Home » 2013 Categories » 2013 – Durham papers

Category Archives: 2013 – Durham papers

The Impact of Neighborhood Organizations on Housing Price

 by Stuart Price DP_Price


In Durham, NC, Neighborhood Organizations can formally register themselves with the Durham City- County Planning Department through an application process. Although this action provides no formal benefits to the neighborhoods, it gives them an intrinsic sense of stature for engaging public officials and addressing local issues affecting their community. In this sense, according to Cunningham and Kotler (1983,8), “neighborhood organizations are able to wage responsible wars of pressure and advocacy to ensure that a just share of available resources goes to the neighborhood.” Moreover, neighborhood organizations enable an improved sense of institutional infrastructure and stability, which Briggs and Mueller (1997) support in their finding that neighborhood organizations can cause positive effects on an area’s social environment.

Given these reported benefits of neighborhood organizations, this paper explores the impact that neighborhood organizations have on housing prices. In their study of households’ valuation of neighborhood amenities in the Baltimore metropolitan area, Dubin and Goodman (1982), found that the presence of neighborhood organizations (one of the thirty-three variables in their hedonic pricing model) positively impacted housing prices. Using a simplified hedonic pricing model, this paper attempts to determine whether the same positive impact holds for implicit valuations of homes in Northgate Park and Colony Park neighborhoods when compared to similar housing. Comparable housing was chosen from adjacent areas that were not incorporated under an official Durham Neighborhood Organization.


Based on the study by Dubin and Goodman and the positive externalities associated with neighborhood organizations as shown by Cunningham and Kotler and Briggs and Mueller, it is expected that the presence of a neighborhood organization should positively impact housing prices. Thus when comparing a property that is a part of a neighborhood organization to a similar property which is unaffiliated, a dummy variable constructed for the neighborhood organization variable should be significant and positively correlated to housing price (assuming 1 is assigned to neighborhood organization and 0 is assigned to no neighborhood organization).

Methodology and Model Description

This paper analyzes the effect of both the Northgate Park Neighborhood Association and the Colony Park Association on their respective community’s housing prices. Hedonic pricing models provide an implicit evaluation of households “through the analysis of a house as a bundle of structural and neighborhood characteristics….This hedonic price is interpreted as the added value in the market of a unit of the characteristic (Dubin & Goodman, 167).”

Within this paper, there are two different hedonic models that are used to predict housing value. For the Colony Park Association, the following regression model was used:

Ln (Zillow Price Estimate) = α0 + β1#Bedrooms + β2 #Bathrooms + β3 Square Footage + β4 Lot Size+ β5 Year Built + β6 Left Neighbor Price Estimate + β7 Right Neighbor Price Estimate + β8 Neighbor + ε

And for Northgate Park Neighborhood Association the following regression was used:

Ln (Zillow Price Estimate) = α0 + β1#Bedrooms + β2 #Bathrooms + β3 Square Footage + β4 Lot Size+ β5 Year Built + β6 Left Neighbor Price Estimate + β7 Right Neighbor Price Estimate + β8 Neighbor + β9 Elementary School + β10 Middle School + β11 High School + ε

Both separate regression models relied on sixty observations. These observations were randomly selected and included thirty properties within the Northgate Park and Colony Park neighborhoods as well as thirty observations of comparable, adjacent properties. Maps of the neighborhoods and the areas of corresponding comparable housing (outlined in red) are provided below:





Northgate Park


Colony Park


All data present in these two regression analyses was gathered from Zillow.com. The dependent variable in both regressions reflects Zillow.com’s most recent Zestimate, which is computed using a proprietary formula and represents the estimated market value of different properties. The variable Neighbor is a dummy variable constructed to test whether the Neighborhood Organization designation impacts housing prices. Lastly, the regression model for the Northgate Park Neighborhood Association includes the three extra dummy variables, Elementary School, Middle School, and High School, which are used to determine the impact the impact that public school assignment has on housing price. In the case of the Colony Park analysis, these variables are unnecessary as all sixty properties are assigned to the same elementary, middle, and high schools.

The regression models themselves are nonlinear. This paper utilizes the semilog functional form because, “it has become perhaps the most widely used functional form in hedonic studies (Coulson, 24).” In such a model, each variable’s coefficient is known as a semi-elasticity and “roughly speaking, the coefficients give the percentage increase in price due to a unit increase in X (Coulson, 25).”

Data Analysis

Multivariate regression analysis on the property observations produced the following Ordinary Least Squares (OLS) estimates:

                      Colony Park                                          Northgate Park


For the Colony Park Association and surrounding comparable housing the hedonic price function estimate is as follows:

Ln (Zillow Price Estimate) = -22.78 + .065*#Bedrooms + -.114*#Bathrooms+ .00027*Square Footage + 3.6503e-6*Lot Size+ .0177*Year Built + -1.319e-6*Left Neighbor +-3.878e-7 Right Neighbor+0.0482 Neighborhood

And for Northgate Park Neighborhood Association the hedonic price function estimate is as follows:

Ln (Zillow Price Estimate) = 1.869 + .0039*#Bedrooms + 0.09*#Bathrooms+ .00029*Square Footage + 1.561e-6*Lot Size+ .0046*Year Built + -1.125e-6*Left Neighbor + 1.99e-7 Right Neighbor+0.127 Neighborhood +0.05Elementary School + 0.0016Middle School + -0.0197High School

                     Colony Park                                          Northgate Park


The Colony Park hedonic model has a low R2 coefficient of determination. Only 45.80 percent of the total variation in Ln (Zillow Price Estimates) can be explained by the regression equation. Assuming a significance level of .05, only the covariates Square Footage and Year Built are significant. The estimate for the Neighborhood covariate indicates that inclusion in the Colony Park Association leads to a 4.82 percent increase in the value of a property in the area. The t Ratio for the coefficient remains low, however, at value of .57, indicating a high level of variability in using membership in the Colony Park Association as an indication of increase property value.

On the other hand, the Northgate Park hedonic model has a higher R2 coefficient of determination. 79.60 percent of the total variation in Ln (Zillow Price Estimates) can be explained by the regression equation. Assuming a significance level of .05, the coefficients Square Footage, Year Built, and Neighborhood are significant. The estimate for the Neighborhood coefficient indicates that property located in the in the Northgate Park Neighborhood Association as leads to a 12.7 percent increase in the value of a property in the area. Moreover, our model predicts a t Ratio of 2.46 associated with the Neighborhood coefficient. This indicates that membership in the Northgate Park Neighborhood Association is a significant indicator of increased property value.

Such a conclusion in the case Northgate Park, however, comes with serious reservations. This paper’s hedonic models for both Neighborhood Organizations are not comprehensive as indicated by low adjusted R2 values. Given the scope of this paper, it relied on Zillow.com as its primary source for information. There are several problems with this approach. First, Zillow’s Zestimate is a mere proprietary tool to estimate the true value of a house. The amount of information known about a specific home affects the accuracy of the Zestimate. Thus missing or incorrect information can skew the pricing data used in this paper. Moreover, because of this incomplete information problem, this paper’s hedonic models could be missing attributes that affect its price. For example, potential variables s such as environmental amenities, age of appliances, and street-level crime data could influence the value of properties. For the purpose of this paper, since the comparable housing was adjacent to each Neighborhood Organizations, it was assumed crime rates were constant for the incorporated and unincorporated properties. This assumption, however, was not necessarily correct but simply necessary because street-level crime data was not accessible.

In addition to missing covariates, this regression analysis also suffered from the phenomenon of multicollinearity. One assumption underlying the OLS model states that random components are not related across observation. A principal component analysis (shown for the Northgate Park regression), however, indicates a correlation between different predictor variables.


This evidence helps to explain some seemingly inaccurate estimates of different coefficients. For example, an estimation that membership in the Northgate Park Neighborhood Association would lead to a 12.7 percent increase in housing price seems very high. Rather, this coefficient estimate could be a linear combination of several different independent variables.



Despite limitations to this paper’s hedonic pricing model, there is evidence that membership in a Neighborhood Organization has some positive effect on housing prices. In regression model for both the Colony Park and Northgate Park neighborhoods, membership positively affected housing prices. This positive affect was significant in the case of Northgate Park.  In order to determine a more accurate estimate of these effects, however, a more comprehensive hedonic pricing model is necessary in order deal with challenges such as omission error and multicollinearity.






Urban planners and economists often debate the merits of cul-de-sacs, or circular, dead end streets that serve adjacent dwellings. Proponents claim that cul-de-sacs reduce urban congestion, improve community relations and reduce crime. However, economists occasionally challenge such claims, asserting that cul-de-sacs do not provide additional safety benefits or that such benefits are negligible. This research explores the relationship between crime and the existence of cul-de-sacs in the city of Durham, North Carolina. By utilizing Geographical Information Systems (GIS) and publically available crime data, I compare the spatial differences in crime rates between communities built on cul-de-sacs and two-way streets. Though crime in cul-de-sacs appears to be markedly lower than crime on nearby two-way streets, further economic analysis is necessary to separate the spatial effects of cul-de-sacs from other socioeconomic factors.



The cul-de-sac is a hallmark of suburban sprawl. The term itself comes from a French expression that means “the bottom of the bag” (Lonngren). It is a dead end street with only one entrance for vehicle traffic. Further, cul-de-sacs tend to differ from typical dead end streets such that they end in a circular turn-around space that permits vehicles to exit by making a wide U-turn. An aerial view of a typical cul-de-sac is provided in the appendix (figure 1).

Cul-de-sacs have a number of important benefits. For suburban developers, cul-de-sacs allow them to place more homes in oddly shaped tracts of land (Nielsen and Lonngren).  For residents, cul-de-sacs provide privacy and limit the noise from traffic while still remaining a part of the larger suburban community. Their one entrance and exit naturally help to constrain the speed of traffic, as well as the frequency of unknown vehicles passing through. Because of lower traffic, they encourage walking, bicycle use and outdoor activity by children. Unsurprisingly, homebuyers perceive houses surrounding cul-de-sacs to be safer than those located on two-way streets, and are willing to pay a premium for them (Neilsen).

However, others criticize cul-de-sacs. By definition, these communities are not well connected to other streets and they are often far from the central business district (CBD) and other areas of economic activity and community participation (Jagannath). On one hand, they encourage automobile use, as public transportation services are unable to accommodate their select residents. This, in turn, produces urban congestion in other parts of the community as well as a host of environmental problems associated with increased vehicle use. On the other hand, they create a number of inefficiencies with respect to the provision of public goods and services. It is more difficult to sweep streets or plow snow in cul-de-sacs. Further, it is more difficult to patrol cul-de-sacs and emergency vehicle access can sometimes be limited (Lonngren).

Much of the common knowledge regarding the popular appeal of cul-de-sacs is based on the idea of safety. As the logic goes, criminals try to avoid areas that lack easy entrance and exit. Further, if a crime is committed in a cul-de-sac, criminals will have a more difficult time escaping. Since residents of cul-de-sacs tend to be more familiar with each other, they are more likely to report and deter suspicious activity on behalf of their neighbors. Further, since fewer people frequently pass through cul-de-sacs, potential criminals would be unaware of opportunities in those communities. However, as noted by Lonngren, it is more difficult to police cul-de-sacs; these homes may actually provide better targets for criminals. Neilsen further notes that cul-de-sac statistics reveal some of the highest rates of traffic accidents involving young children. According to William Lucy, a professor of environmental studies at the University of Virginia, “the actual research about injuries and deaths to small children under five is that the main cause of death is being backed over, not being driven over forward” (Neilsen).

