Public Sector Unions and the Costs of Government · Public Sector Unions and the Costs of...
Transcript of Public Sector Unions and the Costs of Government · Public Sector Unions and the Costs of...
Public Sector Unions and the Costs of Government
Sarah F. Anzia Goldman School of Public Policy University of California, Berkeley
and
Terry M. Moe Department of Political Science
Stanford University [email protected]
This Draft: August, 2012
Abstract: As recent political battles in Wisconsin, Ohio, and a number of other states attest, public sector unions are among the most active interest groups in American politics. They are also different from other interest groups in two key respects: they engage in collective bargaining, and are thus in a position to shape the organization of government in ways that other groups are not, and their members are the government’s own employees—its bureaucrats—who not only influence government from the inside through their official roles, but also from the outside through their unions. For all of these reasons, public sector unions are eminently worthy of scholarly attention, and yet political scientists have almost never studied them. This paper is an attempt to make some headway. Our focus is on how unions and collective bargaining in the public sector affect the costs of government. We present three separate studies, using different datasets from different historical periods, and we examine a range of cost-related outcomes: wages and salaries, health benefits, employment levels, and pension liabilities. In all three studies, our findings show that strong unions and collective bargaining do tend to increase the costs of government, and the impacts are both substantively and statistically significant. In presenting these findings, we hope to encourage other scholars to view public sector unions as important subjects of analysis.
1
For America’s public sector unions, the past few years have unleashed a perfect storm.
States and cities—which employ more than 90% of their members—are in financial crisis,
starved for revenues and forced to cut and save. Public pensions are underfunded by roughly
three trillion dollars, heightening demands for retrenchment. And Republicans, propelled by
huge gains in the 2010 elections and with Tea Partyers in their midst, are seeking remedies
through unprecedented cut-backs to collective bargaining rights for public workers.
In 2011 Wisconsin became ground zero in a battle so intense that Americans throughout
the country literally watched it unfold over a period of weeks on the nightly news. Led by
Governor Scott Walker, the state legislature weathered demonstrations by some 100,000 people
to enact sweeping reforms that weakened public sector bargaining. Ohio Republicans followed
suit (although the bill was later overturned via a union-led ballot measure). New Jersey, under
Republican Governor Chris Christie, prohibited public sector bargaining over health benefits.
Republican-controlled governments in Michigan, Indiana, Idaho, and Tennessee focused on the
teachers unions, whose 4 million-plus members make up more than half of all public sector
union members in the country, and severely limited collective bargaining in public education.1
The arguments on both sides were familiar ones. Republicans claimed that collective
bargaining increases governmental costs, especially via outsized health and pension benefits, that
restrictive work rules (such as seniority provisions) undermine effective organization—and thus
that bargaining should be restricted. Democrats countered that collective bargaining is a
fundamental right, that public workers are underpaid, that all workers should have the kinds of
1 See, e.g., Steven Greenhouse, “Ohio’s Anti-Union Law Is Tougher than Wisconsin’s,” New York Times, March 31, 2011; Angela Delli Santi, “New Jersey Anti-Union Bill Approved, Governor Christie Signs into Law,” Huffington Post, June 28, 2011; Richard Locker, “Tennessee Legislature Ok’s Ban of Teacher Bargaining,” The Commercial Appeal, Memphis, TN, May 20, 2011;.
2
pensions and health benefits that governments tend to provide—and thus that bargaining should
be valued and protected.
Looming above these arguments was a political reality that heightened the stakes
considerably. Public sector unions are a bulwark of the Democratic Party, and collective
bargaining is the unions’ power base, providing members, money, and activists. When
Republicans weaken collective bargaining, then, they are weakening the Democratic Party.
Labor reform is not just about costs or effective government. It is a political game-changer.2
For political scientists, public sector unions raise issues of far-reaching importance.
What is the impact of collective bargaining on the costs of government? What are its impacts on
government organization and the effective performance of public services? What are the
connections between union power, Democratic strength, and the substance of public policy?
These sorts of questions couldn’t be more basic. They also put the spotlight on interest
groups that are uniquely worthy of attention—for the unions are not only powerful, but also
different in key respects from other groups. They are different because they wield power in
collective bargaining as well as politics. And they are different because, rather than representing
interests that arise outside of government, they represent the interests of the government’s own
employees.
Political science, however, has little to say about public sector unions and their
consequences for American government. There is a large research literature on American
interest groups, but it pays little attention to public sector unions (Cigler and Loomis, 2011).
There is also a large literature on American bureaucracy, but it centers on the official roles that
bureaucrats play in carrying out public policy; and to the extent it deals with power, its focus is
2 See, e.g., Moe (2011) for comprehensive evidence on the connections between the teachers unions and the Democratic Party. More generally, see West (2008) and Troy (1994).
3
on the information asymmetry (arising from expertise) that gives them leverage with political
superiors (Epstein and O’Halloran, 1999). The power that bureaucrats exercise outside their
official roles through public sector unions—and their massive political spending, armies of
political activists, and finely tuned organization in elections and lobbying—is barely reflected in
the discipline’s understanding of what bureaucrats do to shape government, politics, and policy.3
New research is clearly needed along many fronts. In this paper, which is just a
beginning, we focus on one fundamental issue: the impact of public sector unions and collective
bargaining on the costs of government. To explore it, we present three separate studies based on
different data sets from different historical periods. The first uses data from 1972 through
1987—when many cities were getting collective bargaining for the first time—to explore the
impact of bargaining on city payrolls for police and fire departments. The second uses more
refined data from 1992 through 2010 to explore whether cities with collective bargaining had
higher wages, health benefits, and employment levels for police officers and firefighters than
other cities. And the third is a study of the current crisis in public sector pensions, which uses
financial data from 2009 to explore whether states with stronger unions are beset by more severe
pension problems.
Our findings across all three studies show that strong unions and collective bargaining do
tend to increase the costs of government. And as we discuss, this is what ought to be expected
theoretically. But much more ground needs to be covered—on costs, organization, politics—if
the role of public sector unions in American government is to be well understood. Our hope is
that the research we report here will not only begin to fill a yawning gap in the literature, but will
also encourage other scholars to view public sector unions as important subjects of analysis.
3 For exceptions, see Carpenter (2001), Johnson and Libecap (1994), and Moe (2006, 2011).
4
Background
During the 1950s, private sector unions were in their heyday. Aided by the National
Labor Relations Act (NLRA), they had organized 35% of the private workforce and seemed
headed for more. In the public sector, by contrast, labor was getting nowhere. Collective
bargaining was almost nonexistent—indeed, often illegal—and few workers belonged to unions.
As the decade came to a close, however, big changes were afoot, propelled by the newly
powerful alliance between organized labor and the Democratic Party, whose joint interest lay in
unionizing the public sector.
In 1959, Wisconsin became the first state to adopt a public sector collective bargaining
law (fashioned after the NLRA), and during the 1960s and 1970s most states followed suit. The
result, during those decades, was an explosion of organizing among teachers, police officers,
firefighters, nurses, bus drivers, and other public workers, and the expansion of collective
bargaining to state and city governments outside the South (which remained resistant).4
By the early 1980s, union membership had soared to 37% of the public workforce, where
it stabilized in an equilibrium that still prevails. In the meantime, unions in the private sector—
beset by rising competition, globalization, and structural change in the economy—were in
decline. They had peaked in the 1950s, and in the years that followed their membership
plummeted in a perpetual downward spiral to just 7% of the private workforce in 2010. Today,
although public employees make up just 17% of the total labor force, they constitute more than
half of all union members, and the leadership of the union movement—in politics, in the
Democratic Party, in collective bargaining—rests with public sector unions.5
4 On the rise of public sector unions, see, e.g., Freeman (1986); West (2008); Kearney (2009), and Troy (1994). 5 For figures on union membership after 1983, see www.unionstats.com, whose data are compiled by Barry Hirsch and David Macpherson from the Current Population Survey. For earlier data, see Troy and Sheflin (1985).
5
When these unions burst onto the American scene decades ago, economists rushed to
study them (and political scientists did not). For a short time the literature was lively and
promising. Yet the excitement was largely over by the early 1990s, and research waned. The
perspective from today is that, on the whole, this body of work is quite dated in most respects.
And quite incomplete.6
Theory
In this paper, we explore one aspect of public sector unions: their impact on the costs of
government. The most basic expectations that guide the analysis are straightforward. It is clear
that public sector unions, like those in the private sector, seek higher wages, better benefits, and
job protections for their members—and that, to the extent they can wield a measure of power,
they should tend to increase these components of governmental costs. Their power may vary, of
course, and they cannot always get what they want. But because unions are able to mobilize
money and manpower and wield credible threats in ways that the unorganized obviously cannot,
there is surely good reason to believe that workers will exercise greater power when they are
unionized and have collective bargaining—and that their clout will result in higher costs.7
The bigger picture is that the power and ultimate impacts of public sector unions—on the
organization and operation of government generally, not just on costs—are deeply rooted in
politics and the governmental context. In our own project, it has been quite a challenge even to
get good data on unions across cities and time, and as a practical matter it has simply not been
possible to delve into the distinctive political processes, calculations, and pressures at work in
6 For a recent review, see Kearney (2009). 7 It is possible that unions increase wage and benefit costs per worker, but that they also increase productivity. Productivity is often difficult to measure in government settings, though, and the literature is sparse and inconclusive about how it is affected by unionization. In a review that covers both public and private sector studies, although the literature is heavily weighted toward the latter, Hirsch (2004, p. 429) notes that positive union effects “are largely restricted to the private, for-profit sectors. Notably absent are positive productivity effects for school construction, public libraries, government bureaus, schools, and law enforcement.”
6
each of these hundreds of cities. This is the kind of detailed work that remains for the future.
But even though our tests cannot directly explore the political foundations of union power, we
think it is helpful to preface our analysis with at least brief attention to what they are and what
they imply. Here are some of the essentials.8
A key feature of the governmental context is that the “employers” are elected officials.
This means that, to the extent unions can wield political power in elections, they can literally
select the employers they will be bargaining with and who will be making decisions about
funding, programs, and policy. Unlike in the private sector, then, management is not
independent of the unions. Indeed, many public officials have incentives to promote collective
bargaining, give in to union demands, and protect jobs even if they know the result will be higher
costs and inefficiencies.9
The governmental context is also noncompetitive. If governments are burdened by high
labor costs or ineffective work rules, agencies won’t go out of business or lose jobs, money, or
functions to competitors—because there typically aren’t any competitors. The functions they
perform, moreover, from public safety to education to mass transit, are often so critical to
citizens that the threat of service disruption gives the unions still more leverage.10
The governmental context thus confers advantages on unions. But it also has its
downsides. State and city budgets are often highly constrained, for example, and citizens averse
8 The economists who contributed to the early research on public sector unions were well aware of the politics of what they were studying, and the elements we highlight below are common points of theoretical departure in that literature. See, e.g., Freeman (1986), Wellington and Winter (1971); Courant, Gramlich, and Rubinfeld (1979); DiLorenzo (1984); Bush and Denzau (1977); Hunter and Rankin (1988); Bellante and Long (1981). 9 For a recent development of this argument and its implications for the usual principal-agent approach to the understanding of bureaucratic power, see Moe (2006). 10 It is true, of course, that citizens can move from city to city or state to state to get better governments, and that business can do the same; but this kind of Tiebout competition within government is muted by massively high transactions costs, and it is unlikely to discipline decisions and inflict consequences in the same way the economic marketplace does—especially given the short term time horizons of most public officials. For a review of the Tiebout literature and a discussion of Tiebout inefficiencies, see Epple and Nechyba (2004).
7
to raising taxes, which makes it difficult for unions to squeeze big wage increases out of
government operating budgets—especially given that strikes by public sector workers are usually
illegal (although in practice this is sometimes ignored). Also, budgets and wage settlements—
which determine the cost outcomes that are the focus of this paper—are intensely political and
open to public scrutiny: which means that large wage gains, even if possible, can set off political
shock waves that even friendly politicians may be eager to avoid.
These downsides set the stage for tradeoffs. When governments can’t meet wage
demands, whether the reasons are financial or political, they can compensate unions in other
ways that are neither expensive nor visible. Notably, they can compensate them through labor-
friendly work rules (e.g., job protections, seniority provisions) and health and pension benefits.
Consider the tradeoff between wages and benefits. Because benefit costs have
historically been only a fraction of wage costs, and because they tend to be much less visible to
the public and more difficult to understand, it has long been cheaper and politically easier for
governments to increase benefits than wages. These incentives have lessened in recent years as
health and pension costs have soared, but traditionally the incentives have been stacked in favor
of benefits. A second point is that public officials have strong incentives to win union favor by
promising benefits that will mainly be paid for in the future, have little or no impact on current
budgets, and ultimately be the responsibility of other politicians. This is the case for pensions as
well as retiree health benefits.11
Finally, there is the tradeoff between compensation and employment. Higher
compensation puts downward pressure on employment: as labor becomes more costly, less of it
will be purchased. Through political action, the unions can try to increase compensation and
11 On the political attractiveness of promising future benefits, see especially Hunter and Rankin (1988); Bartel and Lewin (1981); Bellante and Long (1981); and Ichniowski (1980).
8
employment—by pushing for bigger budgets, say. But even when this is possible, there is still a
tradeoff. On the one hand, for any given budget, existing union members can be better paid if
there are fewer workers to be compensated. So the unions have incentives to keep their numbers
down and limit supply. On the other hand, higher employment means more members and money
for the unions, and thus more power—which they can use to push for higher compensation (and
other goals). So in this respect the unions have incentives to emphasize higher employment,
even if it means lower pay raises per member. How unions should balance these conflicting
incentives is unclear, and various strategies are presumably rational. It is reasonable to suggest,
however, that the unions will be under pressure from their existing members to focus on
compensation, and that there will be much less pressure to increase employment, whose benefits
to existing members are indirect and uncertain.12
Research
Empirical research on the impacts of public sector unions and collective bargaining was
in vogue during the 1970s and 1980s, but then tailed off during the 1990s. Almost all of this
early work was carried out on governments—usually city governments, sometimes departments
within them—and sought to explore union impacts on wage and salary expenditures, budgets,
employment, and other measures of government activity (see Kearney, 2009, for a review).
There is also a more recent, more general literature on employee earnings—one that has
not tailed off—that uses national surveys (such as the Current Population Survey) of individuals
to estimate the wage premiums associated with union membership in both the public and private
sectors. This work does not tie individuals or unions to specific units of government, and so
does not explore the governmental issues we aim to address here. That said, the wage premium
12 On how unions approach employment levels as opposed to compensation, see, e.g., Zax (1989); Zax and Ichniowski (1988); Freeman and Valetta (1988); Trejo (1991); Valetta (1993).
9
literature does show (among other things) that comparable public employees—those with the
same employment-relevant characteristics—do tend to have higher earnings when they are
members of unions (e.g., Gregory and Borland, 1999; Blanchflower and Bryson, 2004; Bahrami,
Bitzan, and Leitch, 2009; Bitzan and Bahrami, 2010).
This literature dovetails with the research on governments, which shows that government
expenditures on wages and salaries tend to be higher (per capita, per worker) due to unions (e.g.,
Ashenfelter, 1971; Ichniowski, 1980; Ehrenberg, 1973; Gallagher, 1978; Trejo, 1991; Zax, 1989;
Zax and Ichniowski, 1988). Whether the units of analysis are governments or individuals,
however, the study of union impacts on compensation is almost always limited to employee
pay—with no attention to fringe benefits. Only a few studies take benefits into account, because
benefits are much more difficult to measure and good data sources are elusive; but such studies
suggest that public sector unions have much bigger impacts on benefits than wages (Ichniowski,
1980; Bartel and Lewin, 1981; Hunter and Rankin, 1988; Zax, 1988). Thus, it appears that by
focusing on earnings alone the literature underestimates the impact of unions on total
compensation. All the more so because public employees tend to have benefits that are more
valuable and costly than their counterparts in the private sector. To ignore benefits is to ignore
much of what public employees are actually “paid,” as well as what the impact of their unions
actually is.13
On other basic counts, the literature also fails to arrive at clear conclusions. Some
studies, for example, show that unions lead to higher government spending overall; others show
that spending is higher in unionized departments but lower in nonunionized departments, with no
tendency for governments to spend more in total (Zax and Ichniowski, 1988; Marlow and
Orzechowski, 1996; Valetta, 1988). Studies of employment levels lead to ambiguous results as 13 See, e.g., the discussion and analysis in Biggs and Richwine (2011).
