Institutional Fragmentation and Metropolitan Coordination in Latin American Cities: What Consequences for Productivity and Growth?

This paper provides empirical evidence on the impact of institutional fragmentation and metropolitan coordination on urban productivity in Latin American cities. The use of night-time lights satellite imagery and high-resolution population data allow the use of a broader definition of metropolitan area. Thus, metropolitan area consists of the urban extent that results from the union between the formally defined metropolitan area and the contiguous patches of urbanized areas with more that 500,000 inhabitants. The initial results suggest that the presence of multiple local governments within metropolitan areas generates opposite effects on urban productivity. On the one hand, smaller governments tend to be more responsive and efficient, which increases productivity. But, on the other hand, multiple local governments face coordination costs that reduce productivity.


Policy Research Working Paper 8696
This paper provides empirical evidence on the impact of institutional fragmentation and metropolitan coordination on urban productivity in Latin American cities. The use of night-time lights satellite imagery and high-resolution population data allow the use of a broader definition of metropolitan area. Thus, metropolitan area consists of the urban extent that results from the union between the formally defined metropolitan area and the contiguous patches of urbanized areas with more that 500,000 inhabitants. The initial results suggest that the presence of multiple local governments within metropolitan areas generates opposite effects on urban productivity. On the one hand, smaller governments tend to be more responsive and efficient, which increases productivity. But, on the other hand, multiple local governments face coordination costs that reduce productivity.
This paper is a product of the Social, Urban, Rural and Resilience Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at nlozano@worldbank.org and prestrepocadavid@worldbank.org.

Introduction
for the coexistence of multiple small and coordinated local governments; the centrist model, which argues that a single governance body takes advantage of reduced transactions costs and economies of scale and scope in providing public goods and services; and the regionalist view which recognizes the benefits of local governments while highlighting the importance of coordinated governance. This paper will examine the interaction between local governance and economic performance, looking specifically at the effects of the fragmentation of functional urban areas across administrative entities. It will empirically test whether the economic performance of LAC is affected by the degree of institutional fragmentation that they exhibit. The paper will also seek to explore which, if any, of the different forms of metropolitan coordination observed in the region leads to a reduction of the penalties/advantages imposed by institutional fragmentation on economic performance. Finally, the paper explores whether results vary across city sizes (i.e. is there a specific threshold after which fragmentation penalties appear?) and whether they depend on the level of population concentration in a city's center or the overall spatial population distribution across administrative units (i.e., does it matter if most of the metropolitan population is concentrated in the city center?).
Understanding these links is also important from a policy perspective. Improving the understanding of the strength and direction of these links can suggest whether national governments should support or incentivize metropolitan coordination mechanisms or entities, or whether they should support the consolidation of local governments. Further, there is also the question of whether national governments should support the fragmentation of local governments, or on the contrary, whether there are specific sectors (i.e. transport, environment, local economic growth) for which it makes sense to align fragmented local governments.
The rest of the paper is organized as follows. Section 2 provides a literature review. Section 3 presents the empirical models. Section 4 describes the source data. Section 5 presents the empirical results.
Finally, section 6 presents our conclusions and ideas of future work.

