Stylized Facts on Public Good Agglomeration Economies

Get Complete Project Material File(s) Now! »

Institutional Background

Deprived peripheral neighborhoods in France : a persisting issue for urban policy

Urban segregation in French cities has been a perennial issue for urban and social policy-makers for decades. The post-war housing crisis lead to haphazard development of large housing compounds at the fringe of every French city in the 1950s and 1960s : Les Grands Ensembles (The Compounds), built for car-owners of the then-emerging middle-class with limited connections to urban transport networks or city centers (Newsome, 2004). Those housing complexes experienced important shifts in social composition in the 1980s when the middle class moved to city centers, followed by progressive relocation of low income pop-ulations in those ageing complexes, which left French cities facing large, spatially isolated deprived neighborhoods at the fringe of the major metropolitan areas.9
The Priority Neighborhood : a zoning policy Facing rapidly increasing urban inequali-ties, French policymakers have historically relied on zoning to define place-based policies. Among the many zoning policies defined since the 1990s, the most important and partic-ularly well evaluated one, since it was used for local tax incentives and enterprise zones10, was the « Sensitive Urban Zone » (Zone Sensible Urbaine – ZUS) defined in 1996. It was replaced in 2016 by the « Priority Neighborhood » (Quartier Prioritaire de la Ville – QPV).
Contrary to the ZUSes that were determined jointly by local and state officials based on qualitative criteria (presence of Grands Ensembles, unemployment level, etc.), the QPVs are defined on quantitative criteria that aimed at both avoiding political interference and ensure high similarity among the neighborhoods to allow for the evaluation of local policies. 1296 « priority neighborhoods » were defined by the French National Statistical Institute (INSEE) on the basis of the 2010 census. They regroup 5 million inhabitants in 702 different munici-palities. They were defined using 2010 census data by
• a median income below a threshold defined by :
– S = 0, 6 [(0, 7 National MedianIncome) + (0, 3 CityMedianIncome)] for urban units larger than 5 million inhabitants
– S = 0, 6 [(0, 3 National MedianIncome) + (0, 7 CityMedianIncome)] for the others
• a population larger than 1,000 inhabitants
However different in their definitions, the similarities of these two zonings are striking : if QPVs are smaller and more numerous than ZUS, they cover very similar zones. In our cities of interest, 92% of the ZUS (299 of 326) defined in 1996 have a 2010 QPV in their perimeter as can be seen in Figure 1-1. Permanence of urban segregation and spatial isolation (Briant et al., 2015) may explain such persistence.
The inhabitants of Priority Neighborhood Population composition is very similar across these compounds due to the zoning definition. These urbanites face multiple obstacles in both the labor market and housing market. First and foremost is geography: most of these territories are located on the outskirt of cities and are often surrounded by physical barriers such as railway lines or highways as exemplified by Briant et al. (2015). Such spatial iso-lation, coupled with lower car ownership rates, translates into lower mobility and higher reliance on public transportation (Nicolas et al., 2018). A second obstacle is social. These neighborhoods’ inhabitants are more likely to have immigrant backgrounds and suffer from a discrimination in the labor and housing market due to their origins and the bad reputa-tion of their neighborhoods (Mathieu et al., 2016; Bunel et al., 2017). Finally, as documented by descriptive statistics in Section III, they are also less educated and skilled than the rest of the population. These difficulties translate into higher poverty rates, unemployment and crime prevalence that are higher in those neighborhoods than anywhere else in metropolitan France. More specifically, at the beginning of our period of interest in 2005, 22.1% of deprived neighborhoods’ residents were unemployed compared to 11% at the national level.
Besides zoning and place-based policy, another lever to reduce spatial inequalities for local authorities has been transportation policy, which aims at reducing the isolation of de-prived neighborhoods.

