The essay of ECo

shazuanzhe

9.TheShortandLong-RunEffectsofResourcesonOutcomes.pdf

Home >Business & Finance homework help >Economics homework help >The essay of ECo

NBER WORKING PAPER SERIES

THE SHORT-RUN AND LONG-RUN EFFECTS OF RESOURCES ON ECONOMIC OUTCOMES: EVIDENCE FROM THE UNITED STATES 1936-2015

Karen Clay Margarita Portnykh

Working Paper 24695 http://www.nber.org/papers/w24695

NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue

Cambridge, MA 02138 June 2018

The authors thank participants at the 2013 Economic History Association Meetings, the 2014 Queens Economic History Workshop, 2015 and 2017 AERE sessions at the ASSA, Brown University, Miami University, Vanderbilt University, Clemson University, and Michael Alexeev, Hunt Allcott, Lee Alston, Daniel Berkowitz, Ian Keay, Marvin McInnis, and Gavin Wright for helpful comments. Ellis Goldberg, Erik Wibbels and Eric Mvukiyehe generously shared their data. Avery Calkins and Alex Weckenman provided excellent research assistance. The authors gratefully acknowledge financial support from Heinz College at Carnegie Mellon University. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.

© 2018 by Karen Clay and Margarita Portnykh. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.

The Short-Run and Long-Run Effects of Resources on Economic Outcomes: ¸˛Evidence From the United States 1936-2015 Karen Clay and Margarita Portnykh NBER Working Paper No. 24695 June 2018 JEL No. J2,N12,O4,Q24,Q43

ABSTRACT

This paper draws on a new state-level panel dataset and a model of domestic Dutch disease to examine the short-run and long-run effects of oil & natural gas, coal, and agricultural land endowments on state economies during 1936-2015. Using a flexible shift-share estimation approach, where the shift is national resource employment and the share is state resource endowment, we find that different resources had different short-run effects in different time periods, across increases and decreases in resource employment, and across different outcomes. Using long differences, we find that long-run population growth was an important margin of adjustment over 1936-2015. States with larger coal and agricultural endowments per square mile experienced significantly slower population growth than states with smaller endowments per square mile. Resource endowments had no effect on long-run growth in per capita income.

Karen Clay Heinz College Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA 15213 and NBER [email protected]

Margarita Portnykh Heinz College Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA 15213 [email protected]

1. Introduction

What are the short-run and long-run effects of resources on economic outcomes? The

effects of resources on outcomes are widely debated, because many countries, states, and

counties have substantial endowments of natural resources. The salience of the relationship

between resources and outcomes has led to a large amount of research. Despite the large

literature, there is relatively little consensus regarding the answer to the question. In both the

U.S. and the international contexts, different papers reach different conclusions about the effects

of resources on outcomes.1

To address this question for the United States, we use a new state-level panel dataset and

a model of domestic Dutch disease. The data set spans 1936-2015 and covers the three most

valuable natural resources – oil & natural gas, coal, and agricultural land. The model, which is

from Allcott and Keniston (2018), provides short-run and long-run predictions regarding the

effects of resources on population, wages and employment. Our empirical analysis examines the

extent to which these predictions hold for state population, wages and employment. Because

some papers in the literature focus on per capita income, the empirical analysis also examines per

capita income. The long time period allows us to examine the effects of resources over different

time periods. The use of three resources facilitates comparisons across coal and agriculture,

which had declining employment, and oil & natural gas, which had increasing employment. It

also facilitates comparisons across oil & natural gas and coal, which are non-renewable, and

1 In the international context, Sachs and Warner (1995, 1997), Sala-i-Martin and Subramanian (2003), Papyrakis and Gerlagh (2004) and other papers find a negative relationship between resources and outcomes, but Alexeev and Conrad (2009) and Cavalcanti, Mohaddes and Raissi (2011) do not. In the United States context, Black et al (2005), Papyrakis and Gerlagh (2007), Goldberg et al (2008), James and Aadland (2011) and Jacobsen and Parker (2014) find a negative relationship between resources and outcomes, but Boyce and Emery (2011), Michaels (2011), Weber (2012, 2014), Feyrer et al (2017) and Allcott and Keniston (2018) do not.

agricultural land, which is renewable and can be used to produce different products at different

times.

To examine the short-run relationships between resources and outcomes, we use a

flexible shift-share approach, where the shift is changes in national employment for that resource

and the share is state endowment of a resource per square mile. Our primary measure of

endowment is state endowment in 1935 per square mile based on 1935 knowledge of reserves,

but we present results for alternative measures of endowment including endowment in 1935 per

square mile based on 2015 knowledge of reserves and endowment per capita based on population

in 1929. Our estimation approach is flexible in that it allows for different effects across increases

and decreases in resource employment. Our main results focus on short-run effects over 5-year

time intervals, rather than the 1-year time intervals that are more common in the literature, to

allow time for spillovers to develop.

The paper has two main findings. First, the paper finds that different resources had

different short-run effects in different time periods, across increases and decreases in resource

employment, and across different outcomes. For growth in population and for growth in per

capita income, the coefficients for a given resource in a given time period were not necessarily

the same in sign or significance. This is relevant, because growth in population and growth in per

capita income are frequently used as proxies for welfare. Across a hypothetical boom-bust cycle

for a given resource, states in many cases could be worse off after the cycle than before the

cycle.

Second, the paper finds that the primary margin of long run adjustment has been larger

long-run relative population declines in states with larger coal and agricultural endowments.

States with larger coal endowments had larger relative population declines for the full sample

period and for the 1936-1974 and 1975-2015 sub-periods. States with larger agricultural

endowments had larger relative population declines in the later period. As a result of population

adjustments, resources had no effect on growth in state per capita income in any of the periods.

Employment in coal and agriculture peaked in the early twentieth century. To evaluate whether

the long-run effects were different when employment in coal and agriculture was rising, we

examine the long-run relationships between resource endowments in 1935 and population

growth during 1880-1935. During this earlier period, states with larger coal endowments had

larger relative population increases.

This paper contributes to the U.S. literature on the relationship between resources and

economic outcomes by examining effects of multiple resource sectors on multiple outcomes over

an 80-year time period and using a flexible estimation approach that allows increases and

decreases in resource employment to have different effects. Our findings on the effects of

increases and decreases in resource employment are related to Black et al (2005) and Jacobsen

and Parker (2015), who find negative effects of boom-bust cycles for coal and non-coal counties

in Appalachia over the period 1970-1989 and for oil and non-oil counties in the Western U.S.

over the period 1969-1998. Our analysis is also related to Allcott and Keniston (2018), which

examines the effects of oil & natural gas on county outcomes from 1969-2014. They find over

the boom-bust cycle of the 1970s and 1980s that there is no long-term effect of oil & natural gas

endowment on a range of county outcomes, which is similar to what we find. Finally, our work is

related to Hornbeck and Keskin (2015), who look at spillovers generated by agriculture by

comparing agricultural counties that differ in their access to the Ogallala aquifer. Our analysis

complements time series work by other scholars at the U.S. county level, which tends to focus on

individual resources, shorter time periods, and 1-year time intervals over which resources affect

outcomes.

The paper also contributes to the literature in American economic history that examines

the long-run role of resources on economic outcomes during the mid- and late twentieth century.

Our finding that there were no long-run effects of resources on per capita income and that

population was the primary margin of adjustment are most closely related to Mitchener and

McLean (2003), Michaels (2011), Hornbeck (2012), and Matheis (2016). Mitchener and McLean

(2003) find that state resources, as measured by the share of the workforce in mining, were

related to worker productivity through 1940, but not in 1960 or 1980. Michaels (2011) examines

southern counties with and without oil in Texas, Louisiana, Oklahoma, and nearby states using

data from 1940-1990. He finds population increases in oil counties relative to counties without

oil and higher, but declining, differences in per capita income and median family income.

Hornbeck (2012) finds that population loss was the primary margin of adjustment to erosion in

Dust Bowl counties from 1930 to 1940 and that population declines continued through the

1950s. Matheis (2016) studies the short- and long-run effects of coal production on county

population and manufacturing. He finds large positive effects of coal production in the previous

decade on population pre-1930 and smaller effects in later periods.

Our results speak indirectly to the economic history literature on the importance of

resources during the nineteenth and early twentieth centuries. We find that the long-run effects of

state endowment of coal on state population during 1880-1935 were positive. This is consistent

with Habakkuk (1962), Wright (1990), and Wright and Czelusta (2004), who argue that mineral

resources had important benefits for the American economy during the nineteenth and early

twentieth centuries.2 And it is consistent with the European literature on coal and economic

development (Pomeranz 2001, Allen 2009, and Fernihough and O’Rourke 2014).

2. Resources

This section briefly discusses the literature on the relationship between natural resources

and outcomes in the United States, measures of resources used by different authors, and the

measures of resources used in this paper.

Resources and Economic Outcomes in the United States

A large number of papers have examined the effects of resources on outcomes. Nearly all

papers that apply cross sectional analysis find that resources had a negative effect on outcomes,

including Boyce and Emery (2011), Goldberg, Wibbles, and Mvukiyehe (2008) James and

Aadland (2011), and Papyrakis and Gerlagh (2007). An important exception is Mitchener and

McLean (2003), which examines earlier time periods and focuses on price adjusted income per

worker. They find resources were positively related to outcomes in 1880, 1900, 1920 and 1940

and had no effect in 1960 and 1980.

The results are somewhat mixed for papers that use time series analysis. These papers

predominatly focus on the short run. Using state data, Goldberg, Wibbles, and Mvukiyehe

(2008) find resources are negatively related to growth. Boyce and Emery (2011) find resources

are negatively related to growth, but positively related to income. Using county data, Hornbeck

and Keskin (2015) show that having access to the Oglalla Aquifer, which was used for irrigation

after World War II, is positively related to a range of outcomes. Hornbeck (2012) shows that

counties with larger land erosion during the Dust Bowl experienced larger population outflows.

2 The Canadian literature also emphasizes the importance of resources (Chambers and Gordon 1966, Lewis 1975, and Keay 2007).

Matheis (2016) finds that coal is positively related to a population and manufacturing. Allcott

and Keniston (2018) and Michaels (2011) find that oil & natural gas are positively related to a

range of outcomes. Feyrer et al (2017) and Weber (2012, 2014) examine the recent effects of

hydraulic fracturing and find positive effects on outcomes. Using county data, Black et al (2005)

and Jacobsen and Parker (2014) find that the boom is smaller than the bust, leaving coal and oil

& natural gas counties worse off after the boom-bust cycle than before.

A strand within U.S. and European economic history argues that natural resources were

important drivers of long-run growth. Some examples include Habakkuk (1962), Wright (1990),

Pomeranz (2001), Wright and Czelusta (2004), Mitchener and McLean (2003), Keay (2007),

Allen (2009), and Fernihough and O’Rourke (2014). Other authors such as Mokyr (1976, 1992),

Clark and Jacks (2007), McCloskey (2010) have argued that natural resources were not key

drivers of growth, instead stressing other factors. In contrast to the broader literature, however,

they generally do not argue that resources had a negative effect on outcomes.

Measures of Resources in the Literature

The definition of resources varies considerably across papers in the U.S. literature. For

example, Papyrakis and Gerlagh (2007) define resources as “The share of the primary sector’s

production (agriculture, forestry, fishing, and mining) in GSP for 1986.” In their study of U.S.

counties, James and Aadland (2011) use percent earnings from agriculture, forestry, fisheries,

and mining as their measure of resources. Other papers focus primarily on fossil fuels. The

dependent variables are mining (which includes oil, coal and other minerals) in Mitchener and

McLean (2003); coal in Black et al (2005); oil and coal in Goldberg et al (2008); oil and natural

gas in Michaels (2011); mining in Boyce and Emery (2011); natural gas in Weber (2012, 2014);

oil and natural gas in Jacobsen and Parker (2014); coal in Matheis (2016); natural gas in Feyrer

et al (2017); and oil and natural gas in Allcott and Keniston (2018).

