SciELO - Scientific Electronic Library Online

 
 número70Evaluation of an Active Labour Market Programme in a Context of High UnemploymentNegative income Tax in Uruguay: An alternative redistribution policy? índice de autoresíndice de assuntospesquisa de artigos
Home Pagelista alfabética de periódicos  

Serviços Personalizados

Journal

Artigo

Indicadores

Links relacionados

  • Em processo de indexaçãoCitado por Google
  • Não possue artigos similaresSimilares em SciELO
  • Em processo de indexaçãoSimilares em Google

Compartilhar


Desarrollo y Sociedad

versão impressa ISSN 0120-3584

Desarro. soc.  no.70 Bogotá jul./dez. 2012

 

The Economic Effects of Geography: Colombia as a Case Study

 

Los efectos económicos de la geografía: Colombia como caso de estudio

 

Juan Mendoza *
Andrés Rosas **

* Universidad del Pacífico, Lima, Perú. Correo electrónico: mendoza_jj@up.edu.pe.

** Pontificia Universidad Javeriana, Bogotá, Colombia. Correo electrónico: Andres.Rosas@javeriana.edu.co.

Este artículo fue recibido el 24 de noviembre de 2011; modificado el 22 de agosto de 2012 y, finalmente, aceptado el 18 de octubre de 2012.


Abstract

This paper quantifies the economic impact of geographical features using Colombian data at the municipal level. We use the proportion of slave population in 1835 as an instrument of current institutions. We find that, controlling for institutional quality, geographical characteristics, such as the percentage of flat terrain, the proximity to the marketplace and the proximity to the main rivers are statistically-significant determinants of income per capita and have large economic effects.

Key words: Geography, climate, development, institutions.

JEL classification: O11, O54, P16, R11.


Resumen

Este artículo cuantifica el impacto económico de las características geográficos utilizando datos de Colombia a nivel municipal. Utilizamos el porcentaje de población esclava en 1835 como instrumento de las instituciones actuales. Encontramos que controlando por la calidad institucional, características geográficas, como el porcentaje de terreno plano, la proximidad a los mercados y la proximidad a ríos importantes, son determinantes estadísticamente significativos del ingreso per capita y tienen efectos económicos significativos.

Palabras clave: geografía, clima, desarrollo, instituciones.

Clasificación JEL: O11, O54, P16, R11.

We have received helpful comments from Herschel Grossman, Sean Campbell, and an anonymous referee. We thank Guillermo Sinisterra for excellent research assistance.


Introduction

At some level, it is evident that geography matters for economic performance. Except for a few oil-producing and island states, the tropics are poor and backward while the temperate regions of the globe are rich and prosperous. In effect, the terms north and south are embedded in economics as signifiers of a country's stage of development or underdevelopment. This paper quantifies the impact of geographical characteristics on income using data from Colombia at the municipal level. A by-product of the paper is a quantitative assessment of the effects of institutional quality on income per person.

The study of the economic effects of geography is not marred by considerations of reverse causality. The reason is that geographical features can be safely assumed to be exogenous with respect to income per person. But, until recently, the economic literature had neglected the role of geographical features in explaining the variability of income. As a matter of fact, there is no consensus on the relative importance of the mechanisms governing the influence of geography on economic development.

In our reading, the literature recognizes various direct channels by which geography affects development. The direct channels emphasize the effects of geography on productivity and offer several reasons to explain this influence. First, as one would naturally expect, geographical characteristics might account for the extent and productivity of primary activities such as agriculture, farming, mining, fishing, and forestry. Second, going back to Marshall, many economists have stressed how extreme climes and temperatures might diminish labor productivity because they affect nutrition, health, and work effort. Third, higher productivity might be the consequence of spillovers and positive externalities from clustering, agglomeration or proximity to the marketplace. Examples of proponents of the direct channels of geography include Mellinger, Sachs and Gallup (2000), McArthur and Sachs (2001), Sachs (2001), and Fujita, Krugman and Venables (1999). If the direct channels are correct then geography has contemporaneous effects on income per person.

The literature also recognizes that geography might affect income via the indirect channel of institutional quality. In this context, institutions are good if they limit the scope of predatory government behavior, serving as catalysts for the establishment of secure property rights. In a leading article, Engerman and Sokoloff (1997) tell us how the adoption of exploitative institutions by Europeans in the Americas (slavery and forced labor) was largely determined by geographical endowments. In the same spirit, Acemoglu, Johnson and Robinson (2001) theorize that weak and extractive institutions were more likely to be established in places dominated by unfavorable geographical characteristics. Hall and Jones (1999) claim that geography affected institutional quality because fifteenth-century Europeans had an incentive to settle in sparsely populated areas with climates similar to Europe. García-Jimeno (2005) finds that some colonial institutions in Colombia, such as slavery and forced labor, have had persistent and significant effects on economic development across the country. Proponents of the indirect channel underscore that geography matters for development because of its past effect on institutional quality under that assumption that there is persistence in the evolution of institutions.1

From an econometrical standpoint, there is substantial agreement on the statistical significance of the institutional channel of geography.2 In contrast, recent work disputes the empirical relevance of some of the direct channels. Acemoglu et al. (2001) find that health-related variables, which one would expect to affect labor productivity, have no significant impact on income per person. Mendoza and Rosas (2012) report that, controlling for institutional quality, infant mortality and malaria incidence do not have a significant negative effect on income per person in a sample of Colombian municipalities. Rodrik, Subramanian and Trebbi (2004) go even further and conclude that geographically-related variables are, at best, weak direct determinants of income per capita again controlling for institutions, using various cross-country samples and econometric specifications. Easterly and Levine (2003) do not even find evidence that geographical endowments matter for development other than through institutions.

