Rainfall Risk and Religious Membership in the Late Nineteenth-Century US by Philipp Ager ** Antonio Ciccone

Documento de Trabajo 2013-17

December 2013

* UPF. ** ICREA-UPF, Barcelona GSE, CREI and FEDEA.

Los Documentos de Trabajo se distribuyen gratuitamente a las Universidades e Instituciones de Investigación que lo solicitan. No obstante están disponibles en texto completo a través de Internet: http://www.fedea.es. These Working Paper are distributed free of charge to University Department and other Research Centres. They are also available through Internet: http://www.fedea.es. ISSN:1696-750

Philipp Ager and Antonio Ciccone

November 2012

Abstract

Insurance among the members of religious organizations should be more valuable in communities facing greater risk, making membership in religious organizations more attractive in high-risk environments. We examine the link between rainfall risk and church membership as well as seating capacity across US counties in the second half of the nineteenth century. Our results indicate that church membership and seating capacity were signiÖcantly larger in counties likely to have been subject to greater rainfall risk. This link is present among the most agricultural counties and among counties with low population densities, but not among less agricultural or more densely populated counties. Among the most agricultural counties, a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 32 percent in 1890 and 65 percent in 1860.

UPF (philipp.ager@upf.edu) and ICREA-UPF, Barcelona GSE, and CREI (antonio.ciccone@upf.edu). We thank Richard Hornbeck for help with the soil data in the early stages of the pro ject and Hayk Yeritsian for professional GIS assistance.

1 Introduction

One aspect of todayís major religions is that they spiritually reward mutual aid and charity (McCleary and Barro, 2006). This could be a reason why religious organizations have often been at the center of community assistance and mutual aid networks (e.g. Bremner, 1994; Parker, 1998; Pullan, 1998, 2005; Belcher and Tyles, 2011). The access to such networks through religious membership may help explain the historical spread of religion in agricultural societies subject to much uncertainty, as well as recent stagnation in countries where substantial social insurance is supplied by government (Gill and Lundsgaarde, 2004; McCleary and Barro, 2006; Franck and Iannaccone, 2009).

The US in the second half of the nineteenth century provides an interesting opportunity for examining the link between the economic uncertainty faced by communities and membership in religious organizations. The US started the second half of the nineteenth century as a mostly agricultural economy with little social insurance supplied by the government. In 1850, the agricultural sector produced around 40 percent of GDP, and the agricultural share of GDP exceeded the manufacturing share until the 1880s (Gallman, 1986). Government social spending was low over this period, below 0.3 percent of GDP in 1880 and around 0.5 percent of GDP in 1890 (Lindert, 1984). As much of US agriculture was rainfed, output was subject to rainfall risk (USDA, 1923, 1925). There is data to quantify the temporal variability of rainfall at the county level since 1895, and it is therefore possible to measure the rainfall risk local communities were likely to be subject to in the nineteenth century. Moreover, the US Census collected data on the seating capacity of churches at the county level from 1850 to 1890 and their membership in 1890. Hence, we can examine whether membership in religious organizations in the late nineteenth-century US was greater in local communities likely to have been facing more rainfall risk.

Historically, religious organizations have often been at the core of local and cross-regional mutual assistance and social aid networks; for US evidence see Trattner (1974), Bodnar (1985), and Gjerde (1985) for example. Joining such networks can be more attractive in communities facing greater aggregate uncertainty as the value of partially insuring idiosyncratic shocks may increase with aggregate background risk. We show this in a theoretical model where

ñ as in our empirical setting ñ aggregate background risk is driven by rainfall risk common to all members of the community; the condition for partial insurance against idiosyncratic risk to become more valuable as common rainfall risk increases turns out to be that relative risk aversion is above unity, which is plausible empirically (e.g. Attanasio and Weber, 1989; Vissing-Jorgensen and Attanasio, 2003).¹ The assistance networks around churches could also prove valuable because these networks often extended across local communities and could therefore provide some inter-regional insurance; for US evidence on such inter-regional assistance see Overacker (1998) and Szasz (2004) for example.

¹ Another implication of this framework is that (religious) communities with better mutual insurance would Önd places with greater economic risk relatively more attractive.

Our empirical analysis of the link between rainfall risk and membership in religious organizations in the late nineteenth-century US is cross county. For 1890, where we have data on both church seating capacity and church membership, the necessary data are available for around 2650 counties. Our analysis yields a statistically signiÖcant link between rainfall risk and church seating capacity as well as membership, controlling for a range of county characteristics likely to a§ect agricultural productivity as well as county size. A one-standard-deviation increase in rainfall risk is associated with an increase in church membership and church seating capacity of around 12 percent. The link between rainfall risk and membership in religious organizations is similar when we use the 1860 and 1870 data on church seating capacity (the 1880 data were never published).

If rainfall risk a§ects membership in religious organizations through economic risk ñ as in our theoretical model ñ we would expect the link between rainfall risk and religious membership to be stronger among nineteenth-century agricultural counties as their economies were more dependent on rainfall. We therefore examine the link between rainfall risk and religious membership separately among counties with population densities below the median and among counties with population densities above the median, and also split counties into those with value added in agriculture relative to manufacturing above and below the median. We Önd a statistically signiÖcant link between rainfall risk and religious membership among counties with population densities below the median and also among the more agricultural counties. On the other hand, the link between rainfall risk and membership in religious organizations among more densely populated counties and among less agricultural counties turns out to be statistically insigniÖcant. The link between rainfall risk and religious membership in 1890 is strongest among counties with agriculture/manufacturing value added above the median (the median is 6.7 which translates into a share of agriculture over agriculture plus manufacturing of 0.87). There, a one-standarddeviation increase in rainfall risk is associated with an increase in church membership of around 23 percent and an increase in church seating capacity of around 32 percent.

There continues to be a strong link between rainfall risk and membership in religious organizations among local communities likely to depend more on rainfall in 1860 and 1870. Rainfall risk is a statistically signiÖcant determinant of 1860 and 1870 church seating capacity among counties with population densities below the median and among counties with agriculture/manufacturing value added above the median, but not among more densely populated counties or among less agricultural counties. The link between rainfall risk and church seating capacity remains strongest among counties with agriculture/manufacturing value added above the median (the median share of agriculture over agriculture plus manufacturing is 0.89 in 1870 and 0.91 in 1860). In 1870, a one-standarddeviation increase in rainfall risk is associated with an increase in church seating capacity of around 50 percent among these agricultural counties. In 1860, the e§ect is close to 65 percent. In 1850, we do not Önd a statistically signiÖcant link between rainfall risk and church seating capacity. We argue that this is due to the smaller number of counties with the necessary data. In 1850, there is data for around 1450 counties compared to more than 1800 counties in 1860 and around 2650 counties in 1890. Moreover, most of the counties lost in 1850 compared to 1860 or 1890 are counties with low population densities and high agriculture/manufacturing value added ñ the type of counties where the link between rainfall risk and membership in religious organizations was statistically signiÖcant.

