Agriculture, Transportation and the Timing of Urbanization

Agriculture, Transportation and the Timing of Urbanization Global Analysis at the Grid Cell Level Mesbah Motamed Raymond Florax William Masters Department of Agricultural Economics Purdue University SHaPE Seminar August 31, 2012

Introduction This paper addresses the link between a location s geography and its date of economic transition We use newly available high resolution grid cell data covering geographic variables related to agriculture and transportation and match it to historic population data We find statistically significant and economically important relationships between a location s geographic features and its date of urbanization

Previous literature From Malthusian stagnation to modern growth Two-sector model Economic development Related to literature on historic roots of development: Diamond s Guns, Germs and Steel, and work on geography and institutions Growth literature focusing on contemporary effects of geography We focus on urbanization, with subsequent effects of economic growth

Previous literature From Malthusian stagnation to modern growth Two-sector model From Malthusian stagnation to modern growth Different locations enter their modern growth paths at different times Agricultural productivity is seen as the key to overall economic growth 20000 80 1990 international dollars 15000 10000 5000 Urban Population Density 60 40 20 0 1600 1700 1800 1900 2000 year Western Europe Africa Latin America Asia (Excluding Japan) Regional per capita GDP, 1600-2000 AD, Maddison (2001) 0 1500 1600 1700 1800 1900 2000 Year High Suitability Land Low Suitability Land Urban density growth conditional on cultivation suitability level

Previous literature From Malthusian stagnation to modern growth Two-sector model A simple two-sector model of ag-driven urbanization Matsuyama s (1992) two sector, closed economy Farmers N a and manufacturers N m N a + N m = 1 Each activity s output is a function of labor and productivity Y a = A a N a and Y m = A m N m Urban producers profits are: π m = (1 t) A m N m wn m Preferences are non-homothetic U (C a, C m ) = ln (C a a) + ln (C m )

Previous literature From Malthusian stagnation to modern growth Two-sector model A simple two-sector model of ag-driven urbanization Labor is mobile, so the wage paid to workers is equal across activities Resulting equation for urbanization in equilibrium is: N m = 1 t 2 t (1 aaa ) Geography appears in A a and t A a : cultivation suitability, frost in winter t: distance to coast, navigable river, elevation

Previous literature From Malthusian stagnation to modern growth Two-sector model A simple two-sector model of ag-driven urbanization Improvements in agricultural productivity drive workers into urban settings N m / A a > 0 Reductions in transport costs also drive urbanization N m / t > 0 if A a > a t low t high N m A a

High resolution grid cell data Observing data in a new way Urbanization data Geographic data Summary statistics High resolution 62,290 observations covering Earth s land surface Observational area ranges from 1,000 to 3,000 sq. km. (depending on latitude). Regular grid cells 0.5 latitude by 0.5 longitude not subject to endogeneity of administrative boundaries Therefore, we can control for Country fixed effects Neighborhood effects, using spatial econometric techniques

Observing data in a new way Urbanization data Geographic data Summary statistics Economic transition and urbanization We capture transition by observing urbanization. Advantages include measurable level, identifiable location, persistent effects Klein-Goldewijk s (2005) time series of grid cell-level urban and rural populations Constructed from historical records and estimates of population growth rates We use these data to identify the date at which a location s urban population reaches a particular threshold Urban population densities of 1 and 5.67 inhabitants per sq. km Urban population fractions of 10%, 25%, and 50%

Agricultural geography Observing data in a new way Urbanization data Geographic data Summary statistics Cultivation suitability index due to Ramankutty et al. (2002) Based on soil ph, carbon density, potential and actual evapotranspiration, and growing degree days Winter frost, following Masters and McMillan (2001) 30 year average monthly frost reported by IPCC (2002) Captures a location s pest and disease resistance Reported as a dummy variable for a location that has 2.11 days of winter frost following a frost-free summer

Transportation geography Observing data in a new way Urbanization data Geographic data Summary statistics Distance to nearest coast distance in km from each grid cell to the nearest ocean or unbounded sea coast Presence of a navigable river Strahler Index of river orders from 1 to 6 (Vorosmatry et al., 2000) Dummy variable coded for river orders greater than 4 Elevation Captures topographical barriers to agriculture and transport Digital elevation map reported in meters by Hijmans et al. (2005)

