firms increases as the size of the geographical area decreases. Therefore, we would expect local governments to be more constrained by competitive forces at the county or metropolitan level than at the state or national level. The two empirical studies by Oates and Nelson have looked for the constraining effect of competing jurisdictions only at the state level. Oates proposes and tests the hypothesis that the size of the public sector should vary inversely with the extent of fiscal decentralization, other things being equal. He uses the number of jurisdictions within each state as a measure of decentralization. Using state-level aggregates, however, he finds no significant relationship between state and local expenditures as a per cent of state personal income and the number of jurisdictions. In a reply to Oates’ paper, Nelson suggests two modifications. The first is to distinguish between general-purpose jurisdictions (such as municipalities) and single-purpose jurisdictions (such as school districts and mosquito-abatement districts). Nelson argues, and rightfully so, that the two types of districts are not compara ble and consequently should not be lumped together. The multiplicity of special districts within a metropolitan area does not necessarily indicate that consumers have a choice, but rather that residents are provided several serv ices, each by a different district. In addition, since many special districts pro vide only minor services and since nearly half of them lack the authority to levy taxes, Nelson argues that there may be little incentive for individuals to choose between these districts. The second modification is to include statemandated programs in the analysis to account in some way for differences in functional respon sibilities among jurisdictions. W ith these modifi cations, Nelson finds the desired systematic relationships, but the precision of the estimates is below the usual acceptable confidence level.2 II. M arke t Structure of Local Governm ents As mentioned previously, one of the prerequi sites for competition is a sufficient menu of choices offered to consumers. Tallying up the number of local governments in the United States casts little doubt on the potential for choice. According to Aronson and Hilley (1986), 79,862 governmental units below the state level existed in 1977. These units tend to fall into two categories: general-purpose and single-purpose governments. General-purpose governments, such as municipalities and counties, provide a variety of services ranging from fire protection to health care. As shown in table 1, municipalities num bered more than 18,000 in 1977, or 24 percent of all governmental units; counties totalled 3,042, or less than 4 percent. Single-purpose units, consisting primarily of school districts and special districts, comprise the majority of local government jurisdictions. As noted in table 1, over 40,000 governmental units have been established to provide only a single function. More than half of these units are special districts, which include sanitary districts, drainage dis tricts, and soil-conservation districts. ■ 2 N els on does find the desired statistically significant relationship betw een the n um ber of general-purpose g ove rnm e nts and the size of the local public sector using state-level data. How ever, in w hat w e take as N e ls o n ’s m ost preferred specification, equation (3) and dependent variable G * , the coefficient on the genera l-p u rpo se -gove rnm e nt variable has a t-value of only 0 .9 1. T h u s , although w e are in total agreem ent with N e ls o n ’s m ethodological ch an ges, w e do not believe that a clear vindication of the decentralization claim s utilizing the state sam ple has been established. The overlapping structure of local govern ments is far from static. Between 1957 and 1977, the number of local governments fell by 22,514, primarily from a conscious attempt to consoli date local school districts. The reduction in the total number of units would have been much I M M M H T A B L E 1 Num ber and Type o f Lo cal Governm ental U n its in th e United S ta te s fo r Selected Years Number of Units Type of Government in local labor markets, rather than on the aggre gate of the state and local public-goods sectors. Consistent with this focus, we adopt two levels of aggregation as the geographical unit of obser vation: the county and the metropolitan area. In addition, as a point of reference to the previous two studies, we also estimate the relationship at the state level. Our data set consists of observations on local public-sector characteristics and relevant demo graphic features of more than 2,900 counties and 280 SMSAs in 1977. This year was chosen for two reasons. First, it is consistent with the studies by Oates and Nelson. Second, some information, such as state mandates, was avail able only during this period. We have analyzed more current data on local-government expen ditures for 1985, while still using state mandates from the earlier period, and find no qualitative differences in the results. 1957 1967 1977 1982 3,047 3,049 3,042 3,041 Municipality 17,183 18,048 18,862 19,076 Township and town 17,198 17,105 16,822 16,734 School district 50,446 21,782 15,174 14,851 Special district 14,405 21,264 25,962 28,588 Variables 102,279 81,248 79,862 82,290 Local government performance is measured by expenditures on the major local public services as a percentage of personal income in either the county or the SMSA, whichever is appropriate. We include local expenditures on local schools, public welfare, fire and police protection, sanita tion, and local parks.3 The key explanatory variable is market struc ture, which is measured by the number of local governments within the appropriate unit of observation. Local governments are divided into the two classes described earlier: generalpurpose and single-purpose jurisdictions.4 Three different measures of the number of local governments are used in the analysis. The first measure is simply the total number of each class of local governmental units found within the appropriate unit of analysis (county or metro politan area). The second method normalizes the number of units by the size of the popula tion served by all of these local governments. The third method divides the number of juris- County Total SOURCE: Num bers obtain ed from A ronson and Hilley (1986), Table 4-1, p. 76. greater during this time if it were not for the creation of more than 11,000 special districts. Between 1977 and 1982, the proliferation of special districts continued, while the number of other types of governmental units remained relatively constant. As expected, local governmental units are concentrated in metropolitan areas. We find that counties in Standard Metropolitan Statistical Areas (SMSAs) have almost twice as many gov ernmental units as do non-SMSA counties— an average of 40 compared to 21. The ratio is even higher for single-purpose units (2.3 to 1), but it is smaller for general-purpose governments (1.6 to 1). In addition, we find that only 25 percent of the metropolitan areas had fewer than 10 general-purpose units and 14 single-purpose dis tricts. O n the other hand, 50 percent of the SMSAs contained more than 21 general-purpose units and 29 single-purpose districts. ■3 III. The Em pirical Test ■4 The basic relationship to be tested is between government performance and market structure. The specification and analysis in this section follow the lines initiated by Oates and Nelson. The principal difference in our study is that we focus solely on local government expenditures Nelson did not include police protection in his estim ation. W e find , how ever, that the results are not sensitive to its inclusion or exclusion. Th e n um ber of general-purpose g ove rnm e nts is the sum of the n um ber of co un ty and m unicipal g o ve rn m e n ts, except in Pen nsylvan ia, N e w Jersey, and the N e w En g la n d states, w here tow nships are also included. Th e n um ber of single-purpose g ove rnm e nts is the su m of the n um ber of to w n s h ip s, school districts, and special districts, except in the aforem entioned states, w here tow n ship s are not included. T h e reason for the exceptions is that the functional responsibilities closely resemble municipalities in these states. dictions by the total land area in the county or SMSA. This last method accounts to some degree for the ease of mobility among the various governmental units. The other explanatory variables include state mandates, per-capita personal income, popula tion, and intergovernmental grants as a percent age of total local tax revenues. The first three variables may be considered proxies for the demand for local public services. As Nelson notes, state mandates may impose binding m ini m um constraints on certain local government activities. As defined by the Advisory Commis sion on Intergovernmental Relations (ACIR), which collected the data, a state mandate is a legal requirement imposed by the state that a local government must undertake a specified activity or provide a service that meets m ini m um state standards.5 The presence of such restrictions would, therefore, be positively associated with the relative size of the local public sector. The demand for local public services should be positively related to personal income, accord ing to traditional consumer demand theory. However, the relationship between income and government spending as a percent age of personal income has been subjected to considerable empirical scrutiny. Investigation of Wagner’s “law” or, perhaps more correctly, Wagner’s hypothesis of a positive correlation between income and government’s relative claims on that income, has sparked much research and has kindled considerable contro versy.6 To our knowledge, the empirical studies have all involved national samples. Our study will provide a simple test of Wagner’s “law” at the local level. An increase in population, holding other variables constant, would also be associated with a larger local public sector. This result in some ways follows the thinking of Wagner, w ho saw an increase in population density and urbanization leading to increased public expen ditures on personal protection and economic regulation (Bird [1971]). The ratio of intergovernmental grants to local tax revenues measures the extent to which local per capita ■5 governments rely on higher-level governments for funds. Because of the matching provisions of many federal and state grants, we would expect the grants to stimulate local government expenditures.7 Results Fourteen separate models were estimated: one for each level of aggregation and for each mea sure of decentralization. The estimates displayed in table 2 for one of the models are typical of the results found for the other models. We find that an increase in decentralization of governments, measured by any one of the three measures, is statistically significantly re lated to a decrease in the size of the local public sector. This finding supports the decentraliza tion hypothesis: an increase in jurisdictional fragmentation is associated with a decrease in local budget share. O n the other hand, we find that an increase in the number of units increases the local budget share. This suggests that the costs of providing services through special dis tricts outweigh the constraining effects that competition may impose on spending or the savings that result from economies of scale. Thus, our results support the argument that the proliferation of special districts has increased local spending. The negative and significant coefficient on per capita income is evidence against the rele vance of Wagner’s hypothesis applied to the local government sector. At the state level, we find a positive relationship, as does Oates. A negative correlation between local public-expenditure share and income is not unexpected, however. Most studies of local public-expenditure demand find income elasticities that are significantly less than unity, which implies a decline in aggregate budget share as average community income rises.8 The positive coefficients on the population and intergovernmental transfer variables are consistent with our earlier discussion. general-pur pose single-purpose Th e A C IR surveyed local gove rnm e nts about 7 7 functional s u b c o m p o n e n ts in five broad areas: state personn el, other than police, fire, and education (15 c o m p one nts); public safety (31); environm ental protection (8); social services and m iscellaneous (10); and education (13). ■7 King (1984) offers a com p reh en sive s u m m a ry and critique of the effects of grants on local gove rnm e nt spen ding. ■ 6 Bennett and Jo h n s o n (1980) provide a com prehensive s u m m a ry of the debate and a co m p endium of the em pirical results. Ra m (19 8 7) ■ 8 Inm an (19 79 ) includes a s u m m a ry of studies of the dem a nd for appears to have m ade the m ost recent contribution to the literature. local public services. T A B L E 2 Regression Results a t th e S M S A Le v e l, 1 9 7 7 Variables_______________ Mean (Standard error) Coefficient (T-statistic) Number of generalpurpose units 28.8 (40.83) -.015 (4.48) Number of singlepurpose units 54.1 (80.55) .005 (2.79) Per capita income ($ 1,000s) 6.67 (.98) -.317 (2.87) Ratio of transfers to local taxes 1.18 (.53) .559 (3 .02) Population in SMSA (100,000s) 5.53 (10.04) .45 (3.85) Total state mandates 37.0 (11.92) .083 (H-59) Constant 5.23 (6.17) Dependent variable: local expenditures per personal income 6.94 (1.80) Number of observations R-square 289 .43 SOURCE: G overn m en t ex pen d itu re data from Census o f G overnm ents, 1977; personal in c o m e and p op u la tion data from the Bureau o f E c on om ic Analysis; state mandates c o m p ile d b y the ACIR. Various M easures of De ce n traliza tion The conclusion that increased decentralization of general-purpose governments is associated with a smaller local public sector is supported by our analysis regardless of which measure of decentralization is used. As seen in table 3, not only are the coefficients statistically significant at the 1 percent level for SMSAs and counties, but the magnitudes of the elasticities are also of similar magnitudes, with few exceptions. For example, at the SMSA level (column 1), we find that a 10 percent increase in the number of general-purpose jurisdictions reduces the local public sector’s share of personal income by 0.6 percent. In the case of SMSAs, a 10 percent increase in general-purpose governments would mean only an additional three units. However, when state-level data are used, the statistical significance of the estimates falls below the 10 percent confidence level. The only excep tion is the effect of the number of generalpurpose governments, which is statistically sig nificant right at the 10 percent level. Table 3 also reveals that the size of the local public sector at the SMSA level is slightly more responsive to a change in the number of generalpurpose governments than to a change in the number of single-purpose governments. This relationship holds no matter which decentraliza tion measure is used, but is less consistent at the county level. IV. Conclusion We have found a significant relationship be tween governmental structure and government size. Two basic relationships emerge from the analysis. First, an increase in the number of general-purpose government units within a metropolitan area or county boundary reduces the share of personal income going to the local public sector. Second, an increase in single purpose government units has the opposite and equally significant result of increasing the size of the local public sector. The difference in behavior between the two types of governments underscores our conclu sion that competition among local general-pur pose governments constrains local government spending. Recall that suppliers are disciplined by the presence of other suppliers only when they provide similar services to the same mar ket. General-purpose governments meet this requirement more closely than do single-pur pose governments. Typically, a single-purpose government is the sole supplier of a specific service within a local market, whereas each general-purpose district provides a similar array of services. Thus, the existing structure of government creates two opposing forces of government behavior. Competition among general-purpose units, such as municipalities, constrains local government spending. O n the other hand, the overlapping labyrinth of single-purpose govern ments stimulates local government spending. Much of the current arrangement of local governments resulted from attempts by states and localities to respond to changing conditions within the various constraints imposed on them. As a practical matter, states and munici palities have limited ability to respond to chang ing conditions. States are constrained by local loyalties, vested interests, and the inertia of the T A B L E 3 Relationship Betw een Various M easures of Lo cal Governm ent Com petition and Lo c