An American Resource Curse?

Since observers first noted that many countries blessed with rich natural resource deposits were, counterintuitively, fairly poor, substantial evidence has accumulated about when and why this is true. Applying this evidence to the case of coal mining localities in the United States is not always straightforward. But as a starting point, the basic economic pattern that indicates a ‘resource curse’ is present in US coal localities. Douglas and Walker (Reference Douglas and Walker2017) find that Appalachian counties endowed with greater coal resources experienced lower income growth than otherwise similar but less coal-rich counties. Matheis (Reference Matheis2016), examining a longer time period and larger set of counties, shows similar findings: after an initial boost to the population and economy, coal mining led to less growth over time. So, there is direct evidence that coal-mining-dependent localities in the United States experienced slower long-run economic growth than they would have had without mining.

In fact, there is reason to think the economics of the resource curse in the case of US coal operates much as it does elsewhere: by crowding out investment in other sectors and preventing governments, firms, and workers from making long-term investments in other industries (Badeeb et al. Reference Badeeb, Lean and Clark2017; Douglas and Walker Reference Douglas and Walker2017; Van der Ploeg and Poelhekke Reference Van der Ploeg and Poelhekke2009). Matheis (Reference Matheis2016), for example, finds that coal growth crowds out investment in manufacturing over the long term, and Douglas and Walker (Reference Douglas and Walker2017) find that educational attainment – an indicator of investment by government or citizens in skills applicable outside coal mining – is lower in mining areas than otherwise-similar parts of Appalachia.

But these economic pathways alone are increasingly seen as insufficient to explain lower economic growth in resource-endowed areas. Resource-rich areas become poor because, and only when, weak institutions allow resource extraction to preserve incumbent elites’ power (especially authoritarians), increase corruption, and spur civil conflict (Ross Reference Ross2015). Could this political resource curse explain any distinctive features of local institutions in coal mining areas of the United States?

Though a variety of mechanisms for these political effects have been proposed, almost all rely on one condition: that governments profit from resource extraction. Governments earning substantial revenue (through taxation or, especially, direct ownership of the resource) from resource extraction can use this money to buy support, with bribes or with public goods, without raising taxes from their citizens. The ‘rentier effects’ that result are considered the curse’s primary mechanism in the theoretical literature (Deacon and Rode Reference Deacon, Rode, Congleton and Hillman2015; Robinson et al. Reference Robinson, Torvik and Verdier2006), and some empirical work finds that resource extraction hurts institutions most, or even only, when resources are government-owned (Andersen and Ross Reference Andersen and Ross2014; Luong and Weinthal Reference Luong and Weinthal2010).

In the eastern United States, though, virtually no coal resources were owned by governments (David and Wright Reference David and Wright1997). What’s more, local governments had no ability to tax resource extraction and little ability to tax resource wealth until the mid-to-late twentieth century.Footnote ¹ In other words, there was no excess resource revenue ‘cursing’ officials in mining areas with the ability to use poorly monitored public funds for personal or political gain.

The absence of resource rents accruing to governments and officials in the eastern United States closed off two pathways for their development: a ‘cursed’ path, in which graft-motivated officials freely spent torrents of coal wealth to maintain their hold on power, undermining institutional development and public accountability; but also a ‘blessed’ path, in which strong institutions channeled excess coal wealth into public programs encouraging long-term growth, perhaps even strengthening citizens’ relationships with their government.

In other words, while the resource curse may have undermined the long-term economic growth of mining areas, any further pathologies of their governments cannot by explained by rentier effects specific to resource industries. Instead, I will argue, local governments suffered from capture by an economic elite – a possibility whenever important checks on power are absent, regardless of the economic interest at stake. That the economic patterns associated with the resource curse appear in this region, absent their political accompaniments, calls into question the role of institutions in the resource curse elsewhere.

Local Government Capture

Mining areas’ governments did not enjoy the rents that could have led to a classical expression of the resource curse. Yet I argue there are still reasons to expect these governments to be troubled – and in particular, to be low in capacity. These effects are not a special result of resource extraction, but instead a case of concentrated economic interests capturing local governments. Because coal’s dominance was strong enough to plausibly produce observable differences in policy, these places represent an opportunity to test the proposition that sufficiently powerful business interests can shape the capacity of local governments.

