PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT STEPHEN J. TREJO*
Using 1980 data for a large sample of U.S. cities, the author reexamines recent empirical findingsof a positive association between public sectorunionizationand municipalemployment.Several researchers have interpretedthis correlationas evidence that public employee unions successfullyexert political pressure to raise the demand for municipal services. Structural estimates of labor demand and the determinants of police and fire unionization reveal, however, that economies of scale in union formationare at least partlyresponsiblefor any positive association between public sector unionization and municipal employment. The author concludes that previous studies overstate the amount of political clout wielded by municipal labor unions.
sectorlabor unions are currently the focus of an enormous amount of research. The main impetus for this research has been the explosive growthof public sector unionization over the past thirtyyears. These union gains in the public sector are all the more remarkable because theyoccurred during a period in which private sector unions suffereddramatic declines in membership, and also because public sector unions have been successful at organizing the types of white-collarworkerswho forthe mostpart have resistedjoining privatesectorunions. Many studentsof collectiveorganization by government employees have focused on unique legal, institutional,and ecoPUBLIC
* The author is AssistantProfessorof Economics at the Universityof Californiaat Santa Barbara. He thanksGeorge Borjas, Jan Brueckner,JeffreyGrogger, Kristin McCue, Kevin Murphy, Wayne Pierce, and Sherwin Rosen for advice and comments. The data used in this paper are publiclyavailable fromthe U.S. Bureau of the Census. For information about the computer programs used to analyze these data, write the author at the Department of Economics, Universityof California,Santa Barbara, CA 93106.
nomic features of the public sector bargaining environment that distinguish these unions from their private sector counterparts.Freeman (1986) and others have argued that,because of the political contextin which governmentoutput decisions are made, public employee unions may be able to influence the demand for their labor as well as the supply of it. In supportof thisargument,recentempirical studies have documented a positiveassociation across cities between unionization and employmentlevels in certain municipal services.Given available evidence that public sector unions also raise wages, several researchers have interpreted the observed relationship between unionization and municipal employmentas implying that these unions effectivelymarshal political pressure to raise the demand for local governmentservices. Alternativeinterpretationsof this evidence are available, however, including the possibilitythatlarge public sectorwork forces are more likely to be organized. Using 1980 data fora large sample of U.S. cities,I reexamine the employmenteffects of municipalunions and attemptto empir-
Industrialand LaborRelationsReview,Vol. 45, No. 1 (October 1991). ?Dby Cornell University. 0019-7939/91/4501$01.00
166
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve, and extend access to Industrial and Labor Relations Review www.jstor.org
®
PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT ically distinguishunion effectson labor demand frompossible economies of scale in union formation. Unions and Employment Several studies have uncovered a positive correlationacross cities between employmentand measures of union power for selected categoriesof municipal workers. Benecki (1978) estimated reduced form employment equations and found that total municipal employment increased with percent unionized for cities under 25,000 in population. Victor (1977) and Spizman (1980a, 1980b) estimated labor demand functions and provided evidence that unions shifted out the demand for teachers, police, fire, and public welfare workers. Freeman and Valletta (1988) and Valletta (1989) reported that coverage under a collective bargaining contract increased municipal employmenton the order of 20%. Finally, Zax and Ichniowski(1988) estimatedthat the presence of a bargaining unit increased municipal employment by 11%, and Zax (1989) found that municipal departmentsrepresented by a bargaining unit experienced more rapid employment growththan did departmentslackingsuch representation. These findings are very provocative because they contradictstandard theory, in which the union is constrained to operate along a downward-sloping derived demand for labor curve that is imperviousto union influence. The standard monopoly model of union behavior, coupled with the small elasticities of demand typically estimated for many types of public sector labor (Ehrenberg and Schwarz 1986), could conceivably explain reduced form union employment effectsnear zero, but it cannot account for the large positive effectsof unionization on employment or labor demand that have been estimatedfor certaingroups of municipalworkers.' 1 It should be noted that surprisinglyfew studies have investigated the impact of labor unions on employmentin the privatesector,and the empirical
167
The fact that municipal employment is sometimes positively associated with unionization has been interpretedas evidence thatpublic sectorunions can exploit politicalchannels to enhance theiropportunities(Spizman 1980a, 1980b; Freeman 1986; Zax 1989). Specifically, through voting and lobbying campaigns, public employees might be able to shiftout the demand for public services, and in the process raise the derived demand for public sector labor. Public employees could be particularlyeffectiveat influencing political outcomes because they have been shown to possess votingparticipation rates much higher than average (Bennett and Orzechowski 1983), and also because the nature of their jobs provides them witha heightened awareness of the political process. Unions are usually viewed as the mechanism through which public employee political pressure is effected, because substantial free rider problems discourage collective action by unorganized employees. If public sector unions succeed in raising the demand for their labor, then it becomes possible for unionization to increase wages and employment simultaneously. On the other hand, findingsthatunionization and employment vary together across cities need not signal the existence of political pressure by municipal unions. Rather than unionization leading to increased employment,the causality could run in the opposite direction: economies of scale in union organizing might mean that large work forcesare more attractive targets for unionization.2 Many of the support for negative union employmenteffectsin this sector is far from overwhelming.Lewis (1964) found that unionism significantlyreduced both employmentand man-hours worked, whereas Pencavel and Hartsog (1984), who replicatedand refined Lewis's analysis with an extended time series, concluded that the evidence regarding negative union man-hours effectswas mixed. Montgomery (1989) reported that increases in union strength across SMSAs led to small but statistically significant reductions in the probabilityof employment and larger reductions in the usual weekly hours of workers. emnployed The literature on pareto efficientwage and employment contracts provides another potential
168
INDUSTRIAL
AND LABOR RELATIONS REVIEW
costsinvolvedin organizingand maintaining a union are roughly independent of the size of the union, so if marginal variable costs are constant or gradually rising, the substantial fixed cost component will cause cost per member to fall as the union gets larger. Of course, this argument considers only one side of the union formationprocess, and fixed costs leading to economies of scale may also exist for management's effortsto resist unionization. Moreover, the per member gains from unionization probably vary with the size of the bargaining unit. Therefore, the size of a municipaldepartment is likelyto influencewhetheror not a union formsand how comprehensiveit becomes, although the direction of this effectcannot be determineda priori. The empirical work conducted to date on this issue suggests that unionism is more prevalentin larger firms.For example, Hirsch (1982) found unionization in manufacturingindustriesto increase with average firm sales. Hirsch and Berger (1984) reported that manufacturingproduction workers were more likely to be union members the larger the average establishment size in the industry, and Robinson and Tomes (1984) obtained a similar result for hourly paid workers in both the public and private sectors in Canada. Finally, Zax (1989:24) cited a workingpaper by Zax and Ichniowski in which the probability that a municipal department forms a bargaining unit was found to increase with the level of employmentin the department. explanation for the observed positive correlation between unionization and municipal employment. Wage and employmentcombinations on the labor demand curve are inefficientin the sense that both employerand union can in general be made better offby agreeing to a point offand to the rightof the demand curve. Under efficientcontracting,it is possible for unions to raise both wages and employment above competitivelevels, and McDonald and Solow (1981) showed that this must be the case for certain plausible specificationsof union and firm preferences.Ebertsand Stone (1986) found evidence of efficientcontractingby public school teachers in New York state,but Brueckner and O'Brien (1989) testedand rejectedthe efficientcontractsframework using data on police, fire, and sanitation workers similarto the data employed here.
