ArticlePDF Available

Abstract and Figures

We quantify the sensitivity of investments to policy uncertainty by drawing upon the Northern Pacific’s massive land grant and the ensuing political battle that generated significant uncertainty to title from 1879 to 1894. Focusing on irrigation due to its high asset specificity, our analysis exploits the spatially exogenous extent of the grant to identify causal effects on investment, inclusive of spillovers to secure land because coordinated investment is generally necessary to capture the scale economies of irrigation. We find that the uncertainty significantly deterred and delayed irrigation investment in Montana, lowering the state’s economic activity by up to five percent.
Content may be subject to copyright.
1
Development Derailed: Policy Uncertainty and Coordinated Investment
June 2021
By:
Eric Alston – University of Colorado Boulder*
Steven M. Smith – Colorado School of Mines+
Abstract: We quantify the sensitivity of investments to policy uncertainty by drawing upon the
Northern Pacific’s massive land grant and the ensuing political battle that generated significant
uncertainty to title from 1879 to 1894. Focusing on irrigation due to its high asset specificity, our
analysis exploits the spatially exogenous extent of the grant to identify causal effects on
investment, inclusive of spillovers to secure land because coordinated investment is generally
necessary to capture the scale economies of irrigation. We find that the uncertainty significantly
deterred and delayed irrigation investment in Montana, lowering the state’s economic activity by
up to five percent.
JEL Codes: P48, D23, K11, N51, O13, Q15, H81
* Scholar in Residence, Finance Division, Leeds School of Business, University of Colorado Boulder.
eric.alston@colorado.edu
+ Assistant Professor, Division of Economics and Business, and Faculty Fellow, The Payne Institute for Public
Policy, Colorado School of Mines. ssmith1@mines.edu
For helpful comments and suggestions, we thank Doug Allen, Lee Alston, Jeremy Atack, Asaf Bernstein, Richard
Hornbeck, William Hubbard, Bryan Leonard, Gary Libecap, Paul Rhode, John Wallis, Zach Ward, an anonymous
reviewer and our editor, Dennis Carlton. We also thank participants at the following seminars/conferences: CU
Boulder Environmental and Resource Economics Workshop, Colorado School of Mines Economics Brown Bag, CU
Boulder Finance Division Brown Bag, World Congress of Environmental and Resource Economists, Society for
Institutional and Organizational Economics, Mountain West Economic History Conference, Arizona State
University, Southern Economic Association, Society for Empirical Legal Studies, and the PERC Lone Mountain
Workshop. For research support, Alston acknowledges funding from the Property and Environment Research Center
and the Hernando de Soto Capital Markets Program, and Smith acknowledges funding from USDA-AFRI Grant
67023-29421. We also thank John Stafford and Lindsay Vanlerberg for valuable research assistance. All errors are
our own.
2
“Large numbers of settlers are occupying such [railroad grant] lands, and it is important to them
to know whether they can receive their titles from the United States, or whether they will be
required to purchase from the railroad companies. The prevailing uncertainty necessarily retards
improvements and impairs values.”
- N.C. McFarland, General Land Office Commissioner,
Annual Report of the Commissioner, 1882, p. 11.
1. Introduction
Environments of policy or regulatory uncertainty are expected to depress investment,
especially those that are asset-specific and irreversible in nature (Rodrik 1991, Aizenman &
Marion 1993, Bohn & Deacon 2000, Ishii & Yan 2004, Blyth et al. 2007, Fuss et al. 2008, Bulan
et al. 2009, Fabrizio 2013, Caldera et al. 2020). With strong theoretical underpinnings, empirical
evidence is often consistent with the theory, finding negative “relationships” (Gulen & Ion 2016)
or “associations” (Baker et al. 2016) between uncertainty and investment, but often less than
causal. The challenge is that the regulatory and political processes that can create uncertainty are
seldom exogenous. Instead, they are spatially correlated with political borders (Handley &
Limão 2015) and captured by temporal shocks (Julio & Yook 2012) that are correlated with
numerous other factors affecting the entire polity or regulated industry. To better isolate the
causal effect and cost of uncertainty on investment, our study focuses on the Northern Pacific
(NP) land grant – a rare example of policy uncertainty affecting a spatial extent that was at once
massive and exogenously determined.
Construction of US railroads was heavily subsidized through extensive land grant
legislation from 1850-1871, which provided the companies about 158 million acres (Atack &
Passel 1994, at 436). The largest land grant was to the NP: 42 million checkerboarded acres
extending up to fifty miles on either side of the rail. But construction delays exposed the NP
grant to significant forfeiture risk of the lands they were granted (Ellis et al. 1946). For the NP
grant, although title by the railroad was ultimately retained, this result was subject to significant
uncertainty, most acutely from 1879 to 1894. During this time, the executive, legislative, and
judicial branches all contributed to policy uncertainty and failed to clarify what constituted a
valid claim to the land, eroding settlers’ confidence that they could appropriate the returns of
their investments in resource development.
3
We turn to frontier Montana to examine how insecure property rights generated by this
policy uncertainty influenced investment, specifically irrigation ditches. Montana was the most
extensively impacted, having ceded the most land of any state/territory to a railroad grant and
nearly 50 percent of the territory falling within the limits of the NP grant. We consider irrigation
investment because it exhibits high asset specificity and fixed costs, was central to economic
development on the frontier, and has an observable date of investment.
Because this instance of policy uncertainty is channeled through real property rights, we
are able to draw upon and contribute to the related literature on investment, economic
development and secure property rights. The link between the three has well-established
theoretical and empirical foundations (see Besley 1995, Galiani & Schargrodsky 2011, Lawry et
al. 2014). More recently, scholars have argued that just as title to property matters, so too does
the route to that title (Arruñada 2012, Brooks & Lutz 2016, Allen & Leonard 2021). Our analysis
here considers the potential negative consequences when an initial land allocation process fails
to deliver sufficient certainty to land rights.
Our historical context and set of empirical techniques allow us to causally identify the
immediate and persistent effects of policy uncertainty on asset specific investment. The
uncertainty arises due to a national forfeiture movement across the entire expanse of land grants,
ultimately determined by Congress, as opposed to local contestation over specific, more
valuable, land or the individual investor’s personal political ties. Additionally, while the NP
route itself was certainly endogenous to land quality and value, our main specification is a spatial
regression discontinuity at the grant’s far-flung borders, set by statute and running across the
land with no concern for political boundaries or topography. We find lands subject to the
uncertainty just within the land grant border were one-third less likely to be brought under
irrigation than similar, unencumbered areas. If land subject to uncertainty was irrigated, it was
over 3 years delayed relative to just beyond the grant. Our causal claim that the deterred and
delayed investment emanated from the policy uncertainty and not railroads and their land grants
per se is bolstered by the fact that the effects first emerge in short order after the uncertainty (15
years after the land grant itself) and similar effects are not present along analogous borders and
distances from other railroads never subject to uncertainty over land grants.
While we are the first, to our knowledge, to draw on the borders of the land grant for
identification, the exogenous variation embedded in the checkerboard pattern has also been used
4
to identify persistent consequences to land grant policies (Lewis 2019, Edwards et al. 2019,
Smith 2020). We therefore also leverage the exogenous checkerboarding throughout the entire
NP land grant to deploy a second empirical technique that (like our first test) identifies how
public policies that induce uncertainty of ownership, rather than eventual ownership patterns
themselves, impact investment in resource development. This difference-in-difference
specification reveals a similar negative effect on investment, but causality is weaker as the
interspersed, nominally unaffected sections appear to experience smaller but similar effects.
The result suggests that policy uncertainty can spillover on to economic actors not
directly impacted when coordinated investment is warranted, which is often the case with
irrigation ditches, particularly those extending significant distances from a stream due to scale
economies. To test this hypothesis, we provide a triple-difference specification interacting the
differences with the distance from the nearest stream. Consistent with the hypothesis, the lower
investment on the “certain” sections is absent near a stream where coordination is less necessary,
but grows and converges to that of the uncertain sections as longer and more expensive ditches
are required and coordination desirable. These spillover effects are germane to other instances
where development and use of natural resources entail coordination across space such as wind
farms (Lundquist et al. 2019), natural gas fracking (Leonard & Parker 2020), and fisheries
(Costello & Kaffine 2010).
Given our specific source of uncertainty, our findings also add to the perennial research
question as to the benefits of the railroads and the wisdom, impact, and need for associated land
grants (Ellis et al. 1946, Fogel 1964, Mercer 1982, Donaldson & Hornbeck 2016, Allen 2019).
In contrast to studies focusing on the effect on ownership patterns of the railroad land grants’
checkerboarding (Appleman 1939, Cotroneo 1976, Edwards et al. 2019, Lewis 2019, Smith
2020), we identify how public policies that induce uncertainty of ownership affect
contemporaneous investment in resource development. Given the exceptional scale of the NP
grant, the specific and large cost we identify should be a significant factor on its own when
evaluating land grant policy. In addition, although not all railroad land grants were afflicted by
forfeiture-related uncertainty, many were.
The effects we identify are also persistent, extending well beyond the resolution of the
uncertainty. As a final exercise, we construct counterfactuals to approximate the economic costs
of the lower investment levels using agricultural census data from 1920 and 1930. We estimate
5
that aggregate agriculture land value in Montana was between 5 and 21 percent lower due to the
depressed irrigation investment. The annualized reduction in production amounts somewhere
between one and five percent of Montana’s GDP around 1930.
The results are consequential because while exogenous variation that enables clear
identification of the costs of policy uncertainty is rare, the phenomenon is itself common, with
land policy being no exception (Skyner 2001, de Janvry et al. 2014, Gibb 2017). We also identify
spillover effects onto resources not directly exposed to uncertainty. In other words, insecurity to
a given resource can generate negative outcomes on adjacent and associated resources
themselves otherwise securely held. Our analysis demonstrates that when secure title is desired,
failure of public policy to provide a clear route to it causes significant losses to the economy
through derailing important but costly asset-specific investment.
2. Railroads, Land Grants and Uncertainty in Montana
Between 1850 and 1871, the government encouraged private investment in railroads
through land grant subsidies, providing the railroad title to a set number of either even-numbered
or, more often, odd-numbered sections (1 square mile or 640 acres) of public lands
1
lying on
either side of the railroad right-of-way (Decker 1960 at 83).
2
The result was a “checkerboard”,
with the government retaining ownership of the remaining sections to be patented – the initial
issuance of fee simple title from a sovereign government – under the active land laws (Greever
1951 at 83-84). The routes to patent were numerous, but cash sales and claims made under the
Homestead Act were the most common in addition to the railroad grants (see Hibbard 1924 and
Gates 1968).
3
In practice, this resulted in a set of distinct possibilities for individuals to obtain
title to the land within the boundaries of the railroad grant we consider here: (i) the railroad could
sell the odd-numbered sections, often precedent to securing patents for them, although this sale
was subject to forfeiture uncertainty we describe subsequently; (ii) the government could issue a
patent directly to an individual under the land laws for the odd-numbered sections conditionally
1
The Land Ordinance of 1785 set out the Public Land Survey System. From principal meridians, townships (6 by 6
miles) were established in 36 numbered sections, each 640 acres. These sections were further subdivided into halves
(320 acres), quarters (160 acres) and quarter-quarters (40 acres). See Online Appendix OB, Figure OB1 for an
illustration. A subset of these lands (typically sections 16 and 36) was reserved for “state purposes.”
2
A railroad land grant was interpreted to be a perpetual exclusive option to patent the land granted to the railroad in
a given land grant act, conditional on the railroad meeting the terms specified in the granting act.
3
Cash sales were direct purchases from the government. Homestead claims were “free” but required improvements
and 5 years residence to receive the patent from the government.
6
granted to the railroad, but this title could then be disputed if the railroad completed its grant
terms; or (iii) the government could likewise issue a patent under the land laws to an individual
for the even-numbered sections within the land grant, without being subject to uncertainty.
We draw upon the NP case, because, in the words of one historian, “[i]n the size of its
land grant, but also in its violations, controversies, investigations, and lawsuits, the Northern
Pacific had no peers.” (Daffran 1998).
4
Just months after Montana became a territory, the NP
was incorporated (Act of July 2, 1864, ch. 217, 13 Stat. 365) and was granted 42 million acres, as
much as the next three largest railroad land grants combined (Haney 1908).