Measures of resource intensity also vary. Some use the value of resources produced or

employment divided by income or population or workforce. Others classify geographic units

based on reserves (Michaels 2011) or reserves per square mile (Allcott and Keniston 2018) or

use cutoffs to identify high and low coal counties (Black et al 2005), or high and low oil and

natural gas counties (Jacobsen and Parker 2014).

Measures of Resources in this Paper

This paper examines three resources: oil & natural gas, coal, and agriculture. Why do we

focus on these three resources? Table 1 shows the value in constant 2010 dollars of renewable

and non-renewable resources produced in the U.S. in 1936 and 2015, the first and last years of

our sample.3 Oil & natural gas and coal were the largest nonrenewable sectors in 1936 and in

2015, and agriculture was the largest renewable sector in those years. We treat oil & natural gas

as a single resource, because disaggregated employment is not available for every year.

Examining these three resources facilitates comparisons along two dimensions: i) sectors with

declining vs. increasing employment and ii) non-renewable vs. renewable resources. Coal and

agriculture had declining employment over the sample period, while oil & natural gas had

increasing employment. Coal and oil & natural gas deposits can only produce coal or oil &

natural gas. In contrast, agricultural land is renewable and thus can be used to produce different

agricultural products at different times in response to changing market conditions.

3 The sample includes the 48 contiguous states. In particular, it excludes Alaska, Hawaii, and the District of Columbia. Alaska and Hawaii enter the sample late (1960), and Alaska is an extreme outlier in terms of resource intensity. The federal government dominates economic activity in the District of Columbia. Data were adjusted to 2010 dollars using the US CPI data from Officer and Samuelson’s website Measuring Worth.

State endowments of oil & natural gas, coal, and agriculture are measured in 1935. For

oil & natural gas and coal, endowments are reserves from the Minerals Yearbooks.4 For

agricultural, endowment is the state value of farmland from the 1935 Census of Agriculture.5

Why do we use 1935 and not an earlier measure of endowments? One issue is the low

frequency of data on outcomes for earlier periods. The other issue is endogeneity. 6 Reserves for

oil & natural gas and coal and land values for agriculture are used because they are more

exogenous than production or employment. During much of the nineteenth century, estimated oil

& natural gas and coal reserves are likely to be related to the timing of settlement of states and

state investments in discovery of resources. By 1935, the location and characteristics of oil &

natural gas and coal deposits in the United States were relatively well understood, so this is much

less important than it might have been earlier.7

We construct αir, as a scaled measure of endowment per square mile in 1935 in state i for

resource r.8 Specifically, we divide endowment by the area of the state in square miles, because

states differ both in their endowments and in other attributes such as their area. For example, the

same endowment in Texas, which is 268,580 square miles and in Rhode Island, which is 1,545

square miles would potentially have very different impacts on the state economy.

4 Coal reserves in 1935 are constructed using recoverable reserves in 1950 and coal production from 1935-1950 assuming past losses are equal to production. 5 We use 1935 average state value of farmland multiplied by the number of acres to measure endowment. An alternative approach is to use 1935 average national value of farmland multiplied by the number of acres to measure endowment. This treats all acres as having equal value, wherever they are located. As a robustness check, we present specifications in which each acre has equal value. 6 See Wright 1990, David and Wright 1997, Mitchener and McLean 2003, and Clay 2011. 7 Mitchener and McLean 2003 argue that state level mining can be considered exogenous in 1880. “There were no barriers to the flow of capital and technology across state boundaries, and firms and individuals could take their investment and talents wherever they saw the opportunity for the highest potential return.” 8 This approach is similar to Allcott and Keniston (2018), which also uses endowment per square mile. Reserves are divided by area to address variation across geographic units in area. Some studies examine counties that are roughly similar in size and so simply use reserves. Reserves are not divided by employment or income, because these are likely to change in response to increases in production.

The endowments are then rescaled so that the state with the highest endowment of

resource r per square mile has α = 1. States with zero endowment have α = 0. The lowest

endowment is zero for oil & natural gas and coal. The lowest endowment is positive for

agriculture. The first three panels of Figure 1 present the distribution of agriculture, oil & natural

gas, and coal across states in 1935 (based on available knowledge in 1935). There is considerable

variation in resource endowments. Louisiana, Texas, and Oklahoma have the largest oil &

natural gas endowments. North Dakota, West Virginia and Colorado have the largest coal

endowments. Connecticut, Iowa, and Illinois have the largest agricultural endowments.

We construct two additional measures of endowment. Because there continued to be

resource discoveries and changes in understanding of known deposits that would occur between

1935 and 2015, we construct a measure of 1935 endowment based on knowledge available in

2015. 9 Shares of resources held by different states were generally stable and so can be treated as

the endowment in 1935. The level of economically recoverable reserves would change, of

course, with national changes in technology and economic conditions. If these changes in levels

caused shares to shift between 1935 and 2015, the relationship between 1935 endowment shares

interacted with changes in national employment and outcomes may become more attenuated over

time. For the second measure, we construct per capita endowment in 1935 based on knowledge

available in 1935 using population in 1929. 10

9 For oil & natural gas and coal, we take reserves in 2015 and add back production (in constant 2010 dollars) of oil & natural gas and quantities of coal (in short tons) produced between 1935 and 2015. For agriculture, the alternative measure of endowment is based on the land value in 2015. The correlation between endowment per square mile in 1935 (based on available knowledge in 2015) and endowment per square mile in 1935 (based on available knowledge in 1935) is 0.94 for oil & natural gas, 0.55 for coal, and 0.78 for agriculture. We present specifications using this measure as a robustness check. 10 The correlation between endowment per square mile in 1935 (based on available knowledge in 2015) and endowment per capita (using population in 1929 as a denominator) is 0.57 for oil & natural gas, 0.72 for coal, and 0.27 for agriculture. We present specifications using this measure as a robustness check.

Figure 2 plots the national employment by resource sector over time.11 We see a general

decline in the agricultural employment and in coal mining employment over time. Oil & natural

gas employment was increasing through the early 1980s, declined into the mid-2000s, but has

been increasing since then. For most of the time period, agriculture had the highest employment

and coal had the lowest. Appendix Figure A1 plots resource income in constant 2010 dollars

over time. Agriculture had the highest income and coal had the lowest.12

We examine two sub-periods: 1936-1974 and 1975-2015. The first sub-period, 1936-

1974, is a period of relative income stability for all three sectors. There is a short boom in the

very early period for agricultural income. Employment is also changing relatively smoothly,

particularly for oil & natural gas and coal. The second sub-period, 1975-2015, is much more

volatile in terms of income. The boom-bust-boom cycle in income is evident for all three sectors.

Employment changes more smoothly than income, but the boom-bust-boom cycle is clear,

especially for oil & natural gas employment.

3. Conceptual Framework

To understand the effects of resource booms, we draw on Allcott and Keniston’s (2018)

model of domestic Dutch disease. In this section, we discuss some key aspects of their model.

Allcott and Keniston (2018) use a Moretti (2010) version of the Rosen–Roback spatial

equilibrium framework to investigate the local welfare effects of resource booms. The model

compares two geographic units, one with a resource endowment and one without, across three

11 According to BEA (2015), wage and salary jobs and proprietors’ jobs are counted when constructing employment, but unpaid family workers and volunteers are not. 12 An important factor in the divergence of resource employment and resource income has been improvements in efficiency driven largely by technology and mechanization. Efficiency improvements are discussed later in the paper.

periods.13 In period 0, the geographic units are symmetric and neither produces resources. In

period 1, the unit with the endowment experiences a (temporary) resource boom, in which

production is positive. In period 2, the boom is over and neither produces resources. In addition

to the resource sector, there are two other sectors that require local labor – a tradable sector and a

non-tradable sector. There is also a housing sector that does not require local labor.

In equilibrium, firms and consumers optimize and markets clear. Firms maximize profits

and demand labor. There are two possible types of spillovers across firms over time – learning

by doing spillovers and agglomeration spillovers. Learning by doing spillovers mean that

current productivity is influenced by prior sectoral employment. Agglomeration spillovers mean

that past population influences current productivity. In every period, individuals decide where to

live, supply one unit of labor, and make consumption decisions about housing, tradable goods,

and non-tradable (local) goods subject to the budget constraint.

The model generates predictions regarding the contemporaneous and long-run effects of a

resource boom. Contemporaneously, the model predicts that the resource boom will increase

population and wages. The boom will also increase local sector employment, decrease tradable

sector employment, and increase local sector prices.

Allcott and Keniston (2018) examine the long-run relative welfare effects. They first ask

whether the boom increases cumulative social welfare in geographic unit A vs. geographic unit

B. In the long run, the model predicts that the relative welfare effects can be signed by

examining relative population. They state: “Intuitively, people vote with their feet by migrating

to the county with higher welfare. This equation will be useful empirically, as it will allow us to

sign the relative welfare effect even without a direct estimate of how the resource boom affects

13 In their online appendix, Allcott and Keniston (2018) show the results hold for many geographic units. In their context the geographic units are counties; in our context the geographic units are states.

local prices and amenities. … If there are no productivity spillovers … then the two counties

have equal productivity, population, and wages after t=1, and a resource boom unambiguously

increases relative welfare. If there are productivity spillovers, then local sector relative

productivity will increase, and the signs of both tradable sector relative productivity and relative

welfare will depend on the relative strengths of the learning-by-doing versus agglomeration

spillovers.” 14

While the model considers a single resource sector, empirically one might expect the

relationship between individual resources and outcomes to be heterogeneous across a variety of

dimensions. For example, resources may have different effects across increases and decreases

because of differences in spillovers. Relationships might change over time due to changing

production technology, transportation costs, capital markets and other factors.15 1617 If there are

adjustment costs, the effects over a one-year period may differ from the effects over a five-year

period.

4. Data on Outcomes

14 Allcott and Keniston (2018) p. 11, 13. They also examine the long-run absolute welfare effects (i.e. whether the boom increases cumulative social welfare in geographic unit A relative to the counterfactual in which A has but does not produce resources). The relative and absolute effects differ, because the general equilibrium effects differ. 15 Recessions could affect these relationships. There is a literature on the ‘cleansing’ effects of recessions (Davis and Haltiwanger 1990, 1992, 1999, Caballero and Hammour 1994, 1996). There is also large macroeconomic literature on oil prices and recessions. See Hamilton (2011, 2012) and Kilian and Vigfusson (2014). Kilian and Vigfusson (2014) discuss nonlinearity of the relationships. In unreported regressions, we did not find statistically significant differential effects during periods of recession. 16 Political institutions could also affect these relationships. Political institutions can affect growth, particularly if countries or states with weak institutions are unable to realize gains from resources (Mehlum et al 2006, Cabrales and Hauk 2011, van der Ploeg 2011, Berkowitz and Clay 2011). In the U.S. context Southern states are viewed as having had weaker institutions during certain time periods. From the turn of the century through roughly 1970, a single party dominated state politics in the former Confederate states. Following the Voting Rights Acts of 1965 and its 1970 amendment, political competition began to increase in Southern states. Besley et al (2010) find that these changes led to increases in per capita income. If stronger institutions led to changes in resource production or use of resource income, then the relationship between resources and growth may have changed. In unreported regressions, we did not find statistically significant differential effects for the South. 17 High rates of federal land ownership in western states could affect these relationships. In unreported regressions, we did not find statistically significant differential effects for states with high rates of federal land ownership.

