Download Adapting to the Weather: Lessons from U.S. History Hoyt Bleakley

Adapting to the Weather: Lessons from U.S. History
Hoyt Bleakley †
Sok Chul Hong ‡
Draft in July 2012
Abstract
An important unknown in understanding the impact of climate change is the scope for
adaptation, a long-term process that requires observations on historical time scales. We
consider the impact of weather across much of US history on various measures of
productivity. Using cross-sectional and panel methods, we document economically and
statistically significant responses of agricultural and individual productivity to weather.
We find particularly strong effects of hotter and wetter weather early in US history, but
these effects are almost completely attenuated in recent decades. We evaluate several
possible channels through which these adaptations took place. We contrast our results to
the existing literature that makes inferences from shorter time horizons.
Keywords: climate change, adaptation, hedonics, farm value, farm output value, cohort
effects
†
University of Chicago, Booth School of Business, and NBER, [email protected]
‡
Sogang University, School of Economics, [email protected]
1
1 Introduction
A major issue in forecasting the economic effect of global warming is how much
people can adapt to warmer weather over the long term. If weather affects production or
investment, adaptation can take several forms: adjusting the mix of factors of production,
choosing an alternative technology, or even developing a new technology more suited to
the new climate. On the other hand, long-term changes in weather could bring ecological
changes, which might be amplified by human action, especially if property rights are not
well established (Diamond 1997; Libecap et al.). In any case, understanding the scope of
long-term adjustment is key for measuring the impact of climate change.
While early approaches to this question relied on simulations of agronomic models,
more recent econometric work has used hedonic/cross-sectional or fixed-effect/panel
methods. Hedonic models account for the price of land across areas, with various factors,
including climate variables. The estimated coefficients are interpreted as the discounted
marginal value of these factors to land productivity. Examples of this work are by
Mendelsohn, Nordhaus, and Shaw (1994) and Schlenker, Hanemann, and Fisher (2005,
2006). (We review the literature further in Section 2.) If areas have had enough time to
adapt to permanent features of their climate, the hedonic estimates can measure long-term
effects of climate. A limitation of such cross-sectional comparisons, however, is that they
measure the effect of climate on land value plus the value of any omitted variables that
happen to be correlated with the measures of climate employed. Further, it bears
mentioning that the hedonic approach is not direct evidence of adaptation: hedonic
methods do not actually observe adaptation, but rather assume it. A complementary
approach to this question uses panel and fixed-effect models, thus considering the impact
of across-year variation in weather within an area. Examples of this type of work include
Deschênes and Greenstone (2007), Dell, Jones, and Olken (2008), Schlenker et al. (2008),
Deschênes, Greenstone, and Guryan (2009), and Guitieras (2009). Both of these methods
have their advantages, but are nevertheless limited by the range of technology during their
sample window, and, especially for panel methods, by the scope for adjustment over the
time horizon implicit in their research design. Meaningful adaptation takes place over at
long horizons, and therefore direct evidence of adaptation requires observation over longer
periods of time.
In the present study, we adopt aspects of both methodologies, but from a longer-term
2
perspective. We examine historical data that span a century and a half in the United States.
We employ geospatial data on average weather conditions as well as historical weather
information from measurement stations on temperature and rainfall. These data are
matched to economic variables at the county and state levels for the analysis below.
(Section 3 and Appendix A describe the data.) The US is well suited for this analysis for
two reasons: first, the country has substantial variance of climate across and sufficient
variability of weather across areas to estimate these models; second, there exists
conformable data on both weather and income over a long stretch of time.
We first consider the effect of weather on farm productivity, in Section 4. In the late
1800s, the cross section of farm value per acre versus average temperature was
characterized by an inverse-U shape. Below the median, higher average temperature meant
more valuable farm land, but, above the median, warmer climate implied lower farm-land
values. On the other hand, this penalty for heat and rain was attenuated if not reversed in
more recent decades: in the second half of the 20th century, farm value per acre increase
monotonically with average temperature. A similar pattern is seen for rainfall: above the
median, more average rain predicts lower farm value in the 19th century, but higher farm
value by the middle of the 20th.
Turning to the shorter run, we use panel/fixed-effect methods to measure the response
of farm productivity to variation across years in temperature and rainfall. We particularly
match long-term weather variables---measured at the decade level---with farm value, and
short-term weather variables---measured by the March-to-October average in each census
year---with farm output value. It is estimated that the response of both farm productivity
measures to high temperature or rainfall is much more negative in the early parts of the
sample than in recent years. These results suggest a substantial economic effect of weather
at various horizons, but also point to a change in both the short-run and long-run
relationships between weather and farm productivity, particularly in warm and wet areas.
Next, we examine in Section 5 the relationship between weather early in one’s life
and subsequent income as an adult, among cohorts born in 1860-1960. For those born prior
to circa 1900, being born in a warmer or wetter area is associated with substantially lower
adult income, but this correlation disappears among those born more recently. Similarly,
using a fixed-effect estimator, we find effects on adult income of interannual weather
fluctuations in early childhood. But we find that this effect is strongest among those born
earlier in the sample, and is sharply attenuated among those born in the past 75 years.
3
Once again, we find relationships between weather and income that were strong in the past,
but now are considerably weaker.
Taken together, these results suggest difficulties in forecasting the effects of climate
change with either hedonic or fixed-effect models, particularly if they are estimated on a
relatively narrow time window. We review these issues with a simple, neoclassical model
in Section 6.1. On the one hand, a merit of the cross-sectional (i.e., hedonic) comparisons
is that the observed differences across areas reflect in part adaptation, but a demerit of this
approach is that such adaptation is only within the scope of currently available technology.
This is plainly an incomplete accounting of adaptation. Putting another way, imagine being
in 1920 and being charged with forecasting the effect of a warmer and wetter climate. A
projection based on the US historical experience up to that point would vastly overstate the
(negative) response by virtue of its failing to account for the subsequent evolution of
technology, which improved productivity in warm and wet areas. On the other hand, an
advertised merit of panel/fixed-effect methods is that they provide an upper bound on the
magnitude of the response to climate change. The argument invokes the so-called Le
Chatelier principle: more adaptation is possible in the long run, so the sensitivity to a
change in weather should lessen at longer horizons. But in both examples that we present,
the short-run response to weather is at least an order of magnitude smaller than the change
in hedonic values over the long run. We show that the relationship of panel/fixed-effect
estimates to long-run elasticities also depends on (a) if cumulative exposure to different
weather also changes the local ecology (be it soil fertility or disease transmission, for
example), and (b) if weather's effects on output are by shifting the production function or
the productivity of investment. (One of the examples is indeed human-capital investment,
and in both examples the effects seem to work in some measure through ecological
changes.) In addition, the panel estimates in both cases are sensitive to the time period
examined, the advance of technology having had a buffering effect on the response of
income to weather.
Both for descriptive purposes and perhaps to better target policy, it is also important
to understand what might be the mechanisms of adjustment that undergird the results. We
provide suggestive evidence on this point in Section 6.2. For both examples, we consider a
variety of hypotheses about how areas adapted over time to warm and wet climate. For
each candidate mechanism, we see if it can account for the changing sensitivity to weather.
Out of possible mechanisms reviewed, malaria emerges as a key mechanism in
4
understanding this result. The adaptation across century in wet and hot areas was more
substantial in higher-risk malarial counties than otherwise. We also find that other
agronomic factors such as drainage, irrigation, selection of crop, crop mix, farm ownership,
fertilizer use, wage in agricultural sector, and farm mortgage rate, and their change over
the century can partly account for the adaptation to weather from the long-term perspective.
2 Background and Related Literature
The effect of climate change on agriculture had been extensively studied as the
worldly concern of global warming was increasing in the 1980s (IPCC 1990). The early
approach of the studies was based on agronomic models of production function or crop
response function (Adams 1989; Adams et al. 1990). This traditional approach models a
function of crop production from empirical or experimental results using parametric, nonparametric or semi-parametric methods; climatic components such as temperature and
precipitation are used as key input variables. Researchers simulate the change in crop yield
by varying the input variables to estimate the impact of climate change on agriculture.
Using this production-function approach, various studies have predicted that U.S.
agriculture would be substantially hurt as a result of global warming. This method is useful
and accurate if the response of crop yield to climate is constant across space and time. But
it can be misleading if the response changes over time as new technologies and economic
conditions reduce the negative effects of drought and extreme temperature on crop
production, i.e. adaptation.
The significance of adaptation was pointed out by Mendelsohn, Nordhaus, and Shaw
(1994). They emphasized the variety of the adaptations and adjustments that farmers
customarily make in response to changing economic and environmental conditions. For
example, adaptive and profit-maximizing farmers would switch the main crop from wheat
to corn and subsequently to grazing as local temperature increases. Without considering
the adaptation and substitution, the production-function approach can overestimate the
agricultural damage from climate change.
As an alternative approach, Mendelsohn, Nordhaus, and Shaw (1994) developed
hedonic price model. Based on David Ricardo’s value theory, the hedonic approach mainly
assumes that the economic value of adaptive activities to climate change would be
5
reflected in land rent. In other words, if farms had enough to adapt to climate change, the
land rent will be equal to the net yield of the highest and best use of the land, i.e. land
productivity. In particular, they measure land rent with farm value, which is the present
value of future rent; by running the regression of farm value on climatic, environmental,
economic and agronomic variables, they estimated the discounted marginal value or price
sensitivity of these factors to land productivity. They found a significantly lower impact of
global warming on U.S. agriculture than did the traditional production-function approach.
Although the newly-developed method and findings have influenced later researches on
climate change, the study does not actually observe the adaptation to climate or dynamic
adjustment because of cross-sectional comparisons in 1982, but rather assumes it.
Other limitations of the hedonic approach have been pointed out by some studies.
Cline (1996) questioned the validity of applying cross-section analysis of present-day land
values to predict future global-warming impacts because relative price of grain would
change due to the global reduction in grain yield. Darwin (1999) argues that the model
produces estimates that violate basic agricultural principles in that it inappropriately
captures the effect of irrigation on farmland value. Regarding this comment, Schlenker,
Hanemann, and Fisher (2005) showed that the economic effects of climate change on
agriculture need to be assessed differently in dryland and irrigated areas, emphasizing the
role of irrigation and water supply in the hedonic approach.
The validity of the hedonic regression was carefully examined by Deschênes and
Greenstone (2007). They recognized that unobserved characteristics are spatially
correlated so that the standard OLS formulas used in the hedonic regression can be
incorrect since the error variance is not spherical. In the case, the hedonic approach can
confound climate with other regional factors that happen to be correlated with the
measures of climate employed. As a possible solution to the omitted variable bias issue,
Deschênes and Greenstone (2007) proposed a fixed-effects model that uses the across-year
variation in temperature and precipitation within county to estimate the effect of climate
change on agricultural profits and yields. More recently, Schlenker and Roberts (2009)
employed a large panel of crop yields and daily weather variables that spans most US
counties from 1950 to 2005. Using fixed-effects model, they estimated temperature
thresholds above which crop yields sharply decline; the findings suggest that global
warming may severely damage US crop yields in the century, which is contrast to the
prediction by Deschênes and Greenstone (2007).
6
Each of those previous studies has its own advantage to understand the effect of
climate change on agriculture. A merit of hedonic approach is that it can account for
adaptive behavior like crop choice and developing new agricultural technologies; the
fixed-effects model can solve the omitted variable bias problem. Nevertheless, the big
unknown in the existing studies is the scope of adaptation. This is not actually observed,
but assumed in the hedonic approach; the studies based on the fixed-effects model is
limited by the time window employed in their research design. Meaning full adaptation
takes place over a long horizon, and therefore direct evidence of adaptation requires
observation over longer periods of time. A merit of this study is that we try to take each
advantage of hedonic and fixed-effects methods. But the most strength of this study is that
we extend the time window to the past 150 years throughout which human has experienced
various range of technological adaptation to the weather.