Data By merging computer tape files from the 1980 Annual Survey of Governments and the 1983 County and City Data Book, I was able to create a cross-sectional data set thatincludes most U.S. municipalities with a 1980 population of at least 25,000. Detailed description and published tablesof these data maybe found in U.S. Bureau of the Census (1981a, 1981b, 1983). The Annual Surveyof Governmentsprovides payroll,employment,and labor relations data for the month of October 1980 thatis disaggregatedby functionalcategories of municipalemployment.These data, and other years of these data, have been used in previousstudiesof unions and municipalemployment(Freeman and Valletta 1988; Zax and Ichniowski1988; Zax 1989). Only the data on police and fire departments are analyzed in the present study.3 These data include nonuniformedas well as uniformedemployees. For each municipal departmentwithina city,the average annual salarywas computedbydividingthe monthlyfull-timepayrollbythe numberof full-timeemployees and multiplyingby twelve.Similarly,the unionizationrate was computed fromdata on the total number of full-timeemployees and the number of these employees belonging to labor organizations.4 The 1983 County and City Data Book contains sociodemographic and economic data for each city. In addition, the governmentalformof each municipalityis characterized as being either mayorcouncil, council-manager,or commission, 3 Data are also available on four other categories of municipal employment.Too few cities employed hospital and public welfare workers to provide reliable estimates,and the data on sanitation(other than sewerage) and streetsand highwaysdisplayed no evidence of the positiveunion employmenteffects thatare the focus of the currentpaper. 4 Unionized fire departments are almost exclusivelyrepresentedby the InternationalAssociationof Fire Fighters.Police departments,on the otherhand, are split among several organizations,primarilythe Fraternal Order of Police and the International Conference of Police Associations (Kearney 1984: 32-33).
PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT as defined by the International City Management Association(1984). An average annual salary in retail trade was computed for each city by dividing the annual payroll in that industry by the number of paid employees. A few variables were obtained from other sources. Using U.S. Department of Labor (1979) and Nelson (1980, 1981), I constructeda dummy variable identifying those stateswithlaws in 1980 thatimposed upon municipalities a "duty-to-bargain" withany union representinga majorityof the employees in a bargainingunit. These stateshave legislatedbargainingrightsfor
169
local public employees similarto those the National Labor RelationsAct provides for most private sector employees. Troy and Sheflin (1985, Tables 7.6 and 7.7) report the unionizationrate of the privatesector, nonagriculturalworkforceof each statein 1975 and again in 1982, and they also indicate which states had "right-to-work" laws in 1980. Right-to-work laws prohibit collectivebargainingcontractsthatrequire union membershipor dues paymentas a conditionof employment. Table 1 provides variable definitions. The data on employment,salaries, and unionization are available separately for
Table 1. Variable Definitions and Sources. Variable
Definition 1980 Annual Survey of Governments
Employment Salary Unionization
= full-timemunicipalemployment. = average annual full-timemunicipal salary. = fractionof full-timemunicipal employees unionized. 1983 County and City Data Book
Population Density Female Black Under Age 18 Age 65 and Over High School College Income Grants House Value Renter Unemployment Salaryin Retail Trade DummyVariables Indicating Form of Municipal Government: Mayor Manager Commission Duty-to-Bargain PrivateSector Unionization Change in PrivateUnionization Right-to-Work Dummy Variables Indicating Census GeographicalRegion: Northeast North-Central South West
= = = = = = = = = = = = = =
total citypopulation. population per square mile. fractionof population that is female. fractionof population thatis black. fractionof population under age 18. fractionof population age 65 and over. fractionof population with 12 or more years of education. fractionof population with 16 or more years of education. per capita money income. per capita intergovernmental revenue. median value of owner-occupiedhousing units. fractionof housing units occupied by renters. civilianunemploymentrate. average annual salary in retailtrade.
=
1 if mayor-council,0 otherwise. 1 if council-manager,0 otherwise. 1 if commission,0 otherwise.