This paper contributes to the literature by focusing on two questions: What geographical variables appear to have direct effects on the level of income per capita? and what is the magnitude and economic significance of these effects? To answer these questions we take into account the role of institutional quality as a determinant of income per capita. In contrast to the bulk of existing studies, our empirical strategy is to utilize regional data from a single country. Specifically, we use a dataset of close to nine hundred Colombian municipalities. In an interesting study, Sánchez and Núñez (2000) examine the geographical determinants of municipal income in Colombia. However, these authors do not consider the possible endogeneity of institutional variables.

We believe that Colombia is an excellent candidate to study the economic impact of geography because of its exceptional regional variability in geographical patterns. Hence, the use of data at the subnational level does not sacrifice sample variability in a significant manner. On the other hand, the utilization of subnational observations has two unambiguous advantages with respect to the approach of relying on cross-country regressions. To begin with, we have many more degrees of freedom than any cross-country study because of a much larger sample size. In addition, while we still control for differences in institutional quality, we abstract from sources of variation in institutions commonly found in cross-country studies. Examples of these sources are the degree of ethnolinguistic fractionalization, the origin of the colonizer, the type of judicial system or the prevalence of particular religions. In this sense, this paper tests the power of institutions to understand patterns of development even within a national unit.

To preview our findings, we identify some geographical features, notably the steepness of the terrain and the proximity to the marketplace as significant direct determinants of income per person. But, we also encounter that other geographical features, altitude, temperature and precipitation among them, fail to exert a significant direct influence on income per person. The estimation also highlights the robustness of institutional quality in explaining economic performance.

I. The Geography of Colombia

Colombia is located in the northern section of South America between the Pacific and Atlantic Oceans. The country covers an area of 1,140,050 square kilometers, about the combined areas of Texas and California. The country exhibits exceptional regional variability in geographical, including climatic, patterns. While it occupies less than one percent of the surface of the earth, the country has ten percent of all forms of animal and plant life and ranks first in the world in the number of birds and amphibians, second in the number of plants and reptiles and seventh in the number of mammals.3 As a matter of fact, the biodiversity of Colombia exceeds that of much larger countries like China, the Russian Federation and even the United States.4

The geographical diversity of Colombia stems from its location and geology. Although Colombia lies in the geographical tropics, its climatic patterns are far from being uniformly tropical. The country is divided by three chains of the Andes mountains: the Western, Central, and Eastern Cordillera. These chains, which run in a roughly north-south direction, give rise to a temperate zone of upland valleys and plateaus. The Eastern and Central cordilleras are separated by the Magdalena River, which reaches the Caribbean Sea in the port of Barranquilla. The Central and Western mountain ranges are separated by the Cauca river. The eastern part of Colombia, extending south and east of Bogotá, is comprised by hot, humid grasslands falling within the drainage basins of the Orinoco and the forested Amazon. The types of terrain of the country include flat coastlands, plains, highlands, forests, jungles, and Andean plateaus.

Even within regions of the country geographical characteristics are highly heterogeneous. Consider the Andean region as an example. This region includes peaks over eighteen thousand feet high and valleys as low as one thousand feet above sea level. Within the borders of the Andean region, temperature levels vary greatly from a medium of eighty degrees Fahrenheit in the tropical zones (below 3,500 feet) to a low of twenty degrees Fahrenheit in the inhabited parts of the cordilleras. Although the Andean region is distinctive for its two annual dry and rainy seasons, there is high variability in the amount of precipitation depending on exposure and elevation. In some valleys precipitation levels are negligible, while in others precipitation may reach as much as 140 inches per year.

II. Geography and Productivity: The Direct Channels

The first link emphasized by the direct channels is the effect of geography on the occurrence and productivity of primary forms of production such as agriculture, farming, mining, fishing, and forestry. In turn, this effect might obtain for a number of reasons. We have long understood the influence of the slope of the terrain, the availability of water and the type of soil, among other factors, on agricultural productivity. For instance, tropical soils are of poor quality because much of their mineral content is washed away by everpresent rains. Gallup (1998) estimates that land productivity in wet temperate zones exceeds land productivity in wet tropical zones by a factor of two.5 Some authors have also stressed that different modes of production, such as shifting cultivation, plantation agriculture or yeoman farming are strongly affected by the natural environment. For example, Butler (1980, page 12) attributes the wide prevalence of sheep grazing in western regions of the United States to both "limited rainfall and long distance to markets". Some other authors have advanced the idea that geographical characteristics might help to determine the growth rate of technological change as well as the diffusion of existing technologies. Lastly, the location and development of mining, fishing, and forestry is obviously fundamentally conditioned by geographical endowments.

The direct channels also underline the possibility that geography might impact labor productivity via its effect on work effort, health, and nutrition. Early writers, most prominently Aristotle and Montesquieu, believed that work effort was higher in cold rather than hot climates. More recently, Marshall (1890:1953, page 195) wrote:

A warm climate impairs vigour. It is not altogether hostile to high intellectual and artistic work; but it prevents people from being able to endure very hard exertion of any kind for a long time. More sustained hard work can be done in the cooler half of the temperature zone than anywhere else.

In the words of Butler (1980, page 135), "lower economic efficiency in many tropical countries results from debilitating diseases and poor nutrition...". In fact, Gallup and Sachs (2001) find that income per capita in non-malarial countries is about five times income per capita in malarial countries.