10
well. Some show that unions bring about higher levels of government employment, while others
show that they have no impact on employment at all or even a negative impact (Zax, 1989; Zax
and Ichniowski, 1988; Freeman and Valetta, 1988; Trejo, 1991; Valetta, 1993).
These were pioneering studies, and they were on the right track in exploring how unions
affect government. But the literature lost its momentum and largely petered out before even the
most basic questions were answered with confidence. It is up to today’s scholars to revisit these
issues, build on what the early studies were able to achieve, and breathe new life into a moribund
research enterprise.
Our purpose here is to contribute toward that effort by presenting three studies of union
impact on the cost of government. These studies go beyond the existing literature in important
respects: they are based on better measures of key variables (collective bargaining, employee
benefits), they recognize endogeneity issues (which have to do with why some governments have
collective bargaining and some do not), and they introduce new and more modern data from the
1990s and 2000s that helps to bring the literature up to date.
These studies help to move the ball downfield, yet we have no illusions that they are
definitive. There are difficult data and methodological problems that stand in the way of clear,
confident findings, and we have no perfect solutions. In our view, this exercise in empirical
analysis is as much about the journey as it is about where we ultimately wind up—for the
journey puts the spotlight on problems and issues that are fundamental to this field of study, and
that political scientists (and economists) will need to deal with if research on public sector unions
is to be resurrected and moved ahead.
11
Unionization and City Wages, Employment, and Payroll in the 1970s and 1980s
We begin by focusing on a time period in which public sector employees were first
securing collective bargaining rights: the 1970s and early 1980s. This is arguably the best
context for estimating the causal effect of unionism on governments’ finances since we can
examine the conditions of the same governments before and after their employees unionized.
Thus, unlike a study using more current data, which would have to rely on cross-sectional
variation to estimate the effect of public sector unions, our analysis of historical data enables us
to leverage within-government variation in the union status of public sector employees over time.
In turning to data collected in the 1970s and 1980s, we are in some ways revisiting
territory that was explored by scholars thirty to forty years ago – scholars who grappled with two
of the biggest challenges to estimating the causal impact of public employee organization:
measurement and endogeneity. The measurement challenge is simply that data on public sector
unions are scarce, and the data that do exist have a number of problems (see Freeman,
Ichniowski, and Zax, 1988). As we describe below, we largely adopt the earlier literature’s
conventions for dealing with measurement problems by using data from the U.S. Census of
Governments and managing their shortcomings in a similar fashion. However, our handling of
the potential endogeneity issues improves upon existing work, producing better estimates of
public sector unionization’s effects on cities’ wages, employment, and payroll expenditures.
The main endogeneity concern is that cities whose employees form unions and secure
collective bargaining are different from cities whose employees remain unorganized, and those
differences could be correlated with compensation and staffing levels. For example, large cities
are more likely to have unionized employees than small cities (e.g., Trejo, 1991), and they also
tend to pay higher wages. If we were to ignore the importance of city size, our estimates of the
12
effect of unionization on wages would be biased upward. Thus, to estimate the causal effect of
unionization, we have to partial out the effects of any city characteristics that influence both its
employees’ proclivity to organize and its compensation and employment practices.
The advantage of studying public sector unionization during the 1970s and 1980s is that
we can design an empirical analysis that substantially reduces the potential for omitted variable
bias by isolating governments’ conditions before and after their employees formed unions.
Specifically, in 1972, 1977, 1982, and 1987, the Census of Government conducted a special
Labor-Management Relations Survey, which included questions about whether governments had
collective bargaining and whether certain groups of employees were members of unions. By
assembling all of these data into a panel, we can estimate the effect of unionization by leveraging
within-government variation, partialling out the effects of any time-constant city characteristics
that could be a source of bias. This presents a tremendous opportunity to conduct clear causal
inference – one that cannot be replicated using data from later periods.
Surprisingly, most of the economic studies of the 1980s relied on cross-sectional data
rather than longitudinal data to estimate the impacts of public sector unionization and collective
bargaining (e.g., Brown and Medoff, 1988). Moreover, the few studies that did use longitudinal
data had a narrow temporal focus (e.g., four years) and did not have any within-unit variation in
their unionization measures, which prevented the inclusion of unit fixed effects.14 In fact, the
only study we are aware of that puts the full 1972-1987 Census of Governments panel together to
conduct a within-unit analysis of the effect of unionization is one by Hoxby (1996), whose focus
is solely on the impact of teacher unionization on school district outcomes.
14 For example, the city department-level data analyzed by Freeman and Valletta (1988) and Zax and Ichniowski (1988) only spanned 1977 to 1980 (and Zax and Ichniowski include 1982). Moreover, their measures of collective bargaining did not vary within departments over time, so they could not include city department fixed effects.
13
Thus, our goal in this initial empirical study is to use variation in unionization within
cities over time to estimate the impact of unions on cities’ average wages, staffing levels, and
payroll expenditures. In particular, we focus on two groups of employees that make up a large
percentage of overall city employment: fire protection employees and police protection
employees. Importantly, because our data allow us to include city-level fixed effects, we
eliminate any sources of omitted variable bias that are constant within cities over time.
Of course, there are other potential sources of endogeneity that must also be
acknowledged and addressed. For example, it could be that wages and staffing levels influence
whether a city’s employees get organized. The direction of the bias in that case would likely be
negative, however: public sector employees are probably more motivated to form unions when
wages and employment are low. Nonetheless, the more general concern is that there may be city
characteristics that vary over time that influence both its employees’ propensity to unionize and
its wages and employment—and to the extent that such factors exist, we must incorporate them
into our models. In the following section, we describe our data and models, which explicitly
account for the most likely sources of bias.
Data and Empirical Strategy
Our data come from the U.S. Census of Government Public Employment Files from
1972, 1977, 1982, and 1987,15 which contain the data from the aforementioned Labor-
Management Relations Surveys as well as the regular government employment and payroll
information collected during each quinquennial census. In each of the years, municipal
governments reported to the Census how many of their police and fire protection employees
were members of employee organizations. Because there is a significant degree of measurement
error in these figures, we follow Freeman, Ichniowski, and Zax (1988) in creating dichotomous 15 We downloaded these files from the Interuniversity Consortium for Political and Social Research.
14
measures of unionization. We code a city as having an organized fire (police) department if at
least some of its fire (police) protection employees are in unions.
Dichotomizing the variable does not fully address the issue of measurement error,
however. Examining the data, we find that a number of cities reported that their fire or police
employees were unionized in one year but not in one or more subsequent years. Fortunately, this
too is a pattern that Freeman et al. recognized and addressed. And when they conducted
telephone interviews with 258 of the governments in their data that, according to the Census, had
lost (or lost and regained) collective organization, they found that not a single one had actually
lost it. In every case, city employees had either organized and stayed organized or they had
never organized at all. Most of the time, it was the former.
To minimize the effects of these reporting errors, we therefore made minor adjustments
to our coding of the fire and police organization variables by examining within-city patterns over
time. Cases that were sufficiently ambiguous were dropped. A full description of our coding
decisions is in the online appendix, along with a sensitivity analysis.
Because we want to focus on cities large enough to house their own police and fire
departments, and because we want to make this analysis consistent with the study we present in
the next section, we limit our focus to cities that had at least 10,000 residents as of 1972. This
gives us a dataset of 1,689 city police departments and 1,400 city fire departments tracked from
1972 to 1987 at five-year intervals.16 In total, 368 of the police departments and 241 of the fire
departments first became unionized over the course of this time period.
16 For the fire protection analysis, we drop 59 municipalities for which full-time fire protection payrolls and employment were 0 for at least one year. For the police protection analysis, we drop 8 municipalities for which full-time police protection payroll and employment were 0 for at least one year. There are also a few cities for which the values of one or more of the dependent variables change dramatically over a five-year period. Since these are almost certainly errors in the data, we exclude cities for which any dependent variable more than triples over a five-year time period (37 cities for the fire analysis and 4 cities for the police analysis). Lastly, for the models of average wages, we lose a number of city-years due to the inclusion of the competition variable that we describe below.
15
For both fire and police protection employees, we analyze three dependent variables:
average wage, employment per capita, and payroll expenditures per capita. We adjust average
wage and payroll figures to 1987 dollars, and we take the log of each dependent variable to
reduce skew in their distributions. Our model of each outcome is as follows:
ln
Subscript i denotes the city, and t denotes the year. The αi are city fixed effects, the δt
are year fixed effects, β and ψ are regression coefficients, and εit is an error term. The variable
unionit is a binary indicator variable equal to one if the employee group is organized, and so the
regression coefficient β is the average effect of the treatment (unionization) on the treated. We
use ordinary least squares to estimate the models, and we cluster the standard errors by city to
correct for autocorrelation with cities over time.
Xit is a matrix of time-variant control variables constructed using data from the 1970,
1980, and 1990 U.S. Censuses of Population with values interpolated for 1972, 1977, 1982, and
1987. As we show in the online appendix, the cities where police officers and firefighters
formed unions were different than the cities where they did not: they were larger in population,
higher in socioeconomic status, and had lower percentages of African Americans and Hispanics
than cities that never unionized. They also had more adults employed in manufacturing, lower
poverty rates, smaller percentages of the population enrolled in elementary and high school, and
lower rates of population growth. Because we suspect that these correlates of unionization might
also be associated with wages, payroll, and employment, we control for the following in our
models: the natural log of city population, population growth, socioeconomic status (an average
of logged per capita income and the percentage of city residents with a college degree), percent
16
African American, percent Hispanic, percent living in poverty, percent enrolled in elementary or
high school, and the percentage of employed adults who work in manufacturing.17
It is also possible that city officials consider the pay rates of surrounding cities in
deciding how much to pay their workers. If, for example, a city were to discover that its wages
were lower than the wages of similar cities in the area, perhaps it would increase its wages in the
next year to avoid losing its employees to nearby cities. Of course, we don’t actually know that
cities compete for public sector workers in this way, and city officials probably also see
advantages to keeping employee compensation costs low.18 However, even if there is no such
“competition effect,” there may well have been a “threat effect” in the 1970s and 1980s.
Specifically, if officials in nonunionized cities with unionized neighbors increased their wages to
avoid the dissatisfaction – and potential unionization – of their employees, the result would be a
tendency toward the equalization of wages across union and nonunion cities.
In our models of police and fire protection average wages, we adopt the following
strategy to allow for these effects: For each state and each year of 1967, 1972, 1977, and 1982,
we regress average wage for all city employees on logged city population and logged city per
capita income.19 The residuals from those regressions become our measure of the extent to
which a city’s wages five years prior deviated from the wages of cities similar in size and cost of
living within the same state. We include this variable in our model to test whether city officials
compensate for having below-market wages in the previous period by increasing wages. If its
coefficient is negative, that would be evidence of a competition effect. In a second specification,
17 We exclude a small number of cities (6 for the fire analysis and 12 for the police analysis) that have very large five-year changes in population growth or percent in school since these are almost certainly errors in the data. 18 For example, it might allow them to keep taxes down, or it might free up resources for other budget items. 19 Average wage for police and fire protection employees was not available for 1967, but in the years 1972-1987, 40% of the typical city’s full-time payroll expenditures went to police and fire functions. Therefore, the average wage for all city employees is a reasonable proxy for police and fire average wages. The population figure for 1967 comes from the 1967 Census Public Employment File. We use 1970 per capita income (as reported by the U.S. Census) for 1967, adjusted to 1987 dollars.
17
we also interact the variable with the indicator for unionization. If we find that it is
predominantly the unorganized cities that increase their wages in response to having low wages
in the previous period, that would be evidence of a threat effect.
Empirical Results
The results from our analysis of the Census of Government data are set out in Table 1.
First, in columns 1 and 2, we ask whether the average wage of municipal fire protection
employees increased after they formed unions. Looking at the coefficients on the Union
indicator, it is clear that the answer is yes. On average, the effect of fire department unionization
was a 3.9% increase in the average wage of fire protection employees, an effect that is highly
statistically significant. Given that one of the main reasons unions form is to pressure for higher
wages, this effect is precisely what we should expect: it indicates that firefighter unions were
successful in increasing their members’ pay in the years shortly following their organization.
The results in column 3 demonstrate that the wage premium that accrued to unionized fire
protection employees did not come at the expense of fire department staffing levels, at least in
the short run. To the contrary, we find that per capita fire protection employment increased by
7.6% in cities where firefighters formed unions, an effect that is statistically significant at the 1%
level. And unsurprisingly, since unionization led to both increased wages and increased per
capita employment in fire departments, total per capita fire protection payroll expenditures
increased when firefighters formed unions. As we show in column 4, a city whose fire
protection employees organized for the first time could expect to spend nearly 11% more on fire
protection salaries and wages as a result. This effect holds above and beyond the effects of
national trends in firefighter wages, time-constant city characteristics, and time-varying city
characteristics like city size, socioeconomic status, racial composition, and poverty rates.
18
Columns 5 to 8 demonstrate that the effects of unionization on police departments’
wages, employment levels, and payroll expenditures were smaller in magnitude than those for
fire departments but still strong, positive, and statistically significant. In columns 5 and 6, we
find that police officers who formed unions saw their wages increase by about 2.3% as a result,
an effect that is significant at the 1% level. Police per capita staffing levels also increased within
cities where police organized, as we show in column 7: relative to national trends in police
department size, municipal police departments employed 2.3% more employees per capita after
those employees organized. Together, these increases in wages and employment resulted in an
average 3.7% increase in per capita payroll expenditures for police. Thus, in the years
immediately following unionization, police unions were successful in pressuring their municipal
government employers for better wages, higher employment, and an overall increase in the
amount cities spent on police compensation.
Most of our control variables behave as we expected. As cities get larger, the average
wages of public safety workers increase, but the number of public safety workers per capita tends
to decrease. This results in a negative effect on per capita payroll expenditures for police and no
effect for fire protection employees. Socioeconomic status is positively associated with wages,
employment, and payroll for public safety workers, and the opposite is true of poverty rates. The
racial composition of a city makes some difference to public safety wages as well: an increasing
percentage of African Americans is associated with slightly lower police wages and higher
police employment, while communities with larger Hispanic populations tend to pay higher
wages for both police and fire employees. Increasing reliance on manufacturing is associated
with higher police and fire employment and payroll expenditures but has an insignificant effect
on wages. Communities with more children in elementary and secondary school tend to employ
19
fewer police per capita and spend less on payroll, and city population growth tends to be
positively associated with all of the dependent variables.
Lastly, in our average wage models, we find evidence that city officials did react to
whether their wages were low or high relative to similar cities in the same state. In columns 1
and 5, we find that when a city’s average wage in the previous period was relatively low, city
officials responded by increasing police and fire protection wages. However, it is not clear from
these results whether all cities adjusted in this way – which would suggest a general competition
effect – or whether adjustments were mostly made by nonunion cities in response to the threat of
unionization. In columns 2 and 6, therefore, we interact the lagged deviation variable with the
indicators for union status. Column 2 shows that there was no significant difference between the
two types of cities’ adjustments to fire protection wages, and column 6 shows that if anything, it
was primarily the unionized cities that adjusted police wages upward in response to being lower
than average five years prior. These results suggest that it was a general competition effect at
work, not a threat effect.
Most importantly, though, on the question of how the unionization of municipal police
and fire departments affected their wages, employment, and payroll expenditures, our results are
very clear. Across the board, we find that police and fire unions were quite successful in
increasing wages, staffing levels, and expenditures on employee compensation in the 1970s and
1980s. These results are robust to a variety of alterations in the city sample and model
specification,20 and they should not come as a surprise. After all, they merely indicate that the
20 The results do not change substantively when we include state-year fixed effects, which can account for the independent effect of state-year-specific shocks, such as the passage of a state collective bargaining law. In addition, when we estimate these models using a smaller subset of cities for which we did not make corrections to the unionization variable, our results are the same. When we eliminate the city fixed effects and control only for Census region, our estimates generally increase in magnitude and remain statistically significant. See the online appendix for details.