Capture is an extreme form of business power over politics. All governments must be responsive to the interests of their taxpayers, including the interests of businesses in their jurisdiction. This responsiveness deepens into capture when business elites systematically bend not only single policy outcomes, but the structures and processes of government itself, to suit their interests (Dávid-Barrett Reference Dávid-Barrett2023; McCann et al. Reference McCann, Spencer and Wood2021).

Substantial research describes business interests’ ability to capture specific regulatory processes (Dal Bó Reference Dal Bó2006), and their engagement in local politics on matters relevant to their bottom line (Anzia Reference Anzia2022). It is less common to think of business interests as engaging in ‘state capture’ – shaping ‘core state functions’ (Dávid-Barrett Reference Dávid-Barrett2023) – in the United States (but see Hertel-Fernandez Reference Hertel-Fernandez2019). There are several obstacles that should stand in the way of a single interest capturing the full policy-making process: actors in and structures of the local government itself, oversight from higher levels of government, competition from other business interests, and accountability to the mass public.

Coal mining elites were empowered to capture local governments in the eastern United States through the late nineteenth and early twentieth centuries because each of these guardrails was absent or weak. Pre-existing local governments were meager. Higher levels of government were uninterested in interceding. Other coherent business interests rarely existed, and the vulnerability of the mass public to company control limited the exercise of public oversight.

First, local governments can be captured when existing institutions are weak. Industrial-scale coal mining arrived in Appalachia and the Illinois Basin in the years after the Civil War. Early mining companies found themselves in a sparsely populated region with few incorporated cities and small county governments (see, for example, Erwin Reference Erwin1914; Banks Reference Banks1980). This gave many mining companies the opportunity to avoid municipal governments altogether, settling in unincorporated areas and avoiding legally forming their settlements into towns or cities. Where they could not avoid local governments (including county governments), these institutions were often small and unprofessionalized, leaving them vulnerable to capture. For example, many mining-area tax assessors did not have the resources to fully value mineral wealth, which allowed companies to limit their tax burdens through various legal maneuvers (see “Fair Assessment of Coal Lands Difficult: Kentucky Mining Counties Profit By Inability of Taxing Bodies to Determine Value of Property Known Only to Experts” 1920).

Second, local governments can be captured when they are afforded significant independence by higher levels of government.Footnote ² In this period, there was little reason to expect state governments or the federal government to threaten mining companies’ roles in local politics: mining companies were active lobbyists in these governments, and they received friendly policies on many fronts (Adams Reference Adams2004; “Is Coal Paying Fair Share?” 1965; McAteer Reference McAteer1973; Shalloo Reference Shalloo1933). Companies’ lobbying efforts were more unfailingly successful in mining-dominated states like West Virginia than in more diversified states like Illinois, where the state legislature acted to limit particular excesses of company town control (Beckner Reference Beckner1929). Still, while state governments engaged periodically in specific conflicts – with governors sending in troops to quell unrest, or state courts censuring especially rigged elections – they did little to limit the broader degree of coal company influence.Footnote ³

Third, local governments can be captured when a single coherent business interest is dominant. Here, the economic resource curse plays a political role: because resource extraction tends to dominate the economies of areas that contain resource deposits (Sachs and Warner Reference Sachs and Warner1995), extractive industries are more likely to be the dominant economic interest in their jurisdiction than are other industries. In the region and period of interest here, huge portions of localities’ labor forces were employed in mining. But even beyond those directly employed in the industry, these areas’ professional classes had commercial interests that aligned with mining, limiting the development of alternate bases of economic power (Gaventa Reference Gaventa1982). Though this kind of dominance is especially likely in resource extraction, other industries can and do become dominant in their geographic areas, as when a single manufacturing plant is the only major employer in an area.