Finally, according to the direct channels, there might be a link between geography and productivity because of the existence of externalities and spillovers from clustering, agglomeration, as well as from proximity to the marketplace. The new economic geography underlines how these externalities and spillovers might be a function of transportation costs and the size of the market in the presence of imperfect competition and increasing returns. Several authors have studied the interplay between the spatial economy of a region and its geographical features. For example, Mellinger, Sachs and Gallup (2000, page 182) estimate that

A striking 67.6 per cent of the world's GDP is produced within 100 km of the sea, though that area comprises only 17.4 per cent of the world's landmass. Meanwhile 67.2 per cent of the world's GDP is produced in the temperate climates, though these account for only 39.2 per cent of the world's landmass.

These authors also find that GDP density is much higher in the temperate regions of the globe than in the tropics.

III. The Data

We have a dataset of Colombian municipalities which are the jurisdictional equivalent to counties in the United States. Appendix A describes the data sources in detail. Appendix B presents summary statistics of the variables used in the estimation.

Official Colombian statistics of income per person are available only at the departmental level. Each of the 32 Colombian departments is composed by several municipalities. Following Sánchez and Núñez (2000), we proxy the level of income in each municipality in 1999 with the sum of the municipal property taxes and municipal industry and commerce taxes between 1997 and 1999.6 Using population data for 1999 we computed the log of income per person in each municipality, INCOME. The correlation coefficient between our measure of income per capita, aggregated by departments, and the official figure is 0.96.

To further test the robustness of the variable INCOME we also use the Unsatisfied Basic Needs index (NBI) as a proxy of the level of development in each municipality. This index, computed by the Colombian National Administrative Department of Statistics (DANE), measures the percentage of the population whose basic needs of housing, health and education are below an established threshold. The NBI index is between 0 and 100. A higher value of the index implies that a higher percentage of the population do not have their basic needs met. We use the value of the index computed by the DANE for 1993. The correlation between INCOME and NBI is -0.53.

With respect to geographical characteristics, we first consider climatic variables:

• ALT: Altitude in meters above sea level.

• ALT2: Altitude squared.

• TEMP: Average yearly temperature in degrees Celsius.

• TEMP2: Average yearly temperature squared.

• RAIN: Average yearly precipitation in millimeters.

• ALTRAIN: The product of Altitude and Rain.

• TEMPRAIN: The product of Temperature and Rain.

We expect climatic variables to affect both agricultural and labor productivity because of their possible influence on crop types, size and quality, and on the health, nutrition and effort of the labor force. We consider altitude, temperature and precipitation squared to allow for non-linear effects of climate on performance as suggested by the empirical estimates of Masters and McMillan (2001). We also include the interaction terms ALTRAIN and TEMPRAIN because we believe that the marginal impact of more or less precipitation is likely to be conditioned by both altitude and temperature.

We also consider indicators of the relative steepness of the terrain:

• PLAIN: The percentage of flat terrain in the municipality, where flat is defined as an average gradient of less than ten percent.

• SLOPE: The average gradient of the terrain in the municipality.

As it has been well established by economic geographers, plains and grasslands are much more productive for agriculture than mountainous and steep terrains. Hence, we expect these variables to influence income via their effect on agricultural productivity. Further, the relative steepness of the terrain might matter for performance, according to the direct channels, because of its effect on transportation costs.

Next, we include the following variables of location:

• DMARKETS: Euclidian distance, in kilometers, from the municipality to the country's nearest main city (Cali, Medellín, and Bogotá) or seaport (Barranquilla and Buenaventura), to consider the possible effects of transportation costs and market size on productivity.

• DRIVERS: Average Euclidian distance to the twenty major rivers of the country in kilometers.

The location of a region is likely to affect transportation costs and the effective size of the market. The distance to major rivers might be an indicator of the hydrological potential of a municipality besides being correlated with the type and quality of the soil. The Magdalena and its main tributary, the Cauca, have been the most important means of fluvial transportation. The Magdalena river flows from south to north for 1,540 km ending its course, as mentioned above, in the Caribbean Sea near the port of Barranquilla. The Cauca river starts in the western section of the country and joins the Magdalena in the lowlands of northern Colombia. Our variable DRIVERS is computed by taking the average distance of the center of each municipality to the twenty rivers with higher flows including the Magdalena and Cauca. We enclose a map of the main rivers of Colombia after Appendix C.

Some of the geographical features that we have considered, such as altitude or rain, are first nature, in the sense of being unchanged by the existence of men. Some other, such as distance to markets or population density, are not first nature because they are the consequence of past human behavior. One might argue that some non-first-nature variables might reflect the past effect of first-nature variables. If correct, this observation would not affect the validity of our econometric results, but would rather guide their interpretation. At any rate, we make the plausible identifying assumption that all the variables considered are predetermined with respect to current levels of income.

IV. Regressions without Institutions

Do all the direct channels of geography matter? How much of the variance of income per capita can be attributed to these channels? Tables 1 and 2 present regressions of our measure of development against our set of geographical variables using Ordinary Least Squares. The first two columns of each table use our proxy of the logarithm of income per person in 1999. Columns (3) and (4) use the NBI index in 1993. We use fixed effects at the departmental level to control for sources of variation in development that are constant over time across Colombian departments.

We do not include altitude and temperature in the same regression because they are almost collinear: a correlation coefficient of -0.985. We only present the regressions that use altitude. All the results are invariant to the use of temperature instead of altitude. Because the effects of geography might be non-linear, Table 2 considers quadratic terms for altitude, precipitation, distance to markets, and distance to rivers. We present the regressions that use PLAIN as a measures of the steepness of the terrain. All our qualitative results are robust to the use of SLOPE instead of PLAIN and are available upon request. Appendix C presents the correlation matrix of the geographical variables considered in the analysis.