In 1910, 1920, and 1930, the US Census collected county-level data on the value of crops produced. This allows us to examine the e§ect of rainfall on agricultural productivity using a within-county approach for a period close to the late nineteenth century. The data on the value of crops can also be used to assess the relative importance for agricultural productivity of rainfall during the winter months compared to the months of spring, summer, and fall. Our results indicate that rainfall during the winter months has a weaker within-county e§ect on the value of crops than spring-fall rainfall, which is consistent with data on the agricultural growing and non-growing seasons at the beginning of the twentieth century (Covert, 1912). We use our theoretical model to map the relative importance of winter versus spring-fall rainfall for agricultural productivity into the relative importance of winter relative to spring-fall rainfall risk for religious membership. When we relate membership in religious organizations to winter rainfall risk, spring-fall rainfall risk, and a cross-season covariance term, we Önd that the statistically signiÖcant link is mostly with spring-fall rainfall risk.

Our work relates to a small literature on religion, insurance, and the welfare state as well as a large literature on insurance in historical agricultural societies and developing countries today. Gill and Lundsgaarde (2004) and Franck and Iannaccone (2009) Önd a negative e§ect of country-level welfare spending on church attendance in cross-sectional and panel data respectively. Scheve and Stasvange (2006) show that religiosity has a negative e§ect on preferences for social insurance in individual data and on social spending outcomes in crosscountry data. Dehejia, DeLeire, and Luttmer (2007) show that involvement with religious organizations results in better insurance against idiosyncratic income shocks using 1982-1998 US micro data. Hungerman (2005) Önds that the 1996 US welfare reform decreasing welfare to non-citizens crowded in member donations and community spending of Presbyterian congregations; Gruber and Hungerman (2007) show that the New Deal crowded out charitable spending of Christian denominations and Hungerman (2009) Önds that a Supreme Courtmandated expansion of social security insurance in 1991 crowded out charitable spending of United Methodist churches. The literature on insurance in historical agricultural societies and developing countries today ñ often characterized by high risk and little insurance supplied by governments or markets ñ has found evidence of insurance mechanisms ranging from the scattering of agricultural plots to reciprocal gift exchange, see the surveys of Alderman and Paxson (1994), Townsend (1995), and Dercon (2004) for example. This literature also contains interesting results on how the need for insurance in environments with high rainfall risk e§ects social outcomes. For example, Rosenzweig (1988a,b) and Durante (2010) ñ who also exploits the heterogeneous e§ect of rainfall on agricultural productivity over the growing year ñ show that rainfall risk ends up a§ecting family structure and trust respectively.

The rest of the paper is structured as follows. Section 2 presents a model of the value of partial insurance against idiosyncratic risk and membership in religious organizations in the presence of common background risk. Section 3 discusses our estimation framework and Section 4 the data and empirical results. Section 5 concludes.

2 A Model of Community Rainfall Risk, Idiosyncratic Risk, and Religious Membership

Religious membership may depend on the aggregate risk faced by a community because the value of partially insuring idiosyncratic risk through membership in religious organizations varies with aggregate risk. We show this in a model where farmers in a county are subject to both county-level economic risk ñ driven by rainfall risk ñand idiosyncratic economic risk. The model yields that membership in religious organizations is more attractive in counties subject to greater rainfall risk as long as farmersírelative risk aversion is above unity.

Consider a county c inhabited by a continuum of ex-ante identical farmers of measure one. Agricultural output produced by farmer in the county depends on Öxed county characteristics ; an idiosyncratic shock , and countylevel rainfall

\[\ln Y _ {f c} = \ln Z _ {c} + \ln s _ {f} + \beta \ln R _ {c}\tag{1}\]

where is a weighted average of monthly rainfall levels,

\[R _ {c} = \prod_ {m = 1} ^ {1 2} R _ {m c} ^ {\alpha_ {m}}\tag{2}\]

with . Hence, captures the percentage increase in agricultural output in response to a one-percent increase in monthly rainfall and that rainfall may matter more in some months than others.

Monthly rainfall levels in the county, , are taken to be jointly lognormally distributed. The distribution of the idiosyncratic shocks farmers face is independent of rainfall and log-normal with

The utility function of each farmer over consumption C is of the constant relative risk aversion type, where is the coe¢ cient of relative risk aversion. This implies that the expected utility of a farmer consuming output in (1) is

\[E U _ {c} = \frac {1}{1 - \rho} \left(\left(E Y _ {c}\right) ^ {(1 - \rho)} e ^ {- \rho (1 - \rho) (1 / 2) (\sigma^ {2} + \beta^ {2} R V a r _ {c})} - 1\right)\tag{3}\]

where measures rainfall risk in the county

\[R V a r _ {c} = \operatorname{Var} \left(\ln R _ {c}\right) = \operatorname{Var} \left(\sum_ {m = 1} ^ {1 2} \alpha_ {m} \ln R _ {m c}\right)\tag{4}\]

and is expected agricultural output in the county

\[E Y _ {c} = E \left(s Z _ {c} R _ {c} ^ {\beta}\right) = \delta Z _ {c} E \left(\prod_ {m = 1} ^ {1 2} R _ {m c} ^ {\beta \alpha_ {m}}\right)\tag{5}\]

with

If farmers were able to marginally reduce idiosyncratic consumption risk through mutual insurance, their utility gain would be M From (3) it follows that this marginal utility gain depends on rainfall risk and expected agricultural output in the county,

\[\ln M I B _ {c} = \mu - (\rho - 1) \ln E Y _ {c} + \frac {\rho (\rho - 1) \beta^ {2}}{2} R V a r _ {c}\tag{6}\]

with an unimportant function of preference and technology parameters: Hence, the marginal utility gain of mutually insuring idiosyncratic risk is increasing in rainfall risk in the county if the coe¢ cient of relative risk aversion is strictly greater than unity, : In this case, greater rainfall risk increases the beneÖts of mutually insuring idiosyncratic risk. Intuitively, this is because implies that the marginal utility gain from reducing risk is higher the more risk there is. Most estimates of the coe¢ cient of relative risk aversion in the literature are above unity, see for example Attanasio and Weber (1989), Vissing-Jorgensen and Attanasio (2003), and Chiappori and Paiella (2011). While these estimates all rely on post-World War II data, it seems reasonable to presume that risk aversion in the late nineteenth-century US ñwhen incomes were closer to subsistence levels and less government insurance was available ñwas at least as high.

We close the model by assuming that church membership provides partial insurance against idiosyncratic risk and that farmer cost of becoming a church member is . The cost could be decreasing in expected output as there may be more churches in richer counties or farmers in these counties may be able to a§ord better means of transport for example. But the cost of church membership may also be increasing in expected output because of a higher opportunity cost of church membership in richer counties or because more alternative social activities may be available in these counties. (Our empirical model also accounts for other county characteristics that may a§ect the cost of church membership, population density for example, but we do not include them here for simplicity.) With we capture individual heterogeneity in the cost of church membership. Combining the cost and the beneÖt of mutual insurance through church membership yields that farmers will become church members if and only if ln ln : Hence, church membership in county c will be

\[M _ {c} = G \left(\mu - (\theta + \rho - 1) \ln E Y _ {c} + \frac {\rho (\rho - 1) \beta^ {2}}{2} R V a r _ {c}\right)\tag{7}\]

where with the cumulative distribution function of As , church membership is greater the more rainfall risk there is in the county as long as

The agricultural production function in (1)-(2) allows for heterogenous effects of monthly rainfall. It is commonly assumed in the literature on the e§ect of weather on crop yields that rainfall matters more during the agricultural growing season than the non-growing season (e.g. Schlenker and Roberts, 2009). The US non-growing season varies by crops and states, see USDA (2007) for modern and Covert (1912) for historical data, but generally includes the three winter months.² We therefore examine the implications of a smaller e§ect of winter rainfall on agricultural output for rainfall risk and membership in religious organizations in our theoretical model. To do so, it is useful to collect the winter months in the set W = December, January, February and the months in spring, summer, and fall in the set S. The agricultural production function in (1)-(2) implies that the e§ect of a one-percent increase in monthly rainfall during winter on agricultural output is while the e§ect of a one-percent increase in monthly rainfall during spring-fall is _s where

\[\widetilde {\alpha} _ {w} = \sum_ {m \in W} \alpha_ {m} \text { and } \widetilde {\alpha} _ {s} = \sum_ {m \in S} \alpha_ {m}.\tag{8}\]

Hence, is a measure of the importance for agricultural productivity of rainfall during winter relative to spring-fall.