The data on resource endowments, resource employment and resource income were

discussed in Section 2. This section considers data on outcomes including per capita income,

population and employment in various sectors. Appendix Figure A2 shows the evolution of

population and per capita income in constant 2010 dollars over time. Data on state personal

income are available annually beginning in 1929 from the Bureau of Economic Analysis. One

can see the effects of major events including the Great Depression, WWII, and the Great

Recession. Appendix Figures A3 and A4 plot average wages, total employment, and

employment in specific sectors. Population, employment and wage data are from the Bureau of

Economic Analysis (BEA).18

Figure 3 plots the distribution of the five-year annualized income per capita and

population growth rate. The average income per capita growth rate is around 2.5% per year. The

average population growth rate is 1.2% per year.

Tables 2a and 2b present the summary statistics for the main variables used in the

analysis. Summary statistics for other variables are available in the Appendix Table A1.

5. Identification

The Allcott and Keniston (2018) model has implications for states with higher and lower

endowments if there is variation over time in resource employment such as we observe in Figure

2. The relative effects are denoted τr, where τr is the effect of an increase in resource employment

on the average difference in outcomes between states with higher and lower endowments. τr

captures spillovers from learning by doing and agglomeration and any other general equilibrium

effects.19

18 State specific employment by sectors is not available prior to 1969. 19 Allcott and Keniston (2018) also estimate τa, the treatment on the treated. This is possible, because they use county data and so can measure spillovers. Empirically they find that τr > τa.

To investigate the empirical relationship between resources and various economic

outcomes we estimate the following reduced form equation:20

𝛥𝛥𝛥𝛥𝛥𝛥𝑌𝑌𝑖𝑖𝑖𝑖 = 𝜏𝜏𝑟𝑟𝛼𝛼𝑖𝑖𝑟𝑟𝛥𝛥𝛥𝛥𝛥𝛥𝐸𝐸𝑟𝑟𝑖𝑖 + 𝜔𝜔𝑖𝑖𝛥𝛥𝛥𝛥𝑌𝑌𝑖𝑖0 + 𝜑𝜑𝑑𝑑𝑖𝑖 + 𝜀𝜀𝑖𝑖𝑖𝑖 (1)

Yit is an outcome in state i in year t. αir is endowment of resource r in the baseline period.

Ert is national employment or income for resource r in time t. Yi0 is a baseline value of the

outcome for state i.21 φdt are census region-year fixed effects. Economic outcomes may be

moving for reasons other than shifts in resources. To address this, we interact baseline values

with year fixed effects, as well as control for census region by year fixed effects. We use robust

standard errors that are clustered by state.22

The variables Y and E are logged, so ΔlnYit is approximately equal to the growth rate in

the outcome variable, and ΔlnErt is approximately equal to the growth rate in national resource

employment. The changes are measured over one year (from t to t-1) or five years (from t to t-5).

The variable αirΔlnErt is similar to the shift-share approach used in Allcott and Keniston

(2018). Here the share comes from the cross-sectional variation in the resource endowment per

square mile in 1935, αir. The shift comes from changes in national resource employment, ΔlnErt.

The estimated τr is similar to elasticity, where τr is the differential effect of a one percent increase

in national resource employment in the state with the largest resource endowment per square

mile.

If increases and decreases in resource employment are uncorrelated with unobserved

economic trends, conditional on baseline outcomes interacted with year and census region year 20 The regression could also be estimated using fixed effects, but differencing is more efficient if errors are serially correlated. 21 Per capita income or population in 1929 is included as a control in specifications where the outcome is per capita income or population respectively. In unreported regressions, the estimated effects are similar if the average value in 1929-1934 is included as a baseline instead of the variable in 1929. 22 In unreported regressions, we bootstrapped the standard errors for some specifications. Bootstrapping does not change the statistical significance of the results.

fixed effects, Equation (1) will produce unbiased estimates of τr. Figure 2 shows that the three

resources follow different time paths. Any confounder would have to follow one of the three

time trends and differentially affect states with higher endowments of that resource.

One limitation of equation (1) is that it restricts the effects to be similar for increases and

decreases in resource employment. A number of papers including Black et al (2005), and

Jacobsen and Parker (2014) suggest that there may be differential effects of increases and

decreases in resources. To allow the effects to differ during booms and busts, we estimate the

following equation:

𝛥𝛥𝛥𝛥𝛥𝛥𝑌𝑌𝑖𝑖𝑖𝑖 = 𝜏𝜏𝑟𝑟−𝛼𝛼𝑖𝑖𝑟𝑟1(𝛥𝛥𝛥𝛥𝛥𝛥𝐸𝐸𝑟𝑟𝑖𝑖 < 0)𝛥𝛥𝛥𝛥𝛥𝛥𝐸𝐸𝑟𝑟𝑖𝑖 + 𝜏𝜏𝑟𝑟+𝛼𝛼𝑖𝑖𝑟𝑟1(𝛥𝛥𝛥𝛥𝛥𝛥𝐸𝐸𝑟𝑟𝑖𝑖 ≥ 0)𝛥𝛥𝛥𝛥𝛥𝛥𝐸𝐸𝑟𝑟𝑖𝑖 +

𝜔𝜔𝑖𝑖𝛥𝛥𝛥𝛥𝑌𝑌𝑖𝑖0 + 𝜑𝜑𝑑𝑑𝑖𝑖 + 𝜀𝜀𝑖𝑖𝑖𝑖 (2)

where 1(𝛥𝛥𝛥𝛥𝛥𝛥𝐸𝐸𝑟𝑟𝑖𝑖 < 0) and 1(𝛥𝛥𝛥𝛥𝛥𝛥𝐸𝐸𝑟𝑟𝑖𝑖 ≥ 0) are dummy variables indicating a decline and an

increase in sectoral 𝑟𝑟 employment 𝐸𝐸𝑟𝑟𝑖𝑖 between t and t-5, respectively. The coefficients of interest

𝜏𝜏𝑟𝑟− and 𝜏𝜏𝑟𝑟+ show the differential effects of resources during boom and bust periods respectively.

6. Short-Run Effects

Short-Run Effects of Resources on Population and Growth in Per Capita Income

Table 3 presents estimates of the relationship between resources and growth in

population. Columns 1 and 2 of Table 3 report the estimates of equation (1) for 1-year and 5-year

time intervals, assuming the effect is symmetric across increases and decreases in resource

employment. The 1-year difference specification (column 1) assumes that changes in resource

employment immediately translate into growth in population in states with higher resource

endowments, while 5-year differences (column 2) allow the effects to develop over a longer time

period. In columns 1 and 2, the coefficients are similar in sign and significance, but the

magnitudes are larger over the 5-year period. National increases in oil & natural gas employment

are positively but not significantly related to growth in population in states with higher

endowments. National increases in coal employment are positively and statistically significantly

related to growth in population in states with higher endowments. National increases in

agricultural employment are negatively and statistically significantly related to growth in

population in states with higher endowments. That is, increases in agricultural employment are

associated with relative declines in overall population. Columns 3-5 of Table 3 present the

results for the more flexible boom-bust specification from equation (2) and examine the effects

across three time periods: 1936-2015, 1936-1974, and 1975-2015.

Compared to the results in column 2 of Table 3, the results in columns 3-5 tell a different

and more nuanced story. In column 3, the coefficient on oil & natural gas for increases and

decreases in resource employment are both positive but not statistically significant. The

coefficient on coal for increases in resource employment, which was positive and significant in

column 2, is now negative and statistically significant. The coefficient on coal for decreases in

resource employment is positive and statistically significant. States with high coal endowments

face population declines during both increases and decreases in national coal employment. The

coefficient on increases in agricultural employment is negative and statistically significant, while

the coefficient on decreases is negative and not significant. States with high agricultural

endowments face population declines during increases in national agricultural employment. In

columns 4 and 5 some coefficients for a given resource and direction of change in employment

differ in magnitudes, significance, and for agriculture during periods of decline in both sign and

significance.

Table 4 presents the coefficients for the same specifications, where the dependent

variable is growth in per capita income. We are interested in growth in per capita income,

because in parts of the literature it is used implicitly or explicitly as a measure of welfare. As in

Table 3, the symmetric results in columns 1 and 2 and the asymmetric results in the columns 3-5

have different implications.

Tables A2 and A3 in the appendix present the specifications for population and per capita

income comparing 1935 endowments per square based on 1935 knowledge to the following

alternatives: i) 1935 endowments based on 2015 knowledge; ii) 1935 endowment per capita

based on 1929 population; iii) agricultural land per square mile in 1935 instead of land value; iv)

changes in national resource income instead of employment. The results are qualitatively similar

for oil and coal. For agriculture, the results are sensitive to the specification.

Figure 4 plots by resource the effects implied by estimates in Tables 3 and 4 of a one

standard deviation increase in employment for the state with the highest endowment. Recall that

the endowment of the state with the largest endowment is equal to 1, so a state with X% the

endowment per square mile of the state with the largest endowment would experience an effect

that is X% of that of that state. It is worth noting that declines have been multiplied by a

negative number because employment fell and so have effects that are opposite in sign to the

coefficients on declines in Tables 3 and 4.

Figure 4 highlights four points. First, there is no one relationship between resources and

outcomes. Different resources have different short-run effects in different time periods, across

increases and decreases in resource employment, and across different outcomes. To get a sense

of the differences across resources and outcomes, for example, one can examine increases in

resource employment during 1935-1974. Increases in national oil & natural gas employment

have no effect on population growth or income per capita in states with higher endowments,

while increases in national coal employment have a negative and statistically significant effect

on population growth and a positive and statistically significant effect on income. Increases in

national agricultural employment have a statistically significant negative effect on population

growth but no effect on per capita income.

Second, to the extent that growth in population responds to changes in resource

employment, it is primarily responsive to declines – rather than increases – in resource

employment.23 The model predicts that population would increase during increases in resource

employment. Two of the six effects (oil) are not statistically significant and the remaining four

(coal and agriculture) are negative and statistically significant. Overall employment in coal and

agriculture was declining over the period 1936-2015. Even during periods of increases in

employment in these sectors, population continued to fall, possibly because individuals were

forecasting further declines. The model predicts that population would typically decrease during

decreases in resource employment, although the magnitude would depend on the extent of the

spillovers. Consistent with this four of the six effects are negative and statistically significant,

one is negative and not significant (oil for 1936-1974), and one is positive and statistically

significant (agriculture for 1936-1974).

Third, the coefficients for growth in population and for growth in per capita income for

an increase or decrease resource in a given time period are not necessarily the same in sign or

significance. For oil, two of the four pairs of coefficients differ in significance. For coal, all four

pairs differ in sign or significance. And for agriculture, all four pairs differ in sign or

significance. This is important, because growth in population and growth in per capita income

are frequently used as proxies for welfare in the literature. In the model, population is a proxy for

23 In unreported regressions, the coefficients from the specification with 1-year differences are qualitatively similar, do this is not being driven by the use of 5-year differences.

welfare. All but one of the coefficients on population are either insignificant or negative and

significant.

Fourth, for oil & natural gas and coal across a hypothetical boom-bust cycle, states could

be worse off in relative terms after the cycle than before the cycle. Although the differences are

not always statistically significant, the coefficient on the decline is always bigger in magnitude

than the coefficient on increase in resource employment. In the model, this could occur if in the

short run learning by doing spillovers were larger than agglomeration spillovers.

