Town and Gown in Worcester, MA

Town and Gown in Worcester, MA:

Measuring the Distance to Colleges and its Effect on House Prices

By

Gael Carter

December 2010

I. Introduction

In a recent article in the Worcester Telegram and Gazette (Kotsopoulos, 2010), the citizens of Worcester condemned local college studentsfor holding out of control off-campus parties. In the weeks following, many students from various colleges were arrested, including students from College of the Holy Cross and Assumption College, at various off-campus locations. In the news, it seems that the neighbors who live close to the colleges are the most directly affected by student behavior, but do the homeowners in neighborhoods further away from the schools feel the same effects? This study aims to answer the question:“To what extent does the location of a house, in relation to a Worcester college, affect the price of the house?”This paper also focuses on the separate effects of the colleges and universities in the city and their differing effects on the sales price of a house.

In addition to the stereotype of wild, drunken behavior, students living in overcrowded off-campus residences also cause increased traffic and parking issues from student ownership of cars. Although it is easy to only see the negative impacts of living close to a college campus, there are many positive externalities as well. Colleges often host free cultural events such as speakers, art exhibitions, and movies on campus. They also hold athletic events, run summer camps, and allow access to the facilities such as libraries, swimming pools, tennis courts, and many more. Positive effects can also be seen in the community through student participation in community outreach programs.

This balance of contradicting externalities has been referred to throughout history as “Town and Gown” (Mayfield, 2001). The term “town and gown” dates back to medieval Europe identifying the town as the lay people in the communities and the gown as the universities. This term refers to the separation of town and gown and also implies conflict between the two groups.In 1862, the Morrill Act was passed in the United States, establishing a land-grant for colleges who provided a public service in return for federal aid (Mayfield, 2001). In this situation, farmers would identify an improvement, need, or specific problem and the experts at the university would solve the problem for the farmers. This became one of the first of many kinds of relationships between universities and the surrounding communities.Martin, Smith and Phillips’ (2005) paper explains that the separation of the universities in America began because of geographical separation from the larger towns and cities. Universities were usually located in rural areas and were often secluded. However, as the urban areas began to expand, the universities found themselves in the midst of the economic and social problems of the community around them. Throughout the history of American universities, there have been examples of successful college and community relationships, but for the majority of college towns the relationship continued to decline(Martin, Smith and Phillips, 2005).

Ten years ago, the city of Worcester faced a similar situation in its local college neighborhoods. The debate at hand is whether or not colleges positively or negatively impact the community. In my paper, I use the distance of a given house from the nearest college or university in order measure a potential effect on the sales price of a house and if the effect is positive or negative. I also examine the schools individually to test if there is a difference in the effects between the diverse types of colleges in Worcester.

This study finds that that the location of a house in relation to a college in the city of Worcester is significant. This distance is significant with and without college differentiation. This effect is also significant when the distances for the different colleges are controlled for, and has a greater explanatory power than the original hedonic regression with the variable of interest being the distance to any college. This implies that the differences in the colleges impact local house values in Worcester, holding distance and all other neighborhood and house characteristics constant.

II. Literature Review:

No study exists that exactly examines the impact of colleges in a “college town” on house prices. However, the recent study conducted by Vandegrift, Lockshiss, and Lahr (2009) examines the value of a college on house values at the aggregate level. Their study looks at the entire state of New Jersey and they test to see if the presence of a college within the borders of a municipality causes an effect on home municipality tax and also tested to see if the presence of a college leads to higher house prices. They utilized a version of the hedonic pricing model to regress their data, which is explained in the Hedonic Model section of this paper. The presence of a college in the town corresponds with a tax base that is about 24% higher than that of a town in New Jersey without a college. This implies that the presence of a college within a town border causes a significant positive effect on the tax base per acre. This effect is greater in towns with a four-year college than towns containing a community college. They also conclude that the variables representing the size of the college and the degree to which the college is residential have virtually no effect on the tax base of the town. Vandegrift, Lockshiss, and Lahr also found that the presence of a college in a New Jersey town was associated with house prices that are 11% higher. They discovered that smaller colleges have a larger positive effect on house values and this effect disappears when the enrollment of any particular college reaches 12,500 students. This implies that while smaller colleges have an impact that increases house values in a town, once the enrollment of the college surpasses 12,500 students; this effect reverses to a negative impact on house values. Similar to the influence of the tax base, the effects of a four-year college are higher than that of a community college and the difference is derived from the degree to which the college is residential. That is to say that a school with a larger percentage of the student body living in off-campus housing would have a greater negativeeffect on the prices of the surrounding houses in comparison to a college that has the entire student population in on-campus housing. They conclude that the overall effect for small sized, four-year colleges is a positive impact on house values and the tax base per acre. The effect is similar, but to a lesser degree in two-year colleges and the effect is negative for large schools.