a l Govern m ent Expe nd itu re s as a Fra ctio n of Personal Incom e Measure of Competition SMSA All Level of Aggregation County Non-Metro Metro State A. Number of units General-purpose Single-purpose - .063 .045 .043 .054 .069* .040 .034 .046 .042 .005** - .076 .036 .062 .068 .032** .050 .035 .045 .033 .019** - .065 .018 .022 .016 .055 .005 .028 .023 B. Number of units per capita General-purpose Single-purpose C. Number of units per square mile General-purpose Single-purpose Note: N um bers are ex p ressed as elasticities. All estimates are significant at the 1 p ercen t level unless d en oted b y an asterisk. A single asterisk denotes significance at the 10 p ercen t level but less than 5 percent level. A d ou b le asterisk denotes significance at less than the 10 p ercen t level. T he estimates are d erived b y regressing the local govern m en t expen ditures as a p ercen t o f personal in co m e against m easures o f g ov ern m en t co m p e titio n , p opu lation , per capita in co m e , intergovernm ental revenue, and state program mandates. Estimates o f a typical regression equation are sh o w n in table 2. SOURCE: Authors. status quo. The power of localities to handle public services is often made difficult by state statutes that limit powers to tax and to incur debt. Since the late 1950s, special districts have been established as a means of circumventing these constraints by shifting responsibilities away from general governments. The federal government has further stimulated the creation of special districts through “direct advocacy.” Many federal agencies would rather deal directly with officials of special districts than with offi cials from general governments such as counties or municipalities (Aronson and Hilley [1986]). In the past few years, a number of states have begun to take a systematic look at the current structure of local governments. Several states have established advisory commissions to con sider reorganizing and streamlining the per ceived fragmented system of local governments that dot their landscape. These commissions appear to be particularly concerned about how the large number of special districts affects the provision of services. Our analysis provides some information that may be useful to these reform efforts. First, our results suggest that reform efforts directed toward special districts are well-guided. Clearly, an increase in the number of single-purpose governments, which consist mostly of special districts, increases government spending. Although these results are very strong, we should caution that we have not been able to control entirely for differences in the level of services provided by these governments. It may be the case that part of the observed increase in spending associated with greater numbers of units simply indicates that additional special districts are providing additional services. Second, our results warn against lumping together general-purpose and single-purpose governments when considering streamlining local government structure. We show that the two different types of governments exhibit distinctly opposite behavior. Third, our results suggest that a competitive environment among specific types of local gov ernments can constrain government spending and promote the efficient provision of local public services. R EFER EN C ES Advisory Commission on Intergovernmental Relations. Washington, D.C.: U.S. Government Printing Office, 1978. tures. State Mandating of Local Expendi Financ Aronson, J. Richard, and John L. Hilley. (4th ed.). Washington, D.C.: The Brookings Institution, ing State and Local Governments 1986. The Political Economy of Federal Government Growth: 1959-1978. College Station, TX: Bennett, James T., and Manuel H. Johnson. Texas A&M University, 1980. Bird, Richard M. “Wagner’s ‘Law’ of Expanding State Activity” 26(1971): 1-26. Public Finance. Inman, Robert P. “The Fiscal Performance of Local Governments: An Interpretive Review.” In Mahlon Straszheim and Peter Mieszkowski, eds., Baltimore, MD: Johns Hopkins University Press, 1979: 270-321. Current Issues in Urban Economics. Fiscal Tiers: The Economics of Multi-Level Governments. London: Allen and King, David N. Unwin, 1984. Nelson, Michael A. “Searching for Leviathan: Comment and Extension.” 77 (March 1987): 198-204. American Eco nomic Review. Bureaucracy and Repre sentative Government. Chicago: Aldine- Niskanen, W illiam A. Atherton, 1971. Fiscal Federalism. Oates, Wallace E. New York: Harcourt BraceJovanovich, Inc., 1972. ________ . “Searching for Leviathan: An Empirical Study.” 75 (Sep tember 1985): 748-57. American Economic Review. Ram, Rati. “Wagner’s Hypothesis in Time-Series and Cross-Section Perspectives: Evidence from ‘Real’ Data for 115 Countries.” 69 (May 1987): 194-204. Economics and Statistics. Review of Zax, Jeffrey S. “Is There a Leviathan in Your Neighborhood?” New York: City University of New York, Department of Economics, November 1987. Exit Barriers in the Steel Industry by M a ry E. D eily M a ry E . D eily is a visiting econom ist at the Federal Reserve B ank of Cleveland and an assistant professor of ec ono m ics at Texas A & M University. Th e author w ould like to thank Paul Bauer, Randall E b e rts , Erica G ro s h e n , Stephen Karlson, and M ark Sniderm an for helpful c o m m en ts. Introduction The U.S. steel industry seems perennially afflicted with overcapacity. Even after numerous plant closings, and despite recent high capacityutilization rates, analysts suggest that another 15 to 20 percent of current capacity should close. W hy has overcapacity been a chronic ailment of steel firms during the 1970s and 1980s? W hy haven’t firms closed plants more quickly, since continued operation of these plants depresses profits for the entire industry? The persistent survival of excess capacity is not inexplicable. In theory, a market system reallocates resources from activities yielding lower-than-normal returns to activities with higher returns. In practice, however, firms can be locked into a low-profit activity if large losses are incurred when capital is transferred to new activities. These potential losses form an exit barrier, delaying plant closings, depressing prof its, and prolonging adjustment for the entire industry.1 The primary purpose of this paper is to examine the nature and size of exit barriers in the steel industry. First, the necessity for contrac tion in this industry is summarized. Then basic exit theory is reviewed, and several types of exit barriers that seem most pertinent to the steel industry are described. The potential size of these barriers in the steel industry is assessed. Finally, the possible effects of current trade protection and pension-insurance policies on the size of exit barriers in the steel industry are discussed. This paper argues that high exit barriers have significantly slowed the industry’s contraction by delaying plant closings. These barriers explain why capacity has fallen slowly even though industry profits have been subnormal since the late 1950s. They also help to explain why the industry failed to modernize some plants, even though these increasingly ineffi cient plants continued to operate into the 1980s. I. The N ecessity for Contraction ■ 1 Th e term “ exit barrier" is perhaps un fortun ate, as it carries the conn otations of inefficiency attached to the phrase “ entry barrier.” S uch is not the case: exit barriers are the va riou s cost conditions that m ake lengthy exit a rational response by firm s. The U.S. steel industry has performed poorly during the last 25 years. Profits for the industry have been low compared to the average man ufacturing return in virtually every year since 1958.2 And despite the industry’s recent buoy ant performance— part of which appears to be due to trade protection— long-run trends in steel demand and steel supply point to continued low profits in the future. Structural changes in steel demand have greatly reduced the growth of the market. These changes, which include increased use of steel substitutes such as aluminum and plastic, and reductions in the amount of steel used in con sumer durables, particularly cars, have reduced the U.S. economy’s need for steel. The average annual growth rate of U.S. apparent steel con sumption has fallen from from 4.1 percent dur ing 1960-1969, to 1.9 percent during 1970-1979, to 0.2 percent during 1980-1986. Not all steel firms have fared the same, how ever. The industry basically consists of two parts: integrated mills and minimills. The integrated mills, which produce steel from iron ore, are the traditional steel industry, while the minimills, which produce steel products by recycling steel scrap, are relative newcomers. It is the integrated portion of the industry that has performed so poorly; minimills have flourished, increasing their market share from about 3 percent in I960 to 18 percent in 1985. As their name suggests, minimills produce steel on a m uch smaller scale than integrated plants, reducing the size of the required capital commitment considerably 3 The mills also ben efit from employing workers at lower wages. Though their costs are extremely sensitive to the price of scrap minimills have become very competitive in the product lines in which they specialize, drastically reducing the integrated mills’ sales in these markets.4 In addition, integrated firms in the U.S. faced tough new competition from imports for a share of the market, as fundamental changes in input costs during the 1950s and 1960s altered the comparative advantage in steelmaking. Two studies, by Crandall (1981) and by Kawahito (1972), examine the changes in the relative cost of materials in the U.S. compared to other countries, particularly Japan. Formerly, abundant supplies of coal and iron ore assured U.S. pro ducers of a materials cost advantage that, along with greater U.S. productivity, more than com pensated for higher U.S. wage rates. However, the discovery of rich iron-ore sources in several parts of the world and the decreased cost of ocean shipping began to reduce the traditional U.S. advantage. Also, as Barnett and Schorsch (1983) point out, countries like Japan experienced phe nomenal growth in steel consumption after World War II. Their steel industries were able to build entirely new, large-scale plants, since their rapidly expanding markets could easily absorb the output of the additional capacity. These new plants incorporated the latest technology into an optimal plant layout, resulting in high productiv ity growth. Increased productivity growth, combined with lower wage rates, reduced the unit cost of labor further below U.S. levels. This advantage, added to the favorable changes in materials costs, made foreign steel very com petitive with U.S. integrated production.5 The result has been a decline in the market share of integrated steel firms in the U.S. from more than 90 percent in I960 to less than 65 percent in the 1980s. Given the slow growth of the market, these figures translated into a need to cut integrated steel capacity by closing plants. And, in fact, the industry has closed plants. From its height in the early 1970s of approx imately 155 million tons, annual raw steel capac ity has fallen to about 112 million tons. But the contraction of the industry has taken a long time, even though capital has been earn ing subnormal profits for many years.6 Rather than moving into other activities, firms appear to be clinging tenaciously to capacity by nursing along aging plants, as if the growth in demand for steel might miraculously increase to pre-1970 levels. But as the discussion in the next section shows, this response may wrell be optimal for firms facing high exit barriers. ■ 2 See Crandall (1981), p. 29 , for the rate of return on equity after taxes in steel versus all U .S . m anufacturing for the years 19 5 4 -19 78 . See U .S . D epa rtm en t of C o m m e rc e , Bureau of the C e n s u s , Financial Reports for Manufacturing Corporations, Quarterly va riou s issues, for subsequent years. ■5 In fact, Crandall (1981) co ncludes that a totally new integrated plant w ould be a p oo r investm ent in the U n ited S tates, given his estim ates of the possible reductions in labor and energy savin gs attainable. ■3 M inim ills typically consist of an electric steel furnace, a co ntinu ou s billet caster, and som e kind of finishing mill, usually for bars. See ■ 6 Th e first m ajor plant closings, those of You n g s tow n Sheet & Tube Miller (1984) for a goo d description of this technology. and the United States Steel C orp oratio n at Y o u n g s to w n , did not occur ■4 addition, because capacity is usually m easured as the ability to produce until the late 19 7 0 s , and the next episode did not occu r until 19 8 2. In M inim ills have a cost advantage over all integrated m ills, w hether dom estic or foreign, in the products they can p roduce. See Barnett and raw steel, estimates of capacity reductions m ay be som ew ha t overstated. Crandall (1986) for a detailed com p arison of minimill to integrated mill T h e introduction of continu ou s casters has increased the yield from raw production costs. steel by 10 to 15 percent. II. A M odel of the P la n t Closing Decision The neoclassical prediction for a competitive industry facing an inward-shifting demand curve is that high-cost plants will exit, leaving the lowest-cost plants to produce in the long run. However, as long as variable costs are covered, a firm will continue to operate an exiting plant that has fixed costs, since doing so minimizes the firm’s losses.7 During this period the firm will not make any major reinvestments; instead, it will disinvest from the capital in place. Because most production processes do involve fixed costs, the decision to close a plant usually will involve a period of operation and disinvestment before shutdown. The optimal closing point will not occur until the net reve nue, which is the return to continued operation of the capital in place, equals the return that could be earned on the salvage value. Thus, the speed with which a firm closes a plant depends on how quickly net revenues decline and on the amount of capital that can be salvaged once the plant is shut down. Clearly, one important factor that will affect the timing of plant closings is the general level of economic activity. W hen sales decline during recessions, they increase the probability of plant closings by reducing net revenues. This is especially true for a cyclical industry like steel. Other factors are also important, however. Since the firm will not replace the aging capital with new equipment, one determinant of a plant’s net revenues is the amount of mainte nance the capital in place requires in order to operate (in other words, its durability). The firm will continue to bear maintenance expenditures as long as the capital generates enough revenue to cover both the additional expense and other variable costs. Obviously, the larger the mainte nance expenditures, the more they reduce net revenues, and the less likely they will be worth making.8 A low salvage value may also delay a plant’s closing. The salvage value is the net amount of money the firm will realize when the plant closes. A large positive value means that much of the capital can be extracted without loss from the plant, thus shortening the time to shut down. A negative value extends the time before exit, causing the plant to be operated even though total variable costs are not covered. In this situation, the firm would actually borrow to pay the uncovered variable costs in order to avoid the greater loss of closing.9 In general, the salvage value is determined by a plant’s resale value minus costs incurred dur ing closing. The resale value of the capital depends on its specificity to the production process and on output growth in the industry. The closing costs include the resources neces sary to gather the information to make the closing decision and the time spent planning and executing it. The firm may also face em ployee-related closing expenses, such as sever ance pay, early retirement pay, and pensions, depending on previous contractual agreements or on local plant-closing legislation. Increases in these costs, by raising closing costs, will delay shutdowns.10 Thus, in a contracting industry with durable and specific capital and high closing costs, firms will delay closing plants. The plants exit even tually, but only after a long period of disinvest ment. The result of selective and drawn-out disinvestment is a gradual increase in the average age of the industry’s capital stock and a slowing of productivity growth. Two things are vital to remember, however. First, in an industry with high exit barriers, a slow decline is the optimal rate of closure, despite years of poor earnings by the industry Resources are always being utilized in their highest return activity during a contraction. Second, although an industry may appear to be failing because of lack of reinvestment, the antiquated plants are the result of exit barriers’ prolonging exit and are not the