20
unions that organized city employees were successful in achieving some of their most important
goals. The contribution we make in demonstrating these relationships is in the methodology we
use to generate the estimates: we estimate the effects using within-city variation in unionization
over time, which allows us to rule out the possibility that time-constant characteristics of cities
are driving the effects. This is a significant improvement over the existing studies that generate
estimates from cross-sectional data.
Yet, while examining this critical period in history is the only way to observe cities
before and after their employees unionized, focusing on the 1970s and 1980s has its
disadvantages. Most importantly, we have no way of knowing how much cities spent on non-
salary forms of compensation at the time. If anything, we expect the effects of unionization were
even more pronounced for fringe benefits like health insurance and retirement packages, which
means that our estimates in Table 1 are probably lower bounds on the effects of unionization for
overall employee compensation. Moreover, the results in Table 1 illustrate the effect of
unionization as of 25 to 40 years ago. Clearly, we also want to know how unionization impacts
the cost of government today. In the following sections, therefore, we carry out two additional
studies in which we use more current data that include information on public sector employees’
health benefits and pensions.
Collective Bargaining and Cities’ Expenditures on Salaries and Benefits, 1992-2010
By the mid-1980s, the rapid wave of public sector unionization had subsided, and
American government had settled into a new equilibrium. For the most part, the groups of public
employees that were going to unionize had already done so, and cities without unions were to
remain without them. In this section, we examine the contours of this new equilibrium,
investigating the consequences of public sector collective bargaining for government in the
21
1990s and 2000s. To do this, we have assembled a rich new dataset on the collective bargaining
status, employment levels, and compensation practices of police and fire departments in
American municipal governments. Not only is this dataset more current than the Census of
Government dataset, allowing us to evaluate the longer-term impacts of public sector
unionization, but it also contains data on cities’ expenditures on employees’ health, hospital,
disability, and life insurance benefits. Thus, this second study gives a more complete picture of
how employee organization shapes governments’ employee compensation costs.
Of course, as we noted above, any study that relies on recent data to examine the impact
of public sector unionism must deal with a considerable challenge: very few governments
adopted collective bargaining for the first time after the 1980s. That means that the only
variation in unionism that can be exploited in the current period is cross-sectional; the
independent variable of interest does not change within cities over time. This is an interesting
problem that history has created. And it makes it all the more important that we (and anyone
studying public sector unions) find ways to make appropriate comparisons between cities with
and without collective bargaining. Fortunately, our analysis of the 1970s and 1980s gives us
some sense of what we should expect. But that was 25 to 40 years ago. Clearly, it is also critical
that we understand the difference public sector unions make for governments today.
Data and Empirical Strategy
The city staffing and compensation data for this study come from the annual Police and
Fire Personnel, Salaries, and Expenditures surveys conducted by the International City/County
Management Association (ICMA). Every year since 1992, ICMA has sent questionnaires to all
municipal governments with more than 10,000 in population to ask about the size and nature of
their police and fire departments, their personnel policies, and how much they spend on various
22
budget items.21 We have assembled all available years of data into a panel. Notably, since a
different set of municipalities responds to the survey each year, the panel has significant gaps,
with most municipalities appearing in the dataset in some years but not others.22
For both police and fire protection employees, we focus our analysis on five dependent
variables. First, we calculate the amount the department spends per employee on salaries and
wages, including base salaries as well as supplemental forms of pay like longevity pay, hazard
pay, holiday pay, and overtime pay. Second, we analyze the amount the department spends per
employee on health, hospital, disability, and life insurance benefits. As before, we also analyze
the total number of employees in each department per city resident. We then inspect the total
amount that municipalities spend on employees’ salaries per capita, as well as the total amount
they spend on health benefits per capita. We take the natural log of all the variables. A complete
description of how we assembled and cleaned these variables is available in the online appendix.
We have different numbers of observations for each dependent variable, but the maximum is
16,809 for police departments (in 2,243 unique municipalities) and 8,809 for fire departments (in
1,177 unique municipalities).
Our key independent variables are binary indicators of whether municipal police and fire
protection employees have collective bargaining, which we constructed using three groups of
data. The first is the Law Enforcement Management and Administrative Statistics (LEMAS)
surveys conducted by the Bureau of Justice Statistics in 1987, 1990, 1993, 1997, 2000, and 2003,
which asked U.S. law enforcement agencies whether their sworn police officers have collective
bargaining. Second, we rely on the Labor Management Relations surveys conducted by ICMA
in 1988 and 1999, which asked a series of questions about the collective bargaining status of
21 The sample of municipal governments includes cities, towns, villages, boroughs, and townships. 22 With a typical survey response rate of about 45 to 50%, the average number of observations per year is 1,420. The average city appears in the dataset in 8 of the 19 years.
23
various groups of municipal employees. Lastly, we use a special 1977 survey conducted by the
Census of Governments that documented whether certain groups of employees in each municipal
government were part of a bargaining unit. We describe how we combined these datasets and
coded the collective bargaining indicators in the online appendix.
Our strategy for estimating the impact of collective bargaining on our five measures of
municipal employment and compensation is similar to the one we used in the previous section.
The unit of analysis is again the municipality-year, and we estimate the impact of collective
bargaining using OLS with standard errors clustered by municipality.23 We include all of the
control variables from Table 1 as well as logged population density (since denser cities are more
likely to have organized employees and higher demand for public safety services) and logged
median rent in the city (to account for cost of living differences within and across cities).24 All
of the data for these variables come from the 1990, 2000, and 2010 Censuses and the 2005-2009
estimates from the Census’ American Community Survey, with values interpolated for each year.
In our models of per-employee salary and health benefits expenditures, we also include a control
for the competition effect we observed in the 1970s and 1980s, using a similar strategy.25
The main difference between our empirical strategy here and that of the previous section
is that because there are so few cities where police or firefighters got collective bargaining for
the first time after 1992, we do not include city fixed effects.26 We therefore include three
23 We also run each of our models using robust estimation to ensure that our estimates are not sensitive to potential outliers and leverage points. See online appendix. 24 Population growth in this study is percentage growth from 1990 to 2010. 25 Specifically, for all cities in the same state, we regress the dependent variable on log population and log median rent. The lag of the residuals from that regression are used as an indicator of whether cities’ salaries and benefits were below or above average for similar cities in the same state in the previous year. Not every city appears in the dataset in every year, so incorporating this lagged variable could result in the loss of many observations. To avoid this, if a city is missing a value in the previous year, we use the value from two years prior. If it is missing the value in both the year prior and two years prior, we use the value from three years prior, and so on. 26 For example, in our model of expenditures on fire protection employees’ health benefits, there are only 9 cities that changed from not having collective bargaining to having collective bargaining between 1992 and 1999.
24
additional sets of controls in all of our models that we consider to be important for explaining
between-city differences in collective bargaining norms and employment and compensation
policies. The first is a measure of cities’ political leanings, since more liberal, Democratic cities
are more likely to have unionized public sector workers and more generous compensation
policies: we control for the percentage of the two-party vote that went to Al Gore in the 2000
presidential election in the municipality’s parent county. Since some states just have more
worker-friendly cultures than others, we also control for the rate of private sector union
membership in each state and year using data compiled by Hirsch and MacPherson (2003, 2011).
Finally, to account for cross-regional differences in collective bargaining policies, employment,
and compensation, we include dummy variables for three of the four geographic regions in the
U.S. As in all of our earlier models, we include year fixed effects to explain secular variation in
public sector employment and compensation over time.
Empirical Analysis
We start with an analysis of municipal fire departments, the results of which are set out in
Table 2. In column 1, we estimate the effects of collective bargaining on the amount cities spend
per employee on salaries. The effect we estimate here is larger than the effect we estimated
using data from the 1970s and 1980s: on average, municipal fire departments with collective
bargaining spend about 9% more per employee on salaries and wages. Notably, however, the
differences between fire departments with and without collective bargaining are even greater
when it comes to per-employee expenditures on health, dental, disability, and life insurance. As
we show in column 2, fire departments with collective bargaining spend an average of 25% more
on those benefits for the typical employee. Relative to the average across cities, this amounts to
an extra $1,507 per employee per year in cities where firefighters have collective bargaining.
25
In column 3, we test whether these wage and benefit premiums come at the cost of fire
protection employment levels. We find that they do not: there is no significant difference
between per capita fire protection employment in cities with and without collective bargaining.
Thus, it comes as no surprise that cities where firefighters have collective bargaining spend more
overall on employee compensation, which we show in columns 4 and 5. In column 4, we find
that cities with collective bargaining spend 9% more per city resident on fire protection salaries.
And as column 5 shows, the differences are bigger for spending on health benefits, an area in
which cities with collective bargaining spend over 25% more.
These results clearly indicate that collective bargaining for fire protection employees has
a sizeable effect on the amount municipal fire departments spend on employee compensation.
Moreover, they show that it is absolutely imperative to take fringe benefits into account when
estimating the effect of collective bargaining on public employee compensation: that is the area
in which unions have secured the greatest gains over the years.
Of the remaining variables, there are a few that have significant effects. For example,
fire departments in larger cities spend more on employee salaries but also employ fewer fire
protection workers per capita. Cities with residents of higher socioeconomic status tend to
employ more fire protection employees, pay higher wages, and, as a result, spend more on their
salaries and health benefits. As expected, a higher cost of living is associated with higher per
employee salaries, as is greater population density, but higher cost of living is also associated
with fewer fire protection employees per capita. Our measures of liberalism and union-friendly
political culture generally have positive effects on fire protection employee compensation, as
does the percentage of employed city residents working in manufacturing. However, unlike our
analysis of the 1970s and 1980s, we find no evidence of any competition effect in this more
26
current dataset. In fact, the coefficient on the lagged deviation from the state average in the first
two columns is positive, suggesting that paying less than the state average in prior years is
associated with paying lower salaries in the current period.27
In Table 3, we find that the effects of collective bargaining in municipal police
departments are similar to those of fire departments. On average, cities where police have
collective bargaining spend over 10% more on salaries per police protection employee than cities
where police do not have formal bargaining rights. As with firefighters, the gap between the two
types of cities is even wider when it comes to expenditures on health benefits. The typical
municipal police department with collective bargaining spends about 21% more on health
benefits per employee – or about $1200 – than the typical city without collective bargaining.
Both of these positive effects are statistically significant at the 1% level.
In contrast to our findings for fire departments, however, we find that police departments
with collective bargaining operate at lower per capita staffing levels than non-bargaining
departments. Specifically, in column 3, we find that on average, per capita police employment is
5.8% lower in cities where police have bargaining rights. And because the salary and health
benefits premiums observed in columns 1 and 2 are partially offset by these lower employment
levels, the consequences of collective bargaining for cities’ total per capita expenditures on
police compensation are slightly more muted than for fire protection. In columns 4 and 5, we
find that cities where police have collective bargaining spend 4.3% more on salaries per capita
and about 16.5% more on health benefits per capita.
27 While there is little reason to expect that non-collective bargaining cities would feel the need to increase salaries in response to the threat of unionization in the post-1980s period, we do test for a threat effect as well, interacting the lagged deviation variable with the collective bargaining indicator. As the results in the online appendix show, we find no evidence of any threat effect: even for non-collective bargaining cities, the coefficient on the lagged deviation from the state average is positive – not negative.
27
The size, strength, and consistency of the empirical results presented in Tables 2 and 3
make it very clear that collective bargaining has powerfully shaped employment and
compensation practices in the cities where public sector employees have successfully organized
– and in the direction we expect. Of course, even though we have controlled for a rich set of
demographic variables, the partisan leanings of the cities, the union-friendliness of each state,
and regional and temporal trends, it remains possible – even if unlikely – that our design neglects
some important omitted variable that is correlated with collective bargaining and our dependent
variables. If so, our estimates of the effect of collective bargaining would be biased.
One way of addressing this potential problem would be to find an instrument – a variable
whose effect on compensation and employment works only through collective bargaining.
Indeed, instrumental variables (IV) regression is standard practice in political science and
economics. Yet, many studies that use IV regression provide little justification for the
assumption that the chosen instrument is uncorrelated with the error term of the model. In our
view, it is not uncommon for studies to use instruments that probably do not satisfy the exclusion
restriction. And if the exclusion restriction is not satisfied, IV estimates are inconsistent, and one
might be better off just using OLS.28 Thus, without a variable that we can be sure is exogenous,
we do not consider IV regression to be preferable to the OLS models presented here.
Instead, we use an alternative strategy to try to partial out important differences between
cities with and without collective bargaining. The problem is that certain cities are more inclined
to pay higher salaries for reasons we cannot observe, but in ways that could be correlated with its
28 One might think, for example, that state-level church attendance rates could serve as an instrument for collective bargaining, since cities in more religious states tend not to have collective bargaining, and since religion probably does not cause lower public sector wages. However, since church attendance could be correlated with conservatism, which we may not have measured perfectly in our model, it could still be correlated with the error term. As it happens, when we carry out IV regression using church attendance as an instrument, our estimates of the effect of collective bargaining become quite large – in our view, implausibly large.
28
employees’ propensity to unionize. If those cities have been so inclined for many years, even
before collective bargaining was an option, then we would not want to attribute those higher
salaries to the presence of collective bargaining. Therefore, in an attempt to partial out the
effects of any differences between bargaining and non-bargaining cities that predated the onset
of public sector bargaining in the 1960s, we add two new control variables to each model: cities’
total per capita payroll in 1957, and cities’ total fire (or police) employment per capita in 1957.29
Because the inclusion of these variables forces us to drop a non-negligible number of
observations,30 we leave the full presentation of the results to the online appendix. Most
importantly, though, the inclusion of these variables has only a small effect on our estimates of
the impact of collective bargaining. We still find that per-employee fire protection salary
expenditures are 8.5% higher in cities with collective bargaining, and per-employee fire
protection health benefits expenditures are 25% higher. For police protection employees, the
estimated effects are 9.4% and 21%, respectively. We estimate that in total, cities with collective
bargaining spend 21% more per capita on fire protection health benefits and 18% more on police
health benefits. The only substantive difference is that the estimated effect of collective
bargaining on fire protection employment is negative using this approach, and so its effect on per
capita fire protection salary expenditures is indistinguishable from zero. For police protection
employees, we still estimate a 4.5% effect of collective bargaining on salaries per capita.
In sum, the cities where public sector employees secured collective bargaining during the
1960s, 1970s, and 1980s have progressed along a markedly different path than the cities whose
employees never pursued or won bargaining rights. Municipal police and fire departments with
29 Data come from the 1957 Census of Governments. 30 Many of the municipalities in our 1992-2010 dataset did not exist in 1957. Moreover, the 1957 Census of Government only tracked these variables for cities with more than 5,000 people, which also eliminates a number of cities in our dataset.
29
collective bargaining spend significantly more on their employees’ salaries than similar
departments without collective bargaining. In police departments, that salary premium has come
with slightly lower per capita employment levels. But most importantly of all, we find that the
biggest gap between bargaining and non-bargaining cities is in the area of health benefits
expenditures. When it comes to health benefits for police and fire protection employees, cities
with collective bargaining are spending 15 to 25% more than cities without collective bargaining.
Unions and Public Sector Pensions, 2009
While the cost of employees’ health benefits has been a significant source of stress on
state and local budgets in recent years, that stress pales in comparison to the size and seriousness
of the problem that has arisen in the area of public sector retirement benefits. In fact, just as we
have argued that studies that focus exclusively on salaries underestimate the effects of public
sector unions on compensation, we, too, would only be telling part of the story if we ignored
public sector pensions.
Until very recently, however, the available data on public sector pension liabilities were
largely inaccurate. The main problem boils down to an actuarial assumption: in order to
estimate how much money has been promised to future retirees in today’s dollars, and to
estimate how much a fund needs in assets today in order to meet those obligations, one must
assume a particular discount rate. Most state and local governments assume a discount rate equal
to the expected rate of return, usually around 8%. However, almost all economists agree that the
discount rate for liabilities should take into account the uncertainty of those liabilities. And since
governments are legally obligated to make promised pension payments to retired state and local
workers – meaning that the risk of default is low to zero – most economists argue that the
discount rate should be lower than the expected rate of return in order to reflect the risk that
30
governments (taxpayers) have taken on with such obligations. Thus, depending on the certainty
with which public pension promises will be paid off (which depends on matters such as whether
states’ constitutions guarantee them), the appropriate rate for discounting pension liabilities
ranges from about 3% (the average U.S. Treasury rate) to 5-6% (the tax-adjusted municipal bond
rate). Today, for example, private sector pension funds typically assume rates around 5%.