Tramways in France : a revival motivated by social equity concerns

Tramways as an urban policy toolbox If electric tramways were common in European and American cities in the early 20th century, they totally disappeared after World War II due to the combination of low fuel prices, the rise of individual cars and a correlated shift of public investment towards road construction (Goddard, 1996). At the end of the 1980s, French cities initiated a Light Rail Transit revival (locally known as Tramways) through an unprecedented nation-wide consistent program11. In contrast with previous public transport infrastructure built in France, this tramway revival appears to have been largely motivated by increasing social concerns linked to urban segregation, which makes it of particular interest for our study.
Indeed, the tramway program represented a pivotal moment for the French doctrine on urban transportation decision-making. France transport infrastructure policy had been car-ried out across the whole territory since the 18th century by the central-state administrative body, the Ponts et Chaussées (Picon, 1992, 1994) whose decisions relied notably on utilitar-ian cost-benefit analysis following the tradition introduced by Dupuit (1844) and continued by Colson (1924-1928)12. The 1982 Deferre decentralisation bill suddenly moved the au-thority on public transportation from the all-powerful hands of central state to municipal authorities, which paved the way for better assessment of local political priorities in in-frastructure projects (Thisse, 2007; Offner, 2001), including social equity concerns. Lévêque (2017) showed for instance in the case of Lyons’ metropolis that connection between de-prived neighborhoods and the city center has been a constant goal of local transportation schemes since the mid-1980s. Similarly, the first tramway networks to be rebuilt in Nantes in 1985 and in Grenoble in 1987 were explicitly aiming to connect deprived peripheries to affluent city-centers. The popular success of these pioneer tramway made it one of the most popular urban planning tools among mayors nationwide to tackle the issues of congestion and urban segregation. Between 1998 and 2018, more than 25 cities in continental France built or extended tramway networks adding up to 600km of tracks and 800 stations (for a complete list of built lines and projected networks see figure A-3 in Appendix A.1.1). In most cities, tramway usage has exceeded the initial target by a wide margin.
Over the period in France, tramway constructions were the most important public invest-ments by local authorities13. This national trend for tramway building is the most striking example of a global movement in which 138 new tramway networks were built worldwide in the last twenty years, more or less explicitly mirroring the French experience tramway : comparative advantages The success of tramway with policy makers and com-muters stems from its relative advantages compared to the bus and the metro. For mid-size, budget-constrained cities, it brings some of the advantage of the latter in terms of comfort and frequency but at a fraction of the cost. Faster, more frequent, and only 3 times more expensive than the bus, tramway circulate on their own tracks and benefit from right of way at crossings, which is particularly advantageous to avoid congestion. Table 1.1 lists theses advantages. Unfortunately, there is no source of data which would allow us to easily com-pute the gain in travel time 14. Anecdotal evidence shows that they often are substantial. A rough approximation can be drawn under the hypothesis that the tramway simply replaces a bus line. Given the difference in speed, traveling from two points on the line would take between 49% and 31% less time than with a bus. The time gain are potentially even more important at peak hours when tramway benefit the most from its corridor.
A common radial design Most of the networks are very similar in their design. Tramway lines are radial; they connect to each other in central district and run separately to the peripheries through large avenues15. The only deviations from this radial design are observed either in the polycentric suburbs of large metropolitan areas such as Paris (Ile de France net-work) and Marseilles (Aubagne network), or in a dense city network with no prominent center, as in the case of Valenciennes, in France’s northern mining basin. Table 1.2 reports some descriptive statistics about network design. Most (77%) tramway works can be consid-ered as new line openings that exhibit a mean 18 stops per line, a 11.4km length and a mean distance of 630m between stops. This length typically corresponds to the distance between the city center and the fringe of the continuously built area. Those dense lines directly con-necting the city center to periphery supposedly create a significant accessibility shock for the periphery neighborhoods and, depending on the existing network, the rest of the city. By contrast, line extensions are generally short and should be excluded from our analysis.
A deviation from a utilitarian objective The peripheries connected by these radial net-works, though, are not random. They follow a common pattern by connecting both the main points of interest (POI) and the most deprived neighborhoods. Connected POI may be either central as the town hall, main hospital, train stations, or located at the urban fringe such as largest commercial malls and, if applicable, universities and airports, as shown by Table 1.3. However, the construction of tramway networks does not only aim at pursued a clear social equity objective: connecting the lower-income neighborhoods to the affluent city centers and to the rest of the city (Pissaloux and Ducol, 2012). Figure 1-2 exhibits maps of constructed net-works that reflect such objectives : we can see that QPV are quasi systematically connected, even at the cost of a deviation from direct center to periphery route or extension to areas that exhibit no specific POI. Table 1.3 similarly suggests that tramway tend to be diverted from utilitarian objectives such as connecting POIs to connect deprived neighborhoods. Thirty years after the beginning of this nationwide trend, many Quartiers de la Politique de la Ville in large French cities were indeed connected to an tramway network as shown by figure 1.

