The Growth of Local Government Christopher Berry Harris School of Public Policy The University of Chicago Jeff Grogger Harris School of Public Policy The University of Chicago Martin West Harvard Graduate School of Education This Draft: June 2015
Government has been growing at least as long as scholars have been studying it. Wagner (1893) observed rapid growth of government in parts of Europe during the 19th century. Fabricant (1952) and Peacock and Wise (1961) found high growth rates in the US and UK during the first half of the 20th century, even after accounting for war-‐related expenditures. Figure 1 shows government spending as a share of U.S. GDP over the past century. It rose from 7 percent in 1903 to 41 percent in 2010. Government has expanded similarly rapidly in many OECD countries (e.g., Lindert 1996; Tanzi and Schuknechy 2000). Most studies of public-‐sector growth have focused on national governments.1 We focus on U.S. local governments, the growth of which is not only of substantive importance but also calls into question conventional explanations of government growth. Local government revenue equaled 7.5 percent of GDP in 2010 and local governments employ more workers than state and federal governments combined. Moreover, in the post-‐WWII period, the local government sector has grown faster than the national government, making it a natural focus for anyone interested in the growth of government. The rapid growth of local government is also at odds with conventional wisdom. It contradicts the notion that Tiebout competition should keep local government in check (Brennan and Buchanan 1980). Nor can it have much to do with growth of the welfare state, since local government spends little on welfare. 1 See Mueller (2003, ch. 21) for a review. An exception is the “local Leviathan” literature that emerged in the late 1980s (see Oates 1989). However, this literature generally analyzed cross-‐sectional variation in the size of local government rather than its growth over time.
1
Thus even if increasing demand for redistribution explains the growth of national governments (Lindert 1996, 2004), it cannot explain growth in local government. Finally, focusing on local government lets us study places that are losing population. Population loss is rare at the national level but quite common at the local level. Surprisingly, local government expands even when population falls. Of 602 counties that experienced sustained population declines between 1960 and 2000, 528 had more public employees at the end of the period than the beginning. I. Background A large literature on the growth of government has yielded many hypotheses but little consensus. We do not attempt to provide a comprehensive survey of the literature but rather focus on the leading explanations that might plausibly apply to local government. Wagner (1893) was one of the first to observe the phenomenon and proposed perhaps the earliest theory the growth of government (see Biehl 1998; Duveral and Henrekson 2011). He noted a positive relationship between the level of economic development in a country and the scope of its government. Wagner posited that government-‐provided infrastructure complements private inputs in the production of economic output, such that government expands disproportionately as the economy grows. Although best thought of as a description of an empirical regularity more than a well-‐formed theory, “Wagner’s Law” is nevertheless regularly cited as a leading explanation for the growth of government and a substantial empirical literature has sought to “test” it (Peacock and Scott 2000).
2
Baumol’s “cost disease” hypothesis (1967) is another integral part of the debate over the growth of government. Baumol suggested that government is essentially a labor-‐intensive service industry and that demand for its services is inelastic. Services have not experienced the same productivity gains over time as have more capital-‐intensive industries, such as manufacturing. Therefore, increasing real wage rates over time imply that the real cost of public sector services will rise relative to private goods. If the relative price of government services increases and demand is price inelastic, total government spending will increase. Scholars generally appear to accept that Baumol’s “disease” is at least a partial explanation for the growth of government, though how much can be explained by this mechanism remains a matter of some debate (Ferris and West 1996). A simple, if more mechanical explanation for the growth of government is the so-‐called ratchet effect attributed to Bird (1971, 1972). The basic hypothesis is that government spending increases apace with the private economy during expansionary periods but declines more slowly than private income during an economic downturn. As a result, government spending as a share of the economy increases over time through the cumulative effects of business cycles. The precise reason why spending does not fall during recessions is not specified. There is no obvious countercyclical fiscal policy enacted at the local level, however, so it is unclear what mechanism would produce a local ratchet. One possibility, which we explore below, is that countercyclical intergovernmental aid props up public employment in declining localities.
3
Another large area of research focuses on the rise of “the welfare state” and the expansion of social spending (e.g., Lindert 1996, 2004). A common theme in this literature is that greater employment and income volatility in the private sector, possibly generated by globalization, gives rise to increasing demands for social insurance programs (Rodrik 1998). While this theory may explain growth in the federal government, it is an unlikely candidate to explain the expansion of the local sector, since local social insurance and welfare spending are negligible. One of the few accounts of government growth to give specific consideration to local government is from Brennan and Buchanan (1980), who model government as a “Leviathan” whose sole objective is to maximize revenue. According to this view, elections do no provide sufficient control over rapacious politicians and constitutional limits on government’s tax power are required to reign in fiscal excess. The authors largely exempt local government from their critique, however. Inspired by Tiebout (1956), Brennan and Buchanan posit intergovernmental competition as a powerful constraint on government expansion. Specifically, they suggest that Tiebout sorting and competition are “partial or possibly complete substitutes for explicit fiscal constraints on the taxing power” (1980, p. 184). The rapid expansion of local government in the U.S. would appear to belie the competition-‐constrains-‐Leviathan hypothesis, however, a topic we will explore further below.
4
II. Why Local Government? Although the local public sector has not received much attention in the literature, it has several features that make it a natural focus for studies of government growth. The local sector is large and expanding rapidly. As Figure 2 demonstrates, local government own-‐source revenue (i.e., excluding intergovernmental transfers) was equivalent to 7.5 percent of GDP in 2010. Moreover, when looking at public employment rather than revenue, the local sector dominates. As seen in Figure 3, local government accounted for 47 employees per 1000 in 2009, compared with 17 for state governments and 9 for the federal government. Simply put, the local sector is where the actual work of government gets done. Local government has also grown faster than the federal government in the post-‐WWII era. Indeed, the federal government workforce has actually shrunk over time (Figure 3) while federal revenue has been stagnant (Figure 2).2 In contrast, local public employment and revenue have grown almost constantly over the past 60 years. Even more striking than the pace of local sector growth is its ubiquity. Figure 4 depicts changes in public sector employment between 1962 and 2002 for all counties with a starting population of at least 25,000, which we henceforth refer to as “large” counties. Public employment is measured as the total number of employees for all local governments within the county. Of the 1186 counties