Using the notation in (8), the rainfall risk measure in (4) can be written in terms of rainfall risk during spring-fall, rainfall risk during winter, and a covariance term,

\[R V a r _ {c} = \widetilde {\alpha} _ {s} ^ {2} R V a r _ {c} ^ {s} + \widetilde {\alpha} _ {w} ^ {2} R V a r _ {c} ^ {w} + \widetilde {\alpha} _ {s} \widetilde {\alpha} _ {w} R C o v _ {c}\tag{9}\]

where and are analogues of (4) for spring-fall and winter respectively

\[{R V a r _ {c} ^ {s}} = {V a r \left(\sum_ {m \in S} \frac {\alpha_ {m}}{\widetilde {\alpha} _ {s}} \ln R _ {m c}\right)}\tag{10}\]

\[{R V a r _ {c} ^ {w}} = {V a r \left(\sum_ {m \in W} \frac {\alpha_ {m}}{\widetilde {\alpha} _ {w}} \ln R _ {m c}\right),}\tag{11}\]

and is twice the covariance of spring-fall and winter rainfall

\[R C o v _ {c} = 2 C o v \left(\sum_ {m \in S} \frac {\alpha_ {m}}{\widetilde {\alpha} _ {s}} \ln R _ {m c}, \sum_ {m \in W} \frac {\alpha_ {m}}{\widetilde {\alpha} _ {w}} \ln R _ {m c}\right).\tag{12}\]

It follows from (7) and the expression for rainfall risk in (9) that the importance of winter rainfall risk for membership in religious organizations relative to the importance of spring-fall rainfall risk is . Hence, if the e§ect of winter rainfall on agricultural output relative to the e§ect of spring-fall rainfall, is small, the e§ect of winter rainfall risk on membership in religious organizations relative to the e§ect of spring-fall rainfall risk should be even smaller.

² According to Covert (1912), the growing season for corn, cotton, and wheat (including winter wheat) went from March to November. The non-growing season would therefore have been the winter months December, January, and February.

3 Estimating the E§ect of Rainfall Risk on ligious Membership

Our empirical investigation of the link between rainfall risk and membership in religious organizations across US counties in the second half of the nineteenth century begins with a log-linearized version of (7),

\[\ln R e l M e m b e r _ {c} = \varphi + \lambda R V a r _ {c} + \gamma \ln E Y _ {c}\tag{13}\]

where is the number of church members or seats, and rainfall risk and expected agricultural output are deÖned in (4) and (5) respectively. The parameter of interest is the link between rainfall risk and membership in religious organizations ñwhich according to (7) should be positive as long as relative risk aversion is above unity.

To estimate the link between rainfall risk and membership in religious organizations using (13) we need measures for rainfall risk in (4) and expected agricultural output in (5) for all counties in our sample. This requires county-level data on rainfall over a su¢ ciently long period of time, as well as values for and . Our main analysis is for the case where monthly rainfall enters the agricultural production function in symmetrically, But we also examine the case where rainfall during the winter months has a di§erent e§ect on agricultural output.

Symmetric e§ects of monthly rainfall When monthly rainfall enters the agricultural production function in symmetrically, the rainfall risk measure in (4) becomes

\[\text { RainfallRisk } _ {c} = \text { Var } \left(\frac {1}{1 2} \sum_ {m = 1} ^ {1 2} \ln R _ {m c}\right)\tag{14}\]

and expected output in (5) can be written as

\[\ln E Y _ {c} = \ln \delta Z _ {c} + \ln E R _ {c} = \ln \delta Z _ {c} + \ln E \left(\prod_ {m = 1} ^ {1 2} R _ {m c} ^ {\frac {\beta}{1 2}}\right).\tag{15}\]

To get a sense for the average e§ect of rainfall on agricultural productivity ñduring the nineteenth century, we estimate the e§ect of rainfall on the countylevel value of crops reported by the US Census in 1910, 1920, and 1930. The availability of multiple observations for each county allows us to take a withincounty approach. Our estimating equation is based on (1),

\[\ln Y _ {c t} = \text { county FE } \& \text { time effects } + \beta \left(\frac {1}{1 2} \sum_ {m = 1} ^ {1 2} \ln R _ {m c t}\right),\tag{16}\]

where is the value of crops per unit of farmland. The county Öxed e§ects (FE) capture all Öxed county characteristics. The time e§ects capture changes over time and are allowed to vary by state. We also control for ln farmland and estimate speciÖcations with lagged rainfall and temperature controls.

Substituting (14) and (15) into (13) yields our estimating equation for the link between rainfall risk and membership in religious organizations

\[\ln R e l M e m b e r _ {c} = \lambda R a i n f a l l R i s k _ {c} + \gamma \ln E R _ {c} + \phi X _ {c}\tag{17}\]

where Rainf allRisk is deÖned in (14); ln captures the e§ect of rainfall on average output and is deÖned in (15) with estimated using (16); and stands for other county characteristics that may a§ect agricultural output or the cost of church membership, like soil quality, elevation, population, and area for example. _c and are calculated as the corresponding moments over the 1895-2000 period (the county rainfall data is only available since 1895).

When rainfall matters less during the winter months To get a sense for the link between membership in religious organizations on the one hand and rainfall risk during the winter months and during the spring-fall months on the other, we reestimate (17) after replacing the term for rainfall risk by

\[\lambda_ {s} R V a r _ {c} ^ {s} + \lambda_ {w} R V a r _ {c} ^ {w} + \delta R C o v _ {c}.\tag{18}\]

The variances and the covariance are deÖned in (10)-(12) and calculated as the corresponding moments over the 1895-2000 period. We take the e§ect of monthly rainfall during winter as well as during summer-fall to be symmetric, for and for

Our theoretical model implies that the importance of winter rainfall risk relative to spring-fall rainfall risk for religious membership in (17)-(18) should be linked to the importance of winter relative to spring-fall rainfall for agricultural productivity, ; see To get a sense for we return to the agricultural production function in (16) but break up the rainfall e§ect into a spring-fall term and a winter term

\[\text { Rainfall effect } = \beta_ {s} \left(\frac {1}{9} \sum_ {m \in S} \ln R _ {m c t}\right) + \beta_ {w} \left(\frac {1}{3} \sum_ {m \in W} \ln R _ {m c t}\right)\tag{19}\]

where and according to (1)-(2). Reestimating the agricultural production function in (16) after substituting (19) for the rainfall e§ect allows us to obtain as the e§ect of winter rainfall, relative to the e§ect of spring-fall rainfall,

4 Data and Empirical Results

4.1 Data

Religious Membership 1850-1890 The decennial US Census in 1850-1890 collected information on churches at the county level. There are two measures of membership in religious organizations, the seating capacity of churches in 1850, 1860, 1870, and 1890 (the 1880 data were never published) and the number of church members in 1890. Our data refers to all denominations listed in the Census. These data are retrieved from ICPSR Öle 2896 (Haines, 2006). For summary statistics see the Appendix Table.