Data and Empirical Strategy


We combine several comprehensive administrative datasets that describe job, transportation and housing markets at individual or city-block levels to assess the effect of an tramway net-work on deprived neighborhoods. This unique database extends over 14 years and exhibits granular spatial precision, which crucially enables us to document change around the arrival of tramway at an infra urban scale.

Unemployment data

Unemployed individuals Our unemployment data set contains the universe of unem-ployed individuals who were registered at their local unemployment office between April 2005 and December 2018 (corresponding to a total population of 123,161 unemployed people in our neighborhoods of interest, and more than 20 million nationwide), their socioeconomic characteristics, education, unemployment history, benefit eligibility, job search sector, max-imal radius of search (expressed in distance or time) as well as postal addresses upon reg-istration, drawn from Pôle Emploi’s (French Unemployment Agency) Fichier Historique (FH) data set and completed with the outcomes of several internal working databases.
Censoring and Outcome of Interest Job seekers are required to notify their job agencies every month that they are still looking for a job to preserve their status. Additionally, lo-cal job agencies are only aware of the fact that job seekers find a job if they declare it when they terminate registration. As a result, a well-known shortcoming of this type of data is that we cannot always know for sure if job seekers stopped registering because they had in-deed found a job or only because they failed/forgot to notify their job agencies. Job seekers entitled to UI benefit are strongly encouraged to remain registered as it is a necessary con-dition to receive their benefits but many are not eligible20. Table 1.4 reports the motives for termination of registration.
We address this shortcoming in two ways. First, to avoid cases in which unemployed people simply forgot to notify their agencies for a given month, we only consider a spell to be terminated if job-seekers do not register again at Pôle Emploi in the following month. Second, we define two individual outcomes that are not affected by censoring and a block-level outcome that corrects it. For each individual spell we compute both the probability to have exited unemployment after 6 months, irrespective of the exit type and the share of days spent registered in unemployment in the two years following a registration. The latter allows us to capture potential longer term effects of the tramway not only on the probability of finding a job but also on the quality and durability of the match. Finally, to take into account the information on the exit types, we estimate a simple Kaplan Meier estimator of survival at 6 months for each quarter of registration – block cells where we set all incomplete spells at the end of the period and unknown destination to right censoring while defining two competing risks for « finding a job » and « exiting to non-employment. » We then define a corrected probability to have found a job with certainty at 6 months for an individual living in place j as P(Job)j = 1 ˆ ˆ Sj where S is the Kaplan Meier survival into joblessness at 6 months.
Geocoding Although crucial to a precise identification of the effect of public transporta-tion on the labor market, infra-municipality data remain scarce in the literature. To the best of our knowledge, we are the first researchers to use metric-scale geocoded data to describe urban labor markets. Using a phonetic string fuzzy matching algorithm on a comprehensive database of postal addresses, we are able to associate up to 85% of spells with the coordinates of the job seeker’s residence21. Figure 1-3 reports the output of this process on a small neigh-borhood. The precision of the coordinates that were found is actually sufficient to identify not only the building but also the staircase of residence of the unemployed individuals.22. Geocoding makes it possible to measure individual distance to the closest tramway stop upon registration. It is to be noted that the data on job-seekers’ addresses only exists for the job seekers registering in unemployment after April 2005. To put it differently, we do not know the location of the job seekers who are already registered for unemployment at this date, which prevents us from studying the effect of the tramway on the stock of job seekers in treated blocks.

Block level Data

To describe the mutations of a neighborhood’s population induced by an tramway, we turn to block-level aggregated variables. We define a city block by the most precise spatial unit available in the French cadaster, the Section Cadastrale.23
Population Composition We retrieve population characteristics at the block level from a fiscal database on the universe of population and housing stock (Fichier des Logements par Communes, henceforth FILOCOM) available every two years over 2000-2014 at the French Ministry of Housing. It provides information on each non-commercial dwelling every two years between 1995 and 2015. It displays the location of each dwelling, its surface and whether it is rented, owner-occupied, or if it is social housing. It also contains the num-ber of people who live in it, their age and income. Because of privacy constraints, the data set does not contain information on blocks of fewer than 10 households.24. It allows us to recover income, age and household composition for each block, to document the evolution of the composition of the populations of deprived neighborhoods following the arrival of the tramway. The cadastral map is not constant over time and sections can be yearly redrawn by municipal authorities. However, the use of dwelling identification makes it possible to follow changes and to construct constant sections. Moreover, with dwelling identification being held constant over time, one can use it to characterize population flow at the block level. Housing market We exploit the administrative PERVAL database from the French Board of Notaries (Chambre des Notaires) to further characterize the evolution of the neighborhoods in the housing market. It records transactions on the housing stock every two years from 2000 to 2014, localized at the block level and with detailed information on both the dwelling’s characteristics and the buyer’s and seller’s status and occupation. Notary records provide a representative sample of the French housing market25.


