Home » 2013 Categories » 2013 – Technical presentations

Category Archives: 2013 – Technical presentations

Racial Housing Price Differential and Racially Transitional Neighborhoods

by Carmen Augustine TP_AUGUSTINECARMEN

            The question of whether black renters and homeowners face a different set of housing prices than their white equivalents is addressed in several studies through the period of integration and beyond. Past studies have drawn conflicting conclusions, some pointing toward a discount for black housing and others suggesting there is a premium paid by blacks. The present study is a response to Follain and Malpezzi’s 1981 review of racial price differentials in Chicago from 1974-1976, which reported significant discounts to black renters and owners in a majority of regions in Chicago. The survey omitted racial composition of each region as well as neighborhood control variables, which Chambers believes may have a significant effect on price differentials. For example, it may be the case that black renters and owners more often live in older or otherwise less desirable neighborhoods, which are factors outside of race that would cause the price of housing to be reduced.

Chambers additionally distinguishes between three types of racial housing differentials. A household differential exists “if black households pay more than white households for the same housing in the same market.” (Chambers, 216) Another cause of differential may be exclusion of black families from certain neighborhoods, raising demand for housing (and with it prices) in neighborhoods supporting a larger black population. The third cause is white households that prefer areas of more homogenous racial composition, causing housing demand to go up in less integrated areas and with it prices.

In order to more thoroughly assess the racial price differential, Chambers uses data from the 1975 Annual Housing Survey for Chicago as well as linked data from the 1979 survey to look at the effect of housing structure, neighborhood characteristics, renter/owner racial characteristics and neighborhood racial characteristics on the price of housing. This is an extension of the 1981 Follain and Malpezzi model that excluded both racial and non-racial neighborhood characteristics.

Presentation of the Model

Chambers first presents the following model to describe changes in housing price, Ei:

Ei = f(Hi, Zj, BLACKi, RCBj)

Where H represents housing structure characteristics, Z represents non-racial neighborhood characteristics, BLACK represents race of the renter or owner (a dummy variable equal to 1 if head of the household is black) and RCB is the percentage of housing units occupied by a black household. The subscript i represents one housing unit and the subscript j represents one neighborhood. H is a vector that includes many variables describing housing—number of rooms, presence of amenities such as air-conditioning and heating, parking and age (for a complete list, see Chambers 228-230). Z is a vector that includes average aggregate variables for a neighborhood, including household income, education level of the household head, neighborhood quality, property tax rate and others (for a complete list, see Chambers 231).

The model is estimated as a linear semi-log equation:

(1) ln(Ei) = a + bHi + cZj + dBLACKi + eRCBj + u1

Where each of the variables b, c, d and e are a coefficient on one of the independent variables and u1 is an error term. In relation to the comparable Follain and Malpezzi equation:

ln(Ei) = a + bHi + dBLACKi + u2

It can easily be seen that omitting the variables Z and RCB would cause the error term, u2, to be much higher than u1 if Z and RCB are significant. This is because if those two variables do in fact affect housing price, their effect will turn up in u2. Additionally, if either Z or RCB is correlated with H or BLACK we can state that the independent variables are endogenous in the Follain and Malpezzi model, which violates an assumption of OLS estimation and implies that our estimated coefficients (b and d) are biased. Chambers further speculates that the effect of omitting Z and RCB is a downward bias on the coefficient d. Z and BLACK are negatively correlated, and we anticipate the coefficient on Z to be positive (as neighborhood quality improves, price increases) yielding a negative bias. Similarly, RCB and BLACK are positively correlated but we anticipate the coefficient on RCB to be negative (as percentage black increases price of housing decreases), causing an additional negative bias. Thus, we expect Follain and Malpezzi to have lower estimates of the effect of race on housing price than the complete model.

Using data available from both 1975 and 1979, Chambers estimates the restricted Follain and Malpezzi model as well as his revised model. He finds that omitted variables bias the coefficients on BLACK and RCB downward. In the complete specification, the effect of race on rental prices almost disappears (drops to 0.3% premium in 1975 and 0.2% in 1979) and the effect of race on housing prices for owners drops significantly in both years, where it was strongly negative in the incomplete model. The effect of RCB on rental and housing price is negative in all cases except that for renters in 1975, suggesting that as the percentage of units occupied by blacks increases cost of housing decreases.

The results suggest that the observed black household discount is likely a result of racial composition and other non-racial amenities rather than race of renter or owner.

Addition of Racial Submarkets

Chambers breaks down housing price further by racial submarkets, speculating that each region may have an independent average price of housing and thus regions should not be aggregated across the entire Chicago area. Submarkets are divided into black ghetto with an average of 81% black occupied housing, black border with an average 27% black occupancy, Spanish submarket which is predominately Spanish but has 3% black occupancy and white interior submarket with 2% average black occupancy.

The updated model includes housing unit i, neighborhood j and submarket k as follows:

(2) ln(Ei) = a + bHik + cZj + dBLACKik + eRCBj + fRCBSQj + gSUBk + u3

The terms added in this iteration of the model include RCBSQ, which is a quadratic term reflecting racial composition, and SUB which is a dummy variable for each submarket. The coefficient g on SUB displays price differentials between submarkets.

The 1975 results suggest that rental prices are higher for black renters than white renters in the ghetto and Spanish submarkets, and lower for black renters in the black border and white interior submarkets. This is somewhat aligned with the theory that increased demand in areas supporting a larger black population cause prices to rise. In 1979 there is no black renter premium in the ghetto submarket, but a premium emerges in the black border submarket, which may reflect an “entrance fee” into communities bordering on racial transition.

For owners, the findings suggest that black households are given a discount in ghetto, black border, and white interior submarkets in both 1975 and 1979. This is not consistent with the “entrance fee” hypothesis, and may instead be a result of black owners purchasing homes in racially transitional neighborhoods within a submarket that are in lower demand.

The relationship between racial composition of the neighborhood and price was inconsistent across years for both owners and renters based on both the RCB and RCBSQ terms. Additionally it is not clear that a price differential exists between submarkets, as evidenced by the changing sign of coefficients on the dummy variables GHETTO and BLKBORD, representing homes in ghetto or black border submarkets.

Addition of Racial Transition Factor

To take into account the hypothesized discount due to racial transitioning within a neighborhood, Chambers adds a factor RTB to model (2) based on change in percentage of neighborhood population that is black from 1975 to 1979.

(3) ln(Ei) = a + bHik + cZj + dBLACKik + eRCBj + fRCBSQj + gSUBk hRTBj + iRTSj+ u3

The coefficient on RTB is negative and statistically significant, indicating that there is a discount on housing based on the level of racial transition within the neighborhood—neighborhoods experiencing more transition over the period had overall lower housing prices for both renters and owners. Looking at the same transition in percent occupied by Spanish residents (RTS), there is also a significant discount in housing price in neighborhoods with higher RTS.

Implications and Extensions

The study suggests that racial housing differentials are overstated in models that fail to include racial and non-racial neighborhood characteristics. The addition of a racial transition factor specifies that neighborhoods experiencing higher rates of racial transition offer an overall price discount, which may account for some of the observed price discount to black households.

The discount in racially transitioning areas implies lower demand for housing, which in turn suggests that integration of black and white households may be encountering some social barriers—whether consciously or not, households prefer to be in racially homogenous areas. One obvious extension of the present model is to look at racial price differentials in more recent years, as well as the 90s and the changes over the entire life of the study (1975-present). Additionally, it would be interesting to extend the survey beyond Chicago and look at other racially integrated urban areas and the difference between cities—for example, whether integration has been more complete in the northern or southern US and whether certain cities are more open to integration than others. I would also be interested in seeing the application of model (3) to Asian minorities in cities receiving a larger volume of Asian immigrants.


D.M. Chambers. 1992 (September). “The racial housing price differential in racially transitional neighborhoods.” Journal of Urban Economics: 214-232.

Beijing Land Market

by David Wang TP_WangDavid


Beijing Land Market

Over the last two decades, Beijing’s housing market has experienced tremendous growth that coincides with China’s overall economic growth and new free market housing policies.  This rapid growth is an interesting case study into the development of modern cities, as old housing is privatized or demolished to make way for new infrastructure, commercial developments, and housing projects.  As the Central Business District expanded, the city began to spread out towards the urban fringes.  Zheng and Kahn seek to examine the classic urban monocentric model in a city experiencing massive new development.  Building upon the monocentric model, Zheng and Kahn also consider Brueckner, Thissé, and Zenou’s (BTZ) theory of amenities to explain the similarities of Beijing to European cities, where high-income residents locate close to the city-center.  Beijing is quite similar to Paris, as the city center (designated to be the Tiananmen Square area) and nearby CBD contain many urban amenities that attract higher-income residents.  In fact, much of Beijing’s current urban form can be explained using the monocentric city model.  In addition, the capitalization of local public goods adds further insights into this developing urban form.