Climate Data Our rainfall data come from PRISM, which provides monthly rainfall data since 1895 on a 4x4 km grid.³ PRISM was developed for the National Oceanic and Atmospheric Administration and the PRISM model is used by the US Department of Agriculture, NASA, and several professional weather channels.⁴ We map the PRISM grid into counties to obtain monthly rainfall at the county level. PRISM also provides data on monthly average temperature which we process analogously to the rainfall data.

Soil and Elevation Data We control for 53 soil types using the US Department of Agricultureís SSURGO database.⁵ We use these data to calculate the fraction of each countyís land area falling into the di§erent soil categories. The source of our elevation data is the Environmental System Research Institute.⁶ We calculate the fraction of each countyís land area falling into the following 11 bins: below 200 meters, 200 to 400 meters; 400 to 600 meters and so on up to 2000 meters; and above 2000 meters.

Other Data The data on land area, population, and value added in agriculture and manufacturing come from the US Census. Value added in manufacturing 1860-1890 is calculated as manufacturing output minus the cost of materials. Value added in agriculture is obtained as output minus the cost of fertilizers in 1890 and as output in agriculture in 1860-1870 as there is no information on fertilizer purchases. The data on the value of all crops produced and total farmland by county 1910-1930 also come from the US Census. The data are retrieved from ICPSR Öle 2896 (Haines, 2006).

4.2 Empirical Results

Table 1 contains our results on the e§ect of rainfall on the value of crops per unit of farmland from the US Census in 1910, 1920, and 1930 using the within-county estimation approach in (16). Our method of estimation is weighted least squares.

³ See www.prism.oregonstate.edu.

⁴ See Deschenes and Greenstone (2007) who also use the PRISM data.

⁵ See http://soils.usda.gov/surveys/geography/ssurgo/.

⁶ See www.esri.com.

We weight counties by their average farmland over the period as within-county changes in the value of crops per unit of farmland should be more closely related to county-level average rainfall when more land is under cultivation.⁷ The value of crops reported in the US Census corresponds to the year preceding the census year so that t in (16) refers to 1909, 1919, and 1929. The rainfall "year t" data we use in column (1) goes from December t 1 to November t. That is, the rainfall year t encompasses the four seasons ending in year t, which facilitates comparisons when we allow for separate e§ects of rainfall during winter months and during spring-fall months.⁸ Column (2) adds a control for the rainfall year t 1 which is deÖned analogously to rainfall year t and therefore goes from December t 2 to November t 1. The results in columns (1)-(2) indicate a statistically signiÖcant e§ect of rainfall in year t while the e§ect of rainfall in year t 1 is statistically insigniÖcant. The e§ect of rainfall in year t implies that a one-percent increase in monthly rainfall in year t raises the value of crops produced by around 0.5 percent. The average temperature controls added in column (3) are statistically insigniÖcant. This probably reáects in part that capturing the e§ect of temperature on agricultural productivity requires data on daily temperatures or even the distribution of temperature within a day, see Deschenes and Greenstone (2007) and Schlenker and Roberts (2009) for example.⁹ Such data are not available for our period of analysis.

Tables 2-5 contain our results on the link between rainfall risk and membership in religious organizations. The estimating equation is (17) and the method of estimation least squares. The left-hand-side variable is either ln church membership (1890) or ln church seating capacity (1890, 1870, and 1860). The right-hand-side control ln ER in (17) is calculated using a value for of 0.52 based on the results in Table 1. Other controls used are ln population and ln land area; the share of land of a given soil type using a 53-category soil classiÖcation system; the share of land at a given elevation using 11 elevation bins; average elevation; average temperature over the period 1895-2000; and state Öxed e§ects.

⁷ There are two reasons. First, idiosyncratic shocks to the output of di§erent units of farmland are more likely to average out in counties with more land under cultivation. Second, our measure of average rainfall refers to the average in a county as a whole, not the average in a countyís area under cultivation. The discrepancy between these two averages should tend to be smaller in counties with more land under cultivation, holding the share of a countyís land under cultivation constant. Moreover, the discrepancy should also tend to be smaller in counties with a larger share of their land under cultivation, and counties with more land under cultivation tend to have a larger share of their land under cultivation in our data. In any case, the unweighted results are similar to those in Table 1 in that all e§ects other than rainfall at t are statistically insigniÖcant; the e§ect of rainfall at t is statistically signiÖcant at the 1% level but smaller than in Table 1, 0.27 as compared to 0.52 in the speciÖcation in column (3). Using the value of 0.27 to obtain ln ER in (17) does not a§ect any of our Öndings on the link between rainfall risk and religious membership (the point estimates change by very little).

⁸ The 12 months of rainfall on the right-hand side of (16) would have also spanned the four seasons if we had used rainfall during the winter that starts in year t instead of the winter that ends in year t. But it seems more reasonable to presume that the value of crops in year t depends on rainfall during the winter that ends in year t than the winter that starts in year t:

⁹ When we allow for separate e§ects of winter and spring-fall weather in Table 6, the t 1 spring-fall temperature enters positively and statistically signiÖcantly. The magnitude of the spring-fall temperature e§ect is such that our theoretical model implies a minor role of temperature risk relative to rainfall risk for religious membership, see the discussion of Table 7 on page 14.

Table 2, column (1) shows that the link between rainfall risk and church membership in 1890 is statistically signiÖcant at the 1-percent level. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church membership of around 12 percent (the crosscounty standard deviation of rainfall risk is 0.054). Columns (2)-(3) split the full 1890 sample into counties with population densities below and above the median. Counties with relatively low population densities are more likely to depend mostly on agriculture.¹⁰ Hence, if rainfall risk a§ects church membership through economic risk in the agricultural sector as in our theoretical model, we would expect a link between rainfall risk and church membership among these counties. Column (2) conÖrms this prediction as the link between rainfall risk and church membership in below-median population density counties is statistically signiÖcant at the 1-percent level. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church membership of around 17 percent. On the other hand, the link between rainfall risk and church membership among counties with relatively high population densities ñwhich are often urban counties that rely mainly on manufacturing and related services ñ in column (3) is estimated imprecisely and statistically insigniÖcant.¹¹

Table 2, columns (4)-(5) show the results when the full 1890 sample is split into counties with value added in agriculture relative to manufacturing above and below the median. The median value-added share of agriculture over agriculture plus manufacturing in 1890 is 0.87 and the average share in counties with above-median agriculture/manufacturing is 0.95. Hence, counties with abovemedian agriculture are essentially agricultural. As the economic repercussions of rainfall variability are stronger in agriculture, we expect a link between rainfall risk and church membership among these agricultural counties. The result in column (4) shows that the link between rainfall risk and church membership among these counties is in fact statistically signiÖcant at the 1-percent level. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church membership of around 23 percent. On the other hand, among the less agricultural counties in column (5), there is no statistically signiÖcant link between rainfall risk and church membership (the average share of agriculture over agriculture plus manufacturing in these counties is 0.43).