Other studies deal with the effects of environmental externalities, open space, and the quality of primary and secondary schools on house values. In Kiel’s(1995) research of the impact of hazardous waste sites on house values, she examines the distance to the closest hazardous waste site in the town of Woburn, MA.In Woburn there are two different hazardous waste sites. Similar to my process, Kiel used the hedonic approach using variables such as price, finished area in square feet, date of sale, year built, architectural style, and minimum distance from house to the Superfund sites. In order to measure the effect of distance from the Superfund sites on house values over time, the data were grouped by time periods. For example, the first period was prior to announcements in the media about the Superfund sites, the second time period represents the discovery of these sites, and the third time period represents the time period in which the EPA added the sites in Woburn to the Superfund list. The possibility of cleaning up the sites was made public during the fourth period, the fifth period was when the official cleanup plan was announced, and the final period is when the cleanup actually began. The regression results show that the information released on the toxic sites did impact the house prices. Unlike the work of Vandegrift, Lockshiss, and Lahr, Kiel’s approach allows for the data to be viewed at a local level to gain the specific effects of the Superfund sites in Woburn over time.

Downes and Zabel (2002) quantified the impact of school characteristics on house prices in Chicago from 1987-1991. In this study, they combined information from the American Housing Survey and the Illinois School Report Cards to form their own data set. They also used the hedonic method to estimate their findings. Due to the complexity of the data set, Downes and Zabel used 33 variables in their regression and found that in the district-level results, the per-pupil expenditures and test scores have similar positive impacts on house values. The school-level results implied that potential buyers respond to the racial composition of a school when they are deciding the price that they are willing to pay for a given house.

My study looks specifically at the city of Worcester and the twelve colleges and universities that it contains. The information that I have collected is on a microeconomic scale. Therefore, the effects of the colleges on the surrounding house prices will be much more accurately identified than in Vandegrift, Lockshiss, and Lahr’s paper. In Worcester the tax base will be constant for the entire city, so I do not include that variable in my study and I will simply focus on house sale prices from the year 2000 and the census level data from the corresponding year. Similar to their study, I also use the hedonic house price model in which I include variables such as percentage of students in off-campus housing, number of undergraduate students, and number of graduate students in my regression in order to account for differences in the various schools that I include in my data set. Based on the study conducted by Vandegrift, Lockshiss, and Lahr, the negative effects of the colleges on house values appear to be on a more local level, or within the given municipality. In their regressions, they used municipal-level data for the entire state of New Jersey to look at the effects on an aggregate level. The positive effects are conspicuous at the aggregate level in the state of New Jersey. However, by looking at the sale prices of single family residences and their proximity to the nearest college, I am able to get a more accurate idea of the positive and negative impacts caused by the colleges on the house sales price.

My economic model is very similar the model from Kiel’s paper. This study uses the the same house price hedonic method, and the price of the house as the dependent variable.However, I use distance as the variable of interest in my regression analysis. Based upon the assumption that effects of colleges on house prices would change very slowly over time, I only use data from a single year in order to determine a generalized effect of the colleges on house sale prices. Thus, by substituting in the distance from a given house to the nearest college in Worcester, I am able to see the effects on the year 2000 sales price of the house.