2 Federal spending has risen over this period, as has the federal deficit (not shown).
5
included in the analysis, all but 10 had more public employees in 2002 than they did in 1962. The ubiquity of local public sector growth might come as little surprise if all counties had grown over time. However, a substantial number of counties experienced sustained population declines over this period. Declining population is rare at the national level, though it is common at the local level. Of our 1186 large counties, 187 have experienced sizeable population reductions. These counties run the gamut from relatively small places to once-‐giant industrial centers such as Cleveland, Detroit, and Pittsburgh. Among declining US counties, local government almost always expands. Figure 5 shows change in population and change in public employment for the same set of large counties. Among growing counties, there is a consistently positive, nearly linear relationship between population and public employment growth. On average, for each 20 people added to the population, one worker will be added to the local government workforce. Among declining counties, however, commensurate cuts to local government are not evident. Of the 187 large counties that declined in population between 1962 and 2002, only 3 shed public sector workers. In Figure 6 we report the same relationship in log form, since we estimate statistical models in logs in the next section. The figure shows that those counties with less proportional population growth had lower proportional growth in their public workforces. Nevertheless, their public workforces grew rapidly. The important point in Figure 6 lies in the y-‐intercept, marked with a large X: among
6
counties experiencing zero population growth, the public sector workforce grew on average by 63 percent. Furthermore, nearly all of the data points, including those to the left of the y-‐axis, are above the x-‐axis. It is important to note that Figure 6 depicts 40-‐year changes. This means that the failure to reduce public employment does not simply reflect slow adjustment due to the difficulty of firing public workers. Over a 40-‐year span, the public workforce can only grow if workers are not just replaced but multiplied. Or course, population is not the only factor that influences the size of the local public sector. If public services are normal goods and income grows, then demand for public services should rise, holding population constant. Changes in the age distribution of the population may also change the demand for public goods, particularly as the baby boom progressed through school and now as the population has begun to age. Numerous other factors could likewise influence growth in the local public sector. In the remainder of the paper, we attempt to explain the growth in local government over the last half of the 20th century. Section III presents our analytical approach, in which we estimate regression models that relate changes in the size of local government to changes in various demand factors. Section IV presents results. We report regressions for total local government revenue and employment, as well as intergovernmental transfers. Section V analyzes the contribution of public sector unionization to the growth of local government. Section VI explores the policy implications of local government growth, with a focus on public sector pension
7
liabilities, while section VII concludes by discussing the implications of our findings for theories of government growth more generally. III. Empirical Strategy and Data To explain the growth of local government, we consider factors such as rising incomes, a changing age distribution in the population, and rising levels of public-‐ sector unionization. We also introduce a number of factors that are specific to local government, such as inter-‐jurisdictional competition and governmental overlap, as well as features of the built environment and intergovernmental transfers. Our data on public employment and finance come from the Census of Governments (COG), which reports data for all local governments at 5-‐year intervals, for years ending in 2 and 7. We use COG finance data for 1957 through 2002 and COG public employment data for 1962 to 2002, beginning with the earliest year available in each case. We aggregated both data sets to the county level. These county-‐level observations represent the aggregation of all local governments in a county area. An advantage of this form of aggregation is that we need not be concerned with shifting functional responsibilities across types of local governments over time.3 Our demographic and housing variables come from the decennial Census of Population. We linearly interpolated values from the 1960 through 2000 population censuses to match the COG years. Altogether, we have data
3 For example, in 2002 New York City’s school system was reorganized and control
given to the mayor, resulting in what would appear to be a massive increase in city employment. That is, teachers formerly deemed to be employees of the independent school district were then classified as city employees. Such shifting of responsibilities will not influence the county aggregate employment numbers.
8
for 3,134 county areas observed at 5-‐year intervals from 1962 through 2002, for a total of 28,064 observations. Our basic regression model takes the form (1)
ln Ect = α c + ηt + β lnYct + γ ln N ct + Xctδ + uct ,
where E ct is local government expenditure in county c at time t; αc and ηt are county and year fixed effects, respectively; Yct is county per capita income at time t, Nct is county population at time t, Xct if a vector of control variables, and uct is an unobservable disturbance term.4 The terms β, γ, and δ are parameters to be estimated. Peltzman (1980) suggests that (1) can be interpreted a demand equation for local public expenditures in an environment where the wage of local public sector workers, which is effectively the “price” of local public spending, varies proportionately with local income. We also estimate regressions like (1) where we replace local public expenditures with local public employment and with inter-‐ governmental expenditures.
Means of the variables used in our regressions appear in Table 1 for selected
years between 1962 and 2002. The first row of Panel A shows that real revenues more than quadrupled over our sample period, rising from $91,425 to $397,028. The share of revenues stemming from intergovernmental transfers, rather than own sources, rose from 30 to 40 percent. Local public employment, measured on a full-‐ time equivalent basis, rose by a factor of 2.5 over the same time period.
4 Versions of (1) where expenditures and income are expressed in per capita terms
yield similar results.
9
The first row of the Panel B tells a familiar story. Real family income rose
rapidly between 1962 and 1972, was roughly constant over the next twenty years, then rose modestly between 1992 and 2002. This suggests that increasing real incomes may explain part of the growth in local public spending during the first and last parts of our sample period, but cannot explain growth during the period from 1972 to 1992, when real local public spending increased by 63 percent. Mean county population rose from 59,426 in 1962 to 91,749 in 2002, a gain of 54 percent. This implies that local public expenditures rose not only in real terms between 1962 and 2002, but in real terms per capita.
The remaining variables in Table 1 serve as controls in our regression
models. Control variables with strong trends are the most likely candidates to explain the growth in the local public sector. The share of population over 65 and the share under 18 are associated with higher demand for local public goods and services (e.g., Poterba 1997). The share of elderly in the population grew over time, consistent with higher demand for public services, but the share of young people fell, which all else equal should lead to a reduction in demand for public services, particularly schooling. The population became more educated over time, whereas the black share of the population and vacancy rates were roughly constant. Persons per housing unit, a measure of population density, fell considerably, from 3.45 in 1962 to 2.59 in 2002. If the cost of providing public services is lower in denser places, then the decline in persons per housing unit could help explain why the local public sector has grown.