cause of the industry’s decline. While some plants will be modernized, those that are exiting will receive little investment. In sum, an important consequence of allow ing the market to reallocate resources from an industry with high exit barriers is that capacity will contract slowly, with old capacity lingering on and plants closing in bunches during dow n turns that lower revenues. ■9 The cost of going bankrupt, instead of continuing to pay uncovered variable co sts, w ould be an up w ard b ou nd on the am ou nt the firm w ould be willing to borrow in this situation. ■7 ■ 10 T h is conclusion d e p e nds on the sim plifying assum ption m ade In this co ntext, fixed costs refer to costs that m ust be paid w hether the plant is open or closed. here that closing costs do not increase over tim e . A s pointed out by Littm an and Le e (1983), if em ployee-related closing costs rise quickly with the seniority of the w ork fo rc e , then a firm m ight accelerate closing to ■ 8 See La m fa lu s sy (1961) for a discussion of these issues. avoid the greater future liability. III. Th e S ize of E x it Barriers in the Steel Industry Clearly, the magnitude of exit barriers in an industry depends on three factors: how long gross revenues are expected to cover variable costs, how specific and durable the capital is, and how high closing costs are.11 This section presents some information about these factors in the steel industry which suggests that exit bar riers are large. T A B L E 1 C osts of Steel Production in the U .S . (Current dollars) 1976 1986 $217.00 $206.00 Total Variable Cost of Finished Stet 1 Materials, Energy, and Labor (per net ton of finished product) 310.28 348.00 Total Cost of Finished Steel (per net ton of finished product) 361.38a 449.00 Total Variable Cost of Raw Steel Materials, Energy and Labor (per net ton) a. T he nu m b er cite d here is slightly low er than the figure re p orted b y the C ou n cil o n Wage and Price Stability, but is calculated as they d escribe in the text. SOURCES: U.S. C o u n cil o n Wage and Price Stability (1977), p. 60; W harton E con om etrics (1987), p. 4.5. A rough idea of the likelihood that gross revenues will cover variable costs— the costs of all variable inputs to production— can be obtained by comparing the average variable cost of a ton of steel to the prices of various steel products. This cost is conventionally measured as the sum of labor, energy, and materials. The U.S. Council on Wage and Price Stability cal culated that the average total variable cost per net finished ton of steel in 1976 was $310.28. W harton Econometrics estimated that this cost equaled $348.00 in 1986. These estimates include the cost of producing raw steel, as well as the average industry cost of finishing it. Both of these studies also include estimates of the financing costs of steel production, taken here to be the average fixed cost of production (see table 1). Table 2 compares estimates of average vari able cost and average total cost for selected steel products to the average realized price per net ton of those products in 1976 and in 1986. In most cases, product prices were above the average variable cost. O n the other hand, almost all of these prices were well below the total cost of finished steel. (Product prices do vary cyclically, causing the size of this shortfall to change over time. See table 3.) Overall, the data indicate that product prices may fall consider ably below the average total cost without making immediate shutdown a firm’s loss-minimizing alternative. How long does a plant that is not covering total cost continue to operate? As stated above, unless prices dip or variable costs rise unexpect edly, a plant’s closing would depend on the durability of its capital, on its resale value, and on the amount of closing costs. O f these three, the high cost of closing appears to be the most important exit barrier currently in the steel industry. W hen closing a plant, a firm records a charge for the costs of dismantling the mill, for the operating loss until closing, for losses involved with contract termi nations, and for a write-down of the assets. It also records the estimated liability for current and future payments to employees for pensions and insurance benefits. The payments due to the work force when an integrated steel plant closes are substantial. For instance, by the provisions of a typical labor contract, qualified union members on layoff because of a permanent closing are eligible for severance pay, supplemental unemployment benefits, pension payments and, in some cases, supplemental pension payments.12 Severance pay for union members with at least three years of seniority equals four to eight weeks’ wages, depending on their years of service. A firm continues to pay life- and medical-insurance premiums for six to 12 months for workers with at least two years of continuous service. Workers may also be entitled to supplemental unemploy ment payments for up to two years. One of the largest parts of the employeerelated closing costs is the estimated liability for future pension payments. O f course, the portion of closing costs represented by the pension liability is not by closing, since the firm caused ■ 11 See C a ve s and Porter (19 76 ) and Porter (19 76 ) for an exhaustive 12 list of various possible exit barriers. T h e types of barriers discussed here ■ are those that seem particularly pertinent to the steel industry. contracts m ade in later years appear to be quite similar. Th e contract described here becam e effective in 1980. Term s of owes retiring workers their pensions if the plant stays open. Nor are all of these charges out-ofpocket expenses. But they do represent pay ments that the firm must fund from some new source, since the cash flow from the plant will cease. This places an increased burden on a firm’s remaining mills.13 T A B L E 2 Price and C ost Es tim a te s fo r Selected Steel Pro d u c ts, 19 76 and 1986 (Current dollars) 1976 Hot-Rolled Sheets Cold-Rolled Sheets Hot-Dipped, Galvanized Sheets and Strip Hot-Rolled Bars Structurals Average Variable Cost Average Realized Price Average Total Cost $282.30 328.94 $229.43 288.43 $333.40 380.04 356.92 286.96 272.97 368.59 311.14 358.94 408.02 338.06 324.07 Average Variable Cost Average Realized Price Average Total Cost $305.00 376.00 $273.04 418.21 $406.00 477.00 419.00 313.00 291.00 537.93 360.03 321.57 520.00 414.00 392.00 1986 Hot-Rolled Sheets Cold-Rolled Sheets Hot-Dipped, Galvanized Sheets and Strip Hot-Rolled Bars Structurals Note: T he co st data from table 1 w ere adjusted for variation in finishing costs am on g p ro d u cts using data from W h arton E con om etrics (1987), p. 4.7. Estimates are industry averages; costs are b o u n d to b e higher in exiting plants. SOURCE: Bureau o f the Census, C u rren t In d u stria l R ep orts: Steel M ill P rod u cts, various issues. In addition, because of the terms of pension agreements in this industry, the pension pay ments are actually higher if workers retire from a closing plant rather than from an operating mill. Under normal circumstances, union members are eligible for pensions after 30 years of service, or at age 65 (with 10 years of service), or at age 60 (with 15 years of service). But for workers laid off by plant closings, the eligibility require ments are eased. For instance, workers over 55, whose age plus years of service equal at least 70, become eligible. Also, some workers receive supplemental pension payments of $400 per m onth until they reach age 62, if they are laid off by a shutdown. By the terms of this typical labor contract, it is clear that the size of the payments depends crucially on the age of workers and on their years of service. A firm might be able to reduce the work force somewhat by attrition before closing a plant, but under a seniority system, the remaining workers would tend to be older, with more years of service, which would drive up closing costs.14 These claims raise the cost of closing steel facilities enormously In 1979, the United States Steel Corporation shut down a variety of mills and parts of mills, laying off more than 11,000 workers. According to the company’s annual reports, the total cost of the closings was approximately $650 million, of which about $415 million represented labor-related expenses, implying a cost per worker of more than $37,000. Bethlehem Steel reported similar fig ures in its annual report, recording a $700 million liability in 1982 when about 18,000 workers were laid off during a restructuring that dealt principally with steel facilities. More recent estimates show that these costs may be higher. One study indicates that the total cost per employee of closing a mill is $75,000, of which $54,000 represents employee-related closing costs (Wharton Econometrics [1987]). Using these figures, the Bethlehem Steel restruc turing would currently cost $1.35 billion. Firms cannot depend on high resale values to cover the large closing costs. The capital is quite specific to the industry and is of little value for any purpose other than steelmaking. Nor are other steel firms particularly interested in buy ing these plants; most integrated firms are reducing their capacity, and minimills are build ing new plants. Furthermore, the equipment in a closed plant is usually in need of major invest ment, since the former owner has disinvested from it before closing.15 ■ 14 It is difficult to evaluate h ow these em ployee-related costs change over tim e. T h e severance paym ent form u la does not appear highly sensitive to the seniority profile of the plant: the m axim um sever ance pa ym ent is earned by w orkers with 10 years of experience. Th e supplem ental pension pa ym ent is m ore com p licated. Th e liability w ould increase if the n um ber of qualifying w orkers rose over tim e (w orkers ■ 13 Th e problem is similar to that of Social S ecurity w hen future qualify if their co m b ined age and years of service is over a certain generations are smaller. W hile in 1 9 7 7 there were 2 .3 w orkers for each m inim um ), and w ould fall if the num ber of qualifying w orkers fell over retiree, currently there are tw o retirees for eve ry steelworker. tim e (w orkers receive the pa ym ent only until age 62). T A B L E 3 A verage Realized Prices of Selected Steel Prod ucts (Dollars per net ton) Year Hot-Rolled Sheets Cold-Rolled Sheets 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 $229.43 254.15 281.10 314.87 317.30 350.12 338.79 325.53 326.01 310.35 273.04 $288.43 320.51 354.31 388.78 395.42 436.77 433.87 437.93 453.18 437.97 418.21 Hot-Dipped Galvanized Sheets & Strip $368.59 392.72 430.35 468.76 487.64 532.31 525.84 525.87 560.16 536.75 537.93 Hot-Rolled Bars Structural $311.14 337.23 364.26 403-38 415.90 445.83 414.94 387.38 393.49 366.89 360.03 $284.46 293.41 326.98 372.02 408.03 428.82 421.90 362.64 358.52 332.57 321.57 SOURCE: Bureau o f the Census, C u rren t In d u stria l R ep orts: Steel M ill P rod ucts, various issues. During the industry’s contraction, there have been few examples of closing plants sold for continued operation as integrated steel mills. (One notable exception is the plant in Weirton, West Virginia. The employees purchased this mill from National Steel and have continued integrated production.) W hen sales do take place, the purchasers are generally interested in the rolling and finishing facilities, and keep steel furnaces closed. For instance, California Steel now imports semifinished steel for finishing at a (formerly integrated) plant in Fontana, which it purchased from Kaiser Steel.16 There are few opportunities to sell individual pieces of equipment. One company reportedly auctioned off some equipment when it went bankrupt, and some used equipment has been sold abroad, but no steelmaking operations have been sold for movement. Inventories of raw materials and parts can be distributed to other plants, but beyond that, the equipment is likely to sit until the price of steel scrap rises enough to pay the junk dealer for dismantling it. ■ 15 Fro m 1960 to 1981, the average annual investm ent per ton of capacity in m ajor pieces of steelm aking equipm ent w as $ 3 4 .0 8 in plants w ho se closing w as an no un ced before 19 8 4 , com p ared with $ 1 2 8 .2 7 for plants rem aining open (Deily [1988]). See Deily also for evidence that steel firm s channeled investm ent away from plants that were least able to com p ete with im ports and minimills, particularly during the period 1971-1981. ■ 16 See W ha rton Econ om etrics (19 8 7), p. 1 .8 , and J . Ern est Beazley, “ Big Steel’s P u sh to E xte n d Im port Q u o ta s D ra w s D eb a te ,” Journal, D ece m b er 3 0 ,1 9 8 7 . Wall Street The last exit barrier, the durability of steel industry capital, also works to delay plant clos ings by allowing the continued operation of aging equipment without major reinvestment. Furnaces and mills are depreciated over 15 to 20 years, but may operate for longer. For example, table 4 indicates that the ages of various pieces of capital were more than 10 years in 1979, and that a significant percentage of the equipment had been operated for more than 20 years. O f course, operation of the equipment still involves noncapitalized maintenance and repair expenditures. In addition, the blast furnaces, which provide the flow of hot metal to the steel furnaces in an integrated plant, require periodic relining. Blast furnaces basically operate on a continual basis for two to eight years, depending on their rate of utilization. But eventually the refractory material that prevents the hot metal from destroying the furnace must be replaced. Figures cited for a somewhat short-term repair process, called gunning, range from $14 million to $18 million. Actual replacement of refractories may cost anywhere from $20 million to $100 million, depending on the extent of the replace ment and furnace rebuilding, though on average the cost will probably fall into the $20 million to $50 million range. Frequently, firms will postpone a reline and leave the blast furnace idle, provided they have another operating furnace. But there are some limits to their ability to escape both operating losses and closing costs by idling entire plants. For instance, after being laid off for two years because of idled equipment, workers eligible for average pensions may claim them. Also, laid-off workers are eligible for supplemental unemployment benefits for up to two years. In sum, integrated steel firms appear to face sizable exit barriers. High closing costs, consist ing principally of payments to employees, cur rently appear to be the largest barrier. Durable capital and low resale values also work to delay plant closings. Average Age of Capacity (years) Capacity Over 20 Years O ld (percent) 17.3 11.0 14.3 19.0 17.5 46.9 2.3 25.3 31.5 33.3 Coke Ovens Basic Oxygen Furnaces Electric Furnaces Hot Strip Mills Aggregate*3 a. As o f January 1,1979. b. Includes data o n o p e n hearth furnaces, plate m ills, w ire r o d mills, c o ld strip m ills, and galvanizing lines. SOURCES: A m erican Iron and Steel Institute (1980), p. 21. Based o n data from The W orld Steel In d u stry D a ta H a n d b o o k , vol. 1, and the A m erican Iron and Steel Institute. IV. Im plications fo r Public Policy The data presented here suggest that the decline of the steel industry has been painful and pro longed because of large closing costs and high exit barriers created by the technology of the production process. Although these barriers have delayed closings, resulting in lower profits and antiquated capital stocks in some plants, the necessary reduction of U.S. integrated steel capacity has been taking place, albeit slowly. Is there any need for policies aimed at raising or lowering exit barriers? Although different firms, workers, stockholders, and communities could gain or lose, it is not at all clear that the economy as a whole benefits from either hasten ing or delaying plant closings. However, public policy in at least two areas of recent concern may have a strong impact on the steel industry’s exit barriers. First, the pension-insurance program affects exit barriers in the steel industry by altering the cost of closing plants. The Pension Benefit Guar anty Corporation (PBGC), a federally chartered agency that insures all workers with definedbenefit pensions, has already assumed some of the industry’s plant closing costs and may ulti mately assume more. As stated previously, pen sion liabilities are a major part of the cost of closing. A firm that desires to close plants, but that cannot afford to do so, may find that declaring bankruptcy is the cheapest way to reduce capacity, because the PBGC becomes responsible for the firm’s pension liabilities.17 Thus, at least potentially, the PBGC could end up paying the pension liability portion of some firms’ closing costs, thereby speeding up plant closings by lowering this particular exit barrier.18 The situation has become more uncertain, how ever, because of the recent and still-unresolved differences between the PBGC and LTV