Summary statistics Introduction Observing data in a new way Urbanization data Geographic data Summary statistics variable mean std. dev. min max number of years since the urban population reached: 1 inhabitant per sq. km. 29.640 109.622 0 2000 5.67 inhabitant per sq. km. 15.859 61.893 0 1600 10% urbanization 34.169 94.903 0 1800 25% urbanization 19.829 53.716 0 1100 50% urbanization 9.690 31.038 0 700 cultivation suitability index 0.264 0.318 0 1 distance to coast (km) 521.984 511.641 1 2514.704 presence of navigable river 0.029 0.168 0 1 presence of frost in winter 0.116 0.320 0 1 land elevation (meters) 594.458 772.698-76.667 5717.111 62,290 observations

Economic transition and urbanization Observing data in a new way Urbanization data Geographic data Summary statistics Number of years since transition to a 10% urban population

Cultivation suitability index Observing data in a new way Urbanization data Geographic data Summary statistics Worldwide distribution of land suitability for cultivation

Observing data in a new way Urbanization data Geographic data Summary statistics Plotting urbanization and geography.5.4 year 2000 year 2000.4 urban fraction of population.3.2 year 1970 year 1950 urban fraction of population.3.2 year 1970 year 1950 year 1900.1 year1900.1 year 1800 year 1800 0 0 0.2.4.6.8 1 0 2 4 6 8 cultivation suitability index natural log of distance to coast Local polynomial regression of urbanization on cultivation suitability index Local polynomial regression of urbanization on natural log of distance to coast

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Specifying a model Strategy: regress the time since a location reached each urbanization threshold on the geographic variables. ln T ij = β 0 + β 1 ln cultiv ij + β 2 ln coast ij + β 3 river ij + β 4 frost ij + β 5 ln elevation ij + δ j + ɛ ij T ij is number of years elapsed since grid cell i in country j reached the threshold cultiv ij is the cultivation suitability index coast ij and river ij are navigable waterway variables δ j captures country fixed effects

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Results from OLS (1) (2) (3) (4) (5) 1 inhabitant 5.67 inhabitants 10% 25% 50% per sq. km. per sq. km. urban urban urban cultiv 0.329 0.229 0.303 0.246 0.158 coast 0.130 0.253 0.131 0.149 0.148 river 0.570 0.493 0.913 0.809 0.647 frost 0.847 0.686 0.821 0.736 0.544 elevation 0.089 0.084 0.097 0.045 0.008 constant 2.116 0.991 1.878 0.552-0.735 R 2 0.355 0.354 0.274 0.248 0.196 n = 62, 290. All coefficients are significant at the p < 0.01 level. All regressions use country fixed effects (not shown).

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Marginal effects from Tobit (1) (2) (3) (4) (5) 1 inhabitant 5.67 inhabitants 10% 25% 50% per sq. km. per sq. km. urban urban urban cultiv 0.446 0.309 0.356 0.297 0.226 coast 0.184 0.130 0.185 0.196 0.199 river 0.493 0.406 0.751 0.697 0.647 frost 0.245 0.414 0.209 0.252 0.292 elevation 0.001 0.026 0.044 0.035 0.326 Note: n = 62, 290 All coefficients are significant at the p < 0.01 level, except elevation in Model (1) which is not significant. For now, only Model (2) reflects country dummies.

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Spatial diagnostics and unobserved neighborhood effects We expect a location s values to be correlated with its neighbors Moran s I statistic on the errors of the OLS regression reveal significant spatial dependence LM tests reveal that the spatial process occurs in the error terms (1) (2) (3) (4) (5) 1 inhabitant 5.67 inhabitants 10% 25% 50% per sq. km. per sq. km. urban urban urban Moran s I 0.402 0.179 0.207 0.172 0.123 LM lag 126,015 55,539 75,567 55,901 31,176 LM error 134,991 83,057 110,694 76,783 39,390 Note: Moran s I statistics are significant at the p < 0.001 level.

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Modeling neighborhood effects To account for spatial dependence in the error term, we use a spatial error model y = βx + µ µ = λw µ + ɛ W is a weights matrix that captures the errors in neighboring locations Spatial range of the weights matrix is 5 orders of contiguity, implying an 11 x 11 neighborhood

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Results from the spatial error model Dependent variable is number of years since transition (1) (2) (3) (4) (5) 1 inhabitant 5.67 inhabitants 10% 25% 50% per sq. km. per sq. km. urban urban urban cultiv 0.315 0.231 0.300 0.254 0.176 coast -0.044-0.074-0.050-0.088-0.097 river 0.635 0.516 0.630 0.617 0.522 frost 0.226 0.262 0.210 0.211 0.237 elevation -0.168-0.155-0.139-0.094-0.074 λ 0.900 0.882 0.897 0.872 0.815 n = 62, 290. All coefficients are significant at the p < 0.01 level. Declining importance of cultivation index. Rising importance of distance to coast.