Finally, local government capture is possible when public oversight is limited. The mineworkers who largely populated coal towns rarely engaged with local government as a potential check on company power. Before these workers gained union protections, they were subject to economic and physical coercion from both private and public police if they showed signs of organizing action against company interests (Wagner and Obermiller Reference Wagner, Obermiller and Brunn2011; Beik Reference Beik1996). Even after they gained basic rights to organize and make collective demands, though, these capacities were rarely directed at local government. Local union chapters were active in pressuring coal owners and operators for better wages and working conditions (Prosser Reference Prosser1973), and the United Mine Workers of America was active in pressuring state and federal governments for protections (Beckner Reference Beckner1929; McAteer Reference McAteer1973).Footnote ⁴ Whether because of the limited resources local governments controlled or because of companies’ ability to keep them to heel, local government simply does not seem to have been the arena in which most unions chose to fight their battles.Footnote ⁵

Local governments in mining areas, then, could not depend on their own institutional strength, the intercession of other governments, competition from other business interests, or public pressure to limit coal company influence. Mining companies were therefore free to capture local governments. Other industries facing similar conditions – a manufacturing firm moving a large workforce into an unincorporated area, for example – could have the same ability.

But this begs a separate question: why would companies want to capture local governments? Seizing and maintaining control over an entire institution of local government is expensive, even when the process is eased by the absence of obstacles, as it was here. Most businesses would see little benefit in such an undertaking. Businesses have structural sources of power that should obviate the need for this kind of intervention: local governments must generate their own tax revenue, and they do so by taxing economic activity. Because resident businesses can leave a locality if they are dissatisfied with its policies, taking their tax dollars with them, localities are incentivized to provide businesses with their preferred policies (Peterson et al. Reference Peterson1981). For an industry providing the bulk of an area’s economic activity, and therefore largely without competing business interests to satisfy, no bribery or vote buying should be necessary to receive accommodations from local governments.

Resource extraction, though, is an unusual case. A coal company whose property lies in a county cannot simply move the mineral deposits elsewhere if it is unsatisfied with the county government’s policies. As Anzia argues in Local Interests, businesses that cannot simply move to another locality may, unlike more mobile ones, ‘feel the need to get deeply involved in the institutions and inner workings of a particular city government’. With this crucial form of coercive power removed from their arsenal, and an immovable interest in a single locality’s policies, coal companies must more thoroughly control the policy-making process to secure their interests. Coal deposits are an especially immovable asset, but other kinds of investments – like large factories, or reliance on localized labor forces or transit networks – could create similar pressures for other kinds of firms.

Observable Traces of Capture

What did powerful mining elites want from the governments they captured? I argue that the political economy of resource extraction set mining elites’ interests against the growth of government capacity.Footnote ⁶ That is, mining elites wanted governments that did not and could not collect much in tax revenue, that employed few people, and that performed limited functions. This is the central observable implication of capture the empirics below will explore: mining elites in eastern coal country wanted and received low-capacity local governments. Mining elites surely used their capture of governments to secure other outcomes – especially the deference of local police forces – but I will focus on capacity here because it is both crucially important and measurable across places and times.

Many coal mines were located in undeveloped, unincorporated areas, so new mining sites required substantial investments in infrastructure to conduct their business and host their workforce. Coal companies starting production, then, faced a (stylized) choice: they could pay taxes into new or existing municipal or county governments to provide this infrastructure, or they could privately fund, build, and maintain it themselves.Footnote ⁷ Like other locally dominant companies in industries around the world, many coal companies chose to do the latter (Méndez and Van Patten Reference Méndez and Van Patten2022).

Why would coal companies prefer privately providing goods to allowing local governments to develop the capacity to do so? In short, as an area’s only significant taxpayer, mining companies would save little money, give up much control, and take on serious risk by funding these goods through taxes.

First, mining companies’ economic interests were at odds with the development of governments’ fiscal capacity (that is, their ability to collect taxes). Coal companies, like any private actor, were interested in minimizing their tax burden (see also Harpole Reference Harpole2021). But elites in some industries can support growing fiscal capacity: while the elites bear a cost from taxation, their business interests stand to gain from public goods like infrastructure and human capital, and growing fiscal capacity allows these elites to ensure that other taxpayers are not shirking on their portion of the goods’ shared costs (Hollenbach Reference Hollenbach2021).