= =
Data fromOther Sources = 1 if state has a "duty-to-bargain" law whichapplies to municipal employees,0 otherwise. = fractionof the state'sprivatesector,nonagriculturalwork force thatwas unionized in 1982. = percentage change in the state's privatesectorunionizationrate between 1975 and 1982. = 1 if state has a "right-to-work" law, 0 otherwise. = 1 if cityis located in the Northeast,0 otherwise. = 1 if cityis located in the North-Centralstates,0 otherwise. = 1 if cityis located in the South, 0 otherwise. = 1 if cityis located in the West, 0 otherwise.
170
INDUSTRIAL
AND LABOR RELATIONS REVIEW
police and fire departments within each city, whereas the duty-to-bargainand right-to-work dummies and the two variables describing private sector unionization varyonly by state; all other variables varyby city.All data are for the year 1980 except as follows:per capital income is for 1979, per capita intergovernmentalgrants are for a fiscal year running from July 1980 throughJune 1981, average annual salary in retail trade is for 1977, the dummy variables indicating the form of municipal governmentare for 1982, and the privatesectorunionizationrates are as previouslydescribed. Table 2 presents means and standard deviations of the variables used in the empirical analysis. Because not all cities supply the same municipal services, the sizes of the police and fire samples are slightlydifferent. Reduced Form Effectsof Municipal Unionization As discussed above, previous studies have estimated reduced form effectsof unionization on municipal employment that often turn out to be positive. In this section, I firstdemonstratethat a similar phenomenon occurs in the police and fire data studied here, and then testwhetherit is valid to treat unionization as an exogenous variable. The firstand thirdcolumns of Table 3 present the coefficientson percent unionized fromreduced formemploymentand salary regressions estimated by ordinary least squares (OLS).5 Standard errors are 5 Throughout this paper I exclusively utilize percent unionized as the measure of union power. Other dimensions of union power, such as the existence of a collective bargaining contractor the presence of a bargainingunit, have also been shown to significantlyaffect municipal labor markets. Unfortunately,as discussed in Valletta (1989), the existenceof a collectivebargainingcontractcan only be imperfectlyimputed for each municipal departmentusing the Annual Surveyof Governmentsdata. More important,the present study focuses on the demand effects of union political influence and possible economies of scale in union formation,and for both of these purposes a continuous measure of union power such as the unionization rate seems
reported in parentheses. Separate equations are estimated for police and fire departments,and the dependent variables are the natural logarithmsof employment and average annual salary for each department.The note to the table lists the control variables included in the regressions.6 To facilitateinterpretationof the regression coefficients,all variables that do not represent either percentages or dichotomous dummies are expressed as natural logarithms.The complete set of coefficientestimatesfor these regressions is reported in the Appendix. The OLS estimates of reduced form employment effects are large, positive, and statistically significantfor both police and fire unions. Fully unionized police departments are about 12% larger than nonunion police departments,and unionization appears to increase the size of fire departmentsby almost 37%. The finding thatthe employmenteffectsof fireunions are substantiallylarger than those generated by police unions is consistent with other studies that estimate separate regressions by department (Victor 1977; Spizman 1980a, 1980b; Zax and Ichniowski 1988). Least squares estimates of union salary premiums are statistically significantand on the order of 7-11%, quite comparable in magnitude to those obtained by Freeman and Valletta (1988) and Zax and Ichniowski (1988) using similar data. These results confirm the finding of previous work that municipal unionization can be associated with both higher wages and increased employment. As withpast researchon thissubject,the OLS estimates just presented assume municipal unionization to be exogenous.7 more revealing. A discrete representationof union power masks the substantialvariation that exists in these data between the polar extremes of no unionizationand complete unionization. 6 The choice of control variables will be justified below when a structuralmodel of municipal labor marketsis estimated.The control variables are very similar to those used in previous studies such as Freeman and Valletta(1988) and Zax and Ichniowski (1988). 7Zax (1989) recognized the potentialsimultaneity of employmentand unionization,and attemptedto mitigatethisbias by includinglagged employmentas
PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT
171
Statistics. Table 2. Descriptive Police Variable Employment Salary Unionization Population Density Female Black Under Age 18 Age 65 and Over High School College Income Grants House Value Renter Unemployment Salary in Retail Trade Mayor Manager Commission Duty-to-Bargain PrivateSector Unionization Change in PrivateUnionization Right-to-Work Northeast North-Central South West
Fire
Mean
Std.Dev.
Mean
Std.Dev.
293.92 18,548.02 .58 106,338.26 3,740.70 .52 .12 .27 .12 .70 .18 7,580.29 156.03 52,243.02 .40 .06 6,653.61 .35 .61 .04 .37 .19 -.27 .32 .15 .30 .28 .27
1291.23 4,003.50 .36 335,172.33 2,983.58 .02 .16 .05 .05 .11 .10 1,863.81 144.33 25,876.80 .12 .03 775.36 .48 .49 .18 .48 .08 .07 .47 .36 .46 .45 .45
190.31 19,340.83 .71 110,917.71 3,737.07 .52 .12 .27 .12 .69 .18 7,494.72 162.36 52,434.89 .41 .07 6,633.49 .37 .59 .04 .38 .19 -.27 .33 .14 .31 .29 .26
595.91 4,322.24 .37 346,410.68 3,017.76 .02 .16 .05 .05 .11 .10 1,843.36 147.03 24,815.44 .12 .03 765.42 .48 .49 .19 .48 .08 .07 .47 .35 .46 .45 .44
Sample size
For the reasons discussed above, the validityof this assumption can be called into question. To investigate the consequences of relaxing this assumption, the second and fourth columns of Table 3 report instrumental variables (IV) estimates of percent union coefficientsfrom regressionsidentical to those in the first an independent variable. This specificationessentiallyamountsto estimatingthe effectof unionization on employmentgrowth,and Zax found thiseffectto significant.Of course, if be positiveand statistically unions prefer to organize large municipal departments,then presumablytheyalso preferto organize departmentsthat are expected to grow rapidly, so Zax's specificationdoes not eliminate the potential for simultaneitybias. Freeman and Valletta (1988) and Valletta(1989) pooled data forseveral municipal departmentsand included city-specificfixed effects to controlforomittedcitycharacteristicsthatmay be correlatedwithboth employmentand unionization, but they did not directlyaddress the issue of union endogeneity.