Before explaining the findings from the test of the monocentric model in Beijing, it is important to explain the three data sets used in the tests.  The first, a housing project data set, is a record of 920 new housing projects, which contain an average of 791 housing units each, in the Beijing market between 2004 and 2005.  These data are representative of the housing units purchased by Beijing homebuyers as there is little re-sale present in the market.  Since the projects are spread geographically around Beijing, their prices can be used to test the relationship between the housing prices and distance from the City Center.  The second, a land parcel data set, includes information about all land parcel auctions from 2004 to June 2006.  These auctions are the first step for developers to lease land to build a new housing project.  Once again, the selling prices of these open-auctions of land are used as a proxy for real estate prices.  The third data set is a detailed analysis of housing projects and their proximity to various local public goods, including public transit (subways, bus stops), high schools, major universities, crime levels, and environmental amenities (air quality, parks).

The testing of the monocentric model in Beijing is a fairly straightforward process. However, the controls used in the equation are extremely important.  The estimation equation is


where j is a parcel or project at location q in year t.  The controls are dummy variables for the region of Beijing in which a land parcel (or housing project) is located and the date of the land parcel auction (or housing project sell date).  These regions are defined by four quadrants using Tiananmen Square as the origin.  The inclusion of these controls is an appropriate way to control for any inherent differences in the regions that a simple distance measure would not be able to capture.  For example, by controlling for quadrant (region of Beijing), the results show that land prices in the Southeast are 41% cheaper relative to the Northeast.  If this region control were omitted, bias could be introduced into the coefficient for the effect of distance on land price.  This estimation equation is run twice, once for the land parcel data and once for the housing project data.  With the estimation on land parcel data, it was found that an extra kilometer of distance from the CBD reduces land price by 4.8% (including both commercial and residential land).  However, when restricting the regression to only residential parcels, the land price gradient falls to 4.3%.  As Zheng and Kahn suggest, this result may be due to agglomeration economies, where “land closer to the CBD is more valuable for non-residential users.”  When estimating the equation for the housing project data, only a slight decrease in price of 2% per kilometer away is found with an R2 value of 0.175.  Although Zheng and Kahn do not explain this result, it is possible that since controls for transportation were not included, there could be limited conclusions to be made about the housing prices.

Another interesting characteristic of Beijing’s urban form is the relationship between the zoning rules provided by the Land Authority under the Beijing Master Plan.  Using a regression similar to (1), but replacing the dependent variable with FAR, the floor-to-area ratio (a measure of density on a parcel of land) declines with distance from the City Center.  However, this decline in density is limited to commercial land parcels and is not significant with residential land parcels.  In other words, tall commercial buildings are located towards the City Center, while residential building heights are relatively flat across the city.  Government urban planning forces may be causing this flat construction density.  With Tiananmen Square, a historic landmark, at the City Center, the city’s urban planning commission set restrictions on the height of nearby buildings.  At the same time, the planners want to increase building height as the distance from Tiananmen Square increases, in order to create a skyline for the city.  Thus, the flat density gradient of the residential buildings may result from the combination of the urban planning pushing up building height and market forces pushing down building heights with distance.  Therefore, to summarize, land and real estate prices decrease with respect to distance from the City Center.  However, the zoned density of residential projects does not fall with distance from the Center.

The final portion of the paper concerns the monocentric model with additional considerations for the capitalization of local public goods in the real estate prices.  Since homeowners in China do not pay residential property tax, the value of the public goods should have a higher effect on property prices.  Normally, a concern with studying the capitalization of public goods is whether they are exogenously or endogenously determined.  Zheng and Kahn argue that due to the former planning economy, Beijing is a good candidate for such a study.  The central or city government, without necessarily any consideration for market forces, planned the locations of public goods, such as schools, parks, and universities. Other local public goods measured include public transportation stops, air quality, crime levels, university distance, and also university quality (through entrance exam scores). Zheng and Kahn use standard hedonic methods to estimate the capitalization effects.  Using ordinary least squares, equation (2) is estimated, where the dependent variable is the log of the price per square meter of housing in project j located in community q at time t:
X1j represents the physical characteristics of an average unit in a new project.  Zheng and Kahn claim that this control succeeds due to the conformity amongst the housing units in the projects, where each unit has similar building structure, internal space, and decoration.  In addition, since these projects are all new construction, they should all be of approximately the same age and this factor should be already controlled.  Another interesting variable added to the regression is a dummy variable controlling for whether the project is built by a state-owned enterprise (SOE).  Since SOEs are owned by the state, they do not have an incentive to aggressive push for sales above expected market values, resulting in projects that are on average 10% cheaper than private housing projects.

Zheng and Kahn examine the individual effects of the public goods by running the regression multiple times, adding a variable for an additional local public good with each subsequent regression. For example, the first regression includes only the variable for the nearest subway stop distance, while the second regression includes both the subway stop distance and also distance to the nearest bus stop.  With the addition of each public good, the regression’s R2 value increases.  The regression’s explanatory power increases when controlling for distance to local public goods.  From the analysis, we find that air quality, parks, universities, and schools affect home prices, while transit and crime have no significant effect.  Crime may not have any effect because it is mostly concentrated around the city fringe, where migrant workers congregate.  These workers have huge demand for living and may outweigh any negative capitalization of crime.

Zheng and Kahn’s testing of the monocentric model in Beijing provides a consistent explanation of the shape of Beijing’s urban development.  Her collection of data is broad and consideration of alternate regressions shows the completeness of her analysis.  Her use of local public goods in Beijing to study capitalization effects is interesting and her assumption that these goods are exogenously determined is quite reasonable, as China was centrally planned.  However, the location of some public goods, such as the old universities, Tsinghua University and Beijing University, were determined even before the central planning.  In the past, these universities perhaps were located due to endogenous reasons, though it is unlikely that any capitalization effects carried over to the current state after the revolution of the twentieth century.  In addition, we cannot easily attribute causality in any of the regressions.  It may be appropriate to run some difference-in-difference estimators to compare directly the differences between land parcels.  This may further parse out causal effects of distance.  It is difficult to think of additional data that could be collected, but perhaps further exploration could be done into the transportation systems.  Since Beijing recently changed the subway system to a flat fee, transportation costs from far distances in the city should be decreased and may affect some of the results in the monocentric model view of the city.  This sudden change in transportation cost may provide an opportunity for some fixed effects estimators of before and after prices.  Nevertheless, Zheng and Kahn’s study provides valuable insights into the urban development of a large city within an economy in transition.  The unique data available after China’s many market reforms gives a great starting point to test the monocentric city model without bias from historical norms.







Works Cited

Zheng, Siqi, and Matthew E. Kahn. “Land and Residential Property Markets in a Booming Economy: New Evidence from Beijing.” Journal of Urban Economics 63.2 (2008): 743–757.


The Holdout Problem, Urban Sprawl, and Eminent Domain

by Lucas Mitchell

Urban sprawl is a well-documented phenomenon that most associate with a negative connotation.  According to Nechyba and Walsh (2004), urban sprawl is “the tendency toward lower city densities as city footprints expand.”  The standard monocentric city model predicts this trend; the city edge outwardly expands at lower density as transportation costs fall and incomes rise.  The Tiebout sorting effects heighten this spatial growth, to the point that many consider sprawl inefficient.  Brueckner (2000) identifies three sources of market failure (unaccounted externalities) that cause this inefficiency: social value of agricultural land as open space, congestion from commuting, and necessary infrastructure for land development.  Miceli and Sirmans introduce the holdout problem as another externality causing inefficient urban sprawl.

Large-scale development projects require the acquisition of land from multiple owners.  Necessary for urban growth, projects both public and private typically depend upon many individual transactions.  When this information is known, individuals realize they can artificially inflate the price of their land.  The holdout problem occurs when these individual owners assemble to extend negotiations and raise the price of multiple parcels, increasing transaction costs.  Because these costs increase with the number of land owners, developers seek to lower costs by purchasing land from fewer owners with larger parcels – pushing development projects down the density gradient away from the central city.

Miceli and Sirmans model a simple game to represent the holdout problem.  A developer attempts to purchase land from two sellers, who can either bargain or holdout.  There are two periods of bargaining, so holding out from bargaining in the first period results in bargaining in the second.  The value of the land together to the developer is greater than to each seller (V > 2v).  There is a cost c to delaying development to the second period.  As a result, the developers payoff D = VcP~ (if both purchased in period 1), D = VcP1P2 (if one purchased in each period), D = VcP* (if both purchased in period 2).  If forced to scrap the project after acquiring one parcel, the developer faces D = vP1The costs associated with delaying development (and the developer’s knowledge of those costs) provide for P* > P1= v, P~  > P* and P2 > P*, though depending on the magnitude of these costs P~  > P2or P~  < P2.  Either way, the sellers are playing a non-cooperative game, though if the latter is true (c < (V – 2v)/2), the game is a prisoner’s dilemma.  If c < (V – 2v)/2, the dominant strategy is holdout; the only Nash equilibrium is (holdout, holdout).  If not, the Nash equilibria are (bargain, bargain) and (holdout, holdout); though the sellers, lacking full information on the delay costs, would most likely be unsure of which game they are playing.  The joint holdout is the resulting equilibrium of the game; empirically we expect a delay of land assembly projects which increases as the number of parcels increase.