What can we learn from these regressions? The regressions tell us that geographical characteristics are significant determinants of income per person with levels of confidence ranging from ninety to ninety-nine percent. For example in the first regression of Table 1 all variables but precipitation are significant at the ninety percent level of confidence. Indeed, except for precipitation squared (RAIN2) in Table 2, each of the geographical features included in the estimation is statistically significant under some specification. The variables PLAIN, DRIVERS, DRIVERS2, DMARKETS and DMARKETS2 are significant across most specifications. The importance of geography is also expressed in the relatively high R-squares of the regressions, higher than twenty-percent even without fixed effects, in spite of the absence of additional controls for income per capita.

There is some evidence of income per capita being a concave function of altitude. In column (3) of Table 2, using the NBI index as a gauge of development, both ALT and ALT2 are significant at the ninety-nine percent level of confidence. Taking into account the interaction term ALTRAIN (when it is statistically significant), precipitation has a positive effect on income in the average municipality. The results also imply that flatter terrains are associated with higher levels of development. Furthermore, our regressions tell us that the predominant effect of lower distances to rivers and markets is to increase income, but that the effect of these geographical features is quadratic.

What is the order of magnitude of the effects of geography on income per capita? Simple calculations suggest sizable economic effects. Consider regression (1) of Table 1 as an illustration. This regression indicates that a municipality with a one-standard-deviation higher altitude than the average municipality, equivalent to 893 meters, would have fifteen percent higher income. The same regression suggests that if the percentage of flat terrain in a mu- nicipality, PLAIN, is one-standard deviation above the mean, then income per person would be nineteen percent higher than in the average municipality. To illustrate the importance of variables of location, consider now regression (1) of Table 2. This regression implies that a one-standard deviation increase in distance to markets, approximately 91 kilometers, would result in as much as sixty percent lower income per person. Also according to this regressions, if a municipality is twelve kilometers further away from a major river, its income per person would be roughly seventeen percent lower. In our sample, the average distance to major rivers is twenty-three kilometers with a standard deviation of twenty-four kilometers.

V. Institutional Variability in Colombia

The fundamental shortcoming of the previous estimates is that they do not consider the possibility of differences in institutional quality across Colombian regions. This neglect is not trivial because Colombia exhibits great regional variability in the efficiency of institutions. As a matter of fact, the history of Colombia suggests that geography has influenced institutional quality in harmony with the theories of Engerman and Sokoloff (1997), Acemoglu et al. (2001), Hall and Jones (1999), García-Jimeno (2005), Nunn (2008), Dell (2010) and others.

The institutional differences of Colombian regions are rooted in the historical evolution of the country. As in other colonial experiences in the Americas, Spanish settlers chose to establish permanent residence in regions endowed with favorable geographical characteristics. Colonizers placed high value on regions of temperate climates and flat terrains and avoided the extreme conditions of the tropics or the Andean slopes and the higher elevations. In the former regions, the administrative structure and political institutions resembled those of the Spanish metropolis. In the latter, there was sparse European occupation and the Spaniards establish exploitative institutions in order to control the means of production and the population. The chief exploitative institutions in Colombia, as in the rest of Spanish America, were slavery and two forms of forced labor called El Corregimiento and La Encomienda.7

But geography not only influenced the type of Spanish settlements during colonial times. Rugged and diverse geography has been an obstacle to interregional trade and migration, and thereby convergence in levels of income, even after the country's political independence. Rafael Reyes, President of Colombia in the second half of the nineteenth century, observed that Colombians only met regularly with nationals from other regions during the country's civil wars.

Isolated regional development has resulted in fragmented polities and institutions. The variability on the quality of institutions is reflected in large regional differences in the presence of the state and in the security of property rights. In some regions, such as Bogotá, the Colombian state provides much more public security and justice than in places such as the Chocó, a tropical region near the Pacific coastline. In places like the Chocó, the effective rulers are local agents, typically landowners or drug lords, who compete for both the allegiance of people and the power to tax them by providing public goods not offered by the Colombian state. Because in many areas of the country the rule of the state has been either nominal or severely constrained by regional idiosyncrasies, local conditions are crucial determinants of the effectiveness of the Colombian state's laws and institutions.8

Furthermore, there is some evidence that indicators of political violence and armed conflict, which one would expect to be associated with weaker institutions, are partially explained by geographical features as documented by Tellez (2012). Indeed, this author finds that the likelihood of guerrilla attacks depends on the proportion of the territory covered by forests and the slope of the terrain.

We have defined good institutions as those which limit the scope of predatory government behavior serving as catalysts for the establishment of secure property rights. But there are no readily available empirical counterparts to this definition. We propose then two measures of institutions which we believe to be correlated with the true, albeit unavailable, measure of institutional quality.

The first measure is LNCOURTSPC, the log of the number of judicial courts per person in 1998. Judicial courts are the cells of the administration of justice in Colombia. Each court has one judge and one secretary besides administrative staff. Their common organization makes judicial courts comparable across regions. We believe that this variable captures the presence of the state and is likely to be correlated with the security of property rights across regions. The administration of justice is one of the archetypical functions of government and, just as the provision of policing, a prime example of a non-excludable public good. However, Mendoza (1999) shows that a self-interested government might shirk from the provision of policing as evidenced by the experience of many Colombian regions. Assuming that the scope of self-interested government behavior is constrained by institutional quality, and abstracting from differences in crime rates, we expect a higher number of judicial courts per person to reflect both better governance and better institutions.