Little academic research exists regarding criminal activity with relation to the spatial layout of cul-de-sacs, and the majority that does remains inconclusive. As Hillier and others admit, it is difficult to untangle additional socioeconomic variables from the spatial layout of such communities (Hillier). However, it remains a useful exercise to conduct natural experiments that try to minimize such socioeconomic concerns in order to better understand our communities. Though it may not be possible to understand the true magnitude of the additional safety benefit provided by cul-de-sacs, it is certainly feasible to look at direction the answers is pointing. As such, it is possible to evaluate the following question: is living on a cul-de-sac in Durham safer than on a two-way street? To better answer this question, it is worth taking a brief look at recent criminal activity in Durham.



According to FBI statistics, the year 2012 marked an all-time low in violent crime (murder, rape, robbery and aggravated assault) and property crime (burglary, larceny and motor vehicle theft) committed in Durham (Durham Police Department). Crime continued its downward trend from the year 2000; this is shown in the appendix, figure 2. In particular, property crime reached its lowest level since 1988, due to a decrease in larceny and burglaries. Perhaps most striking, property crime is down 44% since the year 2000, with burglaries falling by 15% and larceny by 7% since 2011 (Durham Police Department). Still, larceny and burglary are the two largest contributors to Durham crime, constituting 80% of all criminal activity. These statistics are demonstrated graphically in the appendix, figure 3.

Though criminal activity is at a 23-year low, there were still over 15,000 crimes committed in Durham last year, approximately 80% of which are related to property. Consequently, analyzing the differences in property crime between housing communities on cul-de-sacs and two-way streets may prove useful. Thus far, no specific research exists regarding Durham spatial differences and crime rates. I attempt a first pass at demystifying and exploring these differences below.



I begin by identifying communities in Durham that are built around cul-de-sacs using satellite imaging tools available through Google Maps. I selected 36 suitable cul-de-sacs that are relatively similar with respect to spatial layout. I then pair each of these cul-de-sacs with a nearby two-way street comprised of similarly priced houses. While pairings differ in terms of relative pricing, each pair itself reflects a comparatively equivalent level of housing quality. I then compare each pairing with approximately 15,000 pieces of data regarding criminal activity in the city from the year 2012; such data is publically available from the Durham Police Department and their Geographical Information Systems (GIS) software. I then determine the total number of crimes committed in cul-de-sacs and on adjacent two-way streets. Further, I narrow criminal activity to single-family dwellings to remove possible biases from high traffic commercial properties and high-density apartment complexes. I conclude by separating crimes into three categories: crimes related to larceny, crimes related to assault, and crimes related to burglary/breaking and entering. This provides a more detailed assessment of the common types of crime that occur in cul-de-sacs and on two-way streets.

While the approach is simplistic, it has several advantages. First, pairing cul-de-sacs with nearby (often adjacent) two-way streets removes certain spatial biases: geographically close streets create a sensible natural experiment such that we can assume many spatial variables are held constant. Second, it mitigates socioeconomic biases by evaluating housing communities of similar economic qualities. Third, the approach applies uncomplicated economic principles to thinking about the relationship between criminal activity in cul-de-sacs and two-way streets. It ultimately provides a foundation for more detailed future research.



The data is summarized in the table below (table 1):



Table 1: Summary statistics of crime rates between cul-de-sacs and two-way streets

In total, there were 89 crimes committed within the selected communities, with 74 occurring on properties located on two-way streets and 15 within communities located on cul-de-sacs. This yields a 1-to-4.93 cul-de-sac to two-way street crime ratio. Stated differently, for every crime committed in a community based around a cul-de-sac, there are nearly 5 committed in a related community along a two-way street.

Compared to total Durham crime, this sample reflects less than 1% (approximately .60%) of total crime committed in Durham. However, this appears reasonable. Given a number of generous assumptions such that there are 20-30 homes per selected pairing, and approximately 107,000 total households in Durham County according to the U.S. Census Bureau, assuming that crime is evenly distributed among only households we would expect this to approach a 1% crime rate (30*36/107,000 = 1.01%). Knowing that crime is not evenly distributed and that much criminal activity occurs in highly frequented public spaces, dense residential communities and commercial areas, we can generally infer that this sample appears appropriate to provide a picture of residential criminal activity.

It is important to interpret these results in the context of total Durham crime in 2012 (see appendix, figure 3). While larceny was the largest category of criminal activity, constituting 53% of total Durham crime in 2012, it only comprises 25.83% of crime in the 36 selected communities. There are several notable considerations that may explain the difference. According to the Durham Police Department Annual Report, shoplifting constituted 25% of all larcenies (4). Further, the single largest percentage of larcenies at 40% consisted of the theft of motor vehicles and motor vehicle parts (4). Since this survey only compared single-family homes, and not criminal activity regarding places of commerce or high-density vehicle parks, these results appear more reasonable.

Burglary constitutes the majority of criminal activity in the 36 paired communities at 58%. This again appears intuitively reasonable as the sample reflects only single-family homes that, by their very nature, are more spatial isolated. This isolation and lower population density makes these homes better targets for potential criminals. The final 15.7% of crimes come in the form of assault. This is a bit higher than the Durham average for violent crime. However, approximately 80% of these assault offenses are deemed simple assault, a misdemeanor, and are not captured in the aggravated assault numbers that figure 3 highlights. Thus, aggravated assault rates are actually found to be lower in the 36 selected communities than Durham on the whole.



Though it would appear that cul-de-sacs in Durham are less prone to criminal activity than adjacent two-way streets, taking this result at face value would be misleading and overshadows several notable concerns. First, the relative breakdown of types of crime between two-way streets and cul-de-sacs is not markedly different; larceny, assault and burglary are all proportionally similar. As it looks, there is simply more crime by volume on two-way streets, perhaps pointing to housing volume concern.

I assume that the selected pairings of cul-de-sacs and two-way streets have a relatively similar number of housing units. While a handy approximation, this is most likely not the case. Cul-de-sacs tend to have fewer housing options than comparable two-way streets, given the nature of their short circular design. Additionally, housing on two-way streets may be more densely concentrated with smaller lot sizes. Adjusting for lot size is necessary to smooth out these crime rates. This is a systematic limitation of relying on GIS data; it is not specific enough to determine and adjust for property sizes. However, in order to claim that crime differences are completely negligible between cul-de-sacs and two-way streets purely on the grounds of total housing units, it would be necessary to assume nearly five houses on each two-way street for every one in each cul-de-sac. This seems unlikely, as plot sizes appear comparable and such housing communities were carefully selected. Thus, cul-de-sacs do likely maintain some spatial advantage with respect to criminal activity holding housing volume constant and lot size constant.

Johnson and Bowers raise another relevant point in their study of cul-de-sac safety: it is possible that the type of people who live on cul-de-sacs differ from those who live on two-way streets in ways that might increase their risk of victimization (107). Such people may be of a different socioeconomic status, age, marital status or ethnicity. Perhaps such households predominantly have young children. Given these considerations, it is possible that risk profiles differ among residents and may contribute to a difference in criminal activity exposure that is not linked to the spatial design of such communities.

Spatial permeability is another concern. Johnson and Bowers reference a study by Armitage (2007) that showed crime rates were typically lower in communities built on cul-de-sacs, except when such communities were connected to other streets or public areas by footpaths or trails (107). This is a spatial concern that is directly linked to crime rates and is notably absent from the findings above. Future research would need to account for such connectivity differences in cul-de-sacs; available GIS data is again limiting.

Though cul-de-sacs appear moderately safer than two-way streets, ultimately, correlation does not prove causality. While there appears a strong correlation between lower crime rates and spatial housing layout, one does not dictate the other. Further research is necessary to better evaluate such claims.



It is difficult if not impossible to accurately assess the magnitude that spatial differences play in determining criminal activity. A variety of socioeconomic variables, as well as serendipity, are often present and problematic to untangle from pertinent spatial differences. Others biases may yet be at play. For example, residents of cul-de-sacs tend not to be random; they are self-selecting and seek the benefits and style of living that cul-de-sacs provide. Further, they may also be more affluent. Given two ideal communities that exist ceteris paribus, the one located on a cul-de-sac will command a higher economic premium than the one on a two-way street. Therefore, a comprehensive analysis of spatial differences in crime rates remains illusive. Moreover, even if such an analysis were possible, it would be questionable to extend the findings to communities outside of a given region; local political and socioeconomic differences appear too nuanced to realistically do so.

            However, that is not to say that such research is fruitless. The careful public policy strategist or city planner may continue to utilize GIS data to better understand local spatial differences in order to craft germane policy. Further, such spatial criminal research may be used to identify high-crime communities and lead to better policing of public spaces. A good example comes from the East Weaver Street public housing community. Though East and West Weaver Street are over one mile in length, 6 out of every 7 crimes committed in the area occur in a small public housing community that constitutes a tenth of mile. Insights like this may be useful to help local authorities increase policing and to direct city planners’ time to addressing the causes of criminal activity in the area.






Figure 1: A typical cul-de-sac as shown from Google Maps.




Figure 2: Index Crime Rate per 100,000 Residents by Year for Durham, North Carolina (Durham Police Department).





Figure 3: Crime Breakdown for 2012 in Durham, North Carolina (Durham Police Department).




Figure 4: Durham Crime Mapper Software screenshot. Provides a graphical layout of crime locations as well as police and sheriff tables that detail the exact location of criminal offenses. Red dots indicate police responses while yellow stars indicate sheriff responses.




Durham Police Department. Annual Report: 2012 Durham Police Department. Durham, 2012.             http://durhamnc.gov/ich/op/DPD/Documents/2012AnnualRepor0301FINAL.pdf.


Durham Police Department. Crime Mapper Online Software. Durham, 2012.    http://gisweb.durhamnc.gov/gis_apps/crimedata/dsp_entryform.cfm


Hillier, Bill. “Can Streets Be Made Safer?” Palgrave Macmillan 9.1 (2004): n. pag. ProQuest. Apr. 2004.           Web.    http://proxy.lib.duke.edu/login?url=http://search.proquest.com.proxy.lib.duke.edu/advanc  ed?url=http://search.proquest.com.proxy.lib.duke.edu/docview/194522636?accountid=105            98.


Jagannath, Thejas. “Do We Need Cul-de-sacs?” Urban Times RSS. N.p., 25 Jan. 2013. Web. 01 Apr.             2013. <http://urbantimes.co/2013/01/do-we-need-cul-de-sacs/>.


Johnson, Shane D., and Kate J. Bowers. “Permeability and Burglary Risk: Are Cul-de-Sacs Safer?”    Journal of Quantitative Criminology 26.1 (2010): 89-111. Print.


Lonngren, Betty. “Cul-de-sacs Unproven As Deterrent To Crime.” Chicago Tribune. N.p., 25 Apr.      1993. Web. <http://articles.chicagotribune.com/1993-04-25/business/9304250091_1_sacs-      cul-chicago-neighborhoods>.


Nielsen, John. “Cul-de-Sacs: Suburban Dream or Dead End?” NPR. NPR, 07 June 2006. Web. 07   Apr. 2013. <http://www.npr.org/templates/story/story.php?storyId=5455743>.