3 Data
To estimate the historical changes in the economic effects of temperature and
precipitation across US history, this study matches historical weather variables with county
farm productivity measures in Section 4 and with individual income in Section 5. Thus,
three types of variables mainly have been searched for in various historical sources:
weather variables, county-level farm productivity measures, and individual income
variable. This section briefly explains how each variable is constructed or obtained.
Detailed information on data sources and related issues are discussed in Appendix A.
First, the construction of historical county weather variables is based on the
Nineteenth-Century US Climate Data Set Project developed by National Climate Data
Center and the Long-Term Daily and Monthly Climate Records from the Stations across the
Contiguous United States provided by the US Historical Climatology Network. Both
datasets report monthly mean temperature and monthly accumulated precipitation for
thousands of weather stations which existed over the past two centuries. However, the
weather information is not available for all the counties throughout the past two centuries;
the weather variables of the counties without historical weather stations at certain periods
need to be estimated. For the estimation, this study uses a geostatistical technique called
‘Kriging’. As a linear least squares estimation, Kriging algorithms impute the climatic
value at a given (target) location as a weighted sum of data values at surrounding locations.
7
Kriging assigns weights with inverse distance squared among locations; it puts a lower
weight as a surrounding weather station is located in a longer distance from the target
location (Stein 1999). Using this method, we estimate monthly mean temperature and
monthly accumulated precipitation by county and for every month in 1860-2000. Then, we
utilize the monthly estimation results to calculate March-to-October, annual, decadal, or
centurial mean temperature and precipitation at the county- or state-level according to the
research designs in Sections 4 and 5.
Second, this study uses two measures of county farm productivity, which are mainly
found in Historical, Demographic, Economic, and Social Data: The United States, 17902002 (ICPSR #2896): farm value per farmland acre and farm output value per farmland
acre. ‘Farm value’ is defined as the value of all land, housing, and outbuildings in the farm
at the time of census enumeration. Although it is available from the 1850 agricultural
census, we use the variable only in 1870-2000 considering the periods when the quality of
estimated weather variables is acceptable. ‘Farm output value’ is the total value of crop
production and livestock products within the year prior to the enumeration day, i.e. a sort
of annual farm profit. But the variable is not found in the above source for the census years
of 1910, 1920, and 1930. On the other hand, the changes of county boundaries matter
when county fixed effects are considered in Section 5. To partially fix the problem, we
adjust these county variables in 1880-2000 on the 1870 county boundary, using the areaweighted average method.
Third, as a measure of individual income, we use the occupational income score that
is available for a large number of censuses. The variable is an average indicator of labor
earnings by disaggregated occupational categories that were calibrated using data from the
1950 census. Occupation has been measured by the census more than a century, and so the
income proxy is available for a substantial stretch of cohorts. Using the proxy, we
construct a panel of average income by cohort. The units of observation of the panel are
year of birth by state of birth, and we use microsamples from ten censuses (1880 and 19001990). This results in an unbalanced panel spanning the year-of-birth cohorts from 1860 to
1960 for forty six states of birth.
4 Weather and Farm Productivity
4.1 Long-Term Weather Conditions and County Farm Value
8
We first look into the difference in farm value across areas by local weather variables
measured as decadal average. We particularly look at the historical pattern of the
relationships between the county average of farm value and its weather conditions, using
scatter plots. This cross-sectional analysis is intended to see how county farm value has
adapted to permanent features of local weather over time.
For the six selected decades, the upper panel of Figure 1 plots the logarithm of each
county’s average farm value per acre against its decadal average of annual mean
temperature in Fahrenheit, which is calculated for the period 10 years prior to each census
year. We also put a linear trend line and a lowess fit curve on each decade’s scatter plot.
The bivariate diagrams suggest that the relevance of farm value to temperature gradually
changed over decade. In terms of linear relationship, a high level of temperature depressed
farm value in the second half of the nineteenth century; this negative association had been
weakened throughout the early twentieth century; in recent decades, farm value has been
(weakly) increasing in temperature. In addition, a quadratic relationship is clearly found in
the late nineteenth century, whose threshold level is observed around 50 oF. However, this
inverse U shape relationship had attenuated over time, and it looks insignificant in recent
decades.
[Figure 1 Here]
A similar historical pattern is observed between county farm value and decadal
average of annual accumulated precipitation, which is presented in the lower panel of
Figure 1. Most of all, an inverse U shape relationship is observed until the early twentieth
century; for the counties whose decadal average of annual accumulated precipitation was
below about 35 inches, their farm value increased in precipitation, but a high level of
precipitation above the threshold level depressed farm value. The linear pattern indicates
that there was a negative correlation between county farm value and precipitation in the
late nineteenth century, and the negative slope became flat and insignificant throughout the
early twentieth century. The attenuation of the negative relationship and inverse U shape
relationship accelerated toward the late twentieth century; county farm value has been
strongly increasing in the level of precipitation in recent decades.
This bivariate cross-sectional analysis shows that the adaptation to hot and wet
9
weather can be clearly observed if we look at the adaptation at longer time horizons and
that the pattern of the adaptation has occurred very gradually. However, this hedonic
approach does not say about whether this adaptation to permanent weather conditions
resulted from the change in the climate effect or from the change in the effect of other
local‒agricultural, socioeconomic, or demographic‒characteristics that happen to be
correlated with weather variables.
4.2 Fixed-Effects Model
As a complementary approach, we use a fixed-effect model in this subsection and
compare it with the cross-sectional result. Estimating the following equation (1) estimates
the historical pattern of the effect of long-term weather on farm value controlling for stateand year-specific improvements in agricultural technologies and time-invariant county
factors.1
(1) Yijt  Wijt  X ijt  
  DW
2000
t 1880
t
t
ijt
 Dt X ijt  t    t   jt   i   ijt , i : county, j : state, t : year
where the Yijt is the logarithm of county i’s average farm value per acre measured at year t,
the Wijt is the decadal average of each weather variable (temperature or precipitation) in
county i, the Xijt are some vectors of county-level agricultural controls and constant terms,
the Dt are dummies indicating year t, the δt are year fixed effects, the δjt are state-by-year
fixed effects, and the δi are county fixed effects.2 As the county-level controls, we use
county population density, the ratio of white population to county population, the ratio of
farmland to total available county area, and the estimated number of farmers per farmland
acre. (Sources and construction of each control are discussed in Appendix A.2.) Equation
(1) considers the impact of across-decade variation in weather within a county, and it only
focuses on the linear relation between county farm value and weather variables. In the
estimation model, the reference year is 1870. Thus, the coefficients βt estimate how much
1
Agricultural control variables that we can use in the regression model are very limited because those that
appear in some census years are frequently unavailable in other census years. The use of fixed effects models
partially solves this limitation.
2
All of the estimates of this equation below are calculated using ordinary least squares (OLS) regressions.
10
the response of farm value at year t to local temperature or precipitation differed from its
level in 1870.
Each panel in Figure 2 graphically presents the coefficient βt for temperature and
precipitation, respectively. Dotted lines are the results of estimating equation (1) only
including year fixed effects; dashed lines are the results that additionally control for stateby-year fixed effects; solid lines present the estimation results that also use county fixed
effects.3
[Figure 2 Here]
Figure 2 shows that the effect of variation in long-term temperature and rainfall on
farm value rapidly changed until the mid-twentieth century. The over-time increase in βt
during the period implies that a high level of temperature and rainfall became more
beneficial to farm value as time goes on. Then βt remains stable after the period of 19601970, which suggests that the adaptation to hot and wet weather slowed down in recent
decades. The estimated coefficients depend on the use of fixed effects. County fixed
effects estimate smaller effects of across-decade variation in weather on farm value
relative to the level of 1870 than do year or/and state-by-year fixed effects. Accordingly a
slower adaption to hot and wet weather precipitation is estimated when county fixed
effects are used. This means that the effect of weather on farm value is largely explained
by non-weather unobservable county characteristics correlated with county weather
conditions.4
4.3 Measuring the Adaptation to the Weather across Century
So far we tried to measure the historical pattern of the adaptation to hot and wet
weather conditions using panel and fixed-effect models. The key findings are that the
effect of hot and wet weather on farm value substantially changed over the two centuries,
3
All the coefficients shown in Figure 1 are statistically significant at the 99% level of confidence. See Table
B.1 for the results using county fixed effects.
4
In the panel/fixed-effect analysis, we use county-boundary-adjusted data as discussed in section 3. In
Appendix B.1, we compare the result in this section with the estimation result based on data without the
adjustment, and find that the use of county-boundary-adjusted data does not cause critical problems in
estimating fixed-effect models.
11
and that the turnover of the historical pattern mainly took place throughout the first half of
the twentieth century. Thus, if we compare the effect of weather between the beginning
(the late nineteen century) and ending (the late twentieth century) periods of the sample
time window, the magnitude of the adaptation can be more clearly estimated. Moreover,
this comparison also will be useful in our seeking possible mechanisms of the adaptation
in Section 6.5
The model of measuring the adaptation to the weather across centuries is as follows.
(2) Yijt  Wijt  Wijt D19  X ijt   X ijt D19    t   jt   i   ijt , i : county, j : state, t : year .
For the estimation, we select two representative periods: the decades ending with 1870,
1880, 1890 and 1900, and the decades ending with 1970, 1980, 1990 and 2000.
Accordingly D19 denotes the dummies indicating the decades in the late nineteenth century.
The definitions of the other controls and fixed effects are the same as those in equation (1).
Then, the coefficient β in equation (2) measures the across-century difference in the
response of farm value to variation in long-term weather conditions within a county (say,
the across-century weather sensitivity).
Panels A and B of Table 1 reports the coefficient of the across-century weather
sensitivity and its standard error, clustered on county, by specifications. The results show
that high temperature or rainfall was significantly less favorable to farm value in the late
nineteenth century than it was in the late twentieth century. Compared with model (1) that
is cross-sectional, the fixed-effect models---especially model (4) that uses county fixed
effects---estimate smaller coefficients of the across-century weather sensitivity, but still
statistically significant. Again this means that the adaptation of farm value to long-term
weather conditions is partly explained by other local and year-specific factors correlated
with weather conditions.
[Table 1 Here]
On the other hand, the two weather variables are frequently correlated together. For
5
In the section, we will test some possible mechanisms using various kinds of historical variables. But many
variables are available only for the late nineteenth and late twentieth centuries.
12
example, southern counties’ climate is characterized not only by high temperature, but also
by wet or humid weather. In the case, the impact of one weather component on farm value
can be compounded by another. We consider this issue by running a regression of the
logarithm of average county farm value on each weather variable (say, T and P denote
temperature and precipitation, respectively), the demeaned interaction of the two weather
variables ([T−μT]×[P−μP], where μ denotes the sample mean of each variable), and the
interaction terms of these three variables with the dummy of the late nineteenth century
together as follows.
(3) Yijt   T Tijt   P Pijt   TP (Tijt  T )( Pijt   P )  [  T Tijt   P Pijt   TP (Tijt  T )( Pijt   P )]D19
 X ijt   X ijt D19    t   jt   i   ijt , i : county, j : state, t : year.
The coefficients of the interaction terms with D19 are reported in panel C of Table 1.