10
The last three variables in the table measure different aspects of the political institutions within the county; these variables are available on a consistent basis in the COG only beginning in 1972. The first is the number of municipalities per county, which provides an indicator of the level of Tiebout competition. Tiebout competition has grown over time, but very slightly. The second variable is the number of distinct functions performed by the governments in the county.5 One reason the local public sector has grown may be that local governments have expanded the number of services they provide. In the average county, 1.5 new services were added between 1962 and 2002. The final variable is the share of spending in the county attributed to special districts. Berry (2010) studied special-‐ purpose governments, such as school districts and park districts, that overlap both each other and general-‐purpose municipal governments. This overlap gives rise to a common-‐pool problem in the local tax base that causes areas with higher numbers of special-‐purpose governments to have higher levels of local spending. The number of such overlapping governments has grown sharply over time and the share of spending due to special districts doubled, from roughly 5 to 10 percent of the county total.6
5 The COG classifies 37 distinct functions performed by local governments. Where
any government in a county provides a given service, we count it as being provided in the county. If multiple governments provide the same service, we count the service as being provided once. So this index in principle ranges from 1 to 37 and will increase whenever a new service is added by at least one government that was not previously provided by any of the other constituent local governments. 6 The share of spending controlled by special districts is one simple measure of the extent of jurisdictional overlap. See Berry (2010) for a more detailed discussion. Other measures, such as counting the number of special districts or the ratio of special purpose to general-‐purpose jurisdictions, yield comparable results in the models reported below.
11
IV. Results
Table 2 presents our basic regression specification, using the log number of
public employees (full-‐time equivalent) as the dependent variable. We report results for all counties and separately for growing and declining counties. Declining counties are identified as those whose peak population occurred in 1957 or earlier. Growing counties are defined as those whose peak population was attained after 2002. These definitions ensure that we have the same counties in each category over time, so that our results—in particular, our estimates of the unexplained time trends in the models—are not influenced by compositional changes.
Consistent with Figure 6, population change is an important determinant of
local public employment. The elasticity of public employment with respect to population is 0.83 in growing counties and 0.69 in declining counties, although we cannot reject that they are equal. The income elasticity of public employment is lower, 0.26 in growing counties and a statistically insignificant 0.07 in declining counties. In other words, public employment less responsive to income growth in declining than in growing places.
The age composition of the county population also influences the expansion
of the local public sector. Having a larger share of children in the population significantly increases public employment in both growing and declining counties, likely due to demand for schooling, which is one of the largest employment categories for local government. Interestingly, having a larger share of the population over 65 is significantly negatively related to public employment growth, but only in declining counties. This may be because the elderly are fiscal
12
conservatives and particularly resist spending on elementary education once their own children have left school (Poterba 1997), although it is not clear why this would be the case especially in declining counties. Perhaps unsurprisingly, declining counties have a larger share of residents who are over 65, 18.7 percent as of 2002, compared with 13.3 percent for the average growing county.
A growing share of the population with a BA degree is associated with slower
growth in public employment, although significant only for declining counties. This result could arise because the educated place fewer demands on public services or because they are more politically informed. Relatedly, a growing share of the population that is African American is associated with slower growth in public employment, again only in declining places. The mechanism underlying this result is not obvious.
The two housing-‐related variables also influence public employment growth
in the manner predicted. Declining household size (the number of persons per occupied housing unit) is associated with greater expansion of the public workforce, consistent with the hypothesis that it takes more workers to serve a more physically dispersed population. This is true in both growing and declining places. The housing vacancy rate (the share of housing units unoccupied) is also positively associated with public sector employment, possibly because some public services, such as police and fire patrols, must be provided even to vacant units. Somewhat surprisingly, the relationship is significant only for growing counties.
While most of the variables in Table 2 yield the expected effects, perhaps the
most notable result from the analysis is the size of the “unexplained” component of
13
public sector growth, as reflected in the year dummies. The 2002 year coefficient for all counties is 0.61 log points, relative to 1962, the omitted base year, even after controlling for such basic factors as population and income growth. Moreover, the unexplained component of growth is significantly larger in declining than in growing counties. Declining counties experienced a remarkable 0.74 log point (i.e., roughly 110 percent) increase in public sector employment that cannot be explained the variables included in our model.
One component of the expansion of the local public workforce has been well
documented in prior work; namely, the expansion of public education employment associated with significant reductions in class size over roughly the period we are studying (see Hanushek and Rivkin [1997]). If all of the expansion we observe in public employment is confined to education, then we have not yet shown anything new. Table 3 decomposes public employment into education and non-‐education functions and replicates our basic regression models.
In terms of responsiveness to income, population, and the other included
covariates, education employment follows roughly similar patterns to total employment, as seen in Table 2. Of particular note, the unexplained component of growth, captured in the year dummies, is larger for educational employment than for total employment, and nearly equal for growing in declining counties. In short, educational employment expanded dramatically over time in both growing and declining counties and a large share of the growth is unexplained by changes in income, population, or the school-‐age population.
14
The models of non-‐education employment (columns 4 to 6) deliver an
important result. The unexplained component of growth is large and significant for declining counties but not for growing counties. Specifically the 0.64 log point coefficient for declining counties in 2002 indicates that non-‐education public employment grew by roughly 90% beyond what can be explained by population, income, and the other variables in the model. Meanwhile, the year dummies are significantly smaller for the growing counties and most cannot be distinguished from zero. These results imply that there is a large unexplained increase in non-‐ education employment for declining counties but not for growing counties.
Among the other variables, population change strongly predicts non-‐
education employment in both growing and declining counties, while income growth is significant only for the growing counties. Changes in the percent college graduates and percent African American continue to be negatively associated with expansion of the public workforce, while the vacancy rate demonstrates a puzzling negative result in declining counties (model 6). An obvious question springs from the results shown in Tables 2 and 3. How are local governments financing the expansion of their workforces over time? The two basic components of local government revenue are own-‐source revenue, which includes property and sales taxes, fees, and all other forms of revenue raised by local governments themselves, and intergovernmental aid, which includes grants and other transfers from the state and federal governments to local governments. Tables 4 and 5, respectively, present our models of these two revenue streams.
15
The own-‐source revenue models (Table 4) are generally consistent with our
basic models of public employment (Table 2). Population and income growth are again seen to be important drivers of the expansion of the local public sector. The elasticity of own-‐source revenue with respect to population is just over 1 in both growing and declining counties. The income-‐elasticity of own-‐source revenue is lower than the population elasticity, 0.59 for growing counties and 0.29 for declining counties. The difference between the two groups of counties is significant.
The coefficients on the other control variables are generally consistent with
the results from the employment models, with the notable exception of the population under 18 variable, which is perversely negatively signed in the own-‐ source revenue model. That is, counties with a larger share of children raise less revenue from own-‐sources. The reason becomes clear in light of the intergovernmental revenue results shown in Table 5. Counties with more children receive significantly more intergovernmental aid. Apparently, counties with a larger fraction of children substitute intergovernmental revenue for their own revenue in order to finance public employment, presumably in the area of education.