Table 3 reestimates the speciÖcations in Table 2 using church seating capacity in 1890 as a measure of membership in religious organizations. The pattern of results is similar to the results for church membership. The link between rainfall risk and church seating capacity in the full sample in column (1) is statistically signiÖcant at the 1-percent level. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 10 percent. When we split the sample according to population density in columns (2)-(3), we Önd that the link between rainfall risk and church seating capacity among counties with population densities below the median in column (2) is statistically signiÖcant with a p-value of 0.051. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 13 percent. Among counties with relatively high population densities, the link between rainfall risk and church seating capacity is imprecisely estimated and statistically insigniÖcant. Columns (4)-(5) split the full sample into those with agriculture/manufacturing value added above the median and those with agriculture/manufacturing below the median. Among the most agricultural counties in column (4), the link between rainfall risk and church seating capacity is statistically signiÖcant a the 1-percent level. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 32 percent. Among the less agricultural counties in column (5), there is no statistically signiÖcant link between rainfall risk and church seating capacity.

^{1 0} Agricultural value added relative to agriculture plus manufacturing in below-median population density counties is larger than in the full sample, 0.82 as compared to 0.76, but the standard deviation is quite large (0.23).

^{1 1} Our Öndings on the link between rainfall risk and religious membership in Tables 2-5 are not a§ected when we also control for the variance of the annual average temperature over the 1895-2000 period. The temperature variance always enters statistically insigniÖcantly.

Table 4 summarizes our results on the link between rainfall risk and church seating capacity in 1870. This sample is around 20 percent smaller than the 1890 sample. The pattern of results is similar to that for 1890 church seating capacity. The link between rainfall risk and church seating capacity in the full sample in column (1) is statistically signiÖcant at the 5-percent level. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 13 percent, which is somewhat larger than the e§ect we estimated for 1890 church seating capacity. In columns (2)-(3) we consider the sample split according to population densities below and above the median. The link between rainfall risk and 1870 church seating capacity is statistically signiÖcant at the 1-percent level in counties with population densities below the median in column (2). The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 23 percent among these counties. On the other hand, the link between rainfall risk and church seating capacity among counties with relatively high population densities in column (3) is imprecisely estimated and statistically insigniÖcant. Columns (4)-(5) consider the sample split according to value added in agriculture relative to manufacturing above and below the median. The median share of agriculture over agriculture plus manufacturing in 1870 is 0.89. The link between rainfall risk and church seating capacity among the agricultural counties in column (4) is statistically signiÖcant at the 5-percent level. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 65 percent among these counties. Rainfall risk shows a weaker, but still statistically signiÖcant, link with 1870 church seating capacity among counties with agriculture/manufacturing value added below the median in column (5).

Table 5 contains our results on the link between rainfall risk and church seating capacity in 1860. The sample is around 30 percent smaller than the 1890 sample and 10 percent smaller than the 1870 sample. Results are similar to those for 1870 and 1890 church seating capacity. The link between rainfall risk and church seating capacity in the full sample in column (1) is statistically signiÖcant with a p-value of 0.054. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 12 percent, which is similar to the Önding for 1870 church seating capacity and somewhat larger than the estimate for 1890 church seating capacity. The sample split according to population densities below and above the median is in columns (2)-(3). The link between rainfall risk and 1860 church seating capacity is statistically signiÖcant at the 5-percent level in counties with population densities below the median in column (2). The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 13 percent, which is similar to the Önding for 1890 church seating capacity and somewhat smaller than the estimate for 1870 church seating capacity. On the other hand, the link between rainfall risk and church seating capacity among counties with relatively high population densities in column (3) is imprecisely estimated and statistically insigniÖcant. Columns (4)-(5) report the results of the sample split according to value added in agriculture relative to manufacturing above and below the median. In 1860, the median share of agriculture over agriculture plus manufacturing is 0.91. The link between rainfall risk and church membership is statistically signiÖcant at the 10-percent level among the agricultural counties in column (4) with a p-value of 0.082. The point estimate implies that a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of almost 65 percent. On the other hand, rainfall risk does not show a statistically signiÖcant link with church membership among counties with agriculture/manufacturing value added below the median in column (5).

In 1850, we do not Önd a statistically signiÖcant link between rainfall risk and church seating capacity. We attribute this to the smaller number of counties. The necessary data are available for approximately 1450 counties in 1850 compared to around 1820 counties in 1860; around 2070 counties in 1870; and around 2650 counties in 1890. Moreover, most of the counties lost in 1850 compared to 1860, 1870, or 1890 are counties with low population density and high agriculture/manufacturing value added ñ the type of counties where the link between rainfall risk and religious membership was statistically signiÖcant. The consequence of the drop in sample size between 1860 and 1850 can be illustrated by reestimating the link between rainfall risk and church seating capacity in the 1860 subsample of counties for which there is data in 1850. This always yields statistically insigniÖcant estimates.

Table 6 examines how the e§ect of rainfall on the value of crops per unit of farmland in Table 1 changes when we distinguish between rainfall during the winter months on the one hand and during the months of spring, summer, and fall on the other. Column (1) reproduces column (2) of Table 1 where we control for rainfall in year t as well as year t 1. In column (2) we split rainfall during year t as well as during year t 1 into winter rainfall on the one hand and spring-fall rainfall on the other. The speciÖcation of the rainfall e§ects follows (19) but we also allow for lagged e§ects. The estimates can be interpreted as the e§ects on agricultural productivity of a one-percent increase in monthly rainfall during spring-fall and during winter in years t and t 1 respectively. We Önd that both spring-fall rainfall in year t and rainfall in the winter ending in year t have a statistically signiÖcant positive e§ect. A one-percent increase in monthly rainfall during spring-fall raises agricultural productivity by 0.33 percent and a one-percent increase in winter rainfall raises productivity by 0.15 percent. For year t 1, only spring-fall rainfall is statistically signiÖcant and enters positively. A one-percent increase in monthly rainfall during spring-fall of year t 1 increases agricultural productivity in year t by 0.28 percent. The results in columns (3)-(4) show that the e§ects of rainfall on agricultural productivity change little when we control for temperature.¹²

Table 7 summarizes our results on the link between winter and spring-fall rainfall risk on the one hand and membership in religious organizations on the other. The estimating equation is (17) with the rainfall risk term replaced by (18). The control variables are the same as in Tables 2-5. As rainfall during the spring-fall months is a signiÖcant determinant of agricultural productivity according to our results in Table 6, we should expect a link between spring-fall rainfall risk and religious membership. Winter rainfall matters less for agricultural productivity than spring-fall rainfall in Table 6 and we should therefore expect winter rainfall risk to matter less for membership in religious organizations according to our theoretical model. To get an idea of how much less winter rainfall risk should matter, note that (7), (9), and (19) imply that the importance of winter relative to spring-fall rainfall risk for membership in religious organizations can be calculated as with and the e§ects on agricultural productivity of winter and spring-fall rainfall respectively. However, in contrast to our estimating equation, our theoretical model did not feature lagged rainfall e§ects on agricultural productivity. When such lagged e§ects are incorporated into the theoretical model, the relative importance of winter compared to spring-fall rainfall risk for religious membership also depends on the correlation of rainfall in di§erent years. In our data, the correlation of rainfall in di§erent years is close to zero. The relative importance of winter relative to spring-fall rainfall risk implied by the theoretical model is therefore approximately with subscripts t and t 1 denoting year t and year t 1 rainfall e§ects respectively. This ratio is 0.11 according to the statistically signiÖcant rainfall estimates in column (4) of Table 6.¹³ Hence, winter rainfall risk should matter approximately an order of magnitude less for membership in religious organizations than spring-fall rainfall risk according to our theoretical model.¹⁴

^{1 2} Reestimating the speciÖcations in Table 6 without weighting by farmland also yields that rainfall during the winter months matters less than rainfall during the months of springfall. The main di§erences with Table 6 are that rainfall during spring-fall is only statistically signiÖcant in year t 1; and that rainfall during the winter ending in year t enters signiÖcantly positively while rainfall during the winter ending in year t 1 enters signiÖcantly negatively.