In particular, the coefficient βTP shows that the across-century sensitivity to hot weather
was more substantial when the counties were under wet weather conditions, and vice-versa.
When the temperature-precipitation interaction effects are considered, the coefficient of
each weather variable interacted with D19 becomes slightly smaller, but its significance and
the implication remain unchanged.
To have a better sense of what the estimated coefficients imply, we suppose two
counties with different levels of annual mean temperature: 54.9 oF for the average county
and 62.6 oF for a hot county, which is one-standard-deviation higher than the mean
temperature (see Table A.1 in Appendix A for summary statistics of weather variables). We
also assume that the levels of precipitation and other agricultural conditions are identical
between both counties and across centuries. Then, using the coefficient of T×D19 in model
(3) of panel C, the across-century difference in farm value as a ratio (i.e. the level of farm
value in the late twentieth century relative to that in the late nineteenth century) is
estimated 11.01 for the average county and 15.42 for the hot county.6 Because there is
little difference in county farm value by local temperature today, this simulation suggests
6
Say that FV19C and FV20C denote average farm value in a county in the late twentieth and the late twentieth
century, respectively. Accordingly, the cross-century difference in farm value as a ratio (FV20C÷FV19C) in a
county with specific long-term weather condition is measured by exp(−βT·T) regarding temperature(T) or
exp(−βP·P) regarding precipitation(P). Considering the interaction between temperature and precipitation as
discussed in text increase the estimated magnitude of the historical changes by a small amount.
13
that the adaptation more rapidly occurred in hotter areas over time. Similarly, using the
coefficient of P×D19 in model (3) of panel C, the cross-century difference in farm value as
a ratio is estimated 1.60 for a county with average annual precipitation (37.9 inches) and
1.87 for a wet county with 50.6-inch annual precipitation, which is one-standard-deviation
higher than the mean precipitation. The rate of the adaptation seems to be smaller than the
case of temperature. But the adaptation to precipitation more substantially took place in
wetter areas.
4.4 Short-Term Weather Fluctuations and Farm Output Value
So far we have looked at the adaptation to weather using county farm value and longterm weather variables. A merit of using farm value is that the variable is consistently
available throughout the sample period. But weather conditions affect farm value through
various indirect mechanisms rather than directly. Although we have controlled for those
possibilities using some measured local variables and fixed effects, those controls still
might be not enough to overcome the weakness implicit in farm value and to identify the
effect of weather on farm productivity and its historical pattern.
In this subsection, we employ an alternative measure of farm productivity: county
farm output value. This is a better measure in that the growth of crop is directly affected by
local weather conditions.7 But farm output value is not available in the census years of
1910, 1920 and 1930. This allows us to use only the panel regression of equations (2) or
(3) used in the previous subsection, which measures the across-century weather sensitivity.
Additionally the level of agricultural output more depends on short-term weather
conditions rather than decadal climate. We particularly measure the short-term weather
variables as average mean temperature and accumulated precipitation between March and
October in each census year, assuming that the March-to-October weather conditions are
more essential for the growth of crop and other agricultural activities than are those of
other months.
Per equation (3), panel A of Table 2 reports the coefficients of the interaction between
nineteenth-century dummies and each weather variable in the census year, and the
7
Farm output value has been used in various recent studies such as Deschênes and Greenstone (2007) and
Schlenker and Roberts (2009). But those studies utilize daily temperature variables in recent decades, which
are not available for historical periods.
14
temperature-precipitation interaction. We find that hot and wet weather was less favorable
for producing agricultural goods in the late nineteenth century than in the late twentieth
century, and that the adaptation of farm output value to weather rapidly occurred across
centuries. To check out whether this result is correlated with the effect of weather in other
years around the census year, we add short-term weather variables measured in the periods
one year after and before the census year in equation (3), and report the key estimated
coefficients in panel B. The across-century weather sensitivity is strongly estimated for the
current-year weather variables; the estimated effect of weather in adjacent years is
insignificant and negligible. Moreover, the size of the estimated coefficients of the currentyear weather variables in model (3) are very similar with what we estimated using farm
value and long-term weather variables in the previous subsection. Finally, the panel
regression model (4) with county fixed effects estimates smaller across-century weather
sensitivity than does the cross-sectional regression model (1), which implies that the
weather sensitivity is partly explained by unobservable county characteristics correlated
with county weather conditions.
[Table 2 Here]
This section is summarized as follows. First, the adaptation to weather has taken place
over longer periods of time and can be observed from a longer-term perspective. Second,
the adaptation took place more substantially in hot and wet areas. Third, the effect of
weather on farm productivity and its historical pattern is significantly estimated in terms of
farm value and farm output value, using fixed-effect models. In Section 6, we will seek
and evaluate possible mechanisms that can account for the historical pattern of the
adaptation to weather.
5 Early-Life Weather and Adult Income
A large number of studies have revealed that climate and its change directly or
indirectly affect individuals’ activities and life-time health through causing malnutrition
from crop failure (Ó Oráda 2007), spreading infectious diseases (Lafferty 2009), distorting
health conditions over the course of life (WHO, WMO and UNEP 2003; Patz et al. 2005;
Campbell-Lendrum and Woodruff 2006), changing ecological system (Thomas et al. 2003),
15
taking lives due to heat and natural disasters (Davis et al. 2003; Kahn 2005), decreasing
water supply (Vörösmarty et al. 2000), resulting in migration (Brown 2008), affecting
economic sectors, technological innovation and energy use under greenhouse gas emission
regulation (Nordhaus and Boyer 2003; Stern 2007; Tol 2009), and so on. Although some
studies predict that global warming can increase crop production in some areas (Deschênes
and Greenstone 2007) and reduce cold weather-related mortality (Kalkstein and Greene
1997), most studies predict that the overall impact of climate change would be adverse to
human life. From economic aspect, the above channels are closely related to reduction in
individuals’ economic productivity.8 On the other hand, existing literatures have shown
more concerns over developing countries because they have a large proportion of marginal
populations who are much vulnerable to the impact of hot and wet weather. While, the
adverse link has been mitigated or prevented in many developed countries with the help of
effective technologies and policies. From historical perspective, we thus expect that the US
relation between weather and individuals’ economics productivity would have changed
over longer-time horizons along with technological adaptation to weather.
Recently, Deschênes, Greenstone, and Guryan (2009) show that extreme temperature
in utero can increase the risk of having low-birth-weight babies (also see Murray et al.
2000; Lawlor, Leon, and Smith 2005). Although further evidence is required, they point
out that this can be caused by the association between weather in utero and fetal nutrient
intake or stress. The relation between weather in utero and birth outcome is significant in
that adverse birth outcome can affect life-time human capital accumulation such as
schooling, disability and adult income on the basis of fetal origin hypothesis, which has
been supported by many existing studies. Those studies include Barker (1994), Almond
(2006), and Black, Devereux, and Salvanes (2007). In addition, it has been much studied
that maternal exposure to infectious diseases promoted by weather---notably malaria--increases the probability of having low-birth-weight babies (Desai et al. 2007), and that
exposure to the disease environment can reduce cognitive ability and so the acquisition of
human capital (Holding and Snow 2001; Sachs and Malaney 2002).
5.1 Cross-Sectional Evidence
8
For example, the adverse health impacts of hot and wet climate can deteriorate individuals’ productivity at
work and so income. Changes in economic sectors due to climate change can affect industrial structure and
labor market and so their wage.
16
In consideration of this mechanism, our study focuses on the significance of early-life
weather in later economic outcome and how it has been changed throughout the US history.
We first look for cross-sectional evidence that shows the effect of weather on adult income
from a longer-term perspective, using simple bivariate analysis. The variable of adult
income used here is average income by state-of-birth cohorts. Using the county weather
variables imputed by Kriging interpolation technique, we construct the 1860-2000 average
of annual mean temperature and annual accumulated precipitation by state. Therefore, the
weather variables represent the overall weather conditions which each cohort might have
experienced in early life rather than at a point of time.
In Figure 3, we plot average adult income against long-term average weather
variables in the state of birth, by two cohort groups: 1860-1900 and 1930-1960. The figure
shows that adult income of the cohorts born before 1900 decreases by temperature and
precipitation in the state of birth. In other words, if people were born or had spent their
early periods in hotter and wetter states, they earned less income in late life on average
than otherwise. But this relation almost completely disappears for the cohorts born after
1930; there is little difference in adult income by early-life average weather conditions.
[Figure 3 Here]
The bivariate analysis has various limitations for interpretation. Socioeconomic
conditions of the state of birth are not controlled for, and so the significance of this pattern
cannot be asserted. Even if it is statistically significant, the scatter plots do not provide
explanations for the historical pattern. Further investigation is thus required. Nevertheless,
the figure may suggest a possibility that technological adaptation took place throughout the
early twentieth century, which mitigated the adverse impacts of high temperature and
rainfall on individuals’ economic productivity
5.2 The Diminishing Effect of Short-Term Weather Fluctuations
Emphasizing the historical change of the relation between long-term weather in early
life and adult income, we turn our attention to the effect of short-term weather fluctuations
around the year of birth on adult income, using fixed-effects models. The use of long-term
17
average weather not only captures the effect of weather around birth, but also likely
measures the accumulated or life-time effect of weather if a large proportion of each stateof-birth cohort had lived in the same state during considerable time in their life. The use of
short-term weather variables can better identify the effect of early-life weather on adult
income.
In the following estimation model, we search for the specific point of time around
birth whose weather conditions have significant effects on adult income by running a
regression of cohort average occupational income score on annual mean temperature and
accumulated precipitation at the year of birth, at the three years after birth, and at two
years before birth, per equation (4) below. The regression is conducted for cohort samples
defined by state of birth and year of birth in 1860-1960. The model also contains dummies
for each cell of state of birth times year of birth (  jk ), dummies for each cell of year of
birth times census year (  kt ), and dummies for each cell of sate of birth times census year
(  jt ). These fixed effects will partially capture omitted and unmeasured characteristics in
state of birth, year of birth and census year that might commonly affect those in a given
cohort.
3
(4) Y jkt     (  l T j ,k l   l Pj ,k l )   jk   kt   jt   jkt ,
l  2
j : state of birth, k : year of birth, t : census year.
We summarize the key results of the regression per equation (4) in Figure 4. The y
axis in each plot displays the estimated regression coefficient on the interaction of the
indicated weather variable and a dummy for the calendar year minus year of birth, i.e.  l
or  l ; the x axis in each plot denotes the calendar year minus the year of birth. The error
bars reflect 95% confidence intervals for each coefficient.
[Figure 4 Here]
The main finding is that adult income has been highly associated with across-year
weather fluctuations for specific years. Weather conditions before birth do not have
meaningful effects. The year in which temperature has the most substantial, significant
18
effect on adult income is one year after birth, and the effect declines over next two years.
The effect of precipitation is most substantial at the year of birth, and then it becomes
insignificant.
These results for short-term weather have implications different with what we find
from the cross-sectional analysis in the previous subsection that suggests the negative
correlation between long-term weather conditions and adult income. The state-of-birth
fixed effects average out state-specific climatic characteristics. Thus, the above estimation
measures the effect of within-state across-year variation in weather conditions around birth.
Although many literatures suggest the negative effect of hot and wet weather in utero on
birth outcome as discussed above, some studies have found a positive association between
hot/wet weather in early life and condition at birth. Especially, seasonal difference in birth
outcome has been estimated in historical periods (Beardsley 1987; Costa 2004) and for
developing countries (Ceesay et al. 1997). For example, Costa (2004) shows that babies
born in the spring weighted less than babies born in the summer using birth records on
white babies born at Johns Hopkins in the early twentieth century, and suggests that this
might result from the nutritional stress their mothers experienced during the winter months.