Because tramway are dedicated to local transit, stations are closer than those of heavy rail networks and lines may be extended more gradually. Spatial and temporal precision is thus necessary to describe their phased development and its effect on urban labor markets.
Tramway networks To do so, we built a comprehensive GIS database of tramway stops’ openings on a daily basis from 1985 to 2018. Geographic coordinates at a metric level are drawn from annual editions of the French National Geographic Institute (IGN) database BD-TOPO, supplemented and corrected when necessary with archival maps from local transport authorities. Timing of decision, construction and entry into service is very well documented thanks to the administrative process for infrastructure building. If each stop precise opening date can be easily drawn from local transportation authorities archives, we also have high quality data on the whole local decision process, which allows us to identify when the chosen route is known to the public. We built a panel of city blocks covering our period of interest, and compute the distance to the closest tramway stop at each time of the network’s evolu-tion. Figure 1-4 reports descriptive statistics on this development, on which it appears that the cities that pioneered the tramway renewal (Nantes, Grenoble, Strasbourg, Montpellier) exhibit a quasi total coverage of their QPVs by a tramway stop, while more recent networks, which only exhibit few tramway lines, are still in the process of achieving such coverage.
Note: this graph represents the City Share of Priority Neighborhoods located at less than 500m from an tramway stop in 2018 for cities for which tramway and not the metro is the main historical mode of transportation
We included automated light subways built in the cities of Lille, Toulouse from 1983 to 2018 to our sample, even if they are slightly different on a technical basis from the majority of the French tramways that are non-autonomous surface streetcars26.
Time-span and spatial extension of the analysis Considering the historical depth of these datasets, a common period of interest that allows us to compare the effect of an tramway on both unemployment trajectories, housing prices and population composition lies between May 2005 and September 2014. The revival of the tramway started in the 1980s and has lasted until today27, as a result, many neighborhoods among French tramway cities have been connected before or after our period of interest. We do not consider these openings in our analysis.
Moreover, to further ensure comparability between the different tram openings that we studied, we chose to exclude three cities: Paris, Aubagne and Valenciennes. This choice is notably motivated by the specific design of those networks. The Parisian tramway network is not radial and tends to link peripheries between themselves, which complicates the job market analysis. The Aubagne network is a subnetwork, but only a few kilometers long in a peripheral municipality in the larger Marseilles metropolis, which for local political reasons is not connected to the remainder of the metropolis’s network. The Valenciennes network, by contrast, connects two cities of equal size and exhibits a 20km interurban section with no stops between two city centers of equal size, which makes it more similar to a commuter train than to a tramway.