Literature review
As mentioned by Nelson and Foster (1999), there exist three lines of thought when looking at the links between the governance structures of cities and their economic performance: polycentrist, centrist and regionalist. The polycentrist view is in line with Fisher´s argument in favor of dividing a region into sub-regions to facilitate the planning and distribution of resources (Fischer, 1980). At the city level, the polycentrist view argues that institutional fragmentation of cities is equivalent to creating additional layers of decentralization, which can, in fact, enhance economic growth. This is thought to be achieved through two mechanisms. First, decentralized authorities are better informed of local needs and, therefore, can be more efficient in the provision of public goods (Ostrom, 2010).
Second, increased competition between individual local governments constrains their ability to extract monopoly rents, thereby enhancing economic efficiency and, hence, economic growth (Stansel, 2005).
The second line of thought, the centrist, argues that the presence of multiple local governments within metropolitan areas may generate coordination failures that reduce efficiency in the provision of transport infrastructure and land use planning, and therefore affect economic performance (see Ahrend et al., 2014). Fragmentation may also reduce the metropolitan area's ease of doing business because of the additional bureaucracy (Kim et al., 2014). According to Cheshire and Gordon (1996) and Feiock (2009), the presence of administrative boundaries within the functional region generates higher transaction costs and barriers to the diffusion of growth-promoting policies. As additional evidence in favor of the centrist view, Foster (1993) found a negative association between the proportions of population unincorporated to the metropolitan area and income growth.
Finally, the regionalist view can be seen as a middle way between the polycentrist and the centrist views. The regionalist view recognizes the benefits of local governments while highlighting the importance of metropolitan coordination defined as the efforts of governmental institutions to manage and solve problems in common between municipalities. According to Grassmueck and Shields (2010), more important than the existence of multiple local governments is the way in which they interact and perceive each other. Ahrend et al. (2014) found that the presence of a governance body that coordinates municipalities halved the penalty associated with fragmentation. Foster (1993) and Nelson and Foster (1999) also found empirical support for the regionalist view as a positive association between income growth and the presence of overarching decision-making mechanisms such as multi-jurisdictional, multipurpose regional governments. Also, the existence of singlepurpose districts associated with large-scale infrastructure provision (e.g. water and wastewater systems) fosters income growth.
Regarding the measure of institutional fragmentation and metropolitan coordination, Table 1 and   Table 2 present a summary of the variables commonly used in empirical studies. In Table 1 we classify the variables using the five categories proposed by Hendrick and Shi (2015). Based on the conceptualization of fragmentation as those urban extents that spread out over several, and independent, administrative units, the most common variables are those that measure the number of local governments included in the urban extent (sometimes standardized by population or land area).
Other measures focus on the degree of concentration, the dominance relationship between the central city and the periphery. As considered by the centralist view, the potential drawbacks of having institutional fragmentation can be mitigated through proper channels of metropolitan coordination.
This coordination can be reached via institutions (governance), coordinated planning and infrastructure (land use planning and mobility), the presence of overlapping governments with special of general purpose (provision of public utilities), or can be the result of tight linkages between the administrative units, which intensifies human interactions, generates spatial dependence, and facilitates coordination (functional region). Table 2 classifies the potential variables according to those proposed channels of coordination.
The scarcity of empirical studies, in addition to the differences found when taking different approaches and using different economic performance indicators, suggests that there is a need to further test the empirical links between the institutional fragmentation of cities and their productivity/economic growth. This paper makes multiple contributions to the literature. First, and foremost, it will allow for testing of whether existing findings for the US and other selected OECD countries also carry-over to countries in LAC. Second, the paper will allow for the testing of the robustness of results to different institutional fragmentation and institutional coordination measures.
Third, it will allow for an assessment of whether existing metropolitan coordination measures are effective in helping to produce better economic outcomes. Finally, contrary to existing literature, most of the analysis will be supported by spatial data that are readily available at the global scale and for developing countries. This will allow the paper to contribute to the development of a methodology that can be easily replicated for other regions of the world.

Representation
Fragmentation Index (metro level) Authors I. Size of region Total number of local governments Hendrick and Shi (2015); Hill (1974) Differences in population and area of municipalities Barlow (1991) II. Political fragmentation Total local governments per capita Hendrick and Shi (2015); Hill (1974) Morgan and Mareschal (1999) Number of suburban units with more than 10,000 persons, per 100,000 persons in the MSA. Bollens (1986).
Percent of Metro residents in suburbs with more than 10,000 people Bollens (1986).

