Home » 2015 Categories » 2015 Term Paper » Effects of Grocery Store Openings and Closings on Durham Housing Prices by Li Ding

Effects of Grocery Store Openings and Closings on Durham Housing Prices by Li Ding

Introduction and Background

Extending my previous work, this paper explores the effects of opening and closing food retail stores on nearby housing prices in Durham, North Carolina. While the opening of a new grocer provides nearby residents with increased access to food and other goods, the full economic impacts of new stores are not always clear. In particular, there is often resistance to larger retail stores like Walmart being opened in a neighborhood, with arguments that a new supercenter will negatively affect local businesses and wage levels. Additionally, increased traffic, crime, noise pollution, and so on are also potential concerns regarding new stores. As such, it is not apparent how, if at all, new stores affect nearby home prices. Changes in housing prices can be used as a signal to evaluate these various competing effects and so the results of this study will help Durham residents and government officials evaluate whether to support or oppose new stores in their local areas.

Pope and Pope (2014) examined the effects of opening a Walmart on nearby home prices from 2000 to 2006, and this paper will use a similar framework to their study. Using a difference-in-differences specification, the authors found that houses located within 0.5 miles of a new Walmart saw an increase of 2-3 percent in sale price in the following two and a half years, and houses 0.5 to 1 mile away saw a 1-2 percent increase. These results would suggest a positive impact for new stores in Durham as well, though the authors are careful to note that their results only reflect a national average.

Publicly available home sale data for Durham extends ten years back (to 2005), so this study will analyze store openings and closings that have occurred in that time frame. This includes Walmart, Harris Teeter, TROSA Grocery, ALDI, and Save-a-lot stores. Stores like Target and Food Lion are not included in the study because their newest stores were opened 2004 or earlier.

Housing Data

This paper analyzes data provided by the Durham County Tax Administration on home sales in Durham, NC. Through the county’s online record search, I collected the electronic summary record of each single-family residential home sale in Durham since 2005, the earliest year for which data is provided. These searches yielded approximately 27,000 results, and an Excel VBA macro was used to access the details page for each sale and record characteristics of the property (bedrooms, bathrooms, and so on). This data were filtered to remove a small number—around one percent—of entries with incomplete information.

Next, I used the Rest-CSV interface provided through Geocoder.us to find the longitude and latitude of the homes in the dataset. A macro was written to query the Geocoder.us servers and record the geographic coordinates for the homes included in the service’s dataset. Geocoder.us uses TIGER/Line street and highway information provided by the US Census Bureau from 2004, and as a result, not all the addresses were able to be geocoded. Approximately 10,000 of the 27,000 home sales were not able to be located on the 2004 TIGER/Line maps; these are homes that were built on streets that did not exist a decade ago. This introduces a significant potential for bias in the data, and a re-examination of this paper’s analysis with fully geocoded data may prove insightful. Summary statistics for the full and geocoded datasets are provided below.

Table 1: Summary statistics for the full and geocoded datasets


As expected, the geocoded dataset contains homes that are older, and as a result this reduced dataset contains homes with a lower average sale price, heated area, number of bedrooms, and number of bathrooms. The below graph and table show the data in both datasets distributed by sale year, both in terms of absolute sale numbers and as a percentage of the overall dataset.

Figure 1: Distribution of sales by year for the full and geocoded datasets


Table 2: Distribution of sales by year for the full and geocoded  datasets


Even though the geocoded dataset skews older and cheaper, the proportion of sales occurring in each calendar year is still roughly the same as the full dataset, indicating that the geocoded dataset is still very similar temporally to the full dataset. The effects of the subprime mortgage crisis on Durham home sales can be seen in both datasets; sales in 2008 and the years following are clearly depressed compared to 2005 through 2007. There is also evidence for a strong recovery in the past few years, and 2014 sales were at approximately the same level as the 2006 and 2007 peak.

We now run a hedonic regression on the logarithm of sale price for the geocoded data to examine the dataset. The regression used is below:

li_eq_1The sale price, heated area (in square feet), and bedrooms term are straightforward. The β3 is for the “bedroom/bathroom differential” which is the absolute value of the number of bedrooms minus the number of bathrooms for a house. This was motivated by a model developed by the National Association for Home Builders (NAHB) that aims to estimate home values. As Emrath (2006) describes, their data indicated that home buyers tend to prefer a rough balance between the number of bedrooms and the number of bathrooms. If a home has more bedrooms than bathrooms, an additional bathroom will increase the value of the home by a higher percentage than if the differential were smaller. The β4 is associated with a dummy variable for whether a property has a garage or not. The results of this regression are below.

Table 3: Regression on the geocoded dataset


The p-values for each of the variables is virtually zero. All else equal, an additional bedroom will add 15.3% to a home’s value and a garage will add 13.4%. The previously discussed bedroom/bathroom differential shows a fairly strong negative effect as expected. The NAHB model predicted that an additional full bath would add approximately 20 percent to a home’s value which is very consistent with the derived 19.8%. Heated area is measured in square feet and so an additional square foot has a very small positive effect as would be expected.