However, mining elites in the eastern United States would have gained little and lost much from increased taxation and fiscal capacity. In mining areas, there were rarely other economic engines that could share the cost of public goods, so investments in tax collection capacity would mainly serve to collect more from mining elites.

Second, setting aside fiscal capacity, other forms of government capacity could often be replaced by coal company investment, allowing them to better tailor goods to their business interests. Companies could privately fund any investments in public goods, like quality schools or amenities, that served their business interest in attracting and retaining workers. They would not be served by investing in these goods any further: owners of coal land largely lived outside of mining areas, so they did not personally benefit from public services. And early mines depended on maintaining a supply of cheap labor, so additional investments in human capital and infrastructure could increase wage demands and exit opportunities for their workers.

Third, limiting government capacity prevented governments from being able to respond to citizens’ demands for policies the industry opposed. A well-staffed police department would be useful to mining elites if they could ensure it would be used only in their interests – but a well-staffed police department that could fall under the control of a union-sympathizing chief would be a threat to labor control.Footnote ⁸ Elections create the risk that government capacity could be used against company interests.

In this way, limiting government capacity acts as a kind of insurance against popular demand for government action. In a democracy, even one where processes are dominated by economic elites, mass demands create some risk that officials will come into office who will raise tax rates on elites’ wealth, or otherwise limit elites. (Even very powerful elites can rarely perfectly control elections.) Elites who foresee this may take an alternative path to prevent governments from collecting taxes: limiting the capacity of governments to assess and collect tax revenue, build public goods, or enforce regulations (Hollenbach and Silva Reference Hollenbach and Silva2019; Suryanarayan and White Reference Suryanarayan and White2021). Even if these policies are passed, then, governments may not have the personnel or expertise to implement them in practice. In this way, the threats elites face from democracy can lead them to constrain capacity.

In all, mining elites had incentives to undermine the growth of government capacity. Investing in public goods through public institutions would rarely have allowed coal companies to meaningfully share the costs of those goods with other industries, and subjecting those goods to public decision-making processes would have undermined their level of control. What’s more, weak governments cannot meet voter demands for limiting corporate power, nor can they provide public goods that could increase the mobility of the mining workforce.

In practice, this aversion to government capacity took many forms, both passive (in avoiding existing capacity) and active (in preventing governments from developing capacity). Coal companies avoided municipal governments altogether when possible, locating in unincorporated areas (McAteer Reference McAteer1973; Green Reference Green2011). When they could not avoid a local government’s jurisdiction, they strenuously limited their property tax liabilities through capture of the assessment process (Brown and Webb Reference Brown and Webb1941; Harpole Reference Harpole2021; “Is Coal Paying Fair Share?” 1965). When circumstances demanded investments in public goods, coal companies often made those investments themselves: coal companies built schools, fire stations, and jails, and they hired teachers, police officers, and doctors, sometimes using these gifts as leverage in local government decision-making processes (‘Assessing Coal Lands is Mostly Guesswork’ 1921; McAteer Reference McAteer1973; Shalloo Reference Shalloo1933).

And of course, coal companies engaged in a variety of tactics to strengthen control over the government capacity that remained. Coal owners and managers ran in local elections, companies funded local candidates’ campaigns, and firms used company lawyers and inspectors to lobby bureaucrats over policy implementation. They also engaged in more legally dubious tactics: vote buying seems to have been common in coal towns, sometimes through the mid-twentieth century (Roberts Reference Roberts1904; Hevener Reference Hevener2002; Ricketts Reference Ricketts1998; Walls and Stephenson Reference Lee, Walls and Stephenson1972), and companies used the press and legal system to suppress organized opposition (Kiffmeyer Reference Kiffmeyer1998; Ricketts Reference Ricketts1998).

While these examples come from coal’s heyday, I expect that lower government capacity continued even after the coal industry declined. If coal interests were able to keep property tax revenues low, local governments would have been less capable of building strong and effective institutions during a crucial period for their growth and professionalization – not only because of these places’ developmental stage, but also because of the advances in local government capacity taking place across the country during the Progressive Era. This early disadvantage in capacity would limit local governments’ ability to grow once the industry left, taking its tax revenue with it.