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Variability across continents Null Hypothesis Variable Europe Asia Africa NA SA Aus do not reject lnsuit 0.27 0.22 0.23 0.19 0.28 0.31 reject lndist - 0.02-0.01-0.21-0.08-0.26-0.18 do not reject river order 0.24 0.17 0.14 0.19 0.22 0.24 reject frost 0.08 0.03-0.04 0.89 0.35 0.32 reject lnelev - 0.23-0.15 0.04-0.32 0.15 0.05

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Marginal effects in different continents "marginal" effects at sample means in yrs with continent- specific coefficients EUR AFR ASIA NAM SAM AUS cult 0.86 0.22 0.24 0.27 0.28 0.17 per 0.01 dist - 1.61-0.81-0.06-0.83-1.94-1.02 per 100 km elev - 11.00 0.15-0.64-2.41 1.04 0.28 per 100 m streams 0.36 0.05 0.04 0.08 0.09 0.04 per 0.01 frost 1.08 0.96 1.04 2.43 1.42 1.38 1 vs 0 "marginal" effects at sample means in yrs with global coefficients EUR AFR ASIA NAM SAM AUS cult 0.71 0.21 0.24 0.31 0.22 0.12 per 0.01 dist - 7.09-0.36-0.42-0.96-0.70-0.52 per 100 km elev - 6.14-0.53-0.55-0.99-0.89-0.76 per 100 m streams 0.27 0.05 0.05 0.08 0.07 0.03 per 0.01 frost 1.27 1.27 1.27 1.27 1.27 1.27 1 vs 0

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Linking historic date of transition to year 2000 income 11 LUX log income, year 2000 QAT NOR ARE USA SGP AUS HKG CHE AUT BRN CANDNK KWT ISLIRL SWE 10 ISR FIN CYP PRI BHS TWN NZL SVN OMN MUS TTO KOR SAU CZE HUN ARG ESTCHL LBYGAB URY MYS BLR SVK HRV LVA LTU RUS SWZ CRI POL 9 PAN MEX ZAF BWA BGR TKM TUN VEN DOM LBN GNQKAZ THA BLZ DZA IRN TUR CUB COL CPV MKD ROUNAM JAM FJI SURPRY SLV UKR PNG DJI ECU EGY ALB LKA PER JOR GEO GTM GUY PHL IDN ARM MAR KGZ NIC UZB AZE WSM VUT ZWE 8 BIH BOL GIN CMR IRQ PAK MDACIV HND HTI VNM SLB SYR BGD AGO LSO TJK MRT NPL MNG SEN COM STP GHA PRK BEN LAO COG KEN 7 MOZ RWA YEM UGA NGA MLI SDN GMBBFA CAF BTN MWI TGO ZMB MDG GNB NER TZA TCD BDISLE SOM ETH DEU FRA NLD PRT GRC BRA ESP IND GBR BEL JPN CHN ITA 6 ERI KHM LBR AFG COD 0 200 400 600 800 1000 1200 number of years ago a country's first grid cell reaches 25% urbanization Country-level per capita income in year 2000 and country s first transition date (25% urbanization)

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Linking historic date of transition to year 2000 income We test this relationship in a regression setting while accounting for cross-country spatial dependence using the spatial lag model. ln(income i ) = β 0 + ρw ln(income i ) + β 1 trans i + ɛ i

Specifying a model OLS and Tobit results Accounting for dependence across space Differences across continents Transition and modern national income Results from the spatial lag model (1) (2) (3) (4) (5) 1 inhabitant 5.67 inhabitants 10% 25% 50% per sq. km. per sq. km. urban urban urban trans 0.0003 0.0009 0.0007 0.0015 0.0023 constant 1.80 1.88 1.83 1.95 1.94 ρ 0.777 0.76 0.76 0.740 0.746 n = 169. All coefficients but one are significant at the p < 0.01 level. trans in (1) is significant at p < 0.1 level. Marginal effects 1 inhabitant 5.67 inhabitants 10% 25% 50% per sq. km. per sq. km. urban urban urban trans 0.0015 0.0037 0.0031 0.0058 0.0090 Reaching 50% urbanization one year earlier raises modern incomes by nearly 1%.

We show that features of geography influence the date at which a location urbanizes and begins transition. Cultivation suitability s importance declines as urbanization rises. Distance to coast matters more as urbanization rises. We show that the date at which a country first urbanizes influences incomes today.

Implications Deeply impoverished locations owe something to their poor agricultural geography. Research and development in agriculture can help push populations towards more productive urban activities. Investment in transport infrastructure could mitigate the cost of moving goods to markets posed by distance and topography, and thereby grow urban economic activities.