The unexplained component of growth again stands out, and again the trend
is more pronounced for the declining counties, which experienced a 0.65 log point (91 percent) expansion of local own-‐source revenue that cannot be explained by changes in population, income, or any of the other variables in our model. The corresponding figure for growing counties is 0.38 log points (46 percent).7 7 This is relative to 1957, rather than 1962 as previously. Revenue data are
available since 1957, as compared to 1962 for FTE employment.
16
The model of intergovernmental revenue, shown in Table 5, does not contain
any surprises. Population growth is associated with increases in intergovernmental aid. The relationship is roughly the same in growing and declining places, suggesting that declining counties lose some intergovernmental revenue as their populations shrink. Intergovernmental aid is negatively related to income growth, although the relationship does not attain statistical significance. As noted above, a growing share of children in the population leads to increased intergovernmental transfers. The latter relationship is not significant for declining counties, although we cannot reject the hypothesis that the coefficients for growing and declining counties are equal.
The unexplained component of growth is especially large in the
intergovernmental revenue regressions, amounting to 1.8 log points relative to 1957 for growing counties and 1.6 log points for declining counties, reflecting the roughly fivefold growth in intergovernmental revenue during this period, in real terms, already evident from Table 1.
In short, intergovernmental aid increased substantially over this period and
likely explains some of the growth in local public employment. However, there are no significant differences between growing and declining counties in the intergovernmental aid models, suggesting this source of revenue cannot explain the differential growth in public employment between the two groups of counties. Finally, we examine three variables specific to local government institutional organization and functional performance. Because functional categorizations were standardized in the COG beginning only in 1972, we examine a shorter 30-‐year panel of data for this analysis. Table 6 shows that as new governmental functions
17
are added within the county, public employment increases, although the effect is small substantively and only significant for growing counties. The share of spending by special districts is strongly positively associated with public employment, consistent with Berry (2010). An increase from 5 to 10 percent of total spending by special districts—the average increase over this period (per Table 1)—would lead to about a 2 percent increase in public employment.8 The number of municipalities in the county, a measure of Tiebout competition, is actually positively associated with increases in public employment, although only significantly so in declining counties. In sum, the addition of new government functions and the growth of special districts do appear to contribute to the growth of local public employment over time, but (a) these two factors are relatively small contributors to the overall growth of government, and (b) they cannot explain the differential growth in declining vs. growing counties. V. Unions Perhaps the leading potential explanation for the unexplained growth in local government documented in the previous section is the unionization of the public sector. Prior to 1960, collective bargaining was virtually non-‐existent in the local public sector and few government workers belonged to unions. Over the next two decades, however, most states enacted laws sanctioning collective bargaining for state and local workers and, in many cases, imposing on local governments a duty to bargain with unionized employees. These laws generated an explosion of organizing
8 The effect of special districts on own-‐source revenue (not shown) is substantially
larger: a 5-‐percentage point increase in the share of spending by special districts is associated with a 6.8 percent increase in own-‐source revenue.
18
activity among government workers, facilitating the emergence of public sector unions as a potent political force at all levels of American government (Freeman 1986).
One distinguishing feature of public sector collective bargaining is the ability
of unions to influence the demand for labor through the political process. In the private sector, unions typically seek to improve compensation for current members at the expense of growth in employment. Public sector unions, on the other hand, can lobby for larger budgets that in theory would enable increases in both compensation and employment. Higher employment may be an attractive goal for unions to pursue to the extent that it translates into more members and, therefore, additional resources for political activity. For any given budget, however, existing employees can be better compensated if there are fewer of them. Therefore, while there are strong reasons to expect the adoption of public sector collective bargaining to lead to higher wages for public employees, the expected impact on employment levels is theoretically ambiguous (Matsusaka 2009).9 To shed light on the role of unionization in the growth of local government, we incorporate into equation (1) an additional variable characterizing the evolving legal environment for public sector collective bargaining in each state. Our data come from the NBER Public Sector Bargaining Law Data Set (see Valetta and Freeman 1988), which provides annual information for 1955-‐1996 on the rights 9 Early research based primarily on cross-‐sectional comparisons indicated that collective bargaining coverage increased wages in the public sector by a lesser amount than it does in the private sector, but that these higher wages were not offset by the reductions in employment observed in private industry (see, e.g., Freeman and Ichniowski 1988).
19
afforded to each of five employee groups: state employees, teachers, police, firefighters, and other local workers. We use this information to create a dummy variable for each of these groups indicating whether the state had imposed a “duty to bargain” on their employers and calculate the average of this variable across all five groups.10 The mean value of this index increases from zero in 1955 to 0.47 in 1997, with most of the growth occurring between 1967 (µ=0.05) and 1977 (µ=0.40). Because the underlying data on state collective bargaining policies are available only through 1996, we assign the 1996 value of this index for each state to its 1997 observation in the COG but exclude 2002 from the analysis. Columns 1-‐3 of Table 6 present estimates of the effect of state collective bargaining policy on log total employment (full-‐time equivalent) overall and in growing and declining counties. Perhaps surprisingly, we find that the enactment of laws favorable to collective bargaining reduces the level of local public employment. Specifically, the imposition of a duty to bargain for all five categories of public employees is associated with an 8 percent reduction in the number of local employees. Although the point estimate of the effect of bargaining environment is modestly smaller in declining counties, we cannot reject the hypothesis that the effect is the same in both samples. Given that a majority of states adopted duty to bargain laws during this time period, the inclusion of this variable in our model actually increases the unexplained growth in public employment (i.e., the coefficients on the year dummies) over what 10 Previous research exploiting differences in the timing of the passage of state bargaining laws for public employees confirms that the adoption of mandatory bargaining laws substantially increased union membership and collective bargaining coverage (Ichniowski 1988; Saltzman 1985, 1988).