Steel over responsibility for the latter’s pension lia bilities. Since this uncertainty makes it more difficult for firms to evaluate plant-closing deci sions, it is important for policymakers to clarify w ho will ultimately pay these liabilities. Policies to protect the industry from imports, on the other hand, may raise the exit barriers that steel firms face. The industry is currently protected by five-year Voluntary Restraint Agree ments that the Reagan administration has negoti ated with a number of steel-exporting countries. In the short run, the effect of the quotas may be to delay plant closings if the protection causes the industry to upwardly revise the expected revenues of its plants. The long-run effects of the legislation are less clear. Firms are unlikely to reverse their long-run disinvestment from marginal plants unless they are convinced that the profitability of these plants has increased permanently. Such an assurance would require at least that the govern ment make a long-term commitment to trade protection for the industry. But such a com m it ment would be expensive for domestic indus tries that use steel, and would by no means rule out further capacity reductions, since the minimill sector will continue to grow. ■ 1 7 See B uynak (19 8 7) for a description of the limits on the am o u n t of the firm ’s assets that the P B G C can claim to cover unfunded pension liabilities. ■ 18 Indeed, since the m axim u m paym ent the P B G C m akes to w orkers m ay be well below w orkers’ contracted p en sion s, and since supplem ental paym ents for early retirement are not covered, the total cost of closing plants w ould be lower, thou g h at the direct expen se of the em ployees. What public policy should not be doing is forcing reinvestment in the steel industry The most misguided aspect of the trade protection currently in place is its requirement that the industry reinvest its net cash flow from steel businesses back into steel plants (Steel Import Stabilization Act of 1984,19 U.S.C. 2253). The result of this directive may be to force invest ment in plants that will never yield an adequate return, a circumstance that will increase plant owners’ losses when the plants are eventually closed. Lack of reinvestment is not the underlying problem of the steel industry. Although invest ment in the plants that will survive is essential to their competitiveness, it is clear that additional capacity will eventually close. But shutdowns will be delayed as long as steel firms find that exit barriers make continued operation of mar ginal plants less costly than closing. REFER EN C ES Up from the Ashes: the Rise of the Steel Minimill in the United States. Washington, D.C.: The Barnett, Donald F., and Robert W. Crandall. Brookings Institution, 1986. Steel: Barnett, Donald F., and Louis Schorsch. Cambridge, MA: Ballinger Publishing Co., 1983. Upheaval in a Basic Industry. Beazley, J. Ernest. ‘Agency in Crisis: Bankruptcy Filings in Steel Overwhelm U.S. Pension Insurer.” May 21,1987, Wall StreetJournal. pi. Buynak, Thomas M. “Is the U.S. Pension-Insurance System Going Broke?” Federal Reserve Bank of Cleveland. January 15,1987. Economic Com mentary. Caves, Richard E., and Michael E. Porter. “Bar riers to Exit.” In R.T. Masson and P. David Qualls, eds., Cambridge, MA: Ballinger Publishing Co., 1976. Essays on Industrial Organiza tion in Honor of Joe S. Bain. The U.S. Steel Industry in Recurrent Crisis: Policy Options in a Com petitive World. Washington, D.C.: The Brook Crandall, Robert W ings Institution, 1981. Deily, Mary E. “Investment Activity and the Exit Decision.” (1988— forthcoming). Statistics. Review of Economics and TheJapanese Steel Industry, with an Analysis of the U.S. Steel Import Problem. New York: Praeger, 1972. Lamfalussy, Alexandre. Investment and Growth in Mature Economies: The Case of Belgium. Kawahito, Kiyoshi. New York: St. Martin’s Press, 1961. Lawrence, Robert Z., and Robert E. Litan. Saving Free Trade: A Pragmatic Approach. Washington, D.C.: The Brookings Institution, 1986 . Littman, Daniel A., and Myung-Hoon Lee. “Plant Closings and Worker Dislocation.” Federal Reserve Bank of Cleveland. (Fall 1983): 2-18. Economic Review. Miller, Jack R. “Steel Minimills.” 250(1984:5): 32-39. can. Scientific Ameri Porter, Michael E. “Please Note Location of Nearest Exit: Exit Barriers and Planning.” 19 (1976:2): California Management Review. 21-33. The Theory o f Price, Stigler, George J. Third Edition. New York: Macmillan Co., 1966. United States Council on Wage and Price Sta bility. Report to the President on Prices and Costs in the United States Steel Industry. Washington, D.C.: U.S. Government Printing Office, October 1977. United States Department of Commerce, Bureau of the Census. Series MA-33B. Washington, D.C.: U.S. Government Printing Office, various issues. Current Industrial Reports: Steel Mill Products. Pension Agree ment Between United States Steel Corporation and United Steelworkers of America. Effec United States Steel Corporation. tive July 31,1980. Document supplied by the United Steelworkers. Restructuring and Revival: The World Steel Industry, 1987-2000. W harton Econometrics. Volume III, Part II. Bala Cynwyd, Pennsylva nia: Wharton Econometrics, 1987. W hy Do W ages Vary A m ong Em ployers? by E rica L. G roshen Erica L . G rosh en is an econom ist at the Federal Reserve B ank of C leveland. Th e author would like to thank Jo h n T. D u n lo p , Richard B. F re e m a n , and Law re nce H . S u m m e rs for their co m m en ts and suppo rt on earlier versions. T h e paper also benefited from suggestions by M a ry Deily, Randall W . E b e rts , and Law re nce F. Katz. Introduction In neoclassical economics, wage rates— like the price of any traded com m odity— are deter mined by both supply and demand. Despite the simultaneous nature of the wage-setting process, recent empirical investigations of the determi nants of wages have focused primarily on factors affecting labor supply. Demand factors have been relatively neglected. During the 1940s and 1950s, participation in the administration of wage and price controls led a distinguished group of economists to examine employer wage policies. Reynolds, Segal, Dunlop, Myers, Lester, and Lewis studied interindustry, intra-industry, union, establish ment size, and regional differentials.1 In essence, they focused on variables controlled by employers (that is, and mediumrun labor supply. Dunlop (1957) summarizes many of these effects in his work on wage contours. Research on the influence of supply-side fac tors was stimulated by the development of human capital theory (Becker [1964] and Mincer [1974]), and by the availability of household labor demand) surveys, which gather more information on workers than on their employers. Since the 1960s, labor economists have primarily studied variables controlled by employees (that is, longrun factors) such as age, education, and experience. In the Current Population Survey, a house hold survey, regressions of wages on workers’ characteristics typically produce results similar to those shown in table 1. In this example, the explanatory power of human capital variables is enhanced by exclusion of agricultural workers and of the youngest and oldest workers from the sample. Even within this limited population, the narrowly defined human capital variables explain only a quarter of the variation in the log of wages.2 Addition of occupation raises explan atory power by 16 percent, while race, sex, and union variables add another 6 percent. Industry (broadly defined) raises explanatory power to 51 percent of the variation of wages. What accounts for the 49 percent of wage variation that the equation doesn’t explain? Are there other empirical regularities or theories that labor supply ■ 2 Th e explanatory pow er of hum an capital variables reported 1 is actually relatively high co m p are d to that fou n d in m any in table ■ 1 Segal econom ists. (1986) and Kerr (1983) su m m arize the w ork of these sam ples because of exclusion of yo un ger and older w orkers and of agricultural w orkers. explain the residual variation? And, are we cer tain that industry, unionism, race, and sex reflect differing ability on the part of workers? This study reviews empirical evidence in support of demand-related wage differentials, which suggests that different employers pay T A B L E In none of these cases is the presence of employer differentials inconsistent with profit maximization on the part of employers. How ever, the last two theories predict the existence of queues for high-wage employers while, in the other three models, employer differentials are associated with market clearance. 1 Typical C ross-Sectional W age Regression Results in the I. Em pirical Evidence on Two Types of Interem ployer W age Effe c ts C urrent Population Survey Regressors_______________________________________________R ^ Years of Education, Age, Age Squared .26 Years of Education, Age, Age Squared, Occupation (2-digit) .42 Years of Education, Age, Age Squared, Occupation, Race, Sex, Union .48 Years of Education, Age, Age Squared, Occupation, Race, Sex, Union, Industry (2-digit) .51 Note: D ep en d en t variable was log (h ou rly earnings). Mean: 2.0 5, standard deviation: 0.5 5. N um ber o f observations: 150,579. S ource: Current Population Survey O ne-Q uarter Earnings Sample, 1986. Sample includes all p e o p le em p lo y e d in nonagricultural industries for wages and salaries, aged 18-54. different wages to workers of equal ability. This result is at odds with the predictions of the simple competitive model of wage determina tion. Where does the model fail: in assumptions of worker or job uniformity, or of full informa tion in the labor market? These questions are particularly relevant because employer wage dif ferences may be at the root of the observed unequal earnings between men and women, or among racial or ethnic groups. If these wage differentials arise from efficient, profit-maximiz ing behavior by firms in a second-best world, which policies for reducing inequality are likely to be most effective and efficiency-enhancing? The answers to these questions rest in identi fying the source(s) of employer differentials. This paper discusses five sources of employer wage differentials: 1) employers systematically sort workers by unmeasured ability; 2) wages vary because of unnoted compensating differen tials; 3) costly information generates or perpetu ates random variations in wages; 4) the efficient wage for some employers is above the market rate; and 5) workers inside firms exercise a claim on rents. The focus of this paper is on employer wage differentials: wage differences accruing, on aver age, to all employees at an establishment. For example, in a regression, these are estimated as the coefficients on establishment dummies, con trolling for occupation. These differentials cap ture all differences in labor demand among employers.3 Empirical studies of how wages vary among employers can be divided into two groups: between-industry studies and within-industry studies. Wage variation has been the subject of much scrutiny and specula tion. Far less attention has been paid to wage variation by employers even though wage variation among industries implies variation among employers within industry for two reasons. The first reason is that industry is not uniquely defined. By some criteria, the dif ferences among industries are continuous, rather than categorical. Since no industrial classification system captures all relevant dif ferences among product markets, sources of wage variation among industries should be detectable within industry as well. If explana- among industries within industry, ■ 3 Industrial relations distinguishes between two co m p o n e n ts of w age determ ination in an enterprise: (1) form ation of co m p ensation policy (the periodic adjustm ent of w age and benefit schedules and rules) and (2) the adm inistration of policy (day-to-day decisions about hiring, piece rates, o vertim e, layoffs, discipline, pro m otio ns, etc.). T h is research does not distinguish the im pact of w age schedules from that of personnel adm inistration; it reports total (or net) observed effects. T h u s , the differentials investigated could be the product of differences of policy, adm inistration, or both . F o r exam p le , D u n lo p (1982a) notes: “ ... quite apart from periodic ch an ges in the schedule of w ag es , salaries or benefits, the adm inistration of these elem ents of the schedule and other rules of the w orkplace, from day-to-day, will significantly affect average costs and earnings.” In the discussion that follo w s, the term “ w ages” will be used in terchange ably with "h o u rly earnings” , even th ou g h this blurs an im portant distinction. W ag es are the product of policy and apply only to nonincentive w orkers. Ea rn in g s are the product of adm inistration of policy and apply to all w orkers, including incentive w orkers. tions of wage variation based on characteristics of industries are correct, we should see support ing evidence among establishments within industry. Furthermore, tests within industry for these effects may avoid problems of omitted variables (or of confounding influences). Second, we expect wage variation among employers within industries because even welldefined industries are not homogeneous. Some wage-relevant factors vary greatly within indus try even though they do not vary m uch among industries. Size of establishment is a good exam ple. Because of this, explanations of wage levels based on industry aggregate data understate the economy-wide importance of factors that vary primarily within industry. In short, the forces that generate betweenindustry wage variation should also operate within industry. Since looking only among industries to understand employer variation may be misleading because of omitted variable or 1 T A B L E aggregation biases, a full understanding of the association between employer and wages requires study of both inter- and intra-industry wage variation by employer.4 A . Betw een-lnuustry Differentials Table 2 summarizes a selection of the literature on wage differentials among industries. The studies document the existence, persistence and some of the characteristics of industry wage differentials. They also propose and test models for industry differential formation. The two Dickens and Katz studies provide the most recent and exhaustive investigations of the char acteristics of industry wage differentials. They conclude that wage differences among indus tries account for 7 to 30 percent of wage variation among individuals. 2 Sam ple of Em pirical Studies of Industry W age Effe c ts Authors and Year 1. Slichter (1950) Data Survey of laborers in Cleveland; National Industrial Conference survey of wages of skilled and unskilled workers Relevant Conclusions Industry differentials are consistent across skill levels, increase with propor tion male, vary positively with value added, decrease with labor intensity, vary positively with post-tax corporate income, and are fairly stable over time. 2. Garbarino (1950) BLS productivity and labor cost data for entire industries Productivity and concentration are positively correlated with pay changes across industries. 3. Reynolds and Taft (1956) Published data for four industries (three unionized) in the U.S., and various European countries and Canada Wages vary considerably between plants (within industry, region, and occupa tion), depending on the competitive position and wage policies of employers. Unionism decreases these variations, but substantial industry, geographical, and occupation differentials persist. 4. Weiss (1966) I960 U.S. Census merged with information on industry concentration, average establish ment size, and unionization Earnings increase with concentration, but inclusion of personal characteristics and weeks worked diminishes and often eliminates the effect. 5. Rosen (1969) I960 U.S. Census-industry aggregates standardized for occupation In a two-stage least-squares model, size of establishment influences demand price for labor, but not supply price. 6. Wachter (1970) “Employment and Earnings” aggregate industry statistics Coefficient of variation of industry aver age wages (unadjusted for occupational composition) increases with unemploy ment and cost of living. High-wage indus tries increase wages first and allow them to fall last. Sam ple of Em pirical Studies of Industry W age Effe c ts Authors and Year 7. Waehtel and Betsey (1972) Survey of Consumer Finances (1967), Institute for Social Research sample of full time, full-year service and production workers Residuals of human capital wage regres sions (with age, sex, race, job tenure, education, and marital status) are highly correlated with industry-occupation, union status, city size, and region dum mies. Conclude that these structural (demand-side) variables, especially indus try-occupation, are important determi nants of wages because of rigidities in the labor market. 