Table 1: Payoff matrix for the sellers’ bargaining game


Seller 1

Seller 2




P~, P~

P1, P2


P2, P1

P*, P*


From empirical evidence supporting the monocentric city model, land prices increase and lot sizes decrease as distance from the central business district decreases.  Thus, the effect of the holdout problem steepens the slope of development costs versus distance from the city center.  The market inefficiencies causing sprawl may be fixable.  As many developers already know, maintaining secrecy about a project by using dummy buyers reduces the holdout problem.  To counter urban sprawl, the government may implement policy to tax suburban development or subsidize inner city redevelopment.  Miceli and Sirmans also suggest the government employ eminent domain to foster urban renewal projects, though use of such power to aid private development efforts may be controversial.  According to them, the redevelopment of land close to the city center serves a public purpose by combating urban sprawl; thus the holdout problem can be seen as justification for the use of eminent domain.




Works Cited

Brueckner, J., 2000. Urban sprawl: diagnosis and remedies. International Regional Science Review 23, 160–171.


Miceli, Thomas J and C.F. Sirmans. “The holdout problem, urban sprawl, and eminent domain.” Journal of Housing Economics, Volume 16, Issues 3–4, November 2007, Pages 309-319, ISSN 1051-1377, 10.1016/j.jhe.2007.06.004.



Nechyba, Thomas J. and Randall P. Walsh.  “Urban Sprawl.” The Journal of Economic Perspectives, Vol. 18, No. 4, Autumn 2004, pp. 177-200.



Urban growth boundaries: An effective second-best remedy for unpriced traffic congestion?

By Katerina Valtcheva  TP_Katerina


For this technical presentation I have chosen Jan Brueckner’s paper from 2007 titled “Urban growth boundaries: An effective second-best remedy for un-priced traffic congestion?” In this paper he explores the welfare gain from using UGB as a tool for reducing the distortions created by a first-best toll regime in a congested city and, specifically, un-priced traffic congestion. His paper is based on previous scholarly works, which have proven that appropriate urban growth boundaries (UGB) improve welfare: but what Brueckner is interested in his paper is quantifying this welfare gain. Brueckner’s findings show that UGB are not an effective substitute for a first-best toll regime, nor are they effective as a second-best method in a congested city. These findings are relevant for cities that satisfy the assumptions of the standard monocentric city model.

Brueckner uses Anas and Rhee’s 2006 paper “Curbing excess sprawl with congestion tolls and urban boundaries” as a starting point for his work. In their paper, the two economists numerically evaluate congestion tolls and UGB in a congested city. Their findings show something rather different than previous works on the subject. It turns out that when a city substantially differs from the standard monocentric model used by Brueckner, a toll regime is shown to increases welfare, while the use of UGB, in fact, reduces welfare. What make their result so divergent from previous findings are their assumptions that the city has dispersed employment, where trips within the city are explained by commuting as well as shopping, and “consumer location choices are influenced by random idiosyncratic preferences” (Brueckner 2007).

In his paper, Brueckner makes assumptions based on the standard framework used for a congested monocentric city. Some important assumptions utilized include Cobb-Douglass preferences of consumers over housing consumption and the radial symmetry of the city with a constant fraction of the land used for housing and the rest for radial roads. In addition, he follows a paper from 1985 by Pines and Sadka when he assumes that the city is fully closed with each resident receiving an equal fraction of the aggregate residential rent revenue, which allows him “to conduct a straightforward welfare analysis” (Brueckner 2007). He also assumes that the revenue from tolls is equally redistributed to consumers.

The author defines a number of equations, which express the number of residents that live beyond a certain distance from the central business district (CBD) (Fig. 1), the cost per mile of commuting at a certain distance (Fig. 2), the congestion toll per mile at this distance (Fig. 3), and the commuting cost from that distance (Fig. 4), respectively.


Figure 1


Figure 2


Figure 3


Figure 4

By differentiating the equations in figure 1 and figure 4, Brueckner obtains two relationships (Fig. 5).


Figure 5

The first one shows how, as the distance from CBD increases, the population that lives at a greater distance decreases at the same rate; the second relationship shows that as this same distance increases, so do commuting cost at a rate that equals the cost of direct commuting at that distance along with the cost of the toll. Finally, Brueckner shows that income must equal the sum of exogenous income, income from rent and the income from the toll. It is important to note that in Brueckner’s analysis the “density in any one location in the city depends on densities at all other locations” (Brueckner 2007) as in his model his city is congested. As an implication, instead of solving a set of static simultaneous equations, Brueckner uses an iterative procedure that relies on the equations shown on figure 5 to find the equilibrium of the system.

The results from the method previously described allow Bruecker to compare the welfare generated in a laissez-faire equilibrium, a toll-regime equilibrium, and an equilibrium under an optimal urban growth boundary. As he compares the equilibria generated under four cases of different parameters, he finds that UGBs fail to increase central density in congested cities. This failure to noticeably raise densities around centers of employment is the main reason Brueckner gives for the poor performance of the method. As this is unlikely to change even if a city has more than one such center, Brueckner believes that his findings will hold even if he did not use the monocentric city model in his study. This is one area where I see potential for further improvement of the paper and a strengthening of its implications via performance of numerical analysis for a city with multiple employment centers as well as performance of the same analysis when adding other features to the model. Proving that UGBs are not an effective method of reducing distortions caused by a first-toll regime regardless of the assumptions used to model a city would increase the scope of Brueckner’s findings.

Brueckner’s paper is important as it quantifies the results from previous papers and, thus, takes the knowledge about UGB a step further. Without it a city could have wasted time and resources setting a growth boundary in hope of reducing un-priced traffic congestion only to find out that despite the fact that UGBs generally increase welfare, they are ineffective when used for the previously mentioned purpose.


Brueckner, Jan K. “Urban growth boundaries: An effective second-best remedy for unpriced traffic congestion?.” Journal of Housing Economics 16.3 (2007): 263-273.

Anas, A., Rhee, H.-J., 2006. Curbing excess sprawl with congestion tolls and urban boundaries. Regional Science and Urban Economics 36, 510–541.


Linden, Leigh, and Joanh E. Rockoff. 2008. “Estimates of the Impact of Crime Risk on Property Values from Megan’s Laws.”

by Ingrid Zhuang TP_ZHUANGINGRID

Linden, Leigh, and Joanh E. Rockoff. 2008. “Estimates of the Impact of Crime Risk on Property Values from Megan’s Laws.” American Economic Review, 98:3, 1103-1127

Crime rate, victimization, and the fear of crime risk are studied predominantly as local issues.  In response to crime risk, residents either vote for anti-crime policies, or they choose to relocate.  Therefore, local response to crime is particularly discernible in the housing market, since individuals can reduce their exposure to crime without moving great distances (Linden and Rockoff, 2008).   Individuals’ strong distaste for crime, especially sex offenses, indicates an inverse relationship between housing values and proximity to registered sex offenders, as observed by existing studies.  Understanding the relationship between property values and local crime risk is important in determining optimal policy decisions, such as proper level of policing and expenditures for programs that reduce crime.


Linden and Rockoff indicate in their paper that previous literature had potential omitted variable bias in both the cross-sectional and the time series models, and crime rates are likely to co-move with other unobserved characteristics in a neighborhood.  Linden and Rockoff (2008) improves on past estimates through the use of hedonic estimation methodology to measure the impact of crime risk on property values.  They overcome the bias problem by using cross-sectional and time series data on the timing and the exact locations of sex offenders’ arrival based on the implicit assumption that the small neighborhood around a sex offender is relatively homogeneous.  The timing of a sex offender’s move-in allows Linden and Rockoff to confirm that the change in property values is not caused by other preexisting shocks.

Three sets of data are analyzed.  The first is a January 2005 data on registered sex offenders in Mecklenburg County, North Carolina, provided by the North Carolina Department of Justice (NCDOJ).  It contains information on sex offenders’ names, basic characteristics, types of crimes, incarceration dates, addresses of where offenders currently live, and registration dates.  The second set of data is collected from the Mecklenburg County division of Property Assessment and Land Record Management, providing information on all real estate parcels in the county and comprehensive physical characteristics for each parcel.  The third set is a total of 169,577 property sales of a ten-year period (from January 1994 to December 2004) in the Mecklenburg County. Linden and Rockoff choose to limit the time period to a four-year window: two years prior and two years after the offenders’ arrivals.  They match the first dataset with the second dataset to pin down the exact location of registered sex offenders, and merge with the property sales data to exploit the exogenous variation from the move-in to estimate the property value changes.