Empirical Strategy


As showed supra, French tramway developments explicitly targeted deprived neighbor-hoods and especially QPVs in a redistribution-motivated deviation from utilitarian plan-ning. However, as new infrastructure is costly, not every QPV has been connected to the city center during our period of interest. This offers an opportunity to estimate the impact of the connection of a deprived neighborhood to an tramway network by comparing connected and non-connected neighborhoods in a quasi event-study28 specification at the individual, dwelling or block level.
Are these neighborhoods comparable? The selection procedure of the QPVs ensures high comparability between these neighborhoods. Moreover, our period of interest stands in the middle of tramway development roadmaps in most treated cities: potential bias arising from comparing the first connected neighborhoods, which may have been chosen out of local un-observed urgency, with the last connected or never connected ones, is thus tampered. Finally, Figure 1-6 shows that connected and non-connected QPVs during our period of interest are actually very similar in levels and trends over most outcomes before our period of interest. This parallel pre-trends identifying assumption is verified in practice on every outcome of interest as showed infra.
Discussion on Causality Stricto sensu, these estimates would be causal if the connection of a QPV to an tramway during the period of interest is quasi-random and does not correlate with unobserved characteristics of the neighborhoods that would have an effect on our out-comes of interest. Similarity of pre-trends between treated and non-treated QPVs stands in favour of this identifying assumption.
However, tramway developments are not random since we have seen they also aim at connecting train stations, hospitals, city hall, stadiums or large malls altogether. Far from threatening our strategy, this feature can actually be seen as a source of quasi random varia-tion in tramway development since it is easier to provide tramway service to QPVs located between these points of interest, a location arguably unrelated to local neighborhood-specific unobserved characteristics. Table 1.5 shows that connected QPVs are more likely to be on a route to a POI than non-connected ones.
For a QPV, being closer to a convenient route from the city center to a POI increase the probabilities of connection to the tramway. Though, since we have seen the average tramway line is 11.4km long, being on a convenient route from the city center to a POI does not a priori imply better ex-ante access to the POI itself. It thus constitutes a factor of quasi-random variation in QPV connection unrelated to unobserved characteristics. However, it is not clear whether we shall use explicitly this exogenous variation to wield our results in the spirit of the inconsequential units approach developed by Chandra and Thompson (2000), since endogeneity issues may be different on the housing and the labor market. On the housing market, access to POI is not very likely to influence upwards the housing prices, but since we are sure that it influenced the design of the tramways network, we may want to restrict our analysis to the tramway lines that connect a POI and are most likely to connect the QPV located between the city-center and the POI only because of it inconsequential location. However, on the labor market, QPV located on the route to a POI are more likely to be ex-ante privileged since the benefit from a better connection to POIs that often constitute major employment centers (notably malls, airports, universities, hospitals). Thus, to mitigate this bias, one would prefer to restrict the study to the QPV that are not situated on any route to a POI. Moreover, any restriction of this kind would come at the cost of a large reduction of our set of pertinent tramway stops29, which would threaten the precision of our estimators. However, we provide the results of such restrictions in section 1.4 as a robustness check.


We follow transport economics literature to consider a neighborhood and its residents treated when a tramway stop is opening fewer than 500 meters from its border. Our control group is made up of neighborhoods that are not treated at that time (no tramway stops have been built at 1000 meters from the border of the block). We exclude areas yet treated by another rail infrastructure, should it be subways (Lyon, Marseille, Toulouse and Rennes exhibit metro lines) or existing tramways. Figure 1-6 shows treated and never treated groups for the city of Dijon, Burgundy.
To further take advantage of the precision of our data in the case of unemployed people, we consider them treated if and only if at the time of their registration into unemployment they lived in a treated block and their own individual distance to tramway’s next stop be-came lower than 500 meters. For the block analysis, we define as treated a block intersecting a priority neighborhood area and located at less than 500 meters away from a tram stop. This restriction aims at increasing the potential detected effect by focusing on the individuals and blocks which benefited the most from the new infrastructure.
These restrictions leave us with 195 treated blocks and 152 control blocks in 20 cities Zoning discussion The use of zoning defined on the 2010 population census, during our period of interest could be a problem if tramway had a drastic short term impact on location decision of households. Descriptive statistics show that urban geography of poverty being quite persistent over time, treated and non-treated priority neighborhoods were already very similar and quite poorer than other neighborhoods in 2005.