8. Dalton and Ford (1977) 1970 U.S. Census sample Industry earnings increase with con centration up to a ratio of 0.5, after which they are stable. Sex and race differentials are large and significant for high concentration industries, while industry growth rate affects wages only in the more competitive industries. Regional differentials were significant but had changed since I960. 9. Pugel (1980) IRS profits by 3-digit industry, merged with industry average demographic and market data Workers receive 7 percent to 14 percent of total excess profits: some of which buys higher skills, the rest of which is rent. 10. Krueger and Sum mers (1986a,b) CPS, May 1974, 1979 and 1984; Quality of Employment Survey 1977 Industry wage differentials do not disap pear when controlling for measured or unmeasured differences in human cap ital or for compensating differentials. Consistent with efficiency-wage models, lower turnover and better performance are apparently characteristic of highwage industries. 11. Dickens and Katz (1986, 1987) Current Population Surveys respondents for 1983 ■ 4 all nonunion A further exam ple of the co m p lexity of the subject is that this discussion assum e s that m ost establishm ents operate within a single industry and their w ages reflect the patterns of the in dustry alone. This is a sim plification that abstracts from v e ry real e xam ple s. F o r instance, drug shelf stockers in superm arkets are paid the low w ages co m m o n to drug stores rather than higher superm arket rates. In these cases, even the establishm ent is too high a level of aggregation. Divided workers into 12 occupational categories, calculated industry wage differentials in raw data, fixed effects equations (with human capital) and from residuals of human capital equations. Found that industry differentials are large, persistent, and correlated across occupations and countries. They are also correlated with industry characteristics: percent male, average education, quit rates, and measures of product market power and profitability. Conclude that simple competitive models are not con sistent with observed patterns. While evidence on the source(s) of the differ entials remains inconclusive, a strong link between industry differentials and industrial concentration (or profit rates) is found in all studies that search for it (Slichter, Garbarino, Reynolds and Taft, Dalton and Ford, Pugel, and Dickens and Katz), except Weiss. Krueger and Summers find links between differentials and the predictions of efficiency wage models (lower turnover and higher effort). B . W ith in-ln du stry D ifferentials Table 3 summarizes a selection of the empirical literature that provides evidence of the existence of large wage differentials among firms and among plants.5 The first studies are case studies, where many of the issues explored singly below are investigated for a single labor market. The first two studies are particularly valuable because they use data with unusually rich information on both worker and firm characteristics. Both stud ies find significant differentials among firms. Reynolds concludes that firms select the general wage level on which they operate until forced to change. Rees and Schultz estimate the individual and establishment effect on wages for four groups of occupations and find systematic dif ferences among firms that are not consistent across all occupations. Mackay, et al., Nolan and Brown, and Brown, et al. are fairly recent case studies of English and Australian labor markets. They find that wage variation by plant is a large and fundamental component of wage dispersion, and that employer wage differences are persistent over time and are linked to plant performance. Like the English and Australian studies, Groshen (1988a) focuses on the entire employer differential within industry rather than on the portion associated with a particular charac teristic. She finds that a random switch in employer, within detailed occupational category and industry, is associated with an expected wage change of 12 percent. She also finds that employer size, gender composition, and indus try sector are associated with wage level. How ever, it is unlikely that measures of human capital such as experience, tenure, or education explain the observed establishment differentials. Groshen (1988b) finds that these interemployer wage differences are virtually stationary over six years and present within a single metropolitan statistical area. Hodson matches U.S. household survey data with employer information and finds employer characteristics to be strongly signifi cant predictors of wages. Investigations of employer size and gender composition wage differentials, such as those listed in table 3, are a dimension of the work on employer differentials because they select one aspect of establishment differentials for examina- ■ 5 F o r a survey of the literature and the em pirical problem s associated with m easuring a related issue, the relationship between com pensation and firm perform ance, see Ehrenb erg and M ilkovich (1987). tion. The explanations for these phenomena must also come from the theories explored below. The worker-quality differential studies, by Evans and Conant, are of interest because they argue against sorting by ability or human capital. Finally, the last two intra-establishment stud ies suggest that although interoccupational dif ferentials are compressed within establishments, they do have the same patterns. Thus, establish ment effects are fairly, but not exactly, uniform across occupations. In summary, these studies provide strong evidence that within-occupation interemployer differentials exist, and that they are associated with measurable attributes of employers, such as firm or plant size. II. Sources of W age Differentials Am ong Em ployers This section summarizes five models that explain why an employer might pay a wage premium to all of its employees rather than to particular individuals. These theories are based on the rigorous models of particular economic relationships that have been developed since the 1960s. Virtually all of the ideas in the following discussion can be found in the work of earlier economists, but were later formalized by, and are here referenced to, other authors. A . The Role of Em ployers in the Basic M odel of W age Determ ination The point of departure for the models of em ployer wage effects listed below is basic Mar shallian supply and demand. I begin by noting that in a perfectly competitive labor market with costless contracting and information, and with identical workers and jobs, no differentials based on differences in labor demand would arise. Market labor supply is a function of leisure preferences, population supply, and training costs. Market labor demand is the horizontal sum of all employers’ demand curves, that is, the marginal revenue product of hours worked. Under perfect competition in capital and labor markets, equivalent workers at equivalent jobs earn the same wage. An employer whose wages stray from the market rate will be forced out of business by loss of employees (wages set too □ T A B L E 3 Sam ple of Em pirical Studies o f W ith in-ln du stry Em ployer W age Effe c ts Authors and Year Data CASE STUDIES AND MORE GENERAL STUDIES OF INTEREMPLOYER Case study of an urban blue-collar labor 1. Reynolds (1951) market based on worker interviews and data published by other sources DIFFERENTIALS Plant wage-level depends on industry, unusual efficiency of plant or manage ment, secure monopoly or oligopoly control of product market, and history of relative wages. Most wage move ments occur uniformly within clusters of firms. Plants operate within a range of feasible wage rates, but movement within the band is difficult. 2. Rees and Schultz (1970) Personnel records from 75 Chicago establish ments on 13 occupations, white- and bluecollar, skilled and unskilled; interviews with management personnel and workers Industry differentials vary in size and sign across occupations, and are smaller for skilled workers. No positive relation ship between establishment size and wages, within occupation, industry, location, and controlling for work char acteristics. Location differentials are uniform across occupation. 3. Mackay, et al. (1971) Mean earnings and quit rates by occupation from personnel records for blue-collar work ers in 66 engineering plants in Birmingham and Glasgow from 1959 to 1966. W ithin occupation, inter-plant coeffi cients of variation ranged from 16 percent to 23 percent and rank order correlations (from 1959 to 1966) were about 0.9, except for laborers. Wages were negatively correlated with quits, but unrelated to changes in plant size. Investigations of causes led to rejection of sorting by human capital, of random variations, and of working conditions. Concluded that efficiency wages for quit rates and profit-sharing were most likely sources. 4. Hodson (1983) Wisconsin 1975 survey of high-school gradu Corporate structure variables (size, ates from 1957, matched with employer international links, capital intensity) information strongly affect wages. Product market variables (profits, productivity) have little impact. 5. Nolan and Brown (1983) 10-year survey of wage structure for seven occupations in 25 factories in West Midlands, England Employer effects on wage changes dom inate occupation effects. Nevertheless, rankings by employer are relatively stable across occupation over 10 years; rank correlations of 0.8 to 0.9. 6. Brown, et al. (1984) Survey of 44 occupations in 198 plants in Adelaide, Australia Overawards to Australian workers tend to be tied to establishment rather than to occupation. Industrial concentration is highly correlated with size of overawards. BLS Industry Wage Surveys of production workers’ wages in six manufacturing industries W ithin detailed job classification, wage variation between establishments accounts for 30-60 percent of wage variation, generating a standard deviation of 11 percent. Half of the differentials were associated with characteristics of the establishments (size, union affiliation, etc.). T A B L E 3 Sam ple of Em pirical Studies of W ith in-ln du stry Em ployer W age Effe c ts Authors and Year 8. Groshen (1988b) Data BLS Area Wage Surveys of nonsupervisory workers’ wages (blue-collar and white-collar) in one SMSA for six years W ithin detailed job classification, wage variation between establishments accounts for 20-70 percent of wage variation, generating a standard devia tion of 12 percent. Differentials were unchanged over six years and not associated with growth or shrinkage. WORKER QUALITY DIFFERENTIALS, WITHIN OCCUPATION, BETWEEN ESTABLISHMENTS Across establishments, the strongest Private area wage and salary survey of 1. Evans (I960) observed relationship was between clerical workers in Boston wages and length of service. Test scores and education are inconsistent predictors of wages. 2. Conant(1963) Placement test scores and beginning salaries for typists in Madison, W I Test scores accounted for only 10 per cent of the variation in starting wages offered by different employers to entry level typists. ESTABLISHMENT AND FIRM SIZE DIFFERENTIALS BLS Establishment Surveys— Wages and Hour Hourly earnings are higher in large 1. Perlman (1940) firms, within industry, occupation, Statistics for six industries product group, and region. Hourly earnings are not affected by establish ment size, holding region constant. 2. Lester (1967) BLS Industry Wage Survey and Census of Manufactures Except for textiles, apparel and aircraft, earnings increase with establishment size. Differentials increase when fringe benefits are included. 3. Masters (1969) BLS Census of Manufactures Plant size variable is a stronger (larger and more significant) determinant of average wage differences among indus tries than concentration. 4. Buckley (1979) BLS Area Wage Surveys for 29 areas Controlling for industry mix, wages rise with area cost of living, but not with establishment size. 5. Miller (1981) BLS Census of Manufactures Controlling for industry, wages increase with size of establishment. 6. Personick and Barsky BLS National Survey of Professional, (1982) Technical, and Clerical Pay 1980 Pay levels tend to increase with em ployer size, but above-average levels are associated only with large firms. Wage premia attributable to a firm’s size are larger for entry-level than for experi enced professional workers. Corporate size has better explanatory power for professionals while establishment size does better for clerical workers. 7. Mellow (1982) Both plant size and firm size are positively associated with wages, con trolling for personal characteristics and concentration. The effect is propor tionately larger when fringe benefits are included. Industry-plant size inter action variables were insignificant. Current Population Survey 1979 □ T A B L E 3 A Sam ple of Em pirical Studies of W ithin-lndustry Establish m e n t W age Effe c ts Authors and Year 8. D unn (1980, 1984) Data Independent surveys of employee wages, working conditions, and employer size within one industry 9. Brown and Medoff (1987) Variety of public sources Firm and plant size are associated with higher wages, controlling for occupa tion, industry, and working conditions. Differentials are smaller for highergrade occupations. MALE/FEMALE COMPOSITION OF OCCUPATIONS WITHIN FIRMS 1. Blau (1977) BLS Area Wage Surveys INTRA-ESTABLISHMENT OCCUPATIONAL DIFFERENTIALS 1. Ward (1980) BLS Area Wage Surveys 2. Van Giezen (1982) BLS Area and Industry Wage Surveys low) or the loss of capital (wages set too high). The position that employers are price-takers is the theoretical basis for the current focus on labor supply as the only relevant determinant of wages. The employer in a competitive labor market faces a horizontal labor supply curve, as shown in figure 1. In the figure, Employer 1 has labor demand curve D ,, which differs from the labor demand curve of Employer 2 (labeled D 2). However, because they face a flat labor supply curve (Ls), the differences between the two employers affect only their employment levels (E[ versus E2), not their relative wages. Thus, the simple competitive model generates an empirically testable prediction: variations in labor demand should affect only quantity demanded, not wage level. This is true so long as demand differences do not affect worker utility. The empirical work summarized above sug gests that this simple model does not hold. Wages do vary among employers. In order to Large firms pay higher wages and shift premia than small firms, except in the highest-paid occupations. Compensat ing differentials do not appear to be the cause; infers the presence of bargaining. W ithin occupation, establishments tend to be segregated by sex; pay rates are negatively associated with percent age of establishment female. Occupa tional segregation by sex is associated with industry. National occupational wage spreads do not exactly mirror individual firms; pay differentials are smaller within establishments. Area occupational differentials are larger than intra-firm differentials. Intra-firm differentials vary by industry and region, and decrease with estab lishment size, although differences are small. extend the simple model to allow for apparent demand-side effects, any explanation of wage variation by employer must answer two crucial questions: (1) why would one employer choose to pay more than another, and (2) why don’t high-wage employers go out of business? The answer to the first question is usually that a firm paying higher wages employs more productive workers. The advantage of the pro ductivity explanation is that it also answers the second question. The disadvantage is that pro ductivity differentials are usually due to individ uals’ abilities, not to employers’ characteristics, implying the need for more explanation. If prod uctivity differentials are not invoked, costly information or imperfect competition in the pro duct market must be present and, again, operate similarly on all individuals in an establishment. □ F I G U R E 1 The La b o r Supply Curve Facing an Individual Firm is Assum ed to be Infinitely Ela stic 1 . Sorting by A b ility: Innate Differences, Hum an C ap ital, and M atching The first two explanations relax the assumption of uniformity among workers or jobs in the market. Since the labor market is perfectly com petitive, workers earn the marginal product of their work and employers pay equivalent wages per efficiency-unit of work. However, hourly wages may mismeasure either the workers’ units of work (because this varies among workers) or their compensation (because it omits nonpecuniary returns to employment). In order to generate differentials rather than just differentials, the theories must also explain why the marginal product of work ers varies among employers. Sorting models assume that some workers are more productive than others, and employers consistently hire their workers from a single quality stratum, regardless of occupation. The source of quality difference may be innate advantages (for example, genetic or