Assuming living in close proximity to a sex offender has a negative impact on nearby property values, one should expect a fall in prices of homes near the offender’s location subsequent to the offender’s arrival, with the largest impact on homes closest to the offender.  The graphical evidence confirms the hypothesis.  Comparing to pre-arrival price gradient of distance, figures 2B and 3B exhibit a clear decline for sales during the year within 0.1 miles of the offender.  Homes slightly farther away are less affected.



Model and Estimation:

Linden and Rockoff (2008) uses a cross-sectional difference estimator and a difference-in-differences estimator to test the graphical evidence.

Cross-sectional difference estimator:
TP_Zhuang-3, (1)

where the log of the deflated sale price (sale price/CPI) of the house is a function of a measure of distance from the sex offender, a year specific effect (αt), and a random error term (εijt).  D1/10 ijt is an indicator variable: when a property sale occurs within 0.1 miles of a sex offender’s location it equals to one.

Difference-in difference specification:

TP_Zhuang-4, (2)

adds an indicator variable (D3/10ijt) for homes within 0.3 miles of an offender’s address and an interaction of this indicator with an indicator to test whether the sale took place after the offender’s arrival (Postit).  αjt is a neighborhood-year fixed effect, Xi observable property characteristics, and the term π1 gives the estimated impact of close proximity to a sex offender on property values.

Cross-sectional difference estimator is used to confirm the absence of preexisting differences in the characteristics of homes located within 0.1 miles of an offender.  Table 2 shows two regressions, where Panel A estimates all houses within 0.1 miles of the sex offender’s location that are sold BEFORE his arrival.  Panel B has the same estimation but included all houses regardless of whether sold or not, thus they could not estimate the first two columns of Table 2.  The insignificance of the results demonstrates a high degree of homogeneity in the data.


Table 3 presents statistical estimates of the impact of a sex offender’s arrival on the nearby housing values.  Column 1 shows estimates of equation (1), including sales of all homes in the dataset and sale-year fixed effects.  The estimate of the impact (π1, which in this case is simply a measure of the average price difference between houses within 0.1 miles of an offender’s future location and other houses sold within the same year) is approximately 34 percent.  Significant at 5 percent level, the estimate confirms that homes closer to offenders’ locations are relatively cheap compare to other parts of the county.  Column 2 builds onto column 1 by adding neighborhood-year fixed effects and house characteristics to the regression.  The results are not statistically different from zero at any confidence level, implying that control variables in the regression capture almost all of the differences between areas in which offenders move and the rest of the county.  In a simple pre-post comparison, column 3 shows estimates of equation (2) without the indicator variables for houses selling between 0.1 and 0.3 miles from the offenders.  Linden and Rockoff find that a sex offender’s arrival caused on average 4 percent decline in housing prices within 0.1 miles of an offender’s location, 0.7 percent pre-existing difference in prices.  The 3.3% difference is significant at the 10 percent level.  Linden and Rockoff’s difference-in-differences specification results are recorded in column 4.  The estimates show a slightly higher impact of a sex offender’s arrival, with -4.1 percent at the 4 percent significant level.  The impact of offender’s arrival for homes located between 0.1 and 0.3 miles of an offender’s location is curious but statistically insignificant.  The result indicates that homes slightly farther away experienced little to no decrease in property values on average.  Column 5 re-estimates equation (2), using only property sales from areas with sex offenders’ presence.  Instead of controlling for neighborhood-year fixed effects, Linden and Rockoff control offender area-year fixed effects and estimate standard errors clustering at the offender area level.  This approach provides better identification by focusing on houses within 0.1 miles of offenders.  Since the results are consistent with those from columns 3 and 4, they conclude that additional data from sales outside of offender areas did not bias their estimates.  Adding an interaction of the dummy variable indicating a sale within 0.1 miles of an offender after the offender has moved in, with distance from the offender, column 6 shows no changes in results. Robustness and falsification tests are conducted further to attest their conclusion.  Due to the focus of this technical presentation, detailed explanations on these results are omitted.



From the hedonic estimations, Linden and Rockoff conclude that houses within a one-tenth mile area around the home of a sex offender fall by 4% on average (about $5,500).  The result suggests that residents have a significant distaste for living in close proximity to a known sex offender, and they would be willing to pay a high cost for policies that remove sexual offenders from their neighborhoods.  As Linden and Rockoff mentioned in their paper, one possible extension for this study could be adding data on buyer or seller characteristics in order to avoid overestimating or underestimating the average willingness to pay due to the fact that only prices for houses that sell were analyzed.  Another contribution would be to explore whether the recession has changed the relationship (comparing pre-recession reaction to post-recession reaction) or examine whether neighborhoods of various demographics will react differently to the presence of sex offenders.

Technical Presentation on Baum-Snow and Lutz’s “School Desegregation, School Choice, and Changes in Residential Location Patterns by Race”

by Christopher Bradford TP_BradfordChristopher

Baum-Snow and Lutz consider the responses to school choice and residential location following the desegregation of the American public school system. Data include 92 metropolitan statistical areas (MSAs) that underwent large-scale court-mandated desegregation between 1960 and 1990. The study uses these data to determine the extent to which desegregation was a causative agent for changes in public/private school enrollment and for residency in central districts/suburban areas for both whites and blacks.

The authors note that, while the 1954 case Brown v. Board of Ed of Topeka ruled school segregation unconstitutional, most large school districts did not initiate robust school desegregation until specific court orders forced them to do so. As such, the authors devise a model that accounts for this variation in school desegregation over time, Equation 1:


(1)     ln yrjt = αrj + βrt + cDrjt + εrjt


Here r represents a region (e.g. South), j represents a specific MSA, t represents time, βrt accounts for year-specific effects, Drjt represents the central district’s degree of desegregation at time t, and yrjt represents the variable of interest. The three main variables of interest yrjt for which the authors solve are (1) public school enrollment by race, (2) private school enrollment by race, and (3) population in central districts by race.

As a methodological note, the authors comment that identification of the constant c, the parameter of interest that modifies the degree of desegregation at a given time Drjt, necessitates that timing of desegregation is uncorrelated with any time-dependent, causative omitted variables. The authors feel that desegregation implementation occurred with “pseudo-random timing” due to two observations (8). First, the NAACP tended to file cases in the order in which they were most likely to succeed, not according to where the perceived need for desegregation was greatest. Second, the time between court uptake of a case and issuance of the court’s verdict and implementation was highly variable by district. In an attempt to correct for omitted variables in the overall model, the authors allow the term βrt to differ for South and non-South Census regions, as unmeasured (or immeasurable) variables such as differential degrees of discrimination could vary by region.

In order to gauge racial integration following court-mandated desegregation, the authors utilize a “dissimilarity index,” which ranged from 0 (perfect integration) to 1 (perfect segregation). This dissimilarity index is given by:

(St) St = ½ *TP_Bradbit/Bt – wit/Wt]


Here bit and wit represent the number of black and white students at a given school i at time t, and Bt and Wt represent the total number of black and white students in that district at time t. Using data from the 1970, 1980, and 1990 Censuses, the authors find that the dissimilarity index decreased by a national average of 0.15 due to desegregation, representing a reduction of nearly one standard deviation.

In addition to the dissimilarity index, the authors construct an “exposure index,” which gives the percent of black students in the average white student’s school. This exposure index is given by:

(Et) Et = 1/WtTP_Bradwit * bit/tit


Here tit represents the total enrollment of both races in a given school i. Substituting the exposure index into Equation 1 yields the result that desegregation increased interracial exposure in schools by 0.09, or roughly half of one standard deviation.

To account for spatial variation in desegregation, the authors amend Equation 1. Each North and South region r is divided into four location segments s. The modified Equation 2 takes the form:


(2)     ln E (yirjts) = arjs + brts +  ϒrs Drjt


Here i represents a Census tract within a given MSA j. While the parameter of interest in Equation 1 is c, it is ϒrs in Equation 2. This term accounts for spatial variation by region and segment.

Taking data from 77 MSAs that contained a suburban region, the authors find that over 90% of these MSAs were less racially integrated than the MSA central district in 1970. The average Southern suburb had an exposure index value 0.15 less than in the central district of the MSA. In the North this discrepancy was even larger, at 0.26. The implication of these large differences in integration is that departing central district residences for the suburbs allowed whites to escape exposure to black students.

Indeed, the study finds that white enrollment in public schools in central districts declined by a national average of 12% from 1960 to 1990 due to desegregation. The figure in the South is slightly higher than the national average and slightly lower in non-Southern regions. The models predict that in non-Southern regions, desegregation led to a 16% increase in white private school enrollment in the 30-year period. The authors fail to find statistically significant evidence on the effect of desegregation on white private school enrollment in the South. Data supported that white enrollment declines in central districts public schools were offset by increased enrollment in suburban public schools, however. The evidence for non-Southern regions suggests a causal relationship between desegregation and private school enrollment for whites.

