Do Crime Rates Affect Housing Prices Research Methods in Economics
Àâòîð: Stephanie Swift Èñòî÷íèê:http://www.bus.ucf.edu/mdickie/Research%20Methods/Student%20Papers/Crime/Swift-crime%20&%20housing%20prices.pdf Abstract Do crime rates affect housing prices? One would think so, but where is the empirical research? There are several factors that affect the price of a home, crime presumably being just one. The purpose of this research paper is to determine if there is truly a link between housing prices and the occurrence of crime in the area where the house resides. Many factors are observed and recorded as possible attributes to housing prices in this research. The setting of this project is Jacksonville, Florida. There are four zip codes observed and the results show that crime, violent and non-violent, has an affect on housing prices. Introduction The question to be researched is whether crime rates affect housing prices. My hypothesis is that crime has a negative effect on housing prices. This question has sparked interest at all levels. Individuals contemplating buying their own home may be interested in this to get an idea of whether or not they are overpaying based on housing characteristics, school quality, and crime rates. As most of the students researching this semester are upon graduation day this may be very useful for all of them because they are more than likely going to purchase their own home in the near future. Also, policy makers are very intrigued with this question due to its tax effects. If they know that crime is related to housing prices and they can figure the exact numerical relation, then they know whether it is worth their time to increase police surveillance of certain areas. For example, if they know that for every 1 reduction in crime observances housing prices are positively affected by 30% then they will more than likely figure it to be worth their time to actively fight crime in order to increase values and maximize tax revenue. Also, people in the real estate profession may find this information very useful when deciding whether or not to list a property with their services. If they know they can receive a bigger commission selling homes in low crime areas then that is where they will go. The main objective of this research is to find whether crime has an impact on housing prices, and if so, how much? It is also important for this research to find other relevant variables that affect housing prices, such as square footage and school quality. By recognizing and accounting for these other variables, if there proves to be a correlation between crime and housing prices, it should not be so grossly overstated. Another objective of this research is to find whether or not it is important to distinguish between violent and non-violent crime when figuring the numerical attribution of crime to housing prices. The remainder of the paper will be dedicated to previous research and the actual research performed this semester with its results. Literature Review In the 1970’s, Daryl Hellman and Joel Naroff did a study in Boston analyzing the impact of crime on urban residential property values. They wanted to focus on the fact that the cost of crime does not end with bodily harm or lost possessions, but that it carries over into other areas. They claimed “changes in the utility derived from living in certain sections of urban areas caused by variations in the crime rate can affect property values and thus property tax revenue” (Hellman and Naroff 105). They composed a model for analyzing the effect of crime on a central city with fixed boundaries surrounded by easily accessible suburbs. They chose to focus on the central city in order to be able to make the Polinsky-Shavell assumption that the urban area is small and open. They also assumed a minimally accepted level of crime and then recorded deviations from this level and their impact on housing prices. They recognized and accounted for many other factors affecting housing prices, such as income, transportation costs, housing elasticities, and neighborhood characteristics. They used the FBI index crime statistics data for their crime rates. They ran a regression analysis and concluded that at every level of crime reduction, housing prices increased. Some of the problems with their study are that they did not take any measures to account for unreported crime in the area. There is no estimation of the number of crimes that were never reported to the police. They also did not determine the difference in value to housing prices due to different types of crime. However, in the next study to be reviewed this problem is accounted for. The biggest problem is that they “have not controlled for characteristics such as plot size and age of unit” (Hellman and Naroff 110). The importance of these factors remains unknown in this study (Hellman and Naroff 105-112). Another study was done by Allen Lynch and David Rasmussen in Jacksonville, Florida, in the 1990’s. They focused on different types of crimes in a hedonic housing model in this study. They claimed, “When there are more serious crimes in the distribution of a given number of offenses, the number of crimes stay the same but the severity of crime rises and, arguably, public safety has declined” (Lynch and Rasmussen 1982). They also claimed that the fear of crime is relevant to house prices. They used two sources to derive their weights for each type of crime: the National Crime Victimization Survey and a study done by Cohen. Cohen’s study was the first to address the topic that paying attention to the types of crimes, and not just the number, is also very important. Lynch and Rasmussen felt that Cohen’s estimates were superior because he showed that the costs for violent crime are higher