We also use LNCOMMPC, the log of the number of communal organizations per person as of 1998, as a second measure of institutional quality. Putnam (1993, page 91) reports that institutional performance is strongly correlated with various measures of civic engagement such us "...the number of amateur soccer clubs, choral societies, hiking clubs, bird-watching groups, literary circles, hunters' associations, Lions Clubs, and the like in each community and region of Italy". In Colombia, communal organizations employ voluntary labor in local infrastructure projects using private means and some modest support from the state. A World Bank (1995) study has found that government efficacy and accountability correlate with high degrees of community participation across various Colombian regions. These organizations also provide a mechanism of conflict resolution at the neighborhood level as an alternative to the formal judicial system. Buscaglia and Ratliff (2001, page 6) tell us that, according to a survey conducted in four Colombian departaments, fifty-seven percent of household heads had used or knew someone who had used the informal mechanisms of conflict resolution provided by neighborhood councils. We believe that the number of communal organizations reflects the security of property rights and the degree of government accountability because higher community involvement constrains self-interested government behavior.9

The correlation between LNCOURTSPC and LNCOMMPC is 0.61. Further, the dispersion of institutional quality across Colombian regions is not negligible. The coefficients of variation of LNCOURTSPC and LNCOMMPC are 69% and 88%. For comparison, the main measure of institutions used by Rodrik et al. (2004) has a coefficient of variation of 105% across 140 countries.

VI. Regressions with Institutions

To consider the effect of geography on income while accounting for institutional variability we estimate the following equation:

where GEOGRAPHY collects all the geographical features under consideration, and ε is an stochastic error term. Tables 3 and 4 present the estimates of equation (1) using our two measures of institutional quality. All the regressions consider fixed effects at the departmental level. The estimates of Tables 3 and 4 imply that the quality of institutions is an important determinant of income per person in particular when we consider the variable INCOME. Except for regression (4) in Table 3, institutional quality is significant at least at the ninety percent level of confidence. Tables 3 and 4 also tells us that geographical variables continue to explain the level of development. In particular, the variables PLAIN, DRIVERS and DMARKETS are statistically significant in a systematic manner.

Granted, institutions differ across regions. But, institutional quality is also likely to be influenced by higher levels of income per person. In other words, higher income might lead to better institutional quality instead of or in addition to the effect of institutions on income. In order to consider the possible endogeneity of institutional quality we use PORSLAVE, the proportion of slave population in 1835, as an instrument of current institutions. As suggested by the work of Engerman and Sokoloff (1997) and García-Jimeno (2005) and by the historical accounts of Sharp (1976) and Wade (1993) among others, we expect PORSLAVE to capture the extent of exploitative institutions in 1835.10 We believe that some of the present variation in institutional quality is exogenously explained by the prevalence of slavery in 1835 at least, in part, because of the isolated historical development of the country's regions. The correlation coefficient between PORSLAVE and LNCOURTSPC, the log of the number of judicial courts per capita, is -0.25. The correlation between PORSLAVE and LNCOMMPC, the log of the number of communal organizations per capita, is -0.30.

The proportion of slave population in 1835 is clearly exogenous with respect to current income per capita. We also make the assumption that PORSLAVE is uncorrelated with the error term in equation (1). The statistics of the Hansen-Sargan J test of overidentifying restrictions in Mendoza and Rosas (2012) support the use of PORSLAVE as an instrument of institutional quality across Colombian regions. We are aware that a possible criticism to our use of PORSLAVE is that the prevalence of slavery across Colombian regions was, at least partially, influenced by geographical features. However, we believe that this possible shortcoming of our analysis is not crucial because we are already controlling for an array of geographical characteristics as possible determinants of income per person.

Tables 5 and 6 present the results of the estimation. Both tables use PORSLAVE as an instrument of LNCOURTSPC, the logarithm of the number of judicial courts per capita, and LNCOMMPC, the logarithm of the number of communal organizations per capita, our alternative measure of institutions. The columns of the first panel of these tables mimic the specifications considered in Tables 1 and 2 except for the addition of our measures of institutional quality. That is, Table 5 reports the estimates in which geographical features appear linearly while Table 6 contains regressions in which we consider quadratic terms. In the second panel of Tables 3 and 4 we present the first-stage regressions of the excluded instrument, PORSLAVE. Notice how the F-statistics of these regressions provide support for our utilization of the proportion of slave population in 1835 as an instrument of current institutional quality.11

What can we learn from the estimation? The results reveal that institutions are significant determinants of the variance of income per person. Consider, for example, columns (1) and (2) of Tables 5 and 6. The estimated elasticity of income per person with respect to the number of courts per person ranges from 1.16 to 1.27. The estimated elasticity of income with respect to the number of communal organizations is between 1.36 and 1.44. Even the more modest of these estimates imply that if a municipality were to improve its institutional quality by one standard deviation, it would nearly double its income per person.

How do the results differ from the previous regressions without institutions or without accounting the possible endogeneity of institutions? The estimation implies that, controlling for institutional quality, geography matters for development. However, not all geographical variables seem to have a systematic effect on income per person. In particular, the estimation tells us that climatic characteristics, such as altitude, precipitation and their interaction term, are seldom significant.

Nevertheless, the estimation also implies that some geographical variables continue to be statistically-significant determinants of income per person after accounting for institutional quality. In particular, we find three groups of variables that are consistently significant and have the expected signs across most specifications: the relative steepness of the terrain, the distance to markets, and the distance to rivers. Next, we quantify the magnitude of the economic effects of these geographical variables.

According to our estimation, the percentage of flat terrain, PLAIN, has large effects on income per person.12 Using our preferred measure of development, INCOME, the point estimates of their coefficients are consistently higher with measures of institutions than without them in the regressions. The effect on income per capita of a one-standard-deviation increase in the percentage of flat terrain is between 38% and 54% (Regressions 1 and 2, Tables 5 and 6). Distance to markets is significant at the ninety-five percent level of confidence. For example, the estimates of regression 1 in Table 6 implies that a municipality one-standard- deviation further away from major markets would be fifty-four percent poorer. Finally, the average distance to major rivers is significant in most regressions. Using the estimated coefficients of regression (1) in Table 6, a one-standard-deviation increase in distance to rivers would result in a decrease of 14% in income per person.