Table 7, column (1) presents our results on the link between spring-fall and winter rainfall risk on the one hand and 1890 church membership on the other. The link between spring-fall rainfall risk and church membership is statistically signiÖcant at the 1-percent level. The link between winter rainfall risk and church membership is statistically signiÖcant at the 5-percent level with a point estimate that is around 28 percent of the estimate on spring-fall rainfall risk. When we examine the link between spring-fall and winter rainfall risk on the one hand and 1890 church seating capacity on the other, we Önd a link between spring-fall rainfall risk and church seating capacity that is statistically signiÖcant at the 5-percent level. The link between winter rainfall risk and church seating capacity is statistically insigniÖcant. The results for 1870 and 1860 church seating capacity are similar to those for 1890. The link between spring-fall rainfall risk and church seating capacity is statistically signiÖcant, while the link between winter rainfall risk and church seating capacity is statistically insigniÖcant.¹⁵ The covariance term is statistically insigniÖcant in all speciÖcations except for 1870 church seating capacity.

5 Conclusion

One way to look at religious organizations is as mutual assistance networks that are supported by a spiritual framework rewarding aid and charity. Such networks would have been especially valuable in communities facing substantial risk but lacking formal insurance structures, such as historical agricultural societies for example (McCleary and Barro, 2006).

Insurance of idiosyncratic risk provided by religious organizations should be more valuable in communities facing greater aggregate risk, which would make membership in religious organizations more attractive in high-risk environments. We investigate the link between membership in religious organizations and community risk across US counties in the second half of the nineteenth century. The

^{1 3} The same approach can be used to calibrate the importance of spring-fall temperature risk ñthe variance over time of average spring-fall temperature ñ for religious membership relative to the importance of spring-fall rainfall risk. In this case the formula is with the e§ect of t 1 spring-fall temperature on agricultural output. Using the statistically signiÖcant estimates in column (4) of Table 6 yields 0.056 which indicates that temperature risk should be considerably less important for religious membership than rainfal risk. When we add a control for the spring-fall temperature variance over the 1895-2000 period in our religious membership regressions in Table 7, it always enters statistically insigniÖcantly.

^{1 4} Reestimating the speciÖcations in Table 6 without weighting by farmland implies a greater role of rainfall during winter for the value of crops and therefore a greater role of winter rainfall risk for religious membership. In this case the theoretical model implies that the e§ect of winter rainfall risk on religious membership should be about half the e§ect of spring-fall rainfall risk.

^{1 5} Our Öndings on the link between rainfall risk and religious membership are una§ected when we also control for the variance of spring-fall average temperature over the 1895-2000 period.

US started into the second half of the nineteenth century as a mostly agricultural economy with little social insurance supplied by the government. As agriculture was mainly rainfed, local economies were subject to rainfall risk. It is possible to quantify the temporal variability of rainfall at the county level since 1895, and there is county-level data on the seating capacity of churches from 1850 to 1890 and their membership in 1890. This allows us to investigate whether membership in religious organizations in the second part of the nineteenth century was greater in local communities likely to have been facing more rainfall risk.

We Önd that church seating capacity in 1860-1890 and church membership in 1890 were signiÖcantly larger in US counties likely to be subject to more rainfall risk. This e§ect is present among the most agricultural counties and among counties with low population densities, but not among less agricultural or more densely populated counties. Among the most agricultural counties, a one-standard-deviation increase in rainfall risk is associated with an increase in church seating capacity of around 32 percent in 1890 and 65 percent in 1860.