than that of property crimes. The selling price that they used is regressed on house and lot characteristics, neighborhood characteristics, and alternative measures of FBI index crimes occurring on the police beat in which the property was located. They gathered their housing prices from the Multiple Listing Service of the Jacksonville Board of Realtors. They categorized each neighborhood into a one-mile radius to account for any differences in neighborhoods that were close by. One interesting fact that they found based on the data sets was that most house sales occurred in relatively safe neighborhoods. Two regressions were ran and analyzed. The first regression showed that the number of violent crimes significantly reduced house values, whereas the number of property crimes had a positive and significant impact on the sales price. Therefore, this supported the fact that weighing crimes by a measure of their seriousness is preferred to using the number of reported crimes. They make the assumption that residents in good neighborhoods are likely to report more petty crimes than those in bad neighborhoods. This explained the reason why they found that house values were positively correlated with property values. The second regression that they performed suggested property and violent crime had a small impact on sales price. They compared these findings with those of Hellman and Naroff that suggested crime had a substantially negative effect on housing prices. They also accounted for the fact that some of the house characteristics were more important in unsafe neighborhoods compared to safe neighborhoods. For example, they found that a bigger lot, covered parking, and fenced property was more important to those in a high crime area, because they suggested that this puts distance between the homeowner and the street. They also concluded “although crime does not substantially affect the price of the average home, house values decline dramatically in high crime areas” (Lynch and Rasmussen 1988). They found that houses located in the top two crime areas were discounted at a rate of about 39% relative to comparable dwellings. One of the problems with this study is that they homogenized schools in the areas studied based on the fact that Florida has countywide school districts. However, just because policy is the same in those schools does not mean that quality is the same. This could account for a portion of the discounted house value. They also assumed that household characteristics were similar among blocks. This is not a realistic assumption, because people view crime differently and have different levels of fear. It is not wise to assume these feelings away (Lynch and Rasmussen 1981-1989). Another study on the costs of urban property crime was performed by Steve Gibbons in London in the late 90’s. He hypothesized that there would be “a property or land price gradient between residences in high and low-crime localities” (Gibbons F442). He interpreted that as the individual’s willingness to pay to avoid crime. He accounted for housing prices as being determined by crime in the surrounding neighborhood, property and location characteristics, spatially correlated unobserved characteristics, and a random error variable. The crime considered as being in “the surrounding area” was anything within 250m of the home. Gibbons’ findings were that most of the crimes in the area studied are outside the highest priced districts. A 3.8% decrease in price was found when an additional 5 crimes a year were reported. He arrived at the same conclusion as Lynch and Rasmussen, a positive correlation exists between burglary rates and high price areas. He attributed this to the fact that the higher priced areas are higher return areas for the criminal. Once he accounts for the higher returns to crime he found that burglary rates and home prices were unrelated. He also concluded that criminal damage incidents had a strong correlation to housing prices because of the fear that they induced on individuals. One of the problems to be noted with this study is that Gibbons assumed that criminal damage incidents instill more fear on individuals than burglaries. This may depend on many things that cannot be determined for each individual house due to the time constraints and biased information. Another problem to be mentioned is that fact that Gibbons admitted that there is little prior assumptions as to which home amenities are important enough to warrant data collection (Gibbons F441-F463). Ralph Taylor performed a study during the 1970’s in Maryland on the affects of crime on housing prices. He executed this study because he felt that the previous researchers did not take into account how changing crime influences housing prices over time. He gave credit to a group of researchers under Hakim and Buck, as being the only people to study the impacts of crime on housing prices over more than one point in time. Taylor’s study was influential because he recognized and accounted for the fact that at the same time he performed his study the number of residents belonging to poor neighborhoods was rising. Without this realization one may contend that his results were useless. He regressed the housing values from 1980 on the percentile scale from 1970 and retained the residuals. He determined that the residuals would represent unexpected change. He included measures of crime rates and their unexpected change into the analysis and ran a least squares regression. His results were that neighborhoods decreasing in aggravated assault and murder experienced an increase in their home values. The main problem with this study is that Taylor does not account for measurement errors and he does not allow for two-way relationships (Taylor 28-45). Mark