Historic Designation and its Effect on Durham Home Prices

by Bernadette Lowell  DP_LowellBernadette


Historic designation and the process of historic preservation have saved homes and commercial properties across the country from being torn down for newer construction. As space becomes sparse in larger cities, occasionally it is necessary to tear down these properties, however this can cause unpleasant construction and “mismatched” neighborhoods with homes from many different eras. Through national and local historic designation, owners can receive tax breaks and other incentives to keep their home in its original, historic form.

Durham is a prime example of a city that would need historic designation. Although it is not lacking for open land, there is a rich history in many downtown buildings and homes that needs to be preserved. Tobacco factories line downtown streets along with many homes dating back to the 1920s and 30s.

Durham has sought to preserve its local historic resources by “inventorying historically significant structures in the City and County, designating local historic districts and landmarks, establishing and supporting the Historic Preservation Commission (HPC), and nominating properties and districts for listing on the National Register of Historic Places.”(City of Durham) The HPC meets throughout the year to approve changes for any historic home and possibly designate new neighborhoods. Homes in any of the designated neighborhoods are taxed at 50% of the properties’ value. (City of Durham) However, with this local designation come certain restrictions. In order for any change to be made the exterior of the building, the owner must receive approval in the form of a Certificate of Appropriateness after a meeting with the HPC. (City of Durham)

A few homes and neighborhoods in Durham are also nationally designated. According to the National Register of Historic Places website, Durham county has 77 listed historic places and districts. [1]Although this can mean some federal tax breaks, there are little to no restrictions on any changes to the homes.

Historic neighborhoods in Durham

This study primarily looks at homes in two of the city’s seven historic districts. The Lakewood Park Historic District (Fig. 1), in southwest Durham, was listed as a national historic district in 2003. (Lakewood Form) It includes the blocks 2002-2112 Chapel Hill Road; 1601-1907 West Lakewood Avenue; 1406-1602 James Street; and 1809-1819 Bivins Street. (Lakewood Form) According to the application for historic designation, the buildings were built in 3 “generations” from 1902-1920, during the 1920’s, and in the mid 1930’s. These houses progressed from one-story homes with “modestly stylish Queen Anne features” to bungalows and into the “Minimal Traditional style.” (Lakewood Form) Many of these homes retained a high level of integrity throughout the years, making them prime candidates for historic designation and preservation. Each contributing home was constructed before 1952 and maintains enough of the original design and workmanship to be considered historic.

The Holloway Street District (Figure 2) was nationally designated in 1985 and is also considered to be a local historic district. The neighborhood, located much closer to the downtown area, dates back to the 1860’s, though it was reported in the application that many of the homes only date back to the 1880s through 1920’s.  At the time of designation, many homes were “intact but deteriorated,” with some left vacant and vandalized. (Holloway Form)


For the following regressions, I primarily used data from Zillow.com, which listed the number of bedrooms and bathrooms, square footage, lot size, year built, date and price of last sale, and their own “Zestimate.” This number is Zillow’s own estimate of the home’s market value.[2]

The Lakewood Park Historic District has 83 designated properties. Of these, 13 were commercial properties, vacant lots, or did not have enough information on Zillow, and 8 others were considered “multifamily” and were not included in the regression. These 62 homes in the district had an average Zestimate of $163,167 with a median of $154,112 and were on average built in 1931 with a median of 1924. (Figure 3)

The historic neighborhood was compared with the surrounding area, which is not historically designated on the national or local level. Using similar constraints—no multifamily or commercial properties— and removing any home without information on Zillow, 48 properties were chosen. These homes had an average Zestimate of $115,824 with a median of $108,780. On average they were built in 1950. (Figure 4)

The Holloway Historic District has 29 total homes. Seven of these had no information on Zillow and another three were considered multi-family. Of these 18 total homes, the average Zestimate was $165,739 with a median of $139,378. On average, the homes were built in 1931 with a median year of 1928. (Figure 5) The Holloway homes were compared with 84 surrounding homes. These homes had an average Zestimage of $69,912 with a median of $62043. They were, on average, built in 1934 with a median of 1925. (Figure 6)


For the following regressions, I used the formula lnZestimate=F(Historic, Characteristics) where Historic is a dummy variable for historic designation, and Characteristics include number of bedrooms and bathrooms, the square footage, lot size, and year. This hedonic model is in semi-log form, which implies that the coefficients for each explanatory variable are the percentage change in price with each one unit increase in that variable. From these variables, it is clear that there will be correlations between these explanatory variables, designation specifically being negatively correlated with year and positively correlated with other features of the house. Including all available Characteristics variables decreases the bias of the Historical variable.

For my initial regression, I only included Historic as an independent variable. In the Lakewood neighborhood, designation was found to have a positive influence on the Zestimate, with a coefficient of .317 and a t ratio of 5.52. (Figure 7) For homes in the Holloway district, designation had a positive influence on the Zestimate with a coefficient of .738 and a t ratio of 8.66. (Figure 8) Though both historic neighborhoods had a similar mean price, this initial regression reveals a significantly stronger influence of historic designation on the Holloway district. This could also be clearly seen through the mean Zestimates of the neighborhoods, as it is evident that the area surrounding the Holloway neighborhood contains considerably cheaper homes.

Following this regression, I included the rest of the independent variables for the characteristics of the home. In the Lakewood neighborhood, historic designation had a similarly high, statistically significant coefficient of .105 and a t ratio of 2.63. (Figure 9) This implies that, even after taking into account the structure of the home, historic designation brings a 10% increase in property value for the Lakewood neighborhood.

However, in the Holloway neighborhood, Designation had a coefficient of -.0019, though with a t ratio of -0.03. (Figure 10) Unfortunately, it is hard to draw any conclusions from this regression, as it is statistically not significant. Inclusion of a wider set of observations and explanatory variables could ultimately help this regression analysis.


From these regressions, I can conclude that the Lakewood historic designation significantly increases property value compared to the surrounding homes with the given data. Unfortunately, I cannot draw any conclusions after performing regressions on the Holloway District data.

If it is true that historic designation has lead to increased property values, then there are some potential policy issues that Durham will face. If these homes are in poorer areas, then an increase in property value could drive way poorer residents. If a neighborhood is locally designated, there is also an added burden to keep the home in its original form—requiring any change or fix to be approved by the HPC. In order to keep these homes affordable and less burdensome, there will need to be policy to keep residents in their historic home.

In order to conduct a better analysis of the impact of historic designation on home prices in Durham, this study would need more observations. As there are seven locally designated and fifteen nationally designated neighborhoods, there could be a large change in the outcomes with these homes included in the analysis.

This data set also did not include other variables about the features of the homes. Although they might not be significant, if a home has a garage, an attic, or basement could factor in to the property value as well as what the home is made of or even if it has been foreclosed on in the past. With these, the regressions might be more accurate.

One final inclusion to this data set could also be time. This regression does not show how the home prices have changed since the historic designation. A follow up study could look into whether or not these home prices have increased or decreased more rapidly than those homes in the surrounding blocks.



Figure 1



Figure 2















Map taken from: http://durhamnc.gov/ich/cb/ccpd/Documents/Historic%20Preservation%20Information/Historic_Resources_34x44_020312.pdf


Figure 3

Lakewood Historic Homes (N=62)

  Zestimate Bed Bath Sq. Ft. Lot Year built















Figure 4

Lakewood Non-historic Homes (N=48)

  Zestimate Bed Bath Sq. Ft. Lot Year built















Figure 5

Holloway Historic Homes (N=18)

  Zestimate Bed Bath Sq. Ft. Lot Year built















Figure 6

Holloway Non-historic Homes (N=84)

  Zestimate Bed Bath Sq.Ft. Lot Year built
















Figure 7

Lakewood Neighborhood, first regression


Figure 8

Holloway neighborhood, first regression


Figure 9

Lakewood Neighborhood, second regression




Figure 10

Holloway Neighborhood, second regression





“Historic Preservation.” City of Durham. N.p., n.d. Web. 29 Mar. 2013. http://durhamnc.gov/ich/cb/ccpd/Pages/HPC%20Items/Historic-Preservation.aspx

Individual Property Form for Holloway Street District. June 1984. Http://www.hpo.ncdcr.gov/nr/DH0188.pdf.

Leichenko, Robin, N. Edward Coulson, and David Listokin. “Historic Preservation and Residential Property Values: An Analysis of Texas Cities.” Urban Studies 38.11 (2001): 1973-987. Sage Journals. Web. 4 Feb. 2013.

USDI/NPS Registration Form-Lakewood Park Historic District. 07 Mar. 2003. Http://www.hpo.ncdcr.gov/nr/DH2541.pdf.






[1] http://nrhp.focus.nps.gov/natreghome.do

[2] http://www.zillow.com/wikipages/What-is-a-Zestimate/

[3] The following regressions are similar to those used in Historic Preservation and residential Property values: An Analysis of Texas Cities, which was presented in my literature review.

The Effect of Relative Size on Housing Values in Durham

by Michael Ni  DP_Ni_Compressed

Property-Value Movement in Old North Durham

By Valtcheva Katerina DP_ValtchevaKaterina

The part of Old North Durham around Foster St. and Geer St. has historically been an industrial neighborhood. It was generally considered a bad area of Durham despite its proximity to the central business district. In the past few years, this part of the city has been subject to some changes in development patterns, which have begun to transform the neighborhood. I will present my analysis of property-value movement in the area and examine it in the context of the new businesses that have been opening in the area. My hypothesis is that this recent commercial upturn, which was organic in origin, has had a positive impact on residential property prices in the part of Old North Durham that lies on and surrounds W Geer St. In order to confirm this hypothesis and quantify the impact of the new developments on existing structures, I have compared forty houses in Old North Durham (in proximity to Motorco Music Hall) with forty houses in East Durham and Edgemont over the last eight years. I have used Zillow estimates to obtain house value appraisals, collected annually, for the first month of each year starting in January 2006 and ending in January 2013.

Old North Durham initially developed with the construction of tobacco warehouses, as well as related commercial and industrial ventures nearby. North Durham Elementary School has had an influence on the area with respect to social development, while the Farmers market has also been influential in the area’s development as a center of Durham’s alternative lifestyle scene. However, for quite some time the area has been known as a rough part of town, and public perception to this effect has only recently begun to change.  I believe that this trend started with the opening of several bars and restaurants in the area: Fullsteam Brewery was opened on August 13th 2010 on Rigsbee Ave; in 2011 a popular bar and music venue, Motorco Music Hall, was opened right across the street from Fullsteam; and a restaurant called Geer Street Garden, on the corned of Foster St. and Geer St., followed on May 5th the same year. Further, in the end of January 2013 a café called CocoaCinnamon—the area’s most recent addition—opened its doors across the street from Geer Street Garden. Interviews with the owners of these businesses revealed that a big part of what drew them to the area was not government policy, but a combination of its proximity to the city center and its industrial aesthetic: indeed, Fullsteam Brewery occupies a prior soda bottling facility, Motorco a car dealership, and CocoaCinnamon an automotive repair shop. The buildings that were chosen for these additions to Durham were not only located a short walk away from W Main St and the commercial heart of the city, but also happened to be particularly suitable for the type of bars, restaurants, and clubs that their owners were searching for.[1] ,[2], [3]

The area I have chosen to compare Old North Durham to is similarly industrial, with its historical roots in milling and textiles. East Durham and Edgemont’s two textile mills have been turned into a retirement home and office spaces, which made them unavailable at the time when Old North Durham was chosen to host Motorco Brewery.[4] Other reasons why I have chosen East Durham/Edgemont for my control area are that its houses are very similar to the ones around Motorco, built around the same time period, and that it is approximately the same distance away from the central business district as Old North Durham is. As of 2006, the area was more economically depressed than Old North Durham, but there are two reasons why I don’t believe that makes the results of this study less significant. The first is that, for the purpose of this study, I am interested in relative changes of values between the two neighborhoods pre- and post-2010. The second is that the difference would only be marginal to someone who is already taking the risk of opening a new business in Old North Durham.  While the worse-off state the latter neighborhood does not distort the findings of the study, is nonetheless one of the factors that may have contributed to Geer St. and Rigsbee Ave’s housing Durham’s urban revival. However, this result is better explained by the fact that the industrial buildings existing in East Durham/Edgemont were either unavailable or poorly suited for housing breweries and cafes at the time Fullsteam opened.