709
662
and third columns except that predicted unionization rates are used in place of actual unionizationrates.The instruments for municipal unionization include all other independent variablesin the regressions, as well as the unionization rate of the state's private sector, nonagricultural workforcein 1982, the percentagechange in the state's private sector unionization rate between 1975 and 1982, and a dummy variable identifyingstates with right-to-work laws. The latter three variables, whichprovide identification, appear to be ideal instruments because they are highly correlated with municipal unionizationbut would not be expected to directlyaffectmunicipal employmentand salaries.8
8 Bartel and Lewin (1981:67) also argued that the
172
INDUSTRIAL
AND LABOR RELATIONS REVIEW
Table3. Effects of Policeand FireUnionization. (Standard Errors in Parentheses) Fire
Police DependentVariable
OLS
log(Employment)
.122** (.053)
log(Salary)
.108*** (.020)
IV -.393 (.267) .368*** (.107)
OLS .367*** (.059) .066*** (.019)
IV -.063 (.277) .390*** (.101)
Note: Sample sizes are the same as in Table 2. Other independent variables include: total municipal population, population density,percentfemale,percentblack, percentunder age 18, percentage 65 and over, percent high school graduates, percent college graduates, percent renters, median housing value, civilian revenue, average annual salary in retail unemploymentrate, per capita income, per capita intergovernmental law is in effectfor municipalworkers,dummy trade,a dummyvariable indicatingwhethera "duty-to-bargain" variables indicatingthe formof municipal government,and dummy variables indicatingcensus geographical region. The instrumentsfor municipal unionization used in the second and fourth columns include the independentvariableslistedabove, as well as the state'sprivatesectorunionizationrate in 1982, the percentage change in the state'sprivatesectorunionizationrate between 1975 and 1982, and a dummyvariableidentifying states. "right-to-work" Statistically significantat the .10 level; ** at the .05 level; *** at the .01 level (two-tailedtests).
The differencesbetween the OLS and IV estimates are striking.The estimated union employment effects that were strongly positive in the first and third columns become negative and statistically insignificantin the second and fourth columns. Instrumenting for municipal unionization also results in estimates of union salary premiums that are surprisingly large, but this result is consistent with previous research, which finds that accounting for union endogeneity can greatlyincrease estimates of union wage effects in the public sector (Bartel and Lewin 1981; Robinson and Tomes 1984). The specification test developed by Hausman (1978) can be used to formally test for the exogeneity of municipal unionization. Intuitively,Hausman's procedure can be seen as testingwhetherthe different OLS coefficientsare significantly fromthe IV coefficients.Computationally, I reestimatethe reduced formregressions, includingboth unionizationand predicted unionization as right-hand-sidevariables, along with the other independent variables. If the estimated coefficient on difpredicted unionizationis significantly ferentfrom zero, I reject the hypothesis that unionization is orthogonal to the error term of the corresponding least private sector unionization rate constitutesa valid instrumentin thiscontext.
squares regression. Implementation of this test requires that there exist instruments that are correlated with public sector unionization but do not exert an independent influence on municipal employment or salaries. I assume that the level and percentage growthof the state's private sector unionization rate and the state'sright-to-work statusare valid instruments. The results of these specificationtests suggest that municipal unionization is indeed an endogenous variable. In the employmentregressions,union exogeneity is rejected at the 5% level of significance for police departmentsand at the 10% level for fire departments. In the salary regressions, union exogeneity is rejected at the 1% level for both police and firedepartments. The evidence presented in this section indicates that simultaneitybias contaminates OLS estimates of positive employment effectsby police and fire unions. Least squares regressionsreveal large and positive employment effects for these unions, but instrumentingfor unionization reverses the estimated sign of the employmenteffects,and Hausman specification tests reject the maintained assumption of previous studies that unionization is exogenous to the process of municipal employmentdetermination.It is always possible to question the validity
PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT of the instrumentsused, but in this case seeminglysensible instrumentsare available, and thus we are compelled to entertain seriously the possibility that municipal employment and unionization are mutuallydependent. Therefore,in the next section I present structuralestimates of the demand for police and fireworkers and the determinantsof municipal unionization,in an attemptto disentangleunion effectson labor demand from economies of scale in union formation. StructuralEstimates Reduced form employmentequations, such as those reported in Table 3, measure the' total effect of unionization on employment, which is the sum of the direct effectof unionization on employment and the indirecteffectof the higher union wage on employment. In other words, reduced form estimates of union employmenteffectsconfound shiftsin the labor demand curve withmovementsback up along the curve. These estimates thereforedo not provide a clear-cuttestof the hypothesis that public sector unions can alter the demand for their labor, because union wage gains can lead to a negative net effecton employmenteven when unions are successfulin shiftingout demand. A more direct test of this hypothesis was undertaken by Victor (1977) and Spizman (1980a, 1980b), who estimatedunion employmenteffectsholding salary constant. In essence, these studies added unionization as an exogenous variable to municipal labor demand equations. Consider labor demand functionsof the form (1) log(Employmenti)= Nqlog(Salary,) + biUnionization +
Xr1i + Ei
where i indexes the municipal department (i = p for police and i = f for fire). Following previous work (Ehrenberg and Schwarz 1986), the derived demand for labor in a municipal department is presumed to be a functionof its cost and a vector X of citycharacteristicsthat influ-