From the perspective of fetal origin hypothesis, the positive effect of temperature on birth
outcome can lead to a higher level of human capital accumulation and so adult income.
Similarly, some studies find that rainfall in early life can be beneficial for later outcomes
such as health, schooling, and socioeconomic status (Maccini and Yang 2009).
Following this line of research, it is interesting that the meaningful association among
weather, nutrition in utero, birth outcome, and later outcome is frequently found in studies
on historical experiences and developing countries; the association is probably mitigated in
modern developed countries by enough food supply throughout year, public programs and
technological adaptation. This suggests that the coefficients presented in Figure 4 can vary
across year. To check out this possibility, we now look at how the effect of short-term
weather fluctuations on adult income has changed over cohort.
We measure this long-term pattern by estimating equation (4) on a 40-year-wide
moving window. So the sample for each regression includes cohorts born 20 years before
and after the target year. Consequently the target years are between 1880 and 1940. Figure
5 presents the estimated regression coefficient on the interaction of the indicated weather
variable and a dummy for the year chosen according to the results in Figure 4 (i.e. one year
after the year of birth for temperature and the year of birth for precipitation). We also add
19
the 95% confidence intervals for each coefficient.
[Figure 5 Here]
We mainly find that the effect of short-term weather fluctuations had diminished over
cohorts. As it is estimated above, the effect is generally positive among the cohorts born in
the second half of the nineteenth century and its across-cohort variation is large. But as the
target year approaches to 1910-1920, the magnitude, significance and across-cohort
variation of the effect had attenuated. After then, adult income became little relevant to
weather in early life. The pattern of attenuation over time in Figure 5 is similar with what
we observe in the cross-sectional analysis using long-term weather variables, although the
role of weather in early life is differently estimated between two analyses. Moreover, it is
worthy to remind that the diminishing effect of weather was similarly found in the analysis
of farm productivity in Section 4. The findings so far suggest that the adaptation to
weather largely occurred throughout the early twentieth century not only in agricultural
sector but also at the individual level. In the following section, we discuss theoretical
explanation and possible mechanisms through which the economic implications of weather
have changed across centuries.
6 Discussion
* Note that the following section is very incomplete.
6.1 Theory
[To be added]
6.2 Mechanisms
This section revisits the adaptation of farm productivity to weather in Section 4 to
discuss its possible mechanisms. The results in Section 4 suggest that the level of
agricultural technology of adapting to hot and wet weather in the nineteenth century was
inferior to that of modern periods, which led to a lower level of farm value and farm output
value in hot and wet areas. Explanations of this historical pattern can be found from
20
various kinds of technological adaptations, which have been advanced throughout the
twentieth century. The variables of mechanisms and their statistics are summarized in
Table 3.
First, historically drainage was the key technology to improve the productivity of
farmland by preventing the damage from heavy rainfalls and frequent flooding, increasing
areas for cultivation, changing the hydrology of the soil system, and reducing erosion. As
population pressure and a demand for better farmland increased, drainage practice and
wetland conversion had been expanded to the East in the colonial periods and early-19th
century, to the Midwest and Mississippi river valleys in the mid-19th century, and to the
South and West in the late-19th century. But the scale and effectiveness of drainage efforts
in local areas depended not only on size of local population and demand for arable land,
but also on the level of federal and state governments’ policy and supports, costs of land
reclamation, and drainage technology. In particular, a substantial drainage practice in
southern states had been much delayed until the early 20th century when more advanced
and low-cost technology was introduced. This implies that less drained areas in the
nineteenth century would more sensitive to heavy rains and
Therefore, because wetlands and swamps were generally fertile for agriculture, the
extension of drainage was key to they were little developed primarily due to lack of
drainage technology and great costs of reclamation work in the colonial periods. Until the
early twentieth century, a substantial drainage practice Over the course of the nineteenth
century, population pressure and a need for better farmland demanded efforts for
converting wetlands to farmlands (Pavelis 1987). In particular, as immigrants began to
settle in the Midwestern states and Mississippi river valleys whose large proportions were
swampland unsuited to cultivation, a substantial drainage practice became prevalent in
those areas in the 1850s. But a large scale of wetland conversion in the Midwest occurred
since the 1870s when new drainage technology of using steam power and drainage tiles
were introduced. After the Civil War in 1861-1865, western ward expansion took the
nation’s attention, and the railroads opened new lands in the West and South including
wetlands. As agriculture moved A broad scale of drainage in southern states occurred in the
early twentieth century in response to federal and local governments’ financial supports
and improvements in drainage technology.
Drainage was also a key factor that improved local sanitary and ecological conditions.
Because wetlands and swamps were the main source of water-borne diseases such as
21
malaria, yellow fever and cholera, draining those areas could make local communities
healthy. However, drainage with sanitary intention
[more discussion will be added here] …
For instance, the technological improvement for irrigation is one of the notable
technological adaptations to overcome high temperature (Mendelsohn and Dinar 2003).
Drainage has been utilized to prevent the damage from flooding and to increase areas for
cultivation (Pavelis 1987). The selection of crops, crop mix or inventing hybrid seeds more
resistant to hot and wet weather is another key technique to adapt to the change of local
climate (Olmstead and Rhode 2008). The change of farm ownership and the use of
fertilizer are also important for increasing farm productivity, although they are less directly
related to weather conditions than drainage or irrigation (Gallup and Sachs 2000; Garrett
and Xu 2003). The supply and cost of farm labor and farm land mortgage can be correlated
with local climate conditions, and this can affect farm productivity (Bodenhorn and
Rockoff 1992; Engerman and Goldin 1993).
Besides these mechanical and structural advances of farming, farm productivity
largely depends on farm laborers’ health status and their labor productivity. This is more
likely in the periods when disease environments were rampant and medical knowledge for
disease controls was less accumulated. Thus the eradication of infectious diseases that
deteriorated farm workers’ labor productivity can be a good explanation of the historical
pattern of the adaptation across the centuries. We pay a special attention to the significance
of malaria and its eradication that occurred throughout the first half of the twentieth
century.
Besides malaria eradication, other technological improvements in the first half of the
twentieth century could mitigate the negative effects of hot and wet weather on farm
productivity, and account for the historical pattern. The significance of malaria
environments estimated above can be robust by examining the role of other technological
improvements. From the respect, we consider various aspects of agricultural technologies
related with weather such as drainage, irrigation, farm improvement for crop land, crop
mix, selection of crop, farm organization, fertilizer use, and the cost of farm labor and farm
land mortgage. We have searched for proxy variables from various historical resources,
which measure the levels of these agricultural technologies. Each type of technology above
is measured as follows: (1) the ratio of drained and (2) irrigated areas out of total county
areas, (3) the ratio of cropland out of total available farm land, (4) the ratio of cotton areas
22
in farm land, (5) the level of crop concentration measured by Herfindahl-Hirschman index
for the 10 major US crops, (6) the ratio of share croppers and (7) tenants, (8) the fertilizer
cost per farmland acres, (9) annual agricultural wage per farmland acres, and (10) interest
paid on farms per mortgage value of farms. (Sources and constructions of each variable are
discussed in Appendix A.5.)
We examine the role of these technologies by revising the set of K in equation (5) to
K={T, P, M, F, T×P, T×M, P×M, T×F, P×F }. The new variable F denotes each county’s
agricultural factor introduced above, and T×F and P×F are the interaction term between
agricultural factor and (long-term or short-term) temperature or precipitation, respectively,
depending on the type of dependent variable. We control for each agricultural factor as the
value at the ‘pre-existing’ level, which is found in the earlier periods (mostly in the late
nineteenth century) when significant improvements were not yet made. Accordingly, the
coefficients βTFD19 and βPFD19 measure whether the across-century weather sensitivity
of farm productivity depended on the pre-existing level of each agricultural factor.
Tables 4.1 and 4.2 present the estimated coefficients and standard errors clustered on
county of key variables. According to panel A in the tables, the across-century difference in
the response of farm value to long-term hot weather was significantly greater as counties
are characterized by less irrigated areas, less cropland ratio, less crop concentration, less
share-cropper ratio, more tenant ratio, and lower mortgage rate in the earlier periods. The
across-century sensitivity to wet weather was significantly substantial in the counties with
less drained areas, less cropland ratio, less crop concentration, less fertilizer cost, and
lower agricultural wage. Although all the estimated coefficients and their implications are
not consistent with historical fact and economic intuition, it is generally estimated that the
counties where advanced agricultural technologies were not yet introduced experienced the
negative effect of hot and wet weather on farm value at a greater level than otherwise.
Technological improvements and their introduction to hot and wet areas well account for
the historical pattern of the relationship between farm value and weather conditions.
[Table 4.1 Here]
[Table 4.2 Here]
Another key finding is that adding the pre-existing agricultural factors in the
regression model does not hurt the significance of malaria in general. Compared to the
23
results in model (2) of Table 3, the coefficients of malaria (T×Malaria×D19 and
P×Malaria×D19) in Tables 4.1 and 4.2 are still significantly negative and substantial across
the regression models. Moreover, the coefficients of weather variables (T×D19 and P×D19)
do not change much by adding the pre-existing agricultural factors. Consequently, malaria
eradication, a particular technological adaptation, is thought as a considerable explanation
for the historical pattern of the relation between farm value and long-term weather
conditions.
On the other hands, panel B of Tables 4.1 and 4.2 reports the main results that use
county farm output value as the dependent variable and the March-to-October average
weather variables in the census year, using the same format with Table 4. The general
implication is that the counties that had insufficient levels of agricultural technologies in
terms of drainage, crop land, crop mix, farm ownership, fertilizer use, farm labor wage,
and mortgage rate experienced the negative effect of hot and wet weather on farm output
value at a greater level than otherwise. The role of malaria is still significantly estimated
especially for precipitation even adding agricultural factors in the model.
So far we have looked into two aspects of technological adaptations that primarily
occurred in the first half of the twentieth century in order to seek possible mechanisms
through which the adaptation to hot and wet weather substantially took place. This article
paid a special attention to the role of malaria and its eradication in the first part of this
section. However, we also showed that various agricultural factors and their changes are
closely related with the historical pattern in the relationship between farm productivity and
weather. We more discuss the role of agricultural technologies in Appendix B using
alternative measures of technological adaptations such as the selection of crops and the use
of changes in each agricultural factor over the centuries rather than the use of pre-existing
level.
7 Conclusion
[To be added]
24
Reference
Adams, Richard M. (1989), “Global Change and Agriculture: An Economic Perspective,”
American Journal of Agricultural Economics 71(5): 1272-9.
Adams, Richard M., C. Rosenzweig, R. Pearl, J. Ritchie, B. McCarl, D. Glyer, B. Curry, J.
Jones, K. Boote, and H. Allen (1990), “Global Climate Change and U.S. Agriculture,”
Nature 345(6272): 219-24.
Almond, Douglas (2006), “Is the 1918 Influenza Pandemic Over? Long-Term Effects of In
Utero Influenza Exposure in the Post-1940 U.S. Population,” Journal of Political
Economy 114(4): 672-712.
Barker, David J. P. (1994), Mothers, Babies, and Disease in Later Life, London: BMJ
Publishing Group.
Beardsley, Edward H. (1987), A History of Neglect: Health Care for Blacks and Mill
Workers in the Twentieth Century South, Knoxville, TN: University of Tennessee
Press.
Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes (2007), “From the Cradle to the
Labor Market? The Effect of Birth Weight on Adult Outcomes,” Quarterly Journal of
Economics 122(1): 409-39.
Bleakley, Hoyt (2010), “Malaria Eradication in the Americas: A Retrospective Analysis of
Childhood Exposure,” American Economic Journal: Applied 2(2):1-45.
Bodenhorn, Howard, and Hugh Rockoff (1992), “Regional Interest Rates in Antebellum
America,” in Strategic Factors in Nineteenth Century American Economic History: A
Volume to Honor Robert W. Fogel edited by Claudia Goldin and Hugh Rockoff,
Chicago IL: University of Chicago Press.
Brown, Oli (2008), Migration and Climate Change, Geneva: International Organization
for Migration.
Campbell-Lendrum, D. and R. Woodruff (2006), “Comparative Risk Assessment of the
Burden of Disease from Climate Change,” Environmental Health Perspectives
25
114(12): 1935-41.
Ceesay, Sana M., A. M. Prentice, T. J. Cole, F. Foord, L. T. Weaver, E. Poskitt, R.
Whitehead (1997), “Effects on Birth Weight and Perinatal Mortality of Maternal
Dietary Supplements in Rural Gambia: 5 year Randomised Controlled Trial,” BMJ
315: 786-90.
Costa, Dora L. (2004), “Race and Pregnancy Outcomes in the Twentieth Century: A LongTerm Comparison,” Journal of Economic History 64(4): 1056-86.
Darwin, Roy (1999), “The Impact of Global Warming on Agriculture: A Ricardian
Analysis: Comment.” American Economic Review 89(4): 1049-52.
Davis, R. E., P. C. Knappenberger, P. J. Michaels, and W. M. Novicoff (2003), “Changing
Heat-Related Mortality in the United States,” Environmental Health Perspectives
111(14): 1712-18.
Dell, Melissa, Benjamin F. Jones, and Benjamin A. Olken (2008), “Climate Change and.
Economic Growth: Evidence from the Last. Half Century,” NBER Working Paper No.
14132.
Desai, Meghna, F. Kuile, F. Nosten, R. McGready, K. Asamoa, B. Brabin, R. D. Newman
(2007), “Epidemiology and Burden of Malaria in Pregnancy,” Lancet Infectious
Diseases 7(2): 93-104.
Deschênes, Olivier, and Michael Greenstone (2007), “The Economic Impacts of Climate
Change: Evidence from Agricultural Output and Random Fluctuations in Weather,”
American Economic Review 97(1): 354-85.
Deschênes, Olivier, Michael Greenstone, and Jonathan Guryan (2009), “Climate Change
and Birth Weight,” American Economic Review 99(2): 211-17.
Diamond, Jered (1997), Guns, Germs, and Steel: The Fates of Human Societies, W.W.
Northern and Company.
Engerman, Stanley, and Claudia Goldin (2003), “Seasonality in Nineteenth-Century Labor
Markets,” in American Economic Development in Historical Perspective edited by
26
Thomas Weiss and Donald Schaefer, Stanford CA: Stanford University Press: 99-126.
Gallup, John L., and Jeffrey D. Sachs (2000), “Agriculture, Climate, and Technology: Why
Are the Tropics Falling behind?” American Journal of Agricultural Economics 82(3):
731-7.
Garrett, Martin A., and Zhenhui Xu (2003), “The Efficiency of Sharecropping: Evidence
from the Postbellum South,” Southern Economic Journal 69(3): 578-95.
Guiteras, Raymond (2009), “The Impact of Climate Change on Indian Agriculture,”
mimeo, University of Maryland.
Holding, P. A. and R. W. Snow (2001), “Impact of Plasmodium Falciparum Malaria on
Performance and Learning: Review of the Evidence,” American Journal of Tropical
Medicine and Hygiene 64(1-2 Supple): 68-75.
Hong, Sok Chul (2007a), “The Burden of Early Exposure to Malaria in the United States,
1850-1860: Malnutrition and Immune Disorders.” Journal of Economic History
67(4): 1001-35.
____________________ (2007b), The Health and Economic Burdens of Malaria: The
American Case, Ph.D. Dissertation, University of Chicago.
____________________ (2011), “Malaria and Economic Productivity: A Longitudinal
Analysis of the American Case,” Journal of Economic History, Forthcoming.
Humphreys, Margaret (2001), Malaria: Poverty, Race, and Public Health in the United
States, Baltimore: The Johns Hopkins University Press.
Intergovernmental Panel on Climate Change (1990), Climate Change: The IPCC Scientific
Assessment (J.T. Houghton, G.J. Jenkins, and J.J. Ephraums, eds.), New York:
Cambridge University Press.
Kahn, Matthew E. (2005), “The Death Toll from Natural Disasters: The Role of Income,
Geography, and Institutions,” Review of Economics and Statistics 87(2): 271-84.
Kalkstein, L. S., and J. S. Greene (1997), “An Evaluation of Climate/Mortality
Relationships in Large U.S. Cities and the Possible Impacts of a Climate Change,”
27
Environmental Health Perspectives 105(1): 84-93.
Lafferty, Kevin D. (2009), “The Ecology of Climate Change and Infectious Diseases,”
Ecology 90(4): 888-900.
Lawlor, Debbie A., D. A. Leon, and G. D. Smith (2005), “The Association of Ambient
Outdoor Temperature throughout Pregnancy and Offspring Birthweight: Findings
from the Aberdeen Children of the 1950s Cohort,” BJOG: An International Journal of
Obstetrics & Gynaecology 112(5): 647-57.
Maccini, Sharon, and Dean Yang (2009), “Under the Weather: Health, Schooling, and
Economic Consequences of Early-Life Rainfall,” American Economic Review 99(3):
1006-26.
Mendelsohn, Robert, and Ariel Dinar (2003), “Climate, Water, and Agriculture,” Land
Economics 79(3): 328-341.
Mendelsohn, Robert, William D. Nordhaus, and Daigee Shaw (1994), “The Impact of
Global Warming on Agriculture: A Ricardian Analysis,” American Economic Review
84(4): 753-71.
Murray, L. J., D. P. O’Reilly, N. Betts, C. C. Patterson, G. Davey Smith, and A. E. Evans
(2000), “Season and Outdoor Ambient Temperature: Effects on Birth Weight,”
Obstetrics and Gynecology 96(5): 689-95.
Nordhaus, William D., and Joseph Boyer (2003), Warming the World: Economic Models of
Global Warming, MIT Press.
Olmstead, Alan L. and Paul W. Rhode (2008), Creating Abundance: Biological Innovation
and American Agricultural Development, New York: Cambridge University Press.
Ó Oráda, Cormac (2007), “Making Famine History,” Journal of Economic Literature
45(1): 5-38.
Patz, J. A., D. Campbell-Lendrum, T. Holloway, and J. A. Foley (2005), “Impact of
Regional Climate Change on Human Health,” Nature 438(17): 310-17.
Pavelis, George A. (1987), Farm Drainage in the United States: History, Status, and
28
Prospects, Economic Research Service, U.S. Department of Agriculture.
Sachs, Jeffrey, and Pia Malaney (2002), “The Economic and Social Burden of Malaria,”
Nature 415: 680-5.
Schlenker, Wolfram, and Michael J. Roberts (2009), “Nonlinear Temperature Effects
Indicate Severe Damages to U.S. Crop Yields under Climate Change,” Proceedings of
the National Academy of Sciences of the United States of America 106(37): 15594-98.
Schlenker, Wolfram, W. Michael Hanemann, and Anthony C. Fisher (2005), “Will U.S.
Agriculture Really Benefit from Global Warming?: Accounting for Irrigation in the
Hedonic Approach,” American Economic Review 95(1): 395-406.
____________________ (2006), “The Impact of Global Warming on U.S. Agriculture: An
Econometric Analysis,” Review of Economics and Statistics 88(1): 113-25.
Stern, Nicholas (2007), The Economics of Climate Change: The Stern Review, Cambridge:
Cambridge University Press.
Stein, Michael L. (1999), Interpolation of Spatial Data: Some Theory for Kriging, New
York: Springer.
Thomas, Chris D. et al. (2003), “Extinction Risk from Climate Change,” Nature 427(8):
145-8.
Thomson, M. C., F. J. Doblas-Reyes, S. J. Mason, R. Hagedorn, S. J. Connor, T. Phindela,
A. P., Morse, and T. N. Palmer (2006), “Malaria Early Warnings based on Seasonal
Climate Forecasts from Multi-Model Ensembles,” Nature 439(7076): 576-9.
Tol, Richard S. J. (2009), “The Economic Effects of Climate Change,” Journal of
Economic Perspective 23(2): 29-51.
U.S. Surgeon General’s Office (1988), The Medical and Surgical History of the War of the
Rebellion, 1861-65: Vol. 5, Washington, DC: Government Printing Office.
Vörösmarty, C. J., P. Green, J. Salisbury, and R. B. Lammers (2000), “Global Water
Resources: Vulnerability from Climate Change and Population Growth,” Science
289(5477): 284-8.
29
World Health Organization, WMO, and UNEP (2003), Climate Change and Human
Health-Risks and Responses, Geneva: World Health Organization.
30
Appendix A. Data Sources and Construction
A.1 Weather Variables
Raw Data from Weather Stations
Monthly mean temperature and accumulated precipitation are obtained from two
historical sources: the records of the Nineteenth-Century US Climate Data Set Project
developed by National Climate Data Center and the Long-Term Daily and Monthly
Climate Records from the Stations across the Contiguous United States provided by the US
Historical Climatology Network. The former dataset covers the periods of 1822-1900 and
includes 4,056 individual weather stations with different names. But all the stations did not
exist throughout the nineteenth century. The number of weather station was less than 100
prior to 1854, and it increased to about 400 in the 1860s, and to about 1,000 after the 1880s.
The later dataset has been utilized to obtain monthly weather information in 1901-2000.
The number of weather stations in the dataset is 1,221; each weather station existed over
most years of the twentieth century so that each year’s number of stations is over 1,000.
On the other hand, two datasets include the information on weather stations’ latitude and
longitude, which is used to locate them on map.
Spatial Averaging (“Kriging”)
A Kriging technique is used to interpolate monthly weather variables for the counties
whose weather records are not found in the above two sources. The method is based on a
linear least squares estimation algorithm, which minimizes the variance of the prediction
error. Thus a Kriging estimator is a liner combination of the value at nearby locations; the
distance between neighboring points is generally employed as weights. In this study, the
maximum distance between points is set as 200 miles. An enough number of weather
stations within 200-mile radius of the target location are required to get a reliable
estimation result. It was thought that the size of weather stations in the years prior to 1860
is insufficient, and so those years were dropped from the analysis.
Measurement by County
31
To estimate the county-level monthly mean temperature and accumulated
precipitation in 1860-2000, the geographical center of each county was used as the target
location in the Kriging estimation, and the actual weather information of weather stations
within the 200-mile buffer around the target location was utilized. The county-level
decadal (long-term) or within-year March-to-October (short-term) mean temperature is
calculated by averaging monthly estimated results over the specified periods. The decadal
precipitation is the average value of annual accumulated precipitation for the 10 years prior
to each census year; the March-to-October precipitation is calculated by summing up
monthly precipitation over the months within the year. Table A.1 presents the mean and
standard error of the estimated long-term and short-term weather variables throughout the
sample periods. The table also provides the actual census year that we referred to for
measuring long-term and short-term climate variables.