Now I would like to focus on how I selected properties for use in a quantitative comparison of development in these two areas. House values representative of the Motorco area in Old North Durham come from forty homes that are located on one of the following streets: W Geer St., North St., Hargrove St., Glendale Ave, Northwood Cir, and N Mangum St. The average house size of this sample is 1375 ft2 (± 523 ft2.) The average lot size is 6121 ft2 (±1640 ft2.) The year these properties were built ranges between 1910 and 2004. The average year built is 1937 (±20) years. For my control group in East Durham and Edgemont I have chosen houses on one of the following streets: Hart St., S Driver St., Roberson St., Angier Ave, Vale St., S Plum St., E Main St., Clay St., and Ashe St. The average home size in the control area is 1432 ft2 (±476 ft2.) The average lot size is 7047 ft2 (±1413 ft2.) The year the homes were built ranges from 1900 to 1992 and averages to 1929 (± 21) years. Although the area I have chosen for the controlled sample closely resembles North Durham in many aspects, the house lots there were generally bigger than the ones in North Durham. For the purpose of creating a better group of comparable properties, I have excluded some houses with significantly larger lots.  Moreover, the majority of houses in East Durham and Edgemont were built between 1900 and 1910. Therefore, I included in my sample as many of the later-built houses in the controlled area as possible in order to create a sample that more closely resembles the properties I had chosen in Old North Durham. With this changes, the group I control for has an average lot size of about 900 sq. ft. more than the sample of the North Durham area. The statistical analyses discussed below demonstrate that this is not a big enough difference to significantly distort the finding of my study, as the relationship between house prices and house size and amenities is much stronger than the relationship with lot size.  This is especially noticeable in a town such as Durham where land is not a scarce commodity, as opposed to a place such as New York City, where such a difference in lot sizes would have been significant.


Figure 1: Property value information collected from zillow.com

To explore the trends within home values in Old North Durham and East Durham/Edgemont, I first normalized property values of each house by the square footage of the house. I also normalized each property value by the historic average price of housing in Durham at large for January of the corresponding year (fig. 1) in order to ignore the effects of the housing bubble and bust. For the time being, I have ignored the effects of lot size, as normalizing by that value would imply a linear relationship between lot size and overall property value—a false assumption in this environment. I first plotted mean property values in the two areas prior to 2010, in order to see how house prices in the two areas have been moving prior Old North Durham’s recent commercial development. I found the best-fitting line between the points for each group and I observed a positive change in house prices for this time period with a difference in their slopes of .0113 in favor of the control area (fig. 2 and 3). This does not indicate a significant difference in the house price movements between the two areas prior to 2010.


Figure 2


Figure 3

When plotting the data for all eight years, the picture changes dramatically (fig. 4 and 5). House prices continue to rise in Old North Durham through the entire time period (though, with the addition of datapoints through 2013, the slope of the price curve falls from .0206 to .0108), while the overall trend in house values in East Durham/Edgemont has become negative. This means that the recent fall of house values in the area has been so great that the slope of the line describing the overall trend of house prices went from positive .0319 (between 2006-2010) to negative .0275 (when datapoints are included through 2013); a total change five times bigger difference between the two areas prior to 2010. Running a Student’s t test comparing mean home prices on a year-by-year basis, it was observed that the difference in mean home values between the two neighborhoods was not statistically significant until 2010, from which point it became consistently significant (p<<0.001) (fig. 6). These findings suggest that the rise of new businesses in Old North Durham have correlated with a buoying of property values in the surrounding area, while a similar area in Durham has experienced a statistically significant fall in residential property values.


Figure 4


Figure 5


Figure 6

In order to get a fuller idea of how the relationship between house prices in the area have changed, I ran a total of sixteen linear regressions of house prices versus home size, lot size, and home age in each area over the years investigated. These regressions gave me a measure of how much house price depends on these parameters. My findings are summarized in Figure 7.  From the first graph we observe a substantial marginal change in home value in Old North Durham with square footage. While the result in 2009 appears to be an outlier, all other years show a difference between the price dependence on square footage between the two areas that has been stable up to 2009 and then diverging since then (with only a small change in Jan 2013).  The divergence we see since 2009 shows that the new developments in Old North Durham have increased the gap between the willingness of people to pay more for extra square footage in between these two neighborhoods. The graph also illustrated that this divergence is not due to a significant increase in the customers’ willingness to pay for additional sq. ft. of a house around the Motorco Area as much as a decline in their willingness to pay for the same sq. footage in the control area. This is consistent with what we saw in Figures 4 and 5.


Figure 7

The second graph on figure 7 shows the marginal change of house prices with lot square footage. This does not represent a strong or clear relationship in these areas, either within each neighborhood over time, nor between them. One way to interpret this is the possibility that residents’ decision to buy a house in a poorer and worse neighborhood may not be influenced by the lot size as much as the ones in nicer neighborhoods. However, the small correlation between lot size and home value supports the idea that lot size does not really influence a residential property’s price in a city like Durham where land is abundant.

When looking at the third graph on figure 7, one must note that this the regression is performed on year built rather than on age, which means that the graph shows the relative change in home value for each additional year you add to a home’s age. Interpreting the relationship we see that on average older houses in the Motorco, or Old North Durham, area cost more than older houses in the control group. This is an interesting observation that could mean that there are more renovations in Old North Durham, while more renovations mean that the area has been taking a turn for the better since the new businesses came in.

There is more evidence that Old North Durham has begun to transform in the recent years. From my interview with Fullsteam’s owner I learned that he has noticed his clientele becoming more diverse though the years, and that a lot more students are comfortable stopping by than before. This is evidence that Old North Durham has improved in terms of safety and could be attributed to the opening of the new businesses—not only the new bars and restaurants, but also the food trucks, which are now stopping by regularly, make the area around Old North Durham a lot more vibrant than it used to be. The increased attention to this part of town is also evident from a dispute over a nearby area that is currently used as a soccer field. The disagreement comes from the fact there seems to be a conflict of interests between the members of the community over the use of the land after a proposed renovation.  Some residents are using this area to practice soccer and claim Durham does not have enough soccer fields based on national standards. However, others believe that the area should host a smaller soccer field and some playgrounds. “The park has been included in two Capital Improvements Plans, first in 2001 and again in 2005, but the city has not funded any upgrades there, saying money was better spent in other parks.”[5] The increased attention towards the soccer field debate today supports the idea that the area has began to take a turn for the better and could even suggest that gentrification is occurring.

The outcomes of this analysis provide strong evidence towards the positive effect of the new businesses in Old North Durham. The area’s house prices show stable growth in the past eight years, while a similar area such as East Durham/Edgemont has suffered a decline in house prices. This comes to show that risky investments in bad neighborhoods could pay dividends for residents as well as local business owners. The information in this analysis can be used as an example of how strategic reclamation of industrial spaces can prove an economic boon through enticement of the “alternative lifestyle” market. Basing its actions on the positive changes Old North Durham, the local governments may be able to stimulate the revival of other areas in Durham (as well as other cities with a ample vacant industrial space) by creating incentives for entrepreneurs to develop such spaces.


[1] Holloway, Carson. Personal Interview. 28 03 2013.

[2] Wilson, Sean. Personal Interview. 25 03 2013.

[3] Barrera de Grodski, Leon. Personal Interview. 23 03 2013.

[4] Holloway, Carson. Personal Interview. 25 03 2013.

[5] Schwartz, Joe. “Bad blood brewing over Old North Durham Park.” Indyweek.com. N.p., 13 4 2011. Web. 6 Apr 2013. <http://www.indyweek.com/indyweek/bad-blood-brewing-over-old-north-durham-park/Content?oid=2361969>.

Heterogeneous Effects in the Housing Market’s Reactions to Sex Offenders: Evidence from Durham, North Carolina

by Yuan (Ingrid) Zhuang  DP_ZHUANGINGRID

1   Introduction

Since the enactment of the “Jacob Wetterling Crimes Against Children and Sexually Violent Offender Registration Program” in 1994, all states are required to maintain a registry of convicted sexual offenders.  Later in 1996, Megan’s Law amended the Wetterling Act (1994) by requiring law enforcement authorities to disseminate registered sex offenders’ information to the public.  While federal law provides two major information services to the public: sex offender registration and community notification, states are given significant latitude in their implementation of these provisions.  Currently every state has complied with the legislation and most states have websites that provide access to the sex offender registry over the Internet.  Accessible information typically includes offenders’ names, pictures, type of crime(s), incarceration dates, addresses of where offenders currently live, and registration dates.  North Carolina, Florida and Montana are the only states that provide information on offenders’ move-in dates to the public.

Although laws are implemented nationally, 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, 2006).  Understanding the relationship between sex offense risk and property values is useful for measuring the willingness of individuals to pay to reduce their exposure to crime risk.  Some evidence shows that convicted sex offenders have a substantial probability of reoffending.  Hanson et. al. (2003) found in their study that sexual recidivism rates are approximately 14% after five years, 20% after 10 years, and 30-40% after 20 years.  These results may substantially underestimate the actual rates because some sex offenders re-offend and are not caught.  Therefore, with the increase in transparency level on sex offenders’ information, we expect that the move-in of a sex offender will depress the property values in a neighborhood, particularly for houses in close proximity to a sex offender’s residence.

Literature examining the impact of crime risk on the real estate market is abundant.  As observed by a number of existing papers, individuals’ strong distaste for crime, especially sex offenses, indicates an inverse relationship between housing values and local crime rates.  Earlier studies suffer from potential omitted variable bias.  More recent literature improves on past estimations with their exploitation of data, but still lack detailed analysis of the during/post-recession reactions.  My research explores in particular how sex offenders’ residential locations affect property values in Durham, North Carolina.  It will contribute to the existing literature by firstly strengthening the causal effect of sex offenders’ move-ins on housing prices, and secondly observing whether the recession has changed the inverse relationship (comparing pre-recession reaction to post-recession reaction).  Ideally, we hope to examine whether neighborhoods with various demographics will react differently to the presence of sex offenders.  However due to the limitations on annual neighborhood demographics and income data, this extension is postponed for future research.

The rest of the paper is organized as follows.  In the next section, we provide a brief literature review on related studies concerning the relationship between a sex offender’s move-in and nearby property values.  Section 3 describes the sources of data and explains how we tailored the dataset to our specific use, and additional graphical evidence is presented.  In section 4, we describe the regression model we use to estimate the various parameters.  We present our empirical results in section 5 and analyze how well they fit our hypotheses.  Section 6 finally concludes the paper with our findings, implications of this study and suggestions for future research.