5
This included 40
sections (25,600 acres) per completed mile in territories. An 1883 map of the grant in Montana is
shown in Figure 1 along with a stylized representation, including the first indemnity band (10
miles), created by the authors. “Indemnity bands” are lands beyond the primary grant from which
a railroad could select land to substitute for any land within their primary grant unavailable due
to prior settlement or mineral deposits.
6
The NP struggled mightily, including periods of stagnation and bankruptcy, but was
ultimately completed in 1883, 4 years after the statutory deadline (July 1st, 1879) set out in the
terms of the grant. The amount of land granted to the railroads to support development became a
major political controversy during this same time period, especially as railroads missed their
construction timelines, and contempt for the railroad barons grew amid corrupt behavior like the
Credit Mobilier scandal.
7
Fully cognizant of the implications for their right to the lands adjacent
to the railroad, the NP heavily lobbied members of Congress to grant the line a time extension
(Schwinden 1950), but the effort was ultimately unsuccessful, and so the NP had to rely on the
original grant terms even as the movement toward land grant forfeiture gained steam (Clinch
1965), with 1877-1890 the most active in terms of actual forfeitures in Congress (Ellis 1946).
4
The amount in Montana alone totaled 14,739,697 acres and stands as the largest land grant to the railroads of any
state/territory in the country (Gates 1968, 385). Inclusive of the first indemnity band (see footnote 6), the land grant
became interspersed with 26 million non-railroad acres fanning out 50 miles in both directions from the railroad,
encompassing nearly 50 percent of Montana.
5
See Online Appendix OB, Table OB1 for a summary of the NP timeline. See Alston & Smith (2021) for a robust
timeline.
6
The question of whether settlement could occur on odd sections in the indemnity band itself became a national
controversy due to the varying GLO policy toward this land (Julian 1883, at 251-252, Powers 1889). For use to
replace deficiencies within Montana only, the NP received a second indemnity band beyond the first of an additional
10 miles in the territory. This second band was not withdrawn and was generally less known and less controversial.
7
By one account, the scandal defrauded the government of $20 million (Wahlgren Summers 1993). Although the
most salient, the perception of abuses on the part of the railroads was widespread (Roberts 2011 at 126-128), and
were often directly related to the land grants themselves.
7
Figure 1: Map and Stylized Illustration of the Northern Pacific Land Grant in Montana
Although it did not ultimately come to pass, our identification strategy draws on the
looming and credible threat of forfeiture of NP lands in Montana throughout this period. Efforts
to force the forfeiture of the NP grant were relentless with House of Representatives voting to do
so in 1882, 1884, 1886, and 1888, but the Senate never agreeing (Fairweather 1919, Ellis 1946).
Because of this political uncertainty, settling within the land grant faced considerable risk. Lewis
Haney (1908) described the potential folly: a settler makes improvements on grant land with
assurances of the privilege to purchase first, only to find in the final location adjustment or
forfeiture, the land is restored to the public domain and that they then have no title (p. 30).
Settling under the Homestead Act or a cash-sale from the government was an equally
dubious proposition as the NP could ultimately argue the land had been reserved in the grant and
seek to confiscate improvements for their own gains or demand a hefty ransom from the settler.
Indeed, the General Land Office (GLO) commissioner noted that railroad companies
aggressively antagonized claimants to “compel settlers to purchase railroad waivers or
relinquishments of lands to which the companies had not and might never have any color of legal
right; to appropriate the products of coal and other valuable lands” (GLO 1885, pg. 29). With no
overarching policy set forth, “[t]he courts found themselves faced again and again by suits
brought against some settler or another by the NPRR, or initiated by a settler against the NPRR”
(Bechtold 1992, pg. 28). Under the headline “A Montana Question. Trouble growing out of the
8
Northern Pacific’s land grant”, The New York Times reported that “[t]he result was to unsettle
titles and tie up large and valuable properties” (January 12, 1892).
This period of uncertainty by no means brought settlement and investment in Montana to
a full halt. To the contrary, patent, railroad sales, and agriculture and irrigation data suggest that
this period saw robust activity, even within the land grant boundaries. For instance, the NP
employed heavy marketing to sell their lands throughout this period (see Online Appendix OB,
Figure OB2), and conducted many land sales long before their grant was secured through
patents, shown in Panel A of Figure 2.
8
Besides purchasing from the railroad, homesteads and
cash-sales (purchased from the government) were the most frequent means to a patent in
Montana (see Online Appendix OB, Table OB2) according to GLO patent data (BLM 2016).
Panel B of Figure 2 shows that most patents of these types during this period of uncertainty were
within the primary grant and first indemnity band, 25 percent of which were on odd sections that
were at least nominally reserved for the railroad. Illustrative of the confusion and uncertainty,
this indicates that the GLO did not fully restrict access to these odd sections throughout this
period, directly exposing them to potentially conflicting claims.
9
Figure 2: Land patent and sale activity
8
The NP does not receive a single patent in Montana until 1894, before receiving a significant number in 1896. We
caution that the sales data is for the entirety of the NP, but according to GLO land records, the only land patented
under the NP Land Grant Statute prior to the 1890s was 3000 acres in Washington (1880).
9
Explanation for the Government’s issuance of patents to odd-numbered sections within the railroad land grant is
elusive. But uncertainty as to the location of parcels themselves (not surveyed when settled), or simple error amid
the uncertainty or even deliberate contravention of the terms of the granting act are plausible explanations.
Panel A
Panel B
9
Additional evidence, however, reminds us that securing a patent is endogenous and
neither necessary nor sufficient for investment and development. Reported in Table 1, farm
acreage and patents more than quadrupled from 1880 to 1890 before quadrupling again by 1900,
showing tremendous growth. But notably, even under the generous assumption that only
farmland was patented, 63 percent of farm acres in Montana were not patented in 1890. Even as
late as 1919, over 50 percent of the irrigated quarter-quarters (40 acres) have are not matched to
a patent in the GLO records (see Online Appendix OB, Table OB3). Although the irrigation rate
on patented land was twice that of non-patented, the lion share (90 percent) was patented after
irrigation.
10
Table 1: Montana Farm and Patent Acres
The pattern of patents following irrigation furthers the notion that the choice is
endogenous and that irrigation rates are higher on patented land not only because of the
increased security, but that irrigated land is also more likely to be patented to secure the
investment. In fact, the evidence suggests that many of the patents on odd sections within the
land grant are homestead claims of irrigated land, consistent with the position settlers would
have been in if they had already developed the land on an odd section within the NP grant at the
time of the onset of uncertainty and were seeking to protect their investment.
11
However, even
these settlers with patents on odd sections may not yet have felt secure in their title given the
legal dubiousness surrounding its issuance: in 1885 the Commissioner of the GLO lamented that
a patent “from the United States to a settler under an award by adjudication of this department is
not security to his rights against a railroad company” (GLO 1885, at 45). The upshot is that many
10
While the 5-year residence and improvement requirement of homesteads could explain a portion of this, only 8
percent of homesteads were irrigated and 10 percent of those established irrigation within the 5 years prior to patent.
11
The odd sections in the land grant that were homesteaded were disproportionately on parcels already irrigated.
Having made improvements, seeking title more aggressively to mitigate the emerging uncertainty through the
cheapest means permitted fits with standard theory.
Year 1870 1880 1890 1900 1910
Patented Acres 3,458 172,627 721,680 2,907,409 6,164,323
Farm Acres . 405,682 1,964,196 8,291,077 13,545,602
Percent Patented . 42.55% 36.74% 35.07% 45.51%
10
settlers arriving to Montana during the 19th century lacked a secure title but this did not
necessarily restrict investment because the land laws and first-possession ultimately tended to
yield a safe path to title in the West for the individual claimant. However, policy uncertainty
surrounding the land grant made the process to title on that land ambiguous. It is this uncertainty
that we hypothesize did reduce investment.
Two additional points of history bear examination in light of our empirical approach.
First, although the NP was the first railroad to cross Montana, it was soon followed by another
major railroad project, the Great Northern (see Figure 3 and Online Appendix OB, Figure OB3
for maps), which did not receive any land grants in Montana. This makes the area surrounding
the GN a critical counterfactual by which to consider the rate of land development that a railroad
could create.
12
Second, we also draw on the Kansas Pacific land grant in Colorado (see Online
Appendix OB, Figure OB4 for a map). Completed just four years after its incorporation in 1866,
the Kansas Pacific received its 20 sections per mile with little doubt. We draw on this land grant
and Colorado to distinguish the effect of land grants and checkerboarding on irrigation more
generally as compared to the uncertainty induced by the NP land grant.
3. Property Rights and Irrigation
The expectation that policy uncertainty over a path to secure property would adversely
impact investment and development – specifically irrigation – is well-rooted in property rights
theory. Investors will strengthen rights – whether by bearing the costs of obtaining legal title,
self-protecting, or even investing in an asset directly to bolster a de facto claim – when the
benefits outweigh the costs (Demsetz 1967). But the benefits themselves are often the product of
greater tenure security: investors are more willing to invest in ways that enhance economic
productivity when tenure security is strengthened (Besley 1995, Bohn & Deacon 2000, Brasselle
et al. 2002, Johnson et al. 2002, Lanjouw & Levy 2002, Goldstein & Udry 2008, Galiani &
Schargrodsky 2010, Liscow 2013). Where our empirical context provides additional value to the
extant understanding of the effects of property institutions surrounds how: (i) we identify the
effect of title insecurity on the choice to make an initial investment, as opposed to marginal
effects of security to title on ongoing investment; and (ii) our treatment of insecurity is not
12
Earlier economic historians identified this possibility, with one noting the Great Northern’s “construction invites
comparison...with the well-endowed Northern Pacific” (Rae 1952 at 140).
11
contingent on the social status of the land user in that individual settlers were effectively
homogenous in their inability to influence tenure security.
13
The importance of the process itself of land allocation has recently received more
attention in the literature (Arruñada 2012, Brooks & Lutz 2016, Allen & Leonard 2021), a
narrative to which we also contribute here. We are able to isolate a causal effect of insecure
property rights created by the public allocation process itself because: (1) the NP land grant
provides exogenous variation in secure ownership across similarly endowed units of land; and
(2), in contrast to informal systems of tenure where fixed investments can be made to strengthen
claims, the incentive to do so is not present in contexts, like ours, where the determination of
formal title will carry the day (Sjaastad & Bromley 1997).
14
Our analysis relies on irrigation to capture investment and development rather than land
patents for several reasons. First, patents are a measure of secure title, endogenous to the
settlement and investment process. Second, patent dates do not always align with settlement and
investment (see table 1 above): besides the GLO being notoriously short on resources and its
surveying and claim processing efforts lagging occupation (for example GLO 1885, at 221; GLO
1895, at 363), homesteads were entered at least 5 years prior, and often earlier, while cash-sales
could remain undeveloped today. Third, unlike the price of the land itself, which we do not
observe and could capitalize in uncertainty as documented elsewhere (Grainger & Costello
2014), the input costs to establish an irrigation ditch would not readily adjust to reflect
expropriation risk. Finally, because water diversions are highly idiosyncratic investments – their
locations are specific and they have a singular use – certainty to title is critical given that the
construction costs are sunk (Williamson 1979, Bretsen & Hill 2007).
Asset specificity interacts in important ways with economic incentives and therefore
influences contractual forms (Williamson 1983), as well as investment choices directly (Coles &
Hesterly 1998). Empirical work has also shown that asset specific investments are sensitive to
various sources of uncertainty (Aizenman & Marion 1993, Bulan et al. 2009, Handley & Limão
2015, Gulen & Ion 2016), including irrigation investment in particular (McClintock 2009).
13
Our study is thus closer to Bandiera (2007) and Jacoby et al. (2002) in that the variation in tenure security we
identify is a general feature of the political context in a given geographic area.
14
This is not to diminish the role of the initial private local systems that arose on the frontier and helped to shape
public mineral and water rights (see Umbeck 1977, Libecap 1978, Clay & Wright 2005, Alston & Stafford 2018)
and grazing claims (Alston, Harris, & Mueller 2012).
12
While other components of farming involve sunk investments also sensitive to uncertainty, we
focus on irrigation because it is of considerable expense – accounting for over 40 percent of the
costs to first bring an acre under irrigated cultivation in the US west in 1890 (Newell 1894, at 8)
– and data on its precise year and location are available in contrast to farmland data only
available at the county-decade scale.