[Table A.1 Here]
A.2 County-Level Agricultural, Economic, Demographic and Environmental Data
Farm Productivity
Two variables are used to measure US farm productivity: farm value per farmland
acre and farm output value per farmland acre. Farm value is the value of all land, housing,
and outbuildings in the farm at the time of census enumeration; it can be interpreted as
land price because the value of land in the farm takes a large portion of farm value,
following the hedonic-approach studies. Farm output value is the total value of all farm
products such as crop and livestock products within the year prior to the enumeration day.
Both variables are generally obtained from Historical, Demographic, Economic, and
Social Data: The United States, 1790-2002 (ICPSR #2896), which digitized the key
variables in US federal population and agricultural censuses in 1850-1930 and US county
data book in 1940-2000. We tried to construct the variables for the 14 decades in 18702000, but the variables of county-level farm output value are missing for three decades in
1910-1930 from the ICPSR dataset. All the value is converted in constant 2000 US dollar
using the BLS GDP deflator.
32
Controls
The long-period analysis substantially limits the use of various controls that account
for historical changes in the relationship between climate and farm productivity. We
constructed a panel of four variables at the county level in 1870-2000: population density,
the ratio of white population to county population, the ratio of farmland to total available
county area, and the estimated number of farmers per farmland acre. The first three
variables were obtained from the dataset of ICPSR #2896 above. To estimate the number
of farmers per farmland acre, we first calculated the county ratio of farmers from the
IPUMS dataset, which is a subsample of census, and applied the ratio to the actual
population per farmland acre. Sample means of each farm-productivity measure and
controls by census year are reported in Table A.2.
A.3 Boundaries of Consistent Counties
Changes in county boundaries over time can make county-fixed-effects analyses
problematic. As a solution, we adjusted all the county variables in 1880-2000 on the 1870
county boundary in Figure A.1 by tracking the overlap of county boundaries over decade
and using the (overlapped) area-weighted average method. On the other hand, Appendix
B.1 below shows that the use of county-boundary-adjusted data does not cause critical
problems in estimating fixed-effect models.
[Figure A.1 Here]
A.4 Measuring Various Agricultural Technologies and Data Sources
To discuss possible explanations of across-century difference in agricultural response
to climate, we employ various county variables that show the level of agricultural
technologies and economic conditions and its change over the century. The technologies
for drainage, irrigation, selection of crops and crop mix have been developed to assimilate
the permanent pattern of local climate and its change. Although some variables such as
33
average wage in agriculture, farm structure, use of fertilizer, and farm mortgage rate are
less related with climate change, the control of those variables would show whether the
adaptation to weather resulted from the change in agricultural conditions that are
correlated to local climate.
The year and source of each variable is summarized in Table A.3. Because the
variables are not available for all the census years in 1870-2000, we searched for historical
sources in which each variable was first reported. Although one year’s variable cannot
measure technological changes over time, it may represent the overall level of agricultural
technologies in the periods when substantial improvements were not yet observed, i.e. the
pre-existing condition. We also searched for the variables that measure the agricultural and
economic conditions in the second half of the twentieth century when it is thought that
advanced technologies were largely introduced into the most US counties. The difference
between the pre-existing and recent conditions is used in Appendix B.3 to estimate the role
of technological improvements in explaining the historical pattern of the adaptation to
weather.
[Table A.3 Here]
Appendix B. Supplementary Analyses
B.1 Fixed-Effects Analysis Using Data with the County Boundary Adjustment
Throughout this study, we use county-boundary-adjusted data in estimating countyfixed-effect regressions, as discussed in Appendix A.3 above. Using boundary-adjusted
data, models (1) and (3) present the estimated coefficients of the interaction terms between
each weather variable and each year dummy, and their standard errors, clustered on county,
after regressing equation (1) with county fixed effects. These coefficients are also
graphically depicted by solid lines in Figure 2. In models (2) and (4), we conduct the same
regressions, but use data without the county boundary adjustment. No substantial disparity
is found between two estimation results with and without the adjustment.
[Table B.1 Here]
34
B.2 Fixed-Effects Analysis for South and North
In Section 4, we showed that the relationship between weather condition and
agricultural productivity has substantially changed over centuries. We particularly
measured the across-century difference in the response of farm value or farm output value
to weather conditions by running the regressions per equation (3). Hot and wet weather
significantly depressed agricultural productivity in the late nineteenth century, but the
effect has been attenuated over time. Thus, the historical pattern of the adaptation to
weather would be different across region. To see the regional difference, we estimate
equation (3) by two regions: the southern and northern counties. Table B.2 reports the
coefficients that measure the across-century weather sensitivity. As we did in Section 4, we
control for long-term weather variables when farm value is used as the dependent variable,
and short-term weather variables for farm output value. The estimated coefficients of the
interaction terms between each weather variable and the dummy of the nineteenth century
are more negative for the southern counties than for the northern counties; in the case that
farm output value is used as the dependent variable, the coefficients for the northern
counties is near zero and statistically insignificant. Those results suggest that the
adaptation to weather occurred more substantially in the South than in the North.
[Table B.2 Here]
35
Table 1. Across-Century Difference in the Response of County Farm Value to Long-Term
Weather Conditions
Dependent variable: ln(county average farm value per acre)
(1)
Key Control Variables
Temperature(T)*D19
(2)
Panel A: Only Temperature
-0.0592***
-0.0664***
(0.0030)
(0.0029)
(3)
(4)
-0.0603***
(0.0056)
-0.0447***
(0.0052)
-0.0291***
(0.0028)
-0.0170***
(0.0028)
Panel B: Only Precipitation
Precipitation(P)*D19
-0.0459***
(0.0022)
-0.0411***
(0.0022)
Panel C: Temperature and Precipitation
Temperature(T)*D19
-0.0313***
(0.0031)
-0.0484***
(0.0033)
-0.0565***
(0.0057)
-0.0437***
(0.0053)
Precipitation(P)*D19
-0.0411***
(0.0021)
-0.0305***
(0.0022)
-0.0232***
(0.0027)
-0.0124***
(0.0028)
(T-µT)*(P-µP)*D19
0.0008***
(0.0002)
-0.0003
(0.0002)
-0.0005
(0.0003)
-0.0006*
(0.0003)
Year FE
NO
YES
YES
YES
State by Year FE
NO
NO
YES
YES
County FE
NO
NO
NO
YES
Notes: We estimate the across-century difference in the response of county farm value to long-term weather
conditions, per equation (2) for panels A and B, and equation (3) for panel C. The long-term weather conditions are
measured by the 10-year average of annual mean temperature (T) and annual accumulated precipitation (P) prior to
each census year, respectively. µT and µP denote the mean value of temperature and precipitation across counties,
respectively. D19 is the dummy variable that indicates the census years in the 19th century (i.e. 1870, 1880, 1890 or
1900). Standard errors, reported in parentheses, are clustered on county. Single asterisk denotes statistical
significance at the 90% level of confidence, double 95%, triple 99%.
36
Table 2. Across-Century Difference in the Response of County Farm Output Value to Short-Term
Weather Conditions
Dependent variable: ln(county average farm output value per acre)
(1)
Key Control Variables
(2)
(3)
(4)
Panel A: Weather Variables in Current Year
Current
Year
Temperature(T)*D19
0.0055
(0.0039)
0.0064
(0.0042)
-0.0401***
(0.0082)
-0.0389***
(0.0081)
Precipitation(P)*D19
-0.0324***
(0.0027)
-0.0290***
(0.0029)
-0.0218***
(0.0033)
-0.0156***
(0.0029)
(T-µT)*(P-µP)*D19
0.0023***
(0.0004)
0.0018***
(0.0004)
0.0015***
(0.0006)
0.0010*
(0.0006)
Panel B: Weather Variables in Current, Next and Previous Years
Current
Year
Next Year
Previous
Year
Temperature(T)*D19
-0.0613***
(0.0060)
-0.0434***
(0.0100)
-0.1032***
(0.0217)
-0.0432***
(0.0156)
Precipitation(P)*D19
-0.0174***
(0.0021)
-0.0163***
(0.0021)
-0.0203***
(0.0029)
-0.0152***
(0.0024)
(T-µT)*(P-µP)*D19
-0.0001
(0.0003)
0.0002
(0.0003)
0.0005
(0.0005)
0.0004
(0.0004)
Temperature(T)*D19
0.0568***
(0.0104)
0.0439***
(0.0122)
0.0036
(0.0190)
0.0117
(0.0143)
Precipitation(P)*D19
-0.0122***
(0.0018)
-0.0135***
(0.0020)
-0.0024
(0.0028)
-0.0024
(0.0022)
(T-µT)*(P-µP)*D19
0.0019***
(0.0004)
0.0016***
(0.0004)
0.0025***
(0.0006)
0.0009**
(0.0004)
Temperature(T)*D19
0.0229**
(0.0096)
0.0119
(0.0118)
0.0578***
(0.0186)
-0.0064
(0.0137)
Precipitation(P)*D19
-0.0078***
(0.0022)
-0.0066***
(0.0022)
-0.0101***
(0.0025)
-0.0016
(0.0021)
(T-µT)*(P-µP)*D19
0.0013***
(0.0003)
0.0008**
(0.0003)
-0.0007*
(0.0004)
0.0001
(0.0003)
NO
YES
YES
YES
Year FE
State by Year FE
NO
NO
YES
YES
County FE
NO
NO
NO
YES
Notes: We estimate the across-century difference in the response of county farm output value to short-term weather conditions,
per equation (3). The short-term weather conditions are measured by the March-to-October mean temperature (T) and
accumulated precipitation (P) in the census year (denoted by Current Year) and in the periods one year after and before the
census year. µT and µP denote the mean value of temperature and precipitation across counties in the given year, respectively.
D19 denotes the dummy variable that indicates the census years in the 19th century (i.e. 1870, 1880, 1890 or 1900). Standard
errors, reported in parentheses, are clustered on county. Single asterisk denotes statistical significance at the 90% level of
confidence, double 95%, triple 99%.