Moreover, the models predict that desegregation caused an average of 6% nationally of the central district’s white population to relocate outside the central district. Again, these results are more pronounced in the South, where 12% on average of whites residing in the central district relocated due to desegregation. Significantly, these predictions attribute a proportion of residential white flight from integrated urban school districts to suburban districts directly to desegregation.

For blacks, the models predict that desegregation resulted in a 14% increase in public school enrollment in the central district on average. These results were captured in a period of at least 5 years after court-mandated desegregation policies were implemented. Desegregation caused the percentage of blacks enrolled in private schools in central districts to decrease by 20-28% nationwide within five years of implemented integration. This figure was much larger in the South than in other regions. The data suggest that large numbers of black students left private schools to enroll in newly desegregated public schools, particularly in urban central districts.

The population of blacks living in central districts nationwide increased by about 8% due to desegregation. This result was only obtained in the non-South; the effects of desegregation on central district population in the South were indeterminate. The authors note that the effects of desegregation on black schooling and residency patterns display much more cross-regional variation than for whites. Accordingly, unobserved interregional variables may have played a larger role for influencing black schooling and residency patterns post-desegregation.

Using their models, the authors confirm two predictions of Tiebout sorting. The first predicts that public enrollment and relocation will occur more in the periphery of the central city, where wealthy individuals tend to live in. A Tiebout assumption is that the marginal utility of local public goods (such as public schooling) increases with income. Accordingly, high-income residents are predicted to be the most susceptible to decreases in marginal utility due to school quality deterioration, which is an oft-perceived consequence of school desegregation. Indeed, the study finds that both white public school enrollment and total white population decreased the most in the outer fourth of central districts.

The second prediction is that private school enrollment changes will be more dramatic near city centers. Tiebout sorting predicts that residents who use private schooling will tend to live closer to the city center than those who use public schools, as savings on commuting costs can partially defray the cost of private school tuition. This prediction is also confirmed, as the authors observe the increase in private school enrollment for blacks and the corresponding decrease for whites in regions closer to city centers than they observe the enrollment responses for public schools.

In their conclusion, the authors note that the magnitudes of population shifts due to desegregation are insufficiently large to account for the majority of total central district population loss in the past half century. The authors calculate that, had court-mandated desegregation not occurred, the decrease in white central district population would have declined by 10% from 1960 to 1990, whereas the actual decline was 13%. For blacks in the hypothetical no-desegregation scenario, the authors predict a 44% increase in central district population as compared to the 54% actual figure. Accordingly, the authors conclude that other factors must have been responsible for the brunt of city center population changes.

What the study does show is that desegregation contributed to decreased white enrollment in central district public schools and an increase in enrollment in these schools for blacks. It also shows that desegregation was a causative agent for changing both the racial makeup of urban central districts and private school enrollment for whites and blacks.

School integration remains to this day a large policy issue in America. It is also a significant local concern, with the recent struggles of the Wake County district a prominent example. For my term paper, I may explore topics that branch out from this paper, such as how racial integration in public schools correlates with educational achievement. Or I may take a different angle, and investigate residency patterns by race, perhaps by considering whether residential segregation today correlates with the time at which court-mandated school integration began.



Baum-Snow, Nathaniel and Byron F. Lutz. “School Desegregation, School Choice and Changes in Residential Location Patterns by Race.” American Economic Review, 101(7): 3019-46. 2011.

Collins, William and Katherine Shester, 2013, “Slum clearance and urban renewal in the United States

by Gini Li TP_LiGini

Collins, William and Katherine Shester, 2013, “Slum clearance and urban renewal in the United States: American Economic Journal: Applied Economics 5(1): 239-273.

Title 1 of the Housing Act of 1949 and Consequences

After World War II, American cities witnessed a substantial increase in slum growth. In 1941, the Federal Housing Administration and economists Guy Greer and Alvin Hansen published plans for urban redevelopment and slum elimination with federal aid. These objectives were seen as federal responsibilities because of the high financial and legal barriers for private organizations to organize large-scale redevelopment projects. Title I of the Housing Act of 1949 allowed urban renewal through federal subsidies for locally planned redevelopment projects, code enforcement, and rehabilitation efforts. The U.S. Urban Renewal Program faced controversy in issues such as the use of eminent domain, impact on the urban poor, destruction of neighborhoods, and destruction of historic buildings.

In the past, economists have tried to explain the effects of urban renewal programs through spatial equilibrium models. Roback’s (1982) model describes spatial equilibrium in terms of intercity dynamics and freely mobile workers, capital, and goods. In this model, local amenities are valued by both workers and firms, and will raise the equilibrium property values. Hornbeck and Keniston (2011) use a model to describe how cities rebuilt after fires experience higher values for buildings because of local externalities, plot consolidation, private investment, and new public infrastructure. This can happen because large fires, like urban renewal programs, remove the option for property owners to keep old buildings. Lastly, Schall (1976) shows through an intracity model that raising local housing quality might be unsustainable with public renewal projects.

Collins and Shester attempt to empirically analyze the effects of urban renewal programs on economic outcomes at the city-level through exploiting cross-place variation in urban renewal activity. Through their model and the context of the Rosen-Roback framework for spatial equilibrium, they found that cities with urban renewal programs had higher property values, income, and population growth in 1980.


Urban Renewal Programs and Economic Outcomes: the model and method

            This study improves previous studies on the impact of urban renewal projects on city-level economic development with the following added features:

1. Reducing bias in ordinary least square estimates through a variable that legally constrains cities’ ability to participate in the program

2. Utilizing a new data set that ranges from 1950-1980 for all cities with more than 25,000 residents.

3. Examining whether the effect of urban renewal on city-level economic outcomes worked primarily through the displacement of residents with low levels of human capital or through channels of economic growth. It uses data from the federal census for population and housing and from the U.S. Department of Housing and Urban Development’s Urban Renewal Directory for information on renewal efforts.[1]

The model seeks to explain the relationship between intensity of urban renewal projects, as represented by “grants approved” (URij), where i represents the city and j represents the census division, and economic outcomes in 1980 (Yij80). Additionally, it controls for census-division indicators and city-specific qualities through the variable dj.[2] An instrumental-variable strategy addresses endogeniety of funding and measurement error of URij, the measure of urban renewal intensity. Preprogram control variables (X¢ij50) include the quality, ownership, use and age of housing stock in 1950; demographics, size, and median educational attainment of the population in 1950; and employment characteristics, poverty level, and family income of the population in 1950. One would expect a positive b1 if cities that were similar in 1950 (before the implementation of urban renewal programs) had varying economic outcomes in 1980 that were dependent on the intensity of these programs (URij).

(1)                                           Yij80= a + b1URij + X¢ij50b2 + dj + uij80


Enabling Legislation as the Instrumental Variable

The level of deterioration in some cities is unobservable through the control variables in the basic model. If these types of cities implemented a large volume of urban renewal programs, the economic outcome in 1980 (Yij80) could be worse when compared to other cities, skewing the relationship between intensity of urban renewal programs and economic outcomes. In this scenario, the ordinary least squares coefficient is biased and would understate the impact of URij. The opposite could occur in which cities with high rates of investment in the 1950s also pursued many other urban renewal projects. In this case, they might have witnessed great economic outcomes in 1980, but the impact of federal urban renewal funding on that economic outcome is unclear. The authors attempt to address this problem through finding exogenous variation in urban renewal funding that can be attributed to differences in timing of state-enacted legislation. Because legislation that enables local governments to exercise eminent domain to acquire property for development was approved locally, urban renewal programs were not actually implemented uniformly in any given time period.

(2)                                           URij= g + t1Lij + X¢ij50t2 + lj + eij

X¢ij50in this equation is similar to the one in the first; it is a set of city-level characteristics in 1950. lj is similar to dj and acts as a set of census level controls for 1950. Lij represents each city’s “years of potential participation” in the federal urban renewal program. This is calculated by subtracting the year in which local enabling legislation was passed from 1974, the end of the program. If missing legislation actually did restrict cities from receiving federal grants, then t1would be a positive value. Results of this regression show that an additional year of eligibility for participation results in 9.71 additional dollars of grants per capita. The authors argue that this result is independent of local characteristics because adding the other control variables doesn’t significantly affect the value of t1, and it remains statistically significant.

With the assumption that local enabling legislation impacts the value of “intensity of urban renewal” as defined by the amount of federal grants received for urban renewal, the authors test the correlation between city-level economic outcome and enabling legislation in reduced form regressions. To challenge this assumption, they test the correlation between city-level outcomes (Yij80) and enabling legislation data (Lij) in rural areas. If this correlation exists, Lij represents more than just enabling legislation and therefore can’t be used as a proxy for the intensity of urban renewal projects. The results show no evidence of a relationship between urban renewal enabling legislation and outcomes in rural counties. This supports the authors’ assertion that the instrumental variable does not reflect unobserved differences across states in economic trends.