Downes and Zabelpresent an interesting way to connect house prices and schools, but what I have done is different in nature from their study. The difference lies in the fact that homeowners who are choosing a home near a school are choosing based on the school itself. People buy houses located in good school districts because they will most likely have children to send to the schools. The purchase of a house near a particular primary or secondary school permits enrollment in the schools. On the other hand, the purchase of a house near a college or university is different because it does not guarantee enrollment in that particular college. For most colleges, in-state tuition is much less, which could a potential draw to living in close proximity to a certain college but not the sole reason. Also, the funding sources for colleges are much different than those of primary and secondary schools, who draw from local taxes for funding. Thus the decision to live near a particular college would not stem from the desire for the homeowner’s children to attend the college. However, since the study conducted by Downes and Zabel proves that potential home buyers take school districts into account, it is reasonable to assume that the school district that the house is located in is a neighborhood effect that would affect the price of the house. They also include neighborhood variables such as the natural log of the median income in the house’s census tract, median age of individuals in the house’s census tract, proportion of nonwhite individuals in the census tract, and the proportion of blue-collar workers in the census tract, along with many other variables. These neighborhood variables are important for the hedonic method because they control for the differences in different parts of the city, which affect the house prices. I also use census level data for neighborhood variables in my regression, but my houses are divided up by census block groups rather than census tracts.

III. Data

In my research, I used information from the single family house sales from the year 2000 in the city of Worcester provided by the Warren Group. The Warren Groupis a New England based company that provides its clients with comprehensive studies, based on detailed property and market information collected over time( The data set from the Warren Group initially contained 1450 observations of houses. Some of the houses in the data set were located outside of the Worcester city border; therefore they were eliminated from the usable data. In the column for house sale prices, some of the sale prices seem abnormally low. This low price may be a result of houses that were sold within a family for a price below market value. On the other hand, some of the houses were significantly above the mean house price. In order to correct for these outliers to some extent, I dropped the five lowest and the five highest sale prices. Another problem with the data set arose in the column that described the year in which the house was built. Thirty three of the houses in the data set were reported to be built after the year 2000. However, since the data describe houses sold in the year 2000, it is impossible for the homes to be built in any year after the sale year. These houses have removed from the data set.After cleaning the data set, 1394 observations remain. The number of observations remains constant throughout the series of regressions.

Similar to Kiel’s (1995)house price hedonic method in Woburn, MA, I originally had wanted to include variables such as lot size, interior square footage, number of floors, renovation year, style of the house, type of heating fuel, central air conditioning,basement designation, basement area, number of parking spaces, and if the house sale includes a garage. Unfortunately, the data from the Warren Group was limited in the house descriptive variables and only contained information on price, the year in which the house was built, the total number of rooms, number of bedrooms, the number of bathrooms, lot size, and if the house is mortgaged.

Due to the restrictedsupply of house characteristic and information variables, I rely heavily on neighborhood information. In order to incorporate neighborhood statistics, I layered my house price data set with Census level data from the year 2000, collected from the Massachusetts Office of Geographic Information (MassGIS) website. I usedCensus data at the block group level, and each of the remaining houses in my data set was sorted into the Census block groups of the city of Worcester. The U.S Census Bureau defines “A census block group [as] a cluster of census blocks having the same first digit of their four-digit identifying numbers within a census tract” ( Therefore, the block groups are smaller than the census tracts and contain more specific neighborhood characteristics that I control for in my regression analysis. In general, census block groups contain any number of observations between 600 and 3,000 people, with an optimum size of 1,500 people; see Map 1 for a complete map of the block groups of Worcester, MA.

The information in the Census data that I collected included the total population, the median value of owned houses, and the median household income for each block group. I also included variables on the percentage of the population of the block groups broken down by race, age, high school graduate, college graduate, and others. Similarly, I included percentages of the total houses in each block group for variables such as occupied owned homes, occupied rented homes, the type of power in the homes, rented and owned homes without full plumbing, and the rented and owned homes without full kitchens. For a full list of variables and descriptive statistics, see Table 1. I ran a correlation matrix of all of my independent variables, a few notable correlations exist between the following neighborhood variables: the total population in a block group and the total housing units in a block group, the median house value of a block group and the median household income in a block group, and the percent of owned occupied houses and the percent of rented occupied houses. The median household income for a given block group is also highly correlated with the percent of owned occupied houses and the percent of rented occupied houses. The house characteristic variables for the total number of rooms and the total number of bedrooms were also highly correlated. The total number of rooms and the total number of bathrooms is also correlated but to a lesser degree than the variable for bedrooms. These correlations between the house characteristic variables are logical because the size of the house depends on the total number of rooms, a fraction of which are made up of bedrooms and bathrooms. I kept all of the variables in the regression because removing them would have caused omitted variable bias. All other variables had correlations of less than 0.5, and a correlation matrix of the significantly correlated variables is located in Table2.