moti vational), or acquired differences (for example, education or work experience). Each establish ment hires only the best, or only the worst, workers of each job category. , it is not obvious why an establish ment would need or choose to segregate by ability. If all workers were paid their marginal products, the number of workers paid to pro duce a certain product should be irrelevant. For example, employers should be indifferent between two equally productive workers at one wage and a single doubly-productive worker at twice the wage. Any establishment could have a distribution of productivity levels (all rewarded accordingly) within each occupation. In this sort of world, no apparent establishment differentials would arise. In order for innate or acquired productivity differences to generate apparent establishment differentials, employers must choose workers of fairly uniform productivity within occupations, and apply this policy similarly to all occupa tions. That is, this theory must be combined with an explanation for segregation by firm. Two questions arise: why and how? The most convincing reason may be that employers’ technologies are differentially sen sitive to a worker’s ability. In this case, employ ees of high ability w ho are not being rewarded for their higher ability by employers with abilityinsensitive technology have an incentive to seek out employers w ho will pay according to ability. This leads to a positive correlation between the establishment individual Source: Author. B . Five M odels of Em ployer D ifferentials Table 4 summarizes five microeconomic sources of wage variation. Each source is developed from the competitive model by the introduction of transaction costs and/or of heterogeneity among agents. The table also lists the basic assumptions beyond those of the competitive model, and the additional assumptions neces sary for the models to predict the existence of apparent wage differentials, rather than differentials among individuals or among occupations. Each of the models examined predicts the existence of wage dispersion, and can be extended to predict employer-based dispersion, though the extensions usually involve extra twists of varying plausibility. Although none of the five models relaxes the assumption of profit maximization on the part of employers, they are arranged in order of their divergence from com petitive theory in other aspects. In particular, the last two models, efficiency wages and bargain ing, require assumptions of imperfections or lack of competition in the product or labor markets because they imply the existence of job rationing or queues for high-wage employers. employer A priori ability-sensitivity of the employer’s technology and the average quality of their applicant pool. Thus, employers with ability-sensitive technolo gies hire disproportionately more high-ability workers and, therefore, pay higher wages.6 For example, establishments requiring tech nical typing are likely to be highly sensitive to the skills of typists. So, we expect such employ ers to reward an excellent technical typist more than would employers w ho needed only text M icroeconom ic Sources of Em ployer W age D iffe re ntia ls Wage Equation1 Model Costly Factor(s) Source(s) of Heterogeneity Additional Assumptions Necessary for Existence of ______Employer Wage Effects______ SORTING BY ABILITY w = MP Human Capital, Innate Differences, Job Matching Training Innate or acquired worker quality, quality of job match Establishments differ systematically by average quality of workers, or match, consistently across all or most occupations. w = MP Improvement of undesirable terms of employment Management strategies or technologies Undesirable terms of employment are uniform across all or most occupations within establishment. w = MP + e Employer and/or Random draws worker search, from the pool, job mobility intertemporal wage variation Employers vary in the average value of their draws, employers hire for all occupations during growth surges. MP = f(w)w* = MP* Monitoring of workers’ effort, turnover, design of internal wage structure, firmspecific training Management strategies or technologies, corporate size Employers adopt similar strategies (or technology has a similar effect) on the efficient wage across all or most occupations, workers in most occupations develop firm-specific training. w = MP + f(7r,workers’ bargaining power) Monitoring of workers and/or of management Varying rents, Rent capture is achieved and/or ability of workers shared by all or most occupations, to capture rents, and/or managerial altruism COMPENSATING DIFFERENTIALS Working Conditions, Fringe Benefits, Risk of Layoff RANDOM VARIATIONS Information, Search, Lagged Adjustment e~f(0,a2) EFFICIENCY WAGES Monitoring, Turnover, Market Insulation, Corporate Consistency, Morale, Loyalty BARGAINING Insider/Outsider, Rent Capture, Gain-Sharing ■ 1 Th e s ym b o ls in this colum n are defined as: w =w age M P = m a r g in a l revenue product e = r a n d o m error te rm , distributed with m ean of 0 and variance of f ( * ) = s o m e function of * w ,* M P * = t h e unique profit-m axim izing values of w and M P tt= profits a2 □ typing. The higher pay for skills will, in turn, attract other typists with technical skills into the applicant pools for employers needing technical typing. In order to create establishment differen tials, this explanation must be expanded by the assumption that ability-sensitivity in establish ments is highly correlated across occupations. Otherwise, wage variation would occur pri marily by occupation within establishment, not by establishment across all occupations. Thus, in the example, the need for technical typing must be associated with ability-sensitivity in other occupations. The second explanation is not mutually exclusive with the first and could provide a rationale for the correlation in ability-sensitivity across occupations. This model assumes that variation in the quality of workers in an estab lishment imposes negative externalities on the productivity of more able workers. Envision establishments as assembly lines where work stations are indivisible, or where the timing of the output depends on the speed of the slowest operative. Then, the productivity of the slowest worker determines the productivity of all the workers. As workers seek their best-paying job, establishments become segregated by quality.7 Employers maximize profits by hiring or retain ing (through their recruitment and termination policies) only those workers at least as able as those in their existing work force. Job matching provides another approach within the sorting models (Jovanovic [1979]). Here, both worker and employer are unin formed about the worker’s productivity in a particular job, until both have experienced it. The productivity of a worker-job combination is random, with a distribution known to both sides. Workers accept jobs that pay more than their current jobs. Employers offer wages based on the mean of the distribution, and later adjust wages to reflect measured productivity. Accu racy of productivity measurement improves as tenure increases. Employees with bad matches eventually leave in hope of finding a better match elsewhere. Then differences in the dis tribution of productivity across employers could lead to sorting.8 Other explanations for sorting come from the sociology literature on the joint productivity of teams as a product of the uniformity of team members. In all versions, all employers (whether high- or low-wage) earn zero or equal profits in equilibrium. But, high-wage/high-productivity employers are not associated with higher or lower profit levels than their low-wage/ lowproductivity competitors. Only consistency matters. The human capital model, formalized by Becker (1964) and Mincer (1974), provides a rationale for the variance of wages according to acquired training. Training increases productiv ity, raising the demand curve for hours of trained persons’ time over that for untrained people. However, the costs of training, such as forgone wages and tuition, raise the supply curve for trained persons’ time. Thus, the price of trained labor is higher than that of untrained labor and reflects the difference in marginal product between the two. If human capital differences are manifested as employer differentials, employers must be able to predict productivity on the basis of acquired training (education and seniority), and both hire and pay workers accordingly. High-wage employ ers are such because they select the most highly trained workers in each occupational category. Low-wage employers hire (or end up with) work ers with the least training across the board.9 Innate differences in productivity (for exam ple, due to perseverance, or motivation) are less amenable to measurement by all parties, and are not included in the data bases generally available to economists. As such, they can only be investi gated indirectly However, if these innate qualities ■ 8 Fo r instance, s u p po se that all jobs had the sam e expected ■ 6 M odels of self-selection and sectoral choice w here the sectors productivity, but those offered by certain em ployers had a higher va ry in returns to ability in a com petitive labor m arket were introduced in variance. In this case, the high-variance em ployers m ight tend to have a R o y (1951). A m ore recent treatm ent appears in La n g and D icken s (1987). h igh-w age, m ore-productive w ork force. This w ould happen because the w orkers with the go o d draw s w ould stay longer and the w orkers with the ■7 W hen an em ployer pays w ages that reflect actual marginal product, w orkers will be paid the marginal product of the least-productive w orst draw s w ould leave m ore quickly than they w ould in a firm with less variance. worker, rather than according to their ow n individual abilities. W orkers with higher potential will leave for jobs with a m ore productive “ weakest ■9 link” , causing average potential productivity to decline toward that of the particular form of acquired h um an capital: w ork experience. High-w age O n e explanation for sorting by establishm ent applies only to a least-productive w orker. Em ploye rs w ho pay w orkers according to their establishm ents m ay be older and have a relatively old, experienced w ork potential m arginal product will keep their w orkers, but lose m oney. This fo rc e , com p ared to the younger, less-productive w orkers in low-wage argum en t is similar to the “ Jo b s as D a m Sites” idea introduced in Akerlof plants. If s o , differences in age of em ployer w ould be reflected in w ages, (1981). although w age per efficiency-unit of w ork is identical for all em ployers. are correlated with the usual measures of acquired human capital such as age and experi ence, then controls for measures of human capital also control for innate differences.10 Conant (1963), Evans (I960), and Groshen (1988a) all suggest that employer wage dif ferences are not associated with sorting by measured human capital or by ability correlated with human capital. Gibbons and Katz (1987) suggest that the unmeasured ability explanation also faces a number of empirical problems in addition to high correlation in employer differ entials across occupations. One problem is the lower quit rates in high-wage firms and indus tries, which suggests that the high-wage jobs may be rationed, unless high ability has a partic ularly strong association with a tendency for employment stability. Another problem is that workers displaced from high-wage industries do not appear to retain their wage differentials if they switch industries. Finally, the correlation of employer wage differentials with product market power is difficult to explain within this model. 2 . Com pensating D ifferentials The second possibility is compensating differen tials, described by Adam Smith (1776), refined by other economists since then, and summa rized in Smith (1979). The essential problem is mismeasurement of the total return to working. In the case of poor working conditions, mone tary wage overstates the returns to individuals for their work because it ignores the extra costs imposed by working conditions. Working conditions vary among employers, and it is costly to improve them. All else equal, workers prefer jobs with safe or pleasant work ing conditions to those with poor conditions. Thus, employers providing unfavorable condi tions will be unable to meet their labor demand at the going wage. In response, the firms offer ing undesirable jobs must improve the working conditions or raise wages, whichever costs less. If improvement of conditions is costly, wages will be higher in order to attract sufficient labor, but the profitability of each hour worked is higher because of money saved during each hour worked under poor conditions. If workers were identical, the wage differen tial between any two jobs would ensure that workers were indifferent between the two. If workers varied in their tastes, the differential would depend on the tastes of the marginal worker. The allocation of the work force among poor and good jobs depends on the assump tions made about existing production technolo gies. Technology is usually assumed exogenous, so we need a random distribution of differences in costs of improving conditions. If technology is not exogenous, all firms will choose the one that maximizes profits, so only those combina tions of technologies and compensating differen tials that yield the maximum profits will coexist. In all versions of this model, employer (rather than individual) differentials arise only when quality of working conditions is consistent across all or most of the work force in establish ments.11 Many working conditions, such as physical exertion, do not apply because they are occupation-specific. However, high risk of layoff, poor ventilation, minimal fringe benefits, or inconvenient location could presumably affect all or most workers in an establishment. Then, the costs of improvement of these conditions must vary enough among employers to generate the large and persistent differentials. Empirical studies of compensating differen tials have been notably unsuccessful in finding evidence of their contribution to wage disper sion.12 One exception to this generalization is Eberts and Stone (1985), w ho find evidence of compensating differentials only after controlling carefully for characteristics of employers, sug gesting that compensating differentials are second-order effects. That is, type of employer determines overall level of compensation, but there is some substitution between wages and nonpecuniary compensation within groups of otherwise similar employers. ■ 11 In addition, two fairly m echanical versions of com p ensating differentials are possible. T h e first is based on different age-earnings profiles with differing average tenure am on g plants. T h e second is variation in tim ing of annual salary adjustm ent. G rosh en (1988a) presents evidence that suggests that neither of these possibilities is likely. ■ 12 F o r exam p le , see S m ith (1979). M o s t studies have attem pted to identify co m pensating differentials am o n g industries, w here conditions va ry m ost am on g em ployers. N everth eless, such inquiries have been m arked by their lack of success. Fo r w orking co nditions, see Brow n (1980); for layoff risk, see Topel (1984). It is also unlikely that em ployer w age differences com p ensate for differences in fringe benefits. Free m a n (1981), Sm ith and Ehrenb erg (1981), and Atrostic (1983) find that inclusion ■ 10 Jo b m arket signalling (S pe nce [19 73]) is an extrem e exam ple of of fringe benefits exaggerates w age differences am on g em plo yers. Tha t this type of correlation, which blurs the distinction betw een hum an is, high-w age em ployers pay even m ore of total com pensation in the form capital and innate differences. of fringe benefits than do low -w age em ployers. 