If coal dominance was especially impactful because it occurred at a crucial moment of potential growth for local governments, so too would I expect the industry to be far less damaging had it arrived later on. Many parts of this region were unincorporated when the industry arrived, and their county governments were limited; coal companies were able to skirt government capacity by avoiding jurisdictional boundaries or preventing bodies from forming or growing. Coal companies entering a place with existing, strong institutions of government would have far more difficulty capturing and limiting those powers.

In studying government capacity, I will focus on the outcomes of local government employment, revenue, and spending. Though capacity has many other dimensions, these three represent important aspects of a local government’s ability to set public policy: setting and collecting taxes, employing personnel to carry out government functions, and spending money on public policy are all core functions.

The empirical sections that follow use quantitative analyses to test whether developing a coal-dependent economy leads to smaller local governments. These analyses build on – and are no substitute for – extensive work by regional specialists on the relationship between coal mining and government capacity. This is an old topic in qualitative, historical, and public policy work on eastern coal country.

There is a widespread sense, substantiated by local evidence (Appalachian Land Ownership Task Force 1983; Smith et al. Reference Smith, Ostendorf and Schechtman1978), that coal companies pay little in taxes, and that this has led to underinvestment in roads, schools, and healthcare (Caudill Reference Caudill1962; Gaventa Reference Gaventa1982). In fact, in the 1970s, the West Virginia State Tax Department took the official position that ‘the coal industry’s support of local government and schools, through property taxes, has not been realistic given the extent of the industry’s mineral and fee property holdings’ (as quoted by the Appalachian Land Ownership Task Force 1983). This provides valuable support for the notion that coal companies have the will and the means to succeed in keeping local governments low in capacity.

Data and Measures

To test the hypothesis that a coal-dependent economy led to weaker government institutions in the coalfields of the eastern United States, I draw on data on economic and political conditions at the county level. Counties are a mutually exclusive and exhaustive unit of geography – that is, all land in this region is part of one and only one county – and are small enough in the states of interest that county-level data captures fairly local conditions. I limit my analysis here to the eastern coal country region discussed above, which includes the six states of Illinois, Indiana, Kentucky, Ohio, Pennsylvania, and West Virginia.

I measure a county’s dependence on coal using employment data drawn from the full-count Censuses from 1850–1940, the period covering the peak of coal employment in this region (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover, Meyer, Pacas and Sobek2020). For each county, at each Census, I divide the number of people employed in the coal mining industry by the total number of non-farm, non-domestic laborers in the county. This measure captures the extent to which coal provides the livelihoods of county residents and therefore reflects the power of coal, relative to the area’s other industries, in the local economy.

I measure the size of local government institutions using two kinds of data. First, using the same 1850–1940 full-count Census data, I calculate the number of people in each county whose industry of employment is classified as ‘local government administration’. This includes people whose full-time job (or the job that provided most of their income) was provided directly by the local government, including, for example, police officers, firefighters, lamplighters, clerks, assessors, judges, and elected officials.Footnote ⁹

Second, the Census of Governments provides information on the revenue from property taxes, expenditures, and employees of local governments in the United States over the period from 1957 to 2002. Earlier versions of the survey were conducted roughly decennially from 1880 to 1932.Footnote ¹⁰ I sum the property tax revenue, expenditures, and employment of local governments across all local governments in a county – including the county government and any municipal governments – as measures of local government strength.

The sections that follow assess the relationship between coal industry dominance and local government capacity. The first presents a difference-in-differences analysis, showing that counties where the coal industry developed abruptly diverged from non-coal counties in government size at the time the industry arrived. The second tests the relationship between coal industry size and local government employment, revenue, and spending throughout the twentieth century, using tonnage of coal deposits as an instrument for industry size.