Water use in the arid west is governed by the prior appropriation doctrine (see Getches
2009). Under this system, quantified water rights are associated with an appropriation date that
provides security of the right against subsequent (upstream) diversions by “junior” irrigators,
which creates strong incentives for water users to formalize their appropriation date as soon as
feasible. As importantly, western legal doctrines evolved to attach prior appropriative water
rights to productive uses on a specific piece of land. While attention has been given to the
rationale and impact of secure water rights in the West (Burness & Quirk 1979; Leonard &
Libecap 2019; Smith 2021), little formal empirical attention has been paid to the need for
accompanying land rights to induce investment.
15
This suggests a significant potential for
spillover effects between property rights to different natural resources; if viable agricultural
production requires both land and water rights, then the absence of security to one is likely to
deter investment in the other.
Of additional relevance to our empirical analysis, the size of individual farms on the
frontier was typically smaller than the size of organizations needed to fund and operate irrigation
ditch projects, requiring significant levels of coordinated effort as outside capital was rarely
present prior to the involvement of Federal Bureau of Reclamation and advent of the irrigation
district (Teele 1904, Coman 1911, Bretsen & Hill 2007, Leonard & Libecap 2019). In total, the
supply of irrigation projects (being asset specific, large-scale, and costly) was likely more elastic
to property rights defects than individual economic activities along the frontier. Importantly, this
means there are potentially negative spillovers for investment on parcels that are themselves
clear of uncertainty due to the uncertainty prevailing nearby.
15
A contemporaneous public report dealing with the disposition of state-owned lands in the West notes explicitly
how with respect to lessees of state land “the cost of such water rights…is becoming so great that lessees hesitate
and refuse to purchase them for land they do not own,” providing contemporaneous historical evidence of this
phenomenon in practice (Colorado State Board of Land Commissioners 1904, 6).
13
4. Data and Empirical Overview
Our analyses utilize disaggregated spatial data with observations at the quarter-quarter
PLSS unit (40 acres), which are delineated by GIS files obtained from the BLM and Montana
State Government.
16
To relate these units to measures of settlement and development, we have
linked them to historic GLO patent records using PLSS identifiers, and to irrigation records
using state-maintained GIS shapefiles. Both Montana and Colorado maintain GIS shapefiles
detailing what land is irrigated by which water rights.
17
The irrigated lands are linked to the
water rights that serve the land, allowing us to measure the priority date. This provides a better
proxy for settlement than the patent date because priority dates are based on when water is first
diverted – signifying actual investment and development – and, since both states adopted the
prior appropriation doctrine, irrigators were incentivized to ensure the date was correct because
this determined their water availability. Colorado only had railroad land grants in the South
Platte basin (Water Division 1), and we limit the data to this region accordingly. A full
description of the data is provided in Online Appendix OA.
Variables are summarized for Montana and Colorado in Table 2. We have removed
sections reserved for the State (numbered 16 and 36) as well as those that remained in Native
American Reservations as of 1915, as the choice to irrigate on these lands follow distinct
processes.
18
Furthermore, in both states, the measures for the year are conditional on patenting or
irrigation respectively, making those samples smaller. Across the development outcomes,
Colorado settled earlier and developed more irrigation on average. On average, Colorado patents
16
Most homesteads were 160 acres and the sections received by the railroad were typically 640 acres. We opt for the
40-acre unit of observation because claims this small appear in the patent data and agriculture/irrigation decisions
are often made on 40-acre fields. For instance, in Montana over 60 percent of the quarters (160 acres) that irrigate do
so on 40 acres or less. This does not imply scale economies are irrelevant, as irrigating only 40 acres may make
sense for farm production, but irrigation infrastructure may still require a larger scale. On average, in 1919, an
irrigation ditch in Montana served 307 acres across twelve different 40-acre units. Still, standard errors in the
analyses are adjusted for larger spatial patterns. We also conduct robustness checks at 160- (quarters) and 640-acre
(sections) units.
17
These capture whether the land is capable of being irrigated. At any particular time, fewer acres will actually be
irrigated given farm level cropping decisions as well as total water availability to the ditches.
18
Besides the decision process differing, irrigation on reservations often carries the priority date of the reservation
formation due to the Winters v. United States (207 U.S. 564, 1908) decision. For instance, the irrigated acres from
the Flathead Reservation carry a priority date of 1855 even though it began construction in 1908 and continued
through 1960 (Voggessor 2001). Our results are not sensitive to this decision nor does the inclusion of State reserved
sections affect our estimates. In addition, the samples are limited to sections within the 95th percentile of distance
from any stream in Montana (4 miles) to preclude areas that were very unlikely to be irrigated. In 1910, the average
ditch length was just 1.4 miles, and was seldom perpendicular to a stream (Edwards & Smith 2018).
14
are 13 years earlier, while Colorado water rights are nearly 50 years earlier on average. While we
contend the lack of certainty over land ownership in Montana may have created this relative
delay, our empirical strategy is considerably more conservative and focuses primarily on
irrigation outcomes as they were in 1919.
19
Even with this restriction, Colorado still had two
percentage points more of its land under irrigation and tended to do so 24 years earlier. It is
notable that from 1859-1919, the average water right year in Montana is 1896, just after the first
railroad patents were issued.
Table 2
We gather a number of other variables that may influence the proclivity to invest in
irrigation infrastructure. These are included in Table 2 as well. For this analysis, however, the
most important are the variables capturing a land unit’s proximity to railroads and land grants.
For every observation, we calculate the distance to the nearest railroad based on 1911 rail
networks (Atack 2016) and the distance from the main stems of the NP, GN, and KP specifically
to determine whether the observation falls within the land grant buffer (or a placebo buffer in the
19
We use 1919 as our main analysis for two reasons. First, in 1919, the NP had received 90 percent of the grants it
ultimately receives in Montana, indicating some certainty that the railroad had received the land (see figure OB5,
Online Appendix OB). Second, 1919 is the priority year for the Milk River project, a large Bureau of Reclamation
project that brought many acres under irrigation primarily beyond the NP land grant, potentially skewing the results
for independent reasons. Still, we conduct robustness checks using the decadal years (1870, 1880, …, 2010) as cut-
offs, providing a placebo test for before the uncertainty emerges and confirming the 1919 cut off is not driving the
results.
Difference
Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max
Irrigation Development
Irrigated 1,968,465 0.17 0.38 0 1 407,748 0.11 0.31 0 1 0.07
Irrigated by 1919 1,968,465 0.074 0.262 0 1 407,748 0.092 0.289 0 1 -0.02
Water Right Year 308,869 1925.93 33.82 1859 2017 39,633 1876.08 25.08 1859 1979 49.84
Water Right Year by 1919 146,501 1897.25 12.69 1859 1918 37,513 1871.09 14.02 1859 1918 26.16
Patent Data
Railroad Owner 1,968,465 0.04 0.20 0 1 407,748 0.03 0.18 0 1 0.01
Patented 1,968,465 0.32 0.47 0 1 407,748 0.33 0.47 0 1 -0.01
Patent Year 631,188 1916.54 15.84 1776 1999 133,989 1904.97 19.09 1800 2013 11.58
Controls
Land Grant Buffer (Primary) 1,968,465 0.41 0.49 0 1 407,748 0.40 0.49 0 1 0.02
Land Grant Buffer (Indemnity (1)) 1,968,465 0.48 0.50 0 1 407,748 0.60 0.49 0 1 -0.12
Distance to Railroad (miles) 1,968,465 17.47 15.22 0 72.19 407,748 7.62 6.79 0 33.15 9.85
Strahler Stream Order (nearest) 1,968,465 1.63 1.14 1 6 407,748 1.49 0.89 1 5 0.13
Distance to Stream (miles) 1,968,465 1.04 0.89 0 3.89 407,748 1.12 0.95 0 3.89 -0.08
Soil Quality 1,968,465 5.30 1.69 0 8.01 407,748 4.91 1.61 0 7.01 0.39
Elevation (mean) 1,968,465 1214.61 484.28 573 3606 407,748 2022.57 703.36 1031 4267 -807.96
Elevation (st. dev.) 1,968,465 26.71 38.51 0 404.58 407,748 26.90 40.71 0 377.35 -0.19
Second Division Summary Statistics
Montana
Colorado
Note: Descriptive Statistics for 40 acre PLSS units in both Montana and Colorado. Colorado is limited to what is now Water Division 1 (North-East
corner) due to limited irrigation data and railroad grant location. See Appendix A for a full description of the variables.
15
case of the GN).
20
For the NP we take this to be inclusive of the first indemnity band (or within
50 miles of the main line) since that land was also subject to uncertainty and the NP ultimately
made many claims to it (see Online Appendix OB, Figure OB6).
21
In contrast, in Colorado
indemnity lands were not available for the Kansas Pacific to patent. Accordingly, we take “land
grant buffer” in Colorado to be the 20 miles on either side of the rail line that made up the
primary land grant. In subsequent analysis, we explore distinctions between the primary and
indemnity lands within Montana.
In addition to latitude and longitude, we also gather information on the nearest natural
surface water source, including the distance and size.
22
Soil quality provides a measure of the
soil’s suitability to agriculture, scaled from 1 to 8, with lower numbers being better. For
elevation, calculated in meters, both the average and standard deviation (giving a measure of
topography) are calculated at the section level (640 acres).
One difficulty of identifying the effect of land grant uncertainty on investment is that
railroads, which provide greater access to markets and are built near desirable land, are
necessarily collocated. To overcome this, we employ two distinct empirical tests drawing on the
statutorily-determined, rather than attribute-determined, extent and pattern of the grants. Our
preferred analysis from a spatial regression discontinuity narrows in on land just within and just
beyond the statutory limits of the land grant to address selection issues. The analysis focuses on
the 50-mile distance for the NP where the advantage of the railroad transportation (or
alternatively, the irrigation along the nearby larger streams – Missouri and the Yellowstone)
dissipates. Still, smaller streams populate the landscape that appeal to the entrepreneurial
irrigator compelled to settle further afield due to the uncertain status of the land closer to the
railroad. From Figure 3, the intuition is to compare irrigation uptake on sections L and K to those
20
For land grant determination, ESRI (2016) railroad locations were utilized. Results are not sensitive to using the
1898 rail network, modern day networks, or exclusion of the rail distance as a covariate entirely.
21
A contemporaneous map also invokes the 50-mile extent: “[t]hese extremely productive lands stretch out for 50
miles on each side of the Northern Pacific Railroad.” (Rand McNally & Northern Pacific Railroad Co. 1883)
22
Size is imperfectly measured by Strahler order, which scores stream reaches (hydrologically homogenous portions
of a stream) from 1 to 7 (discretely) and increases as stream reaches come together. These locations are based on
modern sources (National Atlas of the United States, 2014) although irrigation may have altered flows, particularly
downstream of reservoirs (Shields Jr. et al. 2000). Typically, stream migration rates due to agriculture development
are small (Micheli et al. 2004) and not solely unidirectional (the stream migrates back and forth). The source data
and calculation rely on centroids, meaning wider and narrower portions due to reservoirs are not influential. The
upshot is that while modern stream locations are utilized, this likely induces measurement bias, and unless the effect
is systematically different within the land grant, no endogeneity bias is expected to emerge. Robustness checks show
the exclusion of the stream covariates is not influential on the main results.
16
on M and N. Some evidence of the effect can be seen in the detailed view in Figure 3, where
irrigated lands appear to stop near the edge of the land grant.
Figure 3: Empirical and stylized land grant overview
The spatial regression discontinuity estimates are well identified, but measure only the
local average effect for land some 50 miles removed from the NP. Meanwhile, the uncertainty
would impact the entire expanse of the land grant including closer to the railroad itself where, all
else equal, development was more attractive. Accordingly, we expand the sample and use the
exogenous variation created by odd and even sections to employ a difference-in-differences
analysis. This considers how the difference between odd (M) and even (N) sections in the land
grant compares to the difference beyond the land grant (L-K). Some suggestive evidence of this
effect is also seen in the detailed section of Figure 3.
However, because of the scale economies associated with irrigation projects, uncertainty
on the odd sections could spill over to investment decisions on even sections. This, in practice,
undercuts the identifying assumptions of the difference-in-differences approach. In fact, we find
a large portion of the effect is on the difference between inside and outside the buffer, regardless
of odd or even assignment. Although we contend the presence of the GN and our covariates
provide a credible counterfactual for this estimate, the fact remains that the location of the land
grant does not exhibit the same exogeneity as the odd/even sections. Accordingly, we augment
this analysis with a triple difference (DDD), allowing the effects to vary with the distance from
17
surface water. If the effect is driven by uncertainty, as the fixed costs of irrigation increase at
greater distances from a stream, so too should the difference in irrigation rates between land
inside and outside the land grant, but the distinction of that difference between odd and even
sections within the land grant should dissipate as coordinated investment becomes increasingly
desirable and the spillover effects emerge.