Short-Run Effects of Resources on Wages and Employment

Table 5 explores the effects of resources on mining (all resource extraction), agricultural,

and manufacturing wages for 1975-2015.24 As predicted by the model, in Panel B, increases in

oil employment and coal employment are associated with increases in wages in both mining and

manufacturing in states with higher endowments. Similarly, increases in agricultural employment

are positively related to agricultural wages and positively and significantly related to

manufacturing wages in states with higher endowments.

Table 6 explores employment effects for total employment, retail, manufacturing,

transportation, and construction for the same period. In Panel B column 1 shows the effects for

total employment. For increases in resource employment, the coefficients are positive and

significant for oil & natural gas, negative and not significant for coal, and negative and

significant for agriculture. For decreases in resource employment, all three coefficients are

positive and significant. During declines in resource employment, total employment falls more in

states with larger resource endowments. These results for oil & natural gas, coal, and agriculture

are similar to the effects on population shown in Table 3 and Figure 4. The effects on retail and

24 Over this period, separate series are not available for oil & natural gas and coal.

manufacturing employment are broadly similar to the effects on total employment for all three

resources. The effects on transportation and construction vary by resource.

7. Long-run Effects of Resources

Table 7 examines the long-run relationship between resource endowments and growth in

population in Panel A and between resource endowments and growth in per capita income in

Panel B over the periods 1936-2015, 1936-1974, and 1975-2015.25 All columns include controls

for initial levels of the outcome, either population in 1929 or income per capita in 1929, and

region fixed effects.26

Figure 5 plots the coefficients from Table 7 by resource for population growth and

growth in per capita income. States with higher coal endowments and agricultural endowments

experienced slower long-run relative population growth. The coefficients on coal endowments in

all time periods and the coefficient on agriculture in 1975-2015 are negative and statistically

significant. Strikingly, endowments have little long-run relationship to per capita income. The

coefficients on endowments are small, positive, and not statistically significant.

Population growth was a key margin of adjustment for states with large endowments of

coal and agriculture. To provide a sense of the magnitudes, consider West Virginia, which had a

large coal endowment, and Illinois, which had a large agricultural endowment. The estimates

from Table 7 column 1 imply that if it did not have any coal, West Virginia’s population in 2015

would have been about 1.3 million higher.27 This is very large effect, given the population of

West Virginia in 2015 was 1.8 million. The estimates from Table 7 column 3 imply that if it did

25 Tables A4 and A5 in the appendix present results for alternative measures of endowment. 26 In unreported regressions, the estimated effects are similar if controls for the former Confederate states or/and federal land are included. 27 For West Virginia, 1.3 million higher equals (1.8*(exp(80*0.0125*0.54)-1)). For Illinois, 3 million higher equals (12.9*(exp(40*0.0077*0.668)-1)).

not have any agricultural land, Illinois’ population in 2015 would have been 3 million higher.

This is also a large effect, given the population of Illinois in 2015 was 12.9 million.

The model tells us that a boom-bust cycle can be welfare enhancing if the geographic unit

sees an increase in population and then returns to its original population. The question is, of

course, over what time frame. The short-run and long-run results for 1936-2015 suggest that

states with larger coal or agricultural endowments had slower relative population growth during

one or both time periods. Was having larger coal or agricultural endowments associated with

faster population growth prior to 1936? If so, then viewed from the perspective of the nineteenth

century, having coal or agricultural resources may have been welfare enhancing.

Coal and agricultural employment peaked before the start of our sample period. Thus, it

is possible that these states experienced faster population growth during the period in which

employment was rising. Figure A5 shows the trajectory for coal employment, which peaked in

1923, and the trajectory for farm population, which peaked in 1935.

Why did employment decline? The decline in coal and agricultural employment was not

a function of output, which continued to increase over time, but rather a function of

technological and organizational improvements that reduced the labor inputs necessary to

achieve a unit of output. Figure A6 plots workers per BTU for coal and workers per acre for

agriculture. Both fell dramatically over time. For coal, Darmstadter and Kropp (1997) and

Ellerman, Stoker, and Berndt (2001) describe the changes in mining technology such as the

adoption of and improvements in long wall mining and the development of highly productive

large-scale mines in the Powder River Basin. For agriculture, Olmstead and Rhode (2000)

demonstrate that improvements were partly driven by the replacement of horses and mules by

tractors, which freed up land, and by improvements in the price and quality of tractors and farm

equipment, which reduced the labor input. In their book, Creating Abundance: Biological

Innovation and American Agricultural Development, Olmstead and Rhode (2008) document the

stream of biological innovations that improved crops and livestock and in many cases resulted in

greater labor productivity.

Transportation costs were falling as well, making it easier for areas to specialize in

specific products and ship them to other markets, thereby realizing the gains from trade.

Redding and Turner (2014) show that rail costs in the U.S. were secularly declining from the late

nineteenth century onward and transportation costs as a share of GDP have fallen since the early

twentieth century. The importance of rail for the U.S. economy has been an active area of

research since Fogel (1964). Recent work by Donaldson and Hornbeck (2016) finds that effects

of the railroad on the U.S. economy in 1890 were larger than originally suggested by Fogel

(1964).28 Further, Costinot and Donaldson (2016) find that there were substantial gains to

economic integration for agriculture between 1880 and 1997. These are all relevant, because they

may have impacted spillovers from resources.

Did states with large coal and agricultural endowments experience faster population

growth prior to 1936? In Appendix Table A6, we explore this using our main 1935 measure of

endowment and the alternative measures of endowment from Appendix Table A2. Over 1880-

1935, seven of the eight coefficients on coal and agricultural endowments are positive. Further,

the coefficient on coal is positive and statistically significant in the main specification and in two

of the three alternative specifications. In sum, although having a larger coal endowment was

associated with statistically significant negative short-run and long-run effects on population

during 1936-2015, it was associated with statistically significant positive long-run effects during

28 See also related work on the effects of railroads on a variety of earlier outcomes including Atack et al (2010) and Haines and Margo (2008).

1880-1935. The effects of having a larger agricultural endowment were qualitatively similar to

the effects of having a larger coal endowment, although the coefficients were less frequently

statistically significant and are somewhat sensitive to how the endowment is measured. It is

worth noting that states with larger agricultural endowments may have experienced faster

population growth prior to 1880.

In contrast to coal and agriculture, oil & natural gas employment is currently at or near an

all time high. Like agriculture and coal, there have been significant improvements in productivity

in oil & natural gas. Bohi (1998) discusses the factors driving productivity in oil discovery and

development. Figure A6 shows that employees per BTU fell into the early 1970s and then

increased during the mid to late 1970s, when oil prices rose. The early period was characterized

by discovery of many of the largest oil & natural gas fields, while the later period was

characterized by the need to focus on efficient extraction of smaller or more difficult to access

fields. We find limited short-run effects and no long-run effects of oil & natural gas endowment

on population during 1936-2015 or during 1880-1935.

In sum, over the long run, the evidence is mixed. Having a larger coal endowment

appears to have been associated with an initial relative population increase during 1880-1935

followed by relative population decline during 1936-2015. Having a larger agricultural

endowment was associated with relative population decline during 1975-2015. Having a larger

oil endowment never had a significant effect on long-run population. When viewed from the

perspective of the nineteenth century, resource endowments may have been welfare neutral or

welfare enhancing.

Our results speak indirectly to the economic history literature on the importance of

resources during the nineteenth and early twentieth centuries. States with larger coal

endowments experienced statistically significantly larger population increases during 1880-1935.

This is consistent with welfare increases in states with larger coal endowments and with

Habakkuk (1962), Wright (1990), and Wright and Czelusta (2004), who argue that mineral

resources had important benefits for the American economy during the nineteenth and early

twentieth centuries. And it is consistent with the European literature on coal and economic

development (Pomeranz 2001, Allen 2009, and Fernihough and O’Rourke 2014).

8. Conclusion

What are the short-run and long-run effects of resources on economic outcomes? To

answer this question, the paper drew on a model of domestic Dutch disease and new state-level

panel datasets spanning 1936-2015 covering oil & natural gas, coal, and agricultural land. The

analysis used a flexible shift-share estimation approach, where the shift was changes in national

employment for that resource and the share was state endowment of a resource per square mile.

In the short run, the paper found that different resources had different short-run effects in

different time periods, across increases and decreases in resource employment, and across

different economic outcomes. These findings suggest that researchers should be cautious about

making general statements regarding the effects of resources on outcomes.

In the long run, the primary effect of resources has been on population growth for some

resources in some time periods. States with larger coal and agricultural endowments experienced

slower long-run growth in population than states with smaller endowments. This held for states

with larger coal endowments for 1936-2015, 1936-1974, and 1975-2015 and for states with

larger agricultural endowments in the later period. As a result of differential population growth

across states, resources had no effect on growth in state per capita income. Even in the long run,

however, one needs to be cautious when making general statements regarding the effects of

resources on outcomes. The robust negative relationship between state coal endowments and

state population in the long run might suggest that the effect of coal had only ever been negative,

and yet when we look back to 1880-1935, the relationship between state coal endowments and

state population was positive.

References Alexeev, Michael, and Robert Conrad. "The elusive curse of oil." The Review of Economics and Statistics 91.3 (2009): 586-598. Allcott, Hunt, and Daniel Keniston. "Dutch Disease or Agglomeration? The Local Economic Effects of Natural Resource Booms in Modern America. " The Review of Economic Studies 85.2 (2018): 596-731. Allen, Robert C. The British industrial revolution in global perspective. Cambridge University Press, 2009. Atack, Jeremy, Fred Bateman, Michael Haines, and Robert A. Margo. “Did Railroads Induce or Follow Economic Growth? Urbanization and Population Growth in the American Midwest, 1850–1860.” Social Science History, 34 (2010):171–197 Berkowitz, Daniel, and Karen B. Clay. The evolution of a nation: how geography and law shaped the American states. Princeton University Press, (2011). Besley, Timothy, Torsten Persson, and Daniel M. Sturm. "Political competition, policy and growth: theory and evidence from the US." The Review of Economic Studies 77.4 (2010): 1329- 1352. Black, Dan, Terra McKinnish, and Seth Sanders. "The economic impact of the coal boom and bust." The Economic Journal 115.503 (2005): 449-476. Bohi, Douglas R. “Changing Productivity in U.S. Petroleum Exploration and Development.”, Discussion Paper 98-38, Resources for the Future, June, Washington. (1998). Boyce, John R., and J. C. Herbert Emery. "Is a negative correlation between resource abundance and growth sufficient evidence that there is a “resource curse”?." Resources Policy 36.1 (2011): 1-13. Caballero, Ricardo J., and Mohamad L. Hammour. "The Cleansing Effect of Recessions." The American Economic Review (1994): 1350-1368. Caballero, Ricardo J., and Mohamad L. Hammour. "On the Timing and Efficiency of Creative Destruction." The Quarterly Journal of Economics (1996): 805-852. Cabrales, Antonio, and Esther Hauk. "The Quality of Political Institutions and the Curse of Natural Resources." The Economic Journal 121.551 (2011): 58-88. Carrington, William J. "The Alaskan labor market during the pipeline era." Journal of Political Economy (1996): 186-218.

Cavalcanti, Tiago V. de V., Kamiar Mohaddes, and Mehdi Raissi. "Growth, development and natural resources: New evidence using a heterogeneous panel analysis." The Quarterly Review of Economics and Finance 51.4 (2011): 305-318. Chambers, E. J., and Gordon, D. F. "Primary Products and Economic Growth: An Empirical Measurement." Journal of Political Economy 74 (1966): 315-32. Clark, Gregory, and David Jacks. "Coal and the industrial revolution, 1700–1869." European Review of Economic History 11.1 (2007): 39-72. Clay, Karen. “Natural Resources and Economic Outcomes” in Economic Evolution and Revolution in Historical Time. Edited by Paul Rhode, Joshua Rosenbloom and David Weiman. Stanford: Stanford University Press, (2011).