The difference-in-differences analysis and the instrumental variables analysis rely on different identification assumptions, outcome measures, operationalizations of coal industry dominance, and tranches of the data. Testing hypotheses using multiple research designs with different strengths and weaknesses is a form of ‘triangulation of evidence’, an especially important exercise in the pursuit of causal inference using observational data where no single approach offers perfect identification (Hammerton and Munafò Reference Hammerton and Munafò2021).

Difference-in-Differences Analysis

One approach to testing the relationship between the coal industry and local government capacity is to take advantage of the timing of the coal industry’s arrival: coal counties’ governments should become noticeably smaller relative to other places’ governments after the coal industry arrived. This section compares government employment in areas without coal industries to government employment in places with large ones, before and after the industry’s moment of arrival.

As the coal industry expanded over the late nineteenth and early twentieth centuries, some coal-producing areas experienced steady and gradual growth in mining employment. In other areas, the industry arrived abruptly. For example, in 1910, nineteen of the 10,500 residents of Harlan County, Kentucky, worked in the coal industry. By 1920, 5,267 worked in coal mining, representing 69 per cent of the county’s non-farm labor force.

In all, there are twenty-two counties in Appalachia and the Illinois Basin that developed dominant coal industries over a short time period. Each of these counties, between two consecutive Censuses in the 1850–1940 period, went from less than 5 per cent of their labor force in coal mining to at least 20 per cent, and eventually reached at least 40 per cent. In all cases, coal became the county’s largest industry at some point in (and often throughout) the period between 1870 and 1940. For the purposes of this analysis, this group of twenty-two counties serves as the ‘treatment group’: they began untreated (no coal industry) and, at an identifiable point in time, received the ‘treatment’ of developing a large, dominant coal industry.Footnote ¹¹

To assess the relationship between coal specialization and local government size, I compare changes in pre- and post-treatment government size in these treated counties to changes in government size at the same time for counties that did not receive the treatment; that is, a difference-in-differences design. I define the untreated counties as those in which no more than 5 per cent of the labor force worked in the coal industry at any point between 1850 and 1940 (n = 280).Footnote ¹²

The outcome of interest in this analysis is the number of employees across all local governments in a county, a measure available over the full lifespan of the coal industry thanks to measures on early decennial censuses. To isolate theoretically relevant variation in the number of government employees in a county-year, I measure the difference between a county’s real number of employees and the number predicted by a model including the interaction between the county’s state, population, and time.Footnote ¹³ To capture the relationship between these variables and government size absent industry intervention, the model used to produce the measure is trained only on control counties. Further details on this measure can be found in Appendix Section A.2.2.

I expect that after the coal industry arrives in treated counties, governments in treated counties (as measured with this residualized figure) will grow more slowly than governments in control counties.

The key identifying assumption of this analysis is that of ‘parallel trends’: the potential outcomes under the control condition of both groups of counties move in parallel over the time period being studied. Here, this means that if the treated counties had never developed a coal industry, their governments would have grown at the same rate as the governments of control counties. Within states and years, and accounting for population size, I argue this assumption is plausible; as long as they are being compared to similarly populated control units with the same state-delegated responsibilities, there is no reason to expect coal counties’ governments to have grown differently from their neighbors absent intervention. Appendix Section A.2.1 presents tests that show no signs of pre-treatment trends towards divergence, and in Appendix Section A.2.6, I construct several alternative control groups that make this assumption even more plausible and find similar results.

The coal industry arrived in each of the treated counties between 1870 and 1920. The earliest time period in which outcome data are available is 1850, allowing for the observation of pre-treatment trends in the outcome even for the earliest-treated units. Outcome data are available at five or ten year increments through 1997 (except the 1940–57 period), and to observe how treated and control units’ governments grew over the course of the twentieth century, I test for differences at every point following treatment.

In some of the treated counties, the coal industry did not survive long after the downturn of the 1920s, and in most, coal employment accounted for less than 10 per cent of the labor force by the 1980s. However, I consider treatment status here to be irreversible – that is, once a county is ‘treated’ by developing a dominant coal industry, it remains treated in all following years, regardless of the industry’s size in later periods. If I am correct that the presence of a dominant coal industry in the early twentieth century shaped the development of local governments durably, even after the industry left, it cannot be considered comparable to a unit that never received the treatment at all.