5. Spatial Regression Discontinuity
Figure 4 below shows that land nearer a transcontinental railroad, be it the NP or the GN,
is more attractive for irrigation investment in Montana. The advantage of the railroad line
seemingly dissipates beyond 10 miles or so.
23
The RD design allows for this selection but
assumes that it was driven by the land closest to the railroad, without direct regard for the fringes
of the land grant. In other words, because selection was on the main line and the extent of the
land grant was set by statute, we expect no systematic differences of land just within the grant
compared to land just beyond the grant.
Figure 4: Propensity to eventually irrigate in Montana.
In testing this assumption for covariates – stream size, distance to stream, soil suitability,
elevation, and elevation variation – we find no statistical distinction at the border except for
stream distance, which is 85 meters closer within the border (results in Online Appendix OB,
Table OB4). Additionally, we are unaware of any other policies that were spatially aligned with
23
Fogel (1966) in his seminal work suggested 40 miles was a reasonable limit beyond which the benefits of
transportation networks were considerably muted.
Panel A
Panel B
18
land grant borders except that even sections demanded a “double-minimum”, making them twice
as expensive. This policy only pertained to cash-sales and did not preclude homesteads, although
they were limited to 80-acre claims by law until 1879. However, the GLO did not uniformly
adhere to either the minimum cash price or the maximum homestead acreage in practice due to
the wide variety of legislative exceptions to these requirements as well as problems surrounding
the administration of the general and local land offices at the time (Conover 1923 at 26, Rae
1938, Gates 1954, Gates 1968 at 366).
24
Still, this could discourage settlement (and subsequently
irrigation) and we choose to focus our RD analysis on the first indemnity land border (50 miles)
where the uncertainty remained but the double minimum did not (Powers 1889, Rae 1938, 219-
226, Gates 1954).
Suggestive evidence that the railroad grants mattered is provided in Figure 5, which plots
the linear fit for the year of irrigation for units within 5 miles of either side of the first indemnity
band border in Montana. There appears a stark change near the border, and actual irrigation
investment occurs years earlier just beyond the railroad grant lands.
25
Figure 5: Spatial discontinuity
24
Specifics on cash sale prices are not available, but Gates (1954) concluded that on one grant, the public domain
sales recovered just $1.39 per acre. For the homestead acreage limit, we observe just 10 homestead patents of 80
acres in the land grant region of Montana prior to 1879, but 664 homestead patents of 160 acres, highlighting the
inconsistent application of the land laws in this region at the time.
25
As one moves beyond the uncertainty of the grant, the graph (panel A) indicates irrigation rates increase despite
being further from the NP. But in our historical context, along the northern edge of the land grant the parcels are
increasingly close to the alternative GN railroad. While all of these observations are over 50 miles from the NP, the
median distance to the GN for the observations is just 31 miles, 25 percent are within 16 miles, and some are within
a 1 mile of the GN, making development and irrigation more appealing.
Panel A
Panel B
19
To formally test for the effect of the uncertainty, we estimate parameters of the following
equation:
!"#$ % &' ()*+$,-./0123$4-()*+$,'5/0123$4-6789
(1)
As our main outcomes, we consider whether the unit is irrigated by 1919 and, if so, when.
()*+$,
is an indicator function equal to one for units within the NP land grant and the coefficient
of interest is
&
, capturing the discontinuous effect at the edge of the grant. Our main specification
is a local linear estimator. In practice, we implement the rdrobust command in Stata (Calonico et
al. 2017) that selects an optimal bandwidth, trading off between bias and variance. In a series of
robustness checks we try alternative polynomials, bandwidths, and inclusion of covariates, most
notably controls for distinct places along the borders.
26
Table 3
Results, considering whether the land is irrigated, and if so, when, are presented in Table
3 for each land grant limit (40, 50, and 60 miles) along the NP. Just within the primary grant
buffer (40 miles), the share of parcels irrigated by 1919 decreases by 0.0090. The share of land
irrigated within the bandwidth of the land grant border by 1919 is only around 0.06. This means
that parcels just outside the primary lands were 15 percent more likely to be irrigated. The results
also suggest irrigation was delayed by one and half years. At the edge of the first indemnity band
26
Some robustness checks employ the linear OLS approximation (
!"#$ % : '
;
0123<=
;
$->'
;
0123<=
;
$?
@AB210C<=$-&'AB210C<=$-DE 'FG-HE ' IJ-6789@
) with observations limited to those within the optimal
bandwidth.
(1) (2) (3) (4) (5) (6)
Discontinuity (within) -0.00899** -0.00788*** 0.00200 1.400*** 3.120*** 2.380***
(0.00407) (0.00263) (0.00337) (0.401) (0.554) (0.701)
Observations 1,338,179 1,338,179 1,338,179 106,679 106,679 106,679
Bandwidth 2.867 5.901 4.144 6.567 5.559 2.971
Primary Indemnity 1 Indemnity 2 Primary Indemnity 1 Indemnity 2
40 Miles 50 Miles 60 Miles 40 Miles 50 Miles 60 Miles
*** p<0.01, ** p<0.05, * p<0.1
Spatial RD Estimates for the Northern Pacific Land Grant
Irrigated (=1)
Priority Year
Distance from Northern Pacific
Note: Results are for the change in the dependent variable (measured in 1919) at 40, 50, and 60 miles away from the
Northern Pacific Railroad. Columns (1)-(3) use an indicator variable for irrigation as the dependent variable. Columns
(4)-(6) use priority year (for those that are irrigated) as the dependent variable. Estimates come from the rdrobust
command in Stata. Observations include only PLSS sections within 80 miles of the Northern Pacific in Montana.
Standard errors in parentheses
20
(50 miles), the parcels just beyond are more likely to be irrigated and brought under irrigation
around 3.1 years earlier than parcels within the first indemnity band. Across Montana during this
period, the standard deviation of priority year is just 15 years, so this delay amounts to 29
percent of one standard deviation. Finally, at the edge of the second indemnity band (60 miles),
there is no distinction in the rate of irrigation, but just beyond the totality of the NP’s possible
claim, irrigation does occur 2.5 years earlier. On net, each discontinuity suggests irrigators
preferred more certainty, choosing to irrigate more and/or earlier just beyond the boundaries,
even when the land and market access were similar. The strongest evidence of delayed and
deterred investment is at the edge of the first indemnity band, and we focus our numerous
robustness checks on this boundary.
Figure 6: RD estimates by decade
First, we relax the presumption that 1919 is the proper snapshot in time to assess the
impact of the NP land grant on irrigation and conduct separate regressions for each decadal year
from 1870 through 2010. The estimated discontinuity coefficients are plotted in Figure 6.
27
The
first thing to note is that the area just inside the land grant buffer was not less likely to be
irrigated in 1870 and 1880. In 1870, few doubts over the land grant were present; the NP was
under construction with the future bankruptcy and subsequent failure to meet the statutorily set
27
Table OB5, Online Appendix OB provides the estimates and standard errors.
21
timeline not yet known or anticipated. After missing the 1879 deadline, doubts may have begun
to creep in by 1880, even though no major railroad forfeiture had yet occurred. But between
1880 and 1890, amid the numerous attempted forfeiture acts, irrigation rates declined sharply
inside the grant relative to just beyond. The time at which the effect emerges is consistent with
uncertainty providing the mechanism, and not the land grant itself, which existed throughout the
1870s. Second, we see that our choice of 1919 does not drive the results, and the effect persisted
even once the uncertainty subsided. We take this persistence to be illustrative of how property
rights to complementary inputs can interact with one another: once the land rights were clarified,
the prior development of water rights nearby left less secure junior water rights, continuing to
dampen the incentive to irrigate. Furthermore, the pre-existing development could further
leverage scale economies, reducing the marginal cost of expanding irrigated acreage off of a
main ditch.
While the timing of the effect emerging helps, to further bolster our claims that the
effects stem from uncertainty and not other factors, we conduct a series of placebo tests for
distances from the NP and other railroads where there is no change in legal certainty (estimates
provided in Table OB6 of Online Appendix OB). First, we consider an arbitrary distance (20
miles) from the NP that is irrelevant to the land grant. Second, we test for any effects 50 miles
from the GN to ensure there is nothing special about being 50 miles away from a transcontinental
railroad in Montana. Finally, we turn to the Colorado data and assess if there is reason to believe
that the edge of land grants induces this variation in irrigation, independent of any legal
uncertainty surrounding the grants. The estimates are all statistically insignificant except for a
one-year delay in irrigation at 20 miles from the NP.
A battery of additional robustness checks at the 50-mile discontinuity is provided in
Online Appendix OB, Tables OB4, OB7-OB14. We test the robustness of the main RD
specification by: including covariates, utilizing different bandwidths (1, 5, 10, 15, and 20 miles),
and deploying various local polynomial orders (2, 3, and 4); leaving out those within a mile of
the border, aggregating observations to quarters (160 acres) or entire sections (640 acres);
allowing for spatial autocorrelation in errors across township-scale neighborhoods; conditioning
the sample on patent status; and using a hazard model-based estimate. Finally, because Montana
is vast and the NP land grant within extends some 500 miles east to west while its distinct north
and south first-indemnity-band borders lay 100 miles apart, we also test for differential effects
22
along the distinct borders and, drawing inspiration from Mian et al. (2015), include indicator
variables for each 10-mile segment east to west to account for distinctions along the long spatial
RD borders. Results are similar across all specifications.
6. Difference-in-Differences
Although the effect of the uncertainty is most cleanly identified near the border of the
land grant, we nonetheless expand our scope to approximate the impact uncertainty had across
the entirety of the land grant through exploring the dynamics of the spillover effect on even
sections. Because the uncertainty applied directly to the checkerboarded odd sections only, we
begin with a difference-in-differences approach and consider how the extent and timing of
irrigation on odd sections within railroad land grants compare to even sections generally and
within the railroad land grants. The regression equation is:
!"#$ % : '()*+$,- > ' ()K00#, - & ' ()*+$,' ()K00#,- DE ' FGL -HE ' IGLJ -6789
(2)
Outcomes for 40 acre-units (u) in section s of township f considered are again whether
the land is irrigated by 1919 and, if so, the year of the first associated water right.
28
The two
coefficients of interest are
:
, which estimates the impact of being within the NP land grant buffer
distance, and
&
, which is the estimated differential impact on an odd section within the NP land
grant buffer. The underlying assumption is that the interspersed odd sections would have been
equally likely to be irrigated as the even sections absent the legal uncertainty induced by the land
grant. The expected value of
>
is zero, since an odd section outside of the railroad land grants
should not be expected to be systematically different from even sections.
29
In addition to the variables of interest, we also include section (
FGL
) and quarter-quarter
unit (
IGLJ
) covariates. Most notably, we directly control for distance (as a second-order
polynomial) from the nearest railroad using the 1911 network. This helps to address the market
access impact of being near a rail line (Donaldson & Hornbeck 2016) that is generally correlated
with railroad land grant locations. In Montana, where the entire state is in the sample, and the
GN also traverses the state but did not receive land grants, identification of the rail access effect
28
Because railroads themselves, independent of the land issues, provide numerous benefits of market access (both
for inputs and outputs), we account for the later arrival of the Great Northern (1893) by adjusting the priority year to
be relative to the completion year of the nearer of the two transcontinental railroads.
29
Generally, rates could differ on even sections owing to the reservation of those sections for the state. These,
however, have been removed from our sample.
23
distinct from the land grant impact is garnered. Furthermore, we again draw on Colorado as a
placebo test as to whether the railroad land grant itself (and the double-minimum on cash-sales)
reduces irrigation investment. Other controls are latitude, longitude, distance to and size of the
nearest stream (with the second-order polynomial of distance interacted with stream size to allow
for non-linearities), and soil quality. We also include the average and standard deviation of
elevation measured at the section-level.