Costinot, Arnaud, and Dave, Donaldson. “How Large Are the Gains from Economic Integration? Theory and Evidence from U.S. Agriculture, 1880-1997” mimeo, Stanford University and MIT, November. (2016)

Darmstadter, Joel, and Brian Kropp. “Productivity Change in U.S. Coal Mining,” Discussion Paper 97-40, Resources for the Future, Washington, D.C., July. (1997).

David, A. Paul, and Gavin Wright. "Increasing returns and the genesis of American resource abundance." Industrial and Corporate Change 6.2 (1997): 203-245. Davis, Steven J., and John Haltiwanger. "Gross job creation and destruction: Microeconomic evidence and macroeconomic implications." NBER Macroeconomics Annual 1990, Volume 5. MIT Press, (1990): 123-186. Davis, Steven J., and John Haltiwanger. "Gross Job Creation, Gross Job Destruction, and Employment Reallocation." The Quarterly Journal of Economics (1992): 819-863. Davis, Steven J., and John Haltiwanger. "Gross job flows." Handbook of labor economics 3 (1999): 2711-2805. Donaldson, Dave, and Richard Hornbeck, 2016. "Railroads and American Economic Growth: A “Market Access” Approach," The Quarterly Journal of Economics, vol. 131(2):799-858. Ellerman, Denny, Stoker, Thomas M. and Berndt, Ernest R. “Sources of Productivity Growth in the American Coal Industry: 1972–95,” in Charles R. Hulten, Edwin R. Dean, and Michael J. Harper, eds., New developments in productivity analysis. Chicago: The University of Chicago Press, 2001, pp. 373–414. Fernihough, Alan, and Kevin Hjortshøj O'Rourke. “Coal and the European industrial revolution.” No. w19802. National Bureau of Economic Research, (2014).

Feyrer, James, Erin T. Mansur, and Bruce Sacerdote. “Geographic Dispersion of Economic Shocks: Evidence from the Fracking Revolution.” American Economic Review (2017): 107.4, 1313-1334. Fogel, Robert W. Railroads and American Economic Growth: Essays in Econometric History. Baltimore: Johns Hopkins University Press. (1964). Goldberg, Ellis, Erik Wibbels, and Eric Mvukiyehe. "Lessons from Strange Cases Democracy, Development, and the Resource Curse in the US States." Comparative Political Studies 41.4-5 (2008): 477-514. Habakkuk, H.J. American and British Technology in the Nineteenth Century. Cambridge University Press, Cambridge, MA. (1962). Haines, Michael, and Robert A. Margo, “Railroads and Local Development: The United States in the 1850s.” 2008. Quantitative Economic History: The Good of Counting, J. Rosenbloom , ed. (London: Routledge, 2008), 78–99. Hamilton, James D. Historical oil shocks. No. w16790. National Bureau of Economic Research, (2011). Hamilton, James D. Oil prices, exhaustible resources, and economic growth. No. w17759. National Bureau of Economic Research, (2012). Henderson, Jason, Brent Gloy, and Michael Boehlje. "Agriculture’s Boom-Bust Cycles: Is This Time Different." Economic Review Q IV (2011): 81-103. Hornbeck, Richard. "The Enduring Impact of the American Dust Bowl: Short- and Long-Run Adjustments to Environmental Catastrophe." American Economic Review, 102 (4) (2012): 1477- 1507. Hornbeck, Richard, and Pinar Keskin. "Does agriculture generate local economic spillovers? short-run and long-run evidence from the ogallala aquifer." American Economic Journal: Economic Policy 7.2 (2015): 192-213. Jacobsen, Grant D., and Dominic P. Parker. "The economic aftermath of resource booms: evidence from boomtowns in the American West." The Economic Journal (2014). James, Alex, and David Aadland. "The curse of natural resources: An empirical investigation of US counties." Resource and Energy Economics 33.2 (2011): 440-453. Keay, Ian. "The engine or the caboose? Resource industries and twentieth-century Canadian economic performance." The Journal of Economic History 67.01 (2007): 1-32. Kilian, Lutz, and Robert J. Vigfusson. "The Role of Oil Price Shocks in Causing US Recessions." manuscript, University of Michigan (2014).

Lewis, Frank. “The Canadian Wheat Boom and Per Capita Income: New Estimates.” Journal of Political Economy 83, no. 6 (1975): 1249-1257 McCloskey, Deirdre N. Bourgeois dignity: Why economics can't explain the modern world. University of Chicago Press, (2010). Matheis, Mike. "Local Economic Impacts of Coal Mining in the United States 1870 to 1970." The Journal of Economic History 76, no. 4 (2016): 1152-1181. Mehlum, Halvor, Karl Moene, and Ragnar Torvik. "Institutions and the Resource Curse." The Economic Journal 116.508 (2006): 1-20. Michaels, Guy. "The Long Term Consequences of Resource‐Based Specialisation." The Economic Journal 121.551 (2011): 31-57. Mitchener, Kris James, and Ian W. McLean. "The productivity of US states since 1880." Journal of Economic Growth 8.1 (2003): 73-114. Mokyr, Joel. Industrialization in the Low Countries, 1795-1850. Yale University Press, (1976). Mokyr, Joel. The lever of riches: Technological creativity and economic progress. Oxford University Press, (1992). Moretti, Enrico. “Local Multipliers.” American Economic Review 100 (2010): 1–7.  Olmstead, Alan L. and Paul W. Rhode. "The Diffusion of the Tractor in American Agriculture: 1910- 1960," NBER Working Paper 7947, October. (2000). Olmstead, Alan L., and Paul W. Rhode. “Creating Abundance: Biological Innovation and American Agricultural Development.” Cambridge: Cambridge University Press, (2008): 480 pp. Papyrakis, Elissaios, and Reyer Gerlagh. "The resource curse hypothesis and its transmission channels." Journal of Comparative Economics 32.1 (2004): 181-193. Papyrakis, Elissaios, and Reyer Gerlagh. "Resource abundance and economic growth in the United States." European Economic Review 51.4 (2007): 1011-1039. Pomeranz, Kenneth. The great divergence: China, Europe, and the making of the modern world economy. Princeton University Press, 2001. Redding, Stephen J. and Matthew A. Turner, 2014. "Transportation Costs and the Spatial Organization of Economic Activity," NBER Working Papers 20235, National Bureau of Economic Research, Inc.

Sachs, Jeffrey, and Andrew Warner. “Natural Resource Abundance and Economic Growth.” National Bureau of Economic Research working paper 5398. Cambridge, MA. (1995). Sachs, Jeffrey, and Andrew Warner. “Natural Resource Abundance and Economic Growth.” Revised version. Unpublished manuscript. Harvard Institute for International Development. Cambridge, MA. (1997). Sala-i-Martin, Xavier, and Arvind Subramanian. Addressing the natural resource curse: An illustration from Nigeria. No. w9804. National Bureau of Economic Research, (2003). U.S. Bureau of Economic Analysis (BEA). 2015. “State Personal Income and Employment: Concepts, Data Sources, and Statistical Methods” available at https://www.bea.gov/regional/histdata/releases/0915spi/spi0915methods.pdf, accessed on May 22, 2018. United States. Bureau of Mines. (Various years). Mineral resources of the United States. Washington: Government Printing Office. United States. Bureau of Mines. (Various years). Minerals yearbook. Washington: Government Printing Office. U.S. Forest Service. (Various years). Agricultural Statistics. U.S. Department of Agriculture. Van der Ploeg, Frederick. "Natural resources: Curse or blessing?." Journal of Economic Literature 49.2 (2011): 366-420. Weber, Jeremy G. "The effects of a natural gas boom on employment and income in Colorado, Texas, and Wyoming." Energy Economics 34.5 (2012): 1580-1588. Weber, Jeremy G. "A decade of natural gas development: The makings of a resource curse?." Resource and Energy Economics 37 (2014): 168-183. Wright, Gavin. “The Origins of American Industrial Success, 1879-1940,” American Economic Review, 80(4) 1990: 651-68. Wright, Gavin, and Jesse Czelusta. "Why economies slow: the myth of the resource curse." Challenge 47.2 (2004): 6-38.

https://www.bea.gov/regional/histdata/releases/0915spi/spi0915methods.pdf

Figure 1 - Resource Endowments in 1935

Oil

Coal

Agriculture

Notes: This figure maps the resource endowments as of 1935. The gradients are based on percentiles, conditional on nonzero value of resources ((0-25, 25-50, 50-75, 75-100)). Oil & natural gas map plots the dollar value of oil & natural gas reserve in 1935, using 1935 oil and natural gas prices. Coal map shows the dollar value of recoverable coal reserves in 1935 using average coal price in 1935. Agriculture map plots the farm value (value of land and buildings in farms) used in agriculture in 1935. Oil & natural gas and coal data are from Minerals Yearbooks. Agriculture data are from United States Department of Agriculture (USDA), Census of Agriculture.

Figure 2 – National Employment: Resource Sectors

Notes: National Employment (in thousands) over time (1935-2015) in different sectors based on 1987 Standard Industrial Classification (SIC) for 1935-2001 and based on North American Industry Classification System (NAICS) for 2002-2015: Agriculture, Oil & natural gas extraction. National employment statistics for oil & natural gas and agriculture sectors for 1935-2015 are taken from the Bureau of Economic Analysis. National coal mining employment is taken from U.S. Bureau of Labor Statistics.

Figure 3 - Income per Capita Growth and Population Growth

Notes: Graph plots the distribution of the main dependent variables: annualized five-year difference in log of income per capita and annualized five-year difference in log of population.

Figure 4 – Effects of a One Standard Deviation Increase in Resource Employment on Population Growth and Income Per Capita Growth

Notes: The figure is based on Tables 3 and 4 and shows the effects of a one standard deviation increase in oil, coal and agriculture employment on population growth and income per capita growth. Vertical bars show the 95% confidence intervals.

Figure 5 – Long-Run Effect of Resource Endowments on Population Growth and Income Per Capita Growth

Notes: Figure plots the coefficients and the 95% confidence intervals from the Table 7.

Table 1 - Resource Production in 1936 and 2015

1936 2015

Agricultural Output 132 345

Fossil Fuels Total 36 211 Coal (Bituminous, Lignite and Anthracite) 16 26

Oil and Natural Gas 21 185

Total Metals 9 24 Iron Ore 2 4

Copper 2 7 Lead 0.6 0.7

Zinc 0.8 1.6 Gold 2 7

Silver 0.8 0.5 Molybdenum 0.2 0.9

Total Nonmetal Minerals 9 48 Cement 3 9

Clay Products 1 1 Lime 0.4 2

Sand and Gravel 1 7 Crushed Stone (including Slate) 2 12

Phosphate Rock 0.2 2 Salt 0.4 2

Sulfur 0.6 0.86

Forest Products

Timber 0.04 0.15 Notes: Value of production/sales in 2010 dollars (in billion). Data for 1936 are from U.S. Bureau of Mines, Mineral Resources of the United, U.S. Bureau of Mines, Minerals Yearbooks U.S. Geological Survey, Minerals Yearbooks (U.S. Department of the Interior). Data for 2015 are from U.S. Department of the Interior, U.S. Geological Survey, Mineral Commodity Summaries. Forest product values are from the U.S. Forest Service, Agricultural Statistics (U.S. Department of Agriculture, 1944 and 2017).