173
ence the demand for municipal services. In order to testfor the presence of public employee political pressure, the department'sunionizationrate is included as one of the variables allowed to shift labor demand. Included in the vectorX are the percent of the city'spopulation that is female,the percent black, the percent under age 18, the percent age 65 and over, the percent of housing units occupied by renters, dummy variables indicating the census geographical region, and natural logarithms of total municipal population, population density,per capita income, per capita intergovernmental grants from state and federal sources, and the median value of owner-occupied housing. The income and grant variables affectmunicipal budgets,and the othervariablesreflect differencesin population size and composition across cities that may influence the demand for police and fire protection.In addition, population size, population density, and the regional dummy variables help control for intercityvariation in the price level.9 Because of the endogeneity of (log) salary, this variable is instrumentedfor using all other exogenous variables in the demand equation, as well as the percentof high school graduates in the city'spopulation, the percent college graduates, the civilian unemployment rate, the natural logarithmof the average annual salary in retail trade, a dummy variable indicating whether the state has passed a dutyto-bargain law that obliges city governments to bargain in good faith with municipal employee unions, and dummy 9 No comprehensivecost-of-living index existsfor U.S. cities, so some previous researchers have estimatedsuch an index using available price data for metropolitanand nonmetropolitanareas (Ehrenberg 1973; Bartel and Lewin 1981). All dollar variablesare then deflated by the estimated cost-of-living index, but these studies find that regression results using the deflated variables are very similar to those obtained with the undeflated variables. Moreover, my estimates do not change appreciably when the census region dummies are replaced by more detailed dummy variables representing the nine census divisions. Therefore, it is unlikely that the results reported here are affected by intercity differences. cost-of-living
174
INDUSTRIAL
AND LABOR RELATIONS REVIEW
variablesindicatingthe formof municipal government. These additional instruments provide identificationof the demand equation. The education variables measure the skill level of the local labor force, the unemployment rate reflects local labor market conditions, and the annual salary in retail trade is meant to proxyfor the opportunitywage of municipal workers.'0 The duty-to-bargainvariable captures interstate differences in public sector labor laws; these laws have been shown to significantlyaffect the bargaining power of both union and nonunion departments(Freeman and Valletta 1988). Finally,previous studies indicate that the form of municipal governmentcan have importanteffectson police and fire salaries (Ehrenberg 1973; Bartel and Lewin 1981). The specification in equation (1) assumes that the markets for separate categoriesof municipallabor are independent of each other withinas well as across cities. Yet, we might expect municipal governments as multiproduct firms to consider the prices of all factors of production when choosing employment for any given department, and it also seems likely that the bargaining and political power of a municipal union depends on the extent of unionization in other departmentsin the same city.These considerations imply that the labor demand equation for any one municipal department should include salary and unionization variables from all other departmentsin the same city.As a practical matter, however, the high correlation among salaries and unionization rates within cities leads to very imprecise and unreliable estimatesof these cross-effects 10This measure of the opportunitywage suffers from its sensitivityto intercityvariation in annual hours of work,which may be substantialbecause of the prevalence of part-timework in retailtrade. For a subsample of cities, data on the average hourly earnings of manufacturingproduction workers are available. Very similar results were obtained using this alternative opportunitywage measure; therefore, the results reported below use the average annual salary in retail trade because it allows the sample size to be increased by 25%.
in the data used here, so I retain the simplerspecification." A related issue is thatthe error termsin the police and fire labor demand equations (EP and Ef)are likelyto be correlated within the same city because of cityspecificshocks that affectemploymentin all municipal departments. Two-stage least squares estimatesdo not account for this within-city correlationyet are consistent,but efficiencygains can be realized by implementing three-stage least squares procedures that combine the police and fire equations into a single system and allow the errortermsto be contemporaneously correlated. Two-stage and threestage least squares produced similar results, so only the two-stageleast squares estimatesare reported below. In this way we can retainin the sample those citiesfor whichthere exist eitherpolice or firedata but not both. Table 4 presents estimatesof equation (1) separatelyfor police and fire workers under the assumptionthat unionizationis exogenous. As expected, for both departmentsthe effectsof unionizationon labor demand reported in Table 4 are larger than the OLS reduced form employment effectsgivenin the firstand thirdcolumns of Table 3. Because fire departmentsare estimatedto have much more wage-elastic demand functions than police departments,they also display a larger increase in the union employment effect when going from Table 3 to Table 4. The estimatesin Table 4 implythat police and fireunions are able to engineer substantial increases in labor demand. Estimatedcoefficientsforthe remaining variables appear reasonable in both sign and magnitude and are broadlyconsistent withprevious empiricalwork surveyedby Ehrenberg and Schwarz (1986) and Ru"
See Zax and Ichniowski (1988) for a recent attemptto sort out the labor marketinterdependencies that may arise between municipal unions in the same city.Interestingly, theyfound thatthe presence of bargaining units in other municipal departments did not significantly affectemploymentin police and fire departments, so equation (1) may provide a workable description of the demand for municipal employment.
PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT Table 4. Demand for Police and Fire Workers. Dependent Variable = log(Employment) Exogenous Unionization (Standard Errors in Parentheses) Variable Predictedlog(Salary) Unionization
log(Population) log(Density) Female
Black Under Age 18 Age 65 and Over
log(Income) log(Grants) log(House value)
Renter North-Central South West Intercept
R2
Police -.060
(.324) .134** (.066)
1.053***
(.027) -.042 (.029) -.066 (1.153)
.438***
Fire -2.217
(.491)
.492*** (.081)
1.068***
(.040) .043 (.043) 2.536 (1.593)
.404**
(.139) .572 (.492) 1.761*** (.471)
(.202) .622 (.758) 1.708** (.696)
(.148) .070*** (.025) -.340*** (.089)
(.215) .153*** (.037) - .245* (.130)
.669***
1.120***
(.194) - .101* (.056) .013 (.077) - .047 (.064) -9.357*** (2.311) .846
.906
1.349***
(.291) -.014 (.084) - .005 (.107) .386 (.112) 5.338 (3.358) .751
Note: Sample sizes are the same as in Table 2. Instrumentsfor (log) salary include all other independent variables in the employment equation, as well as percent high school graduates, percent college graduates,civilianunemploymentrate, average annual salary in retail trade, a dummy variable indicating whether a "duty-to-bargain"law is in effectfor municipal workers,and dummy variables indicatingthe formof municipal government. * Statistically significantat the .10 level; ** at the .05 level; *** at the .01 level (two-tailedtests).
binfeld (1987). Both labor demand functions are downward-sloping,and the demand for police is less elastic than the demand for fire workers. Municipal employmentand population size are roughly
175
proportionalto each other,and per capita income and grant revenue received from state and federal governmentshave positive and statisticallysignificanteffectson municipallabor demand. The demand for police and fire workers is also larger in cities with more blacks, more senior citizens, more renters, and lower house values. The labor demand estimatespresented in Table 4 assume municipal unionization to be exogenous. In order to investigate the consequences of relaxing thisassumption,I posit thatthe extentof unionization in municipal departmenti is determined according to the followingequation: (2)Unionizationi = otilog(Employmenti) + Xilog(Salaryi)+ Z-yi +
Vi,
where Z is a vector of characteristicsthat affect union formation in the public sector. Included in Z are natural logarithms of total city population and the average annual salary in retail trade, the level and percentage growthof the state's private sector unionization rate, and dummyvariables indicatinggovernmental form,whetherthe statehas a public sector duty-to-bargain law, whetherthe state has a right-to-work law, and census geographical region. The salary in retail trade is meant to reflectlocal wages in a predominantlynonunion sector,and the variables describingprivatesector unionizationand the dummy identifying right-to-work states characterizehow receptivethe local environmentis to unionization. Previous work has demonstrated that form of government(Bartel and Lewin 1981) and duty-to-bargainlaws (Zax and Ichniowski 1990) are importantdeterminantsof municipal unionization. Finally,city population is included so that the estimated coefficienton (log) employmentdoes not simply reflect the high correlation between population and municipal employment. Equations (1) and (2) comprise a structural model in which municipal employment and union density are jointly determined.Municipalsalariesare also considered to be endogenous. The model
176
INDUSTRIAL
AND LABOR RELATIONS REVIEW
is estimatedby two-stageleast squares, using as instrumentsall the exogenous variables in the employmentand unionization equations (the vectorsX and Z), as well as percenthighschool graduates,percentcollege graduates,and the civilianunemployment rate. These latterthree variablesare assumed to influence employment and unionizationonly indirectlythroughtheir effecton municipal salaries, and this assumptionhelps to identifythesystem.Identificationis also achieved through exclusion restrictionsin the employment and unionization equations. For example, the variablemeasuringper capita intergovernmental grants enters the labor demand equation but not the unionization equation, whereas the state's private sector unionizationrate directlyaffectsonly municipal unionization,not employment.12 Table 5 reports estimatesof equations (1) and (2) for police and fire departments. The labor demand functionsare presented in the firstand third columns, with the key difference between these estimatesand those presented in Table 4 being that the specification in Table 5 treats unionization as an endogenous variable. This differencehas a dramatic effect on the unionization coefficients. The abilityof police unions to raise the demand for theirlabor falls by 40% from what it was in Table 4, and the corresponding coefficientfor fire unions is a tenthof its previous value. Note, however, that instrumentingfor unionization also greatly increases the standard errors of the unionization coefficients,making it difficultto infer much about the magnitude or directionof union demand effects. The coefficientsof the other variables in the labor demand equations are similarto what theywere in Table 4. The second and fourth columns of Table 5 presentestimatesof the structural equations determining the extent of unionization in police and fire depart12
Althoughguided by theory,decisions regarding exactlywhich variables to include in each equation remainsomewhatarbitrary,so I experimentedwitha varietyof differentspecifications.The main empirical resultsappear to be insensitiveto specification.