The bottom line is that the assumed discount rate makes a big difference to estimates of
pension liabilities. And government officials have good reason for wanting to keep it at 8%: it
makes their pension liabilities look smaller than they actually are, which in turn allows them to
pay less into the pension funds than they ought to be paying to ensure that the latter are fully
funded. Recently, however, scholars have begun to model these liabilities using more
appropriate discount rates, creating new, more accurate measures of how much governments
have promised to public sector workers in pensions – and the degree to which they have fallen
short in funding those promises. These new measures make it clear that the pension crisis facing
state and local governments is far worse than governments themselves report (Elliot, 2010).31
In this final section, we use these new measures to test whether states’ public sector
pension liabilities and the degree to which they are underfunded are related to public sector
unionization. From the outset, we note that our ability to identify the causal relationship between
unions and public sector pensions is limited. Because the vast majority of state and local
governments in the U.S. invest in large, statewide pension funds, our analysis uses aggregate
state-level data, and yet there are many characteristics of states that could be correlated with both
their rates of public sector unionization and their public sector pension liabilities. Moreover,
31 In addition to the assumption about the discount rate, there are a number of assumptions built into models that value pension liabilities and assets, such as assumptions about how future service and wage increases are handled and how asset values are smoothed over time. We focus our discussion here on the discount rate because it has the biggest consequences for estimating pension liabilities and how underfunded they are.
31
while we consider ourselves fortunate to have access to any accurate data at all, it poses a
challenge for the empirical analysis that the data are only available for a single year. Thus, the
analysis we present here is really just a first step. Even so, it is an important first step that we
hope will inspire scholars to find better data and better ways of investigating the politics of
public sector pensions.
Data and Empirical Strategy
For our study, we use the state-level estimates of public pension assets and liabilities as
of June 2009 calculated by Novy-Marx and Rauh (2010). Novy-Marx and Rauh assembled data
on 193 unique pension plans (the 116 major state-sponsored plans plus 77 local plans that had
assets over $1 billion), estimated each plan’s assets and liabilities using the zero-coupon
Treasury rate, and then aggregated the figures by state.32
We are interested in two main quantities from their dataset. The first is the total amount
states have taken on in pension liabilities. In some ways, this is the most straightforward
measure for us to use since it represents the total amount promised in retirement benefits to
public sector workers, which we expect to be higher in states where public sector unions are
strong. Since larger and better economically situated states can be expected to have larger public
pension liabilities, our first two dependent variables are total public pension liabilities per capita
and total public pension liabilities as a percentage of state gross product.
We are also interested in testing whether the extent of the pension problem is related to
the strength of public sector unions. In addition to total liabilities, then, we look at the degree to
which public pensions are underfunded in each state, defined as the difference between pension
liabilities and assets (both per capita and as a percentage of state gross product). There is good
reason to think that pension underfunding would be positively correlated with liabilities: as 32 See also Rauh (2011).
32
pension liabilities mount, they become an increasing burden for state and local governments,
which probably increases politicians’ temptation to shortchange pension funding. If so, then we
would expect public sector unionization to be positively associated with pension underfunding.
There is an alternative hypothesis to consider, however, which is that strong unions might be
relatively more successful in making sure that state and local politicians make the regular
contributions necessary to keep the funds solvent. If so, then pension funding levels in strong
union states could be equal to or even better than those of states with weaker unions. This is an
empirical question, and one that we test directly.
In using the Novy-Marx and Rauh state-level estimates, we are cognizant of the fact that
certain pension funds were excluded from their calculations. This is not an issue for states where
most or all public pensions are managed by large state funds, but in some states, such as Texas,
Florida, and Illinois, there are hundreds of small local pension funds that are not included. Out
of concern that the percentage of pension assets and liabilities included in their figures might be
correlated with states’ levels of unionization, we created a control variable equal to the
proportion of total public sector pension members in a state covered by the plans in the Novy-
Marx and Rauh dataset, using information from the U.S. Census Annual Survey of State and
Local Public Employee Retirement Systems. The proportion ranges from 0.57 to 1, with a
median of 0.97 and a mean of 0.92.33
Since our goal for the empirical analysis is to test whether there is a link between public
sector unions and public sector pension liabilities, we also need a measure of the rate of public
33 To create this variable, we downloaded the complete list of state and local pension funds included in the U.S. Census Annual Survey of State and Local Public Employee Retirement Systems from fiscal year 2009, and we matched each fund in the Novy-Marx and Rauh listing to the larger set of funds in the Census data. Then, using the Census information for each individual fund, we calculated the total number of members and beneficiaries covered by the funds included in the Novy-Marx and Rauh dataset. (There were two funds included in the Novy-Marx and Rauh dataset for which we could not acquire data from the Census survey.) From there, we used the Census estimates of total pension fund membership and beneficiaries by state to calculate the proportion of all fund members and beneficiaries who are covered by the Novy-Marx and Rauh state-level estimates.
33
sector union membership in each state. To create such a measure, we used data from the Current
Population Survey (CPS), which asks a quarter of the individuals in its sample each month
whether they are members of unions.34 After downloading all of the monthly CPS data from
January 2006 to June 2010, we isolated the individuals in each data file who worked full-time for
either state or local government. Using that subsample, we calculated the proportion of
respondents in each state who were members of unions.
In our models of public sector pension liabilities, we use OLS to regress our liabilities
measures on the proportion of public sector workers who are members of unions, controlling for
the percentage of pension fund members accounted for in the Novy-Marx and Rauh data.
Specifically, our model is as follows:
The subscript j indexes the states, unionj is the proportion of public sector workers who are in
unions, coveragej is the proportion of pension members in the state covered by the pension funds
included in our dependent variable, γ, ρ, λ, and ω are regression coefficients, and εj is an error
term. To correct for heteroskedasticity, we use robust standard errors in all of our models.
The matrix Zj is a set of variables that we include to account for state-level factors that
could be correlated with both a state’s level of pension liabilities and its public sector
unionization rate. Perhaps most importantly, since more Democratic states tend to have stronger
unions and also are more inclined to provide public sector workers with good retirement benefits,
we control for Barack Obama’s two-party vote share in the 2008 presidential election. As in our
analysis of the ICMA data, we control for the worker-friendliness of the state by including the
union membership rate among private sector workers in 2009, as measured by Hirsch and
34 The relevant question is: “On this job, is ___ a member of a labor union or of an employee association similar to a union?” Respondents who answer “yes” are counted as union members.
34
MacPherson (2003, 2011). Variation in cost of living could also explain why some states have
higher pension liabilities than others, and states with higher cost of living also tend to have
higher rates of public sector unionization. We therefore include logged per capita income in
2009 as a predictor.35 In addition, we control for the proportion of all employed individuals who
work in the public sector, expecting that the greater the proportion, the higher should be both
public sector union membership and public sector pension liabilities. Lastly, since some states
depend heavily on the federal government for revenue, which might help them to pay heftier
pensions to public sector workers, we control for the percentage of a state’s total general revenue
that came from federal sources in 2008.36
Empirical Analysis
The results are set out in the first two columns of Table 4. In column 1, we ask whether
states’ per capita pension liabilities are related to the rate of public sector union membership.
The coefficient on the unionization variable indicates clearly that the answer is yes: we find that
each 10 percentage point increase in the public sector unionization rate is associated with an
increase of $1,412 per capita in public sector pension liabilities. What this means is that a move
from the least unionized state to the most unionized state in the U.S. is associated with a
whopping $9,857 increase in public sector pension liabilities for every person in the state. This
is indeed a massive effect, highly significant both statistically and substantively.
We find a similar effect in column 2, where the dependent variable is total liabilities as a
percentage of state gross product. The coefficient on the unionization variable in that model
suggests that moving from the least to the most unionized state is associated with an increase in
public pension liabilities equivalent to over 20% of state gross product. In other words, high
35 The per capita income data are from the Bureau of Economic Analysis. 36 This last variable comes from the 2012 U.S. Statistical Abstract.
35
rates of unionization in the public sector are related to increased pension promises equal to a fifth
of the value of the whole state economy. As in column 1, the coefficient is statistically
significant at the 1% level. More importantly, though, it demonstrates that there is a linkage
between public sector unions and the size of states’ pension promises.
Aside from the public sector unionization measure, only one of the other variables in the
model registers as a statistically significant predictor of pension liabilities: the percentage of
members covered by the funds included in our dependent variables. As we suspected, the
coefficient is positive, meaning that when larger percentages of statewide pension fund members
are covered by the biggest state and local funds, the Novy-Marx and Rauh measure of liabilities
gets larger as well. Among the remaining control variables, none are significant predictors of
both measures of liabilities. The results in column 2 suggest that per capita income is negatively
associated with larger pension liabilities as a percentage of state gross product, which probably
has to do with the relationship between per capita income and state gross product: states with
high per capita income also have strong economies, and in states with strong economies,
liabilities as a percentage of state gross product will tend to be lower (since the denominator is
larger). However, the coefficient on per capita income in column 1 is indistinguishable from
zero. Moreover, for percent Democrat, percent in private sector unions, percent public sector
employment, and percent of revenue from federal sources, we find no significant effects.
To test whether the extent of public pension underfunding is related to public sector
unions, we use the same model but include some additional controls to account for the likely
possibility that states in poorer economic condition probably have more poorly funded pensions
than states that are more economically robust. First, as a measure of the overall condition of the
state economy, we include the state’s unemployment rate in 2009, as reported by the Bureau of
36
Labor Statistics. Second, we include two measures of the impact of the recession on the
economic condition of the state: the change in the state unemployment rate from 2004 to 2009,
and the percentage change in per capita income from 2004 to 2009 (in real terms). We expect
unemployment and change in unemployment to have negative effects on the extent to which
pensions are underfunded, and we expect change in per capita income to have a positive effect.
We present the results of these models in columns 3 and 4 of Table 4. Again, we find a
very clear, positive, statistically significant effect of public sector unionization. The coefficient
of interest in column 3 implies that a 10 percentage point increase in public sector unionization is
associated with an extra $626 in unfunded pension liabilities for every person in the state.
Moving from the least unionized to the most unionized state is predicted to increase unfunded
pension liabilities per capita by $4,369. In column 4, we find that such a move is associated with
an increase in unfunded pension liabilities equivalent to nearly 10% of state gross product.
These effects are statistically significant, and they show that the average severity of the pension
crisis is far greater in states with strong public sector unions than in states with weaker unions.
Of the remaining variables, two stand out as strong predictors of states’ public pension
underfunding. The first is the state’s unemployment rate. Unsurprisingly, states with higher
unemployment rates also have greater unfunded public pension liabilities than states with lower
unemployment rates. The coefficients on the other two economic indicators – change in
unemployment and change in per capita income – have the expected signs but are statistically
insignificant. Second, as in columns 1 and 2, we find that our coverage variable is positively
associated with unfunded liabilities.
But by far the most striking results in Table 4 are the coefficients on public sector union
membership. As we have argued, this is precisely what one should expect. In states where
37
public sector workers are well organized, they should be more effective in pressuring state and
local politicians for higher wages, better health benefits, and more generous retirement packages.
And relative to granting increases in salaries or improving benefits, politicians no doubt find it
less costly to promise more generous pensions. After all, by the time those promises come due,
chances are high that the politicians who made them will no longer be in office. Today,
however, the promises of politicians in the past are coming due. State and local governments
have collectively racked up astronomical obligations to public sector retirees, and trillions of
dollars of those obligations remain unfunded. What we have found in Table 4 is that the problem
is most severe in states with strong public sector unions – states that not only have the greatest
pension liabilities but that also are the most delinquent in funding them.
Conclusion
Prior to the 1960s, public sector unions were not a factor in American government. They
had few members; collective bargaining was rare (and often illegal); and private sector unions
monopolized the labor movement, mobilized its political power, and were kingpins within the
Democratic coalition. But during the 1960s and ‘70s, all that changed. Private sector unions
began a devastating decline that sapped their economic and political power, and new state labor
laws set off an explosion of union organizing and collective bargaining among government
workers. When the dust cleared, the leadership and clout of the union movement—in politics, in
the Democratic Party, in collective bargaining—rested with the public sector unions.
Political scientists have long recognized that interest groups are central to any
understanding of American government, and some of the classic works of political science—
Schattschneider’s The Semi-Sovereign People (1975), Lowi’s The End of Liberalism (1969),
McConnell’s Private Power and American Democracy (1970), to name a few—have
38
documented the powerful role that these groups often play. In today’s world, public sector
unions are surely among the most prominent interest groups in the country, especially at the state
and local levels where most of their members work and most of the nation’s public money is
spent. This in itself is good reason to study them. But they are also different from other interest
groups—and particularly worth studying—on two grounds. One, they alone engage in collective
bargaining, and are thus in a position to shape the organization, operation, and costs of
government in ways that other groups are not. And two, their members are the government’s
own employees—its bureaucrats—who influence government from the inside through their
official roles, but also from the outside through their unions.
When public sector unions are taken seriously as subjects of study, the questions that
arise are clearly fundamental, having to do with the costs of government, with its organization
and effectiveness, with the substance of public policy, and with the role of union power in
elections, in the Democratic Party, and in the political process more generally. Even so, political
scientists have barely paid attention to these important political actors. And although economists
made a good start during the 1980s, rushing to study what was at the time a newly emerging
phenomenon, their interest subsided over the years. The result is that public sector unions
remain poorly studied—and their consequences for American government poorly understood.
This paper is an effort to encourage new research and help move the ball downfield. We
carry out three empirical studies of the impact of public sector unions on the cost of government.
Each one, taken by itself, has certain limitations due to the underlying nature of the data. But
considered together they paint a consistent picture. In the first study, we leverage within-city
variation in public sector unionism from 1972 to 1987 and find that the unionization of police
and fire protection employees had the effect of increasing average wages, employment, and total
39
payroll expenditures in municipal police and fire departments across the country. Using more
current data from 1992 through 2010, we find in the second study that municipal police and fire
departments with collective bargaining spend more on salaries, but also spend far more on their
employees’ health, hospital, disability, and life insurance benefits—on the order of 15 to 25%
more. Finally, in a preliminary study of the relationship between public sector unionization and
public sector pensions, we find that states with higher percentages of public sector workers in
unions tend to have significantly higher pension liabilities as well as higher rates of pension
underfunding. The weight of the evidence suggests, then, that public sector unionization and
collective bargaining do lead governments to spend more on their employees—particularly in the
areas of health benefits and retirement compensation.
These studies are a useful step, we think, in moving the literature forward. But much
more ground remains to be covered. If American government is to be well understood, for
example, researchers need to explore how collective bargaining influences the way government
is organized—via, for example, seniority provisions, restrictions on evaluation and dismissal,
restrictions on the allocation of workers, and much more—and how labor contracts influence the
effectiveness of government performance. It is also important to explore how union power in
politics, particularly elections, affects their ability to win favorable contracts and thereby to
shape the organization and costs of government in distinctive ways compatible with their job
interests. More generally, research needs to determine what public sector unions do in the
political process: how much they contribute in elections and to whom, how successful they are in
putting allies into office, what types of policies they pursue (regarding, say, public spending and
taxing), how their political power shapes the ongoing battle between Democrats and
Republicans—and overall, what the implications are for policy outcomes.
40
The aim, long term, is to move toward a body of knowledge about public sector unions
that, in shedding light on these important but long-neglected political actors, contributes to the
larger effort to understand American government.
41
REFERENCES
Ashenfelter, Orley. 1971. “The Effect of Unionization on Wages in the Public Sector: The Case of Firefighters.” Industrial and Labor Relations Review 24: 191-202.