30
Finally, because of the economies of scale of irrigation projects, it is feasible that the
uncertainty spills over onto the even sections. In fact, our results below indicate that this is the
case, and the first-difference is large. To help substantiate the validity of this first-difference,
which lacks the exogeneity of the odd and even assignments, we also conduct a semi-parametric
triple difference (DDD) with the distance from the nearest stream as follows:
!"#$ %
M
()N,'/OP-:P'()*+$,
P->P' ()K00#, - &P'()*+$,'()K00#,4
-DE'FGL -HE ' IGLJ - 6789
(3)
()N,
is an indicator function for discrete distances from the nearest stream. We utilize
quarter-mile bins in practice. If the estimated impact we find across the entire land grant is due to
selection of location, and not the uncertainty, then the vector of coefficients on the land grant
(
:P
) and effect of odd sections within (
&P
) should not vary. Instead if the even sections are being
influenced by the uncertainty of odd-sectioned neighbors, then we expect the effect to correlate
with the distance to the stream, with areas further away exhibiting greater economies of scale due
to the greater necessary fixed investment to reach the land. In other words, we expect the
:P
coefficients to begin small and increase with distance, while the
&P
coefficients will decline with
distance from the stream.
The main DiD regression results are provided in Table 4, where we also provide the same
estimates for the land grant in Colorado to provide some contrast. Looking at column (1), land
within the land grant buffer in Montana was 1.9 percentage points less likely to be brought under
irrigation by 1919. In contrast, Colorado land within its land grant buffer (Column (2)) was much
30
We do not condition on patent status, timing, or type because they do not represent settlement and are endogenous
choices to strengthen property rights influenced by improvements (irrigation) already made and the presence of the
additional uncertainty within the land grant. We nonetheless provide robustness checks in Online Appendix OB,
Table OB15. The evidence shows that the land patented by 1919 also had lower levels of irrigation amid the
uncertainty.
24
more likely to be irrigated (5.3 percentage points). Given that only 7.4 percent of land in
Montana was irrigated in 1919, the reduction of 1.9 percentage points is quite large, about 25
percent less likely. This effect is inclusive of the even sections, but the odd sections within the
land grant buffer in Montana (Column (1)) were even less likely to be irrigated by an additional
0.5 percentage points, while there was no statistical distinction for odd sections in Colorado.
Together, this suggests that even sections in the NP grants were dissuaded from investing in
irrigation, and if they did invest, they often would do so without the neighboring odd sections,
suggesting unrealized economies of scale and increased capital costs.
Table 4
Turning to the priority year, marking the date of investment, water rights tend to be 3.7
years later within the land grant buffer in Montana than the land outside of it. In contrast,
Colorado grants were irrigated earlier than the surrounding land. Likely related to the economies
of scale, conditional on being irrigated, there is no evidence of differential timing for odd
sections within the grants in either state. In other words, irrigation projects were not well-suited
for checkerboard ownership and relied on ownership across the grid being settled, so the
uncertainty on the odd sections was sufficient to also delay investment on the neighboring even
sections. Finally, the hazard model, which considers both the likelihood of irrigation and when it
(1) (2) (3) (4) (5) (6)
Land Grant Buffer -0.0190*** 0.0528*** 3.700*** -1.955*** -0.124*** 0.502***
(0.00180) (0.00432) (0.304) (0.670) (0.0246) (0.0564)
Odd Section 0.00112 -0.000321 0.199 0.0170 0.0180 0.000190
(0.00151) (0.00267) (0.278) (0.732) (0.0254) (0.0629)
Land Grant Buffer x Odd Section -0.00498** 0.00206 -0.567 0.00152 -0.0697** 0.0155
(0.00227) (0.00613) (0.386) (0.799) (0.0338) (0.0765)
Distance to Rail -0.00118*** -0.0256*** 0.369*** 1.226*** -0.0145*** -0.400***
(0.000130) (0.000606) (0.0205) (0.183) (0.00191) (0.0191)
Distance to Rail21.33e-05*** 0.000765*** -0.00439*** 0.0582*** 9.64e-05*** 0.00467**
(2.15e-06) (2.03e-05) (0.000424) (0.0148) (3.61e-05) (0.00208)
State MT CO MT CO MT CO
Model OLS OLS OLS OLS Hazard Hazard
Observations 1,968,465 407,748 146,501 37,513 1,968,465 407,748
R-squared 0.078 0.306 0.116 0.590
*** p<0.01, ** p<0.05, * p<0.1
Difference-in-Differences Estimates, 1919 Irrigation Outcomes
Irrigated (=1)
Priority Year
Failure=Irrigate
Note: Regression Results for PLSS Second Division Land Units (40 Acres). Land Grant Buffer indicates that the unit is within the
designated distance from the railroad to be in the land grant area (50 miles for Montana, 20 miles for Colorado). Odd Section indicates that
the unit is contained in an odd section. Railroad distance is in miles from the nearest railroad line (not necessarily the Northern Pacific or
Kansas Pacific). Additional unreported controls include latitude and longitude, distance (and distance squared) to and size of nearest stream,
soil quality, and the median and standard deviation of the elevation for the section. Robust standard errors, clustered by section, in
parentheses.
25
occurs, provides similar evidence. In Montana, odd sections in the land grant were around 20
percent less likely to be brought under irrigation in any given year relative to land beyond, while
even sections within the grant were still 12 percent less likely to be irrigated.
Like the spatial RD, these results are robust to modeling choices. In Table OB16 of
Online Appendix OB, we show that the results for Montana are robust to the selection of
covariates, exclusion of “small” irrigation, and using a logit (for irrigated) or tobit (fraction
irrigated). The scale of analysis also is not particularly important. In Table OB17 of Online
Appendix OB, we provide analysis of the data aggregated to quarters (160 acres) and entire
sections (640 acres). Furthermore, we show that clustering the standard errors at the township
level, using spatial errors across a similar scale, or by water right (for priority year only) does not
alter the statistical significance.
31
In addition, the timing of these effects is similar to the RD
estimates, emerging only after 1880 when the uncertainty emerged (see Figure OB7 and Table
OB20 in Online Appendix OB).
32
However, throughout the difference-in-differences analysis, the negative effect is
dominantly on the entire land grant, not just the odd sections. One explanation is the economies
of scale for irrigation projects, which meant the uncertainty over the odd sections bled over to the
adjacent even sections, holding up or precluding irrigation development there as well. However,
this fact also means our identifying assumption for this analysis is weaker than supposed, which
is a recognized issue with checkerboard identification strategies when neighboring parcels
influence productivity decisions (Leonard & Parker 2020). An alternative explanation is that
uncertainty prevailed because settlers may have been aware of their distance from the NP, but
absent survey, clueless as to their land’s position in the grid.
33
This is to say the lack of survey
31
We also test for a distinct effect across the primary lands and indemnity lands in Montana by including an
additional indicator for whether the land is within 40 miles of the NP, as well as controlling for this indicator’s
interaction with odd sections (See Online Appendix OB, Table OB18). Across the indemnity and primary lands,
there is no statistical distinction for the amount of the land units brought under irrigation or the hazard through time
of being brought under irrigation. The primary land, however, is slightly more delayed in time for those areas that
are irrigated. We also explicitly control for being within 50 miles of the Great Northern in another robustness check
(Online Appendix OB, Table OB19), finding the odds of irrigating are greater in this band, suggesting railroads,
absent tenure security issues, are indeed attractive.
32
In contrast to the RD results, there is a further decline in relative irrigation rates in 1920. This is due to the
development of the Milk River Reclamation project which happens to be beyond the land grant. This is in part why
we utilize the 1919 cutoff which pre-dates these water rights of 1920.
33
A contemporary article about Montana raised this issue, stating, “[i]f a claim was located, and on the survey it
proved that it was a railroad section, it must be purchased from the company or held under a clouded title. The delay
in surveying the grant, had, accordingly, further retarded the development of the country” (New York Times, 1892).
26
created uncertainty across the entire grant, even on what were ultimately “secure” even
sections.
34
Unable to rest firmly on the randomization of the odd and even sections, it is less
obvious that the NP land grant buffer was itself entirely random or similar to the remainder of
Montana; other correlated factors, not just policy uncertainty, could partially explain the lower
levels of irrigation.
35
We submit that these factors would likely bias us against finding the
negative results we did since the NP would have most likely pushed for a prime settlement route,
as suggested by the Colorado results, as well as the overarching railroad land grant public policy
purpose of western development.
36
Furthermore, our covariates condition on many potential
factors and the negative effect emerging only after 1880, after the uncertainty emerges, is
consistent with the historical record.
Still, to further assess the causal role of the uncertain land grant and spillover effects on
coordination, we turn to the DDD approach and interact the land grant buffer indicators with
distance from the stream. In Figure 7, we present the estimated effect of the land grant on
irrigation rates, captured in 1919 (and 1879), at various distances from a stream for odd sections
and even sections separately (coefficient values are also provided in Online Appendix OB,
Tables OB23 and OB24). What emerges is that at short distances from streams, where gains
from cooperation and economies of scale are small, the irrigation uptake on even sections is only
slightly reduced relative to even sections beyond the land grant.
37
In comparison, the difference
for the odd sections in the land grant at these same distances is twice that of the difference for
even sections. As the distance to stream increases (and costs of irrigation increase), the effect of
the uncertainty increases for both odd and even sections, but odd sections remain statistically less
likely to irrigate than even sections out to 1 mile from a stream and point estimates remain larger,
34
An additional source of uncertainty in Montana could be the extensive Native American reservation lands in
Montana. In 1880, 32 percent of the state was set aside in reservations, although by 1915 merely 6 percent remained
set aside. While interesting to study in its own right (see Taylor 2019), for our purposes we have run a robustness
test, presented in Table OB19, Online Appendix OB, which controls for whether the unit of land is within a
reservation in either year (data provided by Taylor 2019). Interestingly, being in a reservation in 1880 improves the
odds of irrigation by 1919, but this is driven by land no longer included in the reservations by 1915. Our main
estimates remain robust to this inclusion.
35
See Online Appendix OB, Table OB20 for a balance table of variables for NP land grant and the rest of Montana.
On balance, the areas were similar but if anything, the land grant was arguably more suitable for irrigation.
36
We demonstrate below that irrigated acres within the land grant were 50 percent more valuable than irrigated
acreage elsewhere in Montana as of 1930.
37
In an alternative specification interacting the land grant with a quadratic of the continuous stream measure, the
intercept for even sections at zero miles from a stream is not statistically distinct from zero. See Table OB25, Online
Appendix OB.
27
though no longer statistically distinct, through 2 miles.
38
Beyond this distance, the effect on odd
and even sections converge, consistent with uncertainty on the odd sections spilling over to the
even sections, particularly where greater coordination with neighbors is warranted. In contrast, in
Panel B, we provide the same estimates for a snapshot in 1879, just before the uncertainty
emerged. At that time, irrigation was more likely on the land grant at short distances from a
stream and equally likely at further distances. Furthermore, differences between odd and even
sections were nonexistent. All this is to underscore that the anticipated market access of the NP
was a positive draw – equal across the alternate sections – until the policy uncertainty derailed
that development.
Figure 7: Triple difference estimates
7. Costs to the Montana Economy
Both methodologies, exploiting different sources of identification, suggest a substantial
deterrence and delay of irrigation investment within the NP land grant in Montana relative to
similar areas beyond the grant. Meanwhile, along the GN, which received no land grant in
Montana, and along the Colorado land grant areas, no similar effects were found. The effect
along the NP emerges strongly between 1880 and 1890, the decade when the NP land grant
experienced the most uncertainty. Taken together, our results make a compelling argument that
38
We note that beyond the land grant irrigation is most probable within 0.25 miles of a stream, decreasing
significantly beyond that, but leveling off beyond 1 mile.
A: 1919
B: 1879
28
uncertainty stymied irrigation development within Montana. Furthermore, our empirical
examination indicates the effects persisted even after the uncertainty abated.
By combining our empirical results with US agriculture and irrigation census data from
1920 and 1930 (US Bureau of the Census, 1922, 1932a, 1932b), we can approximate by back-of-
the-envelope calculations how costly the delayed and deterred irrigation was to Montana’s
economy. To do so, we extrapolate our coefficients to approximate the acreage affected (181,959
acres not irrigated and 719,364 acres irrigated 3.7 years later) and then monetize this by land
valuation, crop yields, and irrigation costs from the censuses (the complete details are laid out in
Online Appendix OC). Our reduced form model limits our ability to account for general
equilibrium effects, water supply constraints, or alternative land use values, but we do consider
four distinct scenarios, each of which are illustrated in Table 5, with various levels of exposure to
these shortcomings in order to provide a plausible range of costs.
Table 5
All scenarios assume that plot A would have been irrigated.