ments. There is strong evidence that economies of scale in organizing exist for police and fire unions, and this pattern helps to explain whytreatingunionization as an endogenous variable lowers estimated union demand effects. The estimates imply that, all else equal, doubling departmentsize increasesunion densityby about 15 percentage points in police departmentsand 19 percentage points in fire departments, and both of these coefficientsare statisticallysignificantat the 1% level. Consistent with previous work, public sector labor laws and union success in the private sector also prove to be important determinantsof municipal unionization. Summaryand Conclusion This study has reexamined recent empirical findings of a positive association between public sector unionization and municipal employment. Some analysts have interpreted this correlation as evidence that public employee unions successfullyexert political pressure to raise the demand for municipal services.Alternativeexplanationsare available, however, includingthe possibilityof reverse causality due to economies of scale in union organizing. Using 1980 data for a large sample of U.S. cities, least squares regressions confirm previous work by showing that unionizationis associated withboth higher wages and increased employmentamong police and fireworkers.Instrumentingfor unionization reverses the directionof the employmenteffects,however, and Hausman specificationtestsreject the assumption that unionization is an exogenous determinant of municipal employment. Thus, there is evidence that simultaneity bias contaminates previous estimates of positiveemploymenteffectsby municipal labor unions. Using structuralestimatesof labor demand and the determinantsof police and fire unionization, I have attempted to separatelyidentifyunion effectson labor demand and possible economies of scale in
PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT
177
Table 5. Joint Determination of Employment and Unionization. (Standard Errors in Parentheses) Fire
Police Variable
log(Employment) Unionization
Predictedlog(Employment) Predictedlog(Salary) PredictedUnionization log(Population) log(Density) Female Black Under Age 18 Age 65 and Over log(Income) log(Grants) log(House value) Renter
log(Employment)Unionization
.152***
- .241 (.372) .081 (.196) 1.065*** (.028) - .032 (.029) - .222 (1.184) .443*** (.153) .602 (.495) 1.863*** (.483) .713*** (.149) .072*** (.026) -.324*** (.096) 1.130*** (.198)
Manager Commission log(Salary in retailtrade) Duty-to-Bargain PrivateSector Unionization Change in PrivateUnionization Right-to-Work
(.058) .088 (.172)
-
.134* (.071)
- 1.449*** (.526) .049 (.342) 1.046*** (.035) .040 (.040) 3.300** (1.555) .207 (.222) .116 (.761) 1.795*** (.667) .793*** (.200) .168*** (.036) -.388*** (.144) 1.364*** (.274)
.051* (.030) .041 (.062) -.086 (.112) .171*** (.033) 1.029*** (.258) -.558*** (.167) -.110** (.045)
- .110*
- .075*
-.073
South
- .044
- . 149***
- .053
West
- .046
- .008
Intercept R2
(.087)
(.072) - 8.243*** (2.643) .844
(.044) (.057)
(.05A) .938 (1.511) .375
- .159*** (.057)
- .006 (.032) .101 (.065) -.131 (.116) .162*** (.033) .588** (.295) -.913*** (.186) - .119** (.048)
North-Central
(.061)
.187** (.044) .169 (.214)
(.085) (.116) .293***
(.107) .637 (3.513) .771
.057
(.050) - .106*
(.060) .074
(.063) .731 (1.762) .339
Note: Sample sizes are the same as in Table 2. Instrumentsfor (log) employment,(log) salary, and unionizationinclude all other independentvariablesin the employmentand unionizationequations,as well as percent high school graduates, percentcollege graduates,and the civilianunemploymentrate. * Statistically significantat the .10 level: ** at the .05 level; *** at the .01 level (two-tailedtests).
178
INDUSTRIAL
AND LABOR RELATIONS REVIEW
union formation.There is strongevidence that economies of scale in union organizing are at least partlyresponsible for any positive association between public sector unionization and municipal employment. Because previous estimates ignore the interdependence of unionism and employment,they overstate the positive ef-
fectof unionizationon labor demand, and therebyexaggerate the amount of political clout wielded by municipal labor unions. Unfortunately,the instrumentalvariables estimates of union demand effects presented here are measured too imprecisely to justifyany definitiveconclusions about the magnitude of the bias.
Appendix Reduced Form Employmentand Salary Regressions: OLS Estimates (Standard Errors in Parentheses) Police Variable Unionization log(Population) log(Density) Female Black Under Age 18 Age 65 and Over High School College log(Income) log(Grants) log(House value) Renter Manager Commission Unemployment log(Salary in retailtrade) Duty-to-Bargain North-Central South
Employment .122** (.053) 1.051*** (.022) -.046* (.028) - .170 (1.152) .365** (.146) .587 (.571) 1.874*** (.549) -.073 (.356) .208 (.314) .619*** (.147) .071*** (.027) - .341*** (.087) 1.074*** (.202) - .012 (.038) .009 (.087) .308 (.881) .180 (.157) -.012 (.045) - .108* (.062) .021 (.072)
Fire Salary .108*** (.020) .058*** (.009) .022** (.011) - .468 (.442) .144** (.056) - .304 (.219) -.251 (.210) -.066 (.137) -.286** (.120) .194*** (.056) -.018* (.010) .150*** (.034) -.003 (.077) .075*** (.014) -.021 (.033) .277 (.338) .085 (.060) .073*** (.017) .035 (.024) - .117*** (.028)
Employment .367*** (.059) .943*** (.027) -.033 (.034) 2.301* (1.381) .021 (.175) - .224 (.702) 1.560** (.655) -.176 (.438) .339 (.391) .514* * (.179) .227*** (.033) - .591*** (.106) 1.226*** (.256) - .034 (.045) .026 (.101) - 1.477 (1.038) .062 (.192) -.340*** (.054) - .272*** (.076) .018 (.088)
Salary .066*** (.019) .058*** (.008) .028*** (.011) - .052 (.432) .132** (.055) .073 (.220) .014 (.205) .020 (.137) - .299** (.122) .185*** (.056) -.022** (.010) .169*** (.033) .010 (.080) .060*** (.014) .001 (.032) .781** (.325) .070 (.060) .070*** (.017) .068*** (.024) - .077*** (.027) (Continued)