37
Table 4.1. Across-Century Difference in the Response of County Farm Productivity to Weather
and Pre-Existing Agricultural Factors
% Drained
Areas
Key Control Variables
% Irrigated
Areas
% Cropland
Crop HHI
(1)
(2)
(3)
(4)
Panel A: Farm Value and Long-Term Weather Variables
% Cotton
Acres
(5)
Temperature(T)*D19
-0.0283***
(0.0080)
-0.0287***
(0.0085)
-0.0310***
(0.0087)
-0.0111
(0.0096)
-0.0115
(0.0088)
Precipitation(P)*D19
0.0012
(0.0052)
0.0039
(0.0056)
0.0021
(0.0060)
0.0177***
(0.0063)
0.0041
(0.0058)
(T-µT)*(P-µP)*D19
0.0009**
(0.0004)
0.0009**
(0.0004)
0.0004
(0.0004)
0.0011**
(0.0004)
0.0010**
(0.0004)
T*Malaria*D19
-0.1008***
(0.0311)
-0.1017***
(0.0317)
-0.1625***
(0.0359)
-0.0776**
(0.0315)
-0.1236***
(0.0393)
P*Malaria*D19
-0.1034***
(0.0224)
-0.0926***
(0.0230)
-0.0647**
(0.0276)
-0.1009***
(0.0235)
-0.1007***
(0.0268)
T*Factor*D19
-0.0042
(0.0118)
0.1625***
(0.0483)
0.0435***
(0.0128)
0.1232*
(0.0641)
0.0711
(0.1038)
P*Factor*D19
0.0404***
(0.0085)
0.0012
(0.0374)
-0.0112
(0.0091)
0.1735***
(0.0410)
-0.0359
(0.0450)
Panel B: Farm Output Value and Short-Term Weather Variables
Temperature(T)*D19
-0.0349***
(0.0115)
-0.0407***
(0.0118)
-0.0442***
(0.0120)
0.0100
(0.0121)
0.0036
(0.0119)
Precipitation(P)*D19
0.0320***
(0.0086)
0.0367***
(0.0082)
0.0315***
(0.0097)
0.0420***
(0.0087)
0.0403***
(0.0092)
(T-µT)*(P-µP)*D19
0.0037***
(0.0007)
0.0041***
(0.0007)
0.0031***
(0.0007)
0.0042***
(0.0007)
0.0046***
(0.0007)
T*Malaria*D19
-0.1094**
(0.0487)
-0.0932*
(0.0489)
-0.1758***
(0.0508)
-0.0752
(0.0478)
-0.1591***
(0.0558)
P*Malaria*D19
-0.2183***
(0.0329)
-0.2169***
(0.0312)
-0.1877***
(0.0360)
-0.2046***
(0.0314)
-0.2236***
(0.0374)
T*Factor*D19
0.0164
(0.0109)
0.1295
(0.0914)
0.0729***
(0.0181)
0.4672***
(0.0738)
0.2291*
(0.1357)
P*Factor*D19
0.0267***
(0.0068)
-0.2460***
(0.0942)
-0.0060
(0.0151)
0.1512***
(0.0522)
0.0421
(0.0409)
Notes: We estimate the role of various agricultural factors (F)---specified in the headings---in explaining the historical
pattern of the adaptation to weather, per equation (5) where the set of K also includes {F, TF and PF}. Panel A
employs farm value and long-term weather variables (decadal average); Panel B uses farm output value and short-term
weather variables (March-to-October average). This table reports only the coefficients of key variables---particularly the
interactions of each specified agricultural factor with weather variable and 19th-century dummy. Standard errors,
clustered on county, are reported in parentheses. Single asterisk denotes statistical significance at the 90% level of
confidence, double 95%, triple 99%.
38
Table 4.2. Across-Century Difference in the Response of County Farm Productivity to Weather
and Pre-Existing Agricultural Factors
Annual
ln(Fertilizer
Agricultural
% Share
Cost
Wages
Cropper
% Tenant
per Acre)
per Acre
(6)
(7)
(8)
(9)
Panel A: Farm Value and Long-Term Weather Variables
-0.0277***
-0.0209**
-0.0288***
-0.0316***
(0.0084)
(0.0084)
(0.0086)
(0.0082)
-0.0990***
(0.0208)
Precipitation(P)*D19
0.0048
(0.0056)
0.0037
(0.0056)
0.0077
(0.0057)
-0.0013
(0.0056)
0.0244*
(0.0136)
(T-µT)*(P-µP)*D19
0.0007
(0.0004)
0.0008*
(0.0004)
0.0006
(0.0004)
0.0002
(0.0004)
0.0011**
(0.0005)
T*Malaria*D19
-0.1025***
(0.0348)
-0.0899***
(0.0331)
-0.0869***
(0.0318)
-0.0907***
(0.0317)
-0.1211***
(0.0318)
P*Malaria*D19
-0.0972***
(0.0245)
-0.1050***
(0.0238)
-0.1008***
(0.0235)
-0.0810***
(0.0236)
-0.1014***
(0.0250)
T*Factor*D19
0.0361*
(0.0217)
-0.0913**
(0.0423)
0.0006
(0.0012)
-0.0005
(0.0005)
0.9009***
(0.2332)
P*Factor*D19
-0.0071
(0.0153)
0.0374
(0.0245)
0.0013*
(0.0008)
0.0014***
(0.0004)
-0.2102
(0.1307)
Key Control Variables
Temperature(T)*D19
Mortgage
Rate
(10)
Panel B: Farm Output Value and Short-Term Weather Variables
-0.0351***
-0.0309***
-0.0246**
-0.0371***
(0.0116)
(0.0117)
(0.0123)
(0.0113)
-0.0696**
(0.0288)
Precipitation(P)*D19
0.0365***
(0.0089)
0.0345***
(0.0085)
0.0309***
(0.0087)
0.0262***
(0.0085)
0.0901***
(0.0204)
(T-µT)*(P-µP)*D19
0.0038***
(0.0007)
0.0039***
(0.0007)
0.0035***
(0.0007)
0.0029***
(0.0007)
0.0045***
(0.0008)
T*Malaria*D19
-0.0827*
(0.0491)
-0.1166**
(0.0511)
-0.1112**
(0.0495)
-0.0922*
(0.0487)
-0.1005**
(0.0504)
P*Malaria*D19
-0.2174***
(0.0310)
-0.2392***
(0.0339)
-0.2056***
(0.0319)
-0.1950***
(0.0324)
-0.2049***
(0.0354)
T*Factor*D19
0.0795**
(0.0315)
0.0237
(0.0425)
0.0043**
(0.0018)
-0.0014
(0.0011)
0.4893
(0.3501)
P*Factor*D19
0.0006
(0.0216)
0.0904***
(0.0256)
-0.0015
(0.0013)
0.0027***
(0.0009)
-0.7319***
(0.2505)
Temperature(T)*D19
Notes: We estimate the role of various agricultural factors (F)---specified in the headings---in explaining the historical
pattern of the adaptation to weather, per equation (5) where the set of K also includes {F, TF and PF}. Panel A
employs farm value and long-term weather variables (decadal average); Panel B uses farm output value and short-term
weather variables (March-to-October average). This table reports only the coefficients of key variables---particularly the
interactions of each specified agricultural factor with weather variable and 19th-century dummy. Standard errors,
clustered on county, are reported in parentheses. Single asterisk denotes statistical significance at the 90% level of
confidence, double 95%, triple 99%.
39
Table A.1. Summary Statistics: County Weather Variables
Denote
d
census
year
Actual
census
year
Number
of
counties
1870
1869
2,162
1880
1879
2,454
1890
1889
2,791
1900
1899
2,767
1910
1909
2,916
1920
1919
3,026
1930
1929
3,059
1940
1939
3,074
1950
1949
3,071
1960
1959
3,081
1970
1969
3,079
1980
1978
3,078
1990
1988
3,078
2000
1998
3,080
Average for the entire
sample years
Mean temperature
(oF)
Accumulated precipitation
(inches)
Longterm
54.1
55.0
53.8
54.5
54.6
54.5
55.1
56.0
55.3
55.3
54.5
54.5
54.5
54.6
S.E
.
5.7
7.3
7.8
7.9
7.7
8.0
8.3
7.9
7.9
8.1
7.7
7.7
7.7
7.8
Shortterm
61.9
64.9
62.6
63.8
63.0
64.3
63.9
65.1
64.2
64.3
62.9
65.1
64.1
62.7
S.E
.
5.9
6.1
6.2
7.2
7.1
6.8
7.1
7.0
6.4
6.5
6.6
6.6
5.8
6.8
Longterm
41.8
40.2
38.8
36.4
37.7
36.2
37.3
34.8
37.8
36.1
35.7
38.3
38.1
38.8
54.9
7.7
63.9
6.6
37.9
6.6
11.2
12.2
12.3
12.3
13.0
13.8
13.4
13.6
13.4
13.4
14.3
13.1
14.4
Shortterm
29.8
25.7
25.6
24.3
26.5
29.5
28.9
23.5
27.8
28.7
26.7
28.4
23.7
27.3
7.5
7.8
9.7
7.3
10.2
11.1
12.0
9.9
11.1
11.6
9.8
10.1
7.8
9.5
12.7
27.2
9.9
S.E.
S.E.
Notes: County-level monthly weather mean temperature and accumulated precipitation were estimated by Kriging
interpolation methods as discussed in the text. This table presents sample mean and standard errors across counties for
the given census year. The long-term weather variables are calculated by the 10-year average of annual mean
temperature and annual accumulated precipitation prior to each census year; the short-term weather variables are
measured by the March-to-October mean temperature and accumulated precipitation in each census year.
40
Table A.2. Summary Statistics: Farm Productivity Measures and Controls
Productivity Measures
Denoted
census
year
Actual
census
year
Number
of
counties
1870
1880
1890
1900
1910
1920
1930
1940
1950
1960
1970
1980
1990
2000
1869
1879
1889
1899
1909
1919
1929
1939
1949
1959
1969
1978
1988
1998
2,162
2,454
2,791
2,767
2,916
3,026
3,059
3,074
3,071
3,081
3,079
3,078
3,078
3,080
Controls
Farm Value per
Farmland Acres
Farm Output
Value per
Farmland Acres
Population
Density
Ratio of
White
Population
Farm
Acres /
Total
Area
Farmer /
Farmland
Acres
354.2
335.2
476.1
499.1
753.2
720.3
678.1
765.3
681.4
941.3
1147.7
2477.9
1577.7
2452.9
90.3
76.6
88.9
112.6
147.6
243.2
198.9
242.1
464.0
381.8
365.4
0.0523
0.0825
0.0723
0.0870
0.1093
0.1382
0.1185
0.1264
0.1434
0.1013
0.0996
0.1404
0.1506
0.1165
0.8362
0.8389
0.8582
0.8581
0.8646
0.8761
0.8642
0.8862
0.8920
0.8938
0.9001
0.8862
0.8638
0.8537
0.5207
0.5962
0.5821
0.6740
0.6714
0.6656
0.6441
0.6672
0.6909
0.6419
0.6001
0.5611
0.5290
0.5194
0.0202
0.0191
0.0110
0.0161
0.0191
0.0211
0.0308
0.0155
0.0447
0.0201
0.0407
0.0093
0.0088
0.0093
Notes: Farm value and farm output value are converted in constant 2000 dollars. Population density is measured as the
number of county populations per square mile divided by 1,000. Farm output value is not available at the county level in
1910-1930.
41
Table A.3. Year and Source of Variables for Agricultural Technologies and Economic
Conditions
Type of technology
Variable
Drainage
Irrigation
Land improvement
for crops
Crop mix
Selection of crops
Farm structure
Use of fertilizer
Farm labor market
Farm mortgage rate
Year of
early
condition
Year of
later
condition
Ratio of drained areas
1930
-
Ratio of irrigated areas
Ratio of cropland
1890
1880
1980
1980
ICPSR 2896
ICPSR 2896
Crop Herfindahl-Hirschman
index
1880
1980
ICPSR 2896: County-level
Herfindahl-Hirschman
index was calculated using
county acres for 10 major
crops: wheat, corn, barley,
oats, rye, potato, cotton,
tobacco, rice and hay
1880
1880
1960
ICPSR 2896
ICPSR 2896
1880
1960
ICPSR 2896
1880
1960
ICPSR 2896
1870
-
ICPSR 2896
1890
-
ICPSR 2896
Crop acres out of total
farmland
Ratio of crop-share farms
Ratio of tenant (cash rental)
farms
ln(fertilizer cose per
farmland acre)
Annual agricultural wage
per farmland acre
Interest paid on farms per
mortgage value of farms
Source and note
1930 Agricultural Census:
Drainage of Agricultural
Lands
Notes: ICPSR 2896 is the dataset developed by Michael Haines, which is available at Interuniversity Consortium for
Political and Social Research (ICPSR).
42
Table B.1. The Effects of Long-Term Temperature and Precipitation on Farm Value: Using Data
with and without County Boundary Adjustment
Dependent variable: ln(county average farm value per acre)
Key control variables
Weather*D1880
Weather*D1890
Weather*D1900
Weather*D1910
Weather*D1920
Weather*D1930
Weather*D1940
Weather*D1950
Weather*D1960
Weather*D1970
Weather*D1980
Weather*D1990
Weather*D2000
Weather = Decadal average of annual
mean temperature
(1)
(2)
Using countyUsing data
boundarywithout
adjusted data
adjustment
Coeff.