39
Scenario 1 is the most
extreme and presumes that the land not irrigated within the grant is not otherwise farmed nor is
the irrigation investment displaced to beyond the grant. The implied loss per acre is A, the full
value of irrigated cropland within the land grant that would have otherwise occurred. Scenario 2
allows that while irrigation may have been dissuaded, other economic activity may have
occurred in its place, offsetting the losses by that value, B. For calculation purposes, we assume
39
The econometric model does not directly consider water supply constraints created by other irrigation withdrawals
in the same basin, meaning not all “A” parcels would necessarily have water available. However, the estimates are
conditioned on stream distance and size, the latter of which approximates supply, meaning streams of similar size
and distance beyond the land grant have more irrigation, suggesting water supply would be sufficient in many cases.
Scenario Loss
1919 Crop Production and Costs 1930 Farm Value
1A $5,967.35 $3,603.10
2A-B $5,967.35-$3,891.23=$2,076.13 $3,603.10-$2,223.06=$1,380.04
A-C N/A $3,603.10-$1,881.30=$1,721.80
A+D-B-C N/A ($3,603.10+-$1,954.52)-($2,223.06+$1,881.30)=$1,453.27
Irrigated cropland
Dry-land crop land
No crops
Note: Calculations for losses are detailed in Appendix OC.
Counterfactual
Estimated Net Present Values of Loss per Acre (2015 $)
NP
Counterfactual Alternatives
A
D
4
B
C
3
C
A
B
A
Beyond
NP
Beyond
A
Observed
29
dry-land farming is the alternative use.
40
Because non-irrigated acreage yielded only 55 percent
of what irrigated acreage yielded in 1919 (US Bureau of the Census 1922, Montana, Table 18),
these losses could still be substantial compared to the fully irrigated agricultural production that
would have otherwise occurred.
41
In Scenario 3, we consider that irrigation investment was
displaced from within the grant to beyond the grant, offsetting the loss by C, the value of
irrigated cropland beyond the grant. County level analysis of census data find irrigated farms
were 51 percent less valuable per acre beyond the land grant in 1930 (US Bureau of the Census
1932b), meaning displacement will not fully offset the losses of A.
42
Finally, in Scenario 4, we
allow for both substitution of dry-land farming within the land grant and irrigated farming
displaced beyond the land grant.
In terms of estimating values for A, B, C, and D, we take two approaches based on
available data in the 1920 and 1930 censuses. Because we are drawing on the 1919 coefficients,
the 1920 data (which reports 1919 growing season) is preferable, but we are forced to assume
A=C and B=D because the irrigated land value is only reported at the state level. With no spatial
variation we can only estimate scenarios 1 and 2. To do so, we compare the market value of
crops grown on non-irrigated acreage to value of crops grown on irrigated land, accounting for
the additional capital maintenance costs of irrigation ditches.
43
We then convert this annual loss
to an NPV for this cash flow indefinitely into the future and adjust to 2015 dollars.
44
In the 1930
census, crop yields do not appear, but average farm value for irrigated and non-irrigated farms
for each county is reported. To estimate A and B we take the average of per acre farm value for
irrigated and non-irrigated farms respectively in the 14 counties falling entirely within the land
40
For any specific plot of land this could be mining, timber, ranching, or commercial rather than dry-land cropping.
However, we apply dry-land cropping because: 1) our model predicts these parcels to be irrigated absent uncertainty
based on soil quality, elevation, topography, and latitude and longitude, making it likely these particular parcels
were more suitable for farming than timber, mining, or commercial use; and 2) dry-land cropping is likely less
valuable (per acre) than commercial and mining activities but more valuable than ranching, providing a roughly
average value of the alternatives.
41
Reproduction of the table is provided in Online Appendix OC, Figure OC1.
42
Results of these analyses are provided in Online Appendix OC, Table OC1.
43
Agriculture markets retained regional segmentation at this time, meaning crop prices in Montana would be
affected in general equilibrium by more production. If yields were the only difference, percentage terms should not
be sensitive to price changes. However, to the extent that crop prices would decline relative to costs of irrigation,
these percentages still overstate the loss to the economy.
44
We follow Leonard & Libecap (2019) and apply Fogel’s (1964) social discount rate of 7.91 percent to move
between annualized amounts and net present values.
30
grant, repeating the process for the 14 counties beyond the buffer to estimate C and D.
45
All
these per acre estimates are provided in Table 5 above. Notably, the 1930 methodology results in
numbers that are nearly half of the 1919 methodology, but we should note that the per-acre farm
values experienced a sharp decline between 1920 and 1930 – 43 percent lower in Montana – so
we emphasize the percent of total farm value in the relevant year over the actual monetary
amount as a better comparison.
46
In Table 6 we report the aggregate losses of deterred and
delayed irrigation when scaled to the affected land.
Table 6
The largest estimated cost of scenario 1 – in which the deterred irrigation is not
attenuated in any way – is a $2.3 billion reduction of NPV, or 21.6 percent of total farm value
(land and buildings) in Montana at the time. If, instead, the lack of irrigation meant dry-land
farming (scenario 2), the losses are reduced to 8.7 percent of total farm value. Drawing on the
1930 methodology, the effects are smaller in percentage terms, 14.3 percent and 5.5 percent
45
See Table OC2, Online Appendix OC. Because these are in terms of farm acres, we use the average irrigated land
ratios to convert to the implied value of irrigated crop land contained in those farms. Specifics are in Online
Appendix OC.
46
See Table 20, 1930 Agriculture Census Volume 4: Farms and Farm Property (US Bureau of the Census 1932a).
This trend is true for the entire United States (28 percent decline).
Counterfactual
Displaced Investment
1919 1930 1930
Scenario 1 Scenario 1 Scenario 3
Deterred $1,083.64 $654.20 $312.70
(Percent) 10.2% 9.0% 4.3%
Delayed $1,205.55 $388.90 $388.90
(Percent) 11.4% 5.3% 5.3%
Total NPV Loss $2,289.19 $1,043.09 $701.60
(Percent) 21.6% 14.3% 9.7%
Scenario 2 Scenario 2 Scenario 4
Deterred $377.01 $250.39 $263.75
(Percent) 3.6% 3.4% 3.6%
Delayed $547.89 $148.85 $148.85
(Percent) 5.2% 2.0% 2.0%
Total NPV Loss $924.91 $399.24 $412.60
(Percent) 8.7% 5.5% 5.7%
Note: 1919 method scales our estimated impact on irrigated land by the net value of crops on irrigated land. The 1930
method scales our estimates by the value of land in irrigated and non-irrigated farms, spatially differentiated across
Montana counties. Percent is in terms of the total farm value in Montana reported in the 1920 and 1930 censuses
respectively. Details on calculations are provided in the text and Appendix OC.
Dry-land crops
No Crops
Counterfactual Cost Estimates (2015 Millions $)
No Investment
31
respectively. With the 1930 data, we can also estimate the costs in scenarios 3 and 4. Total losses
in these scenarios range from 5.7 percent to 9.7 percent.
47
Overall, we find that the policy uncertainty imposed considerable losses to Montana’s
agriculture sector. Even in the conservative calculations, 5.5% of farm value is absent. In terms
of the wider economy, if we annualize the losses and compare to Montana’s total state income in
1930 ($3.5 billion according to Leonard & Libecap 2019 calculations, Table 3), the annualized
losses range from 1.1 to 4.7 percent of state income. Although their precision is subject to
caveats embedded in the assumptions and data limitations, these calculations demonstrate that
the uncertainty induced by the NP land grant resulted in non-trivial economic losses.
8. Conclusion
Utilizing the massive NP land grant in Montana and its tumultuous history as a natural
experiment, we identify a causal effect of public policy uncertainty (in our case inducing
property rights insecurity) on coordinated, asset-specific investment. Irrigation infrastructure
critical to agricultural development in the region was, depending on the specification, 12 to 25
percent less prevalent within the land grant, where uncertainty prevailed from 1879-1894. Land
that was eventually irrigated tended to be developed over 3 years later. Though the uncertainty
pertained only to odd-numbered sections, we find negative and significant spillover effects on
the adjacent even-numbered sections in the land grant, particularly further from streams where
fixed costs are higher and coordinated investment across land parcels is more desirable. These
results underscore the economic costs of lingering title insecurity along frontiers, historic or
contemporary, once the demand for title has emerged. Although the appropriate role for the
government in defining property rights remains a question worthy of further consideration, our
analysis suggests that the ideal role for the government in securing title is to first do no harm –
government practices that create uncertainty can observably stymy the beneficial development of
property. Unlike loss of property entirely, which means an absolute loss of investment, policy
47
We assume that delayed irrigation was not displaced during the delay as it seems impractical given the asset
specificity of irrigation ditches (that is, irrigators would not have irrigated beyond the land grant for 3.7 years and
then abandoned that infrastructure to dig a new ditch within the land grant once certainty emerged). This means the
delay costs are the same for scenarios 1 and 3 and scenarios 2 and 4 using the 1930 methodology. We also note that
dry-land crops did not ameliorate the losses in the case of displaced irrigation. This is because, according to the
census data, even non-irrigated farms within the land grant were more valuable than irrigated farms beyond.
32
change, unless particularly severe, should only reduce the value of the flow of benefits from the
asset. Our work here identifies the effect of policy uncertainty at a large scale, and thereby
informs theoretical and policy considerations surrounding its long-run effects. In addition, our
specific context and results uncover an economic cost that has not yet entered the retrospective
ledger book of the costs and benefits of the massive railroad land grants.
These general contributions are in addition to more nuanced outcomes that our empirical
context and results complement. First, the fact that the gap in irrigation did not subside once the
uncertainty did suggests that ownership of one resource has important implications for the
development of that resource’s complements. In this case, the RD analysis indicates that even
once land ownership was more certain, the water rights, in terms of expected water availability,
were less secure given the senior rights already developed on the previously more certain land.
However, it is worth noting that the lack of irrigation development may provide a reversal of
fortune in the long-run as researchers have found fixed-investments early on can deter valuable
re-development later (see Allen & Leonard 2021). Second, we show that policy and property
rights issues on one plot can negatively impact development on neighboring plots when
coordination is warranted. This remains pertinent to scaled development of resources today
including wind farms, natural gas fracking, and fisheries, or more generally, any investment
efforts that span individual land parcels.
Our line of investigation has also prompted additional related questions. Were other
natural resources and their development (for example. minerals or timber) similarly affected by
uncertainty around land grants? Idaho and Washington are considerably more forested than the
areas we treat in our analysis, yet, were subject to similar uncertainty surrounding the disposition
of the Northern Pacific’s land grants in those states. More generally, to what extent did the land
grants around the nation temper the gains the railroads themselves provided? We leave these for
additional research, but our results suggest that in a period of well-recognized fraud and
corruption as between the railroads and the government, such a rapid handout of public lands
was not without its losses. However, this is not to say the NP, or its land grant, was a net loss for
Montana. If without the land grant, no line would have been constructed, or only constructed
much later, irrigation development (and broader development) might have been even further
delayed. Rather, we show that the uncertainty of ownership induced by policy implementation
derailed some development the railroad ushered in and, more generally, that insecure property
33
rights can cause a large reduction in asset specific investment replete with spillover effects and
persistent economic consequences.
Sources Cited
Aizenman, Joshua, and Nancy P. Marion. 1993. "Policy uncertainty, persistence and growth." Review of
International Economics 1.2: 145-163.
Allen, Douglas W. 2019. “Establishing Economic Property Rights by Giving Away an Empire.” Journal
of Law and Economics 62.2: 251-280.
Allen, Douglass W. and Leonard, Bryan. 2021. “Property Right Acquisition and Path Dependence:
Nineteenth-Century Land Policy and Modern Economic Outcomes.” The Economic Journal ueab030.
Alston, Eric and Steven M. Smith. 2021. “Short-term Political Incentives and Long-Run Consequences –
The General Land Office & The Northern Pacific.” Working Paper.
Alston, Eric, and John Stafford. 2018. “Muddying the Waters: Sixty-One Years of Doctrinal Uncertainty
in Montana Water Law." Brigham-Kanner Property Rights Conference Journal 7:155-184.
Alston, Lee J., Edwyna Harris, and Bernardo Mueller. 2012. “The Development of Property Rights on
Frontiers: Endowments, Norms and Politics.” Journal of Economic History 72.3:741-70.
An Act granting Lands to aid in the Construction of a Railroad and Telegraph Line from Lake Superior to
Puget Sound, on the Pacific Coast, by the Northern Route. Pub. L. 38-217. 13 Stat 365. 2 July 1864.
Library of Congress. Web. February 8, 2019.