PUBLIC SECTOR UNIONS AND MUNICIPAL EMPLOYMENT
179
Appendix(Continued) Fire
Police Variable
Employment
Salary
Employment
Salary
- .104 .053* -.071 .126*** (.031) (.080) (.031) (.098) - 10.936*** -7.130*** 4.841*** Intercept 5.270*** (.564) (1.486) (.570) (1.804) .847 .539 .810 .574 R2 Note:Samplesizesare thesameas in Table 2. * Statistically at the.10 level; ** at the.05 level; * I* at the.01 level(two-tailed tests). significant West
REFERENCES Bartel, Ann, and David Lewin. 1981. "Wages and Unionism in the Public Sector: The Case of Police." In Peter M. Mieszkowskiand George E. Peterson, eds., Public SectorLabor Markets.Washington,D.C.: Urban Institute,pp. 59-80. Benecki, Stianley. 1978. "Municipal Expenditure Levels and Collective Bargaining." IndustrialRelations,Vol. 17, No. 2 (May), pp. 216-30. Bennett, James T., and William P. Orzechowski. 1983. "The Voting Behavior of Bureaucrats: Some Empirical Evidence." Public Choice,Vol. 41, pp. 271-83. Brueckner, Jan K., and Kevin M. O'Brien. 1989. "Modeling Government Behavior in Collective Bargaining: A Test for Self-InterestedBureaucrats." Public Choice,Vol. 63, No. 1 (October), pp. 15-41. Eberts,Randall W., and Joe A. Stone. 1986. "On the Contract Curve: A Test of AlternativeModels of Collective Bargaining."Journalof Labor Economics, Vol. 4, No. 1 (January),pp. 66-81. Ehrenberg,Ronald G. 1973. "Municipal Government Structure,Unionization, and the Wages of Firefighters."Industrialand LaborRelationsReview,Vol. 27, No. 1 (October), pp. 36-48. Ehrenberg,Ronald G., and Joshua L. Schwarz. 1986. "Public Sector Labor Markets." In Orley Ashenfelterand Richard Layard, eds., HandbookofLabor Economics, Vol. 2. Amsterdam:North-Holland,pp. 1219-68. Freeman, Richard B. 1986. "Unionism Comes to the Vol. Public Sector."Journalof EconomicLiterature, 24, No. 1 (March), pp. 41-86. Freeman, Richard B., and Robert G. Valletta. 1988. "The Effectsof Public Sector Labor Laws on Labor MarketInstitutionsand Outcomes." In Richard B. Freeman and Casey Ichniowski,eds., WhenPublic Sector WorkersUnionize. Chicago: University of Chicago Press, pp. 81-106. Hausman, Jerry A. 1978. "Specification Tests in Vol. 46, No. 6 (NoEconometrics."Econometrica, vember),pp. 1251-71. Hirsch, BarryT. 1982. "The InterindustryStructure of Unionism, Earnings,and Earnings Dispersion." Industrialand LaborRelationsReview,Vol. 36, No. 1 (October), pp. 22-39. Hirsch, BarryT., and Mark C. Berger. 1984. "Union Membership Determinationand IndustryCharac-
teristics."SouthernEconomicJournal,Vol. 50, No. 3 (January),pp. 665-79. International City Management Association. 1984. The MunicipalYear Book,1984. Washington,D.C.: InternationalCityManagement Association. Kearney,Richard C. 1984. LaborRelationsin thePublic Sector.New York: Marcel Dekker. Lewis, H. Gregg. 1964. "Relative Employment Effectsof Unionism." AmericanEconomicReview, Vol. 54, No. 3 (May), pp. 123-32. McDonald, Ian M., and Robert M. Solow. 1981. "Wage Bargaining and Employment." American EconomicReview,Vol. 71, No. 5 (December), pp. 896-908. Montgomery, Edward. 1989. "Employment and UnemploymentEffectsof Unions."JournalofLabor Vol. 7, No. 2 (April), pp. 170-90. Economics, Nelson, Richard R. 1980. "State Labor Legislation Enacted in 1979." MonthlyLabor Review,Vol. 103, No. 1 (January),pp. 22-39. _ 1981. "State Labor Legislation Enacted in 1980." MonthlyLabor Review, Vol. 104, No. 1 (January),pp. 21-34. Pencavel,John, and Catherine E. Hartsog. 1984. "A Reconsideration of the Effects of Unionism on Relative Wages and Employment in the United Vol. States, 1920-1980." JournalofLaborEconomics, 2, No. 2 (April), pp. 193-232. Robinson, Chris, and Nigel Tomes. 1984. "Union Wage Differentials in the Public and Private Sectors: A Simultaneous Equations Specification." Vol. 2, No. 1 (January), JournalofLaborEconomics, pp. 106-27. Rubinfeld,Daniel L. 1987. "The Economics of the Local Public Sector." In Alan J. Auerbach and Martin Feldstein,eds., Handbookof PublicEconomics, Vol. 2. Amsterdam: North Holland, pp. 571-645. Spizman, Lawrence M. 1980a. "Public Employee Unions: A Study of the Economics of Power." Journalof Labor Research,Vol. 1, No. 2 (Fall), pp. 265-73. . 1980b. "Unions, Government Services, and Vol. 8, Public Employees." PublicFinanceQuarterly, No. 4 (October), pp. 427-42. Troy, Leo, and Neil Sheflin. 1985. UnionSourcebook. West Orange, N.J.: Industrial Relations Data and InformationService.
180
INDUSTRIAL
AND LABOR RELATIONS REVIEW
in U.S. Bureau of the Census. 198 la. CityEmployment 1980. Washington,D.C.: GPO. _ Relationsin State 1981b. Labor-Management 1980. Washington, D.C.: and Local Governments: GPO. . 1983. Countyand CityData Book.Washington, D.C.: GPO. U.S. Department of Labor, Labor Management Service Administration.1979. Summaryof Public SectorLabor RelationsPolicies. Washington, D.C.: GPO. Valletta,Robert G. 1989. "The Impact of Unionism on Municipal Expenditures and Revenues." Industrial and Labor RelationsReview,Vol. 42, No. 3 (April), pp. 430-42. Victor, Richard B. 1977. "The Effectsof Municipal
Unionism on Wages and Employment." Ph.D. diss., Universityof Michigan. Zax, JeffreyS. 1989. "Employmentand Local Public Sector Unions." IndustrialRelations,Vol. 28, No. 1 (Winter),pp. 21-31. Zax, JeffreyS., and Casey Ichniowski. 1988. "The Effectsof Public Sector Unionism on Pay, Employment,DepartmentBudgets, and Municipal Expenditures." In Richard B. Freeman and Casey Ichniowski,eds., WhenPublicSectorWorkers Unionize. Chicago: University of Chicago Press, pp. 323-61. . 1990. "Bargaining Laws and Unionizationin the Local Public Sector." Industrial and Labor RelationsReview,Vol. 43, No. 4 (April),pp. 447-62.