Table 2a –Summary Statistics: Resource Endowments and Employment Variable Obs. Mean Std. Dev.

Panel A: State Resource Endowments in 1935 Oil Endowment 3,840 0.082 0.191

Coal Endowment 3,840 0.095 0.178 Ag Endowment 3,840 0.257 0.209 Panel B: Changes in National Resource Employments 1936-2015 ∆OilEmp 3840 0.022 0.054 ∆CoalEmp 3840 -0.020 0.050 ∆AgEmp 3840 -0.016 0.022

Panel C: Changes in National Resource Employments 1936-1974 ∆OilEmp 1872 0.019 0.036 ∆CoalEmp 1872 -0.024 0.051 ∆AgEmp 1872 -0.021 0.022

Panel D: Changes in National Resource Employments 1975-2015 ∆OilEmp 1,968 0.025 0.067 ∆CoalEmp 1,968 -0.016 0.049 ∆AgEmp 1,968 -0.011 0.021 Notes: Panel A reports summary statistics for state level variables used in the analysis. OilEnd, CoalEnd and AgEnd are oil, coal and farm endowments in 1935 scaled so that the state with largest endowment is coded as 1. Oil & natural gas endowment is the dollar value of oil & natural gas reserves in 1935, using 1935 oil and natural gas prices. Coal endowment is the dollar value of recoverable coal reserves in 1935 using average coal price in 1935. Agriculture endowment is the farm value (value of land and buildings) used in agriculture in 1935. Oil & natural gas and coal data are from Minerals Yearbooks. Agriculture data are from United States Department of Agriculture (USDA), Census of Agriculture. Panels B - D report the summary statistic for national changes in resource employments for the whole sample 1936-2015 (Panel B) and two subsamples: 1936-1974 (in Panel C) and 1975-2015 (in Panel D). ∆OilEmp, ∆CoalEmp and ∆AgEmp are changes in the logged national employment in the oil & natural gas extraction, coal mining and agriculture sectors respectively.

Table 2b - Summary Statistics: Outcome Variables Obs. Mean Std. Dev.

Panel A: 1936-2015 ∆IncPC 3,840 0.025 0.029 ∆Pop 3,840 0.012 0.012

Panel B: 1936-1974 ∆IncPC 1,872 0.036 0.036 ∆Pop 1,872 0.013 0.014

Panel C: 1975-2015 ∆IncPC 1,968 0.014 0.012 ∆Pop 1,968 0.010 0.010 ∆TotEmp 1,968 0.017 0.014 ∆MnfEmp 1,966 0.046 0.023 ∆TransportationEmp 1,960 0.014 0.016 ∆ConstructionEmp 1,962 0.015 0.039 ∆RetailEmp 1,968 0.016 0.017 ∆WholesaleEmp 1,968 0.016 0.023 Notes: Summary statistics for the main outcome variables used in the analysis for the whole sample 1936-2015 and two subsamples: 1936-1974 and 1975-2015. ∆ is five- year difference in logged variables. Data are from BEA.

Table 3- Effects of Natural Resources on Population Growth

(1) (2) (3) (4) (5)

1936- 2015

1936- 1974

1975- 2015

∆Pop ∆Pop ∆Pop ∆Pop ∆Pop VARIABLES D1 D5 D5 D5 D5

OilEnd X ∆OilEmp 0.017 0.039

(0.013) (0.025)

CoalEnd X ∆CoalEmp 0.059* 0.153**

(0.035) (0.066)

AgEnd X ∆AgEmp -0.121** -0.234**

(0.055) (0.112)

(OilEmpDecline=0) X OilEnd X ∆OilEmp

0.018 0.014 0.001

(0.049) (0.065) (0.053)

(OilEmpDecline=1) X OilEnd X ∆OilEmp

0.126 0.129 0.188***

(0.089) (0.276) (0.067)

(CoalEmpDecline=0) X CoalEnd X ∆CoalEmp

-0.127*** -0.229*** -0.106**

(0.044) (0.068) (0.051)

(CoalEmpDecline=1) X CoalEnd X ∆CoalEmp

0.234** 0.225** 0.242**

(0.094) (0.099) (0.099)

(AgEmpDecline=0) X AgEnd X ∆AgEmp

-0.521** -0.703** -0.551**

(0.199) (0.324) (0.209)

(AgEmpDecline=1) X AgEnd X ∆AgEmp

-0.203 -0.420** 0.261**

(0.122) (0.161) (0.122)

Observations 3,840 3,840 3,840 1,872 1,968 R-squared 0.399 0.429 0.444 0.394 0.550 Notes: This table presents estimates of equation (1) in columns 1 and 2 and equation (2) in columns 3-5. OilEnd, CoalEnd and AgEnd are oil, coal and farm endowments in 1935, based on available knowledge in 1935, constructed as described in the resources section. ∆OilEmp, ∆CoalEmp and ∆AgEmp are changes in the logged national employment in the oil & natural gas extraction, coal mining and agriculture sectors respectively. ∆Pop is difference in log of population. D1 and D5 represent one and five year differences. Decline is a dummy variable indicating a decline in respective sectoral employment. Estimated effects in columns 3-5 are relative to zero. Decline = 0 means no decline, Decline=1 means decline in employment between t to t-5. All regressions include controls for census region by year fixed effects. Columns 1-4 also include controls for year interacted with natural log of population in 1929, column 5 includes controls for year interacted with natural log of the population in 1969. The effect of oil during the employment decreases is statistically different from the effect during employment increases over the period 1975-2015. The effect of coal during the employment decreases is statistically different from the effect of coal during employment increases across two sub-periods, but not for the whole time period. The effect of agriculture during the employment decreases is statistically different from the effect during employment increases over the whole time period as well as two sub-periods. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table 4- Effects of Natural Resources on Per Capita Income Growth

(1) (2) (3) (4) (5)

1936- 2015

1936- 1974

1975- 2015

∆Inc PC ∆Inc PC ∆Inc PC ∆Inc PC ∆Inc PC VARIABLES D1 D5 D5 D5 D5

OilEnd X ∆OilEmp 0.123*** 0.162***

(0.023) (0.030)

CoalEnd X ∆CoalEmp 0.025 0.059***

(0.050) (0.019)

AgEnd X ∆AgEmp 0.066 -0.061

(0.060) (0.059)

(OilEmpDecline=0) X OilEnd X ∆OilEmp

0.127*** 0.018 0.156***

(0.024) (0.049) (0.024)

(OilEmpDecline=1) X OilEnd X ∆OilEmp

0.306*** 0.382*** 0.298***

(0.053) (0.101) (0.056)

(CoalEmpDecline=0) X CoalEnd X ∆CoalEmp

0.164*** 0.371** 0.071**

(0.033) (0.152) (0.031)

(CoalEmpDecline=1) X CoalEnd X ∆CoalEmp

0.028 0.024 0.029

(0.018) (0.026) (0.034)

(AgEmpDecline=0) X AgEnd X ∆AgEmp

0.089 0.406 -0.010

(0.187) (0.320) (0.219)

(AgEmpDecline=1) X AgEnd X ∆AgEmp

-0.054 -0.013 -0.063

(0.057) (0.066) (0.080)

Observations 3,840 3,840 3,840 1,872 1,968 R-squared 0.727 0.890 0.891 0.900 0.668 Notes: This table presents estimates of equation (1) in columns 1 and 2 and equation (2) in columns 3-5. OilEnd, CoalEnd and AgEnd are oil, coal and farm endowments in 1935, based on available knowledge in 1935, constructed as described in the resources section. ∆OilEmp, ∆CoalEmp and ∆AgEmp are changes in the logged national employment in the oil & natural gas extraction, coal mining and agriculture sectors respectively. ∆Inc PC is difference in log of income per capita. D1 and D5 represent one and five year differences. Decline is a dummy variable indicating a decline in respective sectoral employment. Estimated effects in columns 3-5 are relative to zero. Decline = 0 means no decline, Decline=1 means decline in employment between t to t-5. All regressions include controls for census region by year fixed effects. Columns 1-4 also include controls for year interacted with natural log of income per capita in 1929, column 5 includes controls for year interacted with natural log of income per capita in 1969.The effect of oil during the employment decreases is statistically different from the effect during employment increases over the period 1975-2015. The effect of coal during the employment decreases is statistically different from the effect of coal during employment increases across two sub-periods, but not for the whole time period. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table 5 - Effects of Natural Resources on Wages: 1975-2015 (1) (2) (3)

1975-2015 1975-2015 1975-2015

∆MinWage ∆AgWage ∆MnfctrWage

VARIABLES D5 D5 D5 Panel A. Symmetric Effect OilEnd X ∆OilEmp 0.343** 0.035 0.104***

(0.128) (0.075) (0.018)

CoalEnd X ∆CoalEmp 0.368*** 0.106* 0.126**

(0.097) (0.059) (0.051)

AgEnd X ∆AgEmp 0.787* 0.107 -0.050

(0.449) (0.159) (0.098)

Observations 1,928 1,968 1,966 R-squared 0.552 0.489 0.927 Panel B. Boom-Bust (OilEmpDecline=0) X OilEnd X ∆OilEmp 0.376** 0.018 0.100***

(0.142) (0.072) (0.014)

(OilEmpDecline=1) X OilEnd X ∆OilEmp 0.247* 0.098 0.133**

(0.138) (0.120) (0.054)

(CoalEmpDecline=0) X CoalEnd X ∆CoalEmp 0.905*** -0.154 0.116***

(0.149) (0.199) (0.033)

(CoalEmpDecline=1) X CoalEnd X ∆CoalEmp 0.078 0.244* 0.131*

(0.112) (0.140) (0.069)

(AgEmpDecline=0) X AgEnd X ∆AgEmp 0.123 0.423 0.495***

(0.909) (0.406) (0.183)

(AgEmpDecline=1) X AgEnd X ∆AgEmp 0.840* 0.070 -0.112

(0.427) (0.179) (0.118)

Observations 1,928 1,968 1,966 R-squared 0.559 0.493 0.928 Notes: This table presents the estimates of equation (1) in Panel A and equation (2) in Panel B. ∆MinWage is the difference in log mining wages, ∆AgWage is the difference in logged wage in agriculture, ∆MnfctWage is the difference in logged manufacturing wages. OilEnd, CoalEnd and AgEnd are oil, coal and farm endowments in 1935, based on available knowledge in 1935, constructed as described in the resources section. ∆OilEmp, ∆CoalEmp and ∆AgEmp are changes in the logged national employment in the oil & natural gas extraction, coal mining and agriculture sectors respectively. Estimated effects in Panel B are relative to zero. Decline = 0 means no decline, Decline=1 means decline in employment between t to t-5. All regressions include controls for year interacted with the respective dependent variable in 1969 and census region by year fixed effects. Number of observations is smaller in column 3 because manufacturing wages are not available for Wyoming in 2002. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table 6 - Effects of Natural Resources on Employment Growth: 1975-2015 (1) (2) (3) (4) (5)

1975-2015 1975-2015 1975-2015 1975-2015 1975-2015

∆Total Emp

∆ Retail Emp

∆ Mnfct Emp

∆ Transportation Emp

∆ Construction Emp

VARIABLES D5 D5 D5 D5 D5 Panel A. Symmetric Effects OilEnd X ∆OilEmp 0.175*** 0.129*** 0.193*** 0.188*** 0.529***

(0.042) (0.031) (0.063) (0.034) (0.083)

CoalEnd X ∆CoalEmp 0.122*** 0.128*** 0.066 0.101* 0.333***

(0.035) (0.026) (0.054) (0.056) (0.075)