2   Literature Review

A number of papers have found an inverse relationship between property values and local crime rates.  Earlier studies, such as Richard Thaler (1978) and Steve Gibbons (2004) find respectively a 3 percent and a 10 percent decrease in property values for a one-standard-deviation increase in property crime.  However Linden and Rockoff (2006) points out, these results may suffer from potential omitted variable bias in both the cross section and times series, since crime rates are likely to co-move with other unobserved factors and may change as the characteristics of neighborhoods change.  More recent studies Linden and Rockoff (2006) and Jaren C. Pope (2008) improve on past estimates through the use of hedonic estimation methodology to measure the impact of crime risk on property values.  Linden and Rockoff (2006) overcomes 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.

In their paper, 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.

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, which my paper examines, is whether the recession has changed the relationship between a sex offender’s arrival and nearby property values, comparing pre-recession reaction to post-recession reaction.


3   Data     

Our analysis is based upon four sets of data from 2003 to 2012: sex offenders’ data, housing data, income data, and neighborhood demographics data.  North Carolina Department of Justice (NCDOJ) provides information on convicted offenders’ basic characteristics, type of offense, date of offense, current address, and date of registration at current address.  Because of the strict provisions governing timely registration in North Carolina, we could assume the registration date is the best approximation for sex offenders’ move-in dates to their current locations, as Linden and Rockoff (2006) explained in their data source section.

There are 279 registered sex offenders in Durham, North Carolina as of April 7, 2013.  A total of 20 registered sex offenders are randomly selected from 8 zip codes in Durham.  Each offender is matched with an “arrival date” based on the date him or her registered, and categorized into a “neighborhood” determined by homes within 0.1 miles[1] radius of an offender’s residence.  Some evidence suggests that sex offenders are more likely to commit their offenses locally than other types of criminals.  Larsen et. al. (2003) observe that 85% of sex offenders committed their offense in the same city in which they resided at the time of their arrest, and 65% of sex offenders committed in their own neighborhoods.  Therefore, it is reasonable to assume that houses within 0.1 miles radius of a sex offender parcel are affected the most after the move-in of a registered offender.

The second source of data comes from Zillow, an online real estate database that pin points the exact locations of houses on a map, and provides historical/current estimates and prices of individual properties.  Of the 279 registered offenders in Durham, 153 could not be matched with a parcel on the Zillow database due to the fact that some offenders are currently residing in Durham County Jail, some are homeless, and others simply could not be located on the database, leaving a total of 126 offenders for our analysis.  We merge the matched sex offender data with property prices from January 2003 to December 2012.  Housing prices are normalized to December 2005 dollars using the annual South Urban CPI, and averaged from aggregated annual prices of individual property located within 0.1 miles radius of an offender parcel.

Our third and fourth sources of data are the average income data from 2010 US census[2] and Durham neighborhood demographics (age, race, sex and education level) data from City Data[3].  These two datasets are excluded from our analysis due to the limitations of data availability.  Average annual income data are available only on city level, not neighborhood level, whereas neighborhood demographics data are not recorded yearly.  Therefore, income and neighborhood data could not be matched to the sex offender and the property values data, and thus could not be estimated by the regression model.

3.1   Graphical Evidence

If living close to a registered sex offender induces an increasing fear for crime risk, then close proximity has a negative impact on property values, and we should see prices of homes near the offender’s location fall subsequent to the offender’s move-in.  Figure 3.1 shows an average annual housing price of all neighborhoods in the dataset.  From the graph, we observe a decline in overall housing market around 2006, may be due to the 2007-2008 recession or other factors, which are not the main concerns of this paper.  Figure 3.2 reflects the treatment effect and compares average housing prices before and after sex offenders’ move-in.  There is clearly a decline in housing values after Year 0 (offenders’ move-in year).  This is consistent with our hypothesis that if the decline in property prices coincides with the offender’s arrival and does not reflect preexisting downward trend in prices, then we could support the causal impact of the sex offender’s arrival on the price fall within 0.1 miles radius.


Figure 3.1 Average Property Value by Year


Figure 3.2 Average Property Value Pre- and Post- Offender’s Arrival


4   Model and Estimations

Ideally, if we observe the move-in dates of sex offenders and both time series and cross-sectional data on housing prices, we would be able to estimate a dynamic model of the housing market’s reaction to the presence of sex offenders.  Theoretically, we hope to estimate a static differences-in-differences model, capturing the heterogeneous reactions due to incumbent racial composition, sex ratio, age distribution, and income distribution (Equation 4.1).  However due to data availability on the local level, we could only focus on the impact of an offender’s arrival on nearby housing prices and explore the effect of the recent recession on the relationship (Equation 4.2).  We will apply ordinary least squares with auto-correlated errors to neighborhood level panel data on both level and changes in housing prices, while controlling for neighborhood and year fixed effects and linear time trends.

Hnt=α+β0*SOnt1*Rt2*SOntRt1*racent*SOnt2*sexnt*SOnt3*agent* SOnt+ γ4*incoment*SOnt+X+εnt                                                                             (Equation 4.1)

Hnt=α+β0*SOnt1*Rt2*SOntRt+X+εnt                                               (Equation 4.2)


Equation 4.1 is a theoretical model.  Variables are defined as following: H is the housing price, SO is a dummy for sex offender, R is a dummy for recession, race is a dummy for racial composition of individual neighborhood, sex is a dummy for male and female ratio in a neighborhood, age is a dummy for median age in a neighborhood, income is a dummy for median household income, X represents control variables (neighborhood and year fixed effects, and linear time trends), n stands for neighborhood and t is time.

Equation 4.2 is the actual model we used for our estimations, which extends previous literature by looking at the impact of recent recession.  Because we are not able to find demographics and income data on the neighborhood level, we have to take out the demographics terms, leaving Equation 4.1 for future research. 


5   Results

Table 5.1 Impact of Sex Offenders’ Arrival on Housing Value and Recession’s Effect



Model 1

Model 2











Notes: year and community fixed effects and time trends are controlled but not reported. P-values are reported in parentheses.
* statistically significant at 10% level, ** statistically significant at 5% level
# of observations = 200, Prob>F = 0.0000, Adjusted R2 = 0.9758





β0 represents the sex offender term, and β2 is the interaction term (=offender dummy * recession dummy).

Since it is not absolutely clear if we are still in a recession, we tried running regressions on all different combinations of recession years.  As shown in Table 5.1, Model 1 considers all years after 2007 as the recession period; Model 2 considers only the years 2007 and 2008 as the recession period.  Model 2 is Model 1’s robustness test, and our model is robust to the definition of recession.

Before running our regressions, we predict that the negative impact of sex offender’s arrival on houses in close proximity would be lessened by the effect of recession.  We suspect that during recession time, people have other worries and would be willing to tradeoff dis-amenities such as living close a sex offender for more important amenities and characteristics of a neighborhood.  Therefore, we do not expect to see such a dramatic depression of housing prices affected by sex offenders’ arrival during/post-recession compared to pre-recession years.

In Model 1, we observe a 5.7% decline in housing price following a sex offender’s arrival.  Although the results of the offender and the interaction terms are significant at 5% and 10% level respectively, but the negative effect is almost completely offset by the presence of recession (the slightly higher absolute value of the coefficients on the interaction term is puzzling, and the total effects of these two terms are statistically insignificant from zero, so we cannot conclude anything from this).  Similar results are observed in Model 2.  Note that the adjusted R square (0.9758) is very close to one, meaning that the fixed-effect panel data model has explained most of the variation.  We hoped but could not include relevant demographic control variables due to limitation of data on the very local level, but neighborhood fixed effects should be able to capture much of these characteristics.


6   Conclusion

In our analysis, we employ the hedonic pricing method to estimate the impact of local crime risk on housing values.  Exploiting detailed data on the current locations and move-in dates of registered sex offenders (information that is publicly available on the North Carolina Sex Offender Registry), we find that the arrival of a sex offender in a neighborhood depresses the values of houses within 0.1 miles radius by roughly 5.7%.  Although we could not explain the magnitude of the coefficients of the interaction term, the results observed imply a lessened effect for during/post-recession since overall income goes down and homebuyers are less critical about the dis-amenities of a neighborhood.  Also the adjusted R square being nearly one should be especially noted.  Since our estimations are consistent with existing literature, we believe that our recession results improve on and add a new dimension to the research question regarding the causal relationship between sex offenders’ presence and property values.

Unfortunately, we are unable to test the heterogeneous effects caused by differences in neighborhood demographics.  Theoretically, it is reasonable to predict that neighborhoods with older residents tend to care less about sex offenders’ presence whereas, neighborhoods where the majority of the residents are young families, they would strongly prefer residing in a sex offender free zone.  Our hypothesis could not be statistically tested in this paper.  However when the income and demographics data become available on the local level, we could proceed to test Equation 4.1 in our future research.  And thus understanding the heterogeneous reactions to local crime risk (due to neighborhood demographics effects) becomes important in determining optimal policy decisions, such as proper level of policing and expenditures for programs that reduce crime.




Gibbons, Steve. 2004. “The Costs of Urban Property Crime.” Economic Journal, 114(499): F441-63.

Hanson, Karl R., Kelly E. Morton, and Andrew J. R. Harris. 2003. “Sexual Offender Recidivism Risk.” Annals of the New York Academy of Sciences, 989: 154-166.

Larsen, James E., Kenneth J. Lowery, and Joseph W. Coleman. 2003. “The Effect of Proximity to a Registered Sex Offender’s Residence on Single-Family House Selling Price.” The Appraisal Journal, 71(3): 253-65.

Linden, Leigh, and Jonah E. Rockoff. 2006. “There Goes the Neighborhood? Estimates of the Impact of Crime Risk on Property Values from Megan’s Laws.” American Economic Review, 98(3): 1103-1127.

Pope, Jaren C. 2008. “Fear of Crime and Housing Prices: Household Reactions to Sex Offender Registries.” Journal of Urban Economics, 64(3): 601-614.

Thaler, Richard. 1978. “A Note on the Value of Crime Control: Evidence from the Property Market.” Journal of Urban Economics, 5(1): 137-45.


[1] The use of 0.1 miles radius is based upon the reasoning Linden and Rockoff (2006) gave in their paper.

[2] Data obtained from http://www.census.gov/2010census/data/

[3] Data obtained from http://www.city-data.com/nbmaps/neigh-Durham-North-Carolina.html

Did Southpoint Mall Lower Property Values?

by David Wang DP_WangDavid

Durham Paper: Externalities of Public Parks

by John Travalini DP_Travalini

Predicted impact of Triangle Transit’s light rail project on Durham’s captive bus riders

by Bao Tran-Phu


  1. I.              Introduction


North Carolina’s Triangle Transit Authority’s (TTA) plan to construct a light rail network between Durham, Orange, and Wake Counties has been in the works since 2006, but insufficient funding has repeatedly stopped it in its tracks (Roberts, 2006).  The project finally got its first major green light in November 2012 when Orange County residents passed a half-cent sales tax to fund long-term transit investments, after Durham County residents had approved an equivalent tax hike a year before (Freemark, 2011; Grubb, 2013).