3 . Random Variations Seminal articles by Stigler (1962) and Rothschild and Stiglitz (1976) launched a family of pure information models that use costly job search to explain wage dispersion. Suppose search were expensive for job-seekers. In this case the mar ketplace can sustain a range of wages because the gain from further search becomes uncertain, rather than a known quantity.13 In the typical model, establishments offer wages according to a distribution known to all job-seekers. Workers accept offers that exceed the expected value of further search. Job-rejecters pay to search again. Thus, the only sustaina ble distributions of wages are those where the m inim um wage paid differs from the mean offer by less than the costs of employee search. These models focus on the role of the indi vidual in wage determination. No rationale is offered for variations among employers. A sym metric formulation of the problem from the employers’ point of view posits the existence of a known distribution of reservation wages among a population of potential employees. Employers interview applicants to ascertain their reservation wages, and jobs are offered to workers (at their individual reservation wages) when the expected value of the wage reduction from an additional interview by the employer falls below the employer’s search costs. Em ployer search costs consist mainly of advertising and interview expenses. The employee-cost/employer-distribution model provides no theoretical basis for the existence of employer differentials. Rather, it ex plains only persistence of variance, leaving unan swered the question of why the employers who pay over the mean do not reduce their wages. The converse model, the employer-cost/ employee-distribution model, abstracts from the fact that firms usually set wages for a job rather than for an individual. Indeed, wages are usually attached to jobs before the interviewing proc ess. Exceptions to this rule occur where job responsibilities are not well-defined, such as in very small firms and for highly skilled or very senior employees. In general, two individuals w ho differed only in reservation wage would not be offered different wages at the same plant. Lagged adjustment, a second type of random variations model, is not inconsistent with the information/search models, but provides a basis for the variations (wage shocks) and an addi tional reason for their persistence (internal adjustment costs). These models, coined “geo logical models” by Dunlop (1982a), focus on the employer. Establishments may tend to hire in surges rather than in steady flows. If the costs of redesigning an internal wage structure are high or if workers are immobile, a firm’s internal pattern and general level of wages will reflect the market wage pattern of its most recent expansion.14 In the random variation models, wages approximate the worker’s marginal product, but costs of information introduce an error term. The mean of the error term is zero, and its variance is a positive function of the search and mobility costs for one or (perhaps) both parties. Consequently, establishment differentials result from random variations in the average error terms of employers. But, if establishment differ entials are large, long-lived, and associated sys tematically with characteristics of employers— as suggested by the empirical work cited above— they are not random variations. 4 . Effic ie n c y W ages Efficiency wage arguments posit a causal rela tionship between the wage level and a worker’s on-the-job productivity.15 Efficiency wage employers maximize profits by paying workers a premium above the market-clearing wage, because the resulting increment in productivity yields the highest profits. The increased produc tivity has been modeled as coming from three ■ 14 Fo r exam ple , establishm ents m ay grow by the addition of a second or third shift, rather than by hiring a few new w orkers each m o n th . W ages at the tim e of a hiring surge reflect current labor-m arket conditions. If the m arket is tight, w ages paid to attract new em ployees will be relatively h igh. Later, w hen m arket w ages fall, adjustm ent dow n to the new m arket-clearing level will not be im m ediate. Redesigning the internal w age structure im poses costs (out-of-pocket and m orale) on the em ployer. W age schedules are rarely adjusted m ore often than annually and are rarely adjusted dow n w ard nom inally. U p w a rd adjustm ents will be slow if w orkers face m obility costs. T h u s , the internal pattern and general ■ 13 Originally, the inform ation m odels were form ulated to explain the existence of price or w age dispersion. S u b s e q u e n t w ork uses these ideas level of w ages at any particular tim e reflects the m arket w age pattern of the e m p lo yer’s m ost recent exp an s io n . (H e n c e , the term “ geological.” ) to predict the level of un em plo ym ent. F o r exam p le , see A za ria dis (1983). Since the focu s of the current w ork is w age dispersion, the earlier ■ 15 T h e main versions of these m odels are s u m m arize d in Yellen form ulations of Stigler will be used to characterize the results of this (1984) and Stiglitz (1984). Effic ien cy w ages were originally form ulated as diverse literature. Later versions of these m odels generate term inal w age an explanation for equilibrium u n em p lo ym en t, rather than for w age distributions from initially assum ed distributions. Stiglitz (19 79 ) and dispersion. W ages do not fall to clear the m arket because firm s m axim ize Venables (1983) provide exam ples of these m odels. profits in a labor m arket w here w ag es are high and jobs are rationed. sources: reduced monitoring (or shirking) costs, decreased turnover, and sociological considera tions. The internal labor market literature adds two more possibilities: market insulation and corporate consistency. In the monitoring/shirking version, workers’ effort is costly to monitor (Bulow and Summers [1986], Shapiro and Stiglitz [1984]). An increase in wages decreases a worker’s incentive to shirk, because shirking increases the probability of losing a high-wage job. In comparison to an em ployer paying the equilibrium wage, efficiency wage employers pay higher wages, experience higher worker productivity, and have lower direct monitoring expenses. The turnover version emphasizes employer costs of hiring and training (Salop [1979]). Wages above equilibrium reduce turnover because workers have fewer superior alternatives and/or because the general level of unemployment rises. Thus, workers paid higher wages have longer tenure. Two related search/recruiting ver sions of the model show that firms with high costs of unfilled vacancies will offer high wages to more quickly fill vacancies (Lang [1987] and Montgomery [1987]).16 A third variant of the argument is based on sociological morale, loyalty, or teamwork effects. Group work norms are raised by wages above the m inim um required. Akerlof (1982) terms this the “partial gift exchange” model. The two internal labor-market variants, as described by Doeringer and Piore (1971), focus on the out-of-pocket and morale costs of design ing a compensation package for a group of employees, and on firm-specific human capital. If all wages are to be set constantly at marketclearing levels, shocks to the external labor market will necessitate periodic readjustments of internal pay relationships. Yet, redesign of wage schedules may be expensive for certain types of employers, especially large ones, or for certain groups of employees, such as incentive workers. In addition, any change in relative wage relationships may be perceived as inequita ble or as a breach of implicit contract. Such dissatisfaction could reduce productivity through increased shirking or turnover. An alternative to frequent, disruptive adjust ments in response to market fluctuations is to ■ 16 L a n g (19 8 7) extends the analysis to s ho w that an equilibrium distribution of w ages can be sustained a m o n g otherwise-identical firm s, but there is no reason to expect firm s’ positions in the distribution to persist, unless firm s lock in their position by their choice of technology. This assum e s the existence of a range of tech no logies, each with different capital-intensity (a nd , th u s, cost of unfilled vacancy). set wages above the market level. If, on average, workers receive a premium, then wage shocks that are small relative to the premium will not force a firm to readjust its compensation pack age. Employers save out-of-pocket and produc tivity costs of the adjustment, in return for paying higher wages. Corporate consistency, the second internal labor-market version, is based on the tendency of large firms to promote workers from within whenever possible rather than hire from outside. Presumably, firm-specific human capital makes promotions or transfers among plants efficient. Nevertheless, such a policy requires that internal wages for each occupation in each plant meet two criteria: (1) they cannot be much lower than local wages for the occupation (or the workers will leave the firm), and (2) they cannot be lower than firm-wide wages for that occupation (or workers will refuse transfers to the plant). This implies identical wage structures for each plant within the firm regardless of location, as long as product lines are similar enough for personnel to be transferred among them. Furthermore, each occupation will earn the maximum local rate over all plant locations. O n average, this yields positive establishment differentials that increase with firm size. Efficiency-wage models can be invoked to explain differentials among firms in two ways. First, the profit-maximizing point is, almost by definition, locally flat. This implies the existence of a plane of (almost) iso-profit wage-productivity points for identical firms. That is, variations in wages from the optim um lead to only small profit losses. Firms are close to indifferent among the possible combinations, so a random distribution of strategies results (Bulow and Summers [1986]). A second, more plausible, explanation stems from economically important heterogeneity among employers: differences in technology (vintage effects, for example), or differences in products (such as differentiated quality niches). The productivity of workers at the marketclearing wage may be indistinguishable from high-productivity work under some technolo gies, or may be adequate for one market but not for another. Workers paid the market-clearing wage form a queue for jobs at the elevated wage, while recipients of the high wage avoid job loss or job changes because of the scarcity of equiv alent opportunities. Efficiency differentials can explain establish ment differentials when workers in all or most occupations in the establishment are affected. That is, it is crucial that the heterogeneity among employers affect the efficient wage for all occupations similarly. The plausibility of this assumption depends on the version of the model in question. Few empirical tests of efficiency wage models have been performed, primarily because of the lack of appropriate data. One recent exception, Leonard (1987), finds little evidence to support the turnover or supervisory-intensity versions among electronics companies in California. Another study, Krueger and Summers (1986a) finds some support for efficiency wage explana tions of interindustry wage differentials. Interest in these models suggests that the results of other tests may be available shortly. 5 . Insider/Outsider Bargaining M odels When bargaining between workers and their employers takes place in the context of com petitive markets (in labor, capital, and products), bargained wages cannot differ from the marketclearing wage. Otherwise, the firm would close or the workers would leave. However, if employ ees can exercise a claim on the rents generated by an enterprise, they will bargain (implicitly or explicitly) with their employers. Wage settle ments will reflect both the size of rents and the relative bargaining power of the parties. Thus, the existence of both rents to the firm and employee bargaining power are necessary con ditions for wage bargaining to produce wage variation. Although all versions of bargaining models must assume the existence of rents, the models differ in the identity of the bargaining agents and in the enforcement mechanisms for the bargain ing. The bargaining agent for the workers is most clear in the case of unionism. In the collective bargaining literature, the outcome of negotiation is likened to the Edgeworth Box. Bargaining is a positive-sum game until the contract curve is reached, and a zero-sum game along the contract curve. The outcome is deter mined by the relative bargaining ability and credibility of participants’ threats. The range of possible wages is bounded by the market-clear ing wage on the bottom end and by the worker’s actual marginal product (with labor appropriat ing all rents and capital earning the normal rate of return) on the high end. In a nonunion setting, the bargaining agent for the workers is not obvious. However, econo mists have long noted the existence of informal organization by workers in nonunion settings (Dunlop [1957]). One version is the union-threat effect, where the threat of unionization forces owners to provide nonunion workers benefits similar to those they would receive if unionized (Dickens [1986]). In a second version, the managerial-capitalism or agency-cost version, managers act as mediators between labor and the owners of capital. If the rewards to management are not highly correlated with rents to the owners, or if managers maximize a utility function dependent on worker satisfaction (whether due to manage rial altruism or to the ability of workers to impose on-the-job problems), then management may not act to maximize rents to owners. Implicit bargaining may occur, with manage ment cast in various roles from agent for the workers, to mediator between the two sets of interests, to agent for the owners. The latter role would generate a model all but institutionally indistinguishable from a union bargaining model. For example, Aoki (1984) presents coop erative bargaining models for modern nonunion corporate enterprises with various constituen cies. Edwards (1979) also presents an informal model of nonunion bargaining. Bargaining models easily lend themselves to the prediction of establishment differentials. The only additional assumption necessary is one that binds together workers of different occupations in the establishment. Three possibilities exist. First, workers’ bargaining power may be consis tent across occupations in an establishment. Second, perhaps workers must form large groups in order to exert bargaining power. Third, managerial altruism may extend uni formly across occupations. The persistent link between measures of product-market power and industry wage differ entials provides an empirical basis for further investigation of bargaining theories. More direct evidence is limited by the lack of data, but studies by Abowd (1985) on unionized firms and by Kleiner and Boullion (1987) on both union and nonunion firms provide some support for bargaining hypotheses.17 As with efficiency wage models, more direct tests of these models are certain to be available in the near future. ■ 17 A b o w d (1985) finds evidence that union contract settlem ents dim inish the value of the firm by exactly the ch an ge in the value of the negotiated settlem ent. Kleiner and Boullion (19 8 7) find that firm s’ w ages are strongly positively correlated with the provision of sensitive financial inform ation to em ployees. III. La b o r M a rke t Policy and Em ployer W age Effe c ts The empirical work cited in this paper suggests that employer wage differentials are large. Thus, they may account for many of the observed inequalities in the labor market, such as those among races or between men and women. Exploration of five models of employer differen tials clarifies the point that these differentials are not necessarily inconsistent with profit max imization by firms acting in a competitive labor market. Yet each model suggests the existence of a particular barrier that prevents formation of a single market wage. The link between theories of employer wage effects and labor market policy to reduce income inequality is labor-market segmenta tion.18 W hen labor markets are segmented, workers are separated into distinct markets by institutional barriers that prevent workers or employers from switching between markets. Thus, different wages persist for each sector of the labor market. Although workers in each sector are paid their marginal product, produc tivity varies between sectors according to sectorspecific supply and demand, or sector-specific quality. Obviously, the costs of barrier removal must be high enough to prevent profit-seeking employers from eroding the differences between sectors. Employer differentials will create segmented markets only if employers limit their recruit ment to one sector, so any model must explain why employers hire all (or most) of their employ ees from the same market sector. Each model discussed above introduces a barrier that could create segmentation, with strikingly different policy implications. Thus, it is precisely the identification of the source of the barrier that makes segmentation difficult to cure with policy. For example, under the sorting model, seg mentation will arise if workers of different sex or race have different access to human capital. The model implies a need for the development of human capital among secondary sector workers (for example, lower cost, better education, or job training). Alternatively, compensating differ entials imply no role for policy, since the market actually remunerates all workers equally. Appar ent segmentation arises simply because tastes differ systematically among groups.19 Random variations suggests that search costs are higher for the classes of workers in predominantly lowwage jobs. A possible solution may be expansion of job-service agencies targeted to these groups. Efficiency wages and bargaining imply the existence of queues of workers for high-wage jobs. Thus, any attempt to reduce inequality should rest on regulation of employers’ recruit ment policies, on improvement of placement services for secondary market workers, and on elimination of any minor productivity deficien cies among workers in the secondary sector.20 These five theories of wage determination also diverge from each other in their predictions for the impact of other kinds of policy. For example, Stiglitz (1984) and Bulow and Summers (1986) analyze the effects of efficiency wages on macroeconomic performance and trade policy. Weitzman (1986) offers an analysis of the effects of a particular form of profit-sharing on eco nomic stability and growth. Understanding the source of employer differ entials is clearly important for understanding the distribution of wages, and for formulating policy to affect it. New sources of data must be devel oped to allow research on employer activities such as supervision, recruitment, terminations, and wage-setting. Without further research on these topics, we will remain unable to sort out whether employer wage differentials are signs of inefficiency, of discrimination, or of other mar ket imperfections. ■ 19 F o r instance, co m p are d to m e n , w om en m ay prefer quieter, cleaner, or m ore flexible jobs (Filer [1983]). ■ 20 B ulow and S u m m e rs (1986) dem onstrate h ow efficiency w ages m ay be a source of m arket segm entation. T h e y em ph asize that segm entation requires the existence of a small productivity differential ■ 18 F o r a s u m m a ry of the literature on seg m en ta tion , see Cain (19 76 ) between w orkers of the two sectors, but that the w age difference between and D ickens and La n g (1985). L a n g and D icken s (19 8 7) provide a detailed the two sectors will be far greater than the productivity difference. A investigation of the relationship betw een the literature on segm ented similar argum ent can be m ade for differentials associated with rent- m arkets and neoclassical econom ic theory. s haring, assum ing profit m axim ization on the part of em ployers. REFER EN C ES Abowd, John M. “Collective Bargaining and the Division of Value of the Enterprise.” Cornell University. Unpublished paper, October 1985. Akerlof, George A. “Gift Exchange and Effi ciency Wage Theory: Four Views.” 