AgEnd X ∆AgEmp 0.158 0.268** 0.397** 0.016 -0.083

(0.104) (0.101) (0.173) (0.110) (0.224)

Observations 1,968 1,968 1,966 1,960 1,962 R-squared 0.638 0.772 0.729 0.531 0.645 Panel B. Boom-Bust (OilEmpDecline=0) X 0.106* 0.035 0.157* 0.076 0.379*** OilEnd X ∆OilEmp (0.061) (0.042) (0.088) (0.054) (0.077) (OilEmpDecline=1) X 0.393*** 0.421*** 0.298*** 0.551*** 1.001*** OilEnd X ∆OilEmp (0.061) (0.056) (0.061) (0.097) (0.131) (CoalEmpDecline=0) X -0.028 -0.005 -0.083 -0.018 0.337** CoalEnd X ∆CoalEmp (0.081) (0.071) (0.200) (0.144) (0.166) (CoalEmpDecline=1) X 0.197** 0.193*** 0.143 0.159 0.321*** CoalEnd X ∆CoalEmp (0.084) (0.057) (0.155) (0.140) (0.065) (AgEmpDecline=0) X -0.553* -0.716** -0.801 -0.460 -1.282** AgEnd X ∆AgEmp (0.281) (0.290) (0.487) (0.349) (0.505) (AgEmpDecline=1) X 0.299** 0.465*** 0.569** 0.167 0.183 AgEnd X ∆AgEmp (0.129) (0.132) (0.245) (0.150) (0.172)

Observations 1,968 1,968 1,966 1,960 1,962 R-squared 0.660 0.794 0.735 0.560 0.654 Notes: This table presents estimates of equation (1) in Panel A and equation (2) in Panel B. ∆Total Emp, ∆RetailEmp, ∆MnfctrEmp ∆TransportationEmp and ∆ConstructionEmp are differences in logged total employment, employment in retail, manufacturing employment, transportation and construction employment sectors respectively. OilEnd, CoalEnd and AgEnd are oil, coal and farm endowments in 1935, based on available knowledge in 1935, constructed as described in the resources section. ∆OilEmp, ∆CoalEmp and ∆AgEmp are changes in the logged national employment in the oil & natural gas extraction, coal mining and agriculture sectors respectively. Estimated effects in Panel B are relative to zero. Decline = 0 means no decline, Decline=1 means decline in employment between t to t-5. All regressions include controls for year interacted with the respective dependent variable in 1969 and census region by year fixed effects. Number of observations is smaller in columns 3, 4 and 5 because BEA employment data are not available for all states in all years to avoid disclosure of confidential information. Specifically, manufacturing employment is not available for Wyoming in 2002, transportation employment is not available for Rhode Island and Wyoming in 2001 and 2002, employment in the construction sector is not available for Rhode Island and Wyoming in 2002 and for Delaware in 2005. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table 7 - Long-Run Effects of Resource Endowments (1) (2) (3) VARIABLES 1936-2015 1936-1975 1975-2015 Panel A. Population Oil Endowment -0.0027 -0.001 -0.0061

(0.004) (0.005) (0.005)

Coal Endowment -0.0125** -0.016*** -0.0092*

(0.005) (0.006) (0.005)

Ag Endowment 0.0014 0.008 -0.0077**

(0.004) (0.006) (0.003)

Observations 48 48 48 R-squared 0.5235 0.428 0.5915 (4) (5) (6) VARIABLES 1936-2015 1936-1975 1975-2015 Panel B. Income Per Capita Oil Endowment 0.0016 0.002 0.0010

(0.001) (0.002) (0.001)

Coal Endowment 0.0030 0.005 0.0010

(0.003) (0.004) (0.002)

Ag Endowment 0.0022 0.003 0.0010

(0.002) (0.003) (0.002)

Observations 48 48 48 R-squared 0.7978 0.881 0.5408 Notes: This table presents the estimated long-run effects of resource endowments on average annualized population and income per capita growth for the whole time period: 1936-2015 as well as for the two sub-periods: 1936-1974 and 1975-2915. All columns include controls for the initial conditions: population or income per capita in 1929 and census region fixed effects. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Appendix Figure A1- Income: Resource Sectors

Notes: Income (in millions) from resource sectors, in 2010 dollars. Oil& natural gas from Alaska excluded from oil & natural gas income. Coal and oil & natural gas income data are from the U.S. Bureau of Mines, Minerals Yearbooks and U.S. Energy Information Administration (EIA), Annual Energy Review. Agriculture income data are taken from the United States Department of Agriculture (USDA).

Figure A2 –Income per Capita and Population over Time

Notes: Graph plots income per capita in 2010 dollars and population over time and 5th and 95th percentile. Data are taken from the Bureau of Economic Analysis (BEA).

Figure A3 –Total Employment and Log(Wages)

Notes: Figure shows total employment per 1,000 and wages in log for mining(MinW), manufacturing(MnfcW), and agriculture(AgW) sectors over time. Data are taken from Bureau of Economic Analysis (BEA).

Figure A4 – Employment: Non-Resource Sectors

Notes: The average state employment in non resource sectors: manufacturing, construction, transportation, and retail over 1970-2015. Employment is based on 1987 Standard Industrial Classification (SIC) for 1970-2001 and is based on North American Industry Classification System (NAICS) for 2002-2015. Data are taken from Bureau of Economic Analysis (BEA).

Figure A5 – Coal Employment and Farm Population (in Thousands): 1880-2012

Notes: This figure shows the coal employment and farm population over the period 1880-2012. Data on coal employment are taken from U.S. Bureau of Mines, Mineral Resources of the United States. 1932–1970: U.S. Bureau of Mines, Minerals Yearbooks. 1971–1993: U.S. Energy Information Administration, Coal Data. 1994–2000: Coal Industry Annual. Data on farm population are taken from U.S. Census Bureau, decennial census publications, Leon E. Truesdell (1960), “Farm Population:1880 to 1950, “U.S. Bureau of the Census, Technical Paper No.3 and Census of Agriculture 2012.

Figure A6 – Employment per Unit of Output: 1935-2015

Notes: Figure shows the employment per unit of output. Oil & natural gas and Coal are employments in oil & natural gas and coal sectors per million Btu respectively. Agriculture is the employment in agricultural sector per acre.

Table A1 – Summary Statistics Variable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

Panel A: D1 - One year difference

1936-2015 1936-1969 1970-2015

∆OilEmp 0.02 0.089 0.016 0.051 0.023 0.109 ∆CoalEmp -0.023 0.081 -0.035 0.089 -0.015 0.074 ∆AgEmp -0.016 0.042 -0.025 0.036 -0.009 0.044

Panel B: D5 - Five years difference

1936-2015 1936-1969 1970-2015

∆OilEmp(Decline=1) -0.028 0.028 -0.015 0.008 -0.04 0.034 ∆OilEmp(Decline=0) 0.057 0.038 0.042 0.029 0.071 0.039 ∆CoalEmp(Decline=1) -0.053 0.029 -0.055 0.039 -0.052 0.014 ∆CoalEmp(Decline=0) 0.033 0.024 0.026 0.016 0.04 0.028 ∆AgEmp(Decline=1) -0.026 0.017 -0.027 0.02 -0.024 0.015 ∆AgEmp(Decline=0) 0.01 0.008 0.008 0.008 0.011 0.008

∆OilInc 0.026 0.072 0.049 0.039 0.007 0.086 ∆CoalInc -0.31 1.196 0.009 0.055 -0.574 1.569 ∆AgInc 0.011 0.045 0.032 0.048 -0.006 0.033

Obs 3,840 1,632 1,968 Notes: Summary statistics for the variables used in the analysis for the whole sample 1936-2015 and two subsamples: 1936-1969 and 1970-2015.

Table A2 – Effects of Natural Resources on Population Growth: Alternative Measures (1) (2) (3) (4) (5)

1936-1974 1936-1974 1936-1974 1936-1974 1936-1974

VARIABLES 1936 Knowledge 2015 Knowledge Per Capita Land Income Panel A. 1936-1974 (OilEmpDecline=0) x 0.014 -0.032 0.094* 0.004 0.009

OilEnd x ∆OilX (0.065) (0.072) (0.051) (0.064) (0.050) (OilEmpDecline=1) x 0.129 0.173 0.267 0.234 -0.216

OilEnd x ∆OilX (0.276) (0.251) (0.339) (0.281) (0.276) (CoalEmpDecline=0) x -0.229*** -0.098 -0.277*** -0.183*** -0.176***

CoalEnd x ∆CoalX (0.068) (0.074) (0.094) (0.054) (0.049) (CoalEmpDecline=1) x 0.225** 0.168* 0.215* 0.186** 0.044

CoalEnd x ∆CoalX (0.099) (0.085) (0.123) (0.085) (0.044) (AgEmpDecline=0) x -0.703** -0.717* -0.183 -0.459 -0.052

AgEnd x ∆AgX (0.324) (0.402) (0.317) (0.413) (0.043) (AgEmpDecline=1) x -0.420** -0.619*** 0.637*** 0.697** 0.221***

AgEnd x ∆AgX (0.161) (0.137) (0.167) (0.274) (0.080)

Observations 1,872 1,872 1,872 1,872 1,872 R-squared 0.394 0.423 0.429 0.423 0.378

1975-2015 1975-2015 1975-2015 1975-2015 1975-2015

VARIABLES 1936 Knowledge 2015 Knowledge Per Capita Land Income Panel B. 1975-2015 (OilEmpDecline=0) x 0.001 0.012 0.022 0.010 -0.006

OilEnd x ∆OilX (0.053) (0.047) (0.061) (0.065) (0.042) (OilEmpDecline=1) x 0.188*** 0.226*** 0.119 0.169* 0.077***

OilEnd x ∆OilX (0.067) (0.061) (0.101) (0.087) (0.028) (CoalEmpDecline=0) x -0.106** -0.089 -0.103 -0.070 -0.058*

CoalEnd x ∆CoalX (0.051) (0.057) (0.080) (0.049) (0.031) (CoalEmpDecline=1) x 0.242** 0.229** 0.269* 0.201** 0.225**

CoalEnd x ∆CoalX (0.099) (0.086) (0.147) (0.096) (0.099) (AgEmpDecline=0) x -0.551** -0.305* 0.222 -0.639* -0.085**

AgEnd x ∆AgX (0.209) (0.169) (0.251) (0.350) (0.042) (AgEmpDecline=1) x 0.261** 0.089 0.043 0.411* 0.168**

(0.122) (0.100) (0.152) (0.216) (0.066)

Observations 1,968 1,968 1,968 1,968 1,968 R-squared 0.550 0.563 0.537 0.564 0.525 Notes: This table presents estimates of equation (2). Panel A and Panel B present the results for 1936-1974 and for 1975-2015 respectively. Endowment measures based on 1935 used in column 1 (same specifications as in Table 3 columns 4 and 5), 2015 knowledge of endowments are used in column 2. In column 3 resource endowments are measured per population in 1929, land area in acres used in agricultural sector per square mile, and 1935 knowledge about coal and oil & natural gas used as endowments in column 4, and, in column 5, the national sectoral income used instead of national sectoral employment. The construction of the variables is described in the resources section. All differences are five year differences. Decline is a dummy variable indicating a decline in respective sectoral employment. Decline = 0 means no decline, Decline=1 means decline in employment between t to t-5. Estimated effects are relative to zero. All regressions include controls for census region by year. Panel A also includes controls for year interacted with natural log of population in 1929, Panel B includes controls for year interacted with natural log of population in 1969. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table A3 – Effects of Natural Resources on Income Per Capita Growth: Alternative Measures (1) (2) (3) (4) (5)