Cohen, a previously mentioned researcher, claimed that some of the studies performed overestimated the impact of crime on housing prices. He felt that some of the neighborhoods experiencing high crime rates might have several factors that warrant data inputs that may lower the value of the houses. The factors he suggested were air pollution, proximity to major highways, and industrial land use. He claimed that researchers did not typically control for those factors. Taylor agreed with Cohen on the fact that, “we also need to include other neighborhood covariates, so that the negative impacts of other conditions are not misattributed to crime impacts” (Taylor 45). He also mentioned that researchers did not typically control for unreported crime. This is a problem that has been mentioned in this review several times for each study. He thought taking this into account would lower the per crime rate affect on housing prices (Taylor 37). What is intriguing about the research in this area is that they all have found a negative impact on housing prices because of crime, in rural and industrialized cities alike. However, the difference in the studies lies in the fact that one took it a step further than the others. The study that separated violent and non-violent crime found that there was actually a positive relationship between housing prices and crime. Another main difference in the studies was the attention to changes over time. Only one addressed the affect of crime on housing prices over time. This study was the only to focus on crime and housing prices as being dynamic. All of these researchers seem to build off of one another and try to correct what the previous researcher failed to do right. Another important aspect of the research is that it includes big industrialized cities, such as Baltimore, and smaller industrialized cities, such as Jacksonville. It also extends research into other countries, such as England. This may be helpful in generalizing the results to a larger area. Regardless of the limitations of each individual study, all of them prove my hypothesis that crime rates have a negative effect on housing prices. Data The data collected for this research paper can be divided into two sections: dependent and independent variables. The Y, or dependent variable, is the housing prices. The X, or independent variables, are square footage, if the house has a pool, the FCAT scores for middles schools that the residents can attend, and the number of violent and non-violent occurrences of crime reported within a two-mile radius of the house. Dependent Variable: Housing Prices The house prices are all collected for the same city: Jacksonville, Florida. Jacksonville was chosen because it represents a medium between metropolitan and rural. The housing prices are collected from random zip codes on different sides of town. The first set of prices consists of houses that are located on the east side of Jacksonville all in one zip code. The second set is comprised of houses on the north side once again from the same zip code. The third set is the south side of town within one zip code. Lastly, the fourth set of prices is the west side of town and all share a common zip code. The reason that it is essential for the houses to share a zip code is because this accounts for some of the neighborhood characteristics that differ among zip codes, such as access to groceries, local parks, etc. The prices are gathered based on the latest assessment of the houses. This information is made public off the city’s web page. The reason that the accessed value is used is because it tends to be a lot more current than sale prices. For instance, if the last time that the house was sold were twenty years ago, then this number would not be accurate because the independent variables are current, and you would need to compare values that are from the same time period at least. The city keeps track of assessed house values from year to year so the information is more accurate. The year that the prices are taken from is 2004. The average housing price for the sample is $62,462.29 Independent Variables 1. Square Footage This variable is pretty self-explanatory. The total square footage for the house is used. There is no delineation made between total and living square footage. Also, the city’s web page is updated yearly so any additions made to the house would be updated at the end of the calendar year. For this reason, data for 2004 is used so that the square footage is the most accurate. Screened in porches are not tabulated in square footage however. They are a separate subsection in the house’s profile that is listed as “extras”. The reason that square footage is used is because this is one of the first things listed in adds for houses for sale. The square footage indirectly captures the number of rooms and bathrooms, so that you do not need to include that in the model as well. The average square footage for the houses sampled is 1493.55 feet. 2. Pool/No Pool The next variable used deciphers whether or not the house has a pool. This is a house specific dummy variable in the regression and is also gathered from the city’s web page. The houses that contain pools are denoted with a 1, and those without are a 0. Pools are included in the regression because it is well known to many people that having a pool increases the value of a home. It also appears as an “extra” in the house’s profile. Another important reason that pools are included is because the analysis is done in the state of Florida, where the summers seem to last all year. Therefore, one could rightly assume that the people in the state place some value on being comfortable and staying cool. The size of the pool or any description on whether it is above or below