References

1. Alderman, H. and C.H. Paxson (1994), "Do the Poor Insure? A Synthesis of the Literature on Risk and Consumption in Developing Countries." In Bacha (ed.) "Economics in a Changing World," MacMillan, UK.
2. Attanasio, O.P. and G. Weber (1989), "Intertemporal Substitution, Risk Aversion, and the Euler Equation for Consumption." The Economic Journal, Volume 99 (Conference 1989), pp. 59-73.
3. Belcher, J.R. and C.J. Tice (2011), "Protestant Church Charity: History, Trends, and Implications." Journal of Religion & Spirituality in Social Work, Volume 30, Issue 2, pp. 164-177.
4. Bodnar, J. (1985), "The Transplanted: A History of Immigrants in Urban America." Indiana University Press, USA.
5. Bremner, R.H. (1994), "Giving: Charity and Philanthropy in History." Transaction Publishers, USA.
6. Chiappori, P.A. and M. Paiella (2011), "Relative Risk Aversion is Constant: Evidence from Panel Data." Journal of the European Economic Association, Volume 9, Issue 6, pp. 1021-1052.
7. Covert, J. (1912), "Seedtime and Harvest: Cereals, Flax, Cotton, and Tobacco." United States Department of Agriculture, Bureau of Statistics Bulletin, Washington (DC), USA.
8. Dehejia, R., T. DeLeire, and E.F.P. Luttmer (2007), "Insuring Consumption and Happiness Through Religious Organizations." Journal of Public Economics, Volume 91, Issues 1-2, pp. 259-279.
9. Dercon, S. (2004), "Risk, Insurance, and Poverty: A Review." In Dercon (ed.) "Insurance Against Poverty," Oxford University Press, UK.
10. Deschenes, O. and M. Greenstone (2007), "The Economic Impacts of Climate Change: Evidence from Agricultural Output and Random Fluctuations in Weather." American Economic Review, Volume 97, Number 1, pp. 354- 385.
11. Durante, R. (2010), "Risk, Cooperation, and the Economic Origins of Social Trust: An Empirical Investigation." Mimeo.
12. Franck, R. and L. Iannaccone (2011), "Why did Religiosity Decrease in the Western World During the Twentieth Century?" Mimeo.
13. Gallman, R.E. (1986), "The United States Capital Stock in the Nineteenth Century." In Engerman and Gallman (eds.) "Long Term Factors in American Growth." NBER, USA.
14. Gill, A. and E. Lundsgaarde (2004), "State Welfare Spending and Religiosity." Rationality and Society, Volume 16, Number 4, pp. 399-436.
15. Gjerde, J. (1985), "From Peasants to Farmers. The Migration from Balestrand, Norway to the Upper Middle West." Cambridge University Press, UK.
16. Haines, M.R. (2006), "Historical, Demographic, Economic, and Social Data: The United States 1790-2000 [Computer File]." Inter-University Consortium for Political and Social Research Study Number 2896.
17. Lindert, P.H. (2004), "Growing Public: Social Spending and Economic Growth Since the Eighteenth Century." Cambridge University Press, UK.
18. McCleary, R.M. and R.J. Barro (2006), "Religion and Economy." Journal of Economic Perspectives, Volume 20, Number 2, pp. 49-72.
19. Overacker, I. (1998), "The African American Church Community in Rochester, New York, 1900-1940." University of Rochester, USA.
20. Parker, C.H. (1998), "The Reformation of Community ñ Social Welfare and Calvinist Charity in Holland, 1572-1620." Cambridge University Press, UK.
21. Pullan, B. (1998), "Support and Redeem: Charity and Poor Relief in Italian Cities from the Fourteenth to the Seventeenth Century." Continuity and Change, Volume 3, pp. 177-208.
22. Pullan, B. (2005), "Catholics, Protestants, and the Poor in Early Modern Europe." The Journal of Interdisciplinary History, Volume 35, Number 3, pp. 441-456.
23. Rosenzweig M.R. (1988a), "Risk, Private Information, and the Family." American Economic Review, Volume 78, Number 2, pp. 245-250.
24. Rosenzweig, M.R. (1988b), "Risk, Implicit Contracts, and the Family in Rural Areas of Low Income Countries." The Economic Journal, Volume 98, pp. 1148-1170.
25. Scheve, K. and D. Stasavage (2006), "Religion and Preferences for Social Insurance." Quarterly Journal of Political Science, Volume 1, pp. 255-286.
26. Schlenker, W. and M.J. Roberts (2009) "Nonlinear Temperature E§ects Indicate Severe Damages to US Crop Yields under Climate Change." PNAS, Volume 106, Number 37, pp. 15594-15598.
27. Szasz, F.M. (2004), "The Protestant Clergy in the Great Plains and Mountain West, 1865-1915." University of Nebraska Press, USA.
28. Townsend, R.M. (1995), "Consumption Insurance: An Evaluation of Risk Bearing Systems in Low Income Economies." Journal of Economic Perspectives, Volume 9, Number 3, pp. 83-102.
29. Trattner, W.I. (1974), "From Poor Law to Welfare State." Free Press, USA.
30. USDA (1923), Agriculture Yearbook, United States Department of Agriculture, Washington (DC), USA.
31. USDA (1925), Agriculture Yearbook, United States Department of Agriculture, Washington (DC), USA.
32. USDA (2007), "Usual Planting and Harvesting Dates for US Field Crops." Agricultural Handbook Number 628, United States Department of Agriculture, Washington (DC), USA.
33. Vissing-Jorgensen, A. and O.P. Attanasio (2003), "Stock-Market Participation, Intertemporal Substitution, and Risk Aversion." American Economic Review, Volume 93, Number 2, pp. 383-391.

Table 1: Rainfall and the Value of Crops Produced in 1909, 1919, and 1929

Notes: The left-hand-side variable is the ln value of crops per unit of farmland produced at the county level in 1909, 1919, and 1929. The estimating equation employed in column (1) is (16), see Section 3 for more details. Columns (2)-(4) add controls for lagged rainfall and temperature. See Section 4.1 for the data sources and Section 4.2 pages 9-10 for more details on the specification of the rainfall and temperature controls. The method of estimation is weighted least squares with weights equal to the average farmland of counties over the period, for details see Section 4.2 especially footnote 7. All specifications control for ln farmland; time effects; and county fixed effects. The time effects are allowed to vary by state. Standard errors account for arbitrary heteroskedasticity and are clustered at the county level. ***, **, and * denote significance at the 1%, 5%, and 10% level respectively.

Table 2: Rainfall Risk and Church Membership in 1890

Notes: The left-hand-side variable is ln church membership at the county level in 1890. The estimating equation employed is (17). The right-hand-side measure of rainfall risk is defined in (14) and calculated with 1895-2000 rainfall data. These rainfall data are also used to obtain lnER which captures the effect of rainfall on average output and is defined in (15) with β equal to 0.52 (this value comes from Table 1). See Section 3 for more details on the specification and Section 4.1 for data sources. Other righthand-side controls used are ln population and ln land area of the county (size); the share of the land of a given soil type using a 53-category soil classification system; the share of the land at a given elevation using 11 elevation bins; average elevation; average temperature over the period 1895-2000; and state fixed effects. The method of estimation is least squares. Standard errors account for arbitrary heteroskedasticity and are clustered at the state level. ***, **, and * denote significance at the 1%, 5%, and 10% level respectively.

Table 3: Rainfall Risk and Church Seating Capacity in 1890

Notes: The left-hand-side variable is ln church seats at the county level in 1890. The estimating equation employed is (17). The right-hand-side measure of rainfall risk is defined in (14) and calculated with 1895-2000 rainfall data. These rainfall data are also used to obtain lnER which captures the effect of rainfall on average output and is defined in (15) with β equal to 0.52 (this value comes from Table 1). See Section 3 for more details on the specification and Section 4.1 for data sources. Other right-hand side controls used are ln population and ln land area of the county (size); the share of the land of a given soil type using a 53-category soil classification system; the share of the land at a given elevation using 11 elevation bins; average elevation; average temperature over the period 1895-2000; and state fixed effects. The method of estimation is least squares. Standard errors account for arbitrary heteroskedasticity and are clustered at the state level. ***, **, and * denote significance at the 1%, 5%, and 10% level respectively.

Table 4: Rainfall Risk and Church Seating Capacity in 1870

Notes: The left-hand-side variable is ln church seats at the county level in 1870. The estimating equation employed is (17). The right-hand-side measure of rainfall risk is defined in (14) and calculated with 1895-2000 rainfall data. These rainfall data are also used to obtain lnER which captures the effect of rainfall on average output and is defined in (15) with β equal to 0.52 (this value comes from Table 1). See Section 3 for more details on the specification and Section 4.1 for data sources. Other right-hand-side controls used are ln population and ln land area of the county (size); the share of the land of a given soil type using a 53-category soil classification system; the share of the land at a given elevation using 11 elevation bins; average elevation; average temperature over the period 1895-2000; and state fixed effects. The method of estimation is least squares. Standard errors account for arbitrary heteroskedasticity and are clustered at the state level. ***, **, and * denote significance at the 1%, 5%, and 10% level respectively.

Table 5: Rainfall Risk and Church Seating Capacity in 1860

Notes: The left-hand-side variable is ln church seats at the county level in 1860. The estimating equation employed is (17). The right-hand-side measure of rainfall risk is defined in (14) and calculated with 1895-2000 rainfall data. These rainfall data are also used to obtain lnER which captures the effect of rainfall on average output and is defined in (15) with β equal to 0.52 (this value comes from Table 1). See Section 3 for more details on the specification and Section 4.1 for data sources. Other righthand-side controls used are ln population and ln land area of the county (size); the share of the land of a given soil type using a 53-category soil classification system; the share of the land at a given elevation using 11 elevation bins; average elevation; average temperature over the period 1895-2000; and state fixed effects. The method of estimation is least squares. Standard errors account for arbitrary heteroskedasticity and are clustered at the state level. ***, **, and * denote significance at the 1%, 5%, and 10% level respectively.