20
was observed in Table 2. And the apparently restraining effect of collective bargaining policies favorable to unions is not limited to employment: Parallel analyses (not shown) indicate that the existence of a duty to bargain is also associated with a reduction in own source revenue and is unrelated to revenue from inter-‐governmental grants. Columns 4 of Table 7 estimate the effect of state collective bargaining policy on (log) average wages, calculated by dividing total payroll expenses by the total number of employees. As expected, the results indicate that the existence of a duty to bargain increases average wages by roughly 4 percent. Columns 5 and 6 indicate that this overall result is driven entirely by growing counties: there is essentially no relationship between state collective bargaining policy in declining counties, and we are able to distinguish this estimate from the parallel estimate for growing counties at the 90-‐percent confidence level. This may indicate that union bargaining power is limited in counties with a declining resource base. More generally, the results in Table 7 suggest that favorable collective bargaining policies lead to higher average wages for local public employees but cannot account for the overall growth of government employment and expenditure. VI. Discussion The results of the regression models reinforce conclusions from the preceding graphical analyses while shining new light on differences between growing and declining places. Local public sector employment and revenue increased dramatically over the period we analyze. Our models explain some of the growth in local government, but much remains unexplained. Two results stand out.
21
First, unexplained growth in local government is greater in declining areas than in expanding areas. Perhaps the most surprising finding from our analysis is the difference in the growth of non-‐educational employment between growing and declining counties. After accounting for income growth and the other variables in our model, non-‐educational employment grows by 90 percent in declining counties. In expanding counties, unexplained growth is insignificantly different from zero. Second, the unexplained component of educational employment is large and can be seen in both growing and declining counties. Factors outside our models may explain at least some of the growth in educational employment (see Hanushek and Rivkin 1997), but we are at a loss to explain the dramatic growth of government employment and revenue in declining places. Public sector unions, a seemingly likely culprit, appear to play at best a minor role. The expansion of the public sector in declining places has several worrisome implications. Rising per capita tax obligations required to finance a burgeoning public sector workforce may increase out-‐migration, reduce in-‐migration, and crowd out private investment. Thus, rather than stimulating the local economy, rising public employment and associated taxes may contribute to the spiral of decline in places already suffering an exodus of jobs and population. Moreover, large public payrolls can leave a lasting fiscal legacy . Since most public sector workers participate in defined-‐benefit pension systems, high public sector employment today implies fiscal obligations well into the future. The ratio of retirees to current residents increases naturally with local population losses; today’s population may flee, but yesterday’s workers and their pensions remain on
22
the books. When public employment grows as population declines, the future burdens are compounded. The taxpayers in a city like Cleveland, which has lost roughly half its population, are essentially paying the historical pension obligations of a city twice as large. Some data underscore these concerns. We obtained data on the current pension liabilities of the 25 largest independent local pension plans from Novy-‐Marx and Rauh (2012). We matched these data to the historical population growth rates of each locality. As shown in Figure 7, there is a strong inverse relationship between population growth and pension liabilities. Cities that declined in population have substantially larger pension liabilities per household than cities that grew. This would not necessarily be a problem for today’s taxpayers if cities had historically fully funded their pensions. As Figure 8 shows, however, unfunded liabilities follow roughly the same pattern. In essence, it is possible for a city to grow its way out of a pension mess by attracting new residents and thereby boosting the ratio of taxpayers to retirees. But it is equally possible for a city to shrink its way into a pension mess. And adding employees during times of decline, as many places do, only adds to future pension liabilities. Our findings about the expansion of local government raise questions about the growth of government generally. Several of the leading theories of government growth are hard to square with the evidence we see in the local public sector. While the precise meaning of “Wagner’s Law” is open to debate (Peacock and Scott 2000), one version posits that local public goods complement private inputs in the production of economic output. This version of the law is plainly broken by
23
declining counties, where government expands even as the private economy contracts. “Baumol’s disease” fares only marginally better. Baumol’s theory suggests that the growth of government spending is a price effect arising from low productivity growth in the labor-‐intensive public sector. At the federal level, this is largely what we see: spending has increased substantially over time while the size of the federal workforce has been stable or even fallen. But we see in local government a dramatic expansion in the number of employees, which implies that much of the increase in local government spending is a quantity effect, not a price effect. This is a case where looking at the local sector is especially illuminating. Brennan and Buchanan’s competition-‐constrains-‐Leviathan theory is doubly challenged by our findings. First, the local sector has grown faster then the national government in the post-‐WWII era. Second, when we attempt to explicitly account for local government competition, as measured by the number of municipalities in the county, we find that the variable is either insignificant or positively related to government growth. It was obvious that theories of the welfare state cannot explain the expansion of local government even before seeing any regression results. Given that local government plays a negligible role in social insurance and related welfare state programs, another explanation is needed for the growth of the local public sector. One possibility is that two different explanations account for the growth of local and national governments, respectively. Another is that one explanation underlies both and, if so, it has little to do with the expansion of the welfare state.
24
Whatever the explanation, the issues we have identified are unlikely to disappear any time soon. Among U.S. counties, population decline is common: over the period 1940-‐2000, 217 counties declined every decade and another 2,107 declined during at least one decade. Fewer than one-‐in-‐three counties grew in every decade. Local government decline is not unique to the U.S.: European cities such as Bremen, Liverpool, and Rotterdam have been losing population for decades (Glaeser 2012). Nor is decline a unique phenomenon of local government. Until recently national-‐level populations have generally grown, but with prolonged declines in birth rates in much of the developed world, national-‐level population loss becomes increasingly likely. Decline is a growing phenomenon that requires us to better understand the link between population change and government growth.