With the introduction of light rail, some may worry that its construction and operation may pull resources away from other transit forms, especially local bus systems that are typically disproportionately ridden by the poor (Garrett & Taylor, 1999).  The sales tax hike is expected to cover a large share of the future rail’s costs and even fund bus service expansions.  Nonetheless, the rail could still negatively affect bus funding in the long run, as subsequent investments and expansions might focus more on the publicly popular and environmentally friendly rail.  Triangle Transit has been expanding its bus fleet rapidly in the past years, increasing its size from 60 buses in 2011 to 64 in 2012 (TTA).  Even if the rail project does not force TTA to decommission any of its buses, its bus fleet could cease to grow as quickly in the future.


This paper examines the effect that the light rail’s introduction is expected to have on Durham’s poor.  While the actual effect will depend on legislative decisions made after the rail is constructed, data on other transit systems nationwide is used to demonstrate how increases in rail expenses have historically caused spending on bus transit to change.  Even after controlling for population and economic growth factors, a regression analysis finds no evidence suggesting that increased rail expenses lead to decreases in bus expenditures.  Instead, increased rail expenses have a significant positive effect on bus expenditures, possibly by triggering a positive spillover effect whereby the increase in overall transit ridership leads to bus service expansions that benefit captive riders as well.


  1. II.            Literature Survey


Studies on public transit consistently demonstrate a disproportionate reliance on local buses among the poor.  While the propensity to use public transit increases as income falls (Baum-Snow and Kahn, 2000), the slope of this trend is not uniform across transit modes.  Iseki and Taylor (2001) find that the median household income for transit bus passengers falls between $15,000-$19,999, compared to $30,000-$34,999 for urban rail and $40,000-$44,999 for commuter rail.  Additionally, this propensity to use public transit is not simply a preference for many bus riders.  By looking at data on the L.A. Metropolitan Transportation Authority (MTA), Iseki and Taylor then find that 69.2 percent of MTA bus riders have household incomes below $15,000, while only 20.3 percent of the L.A. County population falls into that category (see Table 1).


Table 1 – MTA Bus Rider Demographics


1995 Household Income

< $7.5k $7.5-15k $15-35k $35-50k $50-75k >$75k

% of MTA Bus Riders

40.2% 29.0% 20.6% 6.0% 2.8% 1.5%

Source: Iseki and Taylor (2001)

Properties inherent to the transit types are blamed for these demographic differences in ridership.  Transit policy-makers typically categorize users as either “choice” riders, who have access to private vehicles, or “captive” riders, who do not (Garrett & Taylor, 1999).  On average, captive riders are more likely to be poor, belong to a minority group, and live in inner cities, while choice riders are more likely to be white and live in suburbs (Garrett & Taylor).  As rail systems are often designed for longer-distance travel connecting the suburbs to the inner-city, they will have an inherent tendency to benefit wealthier populations.


While studies agree that the poor rely on buses, the link between funding for rail and funding for bus has been studied only indirectly.  One of the most direct approaches came from Iseki and Taylor (2001), who use a cost-allocation model to find the average per-trip subsidies of the various transit modes.  Using data on the L.A. MTA, they find average per-trip subsidies of $3.17 for buses, $6.28 for express buses, and $7.85 for light rail.  This provides evidence that the construction of a new rail line would indeed require a substantial re-allocation of subsidies from buses to rails in the absence of any new external sources of funding.  However, in the case of Triangle Transit’s rail project, the half-cent sales tax passed in Durham and Orange Counties and any grants or subsidies that it will receive serve as external sources of funding.  To predict the rail’s impact on bus service expenditure, an analysis must be done on how changes in spending on rail transit leads to changes in spending on bus transit in the short and long run.


  1. III.         Data


To study how the introduction and expansion of rail transit systems in other U.S. cities have affected funding and support for bus systems, I use National Transit Database (NTD) data on total operating expenses by transit system from 1991-2011, separated by transit mode (TS2.1, 2013).  The NTD collects data on all transit systems that receive grants from the Federal Transit Authority.  I also use data from the World Bank on economic indicators such as GDP and per capita GDP.


Of the 896 urbanized areas (UZAs) listed in the NTD, 39 simultaneously operated bus and rail systems at some point between 1991 and 2011.  Of those, 14 have rail systems that opened during that period, thereby providing data on bus expenditure patterns before and after the rail systems’ openings (see Appendix Table A1).  With the exception of Phoenix’s Valley Metro, the opening of a rail system was never coincident with any substantial decline in bus expenditures.  Further, rail openings do not appear to stall the gradual increase in bus expenditures over time that most AZUs exhibit (see Appendix Figure A2).


III.I. Measuring changes in operating expenditures over time


Operating expenditures were studied rather than capital or total expenditures, as they provide a more direct indicator of service expansions: as rail infrastructure investments are being made, its capital expenses may span unpredictably across many years in advance, but operating expenses will only increase once the new lines open.  Operating expenditures also lack the volatile and cyclic fluctuations characteristic of capital expenditures: Triangle Transit’s operating expenditures changed by only 11 percent from 2011 to 2012, while its capital expenditures changed by 48 percent (TTA; see Appendix Figure A1).


Using this NTD data, I calculate each year’s growth in operating expenses for bus and rail (ΔOpexBus and ΔOpexRail) for each AZU.  Given year i and AZU j,


ΔOpexBusij = OpexBusijOpexBusi[j-1]


ΔOpex is preferred over the given year’s actual operating expenses, as using the data to study the relationship between current bus and rail operating expenses primarily would capture between-city differences rather than within-city changes.  Further, it is insufficient to find whether increasing rail expenditures causes decreases in bus expenditures, as a rail system might stall the bus system’s growth without actually causing it to contract.  Using annual change-over-time values for bus and rail operating expenses addresses both of these limitations, as the ΔOpex captures only within-city changes and can itself be compared between cities.


Since ΔOpexRail is usually highest when an AZU has just opened or expanded its rail system, a negative relationship between ΔOpexRail and ΔOpexBus would suggest that Triangle Transit’s rail system will hurt the bus system by at least stalling its growth.  A neutral or positive relationship would suggest that Triangle Transit’s bus system would not only continue to grow after the introduction of the rail line, but its rate of growth would stay the same or even increase.


III.II. Three time intervals to measure short-term and long-term effects


The effect of rail expenditures may not be instantaneous.  A transit system may choose to construct or expand its rail network without making any cuts to its bus services, only to find its newfound deficit unsustainable.  Furthermore, the new rail network may consume future investment resources that the transit system would have otherwise channeled into its bus network.  In either case, the growth of a rail system may have no immediate impact on buses, but it may still have a lagged effect.  Therefore, the empirical analysis is repeated three times, each using a different time interval (Δt):

Δt = 1 year:     ΔOpexj = Opexj – Opex[j-1]

Δt = 2 years:    ΔOpexj = Opexj – Opex[j-2]

Δt = 3 years:    ΔOpexj = Opexj – Opex[j-3]


Each AZU contributes one observation for each data-year where both OpexBusij > 0 and OpexRailij > 0.  This method records all incidents where an AZU ended the one-, two-, or three-year time interval with both bus and rail systems running, regardless of whether the bus or rail systems existed at the beginning of the interval.  Meanwhile, it excludes all incidents where an AZU started and ended the interval with no rail system or no bus system.  In all, there were NΔt=1 = 780 observations for Δt = 1, NΔt=2 = 741 for Δt = 2, and NΔt=3 = 702 for Δt = 3.


  1. IV.          Empirical model


To study rail transit’s effect on buses, I construct an ordinary least-squares regression model:


ΔOpexBusij = β0 + β1ΔOpexRailij + β2Xij + β3Yi + β4YEARi + β3STATEj + uij


            Xij is a set of population factors, including UZA population, population growth rate, area, and population density.  Yi is a set of economic factors, including national per capita GDP and per capita GDP growth rate.  YEARi and STATEj are sets of dummy variables for each year and state, where Yr_11ij and State_WIij are their respective omitted reference variables.  (See Appendix Table A2 for a complete list of variables).

  1. V.             Results


The coefficient for ΔOpexRailij is both positive and significant at the 0.01 level for all three time intervals.  Thus, the rise in bus expenditures illustrated in Figure A2 (Appendix) does not merely reflect the expansion in both rail and bus services in response to some confounding factor, such as population or economic growth.  Even after controlling for population and economic factors, bigger increases in rail operating expenses predict bigger increases in bus operating expenses.  Furthermore, the coefficient increases as the time interval widens, meaning rail expenses’ acceleratory effect on bus expenditures strengthens over time (see Table 2; see Appendix Table A3 for unabridged regression results).


Table 2 – Selected regression results



Coefficient (Standard error)

  Δt = 1 Δt = 2 Δt = 3
ΔOpexRailij 0.66385




1.200903 (0.05365)*
UZA_Popij 6.589327




18.14565 (3.195444)*
UZA_Areaij -18828.6




-72415.86 (13212.49)*
UZA_Densityij 7452.779




23619.98 (6145.469)*
UZA_Growthij 1.02 × 109

(2.58 × 108)*

1.90 × 109

(3.68 × 108)*

2.60 × 109

(4.59 × 108)*

GDPi -682.802




-1537.338 (537.6729)*
GDP_Growthi 1.72 × 108

(6.64 × 107)*

2.67 × 108

(9.29 × 107)*

4.35 × 108

(1.82 × 108)**

Constant -7626747

(1.73 × 107)

-4.17 × 107

(2.48 × 107)***

-3.31 × 107

(2.99 × 107)

* Statistically significant at the 0.01 level

** Statistically significant at the 0.05 level

*** Statistically significant at the 0.10 level


The coefficients for UZA_Popij, UZA_Densityij, and UZA_Growthij (population growth rate) are all also positive and significant at either the 0.01 or 0.05 levels for all three time intervals.  Given that bus expenditures generally increase over time in this data set, this finding means that bus expenditures rise at a faster rate in more populous, dense, and faster-growing UZAs.  In contrast, the coefficient for UZA_Areaij is negative and significant at the 0.01 level for all three time intervals.  Interestingly, bus expenditures rise faster for smaller urbanized areas.


The impact of gross domestic product on ΔOpexBusij varies by time interval.  The coefficient for GDPi­ is significant at the 0.05 level under the single-year time interval and at the 0.01 level under the three-year time interval, but it is not significant at any level under the two-year time interval.  It is curiously negative, suggesting that bus expenditures rise more quickly when GDP is lower.  In a seemingly contradictory manner, the coefficient for GDP_Growthij is positive and significant at the 0.01 level for all three time intervals.  It is possible that GDP_Growthij has captured the effects of acute economic expansions and contractions, while GDPij reflects a longer-term relationship between GDP and public bus usage.

  1. VI.          Discussion


While bus transit remains a public service on which many urban poor rely on, the results from this study’s regression predict that Triangle Transit’s planned rail project will not harm Durham’s poor by triggering any bus service reductions.  Instead, the results suggest that the project will provide a net benefit for the bus system.  One plausible explanation for this is that various transit modes may act as complements to one another.  As rail networks develop, public transit in general becomes a more viable and convenient option for choice riders.  Overall transit ridership subsequently increases in the area, and the heightened bus ridership allows the AZU to expand its bus services.  Through this mechanism, rail and bus networks expand together as they work to foster a stronger transit network and community.  Thus, while a rail project may be targeted at choice riders originally, it could trigger a positive spillover effect that ultimately benefits captive riders as well.


This view is consistent with the finding that bus expenditures rise more quickly when per capita GDP is lower.  During times of economic distress, choice ridership is expected to grow more easily, as private vehicle owners respond more readily to the financial incentives offered by public transit.  It is also consistent with the finding that bus expenditures rise more quickly in denser regions.  Increased population density offers additional incentives for choice ridership, as amenities would be closer to households on average and as traffic would be more congested.