Table 6: Rainfall and the Value of Crops Produced in 1909, 1919, and 1929

Notes: The left-hand-side variable is the ln value of crops per unit of farmland at the county level in 1909, 1919, and 1929. The estimating equation is (16) with the rainfall term in columns (2) and (4) split into rainfall over the winter months and the spring-fall months as in (19), see pages 9-10 and 13-14. All specifications control for ln farmland; time effects; and county fixed effects. See the notes to Table 1 for more details on the specification and data. Columns (1) and (3) are reproduced from Table 1. ***, **, and * denote significance at the 1%, 5%, and 10% level respectively.

Table 7: Rainfall Risk in Spring-Fall and Winter

Notes: The left-hand-side variable is ln church membership or ln church seats at the county level from the US Census in 1890, 1870, or 1860. The estimating equation employed is (17) with the rainfall risk term replaced by (18). The right-hand-side measures of rainfall risk are defined in (10)-(12) with symmetric effects of monthly rainfall during winter and during spring-fall and obtained with 1895-2000 rainfall data, see page 8 below (18). All other right-hand-side variables are defined as in Table 2. The method of estimation is least squares. Standard errors account for arbitrary heteroskedasticity and are clustered at the state level. ***, **, and * denote significance at the 1%, 5%, and 10% level respectively.

Appendix Table: Summary Statistics

Panel A: Full sample

Panel B.1: Counties with population density above the median

Panel B.2: Counties with population density below the median

Panel C.1: Counties with agriculture/manufacturing value added above the median

Panel C.2: Counties with agriculture/manufacturing value added below the median

References

1. 2013-17: “Rainfall Risk and Religious Membership in the Late Nineteenth-Century US”, Philipp Ager y Antonio Ciccone.

References

1. 2013-16: “Immigration in Europe: Trends, Policies and Empirical Evidence”, Sara de la Rica, Albrecht Glitz y Francesc Ortega.

References

1. 2013-15: “The impact of family-friendly policies on the labor market: Evidence from Spain and Austria”, Sara de la Rica y Lucía Gorjón García.

References

1. 2013-14: “Gender Gaps in Performance Pay: New Evidence from Spain”, Sara de la Rica, Juan J. Dolado y Raquel Vegas.

References

1. 2013-13: “On Gender Gaps and Self-Fulfilling Expectation: Alternative Implications of Paid-For Training”, Juan J. Dolado, Cecilia García-Peñalosa y Sara de la Rica.

References

1. 2013-12: “Financial incentives, health and retirement in Spain”, Pilar García‐Gómez, Sergi Jiménez‐Martín y Judit Vall Castelló.

References

1. 2013-11: “Gender quotas and the quality of politicians”, Audinga Baltrunaite, Piera Bello, Alessandra Casarico y Paola Profeta.

References

1. 2013-10: “Brechas de Género en los Resultados de PISA :El Impacto de las Normas Sociales y la Transmisión Intergeneracional de las Actitudes de Género”, Sara de la Rica y Ainara González de San Román.

References

1. 2013-09: “¿Cómo escogen los padres la escuela de sus hijos? Teoría y evidencia para España”, Caterina Calsamiglia, Maia Güell.

References

1. 2013-08: “Evaluación de un programa de educación bilingüe en España: El impacto más allá del aprendizaje del idioma extranjero”, Brindusa Anghel, Antonio Cabrales y Jesús M. Carro.

References

1. 2013-07: “Publicación de los resultados de las pruebas estandarizadas externas: ¿Tiene ello un efecto sobre los resultados escolares?”, Brindusa Anghel, Antonio Cabrales, Jorge Sainz e Ismael Sanz.

References

1. 2013-06: “DYPES: A Microsimulation model for the Spanish retirement pension system”, F. J. Fernández-Díaz, C. Patxot y G. Souto.

References

1. 2013-05: “Vertical differentiation, schedule delay and entry deterrence: Low cost vs. full service airlines”, Jorge Validoa, M. Pilar Socorroa y Francesca Medda.

References

1. 2013-04: “Dropout Trends and Educational Reforms: The Role of the LOGSE in Spain”, Florentino Felgueroso, María Gutiérrez‐Domènech y Sergi Jiménez‐Martín.

References

1. 2013-03: “Understanding Different Migrant Selection Patterns in Rural and Urban Mexico”, Simone Bertoli, Herbert Brücker y Jesús Fernández-Huertas Moraga.

References

1. 2013-02: “Understanding Different Migrant Selection Patterns in Rural and Urban Mexico”, Jesús Fernández-Huertas Moraga.

References

1. 2013-01: “Publicizing the results of standardized external tests: Does it have an effect on school outcomes?, Brindusa Anghel, Antonio Cabrales, Jorge Sainz y Ismael Sanz.

References

1. 2012-12: “Visa Policies, Networks and the Cliff at the Border”, Simone Bertoli, Jesús Fernández-Huertas Moraga.

References

1. 2012-11: “Intergenerational and Socioeconomic Gradients of Child Obesity”, Joan Costa-Fonta y Joan Gil.

References

1. 2012-10: “Subsidies for resident passengers in air transport markets”, Jorge Valido, M. Pilar Socorro, Aday Hernández y Ofelia Betancor.

References

1. 2012-09: “Dual Labour Markets and the Tenure Distribution: Reducing Severance Pay or Introducing a Single Contract?”, J. Ignacio García Pérez y Victoria Osuna.

References

1. 2012-08: “The Influence of BMI, Obesity and Overweight on Medical Costs: A Panel Data Approach”, Toni Mora, Joan Gil y Antoni Sicras-Mainar.

References

1. 2012-07: “Strategic behavior in regressions: an experimental”, Javier Perote, Juan Perote-Peña y Marc Vorsatz.

References

1. 2012-06: “Access pricing, infrastructure investment and intermodal competition”, Ginés de Rus y M. Pilar Socorro.

References

1. 2012-05: “Trade-offs between environmental regulation and market competition: airlines, emission trading systems and entry deterrence”, Cristina Barbot, Ofelia Betancor, M. Pilar Socorro y M. Fernanda Viecens.

References

1. 2012-04: “Labor Income and the Design of Default Portfolios in Mandatory Pension Systems: An Application to Chile”, A. Sánchez Martín, S. Jiménez Martín, D. Robalino y F. Todeschini.

References

1. 2012-03: “Spain 2011 Pension Reform”, J. Ignacio Conde-Ruiz y Clara I. Gonzalez.

References

1. 2012-02: “Study Time and Scholarly Achievement in PISA”, Zöe Kuehn y Pedro Landeras.

References

1. 2012-01: “Reforming an Insider-Outsider Labor Market: The Spanish Experience”, Samuel Bentolila, Juan J. Dolado y Juan F. Jimeno.

References

1. 2011-13: “Infrastructure investment and incentives with supranational funding”, Ginés de Rus y M. Pilar Socorro.

References

1. 2011-12: “The BCA of HSR. Should the Government Invest in High Speed Rail Infrastructure?”, Ginés de Rus.

References

1. 2011-11: “La rentabilidad privada y fiscal de la educación en España y sus regiones”, Angel de la Fuente y Juan Francisco Jimeno.

References

1. 2011-10: “Tradable Immigration Quotas”, Jesús Fernández-Huertas Moraga y Hillel Rapoport.