VII. Relation to the Literature and Limitations

In line with the economic literature we find that institutions are a robust and powerful determinant of income per person. The magnitude of the effect of institutional quality on income is noteworthy because it takes place within a country. Further, our results imply that the statistical significance of many geographical features is a mirage in regressions that do not consider institutional variability. Our estimation is also consistent with the theories of Engerman and Sokoloff (1997), Acemoglu et al. (2001), and Hall and Jones (1999) that view institutional quality as a mechanism by which geography affects development.

The importance of institutional quality has led some authors to claim that institutions dwarf every other factor as a determinant of economic performance. Rodrik et al. (2004) write "we find that the quality of institutions trumps everything else. Once institutions are controlled for... geography has at best weak direct effects". Our findings are in sharp contrast with this maximalist view of the relevance of institutions. Our results support the theories and empirical studies suggesting direct effects of geographical features on income through their influence on productivity. Examples are Mellinger et al. (2000), Sachs (2001), and Fujita et al. (1999).

An important extension of the paper, using other samples and richer data, would be to unravel the particulars of the relationship between geography and productivity. Is it the case that geography affects the productivity of each factor uniformly? Does geography encourage the accumulation of some factors of production in lieu of others? What is the influence of geography on the extent of primary activities? What is the effect of geography on productivity via lower transportation costs? Why are climatic variables not systematically relevant when one would expect agricultural productivity to be naturally affected by them? To what extent are the findings dependent on idiosyncratic elements of the Colombian economy that we have failed to identify? Answering these questions would also help us to understand the general applicability of our findings.

Finally, we would like to remain silent in terms of recommendations for active policy to deal with the economic effects of geography. A necessary condition to formulate such policy is a precise identification of possible externalities or market distortions caused by geographical features. We would also like to remain cautious with respect to the desirability of policies aimed at improving institutional quality. We believe that our two measures of institutions are but proxies of the true, empirically not available, measure of institutional quality. Further, it might be reasonable to presume that institutions evolve slowly. Thus, an increase in the number of courts per person or in the number of communal organizations, perhaps engineered by active policy-making, might not necessarily involve much change in income per person.

VIII. Summary

This paper quantifies the effects of geography on income per person. We use a dataset of close to nine hundred Colombian municipalities. We control for institutional quality by using the proportion of slave population in 1835 as an instrument of current institutions.

The estimation reveals that institutional quality is a powerful determinant of economic development. Further, the estimation also tells us that, controlling for institutional quality, some geographical variables, such as altitude and precipitation, are not significant determinants of income per person in a consistent manner. Nonetheless, we also find that some geographical features have large economic consequences even after accounting for the variability of institutions. In particular, we find that the proportion of flat terrain, the average distance to the marketplace, and the average distance to major rivers have a significant direct impact on the level of economic activity.


FOOT NOTES

1 In empirical studies of American states, Berkowitz and Clay (2003, 2011) present evidence that colonial institutions are significant determinants of current institutional quality. Banerjee and Iyer (2005) stress the legacy of land tenure in Colonial India as determinants of current differences in development. Narotomi, Soares and Assuncao (2007) study the determinants, some of them geographic, of current Brazilian institutions using municipal data. According to Bruhn and Gallego (2008) the variation in levels of development in the Americas is rooted in differences in the institutions established by the colonizers. Dell (2010) presents evidence of the persistent consequences of a colonial form of forced labor, the Mining Mita, on the quality of Peruvian institutions.

2 For example, Acemoglu et al. (2001) use mortality rates faced by European settlers to instrument for current institutions and find statistical support for their hypothesis in a sample of sixty former colonies. Hall and Jones (1999) assume that the quality of a country's social infrastructure, including the security of property rights, the checks and balances in government and the efficacy of the judicial system, is increasing on the degree of influence from Western Europe. These authors report strong effects of a measure of geography, distance from the equator, on social infrastructure. Nunn (2008) concludes that the slave trade, which is correlated with predatory and corrupt institutions, is an important determinant of the level of development in Africa.

3 Sources: Instituto Von Humboldt (1999) and The World Resources Institute (1998, pages 322-325).

4 The following figures on the total number of known species illustrate this point:

5 See the work of Gallup, Sachs and Mellinger (1998) and Sachs (2001) for more evidence on the link between geography and agricultural productivity.

6 We had 874 observations on municipal taxes for 1999. In order to reduce the number of missing observations for 1999 we imputed 134 values using municipal taxes for 1998 and 1997. The imputed values are the predicted values of OLS regressions of municipal taxes in 1999 against municipal taxes in 1998 and 1997. Our results are robust to the exclusion of the imputed observations.

7 Romoli (1941), Earle (2000), Avellaneda (1995) and Henao and Arrubla (1938) provide interesting accounts of the importance of climate and topography in the European colonization of Colombia.

8 Huber and Safford(1995), Kline (1999) and Oquist (1980) focus on the influence of geography and climate in the formation of the Colombian state. Bushnell (1993), Richani (2002), and Safford and Palacios (2002) document the spatial variability in the effectiveness of the state and in the quality of institutions in Colombia.

9 See Buscaglia and Ratliff (2001) and Safford and Palacios (2002) for more on the number and charac- teristics of communal organizations in Colombia.