74 (May 1984): 79-83. American Economic Review. ________ . “Labor Markets as Partial Gift Exchange.” 97 (November 1982): 543-569. Quarterly Journal of Economics. Review of Economic Studies. Aoki, Masahiko. The Cooperative Game Theory of the Firm. New York: Oxford University ________ . “Jobs As Dam Sites.” 48 (January 1981). Press, 1984. Conant, Eaton H. “Worker Efficiency and Wage Differentials in a Clerical Labor Market.” 16 (April 1963): 428-433. Industrial and Labor Relations Review. Dalton, James A., and E. J. Ford, Jr. “Concentra tion and Labor Earnings in Manufacturing and Utilities.” 31 (October 1977): 45-60. Review. Industrial and Labor Relations Dickens, W illiam T. “Wages, Employment and the Threat of Collective Action by Workers.” National Bureau of Economic Research Work ing Paper No. 1856, March 1986. ________ , and Lawrence F. Katz. “Industry Wage Differences and Theories of Wage Determina tion.” National Bureau of Economic Research Working Paper No. 2271, July 1987. Atrostic, B.K. ‘Alternative Pay Measures and Labor Market Differentials.” U. S. Department of Labor Office of Research and Evaluation, Bureau of Labor Statistics Working Paper No. 127, March 1983. ________ and Lawrence F. Katz. “Industry Wage Differences and Industry Characteristics.” National Bureau of Economic Research Work ing Paper No. 2041, September 1986. Azariadis, Costas, “Employment With Asym metric Information.” 98 Supplement (1983): 157-172. ________ , and Kevin Lang. “Labor Market Seg mentation and the Union Wage Premium.” National Bureau of Economic Research Work ing Paper No. 1883, April 1986. Quarterly Journal of Economics. Becker, Gary. Human Capital. New York: NBER, 1964. Equal Pay in the Office. Blau, Francine. Lex ington, Mass: D.C. Heath and Co., 1977. Brown, Charles. “Equalizing Differences in the Labor Market.” 94 (February 1980): 113-134. ics. Quarterly Journal of Econom j ________ and Kevin Lang. ‘A Test of Dual Labor Market Theory.” 75 (September 1985): 792-805. American Economic Review. ________ and Kevin Lang. “Testing Dual Labor Market Theory: A Reconsideration of the Evidence.” National Bureau of Economic Research Working Paper No. 1670, July 1985. Inter nal Labor Markets and Manpower Analysis. ________ , and James L. Medoff. “The Employer Size Wage Effect.” Unpublished paper, October 1987. Doeringer, Peter B. and Michael J. Piore. Brown, William, John Hayles, Barry Hughes, and Lyndon Rowe. “Production and Labor Markets in Wage Determination: Some Aus tralian Evidence.” 22 (July 1984). Dunlop, John T. “Fundamentals of Wages and Labor Markets.” Unpublished paper, Harvard University. August 3 0 ,1982(a). trial Relations. British foum al of Indus Buckley, John E. “Do Area Wages Reflect Area Living Costs?” 102 (November 1979): 24-29. Monthly Labor Review. Bulow, Jeremy I., and Lawrence H. Summers. A Theory of Dual Labor Markets W ith Applica tion to Industrial Policy, Discrimination, and Keynesian Unemployment.” (July 1986): 376-414. Economics. 4 Journal of Labor Cain, Glenn G. “The Challenge of Segmented Labor Market Theories to Orthodox Theory.” 14 (December 1976): 1215-1257. Journal of Economic Literature. Lexington, Mass: D.C. Heath and Co., 1971. ________ . “What Both Sides Should Know About Compensation.” Unpublished paper, Harvard University. August 3 0 ,1982(b). ________ . “The Task of Contemporary Wage Theory.” In G. Taylor and F. Pierson, eds., New Concepts in Wage Determination. New York, 1957. Dunn, Lucia. “The Effects of Firm and Plant Size on Employee Well-Being.” In John Siegfried, ed., Wash ington, D.C.: Federal Trade Commission, The Economics of Firm Size, Market Structure and Social Performance. 1980. ________ “The Effects of Firm Size on Wages, Fringe Benefits, and Worker Disutility.” In Harvey Goldschmidt, et al., eds., New York: Columbia Press, 1984. o f the Modem Corporation. The Impact Eberts, Randall W , and Joe A. Stone. “Wages, Fringe Benefits and Working Conditions: An Analysis of Compensating Differentials.” 52 0uly 1985): 274-280. Southern Economic Journal. Contested Terrain. New Edwards, Richard. York: Basic Books, Inc., 1979. Ehrenberg, Ronald G., and George T. Milkovich. “Compensation and Firm Per formance.” In Kleiner, Morris, et al., eds., Human Resources and the Performance of the Firm. Industrial Relations Research Association, Madison, Wis. (1987): 87-122. Evans, Robert Jr. “Worker Quality and Wage Dispersion: An Analysis of a Clerical Labor Market in Boston.” I960. Industrial Relations Research Association Proceedings. Filer, Randall K. “Sexual Differences in Earnings: The Role of Individual: Personalities and 18 (Winter 1983): 82-99. Tastes”Journal of Human Resources. Freeman, Richard B. “The Effect of Unionism on Fringe Benefits.” 34 (July 1981): 489-509. Relations Review. Industrial and Labor “A Garbarino, Joseph W Theory of Interindustry Wage Structure Variation.” 64 (May 1950): 282-305. of Economics. Quarterly Journal Gibbons, Robert, and Lawrence F. Katz. “Learning, Mobility, and Inter-Industry Wage Differentials.” Massachusetts Institute of Tech nology. Unpublished paper, December 1987. Groshen, Erica L. “Sources of Wage Dispersion: The Contribution of Interemployer Differen tials W ithin Industry.” Federal Reserve Bank of Cleveland Working Paper 8802, January 1988a. ________ “The Size and Stability of Inter employer Wage Differentials W ithin an Area.” Unpublished paper, 1988b. Workers' Earnings and Corporate Economic Structure. New York: Hodson, Randy. Academic Press, 1983. Holmstrom, Bengt. “Equilibrium Long-Term Labor Contracts.” 98 (Supplement, 1983): 23-54. nomics. Quarterly fournal of Eco Jovanovic, Boyan. “Job Matching and the Theo ry of Turnover.” 87 (October 1979): 972-990. Journal of Political Economy '/Jour ________ . “Firm-Specific Human Capital 87 (December 1979): 1246-1260. nal of Political Economy. Kerr, Clark. “The Intellectual Role of Neorealists in Labor Economics.” 22 (Spring 1983): 298-318. Industrial Relations. Kleiner, Morris M., and Marvin L. Boullion. “Providing Business Information to Produc tion Employees: Impacts on Compensation and Profitability.” University of Minnesota. Unpublished paper, 1987. Krueger, Alan, and Lawrence H. Summers. “Efficiency Wages and the Wage Structure.” National Bureau of Economic Research Work ing Paper No. 1952, June 1986(a). ________ . “Reflections on the Inter-Industry Wage Structure.” National Bureau of Eco nomic Research Working Paper No. 1968, June 1986(b). Lang, Kevin. “Persistent Wage Dispersion and Involuntary Unemployment.” Boston Univer sity. Unpublished paper, December 1987. ________ , and W illiam T. Dickens. “Neoclassical and Sociological Perspectives on Segmented Labor Markets.” National Bureau of Economic Research Working Paper No. 2127, January 1987. Leonard, Jonathan S. “Carrots and Sticks: Pay, Supervision and Turnover 5 (October 1987). Economics. ''Journal of Labor “A Industrial and Labor Relations Lester, Richard A. Range Theory of Wage Differentials.” 5 (July 1952). Review. ________ . “Pay Differentials by Size of Establish ment.” 7 (October 1967): 57-67. Industrial Relations. Lewis, L. Earl. “Wage Dispersion in Manufactur ing Industries 1950-1955.” (July 1956): 780-786. Review. Monthly Labor Mackay, Donald I., David Boddy, John Brack, John A. Diack, and Norman Jones. Labour Markets Under Different Employment Condi tions. London: George Allen & Unwin Ltd., 1971. Masters, Stanley H. ‘An Interindustry Analysis of Wages and Plant Size.” 51 (August 1969): 341-345. and Statistics. Review of Economics Medoff, James L., and Katherine G. Abraham. “Experience, Performance and Earnings.” 95 (December 1980): 703-736. Quarterly Journal of Economics. Mellow, Wesley. “Employer Size and Wages.” 64 (August 1982): 495-504. Review of Economics and Statistics. Miller, Edward. “Variation of Wage Rates With Size of Establishment.” 8 (1981). Economic Letters. Schooling, Experience, and Earn Mincer, Jacob. New York: NBER, 1974. ings. Montgomery, James. “Equilibrium Wage Disper sion and Interindustry Wage Differentials.” Massachusetts Institute of Technology. Unpublished paper, December 1987. Nolan, Peter, and W illiam Brown. “Competition and Work Place Wage Determination.” Oxford Bulletin of Economics and Statistics. 45 (August 1983). Perlman, Jacob. “Hourly Earnings of Employees in Large and Small Enterprises.” Temporary National Economic Committee, United States Government Printing Office, Monograph No. 14,1940. Personick, Martin E., and Carl B. Barsky. “W hite Collar Pay Levels Linked to Corporate Work Force Size.” 105 (May 1982): 23-28. Monthly Labor Review. Pugel, Thomas A. “Profitability, Concentration and the Interindustry Variation in Wages.” 62 (May 1980): 248-253. Review o f Economics and Statistics. Workers and Wages in An Urban Labor Market. Chi Rees, Albert, and George P. Schultz. cago: The University of Chicago Press, 1970. The Structure o f Labor Reynolds, Lloyd G. New York: Harper & Brothers, 1951. Markets. The Evolution of Roy, Andrew D. “Some Thoughts on the Dis tribution of Earnings.” 3 (June 1951): 135-146. Oxford Economic Papers. Salop, Steven C. ‘A Model of the Natural Rate of Unemployment.” 69 (March 1979): 117-125. American Economic Review. Segal, Martin. “Post-Institutionalism in Labor Economics: The Forties and Fifties Revisited.” 39 (April 1986): 388-403. Industrial and Labor Relations Review. Seiler, Eric. “Piece Rate vs. Time Rate: The Effect of Incentives on Earnings.” National Bureau of Economic Research Working Paper No. 879, April 1982. Shapiro, Carl, and Joseph E. Stiglitz. “Equi librium Unemployment as a Worker Disci pline Device.” 74 (June 1984): 433-444. American Economic Review. Slichter, Sumner. “Notes on the Structure of Wages.” 32 (February 1950). Review of Economics and Statistics. The Wealth of Nations Smith, Adam. (1776). Penguin English Library Edition, 1982. Smith, Robert S. “Compensating Wage Differen tials and Public Policy: A Review.” 32 (April 1979): 339-352. and Labor Relations Review. Industrial Smith, Robert, and Ronald G. Ehrenberg. “Estimating Wage-Fringe Trade-Offs: Some Data Problems.” National Bureau of Economic Research Working Paper, No. 827, December 1981. Spence, A. Michael. “Job Market Signalling.” 87 (May 1973): 355-375. Quarterly Journal of Economics. Stigler, George J. “Information in the Labor Market.” 70 (October 1962): S94-105. Journal o f Political Economy. ________ , and Cynthia Taft. New Haven, Conn.: Yale Press, 1956. Stiglitz, Joseph E. “Theories of Wage Rigidities.” National Bureau of Economic Research Work ing Paper, 1984. Rosen, Sherwin. “O n the Interindustry Wage and Hours Structure 77 (March-April 1969): 249-273. ________ “O n Search and Equilibrium Price Distributions.” In Michael J. Posken, ed., New York: Aca demic Press, Inc., 1979. Wage Structure. Economy "Journal of Political Rothschild, Michael, and Joseph E. Stiglitz. “Equilibrium in Competitive Insurance Mar kets: An Essay on the Economics of Imperfect Information.” 90 (November 1976): 629-649. ics. Quarterly Journal of Econom nomics of Human Welfare. Eco Topel, Robert H. “Equilibrium Earnings, Turn over and Unemployment.” (October 1984): 500-522. Economics. Journal of Labor Van Giezen, Robert W. ‘A New Look At Occupa tional Wages Within Individual Establish ments.” 105 (November 1982): 22-28. Monthly Labor Review. Venables, Anthony J. “Random Job Prospects and the Distribution of Income.” 98 (November 1983): 637-657. Quarterly Journal of Economics. Wachtel, Howard M., and Charles Betsey. “Employment at Low Wages.” 54 (May 1972): 121-129- nomics and Statistics. Review of Eco Wachter, Michael L. “Cyclical Variation in the Interindustry Wage Structure.” 60 (March 1970): 75-84. Economic Review. American Ward, Virginia L. “Measuring Wage Rela tionships Among Selected Occupations.” 103 (May 1980): 21-25. Monthly Labor Review. Weiss, Leonard A. “Concentration and Labor Earnings.” 56 (March 1966): 96-117. American Economic Review. The Share Economy: Con quering Stagflation. Cambridge, Mass.: Weitzman, Martin. Harvard University Press, 1986. Yellen, Janet L. “Efficiency Wage Models of Unemployment.” 74 (May 1984): 200-205. Review. American Economic E c o n o m ic R e view ■ Quarter I 1987 Concentration and Profitability in Non-MSA Banking Markets by Gary Whalen ■ Quarter III 1987 Can Services Be a Source o f Export-led Growth? Evidence from the Fourth District by Erica L Groshen The Effect o f Regulation on Ohio Electric Utilities by Philip Israilevich and K.J. Kowalewski Identifying Amenity and Productivity Cities Using Wage and Rent Differentials by Patricia E. Beeson and Randall W. Eberts Views from the Ohio Manufacturing Index by Michael F. Bryan and Ralph L Day FSLIC Forbearances to Stockholders and the Value o f Savings and Loan Shares by James B. Thomson ■ Quarter II 1987 A New Effective Exchange Rate Index for the Dollar and Its Implications for U.S. Merchandise Trade by Gerald H. Anderson, Nicholas V. Karamouzis and Peter D. Skaperdas ■ Quarter IV 1987 Learning, Rationality, the Stability o f Equilibrium and Macroeconomics by John B. Carlson How Will Tax Reform Affect Commercial Banks? by Thomas M. Buynak Airline Hubs: A Study o f Determining Factors and Effects by Paul W. Bauer A Comparison o f Risk-Based Capital and Risk-Based Deposit Insurance by Robert B. Avery and Terrence M. Belton F irst Q u arte r W o rkin g P a p e rs W orking Paper Notice The Federal Reserve Bank A s of January 1, 1987, we no of Cleveland has changed its longer send method of distribution for the individuals as part of a mass Working Paper series mailing. Our current produced by the Bank's Research Department. Working Papers to Working Papers will be listed on a quarterly basis in each issue of the Economic Review. Individuals may request copies of specific Working Papers listed by Papers will be sent free of charge to those who request them. A regular mailing list for Papers, maintained for personal subscribers. Libraries and other organizations m ay request to be placed on a mailing list for institutional completing and mailing the attached subscribers and will automatically form below. receive Working Papers published. ■ 8801 ■ 8802 T o b in ’ s q , In v e s tm e n t, and th e En d o g e n o u s A d ju s tm e n t o f Fin a n c ia l S o u rc e s o f W a ge D is p e r s io n : T h e C o n trib u tio n S tru c tu re e n tia ls W ithin In d u s try by W illia m P. O ste rbe rg by E rica L. G roshen o f In te re m p lo ye r D iffe r Please com plete and detach the form below and mail to: Federal Reserve B an k of Cleveland Research D ep a rtm en t P.O . B o x 6 3 8 7 C leve lan d, O h io 44101 Check item(s) Please send the fo llo w in g W o rk in g Paper(s): requested. □ 8801 □ 8802 Send to : Nam e Please p rin t A ddress Working however, will not be as they are