In 1980, urban renewal programs led to higher median incomes (2.4% increase for $100 per capita difference in grants) and median property values (6.9% increase for $100 per capita difference in grants) at a five percent significance level. The impact on employment rate and percentage of families in poverty is more imprecise, but the effects seem favorable. This research does not account for the impact that private investment has made on economic outcomes or the impact federal grants have on the intensity of private investment.


Robustness Tests

Other Urban Renewal Programs

To control for the impact of other urban renewal programs on the economic outcomes in 1980, the authors controlled for the number of units per capita of new public housing built under the Housing Act of 1949, application status under Johnson’s “Model Cities” program, and city level spending per capita on poverty reduction since 1966.

Quality of Local Governments

Using Moody’s city bond ratings for 1950 as a proxy for the fiscal quality and management quality of local governments, the authors created categories of cities with relatively high ratings, relatively low ratings, and no ratings. The results do not provide evidence that government quality affects economic outcomes.

Cross-state Differences in Support for Cities

The authors collected data on state aid given to city level governments in 1952 from a Bureau of Census publication, which expressed aid relative to the urban population’s size. They also controlled for differences in political conservativism through the percentage of votes for Barry Goldwater in 1964. The results are also vague and insignificant.

Shifts in the U.S. Economy

If secular change to the U.S. economy was correlated to the timing of enabling legislation, then the results for the instrumental variable is invalid. The authors created a control variable for three-digit state level industrial composition in 1950 with national level industry growth rates between 1950 and 1980. The results show that the results aren’t driven by shifts in the U.S. economy.

Without the Largest City in Each State

Large cities are more likely to be politically influential and be able to carry out federal grant programs independent of the timing of enabling legislation. Without the largest cities, the results are still similar to the baseline results and are still statistically significant. The same is true when the smallest cities were dropped.

Changing the Data Used in the Instrumental Variable

Even with two other data sets (one constructed from the earliest date of enabling legislation approval and one constructed from the latest date of enabling legislation approval from discordant legislative recordings), the results are close to the base results.


Application of Model in Multiple Time Periods

(3)                               Yijt = qt + p1tLijt + X¢ij50p2t + sjt + vijt

The equation above regresses city level outcomes (Yjit) in the years 1960, 1970, and 1980 on years of eligibility for urban renewal as of year t (Lijt). sjt represents census-division-by-year fixed effects and X¢ij50represents the standard set of control variables for 1950. The coefficients in this regression are allowed to change every year, so p1t shows the responsiveness of economic outcomes to eligibility for urban renewal at a point in time t. Results show that there is a strong relationship between urban renewal eligibility and property value in 1970, but not in 1960. This suggests that it takes time for real estate prices to adjust. However, there is a strong correlation between urban renewal eligibility and income and employment in 1960 and beyond.


Channels of Influence

After establishing that the intensity of urban renewal programs affect city-level economic outcomes, the authors try to figure out through which channel this occurred. One theory is that it drove away people of low human capital and essentially displaced them outside of the city. Another is that urban renewal creates a continuous virtuous cycle of organic economic growth. These two theories are not mutually exclusive, and the authors use the regression model (2) to see if urban renewal affected displacement proxies and growth proxies. If displacement were the main channel through which economic outcomes increased, then adding a displacement variable to the regression would decrease the value of b1. Findings suggest that this impact is small and imprecise when racial and educational attributes were used. However, growth proxies such as property prices, wages, and population is often linked with higher productivity in the Rosen-Roback model. For the authors, this suggests that increases in economic outcomes happen through the growth channel rather than the displacement channel.

In terms of policy, this is an important finding because it shows that federal funding has had a significant and positive impact on economic outcomes in American cities. This bolsters support for further steps in administering countrywide federal funding to address urban problems. However, it would be beneficial to distinguish between the most successful and least successful cities in this study, and figure out what factors increase potency of federal funding. This would be helpful in crafting future policy measures and increasing the efficiency of grant funding.


Works Cited

Hornbeck, Richard and Keniston, Daniel. Creative Destruction From the Great Boston Fire of 1872: Barriers to Urban Growth Illuminated. (draft), 2012.


Roback, Jennifer. “Wages, rents, and the quality of life.” The Journal of Political Economy (1982): 1257-1278.



Schall, Lawrence D. “Urban renewal policy and economic efficiency.” The American Economic Review 66.4 (1976): 612-628.



[1] This includes federal grants approved and disbursed for urban renewal projects and programs up until 1974.

[2] This encompasses 1950 value of Yij, city-level characteristics at the time of the federal program’s implementation, and indicator variables for nine census divisions.

Regression analysis of rail station proximity’s effect on property values

by Bao Tran-Phu

I.          Research question


In “Identifying the Impacts of Rail Transit Stations on Residential Property Values,” Bowes and Ihlanfeldt (2001) study the relationship between transit station proximity and local property values.  Specifically, they use least-square regressions to study the property value effects of having a Metropolitan Atlanta Rapid Transit Authority (MARTA) rail station nearby.  To understand the factors driving this impact, they disaggregate it into four distinct effects: a direct positive effect (from heightened access to transit), a direct negative effect (from physical aspects of the station itself, such as noise and pollution), an indirect positive effect (as stations attract retail development), and an indirect negative effect (as stations serve as a gateway for crime).  Previous studies have investigated this transit-property value impact, but by not quantifying the effects of these individual drivers, they have failed to build a comprehensive picture of what exactly causes the impact.


Given a Tiebout amenities-based approach to evaluating the economic benefits of public transit, the benefits (as well as costs) of having a transit station nearby would be fully capitalized in local property values.  Thus, by analyzing the relationship between transit station proximity and property values, in theory they are conducting a comprehensive accounting of all of the rail’s social benefits.  More importantly, however, they employ regression analysis to better understand the drivers behind this transit-property values link.  This study is important for transit policy decisions for two reasons: first, for proposed transit networks, it can help transit authorities choose optimal locations for their stations and to understand what their effects will be on the surrounding neighborhoods.  Second, for existing transit networks, it can help local governments tailor policy in order to promote the benefits and mitigate the costs caused by the transit stations.

II.       Empirical analysis


This paper employs a series of multivariate least-squares linear regressions.  To briefly illustrate the mechanisms behind ordinary least-squares (OLS) regression, assume a response variable Y, where Y is determined by the following relationship:

Y= α0 + α1X1 + α2X2 + α12X1* X2

X1 and X2 are explanatory variables that are hypothesized to collectively explain Y.  X1*X2 is an “interaction” of the two explanatory variables, capturing any synergistic effects above and beyond the sum of the two effects individually.  Finally, α1, α2, and α12 are coefficients that quantify the linear effects of their respective explanatory variables, while a0 gives the average value for Y when all explanatory values equal zero.  Explanatory variables can either be continuous, discrete (for ex, binary), or mixed.  If explanatory variables are continuous, then their coefficients give the slope of the variable’s effect, holding fixed all other variables.  If they are binary “dummy variables,” then the coefficient quantifies the difference-in-means effect of X=1 (relative to X=0), holding fixed all other variables.  From there, we set up a regression, given i observations:


In the estimated regression, ui is the residual, signifying the variation in Ythat is not explained by the regression model.  Statistical software can then find the coefficient values that minimize the sum of square residuals.   In their paper, Bowes and Ihlandfeldt use three regression models that are estimated separately.


Hedonic price model (OLS regression)


The first model is used is to quantify station distance’s effect on property values, and to disaggregate that effect into the four main effects listed above.

P = α0 + αZZ + βLL + βCC + βRR + γ SS

P is house sale prices, C is neighborhood crime density, R is neighborhood retail employment, S is rail station proximity, and Z and L are sets of other house and neighborhood characteristics respectively that are independent of rail station proximity (for example, the number of rooms in the house).  They opt for a semilog regression, (whereby the response and explanatory variables relate exponentially) over a linear regression, and they find that doing so achieves higher statistical significance.  Also, station proximity (S) is captured by a set of dummy variables for being less than one-quarter mile, between one-quarter and one-half mile, between one-half and one mile, between one and three miles, and greater than three miles away.  They opt for this technique over using a continuous proximity variable based on the intuition that the effects of station proximity will differ depending on how far away the neighborhood is from it to begin with.


Crime and retail employment auxiliary models (random effects regression)


The two auxiliary models are then used to provide a closer look at how exactly crime and retail development indirectly affect property values.

C = ψ0 + θNN + ψSS

R = φ0 + πQQ + φSS

As before, C is crime density, R is retail employment, and S is a set of station proximity dummy variables.  N and Q are respective sets of crime- and retail employment-correlated neighborhood factors that are independent of station proximity.  In both cases, since they use panel data on 206 census tracts from 1991-1994, random effects models are used instead of the standard OLS (“fixed effects”) regression model.  This is a variation of OLS, whereby the census tracts are treated as randomly-selected samples from a large pool, as doing so avoids having to employ 206 dummy variables to control for the tract-level effects.