Additional research could be done to test for this rail-initiated spillover of benefits from choice riders to captive riders.  A fundamental component in that investigation would be to see whether rail projects can effectively attract choice riders, and whether the choice riders cause a significant increase in bus ridership.


Ultimately, the potential for Triangle Transit’s rail project to have a positive effect on Durham’s poor will depend on several factors.  First, the rail network’s construction and operation must be fully funded by new sources of revenue, so as to not pull resources away from TTA’s existing bus services.  In the long run, the rail must be able to draw in more choice riders from Wake and Orange Counties who would then utilize Durham’s bus system for shorter point-to-point transportation.  This increase in Durham bus ridership could lead to bus service expansions that benefit choice and captive riders alike.







Figure A1 – TTA total annual operating and capital expenses from 2004-2012



Table A1 – UZAs with rail lines opened after 1991

UZA Name Transit System Name Rail Opened
Baltimore, MD Maryland Transit Administration 1992
Memphis, TN-MS-AR Memphis Area Transit Authority 1993
St. Louis, MO-IL Bi-State Development Agency 1993
Denver-Aurora, CO Denver Regional Transportation District 1994
Dallas-Fort Worth-Arlington, TX Dallas Area Rapid Transit 1996
Salt Lake City-West Valley City, UT Utah Transit Authority 1999
Kenosha, WI-IL Kenosha Transit 2000
Seattle, WA Central Puget Sound Regional Transit Authority 2003
Houston, TX Metropolitan Transit Authority of Harris County, Texas 2004
Minneapolis-St. Paul, MN-WI Metro Transit 2004
Little Rock, AR Central Arkansas Transit Authority 2004
Charlotte, NC-SC Charlotte Area Transit System 2007
San Diego, CA North County Transit District 2008
Phoenix-Mesa, AZ Valley Metro 2008

Source: National Transit Database (TS2.1, 2013)


Figure A2 – Bus transit operating expenses before and after rail opening



Table A2 – List of regression variables


Variable name




Change in annual operating expenses of bus transit system, from year j – {1,2,3} to year j

ΔOpexRailij Change in annual operating expenses of rail transit system, from year j – {1,2,3} to year j
UZA_Popij AZU population
UZA_Areaij AZU area, in square miles
UZA_Growthij AZU population growth rate
UZA_Densityij AZU population density, in people per square mile
GDPi U.S. per capita gross domestic product
GDP_Growthi U.S. per capita GDP growth rate
YEARi Dummy variables for each year (e.g. Yr_92­i = {1 if i = 1992; 0 otherwise})
STATEi Dummy variables for each state (e.g. State_NCj = {1 if j is in North Carolina; 0 otherwise})

Note: Not all states have an AZU with both bus and rail transit systems, and STATEi dummy variables exist only for those that do.


Table A3 – Unabridged regression results



Coefficient (Standard error)

  Δt = 1 Δt = 2 Δt = 3
ΔOpexRailij 0.6638503




1.200903 (0.05365)*
UZA_Popij 6.589327




18.14565 (3.195444)*
UZA_Areaij -18828.55




-72415.86 (13212.49)*
UZA_Densityij 7452.779




23619.98 (6145.469)*
UZA_Growthij 1020000000


1900000000 (368000000)*


2600000000 (459000000)*
GDPi -682.8017




-1537.338 (537.6729)*
GDP_Growthi 172000000


267000000 (92900000)*


435000000 (182000000)**


State_AZj -1103388




13100000 (15900000)


State_ARj 4812837


8414082 (16800000)
State_CAj -18200000


-40300000 (10400000)* -61300000 (12900000)*
State_COj -7794084




-26400000 (15800000)***
State_CTj 17000000




42700000 (16300000)*
State_FLj -9615875




-15200000 (15700000)
State_LAj 23400000




36700000 (17800000)**
State_MDj 10300000




14400000 (14900000)
State_MAj 2476362




-10100000 (12700000)
State_MIj 20600000




69400000 (16300000)*
State_MNj 12600000




40900000 (14800000)*
State_NJj 27400000




84100000 (16100000)*
State_NMj -10400000


-24600000 (13100000)*** -38700000 (16300000)**
State_NYj 36100000




69400000 (14100000)*
State_NCj -21600000


-36800000 (18100000)** -47700000 (22500000)**
State_OHj 20900000




58900000 (16200000)*
State_ORj -4859887




-25400000 (15700000)
State_PAj 34700000




96700000 (17000000)*
State_RIj 21600000




56200000 (16300000)*
State_TNj 5679868




15200000 (12800000)
State_TXj -8233170




-15900000 (11700000)
State_UTj -8053100


-22100000 (13000000)*** -37400000 (16100000)**
State_WAj 20900000




63000000 (14700000)*
State_WIj (omitted: reference state)
Yr_92i 17200000


Yr_93i -11600000




Yr_94i -18300000




28600000 (11200000)**
Yr_95i -27200000




-11600000 (12300000)
Yr_96i -21100000




-18300000 (12100000)
Yr_97i -12400000




-27200000 (11800000)**
Yr_98i -9379006




-21100000 (11600000)***
Yr_99i 1298334




-12400000 (11400000)
Yr_00i 7351299




-9379006 (11200000)
Yr_01i 15800000




1298334 (11000000)
Yr_02i 11400000




7351299 (10900000)
Yr_03i 16500000




15800000 (10900000)
Yr_04i 20000000




11400000 (10900000)
Yr_05i 36000000




16500000 (10900000)
Yr_06i 33400000




20000000 (11000000)***
Yr_07i 10800000




36000000 (11200000)*
Yr_08i 742124.7




33400000 (11600000)*
Yr_09i 28600000




10800000 (12100000)
Yr_10i -11600000




742124.7 (12500000)
Yr_11i (omitted: reference year)
Constant -7626747







Asterisks denote statistical significance at the 0.01 (*), 0.05 (**) and 0.10 (***) levels.


Works Cited


Baum-Snow, Nathaniel, and Matthew E. Kahn. “The Effects of New Public Projects to Expand Urban Rail Transit.” Journal of Public Economics 77.2 (2000): 241-63.

Freemark, Yonah. “In North Carolina’s Triangle, the Passage of a Sales Tax Increase in Durham Is Just the First Step.” The Transport Politic. The Transport Politic and Yonah Freemark, 9 Nov. 2011. <http://www.thetransportpolitic.com/2011/11/09/in-north-carolinas-triangle-the-passage-of-a-sales-tax-increase-in-durham-is-just-the-first-step>.

Garrett, Mark, and Brian Taylor. “Reconsidering Social Equity in Public Transit.” Berkeley Planning Journal 13 (1999): 6-27.

“GDP per Capita (current US$).” The World Bank | Data. The World Bank Group, n.d. <http://data.worldbank.org/indicator/NY.GDP.PCAP.CD>.

Grubb, Tammy. “Sales Tax Rises a Half-cent Monday.” The Chapel Hill News. The News & Observer Publishing Company, 31 Mar. 2013. <http://www.chapelhillnews.com/2013/03/31/75668/sales-tax-rises-a-half-cent-monday.html>.

Iseki, Hiroyuki, and Brian D. Taylor. “The Demographics of Public Transit Subsidies: A Case Study of Los Angeles.” Presented at the TRB 81st Annual Meeting (2001).

Mildwurf, Bruce. “Triangle Transit Unveils Plans for Durham-Orange Light Rail.” WRAL.com. Capitol Broadcasting Company, Inc., 3 May 2012. <http://www.wral.com/news/local/story/11059619>.

Roberts, Mark. “Triangle Rail Project Won’t Receive Federal Funding.” WRAL.com. Capitol Broadcasting Company, Inc., 6 Feb. 2006. <http://www.wral.com/news/local/story/148290>.

TTA. “Triangle Transit Publications.” Triangletransit.org. GoTriangle, n.d. Web. <http://www.triangletransit.org/news/publications>.

“TS2.1 – Service Data and Operating Expenses Time-Series by Mode.” National Transit Database. Federal Transit Administration, 27 Jan. 2013. <http://www.ntdprogram.gov/ntdprogram/data.htm>

“What Is the National Transit Database.” National Transit Database. Federal Transit Administration, 26 Jan. 2013. <http://www.ntdprogram.gov/ntdprogram/ntd.htm>

Fare-Free Buses: A Comparison

 by Jeff Sinclair   DP_SinclairJeff

            Chapel Hill Transit (CHT) and Durham Area Transit Authority (DATA) both operate in the Research Triangle area of North Carolina, and both share important similarities. They both serve collegiate and healthcare areas, as well as commuters and residential areas. However, they share important differences. DATA serves a larger area both in terms of geography and population. DATA serves a city with mixed development; CHT serves a primarily residential community. However, the biggest difference between the two is CHT’s use of a fare-free system (with one or two exceptions). When implemented, Chapel Hill Transit believed that this fare-free system would substantially increase ridership. However, the alternative hypothesis is that this is not true. In order to explore this question, it is important to figure out the change in the characteristics of each system since the implementation of CHT’s free fare and then to hypothesize the potential factors that influence these changes. In comparing the changes and applying theory, the goal is to roughly determine the effect of a free-fare system on transit.

Chapel Hill implemented its fare free system in 2002. In the year before, the population of Chapel Hill was 48,902 and Durham’s was 192,397 By 2002, when the fare was implemented, the population of Chapel Hill increased by 3.35% to 50,540 and the population of Durham increased to 196,432, a 2.10% change. At the same time, annual passenger miles increased on CHT from 4,394,609 to 10,111,508—a change of 130%. This growth trend has continued, with far less magnitude, in every year since except for a slight decline in the years 2008 through 2010. Annual passenger miles on DATA went from 15,236,582 to 13,821,392—a decline of 10.2%. Durham’s change has been more variable, with declines occurring between 2001 and 2004 followed by a large jump in 2005 and sustained increases until 2009, followed by a slight decline in 2010.

Before continuing further, it is worth noting why annual passenger miles are being used as a measure of consumption. Because their routes vary in length, a good aggregate measure of production for transit companies is miles of transit route. Passenger miles is the aggregate number of miles all passengers have traveled—which is dependent on the miles of transit produced. Alternate measures such as number of boardings are inadequate because they do not account for the additional cost of longer routes, nor for their benefits to those who demand them.             Service increases are also important in the growth of public transit. Between 2001 and 2002, CHT also increased the service it provided (as measured by annual vehicle revenue miles) from 1,633,050 to 1,843,567, an increase of 12.9%. DATA went from 3,018,752 to 2,819,226—a decrease of 7.10%. The annual vehicle revenue miles hovered around this figure until 2008, when they increased by 513,755. Since then they have continued to increase modestly.

With a population increase of 3.35% and a service increase of 12.9%, it is to be expected that CHT would experience increased ridership between 2001-2002. However, ridership more than doubled, far exceeding the modest increase that the population and service increases would suggest. Furthermore, it seems unlikely that the 12.9% increase in service, even if intensively used, would account for the entire 130% increase, or even a large part of it. A more finite analysis would have to include route-by route information. Since then, however, CHT has only experienced moderate growth, and in a few years a slight decline. DATA’s experience in the same years was the reverse—despite posting a population increase of 2.10%, service declined by 7.10% and ridership by 10.2%. The decrease in ridership here is more easily accounted for by the service decline (with the difference potentially made up by the loss of intensively-used routes—though again that requires route-by route data), although in light of the population increase, the fact that the ridership decline is greater than the service decline is interesting.