I implement this analysis in R using the did package (Callaway and Sant’Anna Reference Callaway and Sant’Anna2021a) and the method described in Callaway and Sant’Anna (Reference Callaway and Sant’Anna2021b). Treated observations are divided into cohorts based on treatment timing, and treatment effects are calculated for each subsequent point at which outcomes are observed. These group–time effects are combined into weighted average treatment effects at each time point for which outcomes are observed, where the weights depend on the number of observations in each group.

Figure 1 presents the results. Each point represents the average treatment effect for groups at each time pre- and post-treatment; the bars represent bootstrapped 95 per cent confidence intervals clustered at the county level. The x-axis shows the time relative to treatment. The leftmost two points, then, test for differences in differences between treatment and control groups twenty and ten years before treatment occurs. These estimates are insignificant, suggesting trends in government employment in treatment and control units are not diverging before treatment is received.

The black coefficients in Figure 1 represent the average treatment effect across groups at each time point after treatment. The significant negative coefficient at time 20 suggests that treated units, between the Census twenty years before being treated and the Census twenty years after being treated, gained less in government employment than did control units over those same years. The estimates from further after treatment compare the change in treated and control units between their final pre-treatment period and the current period. Though the statistical significance of the estimates varies, coefficients remain negative until around forty years after treatment, at which point estimates generally become close to zero.Footnote ¹⁴

A pooled test of the treatment effect over the periods from zero to forty years post-treatment suggests a significant effect of −0.32 (−0.55, −0.09) points. This represents a one standard deviation decrease in government size in treated units relative to control units. To put this in perspective, consider two Pennsylvania counties that differed by 0.3 points on the same scale in 1900: Berks and Lancaster, which both had about 159,000 residents at the time. Berks County’s governments employed 117 people, while Lancaster’s employed 156. This difference of about 25 employees per 100,000 residents is similar to the estimated treatment effect of the emergence of a large coal industry on a coal county.

Figure 1.

Differences in differences: government employment in coal and non-coal counties over time. On the y-axis is the estimated effect of coal employment on residualized government employment. The x-axis shows time relative to treatment; −20 represents twenty years before receiving treatment, while 20 represents twenty years afterwards. Bars represent bootstrapped 95 per cent confidence intervals.

These results suggest that the arrival of a dominant coal industry depressed the growth of local governments in coal-producing areas. Several decades after the arrival of the industry, coal-dominated counties’ government employment had returned to near parity with employment in other places. Given the substantial decline in coal employment over the second half of the twentieth century, it is possible the dwindling of the coal effect is because the coal industry had left these areas. However, for decades after the industry arrived, coal-mining areas showed distinctively slow government growth for their population size and states.

Instrumental Variables Analysis

This section takes an alternative approach to estimating the coal industry’s effect on government size: an instrumental variables design. This approach supplements the difference-in-differences design in important ways. Most significantly, it incorporates all the region’s counties rather than relying on the limited comparison of twenty-two ‘overnight’ counties to completely coal-free areas. The instrumental variables approach also allows analysis of two additional measures of government capacity data – property tax revenue and expenditures – which were not collected often enough to allow for a difference-in-differences analysis. The instrumental variables approach also relies on different assumptions for identification.

One way to test the relationship between government capacity and coal industry size is simply to regress the size of a county’s government on the size of its coal industry (controlling, as in the previous section, for differences in population, state, and year). But the size of a county’s coal industry may be endogenous to its other features: for example, coal companies may prefer to develop mines in counties with weaker local governments in order to avoid oversight. The association between industry size and government capacity, then, would reflect both this ‘selection’ process and the industry’s effect on government.

To address this, I instead use the total tonnage of coal deposits in a county as an instrument for the size of its coal industry. Counties need substantial coal deposits to develop an economy that specializes in coal mining; the tonnage of coal deposits in a county is therefore an excellent predictor of the proportion of a county’s workforce employed in coal. Coal deposits are measured using the United States Geographical Survey’s USCOAL database, an effort to measure the amount of coal resources present in all deposits in the United States. Compiled from a variety of surveys in sources between the 1950s and 1980s, USCOAL contains, when possible, an estimate of the total tonnage of coal originally present in each deposit; otherwise, it contains an estimate of the coal remaining in the ground in the 1950s. Aggregated to the county level, these data provide an estimate of the total coal resources available in a county.