Each model is analyzed both with and without interactions.  Including the interactions allows the authors to study how the effects of station proximity vary depending on station characteristics (for example, whether or not the station has a parking lot), neighborhood income, and distance from the CBD.

III.          Results and implications


For the hedonic price model regression, R2 = 0.475, where R2 measures the percentage of the variation in the response variable explained by the regression.  Effects of both crime density and retail employment on property values were statistically significant, providing validation for the two auxiliary models.  Excluding interactions, being less than one-quarter of a mile away from a station causes property values to be 20% lower than equivalent houses more than 3 miles away, but houses between 1-3 miles away are worth the most (holding all other variables constant).  When the interactions are included, Bowes and Ihlanfeldt additionally find that the premium paid for being between 1-3 miles from a station is greater for high-income neighborhoods, and being less than a mile from a station is worth more in neighborhoods further away from the CBD.


For the crime auxiliary model, R2 = 0.824.  Without interactions, it is found that being less than half a mile from a rail station led to higher crime density.  Including the interactions then shows that rail stations that have parking lots increase crime for neighborhoods less than one-quarter mile away, but decrease crime for neighborhoods between one-quarter and one mile away.  This may indicate that station parking lots serve as a magnet for crime, attracting it in the station’s immediate vicinity but pulling it away from areas a bit further away.  Also, the authors find an increase in crime as neighborhood median income rises, but only for tracts less than one-quarter mile from the station.  Finally, the station proximity effect on crime is smaller for neighborhoods further away from the CBD.  This could suggest that in the suburbs, criminals may see themselves as more easily identifiable as an outsider, and the higher perceived probability of being caught could thus deter them.


Finally, for the retail auxiliary model, R2 = 0.670.  Excluding interactions, station proximity has no significant effect on retail employment.  When the interactions are added, however, it is shown that further away from the CBD (greater than 10 miles away, for example), having a station nearby has a significant positive effect on retail employment.


Overall, these three regression analyses demonstrate that station proximity affects property values both indirectly (by affecting crime and retail development) and directly (possibly by improving transit accessibility and/or by imposing negative physical externalities) (see Figure 1). Direct effects are generally found to be larger than the indirect crime or retail effects.  Further, Retail effects are larger in magnitude than crime effects, except for downtown neighborhoods closest to the stations.  Most importantly, though, the study demonstrates that the station proximity effect differs greatly depending on neighborhood income, distance from the CBD, and distance from the station.  The biggest benefactors of rail stations are neighborhoods that are far from the CBD and very close (one-quarter to one-half mile) from the station, in which case the positive retail effect dominates.  The biggest losers are higher-income downtown neighborhoods closest (less than one-quarter mile) to the station, in which case the negative crime and direct externality effects dominate.


Figure 1. The three models demonstrate station proximity’s direct and indirect effects on property values.


In planning public transit, local governments may assume that stations will have a dominantly positive proximity effect on surrounding property values, due to the heightened transit accessibility that the stations provide.  As these results show, however, the reality is more complex.  To accurately predict the effect of a nearby transit station, the underlying drivers as well as the neighborhood’s characteristics must be considered.  Additionally, local governments should seek to maximize the benefits and minimize the costs of existing transit stations.  In low-income neighborhoods, governments should focus on mitigating the stations’ negative externalities, while the focus should be on reducing transit-driven crime in high-income neighborhoods.

IV.          Extensions


In this paper, Bowes and Ihlanfedlt are meticulous to separate the competing indirect effects of rail stations driven by crime-related and retail employment-related factors.  Curiously, a similar effort is not made to disaggregate the direct effects of station proximity, even while two major competing effects (transit accessibility and negative physical externalities) are identified.


In addition, property values are used here as a proxy for the local demand for transit, as theorized by the amenities view of public goods.  However, given that real estate buying decisions are likely temporally and conceptually displaced from household transportation decisions, property values may be only an imprecise proxy.  A possible extension could thus be a similar study, replacing property value with household transportation costs as the response variable.  This study would investigate the effects of nearby transit stations on actualized transportation costs for surrounding households, which would be rooted in actual transit use.  These transportation cost savings could then be aggregated with the other direct and indirect effects studied in this model to create single hedonic model, providing a more complete illustration of the economic impact of public transit.

The full document can be found here: TP_Bao

Works Cited

Bowes, David R., and Keith R. Ihlanfeldt. “Identifying the Impacts of Rail Transit Stations on Residential Property Values.” Journal of Urban Economics 50 (2001): 1-25.

Modeling Downtown Parking and Traffic Congestion

by Chris Bowman  TP_Bowman

Modeling Downtown Parking and Traffic Congestion

A Model by: Anderson, Simon P., and Andre De Palma. “The economics of pricing parking.” Journal of Urban Economics 55.1 (2004): 1-20.



BackgroundWe have all experienced the frustration of trying to find a parking spot in a crowded city.  When on-street parking is free or the same price throughout the city, we try to park closest to our destination.  However, the congestion resulting from everybody trying to park closest to the CBD creates a parking pattern that is less than socially optimal.


In this model, we are dealing only with individuals that do not have assigned parking spaces in the city, such as shoppers.  We will initially make the following assumptions:

I.         All individuals are traveling to a common location at x=0 (the CBD).

II.         The CBD is located at the end of a long, narrow city, and is served by parallel access roads.  Perpendicular to the access roads are side streets that are used for on-street parking.  Cars can park on street at any free location.

III.         There are N individuals located far away.

IV.         Each individual first drives at speed vd into downtown, then begins looking for an empty parking spot on a side street.

V.         Once an available parking spot is found, the individual walks at speed vw to the CBD.

VI.         The more people trying to park, the longer it takes to find a spot.


Once in the city, the individual will stop at some distance, x, from the CBD and search for an open spot while incurring a cost γ per lot inspected.  He will then walk to the CBD.


The total number of parking spots on the interval [x,x + dx] is represented by K(x)  dx , with K(x) = k .  Therefore, the city has width k, with the CBD located at the end. The number of occupied parking spots over the interval [x,x + dx]  is denoted by n(x)  dx  with n(x) £ k.  The probability that a randomly tested spot will be free is given by       q(x) = [k −n(x)]/k .


Basic Model

The expected cost of an individual searching for an open spot at location x is


Adding the cost of walking from location, x, to the CBD,


Where t is the net dollar cost of walking as opposed to driving.

In an equilibrium where parking is unpriced, all parking locations have the same expected cost, c. Rearranging (2):


IfTB_Xis the furthest distance parked, the car parked at this location has the smallest search cost, γ, so that:


By equating the supply and demand for parking, which requires


We can solve for the equilibrium expected cost, c, in implicit form:


Introduction of a Social Planner

If we introduce a social planner who wants to provide the socially optimal parking pattern, she will want to minimize the social cost of getting the individuals to the CBD.


The solution {n(x), xo} to the optimal control problem above involves equating marginal social cost (with respect to n(x) ) for all locations where at least one car is parked.  By differentiating the integrand above, we solve for the marginal social cost λ


Therefore, the optimal number of people parking at x is presented as


The population constraint can then be shown as


And after integrating the left hand side,


Rewriting this last equation using the value of λ in (8), k/t (√λ −√γ )2 = N .


The optimized marginal social cost is therefore λ = (√γ +√Nt/k )2 . This tells us the optimal value of the location of the last parking place, which we can compare to the location of the last parking spot under the model without a social planner. Substituting this value of λ  in Eq. (8) leads to



Comparing the findings from the unpriced parking scenario in Eq. (4) and (6) with the optimal results achieved by the social planner (10), we can see that under the optimal scenario, the parking span is larger.  This means that when there is unpriced parking, parking becomes more tightly spaced than is socially optimal.  This results from the fact that free parking is a common property resource, and that people do not take into account that, by deciding to search for a spot closest to the CBD, they are increasing the search costs of others.  While this may seem intuitive, it sets up a basis for studying urban parking fee structures.  By raising the cost of on-street parking closest to the CBD, a city may be able to reduce some of the congestion associated with parking.  This is added to the model by stating that the optimal parking fee, τ(x), occurs when the parking price is equal to the difference between marginal social and private cost, given as


By inserting the optimal parking density no(x) expression (9) into the above equation, we reach the optimal parking fee


This equation for the optimal parking fee demonstrates that the fee should decrease as distance from the CBD increases.



Because free or wrongly priced unassigned urban parking spots lead to increased congestion and tighter parking closer to the CBD, there are a number of possible investigations or solutions to the problem that ought to be considered.  This model does not take into account the time limits placed on many metered parking spots, or account for the length of stay once parked.  This factor may be important in determining optimum pricing strategies across the urban landscape.  Also, the model should consider an individual’s option of choosing not to search for a spot, but instead to pay a premium to park immediately in a garage.  Shoup (2006)* explores the relationship between the point of indifference between cruising for a spot, and paying to park in a garage.


Because there are so many factors involved in individuals’ parking decisions, it is difficult to build an entirely comprehensive model, but this model provides a base to begin understanding parking price theory.