DATA, unlike CHT, has seem major shocks to its system since then. Between 2007 and 2008, it increased the annual vehicle revenue miles by 17.3%, but only saw a ridership increase of 3.01%. This suggests that an increase in service does not necessarily bring a corresponding increase in ridership, or that the new services took time to catch on. DATA also experienced the opposite between 2004 and 2005, when service was increased by 1.26% but ridership jumped by 52.3%. It has been difficult to find information on what caused those shocks, although possible factors that could have had an influence include any fare changes, gas prices, operational changes, or employer incentives.

Revenue and expenses for these providers also changed between the years 2001-2002 and since then. CHT saw revenue go from $1,751,597 to $1,012,907—a decrease of 42.2%. Before continuing, it is helpful to explain that when the fare-free system was implemented, revenue did not go to zero because UNC-Chapel Hill and the Town of Carrboro both began paying CHT yearly for the free services. These payments essentially replaced the fare and are captured in the “Fare Revenue” section of the NTD profiles. After this, fare revenue went up to $4,117,415, down to $309,722, and back up to $5,357,852. Since then, it has gradually increased every year except for during the recession. There is no clear reason for the decrease to $309,722, which appears to be an outlier, so it will not be strongly considered in the evidence for the cost effectiveness of the program. Overall, the trend for CHT since implementing free fare has been increased “fare” revenue from payments by UNC-CH and the Town of Carrboro. Given the size of the payments by each, it is not surprising that such revenue has increased.

DATA’s fare revenues during 2001-2002 decreased by 7.90% from $2,050,664 to $1,888,826. In the next several years, however, DATA would see a steady increase in revenue marred only by a slight decline during the recession. It is also worth noting that in most years, CHT earned more than DATA in terms of “fare revenue”. This would suggest that the decision by the Town of Chapel Hill to depend on fixed yearly payments UNC-CH and Carrboro turned out to be beneficial for revenue. According to one report, fares collected on buses accounted for only 8% of revenue before the implementation of the free-fare system, which is not major compared to the 42% the fixed payments now account for. By changing from a variable pricing scheme that depended on individuals and trips to a system that relies on two fixed payments from major stakeholders that represent a large number of individuals, CHT achieved important revenue gains.

In light of the data, the implications of fare free service can be considered in two respects: ridership and revenue. First, it makes sense to consider who is using the bus service. According to the 2011 Community Survey put together by Chapel Hill, 50.7% of users ride the bus to get to work. The next largest segment at 27.9% is constituted by people who use the bus to get to school. Following this are social activities, shopping, and medical appointments in that order. There are concerns with this survey, namely that the sample size is very small (353) and that the percentages’ sum is greater than 100%, but it is nonetheless a valuable tool for getting a rough idea of why people ride the bus. Zero car households are also important in driving transit demand, since that is one of their primary means of transit. Zero car households account for 50% or more of the households in most census tracts in Chapel Hill (COA task memorandum). This is certainly true for the areas in which UNC students are housed. Durham also has a significant number of zero car households clustered around downtown, though interestingly enough, Duke’s campus does not appear to have a significant amount of zero-car households. Further information on the demographics of DATA ridership is hard to find, but some inferences can be made based on the availability of other information. First and foremost, by simply looking at a map, it is clear that most bust routes are not convenient to West and Central campuses, and furthermore, the zero-car problem is not very acute, suggesting many have their own cars (Duke has far more parking than UNC, so it’s easier to keep a car on campus). Additionally, Duke mandates that students live on campus for their first three years of school. This diminishes the incentive to use the Durham bus system since all their classes as well as an abundance of food and social options are already on campus. Again referring to maps, DATA seems more oriented towards commuters, with several routes reaching far out into the suburbs, and a high density of routes in neighborhoods with low income and a high number of zero-car households.

Given these data and assumptions about the ridership base, microeconomic theory can be applied to come up with a rough projection of what ridership should look like after a fare change. Utility curves can be assumed to be convex, so as the price of bus transit goes to zero, the quantity demanded should rapidly increase. This assumption is realized in the existing data on CHT when they implemented the free fare system. However, it seems reasonable that there are smaller effects that may mask one another. For example, the demand for transit amongst bus-captive commuters (i.e. zero-car households) will be inelastic—it doesn’t matter if the fare is a dollar or free, they still need to get to work. For non-captive commuters the curve appears to be highly cross-price elastic. Even in the latest community survey, 53.4% out of 607 respondents said they never use Chapel Hill transit—meaning they prefer to walk, bike or drive. But the primary reason for not using transit according to the same survey is that individuals “just prefer to drive” (the survey, of course, having the same faults as earlier but still being a useful barometer). Even when free, transit still suffers at the hands of those who prefer to drive, suggesting that very dramatic cost decreases are needed in order to get these “hardcore drivers” to substitute—decreases that would have to be measured in time and convenience, since the fare is already free. Own price elasticity is more difficult to determine. However, due to the tremendous increase in bus ridership, it can be assumed that this too is elastic. It is harder to find data on students, but it can be assumed that they have an elastic demand curve as well, since so many live within walking or biking distance of UNC, food choices, and social options, with only a few residing further out (grad students). Around 8,900, or almost half, of UNC’s students live on campus, and those that choose to move off generally stay close, as evidenced by density patterns. Because they generally don’t need to travel that far, biking or walking are often substituted for a bus trip, since these are more flexible and very low cost. On the other hand, their curve is not as elastic as that of commuters. Driving is generally not a realistic option, and some do live far enough away from the university that they essentially become bus-captive. Even some who live closer to campus find the bus attractive depending on their destination and schedule, especially if journey is strenuous or they want to travel further off campus to access things such as grocery stores or specialty shops.

Since annual passenger miles, or the quantity demanded, doubled, the change in price is necessary to infer elasticity. Before the implementation of the free-fare system, fares were $0.75 system-wide. With the implementation of free fare, the monetary cost went to zero, but there remained time and convenience costs. Since these are hard to quantify and usually constant, they will not be included. Given these constraints, the price saw a 100% reduction, and a 130% increase in “quantity” consumed. This would suggest that overall demand is in fact elastic. However, this likely masks the fact that demand is relatively inelastic for many in the Chapel Hill community—such as the low income population.

This provides an important example for Durham: Chapel Hill, which has a median income of over $100,000, is far wealthier than Durham, whose median income is about half of that. Chapel Hill has two major factors that allow its demand to be so elastic—a large number of  individuals who both own and prefer to drive their cars (which means they have the economic means to do so) and a large student population that uses the bus system. Durham, on the other hand, has a bus system with a large number of routes in low-income areas and a poorer population over all. Furthermore, student population does not use the bus system as much as that of Chapel Hill. If it is hypothesized that demand for transit in Durham is inelastic because of these factors, then it seems that implementing a free fare system would not significantly increase ridership.

However, there are other reasons that would lend themselves to a fare-free system in Durham. Since the population is lower income, and their transportation costs take up a larger proportion of their budget, a free fare will have a greater impact. A minimum-wage worker will feel the impact of the savings more so than a well-off individual, since he or she makes less money and pays the same fare. This would be one reason for considering a fare-free system in Durham that is not based on estimates of increasing ridership. It is also worth examining where the high-income areas of the county and city are, and how well they are served. Their demand curves will be more elastic, and if they are served well, could generate an increase in ridership.

The implementation of a fare-free system in Chapel Hill had dramatic effects on CHT. Ridership shot upwards even after controlling for population and service growth—and has continued to grow modestly in the last decade. DATA has experienced aggregate growth as well, but in not nearly as astounding proportions and certainly nothing like the jump seen between 2001 and 2002 in Chapel Hill (although there were a few unexplained, moderate increases). Revenue also grew dramatically after the switch to a fare-free system while Durham’s revenue grew much less dramatically in the past decade. Simply looking at this data, it would appear that a free-fare system is a good way to dramatically increase ridership. But using rough data and assumptions about the habits and characteristics of different groups of riders, it is not clear that a fare-free system would drastically increase ridership. This does not mean with any certainty that it would not, but just that there is no reason for believing so based on this data. There could be other reasons for Durham to implement a free-fare system, including social justice concerns, or it may be that the elasticity of demand in Durham is greater than previously asserted due to the growth of the suburbs or a miscalculation in the characteristics of DATA’s ridership. Pursuing this question further would be worthwhile; unfortunately data is difficult to find and often incomplete (one very desirable missing piece is a way to determine how much ridership consists of students). The fare-free system and the elasticity of demand also has important consequences for Chapel Hill, and UNC-CH. Since UNC is the largest provider of funding for CHT, it has to charge students an  $113.50 transit fee (admittedly, this also goes to non-CHT operators). In order to justify such a fee, UNC has to demonstrate that the students get a value from CHT (and the other operators) that meets or exceeds the cost of the transit fee, which is very much like a back-door fare. If most students walk, bike, or drive, then they’re not getting the full value of the fee and UNC would be better off not charging students as much and paying the transit companies less. On the other hand, depending on just how elastic their demand curves are, students will consume a lot more transit at a lower price, and this would seem to justify the decision to charge a fee. The fare free system has also posed additional challenges to CHT, as the Carrboro/UNC funding still hasn’t proven enough to cover costs, and service reductions as well as parking lot fees are being considered.

Both Durham and Chapel Hill have questions to answer regarding fare-free transit. As such transit comes more and more into vogue nationwide, Chapel Hill in particular will be scrutinized. As a neighbor with another major research university, Durham can scarcely afford to ignore the example. And any city that takes such a measure into account must very carefully consider not only others’ experience, but also the characteristics of their own ridership base, which ultimately lend themselves to either the success or failure of such a system.

















CNN. (2013, 04 03). CNN Money. Retrieved from CNN.com: http://money.cnn.com/magazines/moneymag/bplive/2010/snapshots/PL3711800.html

Federal Transit Administration. (2013, 04 03). Retrieved from National Transit Database: http://www.ntdprogram.gov/ntdprogram/

Federal Transit Administration. (2013, 03 31). Fare Policies, Structures, and Technologies: Update. Retrieved from Transportation Research Board: www.trb.org/publications/tcrp/tcrp_rpt_94.pdf

idcide.com. (2013, 04 02). Durham, NC Profile. Retrieved from idicde.com: http://www.idcide.com/citydata/nc/durham.htm

The Transit Cooperative Research Program. (2013, 04 02). TCRP Publications Shopping Cart. Retrieved from tcrponline.org: http://www.tcrponline.org/bin/publications.pl?mode=abstract&cat_id=23&pub_id=1590

Town of Chapel Hill. (2013, 04 01). Retrieved from The Town of Chapel Hill: http://www.townofchapelhill.org/index.aspx?page=1390

Town of Chapel Hill. (2013, 04 02). Retrieved from Town of Chapel Hill: http://www.townofchapelhill.org/Modules/ShowDocument.aspx?documentid=15766

Town of Chapel Hill. (2013, 04 01). Retrieved from The Town of Chapel HIll: townofchapelhill.org/Modules/ShowDocument.aspx?documentid=2

United States Census Bureau. (2013, 03 31). State and County Quickfacts . Retrieved from census.gov: http://quickfacts.census.gov/qfd/states/37/37063.html