Estimation Strategy

Our identification strategy relies on multiple tramway line openings in different cities at different times. We compare the evolution of several outcomes for the unemployed, house-holds and transactions around the arrival of the tramway. We thus estimate the following individual and block level regression with a balanced panel :
Yi,t = å bk Dl(i),t,k + gXi + ll(i) + mj(i),t + ei,t (1.1)
Yl,t = å bk Dl,t,k + ll + mj(l),tei,t (1.2)
Where Yi,t is the outcome of an individual i (unemployed, household, transactions) in period t, k is the difference between t and the date of opening of the tramway in the neigh-borhood, ll(i) and mj(i),t are respectively a block l and a city j – year fixed effects and Xi is a vector of individual controls. Dl,t,k is a dummy that values 1 if a tramway line was opened in the vicinity of block l in k quarters before time t.
We only use tramway openings for which we can observe the treated blocks for the full pre- and post-treatment widow -s and s. This insure that each coefficient bk is estimated using the same set of control and treated blocks. More specifically, the analysis on population composition and housing prices (1.5) focuses on a -6, +6 year window. As we only have data from 2000 to 2014 on housing prices the openings we can study with those two data sets is limited to the openings occurring in 2006 and 2007. Meanwhile, the short-term analysis of unemployment outcomes (1.4) study the effect of the tramway in the 6 semesters before and after the opening of the infrastructure and rely on data available from April 2005 to January 2019. Taking full advantage of our data we carry the short-term unemployment analysis both for the entire period of availability of the data (reported in the main text) and for the common opening of 2006 and 2007 (reported in appendix).
Our estimations rely on within-city comparisons of blocks treated at the beginning of the period of interest with never-treated blocks, treated blocks, as well as to-be-treated blocks when available. Our estimation is akin to a « stacked » difference-in-difference (DiD). This class of two-way fixed effect specification has recently been the focus of a growing literature which highlights a challenge to the estimation of the average treatment effect (De Chaise-martin and d’Haultfoeuille (2020); Borusyak and Jaravel (2017); Goodman-Bacon (2018). In-tuitively, the bk estimated in stacked DiD is a weighted sum of the average treatment effect of several DiD. In the presence of heterogeneous treatment effects across groups treated at different points in time, group average treatment effects are sometimes assigned a negative weight. In practice such bias will only affect our analysis of the labor market outcomes (1.4) for which we do have several treatment dates and our control group is made up of both never treated and to-be-treated job seekers. We re-estimate our bk using the corrected Did estimator proposed by De Chaisemartin and d’Haultfoeuille (2020) which is robust to nega-tive weighting issues. For the analysis carried in the second part of the chapter, the restriction to opening of 2006 and 2007 and the biannual periodicity of our housing data amount to a unique date of treatment and no correction is needed. Hence, in this latter section the iden-tification will only rely on the comparison of treated blocks with never treated blocks 30.
The coefficients bk can be interpreted causally under the common trend assumption: in the absence of the tramway, treated and non-treated blocks and individuals in the deprived neighborhoods would have evolved similarly. The specification allows us to examine such assumption by observing if the outcomes evolved differently between treated and never treated in the periods leading up to the tramway installation.

Descriptive statistics

Individual labor market characteristics Looking at relevant labor-market characteristics of the treated and never treated populations shows inhabitants of Priority Neighborhoods to be very comparable while reflecting the hurdles they face in finding a job. The Table A.2 presents some descriptive statistics for the job seekers entering unemployment before the arrival of tramway (in the second quarter of 2005) in the two groups as well as for the rest of the population living in our cities of interest. As expected unemployed people living in Priority neighborhoods are both less educated and less skilled than the general population. Only 15% and 13% of them hold a University degree and strikingly they were respectively 56% and 55% to have failed to validate their last diploma. At the same time, they are under represented in managing positions and over represented in the unskilled workforce. Finally, they less often hold French nationality than the general population, which hints at but un-derstates the representation of workers of immigrant descent in Priority neighborhoods.
Block level characteristics Housing and income variables also support the comparability of our population and underline stark differences with the rest of the population. Housing prices and median income are very similar in treated and never treated blocks and respec-tively about 20% and 33% lower than for the rest of the population. Furthermore job seekers in our population of interest live in blocks where almost half of the dwellings are social housing.
Outcomes of interest In addition to Table A.2, Figure 1-6 plots the evolution of the different outcomes throughout the period of interest. It confirms the similarity of the two groups and reflects the existing gaps in job market outcomes. The differences are stronger when looking at the block level probability to have found a job with certainty. Once the censoring was corrected in the second quarter of 2005, we find that only 12 and 13% of the job seekers have found a job with certainty whereas 17% of the general population have, which amounts approximately to a 30% difference. Both the never treated and the treated curves are almost conflated whereas the gap with the rest of the population remains big for the entire periods.
The gap between the share of days in unemployment of the two populations is smaller at the beginning of the period but proceeds to grow slightly, notably during the 2008 crisis. Interestingly, registration status at 6 months does not hold the same pattern between the three groups.
Housing outcomes show more interesting patterns, as the difference between treated and never treated groups, which was initially negligible, increases after the opening of the tramway in 2006 and 2007.