25
Government Spending as Percent GDP 1903-2012 Total Government Spending as Pct GDP 10 20 30 40 50
(Federal + State + Local)
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 Year
Figure 1: The Growth of U.S. Government
26
0
Direct Revenue as Pct GDP 5 10 15 20
25
Government Revenue as Percent GDP 1903-2010
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 year Federal Local
State
Figure 2: Government Own-‐Source Revenue by Sector
27
Public Employees per 1000 Capita 10 20 30 40 50
Government Employment by Sector, 1946-2009
1940
1950
1960
1970
1980
1990
2000
2010
year4 State Federal
Figure 3: The Growth of Public Sector Employment
Local
28
County-level Changes in Public Employment 1962-2002 2002 Public Employees per 1000 Capita 0 20 40 60 80 100
Counties with 1962 Population > 25000
0
10
20 30 40 1962 Public Employees per 1000 Capita
50
Solid line = 45 degrees
Figure 4: Pervasive Local Public Sector Growth
29
Change in Pop. vs. Change in Public Employment 1962-2002
0
Change in Public Employment 50000 100000150000200000250000
Counties Over 25000 Population in 1962
-1000000
0
1000000 2000000 Change in Population Declining counties
3000000
4000000
Growing counties
Figure 5: Antigravity Change in Pop. vs. Change in Public Employment 1962-2002 Log Change in Public Employment 0 1 2 3 4
Counties Over 25000 Population in 1962
-1
X
-1
0
1 Log Change in Population
Declining counties
Figure 6: Log Changes
2
3
Growing counties
30
Popula&on)change)(196232007))and)Pension)Liabili&es:) 25)Largest)Independent)Local)Pension)Plans) 11
Lowess smoother San Francisco
Log Liabilities per HH 9.5 10 10.5
New York City Chicago Boston Detroit Hartford
Los Angeles
Milwaukee Baltimore Cincinnati
Philadelphia St Paul Tacoma Memphis Miami
Dallas Jacksonville Houston Nashville Davidson
San Jose
Seattle Tampa
8.5
9
Fresno City
San Antonio
-1
0
1 Log Pop Change
Phoenix
2
3
bandwidth = .8
Figure 7: Local Pension Liabilities
Liabili&es)data)from)Novy3Marx)and)Rauh)(2012))
31
Popula'on)change)(196232007))and)Unfunded)Pension) Liabili'es:)25)Largest)Local)Pension)Plans) 11
Lowess smoother
Log Unfunded Liabilities per HH 9 10
Chicago New York City San Francisco Boston
Detroit Philadelphia Cincinnati Baltimore Milwaukee Hartford St Paul
Los Angeles
Miami Seattle Tacoma Memphis
Dallas Jacksonville Houston Nashville Davidson
San Jose
8
Fresno City San Antonio
Phoenix
Tampa
-1
0
1 Log Pop Change
2
3
bandwidth = .8
Figure 8: Unfunded Pension Liabilities
Liabili'es)data)from)Novy3Marx)and)Rauh)(2012))
32
Table 1: Data means by year 1962
1972
1982
1992
2002
A. Dependent variables General Revenue Inter-governmental Revenue Own Source Revenue Total Employment (FTE)
91,425 27,757 63,669 1,433
181,605 68,508 113,097 2,166
209,168 86,793 122,375 2,484
296,044 111,433 184,611 2,980
397,028 158,942 238,085 3,628
B. Explanatory variables ln(Median Family Income) ln(Population) Pct Over 65 Pct Under 18 Pct BA Degree Pct Black Vacancy Rate Household Size Number of Functions Number of Municipalities in County Share of Spending by Special Districts Duty to Bargain Index
10.45 9.95 0.11 0.37 0.05 0.1 0.13 3.45 0.01
10.8 10 0.12 0.34 0.08 0.09 0.12 3.17 8 19.54 0.05 0.24
10.68 10.12 0.14 0.29 0.12 0.09 0.14 2.88 8.15 20.93 0.08 0.42
10.79 10.14 0.15 0.27 0.14 0.09 0.15 2.69 8.19 20.97 0.08 0.47
10.91 10.23 0.15 0.25 0.17 0.09 0.14 2.59 8.22 20.98 0.09 -
33
Table 2: Local Public Employment
ln(Median Family Income) ln(Population) Pct Over 65 Pct Under 18 Pct BA Degree Pct Black Vacancy Rate Household Size 1962 (Base Year) 1967 1972 1977 1982 1987 1992 1997 2002 Constant
All Counties (1) 0.163*** (0.0415) 0.837*** (0.0331) -0.509* (0.298) 0.709* (0.410) -0.282* (0.153) -0.235 (0.152) 0.220** (0.109) -0.112*** (0.0416) 0 (0) 0.113*** (0.0106) 0.236*** (0.0218) 0.348*** (0.0336) 0.389*** (0.0434) 0.452*** (0.0506) 0.510*** (0.0570) 0.567*** (0.0626) 0.610*** (0.0663) -3.659*** (0.564)
Growing Counties (2) 0.264*** (0.0584) 0.834*** (0.0416) 0.0597 (0.367) 1.042** (0.439) -0.123 (0.173) -0.115 (0.159) 0.356*** (0.118) -0.120* (0.0643) 0 (0) 0.0936*** (0.0137) 0.211*** (0.0235) 0.331*** (0.0320) 0.377*** (0.0391) 0.419*** (0.0471) 0.479*** (0.0520) 0.531*** (0.0564) 0.564*** (0.0589) -4.904*** (0.661)
Declining Counties (3) 0.0741 (0.0529) 0.694*** (0.0791) -1.243*** (0.363) 1.199*** (0.319) -1.183*** (0.388) -0.828** (0.361) 0.108 (0.174) -0.161*** (0.0429) 0 (0) 0.147*** (0.0167) 0.277*** (0.0343) 0.406*** (0.0531) 0.472*** (0.0671) 0.549*** (0.0712) 0.602*** (0.0795) 0.676*** (0.0866) 0.737*** (0.0906) -1.217* (0.694)
Significant Difference? **
***
** *
** *
* ** * * **
Observations 28,064 16,162 5,415 R-squared 0.835 0.891 0.455 Number of counties 3,134 1,796 602 Robust standard errors clustered by state in parentheses. Models include county fixed effects. *** p<0.01, ** p<0.05, * p<0.1
34
Table 3: Education vs. Non-Education Public Employment Education Employment Growing Declining All Counties Counties Counties (1) (2) (3) ln(Median Family Income) 0.112* 0.182*** 0.0430 (0.0596) (0.0597) (0.0705) ln(Population) 0.846*** 0.845*** 0.654*** (0.0317) (0.0357) (0.0749) Pct Over 65 -1.107*** -0.722** -2.147*** (0.324) (0.340) (0.654) Pct Under 18 1.603*** 2.042*** 1.741*** (0.533) (0.487) (0.556) Pct BA Degree -0.644*** -0.639*** -0.921** (0.133) (0.153) (0.360) Pct Black -0.326 -0.276 -0.680 (0.202) (0.167) (0.432) Vacancy Rate 0.162 0.179 0.525* (0.128) (0.133) (0.285) Household Size -0.105** -0.120** -0.172*** (0.0420) (0.0593) (0.0449) 1962 (Base Year) 0 0 0 (0) (0) (0) 1967 0.153*** 0.150*** 0.147*** (0.0141) (0.0153) (0.0251) 1972 0.311*** 0.305*** 0.309*** (0.0274) (0.0278) (0.0377) 1977 0.431*** 0.442*** 0.411*** (0.0431) (0.0382) (0.0641) 1982 0.495*** 0.510*** 0.484*** (0.0564) (0.0461) (0.0882) 1987 0.608*** 0.616*** 0.590*** (0.0651) (0.0555) (0.0919) 1992 0.678*** 0.688*** 0.645*** (0.0752) (0.0623) (0.104) 1997 0.758*** 0.768*** 0.725*** (0.0806) (0.0665) (0.112) 2002 0.820*** 0.823*** 0.784*** (0.0854) (0.0694) (0.120) Constant -3.971*** -4.859*** -1.152 (0.724) (0.680) (1.282)
Sig. Diff.?