III. Spatial fragmentation
Total local governments per square mile Hendrick and Shi (2015) IV. Range of local governments HH Index of percent of different types of local government Hendrick and Shi (2015) V. Suburban domination (or central city domination) Percent of population not in central city Hendrick and Shi (2015) Ratio of population in the city core to that in the periphery Ahrend et al. (2014) Central-city population share Morgan and Mareschal (1999) Percent of metropolitan population held within the borders of a central city Savitch et al. (1993) Central-city area growth Morgan and Mareschal (1999) Central-city elasticity Rusk (1993); Blair et al. (1996)

I. Governance
Governance Body Ahrend et al. (2014) Age of the metropolitan area Nelson and Foster (1999) Number of municipalities with the same political party Pradenas (2006) II. Land use plan and mobility Percent of municipalities covered by integrated transport systems between municipalities and central city Kim et al. (2014) III. Coordination for special purpose Percent of special purpose to general purpose governments Hendrick and Shi (2015) Number of general purpose units Goodman (1980) Number of special purpose units Goodman (1980) IV. Percent of student population that go to the central city Pradenas (2006) Percent resident-job in central city Pradenas (2006) Number of telephone calls per month from the municipality to the central city must be four times greater. Pradenas (2006)

The Model
To estimate the relationship between institutional fragmentation and economic performance of cities, measured through city productivity, we follow the two-step empirical approach devised by . The authors warn about the importance of accounting for individual sorting of highly skilled individuals into cities when estimating productivity differentials across urban areas (Combes et al., 2011). This is necessary in order to account for the tendency of more talented individuals to colocate in cities that may lead to confounding agglomeration benefits with productivity increases from a more skilled workforce. Thus, in the first step, we use data from the Defense Meteorological Satellite Programs -Operational Linescan System (DSMP-OLS) nighttime lights (NTL) imagery to identify urban areas as well as survey-based micro-data for the period 2000-2014 to estimate productivity differentials across urban areas, net of individual and employment characteristics observables. The estimation on this first stage is then: where is the real wage for individual i at time t; are municipalities fixed effects; is a vector of demographic characteristics, indexed by , that include indicators of education; is a vector of job characteristic, indexed by , that includes industry code and indicators of formality and job benefits; and is an error term. The coefficient captures the productivity differential across cities, after controlling by individual and employment characteristics.
In the second stage, we use the estimated productivity differentials, , as the dependent variable in the following expression: , where is a vector of variables for institutional fragmentation; is a vector of variables for metropolitan coordination; is a vector of control variables, included the intercept; and is the error term.

Study region
In this work, we analyze Latin American and Caribbean metropolitan areas with more than 500,000 people in 2010 ( Figure 1).

Metropolitan areas delineation from DMSP-OLS images
We use data from the Defense Meteorological Satellite Programs -Operational Linescan System (DSMP-OLS) nighttime lights (NTL) imagery to identify urban areas and metropolitan conurbations.
The NTL data are based on nighttime imagery recorded by the Defense Meteorological Satellite Programs -Operational Linescan System (DMSP-OLS), and reports the recorded intensity of Earth's surface lights. Nighttime lights products have high correlation to human activities (Hsu et al., 2015), and have been previously used for regional and global analysis of population modeling (Anderson et al., 2010;Lo, 2001), economic performance (Cao et al., 2016;Forbes, 2013), and urbanization (Cheng et al., 2016;Pandey et al., 2013;Sutton et al., 2006;Zhang and Seto, 2011;N. Zhou et al., 2015;Zhou et al., 2015).
There are two different nighttime light products from DMSP-OLS that can be used to delineate urban areas: the stable or ordinary product (NTL), and the radiance-calibrated (NTL RC) product. We decided to use the latter, since it is aimed to correct the saturation issue in bright areas such as city centers, where the NTL might be brighter, but the recorded digital number (DN) values are truncated at 63; and the RC product gives better correlations with socioeconomic variables than the stable products (Hsu et al., 2015;Ma et al., 2014). Another known issue of the DMSP-OLS products is the "overglow" effect: dim lighting detected from light in surrounding areas of cities because of the scattering of lights in the atmosphere . A novel deblurring process was applied to address the issue of over glow in the radiance-calibrated products. This process involves the use of two sequential filters, a standard deconvolution and the frequency of illumination maxima, to withdraw the light from the surroundings back and restacking it vertically on its source pixels at city centers (Abrahams et al., 2016).
Deblurred DSMP-OLS RC annual composites for the years 1996, 2000 and 2010 were previously inter-calibrated and corrected for a multi-temporal analysis of urban form and city productivity in Latin America (Duque et al., 2017). In that work, the three nighttime images were used to delineate urban extents in each year for most of the Latin American and Caribbean cities that had more than 50,000 people in 2010. We used those delineated urban extents for the year 2010 to identify the metropolitan areas in the region. We consider the presence of a metropolitan area when more than one municipality or equivalent administrative unit intersects a single urban extent with more than 500,000 people in 2010. We use the administrative unit boundaries from the World Bank Latin American and the Caribbean Spatial Framework Database (Branson et al., 2016) for this purpose.
Metropolitan area boundaries were obtained by aggregation of all of the administrative units that intersected the same urban extent. We verified each obtained metropolitan area with ancillary information from official sources to include those municipalities that are part of the official metropolitan area denomination but were not intersected by the urban extent.