Bahrami, Bahman, John D. Bitzan, and Jay A. Leitch. 2009. “Union Worker Wage Effect in the Public Sector.” Journal of Labor Research 30: 35-51. Bartel, Ann, and David Lewin. “Wages and Unionism in the Public Sector: The Case of Police.” Review of Economics and Statistics 63: 53-9. Bellante, Don, and James Long. “The Political Economy of the Rent-Seeking Society: The Case of Public Employees and their Unions.” Journal of Labor Research 2: 1-14. Biggs, Andrew G., and Jason Richwine. 2011. “Assessing the Compensation of Public-School Teachers.” A Report of the Heritage Center for Data Analysis (Washington, D.C.).
Bitzan, John D., and Bahman Bahrami. 2010. “The Effects of Unions on Wages by Occupation in the Public Sector.” International Business and Economics Research Journal 9: 107-119.
Blais, Andre, Donald E. Blake, and Stephane Dion. 1997. Governments, Parties, and Public Sector Employees (Pittsburgh: University of Pittsburgh Press). Blanchflower, David, and Alex Bryson. 2004. “What Effect Do Unions Have on Wages Now and Would Freeman and Medoff Be Surprised?” Journal of Labor Research 25 (Summer): 383-414.
Brown, Charles C., and James L. Medoff. 1988. “Employer Size, Pay, and the Ability to Pay in the Public Sector.” In Richard Freeman and Casey Ichniowski, eds., When Public Sector Workers Unionize (Chicago: University of Chicago Press): 195-216.
Bush, Winston C., and Arthur T. Denzau. 1977. The Voting Behavior of Bureaucrats and Public Sector Growth. In Thomas Borcherding, ed., Budgets and Bureaucrats. Durham, NC: Duke University Press. Carpenter, Daniel P. 2001. The Forging of Bureaucratic Autonomy (Princeton, NJ: Princeton University Press). Cigler, Allan J., and Burdette J. Loomis. 2011. Interest Group Politics (Washington, D.C.: CQ Press). Courant, Paul N., Edward Gramlich, and Daniel Rubinfeld. 1979. “Public Employee Market Power and the Level of Government Spending.” American Economic Review 69 (December): 806-17. Delli Santi, Angela. 2011. “New Jersey Anti-Union Bill Approved, Governor Christie Signs into Law.” Huffington Post, June 28, 2011. DiLorenzo, Thomas J. “Exclusive Representation in Public Employment: A Public Perspective.” Journal of Labor Research 5: 371-83. Ehrenberg, Ronald. 1973. “Municipal Government Structure, Unionization, and the Wages of Fire Fighters.” Industrial and Labor Relations Review 27: 36-48.
Elliott, Douglas J. 2010. “State and Local Pension Funding Deficits: A Primer.” The Brookings Institution.
42
Epple, Dennis and Thomas Nechyba (2004), “Fiscal Decentralization.” In J. Vernon Henderson and J. Thisse, eds., Handbook of Regional and Urban Economics Vol. 4 (Elsevier Science). Epstein, David, and Sharyn O’Halloran. 1999. Delegating Powers. New York: Cambridge University Press. Freeman, Richard B. 1986. “Unionism Comes to the Public Sector.” Journal of Economic Literature XXIV (March): 41-86. ..............., and Robert Valletta. 1988. “The Effects of Public Sector Labor Laws on Labor Market Institutions and Outcomes.” In Richard Freeman and Casey Ichniowski, eds., When Public Sector Workers Unionize (Chicago: University of Chicago Press). …………, Casey Ichniowski, and Jeffrey Zax. 1988. “Appendix A: Collective Organization of Labor in the Public Sector.” In Richard Freeman and Casey Ichniowski, eds., When Public Sector Workers Unionize (Chicago: University of Chicago Press). Gallagher, Daniel G. 1978. “Teacher Bargaining and School District Expenditures.” Industrial Relations 17: 216-39. Greenhouse, Steven. 2011. “Ohio’s Anti-Union Law Is Tougher than Wisconsin’s.” New York Times, March 31, 2011. Gregory, Robert, and Jeffrey Borland. 1999. “Recent Developments in Public Sector Labor Markets.” In Orley Ashenfelter and David Card, eds., Handbook of Labor Economics (New York: Elsevier Science, North Holland): 573-690.
Hirsch, Barry T. 2004. “What Do Unions Do for Economic Performance?” Journal of Labor Research 25 (3): 415-455.
Hirsch, Barry T., and David A. Macpherson. 2003. “Union Membership and Coverage Database from the Current Population Survey: Note.” Industrial and Labor Relations Review 56 (2): 349-354.
Hirsch, Barry T., and David A. Macpherson. 2011. Union Membership and Coverage Database from the CPS, available at http://unionstats.com/.
Hoxby, Caroline Minter. 1996. “How Teachers’ Unions Affect Education Production.” Quarterly Journal of Economics 111 (3): 671-718.
Hunter, William J., and Carol H. Rankin. 1988. “The Composition of Public Sector Compensation: The Effects of Bureaucratic Size.” Journal of Labor Research 9: 29-42.
Ichniowski, Casey. 1980. “Economic Effects of the Firefighters’ Union.” Industrial and Labor Relations Review 33: 198-211.
Johnson, Ronald J., and Gary D. Libecap. 1994. The Federal Civil Service System and the Problem of Bureaucracy. Chicago: University of Chicago Press. Kearney, Richard. 2009. Labor Relations in the Public Sector. 4th ed. (New York: CRC Press). Locker, Richard. 2011. “Tennessee Legislature Ok’s Ban of Teacher Bargaining.” The Commercial Appeal, Memphis, TN, May 20, 2011.
43
Lowi, Theodore. 1969. The End of Liberalism (Norton). Marlow, Michael L., and William Orzechowski. 1996. “Public Sector Unions and Public Spending.” Public Choice 89: 1-16.
McConnell, Grant. 1970. Private Power and American Democracy (Random House).
Moe, Terry M. 2011. Special Interest: Teachers Unions and America’s Public Schools (Washington, D.C.: Brookings Institution). ...............2006. “Political Control and the Power of the Agent.” Journal of Law, Economics, and Organization 22 (Spring): 1-29 Novy-Marx, Robert, and Joshua D. Rauh. 2010. “Public Pension Promises: How Big Are They and What Are They Worth?” Journal of Finance 66 (4): 1207-1245. Rauh, Joshua D. 2011. “The Pension Bomb.” The Milken Institute Review, 2011Q1. Schattschneider, E. E. 1975. The Semi-Sovereign People (Wadsworth). Trejo, Stephen J. 1991. “Public Sector Unions and Municipal Employment.” Industrial and Labor Relations Review 45 (October): 166-80. Troy, Leo. 1994. The New Unionism in the New Society: Public Sector Unions in the Redistributive State (Fairfax, VA: George Mason University Press).
..............., and Neil Sheflin. 1985. Union Sourcebook: Membership, Structure, Finance, Directory (West Orange, N.J.: Industrial Relations Data and Information Services).
Valetta, Robert. 1993. “Union Effect on Municipal Employment and Wages: A Longitudinal Approach.” Journal of Labor Economics 11 (July): 545-74.
...............1988. “The Impact of Unionism on Municipal Expenditures and Revenues.” Industrial and Labor Relations Review 42: 430-42.
Wellington, Harry H., and Ralph K. Winter. 1971. The Unions and the Cities (Washington, D.C.: Brookings Institution).
West, Martin. 2008. “Bargaining with Authority: The Political Origins of Public Sector Bargaining,” paper presented at the 2008 Policy History conference, St. Louis, MO, May 29–June 1. Zax, Jeffery. 1989. “Employment and Local Public Sector Unions.” Industrial Relations 28 (Winter): 21-31.
Zax, Jeffrey S. 1988. “Wages, Nonwage Compensation, and Municipal Unions.” Industrial Relations 27: 301-317.
..............., and Casey Ichniowski. 1988. “The Effects of Public Sector Unionism on Pay, Employment, Department Budgets, and Municipal Expenditures.” In Richard Freeman and Casey Ichniowski, eds., When Public Sector Workers Unionize (Chicago: University of Chicago Press).
44
Table 1: Effect of Public Employee Organization on Wages, Employment, and Payroll, 1972-1987
Fire Protection Employees Police Protection Employees (1) (2) (3) (4) (5) (6) (7) (8)
Average Wage Employment Payroll Average Wage Employment Payroll Union 0.038 0.038 0.073 0.101 0.023 0.023 0.023 0.036
(0.011)*** (0.011)*** (0.020)*** (0.021)*** (0.010)*** (0.010)*** (0.010)** (0.012)*** Ln(Population) 0.167 0.167 -0.096 0.081 0.138 0.137 -0.221 -0.091
(0.029)*** (0.029)*** (0.059) (0.065) (0.024)*** (0.024)*** (0.026)*** (0.034)*** SES 0.058 0.058 0.037 0.089 0.062 0.063 0.075 0.131
(0.021)*** (0.021)*** (0.033) (0.039)** (0.016)*** (0.016)*** (0.019)*** (0.023)*** % Black -0.094 -0.096 0.13 0.039 -0.138 -0.139 0.262 0.146
(0.082) (0.082) (0.198) (0.221) (0.083)* (0.082)* (0.119)** (0.149) % Hispanic 0.281 0.281 -0.964 -0.68 0.274 0.275 -0.01 0.254
(0.124)** (0.124)** (0.298)*** (0.339)** (0.101)*** (0.101)*** (0.123) (0.154)* % Manufacturing 0.145 0.145 0.371 0.484 0.054 0.053 0.442 0.57
(0.102) (0.102) (0.213)* (0.214)** (0.091) (0.091) (0.112)*** (0.135)*** % in School 0.149 0.154 -0.465 -0.45 -0.015 -0.004 -0.738 -0.691
(0.217) (0.217) (0.514) (0.564) (0.186) (0.186) (0.206)*** (0.256)*** % in Poverty -0.413 -0.414 -0.372 -0.769 -0.353 -0.353 -0.176 -0.499
(0.073)*** (0.073)*** (0.152)** (0.167)*** (0.065)*** (0.065)*** (0.080)** (0.100)*** Population Growth 0.022 0.022 0.251 0.279 0.059 0.058 0.183 0.24
(0.035) (0.035) (0.058)*** (0.066)*** (0.030)** (0.030)* (0.030)*** (0.038)*** Lagged Deviation from State Avg. -0.078 -0.054 -0.071 -0.038
(0.020)*** (0.037) (0.017)*** (0.026) Union X -0.038 -0.053 Lagged Deviation from State (0.044) (0.032)* Observations 5149 5149 5600 5600 5959 5959 6756 6756 Unique municipalities 1316 1316 1400 1400 1527 1527 1689 1689 R-squared 0.86 0.86 0.92 0.91 0.86 0.86 0.87 0.87 Notes: Robust standard errors clustered by municipality in parentheses. All models include municipality fixed effects and year fixed effects. Dependent variable in columns 1-2 and 5-6 is logged average wage. Dependent variable in columns 3 and 7 is logged per capita employment. Dependent variable in columns 4 and 8 is logged per capita payroll expenditures. Hypothesis tests on Union are one-tailed in columns 1-2, 4-6, and 8; all other tests are two-tailed. * significant at 10%; ** significant at 5%; *** significant at 1%
45
Table 2: Collective Bargaining and Fire Protection Compensation and Employment
Salary expenditures / employee
Health expenditures / employee
Employment Salary expenditures per capita
Health expenditures per capita
(1) (2) (3) (4) (5) Collective Bargaining 0.086 0.224 -0.007 0.086 0.227
(0.016)*** (0.023)*** (0.031) (0.031)*** (0.042)***
Ln(Population) 0.054 0.027 -0.039 0.016 -0.014
(0.006)*** (0.010)*** (0.012)*** (0.013) (0.017)
SES 0.024 -0.021 0.112 0.14 0.086
(0.010)** (0.019) (0.025)*** (0.025)*** (0.035)**
Ln(Median Rent) 0.386 0.356 -0.244 0.103 -0.003
(0.041)*** (0.067)*** (0.075)*** (0.070) (0.106)
% Democrat 0.186 0.088 -0.005 0.234 0.264
(0.054)*** (0.097) (0.115) (0.110)** (0.161)
Ln(Population Density) 0.034 0.023 -0.032 0.01 -0.013
(0.010)*** (0.017) (0.020) (0.020) (0.030)
% in Poverty -0.258 0.082 0.269 -0.052 0.006
(0.131)** (0.255) (0.257) (0.247) (0.356)
% Black -0.071 -0.266 0.498 0.442 0.315
(0.050) (0.101)*** (0.094)*** (0.097)*** (0.138)**
% Hispanic 0.144 -0.007 -0.192 -0.043 -0.206
(0.053)*** (0.089) (0.123) (0.130) (0.168)
% in School 0.089 0.62 -1.259 -1.13 -0.563
(0.204) (0.354)* (0.407)*** (0.408)*** (0.539)
% Manufacturing 0.213 0.861 0.536 0.78 1.415
(0.082)*** (0.141)*** (0.167)*** (0.166)*** (0.240)***
% Private Sector Union 1.658 2.646 0.126 1.702 2.74
(0.178)*** (0.343)*** (0.406) (0.410)*** (0.551)***
Population Growth -0.025 -0.017 -0.094 -0.114 -0.089
(0.013)* (0.019) (0.020)*** (0.022)*** (0.030)***
Lagged Dev. from State Avg. 0.13 0.402
(0.025)*** (0.038)***
Observations 6897 6009 8809 8471 7530
R-squared 0.51 0.55 0.28 0.31 0.34
Notes: Robust standard errors clustered by municipality in parentheses. All models include region and year fixed effects. Dependent variables are as follows: (1) logged total expenditures on fire protection salaries and wages per employee, (2) logged total expenditures on fire protection health benefits per employee, (3) logged full-time fire protection employment per capita, (4) logged total expenditures on fire protection salaries and wages per capita, (5) logged total expenditures on fire protection health benefits per capita. Collective Bargaining = 1 if fire protection employees in the municipality have collective bargaining. Hypothesis tests on Collective Bargaining are one-tailed in all but column 3; all other tests are two-tailed. * significant at 10%; ** significant at 5%; *** significant at 1%.
46
Table 3: Collective Bargaining and Police Protection Compensation and Employment
Salary expenditures / employee
Health expenditures / employee
Employment Salary expenditures per capita
Health expenditures per capita
(1) (2) (3) (4) (5) Collective Bargaining 0.098 0.19 -0.056 0.042 0.153
(0.010)*** (0.016)*** (0.019)*** (0.021)** (0.029)***
Ln(Population) 0.044 0.006 -0.03 0.01 -0.025
(0.004)*** (0.008) (0.008)*** (0.008) (0.013)**
SES 0.024 -0.012 0.026 0.054 0.017
(0.007)*** (0.013) (0.015)* (0.016)*** (0.023)
Ln(Median Rent) 0.311 0.249 0.039 0.338 0.265
(0.027)*** (0.041)*** (0.043) (0.048)*** (0.064)***
% Democrat 0.089 0.02 0.191 0.288 0.261
(0.034)*** (0.065) (0.069)*** (0.072)*** (0.105)**
Ln(Population Density) 0.036 0.045 0.002 0.036 0.051
(0.006)*** (0.010)*** (0.012) (0.012)*** (0.018)***
% in Poverty -0.354 -0.031 0.344 -0.054 0.192
(0.069)*** (0.134) (0.174)** (0.175) (0.240)
% Black 0.017 -0.22 0.727 0.767 0.502
(0.030) (0.064)*** (0.067)*** (0.071)*** (0.100)***
% Hispanic 0.13 -0.046 0.17 0.317 0.109
(0.033)*** (0.060) (0.065)*** (0.069)*** (0.095)
% in School -0.159 0.242 -1.43 -1.677 -1.166
(0.117) (0.220) (0.248)*** (0.264)*** (0.376)***
% Manufacturing 0.118 0.532 0.037 0.162 0.698
(0.055)** (0.093)*** (0.102) (0.109) (0.159)***
% Private Sector Union 1.604 2.872 -0.36 1.227 2.423
(0.119)*** (0.230)*** (0.260) (0.270)*** (0.354)***
Population Growth -0.018 -0.015 -0.05 -0.07 -0.069
(0.006)*** (0.014) (0.010)*** (0.013)*** (0.014)***
Lagged Dev. from State Avg. 0.188 0.487
(0.026)*** (0.048)***
Observations 13,227 11,531 16,809 15,865 13,976
R-squared 0.55 0.59 0.25 0.37 0.37 Notes: Robust standard errors clustered by municipality in parentheses. All models include region and year fixed effects. Dependent variables are as follows: (1) logged total expenditures on police protection salaries and wages per employee, (2) logged total expenditures on police protection health benefits per employee, (3) logged full-time police protection employment per capita, (4) logged total expenditures on police protection salaries and wages per capita, (5) logged total expenditures on police protection health benefits per capita. Collective Bargaining = 1 if police protection employees in municipality have collective bargaining. Hypothesis tests on Collective Bargaining are one-tailed in all but column 3; all other tests are two-tailed. * significant at 10%; ** significant at 5%; *** significant at 1%.