ground is not included in the profiles. On average, most houses do not have a pool. 3. School Quality School quality is another variable included in the regression. The most common way to measure the quality of the school is by the productivity of its output. The most attainable measure of this for the state of Florida is the FCAT scores of the students. This information is made publicly available via the education and FCAT web page for Florida. Since the FCAT is given to all grade levels and not all could be included it is important to discuss which are used and why. The scores that are chosen for this regression are those of the middle school students ranging from grade 6-8 for 2004. These scores are chosen because they appear to be a good place to look at what the children have learned so far and whether they are being prepared accurately for what is to come. They have spent sufficient time in school already, and they will be attending their last formal years of schooling next. High school scores are the next best to be used, but are not used simply because not all students are prepping for continued learning. Also, all of the magnet schools in Jacksonville are located in one area of town with only one average high school in the same area. This may tend to overstate the FCAT scores for that area, so high schools are not used. The housing values are based on zip code in order to have uniformity for school quality as well. The school districts in Jacksonville can be divided amongst zip codes; therefore the district for each code is identified. Then, all of the schools within each district are gathered along with each schools FCAT scores. There is a score for math, reading, and writing. They are all based on a scale of 500 and averaged so that one score could be used for each zip code. The average FCAT score is a 308.43. 4. Violent Crime Occurrences This information is by far the most tedious to gather. Since most crime statistics are gathered based on FBI data, there is a great challenge in getting such micro level crime occurrences. The FBI only denominates crimes on a citywide basis. This information is not specific enough because crime is surely not the same for the entire city, the same way that school quality is not. Therefore, through a friend at the Jacksonville Police Department, a department web page was located that gave specific crime data based on a numeric radius. For each zip code the violent crimes occurrences were summed into one number for the entire calendar year of 2004. All of the occurrences were within and around a two-mile radius of the zip code, including all crimes within the zip code as well. The zip codes were separated far enough apart so as not to overlap. Violent crime occurrences include such things as sexual battery/assault, battery/assault, murder, and lewd and lascivious acts. All of the figures are based on reported crime to the Jacksonville Police Department. These numbers are not percentage numbers and are based directly on data inputted by the department. The average number of violent crimes is 39.15 occurrences. 5. Non-Violent Crime Occurrences The data for non-violent crime is gathered in much the same way as the data for violent crime. The main difference lies in what is included in non-violent crimes. This section includes robbery without a lethal weapon, larceny, trespassing, and vandalism. The occurrences observed here are greater in number than that of violent crime, indicative of the fact that either more of this crime takes places, or people are just more likely to call it in. That is the main reason for the separation of the crime. It is easier to prevent victimization of non-violent crimes than violent ones. Homeowners could install security systems, gates, or employ large dogs on their property to deter these types of crimes. However, the victims of violent crime are usually much more random and such safety measures cannot always reduce your chances because these acts are not taken against your home, but against you. Also, the effects of non-violent crime are usually issues that can be settled through monetary compensation, whereas with violent crime the stakes are usually a lot higher and most of the time cannot be completely settled through monetary compensation. Due to all the differences between the two categories of crime, each warrants their own analysis. Another reason for the separation of the two types of crimes is the fact that I want to eliminate the correlation between the occurrences. For example, the violent crimes go up about the same percentage amount that the non-violent ones do for each area. The average number of non-violent crimes is 84.88 occurrences. Methods The methods used in this research paper begin once all of the data collection is accurate and complete. There are two simple regressions ran, due to differences in types of crime that have been explained above. The first regression to run utilizes all of the data collected on housing prices complemented by their square footage and whether or not the houses have a pool. The housing prices are ran as a natural logarithm of their value in order to make the numbers more manageable. Then the data for school quality is matched with the appropriate zip 14 code and included as part of the regression. Lastly, the violent crime occurrences are entered in alignment with the zip codes for which they apply. The second regression consists of the same variables with the exception that the non-violent crime occurrences are substituted for the violent occurrences. The regressions are ran using the excel analysis program after the data is entered into the system. Works Cited 1. Cohen, Mark. “A Note on the Costs of Crime to Victims.” Urban Studies v.27 5pp 1990 Florida FCAT homepage Íàçàä |