Estimated productivity differentials,
As presented in section 3, the first step to estimate the relationship between institutional fragmentation and economic performance of cities consists of extracting the productivity differentials between functional areas by extracting first the sorting effect that causes that more skilled workforce have a tendency to live in larger cities Combes et al., 2011). The vector of coefficients in equation 1, which becomes the dependent variable in the second stage (equation 2), was provided by Quintero and Roberts (2017) who studied the spatial variations in productivity premiums in 16 LAC countries. In their study, the authors use micro data on real hourly wage in the main occupation.
As independent variables the authors use: (1)

Measuring institutional fragmentation and metropolitan coordination
Based on the literature desk review and the available data, we construct a database with a series of variables to characterize the functional areas included in this study, in terms of institutional fragmentation and metropolitan coordination. We use the administrative boundaries of local governments that conform the metropolitan areas and distributed population data to calculate institutional fragmentation measures using geoprocessing tools in ArcGIS. Administrative boundaries were obtained from OpenStreetMap 3 (April 25 th 2017) and the World Bank LAC Spatial Database (Branson et al., 2016). We projected the administrative boundaries and the urban extents to the UTM Metropolitan coordination variables were obtained through a number of official information sources to account for the presence of a metropolitan governance body and public services single purpose districts (see Table A1). Tables 3 and 4 present the list of available variables for institutional fragmentation and metropolitan coordination respectively. The descriptive statistics are presented in Table A2.