47
Table 4: Public Sector Pensions and Public Sector Unions
(1) (2) (3) (4)
Liabilities per capita Liabilities / GSP
Unfunded liabilities per capita
Unfunded liabilities / GSP
% Public Sector Union 14,120 0.293 6,258 0.14 (5,835)*** (0.099)*** (2,879)** (0.065)**
% of Funds Covered 10,089 0.244 5,988 0.142 (6,443) (0.125)* (3,238)* (0.070)**
% Democrat -11,539 -0.027 -2,165 0.059 (9,859) (0.206) (6,094) (0.148)
% Private Sector Union 5,292 -0.092 -11,551 -0.393 (26,754) (0.566) (19,813) (0.430)
Ln(Per Capita Income) 2,502 -0.314 6,802 -0.077 (5,950) (0.123)** (4,714) (0.092)
% Public Sector Employment 36,734 0.612 9,015 0.164 (25,060) (0.479) (12,797) (0.296)
% Federal Revenue -1,350 0.22 7,987 0.29 (15,316) (0.277) (8,528) (0.200)
Unemployment 1,073 0.024 (386)*** (0.009)***
Change in Unemployment -777 -0.016 (574) (0.012)
Change in Per Capita Income 11,694 0.322 (12,670) (0.259)
Constant -27,747 3.146 -79,756 0.549 (64,040) (1.288)** (50,995) (0.993)
Observations 50 50 50 50 R-squared 0.45 0.36 0.44 0.39
Notes: Robust standard errors in parentheses. Tests on % Public Sector Union are one-tailed in columns 1 and 2; all other tests are two-tailed. * significant at 10%; ** significant at 5%; *** significant at 1%.
48
Supplemental Information Appendix
This document provides additional description of the datasets used in the paper, “Public
Sector Unions and the Costs of Government,” and also presents empirical results that are
described but not shown in the main text.
1. Unionization and City Wages, Employment, and Payroll in the 1970s and 1980s Most of the data for the first study we present in the paper come from the U.S. Census of
Governments employment files for the years 1972, 1977, 1982, and 1987, which we downloaded
from the Interuniversity Consortium for Political and Social Research (ICSPR studies 69, 8117,
8395, and 6069). Our focus is on municipal governments, defined by the Census as incorporated
places with general-purpose government.37 To make our analysis comparable to the analysis of
the ICMA data described in the next section, and because we wanted to focus on cities large
enough to house their own police and fire departments, we limited our analysis to municipal
governments that had at least 10,000 residents as of 1972.
We used five variables from each of these datasets to construct the dependent variables
for our analysis. First, for both fire and police protection employees, we used the municipality’s
total payroll expenditures (meaning monthly gross pay including salaries, wages, fees, and
commissions) for full-time employees as of October of the year of the survey. In addition, we
used the municipality’s population as well as its total number of full-time fire protection
employees and police protection employees in each year. The first dependent variable in the
analysis – average wage – is total payroll expenditures for full-time employees divided by the
number of full-time employees. Employment per capita is the number of employees divided by
population, and payroll expenditures per capita is total payroll divided by population. We
37 This includes cities, villages, boroughs outside of Alaska, and towns outside of New England, New York, and Wisconsin.
49
identified likely data entry errors by examining within-city changes in each variable over time.
We excluded cities that reported that one or more of these dependent variables more than tripled
over a single five-year period. For the fire protection analysis, this resulted in the exclusion of
37 cities (148 observations), and for the police protection analysis, it resulted in the exclusion of
4 cities (16 observations).
The datasets also included information on the number of full-time fire and police
protection employees who were members of employee organizations in each year, which we
used to create our main independent variables. If a municipality had any organized fire
protection employees in a given year, we coded the fire union variable as a one, and if it reported
having no organized fire protection employees in a given year, we coded it as a zero. We used
the same rule for police protection employees.
We then inspected the within-municipality patterns of fire and police protection
organization over time to identify cases in which municipalities reported having organized fire or
police protection employees in one year but not in one or more subsequent years. Based on
Freeman, Ichniowski, and Zax’s (1988) investigation of governments that reported losing (or
losing and regaining) collective bargaining status during this time period, we assume that these
cases are reporting errors. We address these errors using two different approaches and ensure
that the results we present in the paper are not sensitive to which approach we use.
The first approach was the more conservative of the two: we only included in the
analysis municipalities where fire (or police) protection workers were organized in every year
from 1972 to 1987, were organized in none of the years from 1972 to 1987, or went from being
unorganized to organized at some point between 1972 and 1987. We eliminated any
50
municipalities that did not appear in the dataset in all four years as well as municipalities that
reported having organized fire (or police) employees and then losing that organization.
The second approach, which is the one we used for the paper, involved making
adjustments to the employee organization variables in some municipalities that reported having
and then losing employee organizations. For most cities, we did not make any changes to the
coding of the Union variable because there was no ambiguity. However, for the cities that
reported having unions in one year but not in subsequent years, we generally relied on Freeman,
Ichniowski, and Zax’s finding that cities did not go from organized to unorganized during this
period. Thus, in most cases, if a city reported having unions in one year, we assumed that it had
them in later years and recoded the Union variable accordingly. There were a few cases that
were difficult to classify, however, and we handled them as follows:
If 1972 or 1977 was the only year a city reported having unions, that city was recoded as nonunionized for all years.
If 1982 was the only year that a city reported having unions, and fewer than 50% of
employees were in unions in that year, the city was recoded as nonunionized for all years. Otherwise, it was coded as unionized for 1982 and 1987.
A few cities reported having no unions in 1972, unions in 1977, no unions in 1982, and
then unions again in 1987. We coded most of those cities as unionized as of 1977. However, if a city reported that fewer than 50% of its employees were in unions in either 1977 or 1987, we recoded the city as nonunionized in 1977. Of those, we also coded cities as nonunionized in 1987 if they reported having fewer than 50% of its employees in unions that year.
We excluded cities that reported having unions only in 1972 and 1987, only in 1972 and
1977, or only in 1972 and 1982.
By recoding certain cases in this fashion, we were able to include a larger number of
municipalities in our analysis. However, we also carried out our analysis using the more
conservative approach which did not involve any recoding, and the results are presented in Table
A2 below.
51
We completed the dataset by adding a series of control variables from the Censuses of
Population from 1970, 1980, and 1990: city population, population growth (the percentage
growth in population over the next five years), the percentage of adults with a bachelor’s degree
or higher,38 per capita income, the percentage of employed individuals working in
manufacturing, and the percentages of the population that were African American, Hispanic,
living in poverty, and enrolled in elementary or high school. We imputed the values within each
city to get the values corresponding to 1972, 1977, 1982, and 1987. By inspecting changes in
each variable within cities over time, we identified one city that contained a likely error in the
population and population growth figures, and we eliminated it from our analysis. There were
also a few cities that had implausibly large changes in the elementary and high school enrollment
variable – more than 10 percentage points over a five-year period – which we excluded from our
models (5 for the fire protection analysis and 11 for the police protection analysis). Since per
capita income and percent with bachelor’s degree or higher are highly correlated, we average
them using factor analysis and create a single variable characterizing the socioeconomic status of
city residents.
In Figure A1, we plot the distributions of these variables in 1972 for two types of cities:
cities in which fire protection employees formed unions (either prior to 1972 or between 1972
and 1987) and cities in which fire protection employees never formed unions. The distributions
of the treatment and control cities are clearly different for all of the variables, and those
differences in distributions are confirmed by Kolmogorov-Smirnov tests (not shown).39 The
same is true for cities with and without unions of police protection employees, as shown in
38 For 1970 and 1980, the college education variable is the percentage of the population that had at least four years of college education. For years after 1980, it is the percentage that had a bachelor’s degree. 39 We conduct the Kolmogorov-Smirnov tests using the Matching package in R (Sekhon, 2011).
52
Figure A2: the distributions of the variables differ between cities where police never formed
unions and cities where they did.
Table A1 compares the means of each variable in 1972 in cities with and without unions,
for fire and police protection employees separately. On average, in 1972, cities where fire
protection employees eventually formed unions were larger in population, higher in
socioeconomic status, and had lower percentages of African Americans and Hispanics than cities
where fire protection unions never formed. They also had more adults employed in
manufacturing, a smaller percentage of the population enrolled in school, lower poverty rates,
and lower rates of population growth between 1972 and 1977. Most of the same is true of cities
with police unions as compared to those without police unions, except that average percent
Hispanic and average percent in school in 1972 are statistically indistinguishable between the
two types of cities. In general, then, because we suspect that some of these correlates of
unionization are also correlated with cities’ police and fire employment and compensation, we
control for them in all models that feature cities at the unit of analysis.
Tables A2 to A4 below present a series of empirical results using this dataset that are
described but not shown in the paper. First, in Table A2, we use the more conservative coding of
the Union variable, based on the first approach described above. Note that the number of
municipalities included in each model is smaller, but the results are substantively the same as
those presented in Table 1 of the paper. We continue to estimate an approximate 3% effect of
unionization on fire protection average wages. The effects of unionization on fire protection
employment levels and payroll expenditures per capita are even larger using this smaller sample.
While the effect of police unionization on average wage is slightly smaller in Table A2 than in
Table 1, it is still strong and significant. And the effects of police unionization on employment
53
and total payroll expenditures are slightly larger in Table A2 than the estimates we presented in
the main paper. Thus, our results are not sensitive to the corrections we made to the coding of
the police and fire unionization variables.
In Table A3, in addition to the municipality fixed effects, we include a complete set of
state-year fixed effects. State-and-year dummy variables should account for the independent
effects of any shocks specific to cities in the same state and year, such as the passage of
statewide collective bargaining laws. Our estimates in Table A3 are substantively similar to
those presented in Table 1 of the paper. For both police and fire protection average wage, the
effects of unionization in Table A3 are actually larger than those shown in the paper. For fire
protection employment and payroll expenditures, the effects of unionization are slightly
dampened by the inclusion of state-and-year fixed effects, although we still estimate large and
statistically significant effects. The effect of unionization on police protection employment in
column 7 is not statistically distinguishable from zero when we include state-year fixed effects,
although it is still positive. Finally, we estimate a 3.5% effect of unionization on police
protection payroll per capita.
Finally, in Table A4, we exclude the municipality fixed effects and instead include only
regional dummy variables and year fixed effects. Across the board, our coefficient estimates on
the Union variable are considerably larger than in Table 1. The one exception is column 7:
When we do not account for time-constant unobserved characteristics of municipalities, we
actually estimate a statistically insignificant effect of police unionization on per capita police
protection employment levels. Other than that, the effects of police and fire unionization are
positive and significant in all models.
54
2. Collective Bargaining and Cities’ Expenditures on Salaries and Benefits, 1992-2010 As we describe in the paper, the dependent variables for the second empirical study come
from the annual Police and Fire Personnel, Salaries, and Expenditures Surveys conducted by the
International City/County Management Association (ICMA). For our analysis of municipal fire
and police departments, we use the datasets corresponding to the surveys from 1992 to 2010.
Since the response rates to the surveys were typically 45 to 50%, most cities appear in the dataset
in some years but not others.
Prior to carrying out the analysis, we corrected some obvious errors in the original
datasets. We identified most of these errors by examining within-city, over-time patterns in
salary and health benefits expenditures. The most common type of error was one in which a
single observation within a city was missing a digit at the end.40 Whenever it seemed clear that
this was the case, we added a zero to the end of the observation with the missing digit. We also
fixed observations for which it was apparent that a data point was missing the first digit. In
situations where our inspection of within-city, over-time patterns in salary and health benefits
expenditures identified cases that were clearly erroneous but for which we could not determine
how to correct the data, we excluded the erroneous observations from our analysis. We also
excluded cases for which the ratio of health benefits expenditures to salary expenditures was
implausibly high: our models do not include any city-year observations for which expenditures
on health benefits were more than 60% of expenditures on salaries. All together, these data
corrections affected 5-6% of the observations.41
40 An example of this would be a city that reported spending $2 million on salaries in 1996, $210,000 in 1997, and $2.3 million in 1998. 41 Specifically, we adjusted 811 values of police salary expenditures (40 of which were recoded as missing) and 859 values of police health benefits expenditures (280 of which were recoded as missing). For fire protection, we
55
To construct the main independent variables (whether a city’s police protection workers
or fire protection workers had collective bargaining in a given year), we relied on three groups of
datasets. The first was a series of surveys conducted every three to four years by the Bureau of
Justice Statistics: the Law Enforcement Management and Administrative Statistics (LEMAS)
surveys. This survey samples state and local law enforcement agencies, many of which are
municipal police departments. In 1987, 1990, 1993, 1997, 2000, and 2003, the LEMAS
questionnaires asked the agencies whether their sworn police officers had collective bargaining.42
The second set of surveys comes from ICMA. In 1988 and 1999, ICMA conducted a
special Labor Management Relations survey of cities with at least 10,000 in population. Those
surveys asked the chief administrative officer of each city a series of questions about union
membership and the collective bargaining status of various groups of city employees, including
police protection employees and fire protection employees.
Lastly, we took advantage of a special 1977 survey conducted by the Census of
Governments. That year, for each government in the U.S. that indicated that it had at least one
collective bargaining unit, the Census Bureau documented which particular groups of employees
were part of those bargaining units.43
By piecing these various survey datasets together, we were able to determine the
collective bargaining status of police and fire protection employees in most cities and years for
which we had ICMA police and fire personnel data. Starting with the LEMAS surveys, we
appended the six years of data together and eliminated law enforcement agencies that were not
adjusted 415 cases for salary expenditures (14 were recoded to missing) and 246 cases for fire health benefits (89 were recoded to missing). We also corrected two obvious errors in fire protection employment and one in police protection employment. Lastly, we excluded one police protection employment observation because it had a very large influence on our estimates. 42 Specifically, they asked, “Is collective bargaining authorized for your employees?” 43 The data are available through ICPSR, study 8179.
56
municipal police or public safety departments.44 We then merged in each municipality’s Census
FIPS code by matching on the municipalities’ names. Finally, we created a variable equal to one
if the municipal police department in a given year had collective bargaining and zero otherwise.
Next, we appended together the 1988 and 1999 ICMA Labor Management Relations
datasets. We used the responses to four different questions to create the collective bargaining
variables for police and fire protection employees in those years. Specifically, we coded the
police (fire) collective bargaining variable equal to one if one or more of the following was true:
The city reported that a positive number of sworn police officers (firefighters) were covered by a collective bargaining contract,
The city reported the year of first contract with police officers (firefighters),45
The city reported that police (fire) wages were incorporated into the city’s collective
bargaining contract(s), or
The city reported that police (fire) benefits were incorporated into the city’s collective bargaining contract(s).
In addition, for city departments that provided a year of first contract, we coded the collective
bargaining variable in that year and in subsequent years as a one. For years prior to that year, we
coded the collective bargaining variable as zero. For any city that reported having zero police
(firefighters) covered by a contract, that did not report the year of first contract, and that reported
that police (fire) wages and benefits were not incorporated into contracts, we coded the collective
bargaining variable as zero. In addition, in the 1999 survey, if the city reported that there were
no unions in the city, we coded both collective bargaining variables as zero for that year.