Table of contents :

General Introduction 
1 A Streetcar Named Opportunity: Can Rail Foster Social Integration 
1.1 Introduction
1.2 Institutional Background
1.2.1 Deprived peripheral neighborhoods in France
1.2.2 Tramways in France
1.3 Data and Empirical Strategy
1.3.1 Data
1.3.2 Empirical Strategy
1.3.3 Descriptive statistics
1.4 An access to jobs ? Unemployment trajectories with a new transit option
1.4.1 Dynamic setting
1.4.2 Heterogeneity
1.4.3 How precise is this zero ? A comparison
1.4.4 Robustness checks
1.4.5 Long-term effects
1.5 Capitalization and population displacement
1.5.1 Estimation strategy
1.5.2 An accessibility shock that capitalizes into prices
1.5.3 Population change through migrations
1.5.4 The ambiguous role of social housing in preserving social mixity
1.5.5 Anticipation or nuisance ? Effect of tramway construction works
1.6 Conclusion
2 Optimal Spatial Policies with Local Public Goods and Location Preferences 
2.1 Introduction
2.2 Data
2.3 Stylized Facts on Public Good Agglomeration Economies
2.3.1 Raw Patterns
2.3.2 Descriptive Regressions
2.3.3 Preliminary Comments on Welfare Implications
2.4 Economic Geography Model with Local Public Goods
2.4.1 Central Government
2.4.2 Demand for Cities
2.4.3 Demand for Private Goods
2.4.4 Supply and Ownership
2.4.5 Demand for Public Goods and Tax Competition
2.4.6 Equilibrium
2.5 Optimal Policies
2.5.1 Intuition in a Two-Region Example
2.5.2 Efficient Allocations
2.5.3 Optimal Transfers
2.5.4 An Efficiency Test
2.5.5 Model Calibration
2.5.6 Efficiency of Observed Transfers
2.6 Equity and Density
2.6.1 Compensation and Responsibility
2.6.2 Revealed Social Preferences
2.7 Conclusion
3 The Carbon ‘Carprint’ of Urbanization: New Evidence from French Cities 
3.1 Introduction
3.1.1 Theoretical framework
3.1.2 Main contributions
3.2 Data on fuel consumption and urban form
3.2.1 Fuel consumption: a household measure
3.2.2 A set of quantitative measures of urban form
3.3 Empirical strategy and results
3.3.1 Urban form and fuel consumption: Baseline estimations
3.3.2 Urban form and fuel consumption: Causal estimations
3.3.3 Robustness checks
3.4 CO2 car emissions and city-size: a bell-shaped curve
3.4.1 CO2 car emissions of the sample-mean household across MAs
3.4.2 Driving footprint and city-size: A bell-shaped curve
3.5 Conclusion
General Conclusion 
A A Streetcar Named Opportunity
A.1 French Tramways
A.1.1 A brief history of Tramways
A.1.2 Descriptive statistics
A.1.3 The French public transport infrastructure decision process
A.2 Unemployment analysis
A.2.1 Geocoding procedure for unemployed’s addresses
A.2.2 Descriptive statistics
A.2.3 Heterogenity Analysis
A.2.4 Results on Common Openings
A.3 Housing market
A.3.1 Occupation of buyers and sellers
A.3.2 Robustness checks on housing market
A.3.3 Heterogeneity
B Optimal Spatial Policies with Local Public Goods and Location Preferences
B.1 Model
B.1.1 Local Public Good Demand
B.1.2 A Two-Region Example
B.1.3 Planner’s Problem
B.2 Calibration
B.2.1 Public Good Demand Calibration
B.2.2 Constancy of Expenditure Shares
B.2.3 Calibration of Public Good Index
B.3 Complementary empirical results
C The Carbon ‘Carprint’ of Urbanisation 243
C.1 Statistical zonings
C.1.1 French Metropolitan Areas
C.1.2 A monocentric typology of French Municipalities
C.2 Fractality
C.2.1 What is fractality?
C.2.2 The box-counting algorithm
C.3 Additional descriptive statistics
C.4 European Soil Data Base
C.5 Other specifications to take spatial sorting into account
C.5.1 Uncoupling socioeconomic and spatial effects
C.5.2 Absence of controls for socioeconomic characteristics
C.6 Other specifications with city-level and neighborhood level variables
C.6.1 with MA-centered 3Ds
C.6.2 with non MA-centered 3Ds
C.7 Alternate projections
C.7.1 Projection with MA-centered 3D variables
C.7.2 Projection with city-level and neighborhood level variables
C.8 Additional regression results
C.9 Complementary results on the bell-shaped curve
C.9.1 The mean household ’carprint’ when income varies across cities
C.9.2 Robustness checks on the bell-shaped curve


Related Posts