** **
All Counties (4) 0.296*** (0.0804) 0.832*** (0.0520) 0.389 (0.510) -0.814 (0.507) 0.0569 (0.331) -0.195 (0.225) 0.266** (0.120) -0.143*** (0.0509) 0 (0) 0.0257 (0.0235) 0.0961** (0.0403) 0.198*** (0.0536) 0.211*** (0.0637) 0.188** (0.0757) 0.225*** (0.0818) 0.245*** (0.0910) 0.258*** (0.0950) -5.463*** (0.984)
Other Employment Growing Declining Counties Counties (5) (6) 0.461*** 0.105 (0.113) (0.101) 0.828*** 0.738*** (0.0652) (0.175) 1.543** -0.764 (0.665) (0.701) -0.450 -0.191 (0.554) (0.700) 0.471 -1.866** (0.348) (0.701) 0.0351 -1.191** (0.303) (0.550) 0.464** -0.331* (0.229) (0.192) -0.156* -0.171** (0.0841) (0.0734) 0 0 (0) (0) -0.0214 0.119*** (0.0286) (0.0281) 0.0352 0.214*** (0.0479) (0.0503) 0.133** 0.384*** (0.0559) (0.0828) 0.150** 0.439*** (0.0620) (0.111) 0.0858 0.476*** (0.0737) (0.124) 0.123 0.518*** (0.0787) (0.138) 0.131 0.580*** (0.0886) (0.151) 0.126 0.646*** (0.0913) (0.156) -7.471*** -2.355* (1.266) (1.328)
Observations 27,962 16,139 5,401 28,056 16,157 R-squared 0.801 0.877 0.330 0.644 0.734 Number of counties 3,127 1,796 602 3,134 1,796 Robust standard errors clustered by state in parentheses. Models include county fixed effects. *** p<0.01, ** p<0.05, * p<0.1
Sig. Diff.? **
***
*** * **
*** *** *** *** *** *** *** ***
5,414 0.272 602
35
Table 4: Own-Source Revenue
ln(Median Family Income) ln(Population) Pct Over 65 Pct Under 18 Pct BA Degree Pct Black Vacancy Rate Household Size 1957 (Base Year) 1962 1967 1972 1977 1982 1987 1992 1997 2002 Constant
All Counties (1) 0.437*** (0.0746) 1.143*** (0.0590) -0.484 (0.451) -1.457** (0.605) 0.337 (0.277) -0.480 (0.308) 0.665*** (0.158) -0.193*** (0.0610) 0 (0) 0.154*** (0.0209) 0.235*** (0.0339) 0.324*** (0.0503) 0.241*** (0.0637) 0.296*** (0.0771) 0.386*** (0.0939) 0.422*** (0.0985) 0.473*** (0.102) 0.495*** (0.110) -5.295*** (0.921)
Growing Counties (2) 0.586*** (0.0960) 1.078*** (0.0604) 0.565 (0.589) -0.928 (0.676) 0.742*** (0.265) -0.451 (0.285) 0.752*** (0.153) -0.208** (0.0934) 0 (0) 0.102*** (0.0267) 0.146*** (0.0422) 0.227*** (0.0586) 0.143** (0.0650) 0.196** (0.0751) 0.278*** (0.0911) 0.323*** (0.0957) 0.363*** (0.0933) 0.378*** (0.0982) -6.579*** (1.038)
Declining Counties (3) 0.284*** (0.0914) 1.160*** (0.182) -1.717*** (0.578) -1.379*** (0.505) -0.209 (0.663) -1.207 (0.911) 0.623* (0.325) -0.255*** (0.0911) 0 (0) 0.214*** (0.0291) 0.362*** (0.0503) 0.451*** (0.0826) 0.374*** (0.107) 0.449*** (0.110) 0.542*** (0.124) 0.548*** (0.139) 0.636*** (0.154) 0.649*** (0.168) -3.151* (1.706)
Significant Difference? ***
***
*** *** ** * ** **
* *
Observations 31,144 17,955 6,013 R-squared 0.833 0.879 0.533 Number of counties 3,134 1,796 602 Robust standard errors clustered by state in parentheses. Models include county fixed effects. *** p<0.01, ** p<0.05, * p<0.1
36
Table 5: Intergovernmental Revenue
ln(Median Family Income) ln(Population) Pct Over 65 Pct Under 18 Pct BA Degree Pct Black Vacancy Rate Household Size 1957 (Base Year) 1962 1967 1972 1977 1982 1987 1992 1997 2002 Constant
All Counties (1) -0.135 (0.111) 0.723*** (0.0470) -0.523 (0.681) 1.173** (0.528) 0.354 (0.433) 0.535** (0.238) -0.0855 (0.152) -0.0670 (0.0436) 0 (0) 0.304*** (0.0410) 0.649*** (0.0538) 0.945*** (0.0707) 1.209*** (0.0773) 1.179*** (0.0742) 1.322*** (0.0822) 1.422*** (0.0899) 1.563*** (0.0912) 1.706*** (0.0948) 2.852** (1.276)
Growing Counties (2) -0.169 (0.147) 0.758*** (0.0410) -0.347 (0.722) 1.886*** (0.644) 0.228 (0.423) 0.322 (0.260) -0.0823 (0.252) -0.0957 (0.0572) 0 (0) 0.332*** (0.0559) 0.689*** (0.0727) 0.991*** (0.0894) 1.260*** (0.0900) 1.223*** (0.0827) 1.393*** (0.0943) 1.491*** (0.103) 1.638*** (0.101) 1.790*** (0.105) 2.762* (1.522)
Declining Counties (3) -0.188 (0.143) 0.796*** (0.124) -1.035 (0.841) 1.364 (0.811) 0.703 (0.749) 1.013 (0.618) -0.153 (0.345) -0.225*** (0.0744) 0 (0) 0.270*** (0.0419) 0.635*** (0.0640) 0.915*** (0.0973) 1.152*** (0.126) 1.132*** (0.128) 1.248*** (0.142) 1.364*** (0.159) 1.520*** (0.158) 1.625*** (0.169) 2.946* (1.593)
Significant Difference?