44 The 1987 LEMAS data did not contain city names for the various municipal police departments, so we filled in the city names by matching the agency identification numbers from 1987 with those of future years. For cases where that did not work, we were sometimes able to fill in the city name using the name of the agency. However, we still had to drop several observations in the 1987 LEMAS data because we could not figure out what cities they were in. 45 We eliminated implausible cases that reported that the first collective bargaining contract was signed in 1900 or 1941.
57
We then combined the collective bargaining information from the LEMAS datasets and
the ICMA Labor Management Relations datasets. In inspecting the patterns of collective
bargaining status for police and fire employees within each city, we found a few cities that were
coded as having collective bargaining in one year and not a later year. Rather than investigate
the collective bargaining histories of these cities as Freeman et al. had done with a sample of the
Census of Governments data, we took a conservative approach and dropped them from the
analysis. For all of the other cities, the coding of cities’ collective bargaining status was clear:
they either had collective bargaining from 1992 to 2010 or did not, and in a very small number of
cases, they acquired it sometime during that time period. For these unambiguous cases, we were
able to fill in the collective bargaining status of police and fire employees in most of the years:
If a city’s police or fire employees did not have collective bargaining in one year, we can assume
that they did not have it in years prior to that. If a city’s police or fire employees did have
collective bargaining in one year, we can assume that they also had it in subsequent years. In
addition, since most unionization activity occurred prior to the mid-1980s, if a city did not have
collective bargaining in the last year it responded to the surveys, it almost certainly did not have
it in subsequent years, and we coded it accordingly.
In a final step, we used the special Bargaining Unit Survey conducted as part of the
Census of Governments in 1977 to create lists of all cities that had bargaining units for fire
employees and law enforcement or security employees as of that year. For all city police and fire
departments on those two lists, we coded the collective bargaining variables in our LEMAS-
ICMA dataset as one for all years. The final product was a city-by-year dataset with two critical
variables: whether the city’s police protection employees had collective bargaining, and whether
58
the city’s fire protection employees had collective bargaining. We merged this dataset with the
ICMA police and fire personnel data.
Most of the control variables for the study came from the U.S. Census. We extracted the
following city-level variables from both the 1990 and 2000 Censuses of Population: population,
population density (population per square mile), percentage of adults with a college degree,
income per capita, median rent, the percentage of the population living in poverty, percent
African American, percent Hispanic, the percentage of the population enrolled in elementary or
high school, and the percentage of employed persons working in manufacturing. We were able
to obtain some of the same variables from the 2010 Census of Population as well: population,
percent African American, and percent Hispanic. For the remaining variables, we used the
Census Bureau’s American Community Survey five-year estimates from 2005 to 2009. We then
interpolated the values of each variable for years between 1992 and 2010.46 The population
growth variable in our models is the percentage growth from 1990 to 2010.47
To account for the likelihood that more liberal cities are more likely to have collective
bargaining and to better compensate public sector workers, we merged in the county-level two-
party vote share for Al Gore in the 2000 presidential election. In addition, since state-level
factors influenced whether states eventually got laws permitting or requiring collective
bargaining, which in turn influenced which cities got collective bargaining, we merged in a
measure of the worker-friendliness of each state: the percentage of private sector workers who
were members of unions in each year, as documented by Hirsch and MacPherson (2003, 2011).
46 For the post-2000 values of the variables we obtained from the American Community Survey, we set the 2005-2009 five-year estimates as the values for 2007 and interpolated accordingly. 47 We excluded one city from the police protection analysis because it had an implausibly large value of population growth, most likely due to an error in the data.
59
As we describe in the paper, we also add to the dataset measures of city-level
expenditures and employment from the Census of Governments of 1957: total city payroll, total
police employment, and total fire protection employment. Importantly, however, not all of the
cities in our 1992-2010 dataset existed in 1957. Moreover, the variables we use from the 1957
Census were only collected for cities that had at least 5,000 in population in that year. Therefore,
we are missing these variables for a number of cities that we use in our analysis.
In Tables A5 to A7 below, we present empirical results that we describe but do not show
in the paper. First, in Table A5, we test whether there is evidence of a “threat effect” in the
modern period. Specifically, do nonunion cities with salaries or health benefits that are low
relative to similar cities in the same state increase their salaries in subsequent years in order to
stave off unionization? To test this, we interact the collective bargaining indicator with the
lagged deviation variable we describe in the paper. The results in Table A5 show that there is no
evidence of a threat effect: the coefficient on the lagged deviation from the state average is
positive – not negative – and there is no significant difference in the effects for cities with and
without collective bargaining.
In Table A6, we present the results of the same models as those in Table 2 of the paper
but including two additional variables: logged total city payroll per capita in 1957, and logged
fire protection employment per capita in 1957. The inclusion of these two variables changes our
estimates of the effect of collective bargaining only modestly, as we discuss in the paper. The
same is true for Table A7, where we add two variables to the models from Table 3: logged total
city payroll per capita in 1957, and logged police protection employment per capita in 1957. The
estimates of the effect of collective bargaining are substantively similar to those in Table 3.
60
Lastly, to ensure that our estimates of the effect of collective bargaining are not strongly
influenced by outliers or leverage points, we run the models using robust estimation rather than
OLS.48 We use the lmRob function in the robust package in R, which chooses an appropriate
algorithm to compute a final robust estimate with high efficiency and a high breakdown point.
For our models, the initial estimation is done with M-S estimation since there are factors in our
prediction matrix, and the final estimates are determined by MM-estimation.
The MM-estimates presented in Tables A8 and A9 are very similar to those generated
using OLS. The effect of collective bargaining on average fire protection salary is slightly
higher in Table A8 than in Table 2 of the paper, and the effect on per employee health benefits is
slightly lower. Unlike the OLS estimates, we estimate a small negative effect of collective
bargaining on fire protection employment when using MM-estimation, but the effects on per
capita salary and health benefits expenditures are still large, positive, and significant.
For police, the robust estimates of the effect of collective bargaining on per employee
salaries, per employee health benefits, and per capita employment are slightly greater in
magnitude than the corresponding OLS estimates. Because the effect on per capita employment
is more negative in Table A9 than in Table 3 of the paper, the robust estimates of the effect of
collective bargaining on per capita salary and health benefits expenditures are slightly smaller
than the OLS estimates. Thus, our OLS estimates presented in the paper are not heavily
influenced by outliers or leverage points.
48 For an overview of robust estimation, see Andersen (2007).
61
REFERENCES
Andersen, Robert. 2007. Modern Methods for Robust Regression. Los Angeles: Sage Publications. Freeman, Richard B., Casey Ichniowski, and Jeffrey Zax. 1988. “Appendix A: Collective Organization of Labor in the Public Sector.” In Richard Freeman and Casey Ichniowski, eds., When Public Sector Workers Unionize. Chicago: University of Chicago Press. Hirsch, Barry T., and David A. MacPherson. 2003. “Union Membership and Coverage Database from the Current Population Survey: Note.” Industrial and Labor Relations Review 56 (2): 349-54. Hirsch, Barry T., and David A. Macpherson. 2011. Union Membership and Coverage Database from the CPS, available at http://unionstats.com/. Sekhon, Jasjeet S. 2011. “Multivariate and Propensity Score Matching Software with Automated Balance Optimization: The Matching package for R.” Journal of Statistical Software 42 (7): 1-52.
62
Table A1: Differences Between Union and Nonunion Cities, 1972
Cities with Fire Unions
Cities without Fire Unions
T-test p-value
Cities with Police Unions
Cities without Police Unions
T-test p-value
Ln(Population) 10.448 9.845 0.000 10.348 9.871 0.000 SES 0.046 -0.096 0.000 0.094 -0.078 0.000 % African American 0.075 0.133 0.000 0.065 0.131 0.000 % Hispanic 0.046 0.060 0.093 0.046 0.054 0.290 % Manufacturing 0.247 0.232 0.030 0.253 0.227 0.000 % in School 0.221 0.228 0.005 0.225 0.228 0.158 % in Poverty 0.313 0.378 0.000 0.299 0.370 0.000 Population Growth, '72-'77 0.045 0.076 0.000 0.052 0.069 0.033
63
Table A2: Census of Governments Data Analysis -- Conservative Coding of Union
Fire Protection Employees Police Protection Employees (1) (2) (3) (4) (5) (6) (7) (8)
Average Wage Employment Payroll Average Wage Employment Payroll Union 0.03 0.03 0.083 0.103 0.018 0.018 0.027 0.038
(0.013)*** (0.013)*** (0.024)*** (0.025)*** (0.011)* (0.011)* (0.012)** (0.014)***Ln(Population) 0.13 0.129 -0.067 0.093 0.107 0.106 -0.221 -0.115
(0.035)*** (0.035)*** (0.073) (0.081) (0.028)*** (0.029)*** (0.032)*** (0.040)***SES 0.078 0.078 0.053 0.128 0.077 0.078 0.095 0.164
(0.026)*** (0.026)*** (0.039) (0.045)*** (0.019)*** (0.019)*** (0.022)*** (0.027)***% Black -0.022 -0.025 0.257 0.24 -0.123 -0.125 0.368 0.27
(0.104) (0.104) (0.194) (0.224) (0.104) (0.104) (0.127)*** (0.161)* % Hispanic 0.435 0.435 -0.548 -0.113 0.368 0.37 0.111 0.464
(0.151)*** (0.152)*** (0.236)** (0.276) (0.122)*** (0.122)*** (0.144) (0.180)** % Manufacturing 0.151 0.149 0.401 0.458 0.105 0.102 0.396 0.585
(0.117) (0.117) (0.260) (0.262)* (0.104) (0.104) (0.131)*** (0.153)***% in School -0.053 -0.046 -0.531 -0.689 -0.17 -0.162 -0.594 -0.715
(0.262) (0.262) (0.549) (0.596) (0.218) (0.219) (0.235)** (0.292)** % in Poverty -0.361 -0.362 -0.416 -0.784 -0.352 -0.353 -0.111 -0.407
(0.091)*** (0.091)*** (0.169)** (0.188)*** (0.079)*** (0.079)*** (0.086) (0.107)***Population Growth 0.019 0.018 0.216 0.24 0.036 0.035 0.18 0.233
(0.042) (0.042) (0.069)*** (0.080)*** (0.035) (0.035) (0.036)*** (0.045)***Lagged Deviation from State Avg. -0.062 -0.036 -0.058 -0.031
(0.026)** (0.045) (0.021)*** (0.030) Union X -0.043 -0.049 Lagged Deviation from State Avg. (0.055) (0.038) Observations 3565 3565 3912 3912 4174 4174 4724 4724 Unique municipalities 913 913 978 978 1072 1072 1181 1181 R-squared 0.87 0.87 0.94 0.93 0.87 0.87 0.89 0.89 Notes: Robust standard errors clustered by municipality in parentheses. All models include municipality fixed effects and year fixed effects. Dependent variable in columns 1-2 and 5-6 is logged average wage. Dependent variable in columns 3 and 7 is logged per capita employment. Dependent variable in columns 4 and 8 is logged per capita payroll expenditures. Hypothesis tests on Union are one-tailed except in columns 3 and 7; all other tests are two-tailed. * significant at 10%; ** significant at 5%; *** significant at 1%
64
Table A3: Census of Governments Data Analysis - With State-Year Fixed Effects
Fire Protection Employees Police Protection Employees (1) (2) (3) (4) (5) (6) (7) (8)
Average Wage Employment Payroll Average Wage Employment Payroll Union 0.045 0.045 0.06 0.093 0.026 0.026 0.016 0.034
(0.011)*** (0.011)*** (0.020)*** (0.021)*** (0.009)*** (0.009)*** (0.010) (0.012)***Ln(Population) 0.127 0.127 -0.077 0.068 0.101 0.1 -0.271 -0.175
(0.034)*** (0.035)*** (0.066) (0.070) (0.028)*** (0.028)*** (0.029)*** (0.034)***SES 0.058 0.058 0.111 0.164 0.05 0.051 0.096 0.14
(0.022)*** (0.022)*** (0.037)*** (0.042)*** (0.018)*** (0.018)*** (0.021)*** (0.025)***% Black 0.003 0.002 -0.386 -0.367 -0.098 -0.1 -0.013 -0.069
(0.093) (0.093) (0.212)* (0.235) (0.090) (0.090) (0.133) (0.168) % Hispanic 0.223 0.223 -0.414 -0.196 0.197 0.197 0.099 0.263
(0.144) (0.145) (0.284) (0.333) (0.112)* (0.112)* (0.125) (0.156)* % Manufacturing 0.012 0.013 0.074 0.049 -0.018 -0.021 0.098 0.154
(0.114) (0.114) (0.227) (0.235) (0.100) (0.100) (0.126) (0.147) % in School 0.372 0.377 -0.296 -0.113 0.136 0.15 -0.643 -0.49
(0.210)* (0.210)* (0.517) (0.559) (0.183) (0.183) (0.209)*** (0.248)** % in Poverty -0.232 -0.233 0.27 0.054 -0.155 -0.156 0.063 -0.075
(0.082)*** (0.083)*** (0.180) (0.194) (0.075)** (0.075)** (0.087) (0.108) Population Growth -0.02 -0.02 0.302 0.291 0.005 0.004 0.194 0.202
(0.034) (0.034) (0.059)*** (0.065)*** (0.030) (0.030) (0.030)*** (0.037)***Lagged Deviation from State Avg. -0.077 -0.059 -0.07 -0.036
(0.020)*** (0.036) (0.017)*** (0.026) Union X -0.027 -0.055 Lagged Deviation from State Avg. (0.044) (0.032)* Observations 5149 5149 5600 5600 5959 5959 6756 6756 Unique municipalities 1316 1316 1400 1400 1527 1527 1689 1689 R-squared 0.87 0.87 0.93 0.92 0.87 0.87 0.89 0.88 Notes: Robust standard errors clustered by municipality in parentheses. All models include municipality fixed effects as well as state-year fixed effects. Hypothesis tests on Union are one-tailed except in columns 3 and 7; all other tests are two-tailed. * significant at 10%; ** significant at 5%; *** significant at 1%.