Control variables
In order to isolate the predictive power of the variables describing urban form and to reduce omittedvariable bias, we include in the model a number of control variables including city size, locational variables, natural and urban amenities, as well as country fixed effects. A number of different data sources are used to compute control variables. Population data in gridded format for 2010 were obtained from the GHS layers. Natural amenities were calculated using several GIS layers: Water bodies were used to calculate dummy variables for location near the sea (coast). We used the 250 meters resolution raster MODIS Water Mask (Carroll et al., 2009) for this purpose. Finally, we used the Lloyd's lists of maritime and fluvial ports to account for the presence of ports. Table 5 presents the control variables. Administrative unit boundaries and population count at pixel level from GHS (GHS_POP_ PW42015_GLOBE_R2015A_ 54009_250_v1_0 at 250 meters of spatial resolution). III. Central city domination cc_pop_2010 Central-city population share_2010 Administrative unit boundaries and population count at pixel level from GHS (GHS_POP_GPW42015_GLOBE_R2015A_ 54009_250_v1_0)  Table A1 II. Land use plan and mobility its_cov Percent of municipalities covered by integrated transport systems between municipalities and central city (metro, bus) See Table A1 III. Coordination for single purpose districts spd_water Existence of a single-purpose districts for water See Table A1  spd_energy Existence of a single-purpose districts for energy See Table A1  spd_waste Existence of a single-purpose districts for waste collection See Table A1 spd_sum s p d_water + spd_energy + spd_waste See   Table 6 and Table 7 present the estimates of the relationship between metropolitan fragmentation/coordination and city productivity premium. The results are pretty similar in both cases. Following Ahrend et al. (2014) we report in the first column the positive and significant impact of population on productivity premium, which implies that productivity is higher in larger cities. In this regard Ahrend et al. (2014) reported estimated elasticities that range from 0.016 (for the United Kingdom) to 0.063 (for the United States). In our study, we obtained an estimated elasticity of 0.08 for LAC cities. We also report the coefficients associated to the logged population density (i.e., elasticity of productivity with respect to population) and surface (i.e., elasticity of productivity with respect to area). The results show that an increase in population, while holding the area constant, and an increase in area, while holding population density constant, both have a positive and statistically significant impact on productivity. Finally, the difference between these two coefficients indicates that an increase in area, while holding the total population constant, generates elasticities from 0.04 to 0.07. This range is 0.02 higher than the elasticity range reported by Ahrend et al. (2014), 0.02 to 0.05. (3) of Table 6 and Table   7. The results show opposite effects from fragmentation: on the one hand, the negative and statistically significant coefficient for the logged number of administrative units indicates that the presence of multiple local governments affects economic performance because of factors such as higher transactions costs, barriers to the diffusion of growth promoting policies, and other coordination failures (which is consistent with the centralist view). But, on the other hand, the positive and statistically significant coefficient for political fragmentation (no_au_100th_2010) indicates that the presence of multiple local governments may lead to more responsive government to public needs (Nelson and Foster, 1999). Also, "smaller [local governments] make participation easier, make citizens feel more empowered and interested in their communities, and bring neighbors together" (Oliver, 2010, 65). Finally, the negative and statistically significant coefficient for cc_pop2010_ue,

The block of indicators for institutional fragmentation is reported in column
indicates that an increase in central city domination may affect economic performance.
The fourth column in Table 6 and Table 7 includes the block of variables for metropolitan coordination. None of the coordination variables appears significant, which does not provide evidence in favor of the convenience of the regionalist model. It is important to note that these results are conditioned to the variables used to measure the degree of metropolitan coordination. Finally, column five shows that the above conclusions remain the same after including a series of control variables.

Conclusions
This paper studies the impact of metropolitan fragmentation/coordination on economic performance of 73 metropolitan areas in the Latin America and Caribbean region. This contribution offers complementary evidence on the relationship between institutional fragmentation/coordination and economic performance, since most of the available literature in this topic is concentrated on developed countries.
Following the latest contributions in the literature, we implemented a two-step econometric approach in which we control for individual sorting of highly skilled individuals into cities. We also take advantage of recent developments in remote sensing science and free geospatial libraries to delineate urban extents and identify metropolitan areas in an automatic and highly standardized way, which guarantees comparability across LAC cities.
The available literature has not arrived to a definitive answer on this topic, and there is evidence in favor of the three potential models: polycentric, centralist and regionalist. Our results show that there may exist an optimal level of fragmentation in which the benefits of more responsive government are in balance with the higher costs associated to the presence of multiple local governments within the same functional area. This may indicate that in LAC cities the right model is somewhere in between the polycentric and the centralist governance structures. We found no evidence in favor of the regionalist view, since our results show that the presence of a governance body or integrated public services does not necessarily foster increased productivity.
In line with previous contributions we found that economic performance increases with city size.
While evidence for OECD countries indicates that doubling city size may increase economic performance between 2% and 5% ; for LAC cities, we found an impact that ranges from 4% to 7%.

References
Abrahams