Appleman, Roy E. 1939. "Timber Empire from the Public Domain." The Mississippi Valley Historical
Review 26.2: 193-208.
Arruñada, Benito. 2012. Institutional foundations of impersonal exchange: Theory and policy of
contractual registries. Chicago and London: University of Chicago Press.
Atack, Jeremy, and Peter Passell. 1994. A New Economic View of American History: From Colonial
Times to 1940. 2nd ed. W.W. Norton and Company.
Atack, Jeremy. 2016. Historical Geographic Information Systems (GIS) database of U.S. Railroads for
1826-1911. (computer file)
Baker, Scott R., Nicholas Bloom, and Steven J. Davis. 2016. Measuring economic policy uncertainty. The
Quarterly Journal of Economics 131.4: 1593-1636.
Bandiera, Oriana. 2007. "Land tenure, investment incentives, and the choice of techniques: Evidence
from Nicaragua." The World Bank Economic Review 21.3: 487-508.
Bechtold, Timothy M. 1992. "Now v. Forever: The Conflict Between Business and Forestry in the
Management of Plum Creek Timberlands in Montana" Graduate Student Theses, Dissertations, &
Professional Papers no. 3108. University of Montana.
Besley, Timothy. 1995. “Property Rights and Investment Incentives: Theory and Evidence from Ghana.
Journal of Political Economy 103: 903-937.
Blyth, William, Richard Bradley, Derek Bunn, Charlie Clarke, Tom Wilson, and Ming Yang. 2007.
"Investment risks under uncertain climate change policy." Energy Policy 35.11: 5766-5773.
34
Bohn, Henning, and Robert T. Deacon. 2000. "Ownership risk, investment, and the use of natural
resources." American Economic Review 90.3: 526-549.
Brasselle, Anne-Sophie, Frédéric Gaspart, and Jean-Philippe Platteau. 2002. “Land Tenure Security and
Investment Incentives: Puzzling Evidence from Burkina Faso.Journal of Development Economics,
68: 373-418.
Bretsen, Stephen N., and P.J. Hill. 2007. “Irrigation Institutions in the American West.” UCLA Journal of
Environmental Law and Policy, 25.2: pg. range
Brooks, L. and B. Lutz. 2016. “From today’s city to tomorrow’s city: An empirical investigation of urban
land assembly.” American Economic Journal: Economic Policy 8(3): 69–105.
Bulan, Laarni, Christopher Mayer, and C. Tsuriel Somerville. 2009. "Irreversible investment, real options,
and competition: Evidence from real estate development." Journal of Urban Economics 65.3: 237-
251.
Bureau of Land Management (BLM). General Land Office Records. 2016.
https://glorecords.blm.gov/search/default.aspx?searchTabIndex=0&searchByTypeIndex=0
Burness, H. Stuart and James P. Quirk. 1979. Appropriative water rights and the efficient allocation of
resources. The American Economic Review 69.1: 25-37.
Caldara, Dario, Matteo Iacoviello, Patrick Mollig, Andrea Prestipino, and Andrea Raffo. 2020. The
economic effects of trade policy uncertainty. Journal of Monetary Economics 109: 38-59.
Calonico, Sebastian, Matias Catteneo, and Max Farrel. 2017. "rdrobust: Software for Regression-
Discontinuity Designs." The STATA Journal 17.2: 372-404.
Clay, Karen, and Gavin Wright. 2005. "Order without law? Property rights during the California gold
rush." Explorations in Economic History 42.2: 155-183.
Clinch, Thomas A. 1965. "The Northern Pacific Railroad and Montana's Mineral Lands." Pacific
Historical Review. 34.3: 323-335.
Coles, Jerilyn W., and William S. Hesterly. 1998. "The impact of firm-specific assets and the interaction
of uncertainty: an examination of make or buy decisions in public and private hospitals." Journal of
Economic Behavior & Organization 36.3: 383-409.
Colorado. State Board of Land Commissioners. 1904. Biennial report of State Board of Land
Commissioners of Colorado. Denver, CO: The Smith-Brooks Printing Co., State Printers.
Coman, Katherine. 1911. “Some Unsettled Problems of Irrigation.” American Economic Review 1: 1-19.
Conover, Milton. 1923. The General Land Office: Its History, Activities, and Organization. No. 13.
Baltimore, MD: Johns Hopkins Press.
Costello, Christopher and Daniel T. Kaffine. 2010. “Marine protected areas in spatial property-rights
fisheries.” Australian Journal of Agricultural and Resource Economics 54.3: 321-341.
Cotroneo, Ross R. 1976. "Timber Marketing by the Northern Pacific Railway, 1920-1952." Journal of
Forest History 20.3: 120-131.
Daffran, George. 1998. “Taking Back Our Land: A History of Railroad Land Grant Reform.” Railroads &
Clearcuts Campaign, www.landgrant.org.
35
de Janvry, Alain, Marco Gonzalez-Navarro, and Elisabeth Sadoulet. 2014. "Are land reforms granting
complete property rights politically risky? Electoral outcomes of Mexico's certification
program." Journal of Development Economics 110: 216-225.
Decker, Leslie E. 1960. "The Railroads and the Land Office: Administrative Policy and the Land Patent
Controversy, 1864-4896." The Mississippi Valley Historical Review 46.4: 679-699.
Demsetz, Harold. 1967. “Toward a Theory of Property Rights.” American Economic Review 57.2: 347-
59.
Donaldson, Dave and Richard Hornbeck. 2016. “Railroads and American Economic Growth: A Market
“Access” Approach.” The Quarterly Journal of Economics 799–858.
Edwards, Eric C. and Steven M. Smith. 2018. “The Role of Irrigation in the Development of American
Agriculture.” The Journal of Economic History 78.4:1103-1141.
Edwards, Eric C., Trevor O’Grady, and David Jenkins. 2019. “The Effect of Land Ownership on Oil and
Gas Production: A Natural Experiment.” CEnREP Working Paper No. 19-017.
Ellis, David Maldwyn, Richard C. Overton, Robert E. Riegel, Herbert O. Brayer, Chester McArthur
Destler, Stanley Pargellis, Fred A. Shannon, and Edward C. Kirkland. 1946. "Comments on" The
Railroad Land Grant Legend in American History Texts"." The Mississippi Valley Historical Review
32.4:557-576.
Ellis, David Maldwyn. 1946. "The Forfeiture of Railroad Land Grants, 1867-1894." The Mississippi
Valley Historical Review 33.1: 27-60.
ESRI database. 2016 USA Railroads. (computer file).
Fabrizio, Kira R. 2013. "The effect of regulatory uncertainty on investment: evidence from renewable
energy generation." The Journal of Law, Economics, & Organization 29.4: 765-798.
Fairweather, Hanford W. 1919. "The Northern Pacific Railroad and Some of Its History." The
Washington Historical Quarterly 95-99.
Fogel, Robert W. 1964. Railroads and American economic growth. Baltimore, MD: Johns Hopkins Press.
Fogel, Robert W. 1966. "The new economic history." The Economic History Review 19.3: 642-656.
Fuss, Sabine, Jana Szolgayova, Michael Obersteiner, and Mykola Gusti. 2008. Investment under market
and climate policy uncertainty. Applied Energy 85.8: 708-721.
Galiani, Sebastian, and Ernesto Schargrodsky. 2010. "Property rights for the poor: Effects of land
titling." Journal of Public Economics 94.9-10: 700-729.
Galiani, Sebastian, and Ernesto Schargrodsky. 2011. "Land property rights and resource allocation." The
Journal of Law and Economics 54, no. S4: S329-S345.
Gates, Paul W. (with a chapter by Robert W. Swenson). 1968. History of public land law development.
Vol. 62. Washington, D.C.: US Government Printing Office.
Gates, Paul W. 1954. "The Railroad Land-Grant Legend." The Journal of Economic History 14.2: 143-
146.
Getches, David H., 2009. Water law in a nutshell. 4th ed. St. Paul, MN: Thomson/West
Gibb, Ryan. 2017. "The Land Reform Deception: Political Opportunism in Zimbabwe's Land Seizure
Era." African Studies Quarterly 17.2: 100-101.
36
Goldstein, Markus, and Christopher Udry. 2008. “The Profits of Power: Land Rights and Agricultural
Investment in Ghana.Journal of Political Economy 116: 981-1022.
Grainger, Corbett A., and Christopher J. Costello. 2014. “Capitalizing Property Rights Insecurity in
Natural Resource Assets.” Journal of Environmental Economics and Management 67.2: 224–40.
https://doi.org/10.1016/j.jeem.2013.12.005.
Greever, William S. 1951. "A Comparison of Railroad Land-Grant Policies." Agricultural History 25.2:
83-90.
Gulen, Huseyin, and Mihai Ion. 2016. "Policy uncertainty and corporate investment." The Review of
Financial Studies 29.3: 523-564.
Haines, Michael, Fishback, Price, and Rhode, Paul. 2018. United States Agriculture Data, 1840 - 2012
(computer file). Ann Arbor, MI: Inter-university Consortium for Political and Social Research.
https://doi.org/10.3886/ICPSR35206.v4
Handley, Kyle, and Nuno Limao. 2015. "Trade and investment under policy uncertainty: theory and firm
evidence." American Economic Journal: Economic Policy 7.4: 189-222.
Haney, Lewis Henry. 1908. A congressional history of railways in the United States, 1850-1887. No. 1-2.
University of Wisconsin.
Hibbard, Benjamin Horace. 1924. History of the Public Land Policies. New York: Macmillan
Ishii, Jun, and Jingming Yan. 2004. "Investment under regulatory uncertainty: US electricity generation
investment since 1996." Working Paper no. 127. Center for the Study of Energy Markets.
Jacoby, Hanan G., Guo Li, and Scott Rozelle. 2002. "Hazards of expropriation: tenure insecurity and
investment in rural China." American Economic Review 92,5: 1420-1447.
Johnson, Simon, John McMilllan, and Christopher Woodruff. 2002. “Property Rights and Finance,”
American Economic Review 92: 1335- 1356.
Julian, George W. 1883. "Railway Influence in the Land Office." The North American Review 136.316:
237-56.
Julio, Brandon and Youngsuk Yook. 2012. Political uncertainty and corporate investment cycles. The
Journal of Finance 67.1: pp.45-83
Lanjouw, Jean O., and Phillip I. Levy. 2002. “Untitled: A Study of Formal and Informal Property Rights
in Urban Ecuador.Economic Journal 112: 986-1019.
Lawry, Steven, Cyrus Samii, Ruth Hall, Aaron Leopold, Donna Hornby, and Farai Mtero. 2014. “The
Impact of Land Property Rights Interventions on Investment and Agricultural Productivity in
Developing Countries.” Campbell Systematic Reviews 1: pg number.
https://doi.org/10.4073/csr.2014.1.
Leonard, Bryan and Gary D. Libecap. 2019. “Collective Action by Contract: Prior Appropriation and the
Development of Irrigation in the Western United States.Journal of Law and Economics 62.1:67-
115.
Leonard, Bryan, and Dominic Parker. 2020. “Private vs. Government Ownership of Natural Resources:
Evidence from the Bakken,” Economic Journal (forthcoming).
Lewis, Eric. 2019. “Patchwork Policies, Spillovers, and the Search for Oil and Gas.” American Economic
Journal: Economic Policy 11.1: 380–405. https://doi.org/10.1257/pol.20160373.
37
Libecap, Gary D. 1978. "Economic variables and the development of the law: The case of western
mineral rights." The Journal of Economic History 38.2: 338-362.
Liscow, Zachary D. 2013. “Do Property Rights Promote Investment but Cause Deforestation? Quasi-
Experimental Evidence from Nicaragua.” Journal of Environmental Economics and Management
65.2: 241–61. https://doi.org/10.1016/j.jeem.2012.07.001.
Lundquist, J. K., K. K. DuVivier, D. Kaffine, and J. M. Tomaszewski. 2019. “Costs and consequences of
wind turbine wake effects arising from uncoordinated wind energy development.” Nature Energy 4.1:
26-34.
McClintock, Anthea. 2009. “Investment in irrigation technology: an application of real options analysis.”
No. 420-2016-26724.
Mercer, Lloyd J. 1982. Railroads and land grant policy: a study in government intervention. Amsterdam,
Netherlands: Elsevier.
Mian, Atif, Amir Sufi, and Francesco Trebbi. 2015. "Foreclosures, house prices, and the real
economy." The Journal of Finance 70.6: 2587-2634.