10 Sharp (1976) and Wade (1993) describe the poor social institutions of Colombian regions in which slavery was prevalent during colonial times. These institutions were characterized by insecure property rights as well as by lack of trust, reciprocity, and solidarity among the population.

11 In discussing single-equation instrumental variables regressions, Staiger and Stock (1997) recommend a threshold value of 10 for the first-stage F-statistic. Notice that all the F-statistics in Tables 5 and 6 are higher than this threshold value.

12 All of our qualitative results are robust to the use of SLOPE, the average gradient of the terrain, instead of PLAIN.


References

1. ACEMOGLU, D., JOHNSON, S. y ROBINSON, J. (2001). "The colonial origins of comparative development: An empirical investigation", American Economic Review, 91(5):1369-1401.         [ Links ]

2. AVELLANEDA, N. y IGNACIO, J. (1995). The conquerors of the New Kingdom of Granada. Albuquerque, University of New Mexico Press.         [ Links ]

3. BANERJEE, A. y IYER I. (2005). "The legacy of colonial land tenure systems in India", American Economic Review, 95(4):1190-1213.         [ Links ]

4. BERKOWITZ, D. y CLAY K. (2003). Initial conditions, institutional dynamics and economic performance: Evidence from the American States (Working Paper 615). William Davidson Institute.         [ Links ]

5. BERKOWITZ, D. y CLAY, K. (2011). The evolution of a nation: How geography and law shaped the american states. Princeton, Princeton University Press.         [ Links ]

6. BRUHN, M. y GALLEGO, F. (2008). Good, bad, and ugly colonial activities: Studying development across the Americas (Policy Research Working Paper, June). World Bank.         [ Links ]

7. BUSCAGLIA, E. y RATLIFF, W. (2001). War and lack of governance in Colombia: Narcos, guerrillas, and U.S. Policy (Essays in Public Policy 107). Hoover Institution Press.         [ Links ]

8. BUSHNELL, D. (1993). The making of modern Colombia: A nation in spite of itself. Berkeley, University of California Press.         [ Links ]

9. BUTLER, J. (1980). Economic Geography. Spatial and environmental aspects of economicactivity. Nueva York, John Wiley & Sons.         [ Links ]

10. DELL, M. (2010). "The persistent effects of Peru's mining mita". Econometrica, 78(6):1863-1903.         [ Links ]

11. EARLE, R. (2000). "A grave for Europeans"?, disease death, and the Spanish-American revolutions", en C. I. Archer (ed.), The wars of independence in Spanish America (pp. 283-298). Wilmington, SR Books.         [ Links ]

12. EASTERLY, W. y ROSS L. (2003). "Tropics, germs and crops: How endowments influence economic development". Journal of Monetary Economics, 50(1):3-39.         [ Links ]

13. ENGERMAN, S. y SOKOLOFF, K. L. (1997). "Factor endowments, institutions, and differential paths of growth among new world economies", en S. Harber (ed.), How Latin America fell behind (pp. 260-304). Stanford, California, Stanford University Press.         [ Links ]

14. FUJITA, M., KRUGMAN, P. y VENABLES, A. (1999). The spatial economy. Cambridge, MIT Press.         [ Links ]

15. GALLUP, J. L. (1998). Agricultural productivity and geography. Manuscript. Harvard Institute of International Development.         [ Links ]

16. GALLUP, J. L., SACHS, J. D. y MELLINGER, A. (1998). Geography and economic development (Working Paper 6849). National Bureau of Economic Research.         [ Links ]

17. GALLUP, J. L. y SACHS, J. D. (2001). "The economic burden of malaria", American journal of tropical medicine and hygiene, 64(1-2):85-96.         [ Links ]

18. GARCÍA-JIMENO, C. (2005). Colonial institutions and long-run economic performance in Colombia: Is there evidence of persistence? (Documento Cede 2005-59). Universidad de los Andes, Facultad de Economía.         [ Links ]

19. HALL, R. y JONES. C. I. (1999). "Why do some countries produce so much more output per worker than others?", Quarterly Journal of Economics, 114(1):83-116.         [ Links ]

20. HENAO, J. M. y ARRUBLA, G. (1938). History of Colombia, en J. Fred Rippy (translated and edited). Chapel Hill, The University of North Carolina Press.         [ Links ]

21. HUBER, E. y SAFFORD, F. (editors) (1995). Agrarian structure & political power: Landlord & peasant in the making of Latin America. Pittsburgh, University of Pittsburgh Press.         [ Links ]

22. INSTITUTO ALEXANDER VON HUMBOLDT (1999). Boletín 11 del Instituto de Investigación de Recursos Biológicos. Enero, Bogotá, Instituto Alexander Von Humboldt.         [ Links ]

23. KLINE, H. (1999). State building and conflict resolution in Colombia, 1986-1994. Tuscaloosa, University of Alabama Press.         [ Links ]

24. MASTERS, W. A. y MCMILLAN, M. (2001). "Climate and scale in economic growth", Journal of Economic Growth, 6(3):167-186.         [ Links ]

25. MARSHALL, A. (1953). Principles of economics (1st ed. 1890). Nueva York, The MacMillan Company, Eight Edition.         [ Links ]

26. MELLINGER, A., SACHS, J. y GALLUP, J. (2000). "Climate, coastal proximity and de- velopment", en C. gordon, M. Feldman y M. gertler (eds.), The Oxford Handbook of Economic Geography (pp.169-194). Oxford, Oxford University Press.         [ Links ]

27. MCARTHUR, J. W. y SACHS, J. D. (2000). "Institutions and geography: Comment on Acemoglu, Johnson and Robinson" (Working Paper 8114, 2001). National Bureau of Economic Research.         [ Links ]