1936-1974 1936-1974 1936-1974 1936-1974 1936-1974

VARIABLES 1936 Knowledge 2015 Knowledge Per Capita Land Income Panel A. 1936-1974 (OilEmpDecline=0) x 0.018 0.063 0.084* 0.016 0.025

OilEnd x ∆OilX (0.049) (0.062) (0.042) (0.046) (0.031) (OilEmpDecline=1) x 0.382*** 0.400*** 0.531*** 0.384*** -0.323***

OilEnd x ∆OilX (0.101) (0.097) (0.079) (0.099) (0.099) (CoalEmpDecline=0) x 0.371** 0.124 0.548*** 0.342** 0.116

CoalEnd x ∆CoalX (0.152) (0.087) (0.065) (0.159) (0.100) (CoalEmpDecline=1) x 0.024 0.020 0.076** 0.025 0.213**

CoalEnd x ∆CoalX (0.026) (0.028) (0.029) (0.026) (0.081) (AgEmpDecline=0) x 0.406 0.208 0.893*** 0.797** 0.088

AgEnd x ∆AgX (0.320) (0.243) (0.191) (0.322) (0.079) (AgEmpDecline=1) x -0.013 0.028 -0.050 -0.014 -0.026

AgEnd x ∆AgX (0.066) (0.072) (0.050) (0.062) (0.084)

Observations 1,872 1,872 1,872 1,872 1,872 R-squared 0.900 0.899 0.903 0.900 0.901

1975-2015 1975-2015 1975-2015 1975-2015 1975-2015

VARIABLES 1936 Knowledge 2015 Knowledge Per Capita Land Income Panel B. 1975-2015 (OilEmpDecline=0) x 0.156*** 0.186*** 0.143** 0.156*** 0.102***

OilEnd x ∆OilX (0.024) (0.037) (0.057) (0.021) (0.018) (OilEmpDecline=1) x 0.298*** 0.336*** 0.236** 0.299*** 0.144***

OilEnd x ∆OilX (0.056) (0.068) (0.114) (0.056) (0.029) (CoalEmpDecline=0) x 0.071** 0.052 0.082 0.051* 0.071***

CoalEnd x ∆CoalX (0.031) (0.039) (0.081) (0.028) (0.024) (CoalEmpDecline=1) x 0.029 0.055** 0.022 0.036 -0.024

CoalEnd x ∆CoalX (0.034) (0.024) (0.030) (0.034) (0.027) (AgEmpDecline=0) x -0.010 -0.097 0.557** 0.575*** 0.067

AgEnd x ∆AgX (0.219) (0.161) (0.230) (0.179) (0.090) (AgEmpDecline=1) x -0.063 -0.054 -0.038 -0.001 0.155

AgEnd x ∆AgX (0.080) (0.061) (0.060) (0.066) (0.097)

Observations 1,968 1,968 1,968 1,968 1,968 R-squared 0.668 0.684 0.662 0.671 0.659 Notes: This table presents estimates of equation (2). Panel A and Panel B present the results for 1936-1974 and for 1975-2015 respectively. Endowment measures based on 1935 used in column 1, 2015 knowledge of endowments are used in column 2. In column 3 resource endowments are measured per population in 1929, land area in acres used in agricultural sector per square mile used as endowments in column 4, and, in column 5, the national sectoral income used instead of national sectoral employment. All differences are five year differences. Decline is a dummy variable indicating a decline in respective sectoral employment. Decline = 0 means no decline, Decline=1 means decline in employment between t to t-5. Estimated effects are relative to zero. All regressions include controls for census region by year. Panel A also includes controls for year interacted with natural log of per capita income in 1929, Panel B includes controls for year interacted with natural log of per capita income in 1969. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table A4 – Lon-Run Effects of Resource Endowments on Population Growth: Alternative Endowments Measures

(1) (2) (3) (4)

1936-2015 1936-2015 1936-2015 1936-2015

VARIABLES 1936 knowledge 2015 knowledge Per Capita Land Panel A. 1936-2015 Oil Endowment -0.0027 -0.0038 -0.0020 -0.0032

(0.004) (0.004) (0.005) (0.006)

Coal Endowment -0.0125** -0.0098** -0.0124* -0.0088**

(0.005) (0.005) (0.007) (0.004)

Ag Endowment 0.0014 0.0063* -0.0089** -0.0164**

(0.004) (0.004) (0.004) (0.008)

Observations 48 48 48 48 R-squared 0.5235 0.5461 0.5724 0.6026 (5) (6) (7) (8) VARIABLES 1936 knowledge 2015 knowledge Per Capita Land Panel B. 1936-1974 Oil Endowment -0.001 -0.002 -0.001 -0.003

(0.005) (0.005) (0.006) (0.006)

Coal Endowment -0.016*** -0.011** -0.015* -0.011**

(0.006) (0.005) (0.008) (0.005)

Ag Endowment 0.008 0.015*** -0.017*** -0.023**

(0.006) (0.005) (0.005) (0.011)

Observations 48 48 48 48 R-squared 0.428 0.460 0.506 0.519 (9) (10) (11) (12) VARIABLES 1936 knowledge 2015 knowledge Per Capita Land Panel C. 1975-2015 Oil Endowment -0.0061 -0.0063 -0.0049 -0.0041

(0.005) (0.004) (0.006) (0.005)

Coal Endowment -0.0092* -0.0098** -0.0066 -0.0092

(0.005) (0.004) (0.005) (0.006)

Ag Endowment -0.0077** -0.0036 -0.0109 0.0015

(0.003) (0.003) (0.007) (0.005)

Observations 48 48 48 48 R-squared 0.5915 0.6168 0.6139 0.5597 Notes: This table presents the estimated long-run effects of resource endowments on population growth for the whole time period in Panel A: 1936-2015 as well as for the two sub-periods in: 1936-1975 and 1975-2915 in Panel B and C respectively. In column 1 the endowments of resources are measured per square mile based on 1936 knowledge about the endowments, as in Table 7, in column 2 the endowments of resources are measured per square mile based on 2015 knowledge about the endowments, in column 3 land area in acres used in agricultural sector per square mile, rather than value of that land as a measure of agricultural endowment and 1935 knowledge about coal and oil & natural gas endowments, and in column 4 endowments are measured by population in 1929. All columns include controls for the initial conditions: population or income per capita in 1929 and census region fixed effects. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table A5 – Long-Run Effects of Resource Endowments on Income Per Capita Growth: Alternative Endowments Measures

(1) (2) (3) (4)

1936-2015 1936-2015 1936-2015 1936-2015

VARIABLES 1936 knowledge 2015 knowledge Land Per Capita Panel A. 1936-2015 Oil Endowment 0.0016 0.0030 0.0009 0.0013

(0.001) (0.002) (0.001) (0.001)

Coal Endowment 0.0030 -0.0011 0.0037*** 0.0020

(0.003) (0.001) (0.001) (0.003)

Ag Endowment 0.0022 0.0005 0.0050*** 0.0043**

(0.002) (0.002) (0.001) (0.002)

Observations 48 48 48 48 R-squared 0.7978 0.7905 0.8657 0.8185 (5) (6) (7) (8) VARIABLES 1936 knowledge 2015 knowledge Land Per Capita Panel B. 1936-1974 Oil Endowment 0.002 0.003 0.000 0.001

(0.002) (0.003) (0.001) (0.001)

Coal Endowment 0.005 -0.001 0.004** 0.004

(0.004) (0.002) (0.002) (0.004)

Ag Endowment 0.003 -0.000 0.008*** 0.006**

(0.003) (0.003) (0.002) (0.002)

Observations 48 48 48 48 R-squared 0.881 0.871 0.925 0.893 (9) (10) (11) (12) VARIABLES 1936 knowledge 2015 knowledge Land Per Capita Panel C. 1975-2015 Oil Endowment 0.0010 0.0022** 0.0009 0.0013

(0.001) (0.001) (0.001) (0.001)

Coal Endowment 0.0010 -0.0019* 0.0004 0.0029***

(0.002) (0.001) (0.002) (0.001)

Ag Endowment 0.0010 0.0007 0.0026 0.0020*

(0.002) (0.002) (0.002) (0.001)

Observations 48 48 48 48 R-squared 0.5408 0.5752 0.5641 0.6234 Notes: This table presents the estimated long-run effects of resource endowments on income per capita growth for the whole time period in Panel A: 1936-2015 as well as for the two sub-periods in: 1936- 1975 and 1975-2915 in Panel B and C respectively. In column 1 the endowments of resources are measured per square mile based on 1936 knowledge about the endowments, as in Table 7, in column 2 the endowments of resources are measured per square mile based on 2015 knowledge about the endowments, in column 3 endowments are measured by population in 1929, and in column 4 land area in acres used in agricultural sector per square mile, rather than value of that land as a measure of agricultural endowment and 1935 knowledge about coal and oil & natural gas endowments. All columns include controls for the initial conditions: population or income per capita in 1929 and census region fixed effects. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels.

Table A6 – Long-Run Effects of Resource Endowments on Population Growth: 1880-1935 (1) (2) (3) (4)

1880-1935 1880-1935 1880-1935 1880-1935

VARIABLES 1936 knowledge 2015 knowledge Per Capita Land Oil Endowment 0.0086 0.0137 -0.0029 0.0079

(0.006) (0.009) (0.008) (0.006)

Coal Endowment 0.0149** 0.0038 0.0192** 0.0157**

(0.007) (0.003) (0.009) (0.007)

Ag Endowment 0.0046 0.0087 0.0032 -0.0030

(0.007) (0.009) (0.008) (0.012)

Observations 47 47 47 47 R-squared 0.6043 0.5955 0.6011 0.6025 Notes: This table presents the effects of resource endowments on annualized population growth over the period 1880-1935. In column 1 the endowments of resources are measured per square mile based on 1936 knowledge about the endowments, in column 2 the endowments of resources are measured per square mile based on 2015 knowledge about the endowments, in column 3 endowments are measured by population in 1929, and in column 4 land area in acres used in agricultural sector per square mile, rather than value of that land as a measure of agricultural endowment and 1935 knowledge about coal and oil & natural gas endowments. All columns include controls for the initial conditions: population in 1880 and census region fixed effects. Standard errors are clustered at the state level and are in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent levels. Oklahoma is not in the sample.

Karen Clay, Carnegie Mellon University and NBER0F(
Margarita Portnykh, Carnegie Mellon University
References
Habakkuk, H.J. American and British Technology in the Nineteenth Century. Cambridge
Haines, Michael, and Robert A. Margo, “Railroads and Local Development: The
United States in the 1850s.” 2008. Quantitative Economic History: The Good of
Counting, J. Rosenbloom , ed. (London: Routledge, 2008), 78–99.
Hamilton, James D. Historical oil shocks. No. w16790. National Bureau of Economic Research, (2011).
Mitchener, Kris James, and Ian W. McLean. "The productivity of US states since 1880." Journal of Economic Growth 8.1 (2003): 73-114.
Mokyr, Joel. Industrialization in the Low Countries, 1795-1850. Yale University Press, (1976).
Figure 1 - Resource Endowments in 1935
Figure 2 – National Employment: Resource Sectors
Figure 3 - Income per Capita Growth and Population Growth
Table 2a –Summary Statistics: Resource Endowments and Employment
Table 2b - Summary Statistics: Outcome Variables
Table 3- Effects of Natural Resources on Population Growth
Table 4- Effects of Natural Resources on Per Capita Income Growth
Figure A3 –Total Employment and Log(Wages)
Figure A4 – Employment: Non-Resource Sectors
Table A1 – Summary Statistics