Micheli, E. R., J. W. Kirchner, and E. W. Larsen. 2004. "Quantifying the effect of riparian forest versus
agricultural vegetation on river meander migration rates, Central Sacramento River, California,
USA." River Research and Applications 20.5: 537-548.
National Atlas of the United States. 1:1,000,000-Scale Hydrographic Geodatabase of the United States—
Conterminous United States: Vector digital data. Rolla, MO, 2014. http://nationalatlas.gov/atlasftp-
1m.html
Newell, F.H. Department of the Interior. 1894. Report on Agriculture by Irrigation in the Western Part of
the United States at the Eleventh Census: 1890.
http://usda.mannlib.cornell.edu/usda/AgCensusImages/1890/1890a_v5-14.pdf
Powers, Fred Perry. 1889. “The Guilford Miller Case and the Railroad Indemnity Lands.” Political
Science Quarterly 4.3: 452-479.
Rae, John B. 1938. "Commissioner Sparks and the Railroad Land Grants." The Mississippi Valley
Historical Review 25.2: 211-30
Rae, John B. 1952. “The Great Northern's Land Grant.” The Journal of Economic History 12.2: pp. 140-
145.
Rand McNally, and Northern Pacific Railroad Company. New and Correct Map of the Lines of the
Northern Pacific Railroad and Oregon Railway & Navigation Co. Chicago: Retrieved from the
Library of Congress, 1883. https://www.loc.gov/item/98688749/ .
Roberts, Darwin P. 2011. "The Legal History of Federally Granted Railroad Rights-of-Way and the Myth
of Congress's 1871 Shift." University of Colorado Law Review 82:85-166.
Rodrik, Dani. 1991. “Policy uncertainty and private investment in developing countries.” Journal of
Development Economics 36.2: 229-242.
Schwinden, Ted. 1950. "Northern Pacific land grants in Congress." Graduate Student Theses,
Dissertations, & Professional Papers no. 4807. University of Montana.
Shields Jr, F. Douglas, Andrew Simon, and Lyle J. Steffen. 2000. "Reservoir effects on downstream river
38
channel migration." Environmental Conservation 27.1: 54-66
Sjaastad, Espen and Daniel W. Bromley. 1997. “Indigenous Land Rights in Sub-Saharan Africa:
Appropriation, Security and Investment Demand.” World Development 25.4: 549–62.
Skyner, Louis. 2001. "Political conflict and legal uncertainty: the privatisation of land ownership in
Russia." Europe-Asia Studies 53.7: 981-999.
Smith, Cory. 2020. “Land Concentration and Long-Run Development: Evidence from the Frontier United
States.” Working Paper. Available http://economics.mit.edu/files/18471
Smith, Steven M. 2021. The relative economic merits of alternative water right systems. Journal of
Environmental Economics and Management 105: 102389.
Taylor, Laura. 2019. “American Indian Reservation Boundaries and Land Characteristics: A Spatial
Analysis of Reservation Decline in the American West, 1880-1915.” Working Paper. University of
Arizona, Tucson, A.Z.
Teele, Ray. 1904. “The Organization of Irrigation Companies.” Journal of Political Economy 12: 161-
178.
The New York Times. 1892. “A Montana Question.” January 28.
Umbeck, John. 1977. "The California Gold Rush: A study of Emerging Property Rights." Explorations in
Economic History 14: 197-226.
United States. Department of Interior. Annual report of the Commissioner of the General Land Office to
the Secretary of the Interior. Washington: GPO, 1882. Digital.
———. Annual report of the Commissioner of the General Land Office to the Secretary of the Interior.
Washington: GPO, 1885. Digital.
———. Annual report of the Commissioner of the General Land Office to the Secretary of the Interior.
Washington: GPO, 1895. Digital.
US Bureau of the Census. “1930 Census: Agriculture Volume 4. General Report.,” 1932a.
https://www.census.gov/library/publications/1932/dec/1930d-vol-04-agriculture.html
———. “Fifteenth Census of the United States: 1930: Irrigation of Agricultural Lands,” 1932b.
https://www.census.gov/library/publications/1932/dec/1930e-irrigation.html
———. “Fourteenth Census of the United States Taken in the Year 1920: Volume VII, Irrigation and
Drainage,” 1922. http://agcensus.mannlib.cornell.edu/AgCensus/censusParts.do?year=1920
Voggesser, Garrit. Bureau of Reclamation. 2001. “The Flathead Project: The Indian Projects.
https://www.usbr.gov/history/ProjectHistories/INDIAN PROJECTS FLATHEAD PROJECT.pdf
Wahlgren Summers, Mark. 1993. The Era of Good Stealings. New York: Oxford University Press.
Williamson, Oliver E. 1979. “Transaction-cost economics: the governance of contractual relations.” The
Journal of Law and Economics 22.2: 233-261.
Williamson, Oliver E. 1983. "Credible commitments: Using hostages to support exchange." The
American Economic Review 73.4: 519-540.
Winters v. United States, (207 U.S. 564, 1908)
... Next to the institutional channel, are there any other potential drivers of an economic performance premium upon accession? A leading alternative explanation for the growth success of the USA since the nineteenth century is the large domestic market (Alston and Smith 2019;Fogel 1962). The economic integration of the continent spurred by the expansion of the railroad network throughout the nineteenth century gave firms access to large markets, triggered human capital mobility and stimulated continental knowledge flows. ...
Article
Full-text available
This paper quantifies the economic benefits of joining the USA. Adapting extant static synthetic control models into a dynamic model similar to Arellano and Bond (Rev Econ Stud 58(2):277–297, 1991), we are able to construct the counterfactual growth paths of Texas, California, Arizona, New Mexico, Colorado, Utah, Wyoming and Nevada had they not joined the USA. We show that the real growth path outperforms the counterfactuals substantially in all cases. In the same way, we construct counterfactual growth paths of Puerto Rico, Cuba, the Philippines and Greenland in the scenario where they joined the USA at times in history where this might have been a (remote) possibility. We find counterfactual growth to be substantially higher than the actual growth. Having established the positive economic effects of US membership, we subsequently assess the sources of this added growth, distinguishing between a class of explanations related to internal market access and a class of explanations related to institutional quality. Using a large number of determinants of institutional quality, we find that the institutional quality of the USA as a whole matches the quality predicted for New England most closely. This suggests that upon accession, states imported the institutional quality of New England, which was typically superior to what they would have likely developed by themselves. We show that this institutional bonus accounts for the bulk of the growth benefits of US accession. While we warn against interpreting our results as evidence for the superiority of US culture or institutions in any way, our empirical findings indicate large historical growth and institutional benefits of US statehood.
Article
This paper compares lands settled between 1862 – 1940 under the Homestead Act to lands that sold for cash during the same time. We combine recently digitized individual land patents with modern satellite data and find a negative effect of homesteading on modern land use that cannot be explained by land quality, title characteristics, or unobserved differences in settlers. We test the hypothesis that early homestead settlement put homesteaders “in the way” of future development, creating a path dependence in land use decisions for homesteads, despite the fact that their legal rights were identical to purchased lands.
Article
Property institutions should ideally provide economic actors with certainty that their local choices about investment will not be unsettled by shifting political economic equilibria. We argue that for this to occur, political autonomy, administrative and enforcement capacity, political constraints, and accessible legal institutions are each necessary. A comparison of the evolution of property rights for settlers and American Indians in the United States shows how political and legal forces shape the evolution of property institutions. American Indians, who had property institutions before Europeans arrived, could not defend their land from Europeans and later Americans due to lacking military capacity. Settlers' property rights were relatively secure because the government had sufficient autonomy and capacity to broadly define and enforce their rights, political institutions constrained the government from expropriating settlers' property, and legal institutions provided a forum for settlers to adjudicate and defend their rights in court. Native Americans, in contrast, had systematically inconsistent and often expropriative policy treatment by the government. Although tribes have technically been sovereign since the 1970s, tribal governments continue to lack sufficient political and legal autonomy and capacity to define and enforce property institutions in response to evolving local conditions.
Article
This paper compares lands settled between 1862 – 1940 under the Homestead Act to lands that sold for cash during the same time. We combine recently digitized individual land patents with modern satellite data and find a negative effect of homesteading on modern land use that cannot be explained by land quality, title characteristics, or unobserved differences in settlers. We test the hypothesis that early homestead settlement put homesteaders “in the way” of future development, creating a path dependence in land use decisions for homesteads, despite the fact that their legal rights were identical to purchased lands.
Article
How natural resources are measured and bounded within a property right system can influence their development and productivity. This is especially true for surface water given its fluid, fungible, and stochastic nature. Two quantifications have emerged: The prior appropriation doctrine provides absolute quantities to water allocated based on first use while proportional water rights distribute a set percentage of total water to owners. While theoretical differences have been identified, empirical tests are scarce due to the endogenous choice of water rights. I identify and utilize a natural experiment where acequias (Hispanic-rooted irrigation ditches) developed in Territorial New Mexico are later divided by the formation of Colorado, exogenously compelling that subset to be subject to the priority system while those in New Mexico continue to practice proportional division. Drawing on a broad compilation of archival, administrative, satellite, hydrological, and survey data, I find priority rights curtail investment for later irrigators, but that the marginal product of water is generally lower under that right system. This research advances our understanding of how distinct property right systems affect resource development and use. The results can guide stakeholders devising or reforming rights to a resource, particularly ones with stochastic flows.
Article
Does land fragmentation impair spatially expansive natural resource use? We conduct empirical tests using ownership variation on the Bakken, one of the world's most valuable shale oil reserves. Long before shale was discovered, U.S. policies created a mosaic of private, jointly owned, and tribal government parcels on the Fort Berthold Indian Reservation. We find that all three forms of fragmentation reduced production during the 2010–2015 oil boom, especially joint ownership and the interspersion of small parcels of government and private land. We estimate implied gains from consolidation and discuss implications for the use (or conservation) of other spatially expansive resources.
Article
This paper studies the effects of unexpected changes in trade policy uncertainty (TPU) on the U.S. economy. Three measures of TPU are constructed using newspaper coverage, firms’ earnings calls, and tariff rates. Firm-level and aggregate macroeconomic data reveal that increases in TPU reduce business investment. The empirical results are interpreted through the lens of a two-country general equilibrium model with nominal rigidities and firms’ export participation decisions. News and increased uncertainty about higher future tariffs reduce investment and activity.
Article
The United States has a complex patchwork of mineral ownership, where rights to oil and gas may be owned by the federal government, state governments, or private agents. I show why the policies imposed by one owner have theoretically ambiguous spillover effects on the drilling and production outcomes of neighboring plots of land. Exploiting a natural experiment in Wyoming with exogenous ownership assignment, I find significant spillovers: federal land close to state land has a lower probability of drilling than federal land far from state land.
Book
The Amazon, the world’s largest rain forest, is the last frontier in Brazil. The settlement of large and small farmers, squatters, miners, and loggers in this frontier during the past thirty years has given rise to violent conflicts over land as well as environmental duress. Titles, Conflict, and Land Use examines the institutional development involved in the process of land use and ownership in the Amazon and shows how this phenomenon affects the behavior of the economic actors. It explores the way in which the absence of well-defined property rights in the Amazon has led to both economic and social problems, including lost investment opportunities, high costs in protecting claims, and violence. The relationship between land reform and violence is given special attention. The book offers an important application of the New Institutional Economics by examining a rare instance where institutional change can be empirically observed. This allows the authors to study property rights as they emerge and evolve and to analyze the effects of Amazon development on the economy. In doing so they illustrate well the point that often the evolution of economic institutions will not lead to efficient outcomes. This book will be important not only to economists but also to Latin Americanists, political scientists, anthropologists, and scholars in disciplines concerned with the environment.
Article
We describe a major upgrade to the Stata (and R) rdrobust package, which provides a wide array of estimation, inference, and falsification methods for the analysis and interpretation of regression-discontinuity designs. The main new features of this upgraded version are as follows: i) covariate-adjusted bandwidth selection, point estimation, and robust bias-corrected inference, ii) cluster–robust bandwidth selection, point estimation, and robust bias-corrected inference, iii) weighted global polynomial fits and pointwise confidence bands in regression-discontinuity plots, and iv) several new bandwidth selection methods, including different bandwidths for control and treatment groups, coverage error-rate optimal bandwidths, and optimal bandwidths for fuzzy designs. In addition, the upgraded package has superior performance because of several numerical and implementation improvements. We also discuss issues of backward compatibility and provide a companion R package with the same syntax and capabilities.