Coal deposits are a strong instrument for coal employment (F ranges from 20 in 1880 to 127 in 1920). The other key criterion for a good instrument is the exclusion restriction: that is, coal deposits should be associated with local government size only through their relationship with coal employment. Coal deposits and similar geophysical measures have been used as instruments for economic development in past studies of socio-political outcomes; coal deposits are ‘assigned’ through millenia-old geological processes, and deposits in the ground cannot affect socio-political processes other than through the industry that develops to extract them (Esposito and Abramson Reference Esposito and Abramson2021; Feng et al. Reference Feng, Zhang, Zhao and Zhao2024; Hollenbach Reference Hollenbach2021).Footnote ¹⁵

To estimate the relationship between local government capacity and the size of the coal industry, I regress the three measures of local government size – number of government employees, amount of property tax revenue, and total amount of spending – on the proportion of a county’s wage labor force employed in the coal industry, with controls for population size and fixed effects for state and year.Footnote ¹⁶ I present models using tonnage of coal deposits as an instrument for coal industry size, alongside simple ordinary least squares (OLS) models using the non-instrumented coal industry size for comparison. Standard errors are clustered at the level of the county, and observations are weighted by population.Footnote ¹⁷

For dates between 1880 and 1940, when the coal industry employed substantial proportions of the labor force in many counties in the region, I predict government outcomes in a given year using coal employment from the most recent census. This captures the contemporary relationship between coal dependence and local government at a particular time. For outcomes measured starting in the 1950s, after coal employment had fallen in most places, I predict government outcomes using coal employment in 1920, the industry’s peak. This captures the longer-run relationship between coal dependence and local government after the industry’s strength has waned.

Figure 2.

Coal employment and government capacity. Points represent the coefficient on the proportion of county labor force in the coal industry in a regression of government capacity on coal employment, with controls for total population and state and year fixed effects. Black points represent models with industry size instrumented by coal deposits; gray points are OLS models with endogenous industry size. Left panel (pre-1950) models measure government capacity and coal industry size in the same year; right panel (post-1950) models measure coal industry size in 1920. Bars represent 95 per cent confidence intervals with standard errors clustered at the level of the county.

Figure 2 shows the results, plotting the coefficient on coal industry size (as instrumented by coal deposits in black; naive in gray) in pre- and post-1950 models for each outcome.

The pre-1950 models show that a larger coal industry led to less local government employment and tax revenue. The magnitude of these effects is large: government employment is 11 per cent lower and revenue 10 per cent lower in places with a coal industry in the third quartile of counties compared to places with no coal industry.

The post-1950 models show mixed evidence on persistence: while 1920 coal employment leads to only slightly lower government employment after 1950, it does lead to lower government revenue – of a similar magnitude to the period of coal dominance – and lower government spending. Year-by-year results (presented in Appendix Section A.3.3) show these effects wane in size over time. The much-reduced effect on government employment in the later period mirrors the findings of the difference-in-differences analysis, in which the effects of mining on government employment disappeared 50–60 years after the industry’s arrival.

In general, the instrumental variables estimates of the coal industry’s effects on government size differ little from the ‘endogenous’ OLS models, though the precision of the estimates is substantially reduced. This suggests there is little bias in naive estimates of the relationship between the coal industry and local government introduced by, for example, industry selection into areas with larger or smaller governments. Nonetheless, the instrumental variables design provides confidence this or other unanticipated selection processes are not entirely responsible for this relationship.

Taken together, these results suggest that the coal industry reduces local government capacity: places with larger coal industries have governments that employ fewer people, collect less revenue, and spend less money. Patterns diverge in the later time periods between government employment and government spending and revenue: though employment in coal counties recovers somewhat after the industry declines, more coal-dependent counties continue to collect substantially less in revenue decades afterwards.