65
Table A4: Census of Governments Data Analysis - Regional Fixed Effects
Fire Protection Employees Police Protection Employees (1) (2) (3) (4) (5) (6) (7) (8)
Average Wage Employment Payroll Average Wage Employment Payroll
Union 0.091 0.091 0.339 0.43 0.113 0.113 0.004 0.115 (0.009)*** (0.009)*** (0.048)*** (0.048)*** (0.008)*** (0.008)*** (0.024) (0.025)***
Ln(Population) 0.068 0.069 0.098 0.164 0.052 0.052 0.011 0.06 (0.004)*** (0.004)*** (0.016)*** (0.017)*** (0.004)*** (0.004)*** (0.010) (0.010)***
SES 0.077 0.078 -0.007 0.072 0.085 0.085 0.027 0.11 (0.007)*** (0.007)*** (0.033) (0.033)** (0.005)*** (0.005)*** (0.014)* (0.015)***
% Black 0.278 0.278 0.355 0.645 0.283 0.283 0.95 1.243 (0.033)*** (0.033)*** (0.120)*** (0.124)*** (0.031)*** (0.031)*** (0.063)*** (0.072)***
% Hispanic 0.384 0.383 -0.75 -0.373 0.409 0.408 0.27 0.681 (0.040)*** (0.040)*** (0.182)*** (0.187)** (0.032)*** (0.032)*** (0.058)*** (0.066)***
% Manufacturing 0.252 0.252 0.586 0.87 0.208 0.208 -0.027 0.203 (0.041)*** (0.041)*** (0.158)*** (0.157)*** (0.036)*** (0.036)*** (0.076) (0.081)**
% in School -0.548 -0.544 -1.649 -2.089 -0.461 -0.46 -1.582 -1.973 (0.132)*** (0.131)*** (0.561)*** (0.562)*** (0.109)*** (0.109)*** (0.241)*** (0.248)***
% Poverty -0.952 -0.949 0.968 0.011 -1.008 -1.007 -0.262 -1.292 (0.059)*** (0.059)*** (0.252)*** (0.253) (0.053)*** (0.053)*** (0.119)** (0.125)***
Population Growth 0.012 0.01 0.172 0.182 0.033 0.033 0.347 0.397 (0.031) (0.031) (0.107) (0.110)* (0.025) (0.025) (0.049)*** (0.052)***
Lagged Deviation from State Avg. 0.177 0.138 0.142 0.126 (0.025)*** (0.045)*** (0.019)*** (0.033)***
Union X 0.058 0.025 Lagged Deviation from State Avg. (0.053) (0.040) Observations 5149 5149 5600 5600 5959 5959 6756 6756 R-squared 0.62 0.62 0.17 0.23 0.62 0.62 0.22 0.35
Notes: Robust standard errors clustered by municipality in parentheses. Models include regional fixed effects and year fixed effects (no municipality fixed effects). Hypothesis tests on Union are one-tailed except in columns 3 and 7; all other tests are two-tailed. * significant at 10%; ** significant at 5%; *** significant at 1%
66
Table A5: ICMA Data Analysis and the Threat Effect
(1) (2) (3) (4)
Fire salary expenditures / employee
Fire health expenditures / employee
Police salary expenditures / employee
Police health expenditures / employee
Collective Bargaining 0.087 0.224 0.098 0.19 (0.016)*** (0.024)*** (0.010)*** (0.016)***
Ln(Population) 0.054 0.027 0.044 0.006 (0.006)*** (0.010)*** (0.004)*** (0.008)
SES 0.024 -0.021 0.024 -0.012 (0.010)** (0.019) (0.007)*** (0.014)
Ln(Median Rent) 0.383 0.356 0.312 0.249 (0.042)*** (0.067)*** (0.027)*** (0.041)***
% Democrat 0.187 0.088 0.088 0.02 (0.055)*** (0.097) (0.034)** (0.066)
Ln(Population Density) 0.035 0.023 0.036 0.044 (0.010)*** (0.017) (0.006)*** (0.010)***
% in Poverty -0.265 0.082 -0.354 -0.031 (0.133)** (0.255) (0.069)*** (0.134)
% Black -0.07 -0.266 0.017 -0.221 (0.050) (0.101)*** (0.030) (0.064)***
% Hispanic 0.144 -0.007 0.131 -0.046 (0.053)*** (0.089) (0.033)*** (0.060)
% in School 0.091 0.62 -0.162 0.243 (0.204) (0.354)* (0.117) (0.221)
% Manufacturing 0.214 0.861 0.119 0.531 (0.082)*** (0.141)*** (0.055)** (0.093)***
% Private Sector Union 1.654 2.646 1.604 2.872 (0.179)*** (0.344)*** (0.119)*** (0.231)***
Population Growth -0.024 -0.017 -0.018 -0.015 (0.013)* (0.019) (0.006)*** (0.014)
Lagged Deviation from State Avg. 0.159 0.403 0.172 0.489 (0.065)** (0.049)*** (0.049)*** (0.048)***
Collective Bargaining X -0.044 -0.001 0.023 -0.003 Lagged Deviation from State Avg. (0.069) (0.067) (0.057) (0.081) Observations 6897 6009 13,227 11,531 R-squared 0.51 0.55 0.55 0.6
Notes: Robust standard errors clustered by municipality in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%
67
Table A6: Collective Bargaining and Fire Protection Employment and Compensation - with 1957 Controls
(1) (2) (3) (4) (5)
Salary expenditures / employee
Health expenditures / employee
Employment Salary expenditures per capita
Health expenditures per capita
Collective Bargaining 0.082 0.223 -0.065 0.026 0.189 (0.017)*** (0.027)*** (0.035)* (0.036) (0.046)***
Ln(Population) 0.05 0.013 -0.035 0.015 -0.032 (0.007)*** (0.012) (0.013)*** (0.014) (0.019)
SES 0 -0.032 0.114 0.116 0.06 (0.013) (0.023) (0.032)*** (0.033)*** (0.044)
Ln(Median Rent) 0.513 0.402 -0.387 0.099 -0.092 (0.045)*** (0.079)*** (0.092)*** (0.095) (0.131)
% Democrat 0.14 0.055 -0.181 -0.027 0.101 (0.063)** (0.111) (0.124) (0.123) (0.176)
Ln(Population Density) 0.03 0.021 0.002 0.044 0.03 (0.011)*** (0.020) (0.022) (0.022)* (0.033)
% in Poverty -0.038 0.356 -0.036 -0.111 -0.07 (0.137) (0.282) (0.261) (0.258) (0.381)
% Black -0.148 -0.395 0.489 0.359 0.154 (0.058)** (0.119)*** (0.109)*** (0.112)*** (0.165)
% Hispanic 0.065 -0.088 -0.341 -0.272 -0.439 (0.057) (0.100) (0.140)** (0.145)* (0.179)**
% in School 0.291 0.868 -0.34 -0.128 0.202 (0.221) (0.440)** (0.501) (0.492) (0.650)
% Manufacturing 0.178 0.765 0.219 0.417 1.039 (0.089)** (0.150)*** (0.186) (0.184)** (0.260)***
% Private Sector Unions 1.675 2.311 0.169 1.682 2.318 (0.191)*** (0.358)*** (0.422) (0.424)*** (0.545)***
Population Growth -0.019 0.001 -0.103 -0.116 -0.062 (0.020) (0.022) (0.027)*** (0.036)*** (0.046)
Lagged Deviation from State Avg. 0.109 0.415 (0.025)*** (0.043)***
Ln(1957 Payroll) 0.041 0.106 0.091 0.134 0.187 (0.011)*** (0.024)*** (0.025)*** (0.026)*** (0.040)***
Ln(1957 Fire Employment) -0.003 0.025 0.049 0.038 0.098 (0.012) (0.021) (0.028)* (0.028) (0.039)**
Observations 5255 4564 6649 6427 5688 R-squared 0.55 0.57 0.35 0.37 0.38
Notes: Robust standard errors clustered by municipality in parentheses. All models include regional dummies and time dummies. Hypothesis tests on Collective Bargaining are one-tailed in all but column 3; all other tests are two-tailed.* significant at 10%; ** significant at 5%; *** significant at 1%.
68
Table A7: Collective Bargaining and Police Protection Employment and Compensation - with 1957 Controls
(1) (2) (3) (4) (5)
Salary expenditures / employee
Health expenditures / employee
Employment Salary expenditures per capita
Health expenditures per capita
Collective Bargaining 0.09 0.192 -0.046 0.044 0.164 (0.013)*** (0.020)*** (0.020)** (0.021)** (0.034)***
Ln(Population) 0.043 -0.005 -0.029 0.012 -0.043 (0.005)*** (0.010) (0.009)*** (0.009) (0.017)**
SES 0.01 -0.024 0.041 0.056 0.022 (0.008) (0.018) (0.019)** (0.019)*** (0.036)
Ln(Median Rent) 0.352 0.275 -0.028 0.312 0.202 (0.030)*** (0.055)*** (0.057) (0.055)*** (0.092)**
% Democrat 0.108 0.017 -0.039 0.053 0.028 (0.043)** (0.080) (0.074) (0.072) (0.131)
Ln(Population Density) 0.038 0.064 0.001 0.039 0.054 (0.007)*** (0.014)*** (0.012) (0.013)*** (0.026)**
% in Poverty -0.322 0.02 0.333 -0.011 0.035 (0.079)*** (0.159) (0.165)** (0.166) (0.293)
% Black -0.036 -0.292 0.753 0.741 0.555 (0.038) (0.074)*** (0.069)*** (0.073)*** (0.135)***
% Hispanic 0.079 -0.131 0.138 0.224 0.129 (0.037)** (0.074)* (0.073)* (0.072)*** (0.131)
% in School -0.264 0.388 -0.904 -1.155 -0.846 (0.139)* (0.303) (0.274)*** (0.280)*** (0.544)
% Manufacturing 0.119 0.524 0.034 0.167 0.638 (0.063)* (0.109)*** (0.112) (0.112) (0.198)***
% Private Sector Unions 1.648 2.707 -0.306 1.266 2.447 (0.140)*** (0.265)*** (0.238) (0.237)*** (0.414)***
Population Growth 0 0.014 -0.076 -0.077 -0.076 (0.011) (0.017) (0.021)*** (0.022)*** (0.028)***
Lagged Deviation from State Avg. 0.187 0.504 (0.025)*** (0.065)***
Ln(1957 Payroll) 0.016 0.026 0.079 0.093 0.14 (0.007)** (0.017) (0.014)*** (0.014)*** (0.027)***
Ln(1957 Police Employment) 0.004 0.008 0.105 0.108 0.111 (0.010) (0.021) (0.020)*** (0.020)*** (0.040)***
Observations 8,837 7,745 11,126 10,530 6,265 R-squared 0.58 0.61 0.35 0.51 0.3
Notes: Robust standard errors clustered by municipality in parentheses. All models include region and year dummies. Hypothesis tests on Collective Bargaining are one-tailed in all but column 3; all other tests are two-tailed.* significant at 10%; ** significant at 5%; *** significant at 1%
69
Table A8: Collective Bargaining and Fire Protection Employment and Compensation - Robust Estimates
(1) (2) (3) (4) (5)
Salary expenditures / employee
Health expenditures / employee
Employment Salary expenditures per capita
Health expenditures per capita
Collective Bargaining 0.095 0.205 -0.017 0.058 0.197 (0.006)*** (0.013)*** (0.010)* (0.011)*** (0.018)***
Ln(Population) 0.052 0.023 -0.037 0.026 -0.015 (0.003)*** (0.005)*** (0.004)*** (0.005)*** (0.007)**
SES 0.018 -0.032 0.118 0.119 0.091 (0.005)*** (0.010)*** (0.007)*** (0.008)*** (0.014)***
Ln(Median Rent) 0.399 0.377 -0.343 0.076 0.004 (0.015)*** (0.032)*** (0.024)*** (0.026)*** (0.044)
% Democrat 0.187 0.146 -0.044 0.196 0.238 (0.022)*** (0.048)*** (0.036) (0.041)*** (0.066)***
Ln(Population Density) 0.024 0.019 -0.015 0.017 -0.016 (0.004)*** (0.009)** (0.006)** (0.007)** (0.012)
% in Poverty -0.308 0.115 0.055 -0.195 -0.154 (0.047)*** (0.102) (0.077) (0.086)** (0.139)
% Black -0.046 -0.257 0.5 0.412 0.299 (0.021)** (0.045)*** (0.033)*** (0.037)*** (0.059)***
% Hispanic 0.167 -0.017 -0.195 -0.14 -0.188 (0.021)*** (0.045) (0.035)*** (0.040)*** (0.063)***
% in School 0.059 0.547 -1.228 -0.984 -0.478 (0.081) (0.169)*** (0.132)*** (0.148)*** (0.232)**
% Manufacturing 0.199 0.772 0.496 0.72 1.361 (0.032)*** (0.069)*** (0.054)*** (0.061)*** (0.097)***
% Private Sector Unions 1.731 2.49 0.102 1.906 2.911 (0.081)*** (0.172)*** (0.132) (0.150)*** (0.238)***
Population Growth -0.033 -0.019 -0.07 -0.104 -0.093 (0.005)*** (0.010)** (0.008)*** (0.009)*** (0.013)***
Lagged Deviation from State Avg. 0.135 0.488 (0.008)*** (0.012)***
Observations 6897 6009 8809 8471 7530 Multiple R-squared 0.59 0.54 0.35 0.28 0.33
Notes: Results are MM-estimates generated using the lmRob function in the robust package in R. All models include regional dummies and time dummies. Hypothesis tests on Collective Bargaining are one-tailed in all but column 3. * significant at 10%; ** significant at 5%; *** significant at 1%.
70
Table A9: Collective Bargaining and Police Protection Employment and Compensation - Robust Estimates
(1) (2) (3) (4) (5)
Salary expenditures / employee
Health expenditures / employee
Employment Salary expenditures per capita
Health expenditures per capita
Collective Bargaining 0.102 0.191 -0.059 0.035 0.146 (0.004)*** (0.009)*** (0.007)*** (0.007)*** (0.011)***
Ln(Population) 0.045 0.006 -0.033 0.019 -0.02 (0.002)*** (0.004) (0.003)*** (0.003)*** (0.005)***
SES 0.019 -0.025 0.019 0.033 -0.005 (0.003)*** (0.007)*** (0.005)*** (0.005)*** (0.008)
Ln(Median Rent) 0.324 0.27 0.07 0.402 0.315 (0.009)*** (0.021)*** (0.015)*** (0.017)*** (0.027)***
% Democrat 0.119 0.059 0.11 0.247 0.293 (0.014)*** (0.031)* (0.024)*** (0.026)*** (0.040)***
Ln(Population Density) 0.031 0.032 0 0.041 0.052 (0.002)*** (0.005)*** (0.004) (0.004)*** (0.007)***
% in Poverty -0.351 -0.091 0.342 -0.004 0.162 (0.031)*** (0.067) (0.055)*** (0.057) (0.087)*
% Black 0.011 -0.249 0.74 0.723 0.506 (0.014) (0.031)*** (0.023)*** (0.025)*** (0.039)***
% Hispanic 0.128 -0.048 0.166 0.25 0.104 (0.014)*** (0.030) (0.023)*** (0.025)*** (0.039)***
% in School -0.253 0.03 -1.417 -1.383 -1.344 (0.053)*** (0.114) (0.087)*** (0.094)*** (0.146)***
% Manufacturing 0.137 0.451 0.013 0.169 0.694 (0.022)*** (0.049)*** (0.037) (0.040)*** (0.062)***
% Private Sector Unions 1.549 2.747 -0.395 1.311 2.563 (0.056)*** (0.123)*** (0.091)*** (0.098)*** (0.155)***
Population Growth -0.017 -0.013 -0.06 -0.096 -0.058 (0.003)*** (0.006)** (0.004)*** (0.005)*** (0.006)***
Lagged Deviation from State Avg. 0.237 0.518 (0.008)*** (0.009)***
Observations 13,227 11,531 16,809 15,865 13,976 Multiple R-squared 0.58 0.54 0.27 0.33 0.35
Notes: Results are MM-estimates generated using the lmRob function in the robust package in R. Hypothesis tests on Collective Bargaining are one-tailed in all but column 3. All models include regional dummies and time dummies.* significant at 10%; ** significant at 5%; *** significant at 1%.
71
9 10 11 12 13
0.0
0.4
0.8
Ln(Population)
Den
sity
-2 -1 0 1 2 3
0.0
0.4
0.8
SES
Den
sity
0.0 0.1 0.2 0.3 0.4
04
812
% AfricanAmerican
Den
sity
0.00 0.02 0.04 0.06 0.08 0.10
020
40
% Hispanic
Den
sity
0.0 0.2 0.4 0.6
0.0
1.5
3.0
% Manufacturing
Den
sity
0.05 0.15 0.25 0.350
48
12
% in Elementary orHigh School
Den
sity
0.0 0.1 0.2 0.3 0.4 0.5 0.6
01
23
4
% in Poverty
Den
sity
-0.5 0.0 0.5 1.0
01
23
45
Population Grow th,1972-1977
Den
sity
Figure A1: Distributions of Independent Variables in 1972, Cities with and without Unions of Fire Protection Employees
Notes: Solid lines are distributions for cities in which fire protection employees formed unions either prior to 1972 or between 1972 and 1987. Dashed lines are distributions for cities in which fire protection employees never formed unions.
72
9 10 11 12 13
0.0
0.4
0.8
Ln(Population)
Den
sity
-2 -1 0 1 2 3 4
0.0
0.4
0.8
SES
Den
sity
0.0 0.1 0.2 0.3 0.4
05
10
% African American
Den
sity
0.00 0.02 0.04 0.06 0.08 0.10
010
30
% Hispanic
Den
sity
0.0 0.2 0.4 0.6
0.0
1.0
2.0
3.0
% Manufacturing
Den
sity
0.05 0.15 0.25 0.35
04
8
% in Elementary orHigh School
Den
sity
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
01
23
4
% in Poverty
Den
sity
-0.5 0.0 0.5 1.0
01
23
45
Population Grow th,1972-1977
Den
sity
Figure A2: Distributions of Independent Variables in 1972, Cities with and without Unions of Police Protection Employees
Notes: Solid lines are distributions for cities in which police protection employees formed unions either prior to 1972 or between 1972 and 1987. Dashed lines are distributions for cities in which police protection employees never formed unions.