S.E.
Coeff.
S.E.
0.0243 0.0075
0.0271 0.0071
0.0293 0.0081
0.0323 0.0080
0.0368 0.0086
0.0388 0.0088
0.0457 0.0091
0.0491 0.0092
0.0400 0.0093
0.0429 0.0092
0.0550 0.0096
0.0539 0.0092
0.0555 0.0094
0.0521 0.0089
0.0682 0.0097
0.0648 0.0091
0.0772 0.0099
0.0789 0.0095
0.0738 0.0096
0.0731 0.0094
0.0768 0.0095
0.0741 0.0092
0.0696 0.0094
0.0703 0.0092
0.0685 0.0093
0.0714 0.0090
Weather = Decadal Average of Annual
Accumulated Precipitation
(3)
(4)
Using countyUsing data
boundarywithout
adjusted data
adjustment
Coeff.
S.E.
Coeff.
S.E.
0.0058 0.0057
0.0086 0.0053
0.0064 0.0054
0.0072 0.0055
0.0114 0.0055
0.0100 0.0061
0.0056 0.0055
0.0047 0.0061
0.0041 0.0056
0.0018 0.0058
0.0065 0.0057
0.0037 0.0056
0.0149 0.0057
0.0114 0.0055
0.0101 0.0057
0.0080 0.0056
0.0156 0.0060
0.0162 0.0061
0.0223 0.0058
0.0246 0.0060
0.0214 0.0057
0.0211 0.0059
0.0215 0.0058
0.0212 0.0060
0.0209 0.0058
0.0198 0.0058
Notes: We run the fixe-effect estimation regression that includes all the three fixed effects, per equation (1), for two
datasets with and without county boundary adjustment. The table reports only the coefficients of the interaction terms
between long-term temperature or precipitation and each year dummy. Standard errors are clustered on county. All the
coefficients are statistically significant at the 99% level of confidence.
43
Table B.2. Across-Century Difference in the Response of Farm Productivity to Weather Conditions by
Region
Farm Value
Farm Output Value
(1)
Entire
Counties
(2)
Southern
Counties
(3)
Northern
Counties
(4)
Entire
Counties
(5)
Southern
Counties
Temperature(T)*D19
-0.0437***
(0.0053)
-0.0923***
(0.0088)
-0.0288***
(0.0066)
-0.0389***
(0.0081)
-0.0973***
(0.0161)
0.0046
(0.0074)
Precipitation(P)*D19
-0.0124***
(0.0028)
-0.0389***
(0.0051)
-0.0175***
(0.0045)
-0.0156***
(0.0029)
-0.0460***
(0.0057)
0.0046
(0.0049)
(T-µT)*(P-µP)*D19
-0.0006*
(0.0003)
0.0027***
(0.0005)
-0.0011*
(0.0006)
0.0010*
(0.0006)
0.0045***
(0.0010)
0.0033***
(0.0010)
Key Control Variables
(6)
Northern
Counties
Notes: We run the regressions per equation (3) in the text, respectively, for the entire counties and two regional groups: the
southern and northern counties. This table reports the key estimated coefficients of the interaction terms among weather
variables (long-term weather for farm value analysis and short-term weather variable for farm output value analysis) and the
dummy that indicates the 19th century. Standard errors, clustered on county, are reported in parentheses. Single asterisk denotes
statistical significance at the 90% level of confidence, double 95%, triple 99%.
44
Figure 1. Scatter Plots: Log County Average Farm Value per Acre by 10-Year Average
Temperature and Precipitation
Decadal Average of County Annual Mean Temperature
1900
10
8
6
4
2
0
-2
12
10
8
6
4
2
0
-2
35 40 45 50 55 60 65 70 75
Degree Fahrenheit
8
6
4
2
0
8
6
4
2
0
35 40 45 50 55 60 65 70 75
Degree Fahrenheit
1970
-2
12
2000
Log of Farm Value per Acre
10
10
35 40 45 50 55 60 65 70 75
Degree Fahrenheit
Log of Farm Value per Acre
12
12
-2
1950
Log of Farm Value per Acre
1920
Log of Farm Value per Acre
Log of Farm Value per Acre
Log of Farm Value per Acre
1870
12
10
8
6
4
2
0
-2
35 40 45 50 55 60 65 70 75
Degree Fahrenheit
12
10
8
6
4
2
0
-2
35 40 45 50 55 60 65 70 75
Degree Fahrenheit
35 40 45 50 55 60 65 70 75
Degree Fahrenheit
Decadal Average of County Annual Accumulated Precipitation
1900
10
8
6
4
2
0
-2
12
10
8
6
4
2
0
-2
0
10 20 30 40 50 60 70 80
Inches
0
8
6
4
2
0
-2
12
10
8
6
4
2
0
-2
0
10 20 30 40 50 60 70 80
Inches
8
6
4
2
0
10 20 30 40 50 60 70 80
Inches
0
10 20 30 40 50 60 70 80
Inches
2000
Log of Farm Value per Acre
10
10
1970
Log of Farm Value per Acre
12
12
-2
1950
Log of Farm Value per Acre
1920
Log of Farm Value per Acre
Log of Farm Value per Acre
Log of Farm Value per Acre
1870
12
12
10
8
6
4
2
0
-2
0
10 20 30 40 50 60 70 80
Inches
0
10 20 30 40 50 60 70 80
Inches
Notes: Each weather variable in the x axis is calculated by the 10-year average of annual weather values prior
to each census year. The y axis denotes the logarithm of the county average farm value per farmland acre.
Solid curves are lowess fit curves, and dotted lines are linear fit lines.
45
Figure 2. Trend of the Estimated Effect of Long-Term Temperature and Precipitation on County
Farm Value per Acre
Precipitation
.05
.14
Temperature
Year FE
and State by Year FE
and State by Year FE
and County FE
and County FE
Regression Coefficient
.02
.03
Year
2000
1990
1980
1970
1960
1950
1940
1930
1920
1910
1900
1890
1880
0
2000
1990
1980
1970
1960
1950
1940
1930
1920
1910
1900
1890
1880
0
.02
.01
.04
Regression Coefficient
.06
.08
.1
.04
.12
Year FE
Year
Notes: We run the regressions of the logarithm of county average farm value on the 10-year average of annual mean
temperature or annual accumulated precipitation prior to each census year, standard controls, and their interactions with
the dummies indicating the 1880-2000 census years, per equation (1). Thus, the reference year is 1870. Each panel is the
graphical presentation of regression coefficients of each weather variable. The dotted and dashed lines are the estimation
results that employ year and state-by-year fixed effects, respectively; the solid lines additionally employ county fixed
effects. All the regressions datasets that adjust county boundary changes to the 1870 county boundary.
46
Figure 3. The Relation between Average Weather Conditions in Early-Life and Adult
Income by Two Cohorts
Average Occscore
3
3.25
3.5
Born after 1930
DC
NM
40
CA
OK
LA
AZ
TN NC
AR
FL
TX
SCMS
GA
AL
ND
DC
CT
NY
UT NJ
MA
DE
RIPAOH IL MD
WY
CA
IN MO
LA
KS
WI
WAORNV
NHMI
CO
ID IA NE
MNMT
OK AZ SC
MS
TX
GA
VA
NC
AL
VT SD
TN
ME
AR
WV NMKY
50
60
Average Temperature
70
40
50
60
Average Temperature
DC
AZUT
LA
20
30
40
50
Average Precipitation
UT
WY
NVCOID
AZ MTND SD
NM
DC
NJ
CT
NY
MA
IL
DE
MD
RI
OHIN
PA
CA
MI
KS WI
FLMSLA
WA
NH
MO
SC
IA
NEMN
TXOKOR
VA
NCGA AL
VT
TN
ME
WVKY AR
2.75
2.75
MA
RI
CT
MD NJ
NH
WA NYPA
ME
DE
VT
OK
MI
OH
WI
IA OR IL
MO
IN VA
WVKY SC
TX
FL
GA MS
NCTN AL
AR
NM
10
CA KS
NEMN
70
Born after 1930
Average Occscore
3
3.25
3.5
Average Occscore
3
3.25
3.5
Born before 1900
CO
NVMT
WY
ID
ND SD
FL
2.75
ND
MARI
CO CT
NJ
NV MD
MT
NH NY
WY
PA
WA
KS
ME
DE
MN
NE
VT SD ID
OH
WI MI
IA UT OR IL MO
VA
IN
WV
KY
2.75
Average Occscore
3
3.25
3.5
Born before 1900
60
10
20
30
40
50
Average Precipitation
60
Notes: This figure plots a proxy of adult income against the state-of-birth average temperature and precipitation
for earlier- and later-born cohorts in the U.S. A cohort is defined by year of birth and state of birth. Data on
native adult males are drawn from U.S. censuses of 1880-1990, which span the years of birth 1861-1960. The
outcome variable is the average occupational income score, transformed into natural logarithms. Cohorts are
grouped into those born before 1900 (in the left-hand plots) and those born after 1930 (in the right-hand plots).
The x axis is the state-of-birth average temperature in the upper plots or precipitation in the bottom plots. The y
axis in each plot refers to the cohort group’s average income score. State abbreviations are used to denote the
position of each point. The solid line is the best-fit regression line between the points. Sources and methods of
data construction are described in the appendix.
47
Figure 4. Short-Term Weather Fluctuations around Year of Birth and Adult Income
Temperature
Precipitation
.015
.005
.002
.005
0
0
-.005
-.002
-.015
-.005
-2
-1
0
1
2
3
-2
-1
0
1
2
3
Notes: This figure summarizes regressions of cohort outcome on early-life weather conditions, per equation (4) in the
text. A cohort is defined by year of birth and state of birth. Data on native adult males are drawn from U.S. censuses of
1880-1990, which span the years of birth 1861-1960. The outcome variable is, at the cohort level, the occupational
income score transformed into natural logarithms. The x axis in each plot refers to the calendar year minus the year of
birth. The y axis in each plot displays the estimated regression coefficient on the interaction of the indicated weather
variable (temperature or precipitation) and a dummy for the calendar year minus year of birth. The error bars reflect 95%
confidence intervals for each coefficient. Sources and methods of data construction are described in the appendix.
48
Figure 5. The Diminishing Effects of Early-Life Weather Fluctuations on Adult Income
-.02
0
.02
.04
.06
Temperature
1880
1890
1900
1910
Year of Birth
1920
1930
1940
1920
1930
1940
-.005
0
.005
.01
.015
Precipitation
1880
1890
1900
1910
Year of Birth
Notes: This figure summarizes regressions of cohort adult income on early-life weather conditions, per
equation (4) in the text. The regressions are conducted for a 40-year-wide moving window so that the sample
for each regression includes cohorts born 20 years before and after the target year. Regarding weather variables,
we only control for temperature in the period one year after birth and precipitation at the year of birth according
to the results in Figure 4. The x axis in each plot refers to the average year of birth for cohort samples used. The
y axis in each plot displays the estimated regression coefficient on the interaction of the indicated weather
variable (temperature or precipitation) and a dummy for the year chosen as explained above. The error bars
reflect 95% confidence intervals for each coefficient. Sources and methods of data construction are described in
the appendix.
49
Figure A.1. 1870 County Boundaries
Notes: To resolve the over-time changes in county boundaries, we adjusted all the county variables on the 1870
boundary shown in this figure.
50