28. MENDOZA, J. (1999). "Appropriative conflict in a positive theory of government", unpublished, Brown University, doctoral dissertation.         [ Links ]

29. MENDOZA, J. y ROSAS, A. (2012). "Geography, health, and development: Evidence from Colombia", Manuscript, Universidad del Pacífico and Pontificia Universidad Javeriana.         [ Links ]

30. NARITOMI, J., SOARES, R. R. y ASSUNCAO, J. J. (2007). "Rent seeking and the un- veiling of 'de facto' Institutions: Development and colonial heritage within Brazil" (Working Paper 13545). Massachusetts, National Bureau of Economic Research Cambridge.         [ Links ]

31. NUNN, N. (2008). "The long-term effects of africas slave trades". Quarterly Journal of Economics, 123(1):139-176.         [ Links ]

32. OQUIST, P. (1980). Violence, conflict, and politics in Colombia. Nueva York, Academic Press.         [ Links ]

33. PUTNAM, R. (1992). Making democracy work: Civic traditions in modern Italy. Princeton, NJ: Princeton University Press.         [ Links ]

34. RICHANI, N. (2002). The political economy of war and peace in Colombia. Albany, State University of Nueva York Press.         [ Links ]

35. RODRIK, D., SUBRAMANIAN, A. y TREBBI, F. (2004). "Institutions rule: the primacy of institutions over geography and integration in economic development", Journal of Economic Growth, 9(2):131-165.         [ Links ]

36. ROMOLI, K. (1941). Colombia, gateway to South America. Garden City, N.Y, Doubleday, Doran & Company, Inc.         [ Links ]

37. SACHS, J. D. (2001). Tropical underdevelopment (Working Paper 8119). National Bureau of Economic Research.         [ Links ]

38. SAFFORD, F. y PALACIOS, M. (2002). Colombia: fragmented land, divided society. Nueva York, Oxford University Press.         [ Links ]

39. SÁNCHEZ, F. y NúÑEZ, J. (2000). Geography and economic development: A municipal approach for Colombia (Documento Cede). Universidad de los Andes, Facultad de Economía.         [ Links ]

40. SHARP, W. F. (1976). Slavery on the spanish frontier: The Colombian Chocó, 1610-1810. Norman, University of Oklahoma Press.         [ Links ]

41. STAIGER, D. y STOCK, J. H. (1997). "Instrumental variables regression with weak instruments", Econometrica, 65(3):557-586.         [ Links ]

42. TELLEZ, S. (2012). Geografía, tácticas guerrilleras y conflicto en Colombia (Manuscript, July). Pontificia Universidad Javeriana.         [ Links ]

43. THE WORLD RESOURCES INSTITUTE (1998). World resources 1998-1999. Nueva York, Oxford University Press.         [ Links ]

44. WADE, P. (1993). Blackness and race mixture: the dynamics of racial identity in Colombia. Baltimore, Johns Hopkins University Press.         [ Links ]

45. WEST, R. (1957). The pacific lowlands of Colombia: a negroid area of the American tropics. Baton Rouge, Louisiana State University Press.         [ Links ]


Appendix

Appendix A

All the sources refer to Colombian publications and institutions. To estimate per-capita magnitudes we used population values in 1999 [Source: Departamento Administrativo Nacional de Estadística (DANE), División de Estadísticas Vitales].

• Income: Approximated with the log of per-capita property taxes and industry and commerce tax revenues. Source: Dirección de Desarrollo Territorial. Departamento Nacional de Planeación. 1997, 1998, 1999. Original values are in thousands of pesos.

• NBI: Unsatisfied Basic Needs index. Source: Departamento Administrativo Nacional de Estadística, Bogotá, Colombia. 1993.

• Lncourtspc: Log of the number of judicial courts per capita. Source: Municipios y Regiones de Colombia. Fundación Social. 1998.

• Lncommpc: Log of the number of communal organizations per capita. Source: Municipios y Regiones de Colombia. Fundación Social. 1998.

• Porslave: Slaves as a fraction of the total population in 1835 (at the departmental level). Source: Fernando Gómez, "Los censos en Colombia", Tables 6, 7, 8, in Miguel Urrutia and Mario Arrubla, eds., Compendio de estadísticas históricas, Bogotá, 1970. We matched the current jurisdictional demarcation with the jurisdictional demarcation of 1835 found in the original source.

• Alt: Meters above the sea level. Source: Instituto Geográfico Agustín Codazzi. 1998.

• Rain: Mean annual rain precipitation in mm. Source: Instituto Geográfico Agustín Codazzi. 1998.

• Temp: Mean annual temperature in Celsius degrees. Source: Instituto Geográfico Agustín Codazzi. 1998.

• Plain: The percentage of flat terrain in the municipality, where flat is defined as an average slope of less than ten percent. Source: Departamento de Ecología y Territorio de la Facultad de Estudios Ambientales de la Pontificia Universidad Javeriana. 2003.

• Slope: The average gradient of the terrain in the municipality. Source: Departamento de Ecología y Territorio de la Facultad de Estudios Ambientales de la Pontificia Universidad Javeriana. 2003.

• Drivers: Average Euclidian distance to the twenty major rivers of the country in kilometers. Source: Departamento de Ecología y Territorio de la Facultad de Estudios Ambientales de la Pontificia Universidad Javeriana. 2003.

• Dmarkets: Euclidian distance from the municipality to the nearest main city (Cali, Medellín, and Bogotá) or seaport (Barranquilla and Buenaventura) of the country. Source: Departamento de Ecología y Territorio de la Facultad de Estudios Ambientales de la Pontificia Universidad Javeriana. 2003.

Appendix B. Descriptive Statistics

Appendix C