Observations 31,143 17,955 6,012 R-squared 0.853 0.889 0.719 Number of counties 3,134 1,796 602 Robust standard errors clustered by state in parentheses. Models include county fixed effects. *** p<0.01, ** p<0.05, * p<0.1
37
Table 6: Public Employment (Short Panel)
ln(Per Capita Income) ln(Population) Pct Over 65 Pct Under 18 Pct BA Degree Pct Black Vacancy Rate Household Size Number of Functions Share of Spending by Special Districts Number of Municipalities in County 1972 (Base Year) 1977 1982 1987 1992 1997 2002 Constant
All Counties (1) 0.114*** (0.0206) 0.805*** (0.0303) -0.751** (0.314) 1.135*** (0.305) -0.295* (0.151) -0.374*** (0.131) 0.420*** (0.100) -0.0577* (0.0295) 0.00536*** (0.00117) 0.304*** (0.0288) 0.00241 (0.00257) 0 (0) 0.127*** (0.0136) 0.149*** (0.0205) 0.227*** (0.0243) 0.286*** (0.0293) 0.341*** (0.0315) 0.403*** (0.0338) -2.917*** (0.406)
Growing Counties (2) 0.164*** (0.0329) 0.822*** (0.0386) -0.607 (0.386) 1.565*** (0.319) -0.196 (0.170) -0.253* (0.141) 0.366*** (0.113) -0.109*** (0.0371) 0.00491*** (0.00130) 0.294*** (0.0309) 0.000600 (0.00226) 0 (0) 0.128*** (0.0164) 0.145*** (0.0252) 0.211*** (0.0311) 0.272*** (0.0366) 0.319*** (0.0381) 0.373*** (0.0399) -3.571*** (0.594)
Declining Counties (3) 0.0653** (0.0274) 0.668*** (0.0833) -0.982*** (0.300) 1.297*** (0.294) -0.945** (0.362) -0.759** (0.293) 0.124 (0.204) -0.0851*** (0.0306) 0.00252 (0.00175) 0.389*** (0.0694) 0.00800** (0.00300) 0 (0) 0.128*** (0.0168) 0.176*** (0.0266) 0.255*** (0.0297) 0.305*** (0.0346) 0.365*** (0.0356) 0.439*** (0.0385) -1.025 (0.845)
Significant Difference? **
*
*
Observations 21,479 12,379 4,186 R-squared 0.744 0.831 0.291 Number of counties 3,076 1,769 598 Robust standard errors clustered by state in parentheses. Models include county fixed effects. *** p<0.01, ** p<0.05, * p<0.1
38
Table 7: The Effect of Collective Bargaining Rights on Local Public Employment and Wages Local Public Employment Average Wages Growing Declining Significant Growing Declining Counties Difference? All Counties Counties Counties All Counties Counties (1) (2) (3) (1) (2) (3) ln(Median Family Income) 0.124*** 0.208*** 0.0430 ** 0.184*** 0.209*** 0.141*** (0.0381) (0.0534) (0.0477) (0.0381) (0.0407) (0.0518) ln(Population) 0.838*** 0.846*** 0.701*** 0.0839*** 0.0407** 0.288*** (0.0289) (0.0352) (0.0891) (0.0188) (0.0179) (0.0640) Pct Over 65 -0.516* -0.0249 -1.049** ** -0.346 -0.190 0.0662 (0.299) (0.334) (0.390) (0.212) (0.209) (0.307) Pct Under 18 0.639 0.872* 1.414*** -0.305 -0.234 0.0803 (0.436) (0.486) (0.338) (0.191) (0.216) (0.361) Pct BA Degree -0.273 -0.160 -1.069** * -0.00809 0.106 -0.483 (0.165) (0.183) (0.486) (0.124) (0.117) (0.342) Pct Black -0.263 -0.148 -0.845* 0.440*** 0.382*** 0.618** (0.185) (0.183) (0.483) (0.0969) (0.0939) (0.296) Vacancy Rate 0.185* 0.310** 0.0775 -0.0685* -0.121* 0.276* (0.101) (0.118) (0.170) (0.0395) (0.0672) (0.138) Household Size -0.100** -0.106* -0.169*** 0.00393 -0.00657 -0.0222 (0.0395) (0.0623) (0.0446) (0.00911) (0.00758) (0.0354) -0.0803*** -0.0800*** -0.0584** 0.0429** 0.0562*** 0.0111 Duty to Bargain? (0.0195) (0.0239) (0.0287) (0.0188) (0.0192) (0.0257) 1957 (Base Year for Wages) 0 0 0 (0) (0) (0) 1962 (Base Year for Employment) 0 0 0 0.210*** 0.206*** 0.230*** (0) (0) (0) (0.0160) (0.0178) (0.0210) 1967 0.123*** 0.107*** 0.153*** 0.273*** 0.270*** 0.302*** (0.00994) (0.0135) (0.0165) (0.0219) (0.0230) (0.0330) 1972 0.269*** 0.246*** 0.301*** 0.326*** 0.310*** 0.403*** (0.0206) (0.0238) (0.0364) (0.0302) (0.0319) (0.0455) 1977 0.395*** 0.379*** 0.443*** 0.282*** 0.258*** 0.399*** (0.0343) (0.0366) (0.0580) (0.0335) (0.0341) (0.0473) 1982 0.431*** 0.416*** 0.505*** 0.258*** 0.223*** 0.400*** (0.0436) (0.0430) (0.0721) (0.0327) (0.0321) (0.0412) 1987 0.502*** 0.466*** 0.586*** 0.307*** 0.294*** 0.400*** (0.0491) (0.0497) (0.0748) (0.0356) (0.0385) (0.0450) 1992 0.561*** 0.527*** 0.639*** 0.254*** 0.244*** 0.345*** (0.0555) (0.0550) (0.0837) (0.0341) (0.0369) (0.0495) 1997 0.620*** 0.582*** 0.716*** * 0.207*** 0.199*** 0.314*** (0.0607) (0.0584) (0.0922) (0.0368) (0.0394) (0.0545) Constant -3.263*** -4.399*** -1.042 4.793*** 4.980*** 3.156*** (0.517) (0.583) (0.751) (0.448) (0.455) (0.833) Observations 24,925 R-squared 0.820 Number of id 3,133 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
14,366 0.878 1,796
4,813 0.434 602
28,003 0.586 3,133
16,158 0.647 1,796
Significant Difference?
***
**
***
*
** *** *** ** * **
5,411 0.478 602
39