With the dawn of smart phones, it may soon be possible to virtually assign spots to individuals driving into urban areas, and a bidding scheme could be established to set the price.  It would be a difficult system to enforce, but an interesting thing to consider.            *

* Shoup, Donald C. “Cruising for parking.” Transport Policy 13.6 (2006): 479-486.

Shoup, Donald C. “Cruising for parking.” Transport Policy 13.6 (2006): 479-486.

Job Decentralization and Residential Location: A Concise Technical Presentation of the Work by Leah Platt Boustan and Robert A. Margo

by Chris Whittaker     TP_ChrisWhittaker

Which came first, the chicken or the egg? Urban economists often ask a similar question in relation to the origins of cities: which came first, the jobs or
the workers? Leah P. Boustan and Robert A. Margo set out to provide some insight into this question. Specifically, they question how the spatial
distribution of employment opportunities influences urban Americans’ choice of residential location. The question and research is largely prompted by a
trend of suburbanization in the United States over the past 50 years. Since 1960, the share of metropolitan Americans living in suburbs away from the
central city has risen noticeably. Over the same period of time, the number of metropolitan residents working away from the city has similarly risen. Thus,
urban economists question whether it was workers following jobs or jobs following workers that caused this decentralization of both employment and
population. The pair’s results suggest that the decentralization of employment appears to have been an important (though by no means complete) cause of
residential location and suburbanization. From these results, they conclude that in order to attract residents to central cities, creating new jobs in the
Central Business District (CBD) is more effective at the margin than simply creating jobs in the larger metropolitan area. Their work ultimately adds to
the tools that urban economists have to analyze urban growth, building on such work as the monocentric city model.

Presentation of the model

Boustan and Margo seek to “estimate the impact of employment location on residential location” (3). Though an ideal experiment would randomly place the
location of a given industry randomly throughout a city and look at where its workers choose to live, rarely are opportunities like this available on such
a large or measurable scale. In reality, the variation between locating work and workers may be endogenous as workers seek to reduce commuting costs and
businesses seek to provide compensation in the form of a decreased commute time. Since causality would be difficult to demonstrate, Boustan and Margo use a
natural experiment to evaluate the impact of employment opportunities on residential location. The two choose to look at employment in state government in
a capital city as a reasonable analogue.

According to Boustan and Margo, employment in state government in a capital city has two important advantages that make it a reasonable choice. First, the
location of state government long predates any form of suburbanization; originally most state capital cities formed prior to the 1900’s and workers located
adjacent to the CBD. Second, the majority of buildings in the capital that engage in state government still remain in the CBD (and likely will continue to
do so). Therefore, state capital cities and their government buildings vary spatially based on historical reasons that are unrelated to current residential
development patterns. Therefore, the model focuses solely on worker decisions and avoids “concerns about reverse causality” (3).

However, due to the particular development of the cities, there are possible differences between state workers in capital and non-capital cities (for
instance, level of education). To address this concern, Boustan and Margo compare state workers to both private and public-sector workers in their cities.
In their words, “the key identifying assumption is that state workers in both city types [capital and non-capital] are otherwise identical but for their
‘assignment’ to the state capital” (5). However, as workers are engaged in different types of employment (see table 1, pg. 6), Boustan and Margo begin with
a model to test whether observable characteristics of the state workforce vary between capital and non-capital workers. They estimate:

Xijk= α + β1(state worker × capital city)ijk+ α2(state worker)ijk+ β 3(capital city)ijk+ εijk , (1)

where that X is a vector of individual characteristics, including race, gender, age, and educational attainment and i, j
, and k are used to index individuals, metropolitan areas and class of worker (i.e. state worker or not), respectively. The β1
coefficient is of particular important, as it reflects the interaction between being a state worker and living in a capital city. Boustan and Margo use a
sample that includes more than 700,000 full-time workers in 127 metropolitan areas, 25 of which are state capitals (5).

After running a regression analysis given this first equation (eq. 1), they find that “the main effects of being a state worker and living in the capital
city are large and significant” (6). However, given several dimensions of personal characteristics, the only exception in which state workers in capital
cities differ from state workers in other metropolitan areas is race. That is, state workers in capital cities are less likely to be black. They resolve
this, stating that because African Americans are more likely to live in central cities, “if anything, the racial difference will work against our main
findings” (6). Thus, the similarity of state workers in both capital and non-capital cities along observable dimensions (such as age, gender, education,
race, marital status, etc.) supports their use of a difference-in-differences estimator, or the interaction of state work and capital city in the above
vector equation (eq. 1).

Boustan and Margo then demonstrate that state workers in capital cities are more likely to work downtown. This corresponds to the aforementioned benefit of
state capital formation, where historical government buildings have long existed prior to current suburbanization trends. The regression that underlies
this conclusion is the same as equation 1, but the dependent variable is replaced with a binary indicator variable for either working or living downtown
(8). Now, “if workers based their residential location decisions on their job location we would expect that state workers in capital cities are also more
likely to live downtown” (8). They proceed to show that this is true, that state workers in capital cities are 10 percent more likely to live downtown.

The overall question is motivated by the relationship between employment and population locations. Boustan and Margo use the following estimating equation
to look at this relationship:

1 if live in cityijk = α + β(=1 if work in city)ijk + θX ijk + j + k + εijk. (2)

Again, the difference-in-differences estimator can be used as a viable instrumental variable for working in the central city (9). The two present a fourth
table, utilizing equation 2, that demonstrates the relationship between employment and residential location. I have included this particular table below
because of its significance in the overall conclusion.

They estimate the equation using both ordinary least squares (OLS) and the difference-in-differences estimator as an instrumental variable (IV) for working
in the city center. They ultimately rely on the differences-in-differences estimator to reduce selection bias. They find that the relationship between
working and living in the central city does not vary much between the OLS and instrumental variable treatment, but that “the effect of working in the CBD
on residential location triples in size” (9). This is significant, because working in the CBD has a larger effect on housing location than simply working
in the metropolitan area.

The results from this table, using the instrumental variable, can be extrapolated to predict the overall effect on adding jobs to both the city and to the
CBD. Since they note that 15 percent of jobs are typically located in the CBD, the following equation can be used to calculate the effect of adding 1,000
jobs to the city.

(% jobs in CBD)(residents to CBD) + (% jobs in rest of city)(X) = Y residents in city (3)

Applying the fact that 15 percent of jobs are typically located in the CBD to the coefficients under column 2 (IV) from table 4, we can calculate the
impact of adding 1,000 jobs to the city.

(0.15 jobs in CBD)(356 residents to CBD) + (0.85 jobs in rest of city)(X) = 246 residents in city (4)

We find that X is equal to 227, providing a cumulative effect of adding jobs to the city of 246 new residents. This shows us that adding jobs to the CBD
has a 57 percent greater effect on having residents locate in the central city than simply adding jobs to the city as a whole (=(356-227)/227).

Boustan and Margo conclude by applying their results to the time variation in urban decentralization in the US between 1960 and 2000. As the percent of
workers who worked in the city fell from 59.3 to 42.3 percent, using the instrumental variable estimate we can infer that “the share of workers who lived
in the city would corresponding fall by 4.2 points (or 17% × 0.246; second column of table)” (10). Since the central city metropolitan population fell
by 18.8 points over this same time period, we can infer that employment decentralization accounts for around 22 percent “of the observed suburbanization of
population from 1960 to 2000” (10).

Implications and extensions

The model has several relevant implications for urban economics, policy development and methodological research. From the urban economic policy side, there
are two primary implications. The first is a convenient “Rule of Thumb” for policymakers that want to understand the effects of urban population change
with respect to the effects of proposed policies (18). In general, the model allows a policymaker to assume that one urban resident will be added or
subtracted to the central city for every four jobs created or destroyed, respectively. This is a useful tool for back-of-the-envelope calculations,
allowing policymakers to quickly analyze the impact of potential policy decisions. Second, the findings provide some pushback to the amenities-based theory
for urban growth. While the creation of endogenous amenities (such as museums or art galleries) may lead to a revitalization of downtown urban areas,
Boustan and Margo claim that unless job creation remains the central function of city centers, any amenities-based (or consumption-based) downtown revivals
will fail to reach their full potential (18). Concerning methodological implications and extensions, Boustan and Margo believe that the historical record
will continue to reveal opportunities for natural experiments in urban economics that have often been overlooked due to a prevailing notion that “locations
are rarely determined by exogenous forces” (18). Ultimately, this study demonstrates the usefulness of understanding the relationship between job
decentralization and metropolitan suburbanization trends. Further research could utilize this model with respect to global suburbanization trends,
particularly in Europe and other developed nations.

Original Paper Citation

Boustan, Leah Platt, and Robert A. Margo. “Job decentralization and residential location.” Brookings- Wharton Papers on Urban Affairs Annual 2009:
1+. Academic OneFile. Web. 26 Feb. 2013.