To make sure that there were no abnormal price trends present in the dataset, I established a price index using the data. This index is graphed below along with the Case-Shiller national home price index and the Case-Shiller home price index for Charlotte, NC, the nearest metropolitan area to Durham for which an index is provided. For each month, the Durham index is calculated as the average price per square foot of single-family residential home sales in the past three months; the Case-Shiller indices use a three-month moving average as well. The Case-Shiller indices use repeat sales of the same home to track changes, but the Durham market is not large enough to create an accurate index using this same method. As seen in the graph below, however, the price per square foot measure for Durham roughly approximates the Case-Shiller indices, suggesting that Durham home values have followed the same trends as other North Carolina cities (Charlotte) and the nation as a whole. It should be noted that it is merely a coincidence that the Durham index is similar in magnitude to the Case-Shiller indices.

Figure 2: Indices for home prices


Store Data             

This study analyzes the effects that the openings and closings of Walmart, Harris Teeter, TROSA Grocery, ALDI, and Save-a-lot grocery stores have had on nearby housing prices. The locations to be examined are listed below along with relevant dates. Other grocery stores can be found in Durham today, but the previously listed stores are the ones who have opened or closed stores in the past decade. These locations were manually geocoded using Bing Maps.

Table 4: Locations to be studied and relevant dates



The distance between each home and store was found using the Haversine formula which gives the great circle distance between two points on a sphere:


d = the distance between the two points

r = the radius of the Earth (3961 miles)

lat1, lat2, long1, long2 = the latitudes and longitudes of the two points

Borrowing from Pope and Pope (2014), the hedonic regression used was:


The logarithm of the sale price of a home can be explained through its property characteristics as previously defined and its geographic relation to a given store. Three indicator variables (D0.5, D1, D2) represent whether the given house is within 0.5 miles of a store, between 0.5 and 1 mile, or between 1 and 2 miles, and whether the home is between 2 and 4 miles away is the omitted indicator. Holmes (2011) in his research on Walmart considered a radius of two miles to properly constitute a store’s neighborhood, and this study will do the same for all the stores. The last group of homes 2 to 4 miles away will act as a control group for the homes inside the neighborhood that we are interested in examining.

The variable Post represents whether the sale occurred after the opening or closing of the given store. A sale occurring before the relevant date will have the relevant indicator (D0.5, D1, D2) set to 1, and the interaction between the Post term and the indicator variables inside the parentheses will lead to the corresponding indicator becoming zero. A sale after the relevant date will have both indicator variables at 1, and so the estimates for the second set of spatial estimators ( ) will provide information on whether the store impacted housing prices. This study uses home sales two years before and after each store’s opening or closing.

The difference-in-differences specification is used in order to remove the effects of omitted time-invariant variables that could bias estimates. Comparing homes sold before and after the store opening or closing allows one to more safely disregard neighborhood characteristics that would influence housing prices since these characteristics should more or less influence both the pre and post home sales in the same way. Holmes (2011) created and argued for a model in which Walmart stores are not placed based on characteristics of the nearby markets but rather on distribution and shipping costs. This would imply that the less than two and two to four mile ranges should be similar enough to properly control for broader housing market trends. For time-related variables, Figure 2 presented previously shows that the price per square foot for single-family residential homes in Durham has remained within a $20/sqft range for the past decade. While this is not an insignificant fluctuation, this index shows a reasonably flat trend as a whole so it is unlikely that including time-related variables would significantly influence the analysis.



The results for the first Walmart store located at 1010 Martin Luther Jr Pkwy showed a positive 14.4% effect on homes 0.5 to 1 mile away and positive 8.3% effect on homes 1 to 2 miles away after the store was opened. The Walmart store located at 1525 Glenn School Rd showed no significant effects at any distance, both before and after the store opened. This means that the Walmart was placed into a neighborhood which was similar price-wise to its surrounding area and that the opening of the Walmart did not influence nearby prices. It should be noted that this Walmart featured fewer home sales in its relevant subset of the dataset; its opening was in 2009 during the housing crisis when fewer homes were being sold, and it is located in the northeastern part of the city which is less dense. This Walmart had 1343 sales within four miles in its four year window compared to 2281 for the first Walmart store. This smaller dataset perhaps contributed to the lack of significant coefficients found in the regression.

The Walmart store at 3500 Roxboro Rd closed around the same time the previous Walmart opened. Before the closing, homes 0.5 to 1 mile away sold for 12.4% more than homes 2 to 4 miles away and homes 1 to 2 miles away sold for 23.0% more. The coefficients associated with the post-closing indicator variables were not statistically significant, so the closing of the store did not change nearby housing prices from the previous baselines. Only one of the three Walmart stores openings/closings in Durham caused nearby housing prices to change, indicating that the positive average effects found in Pope and Pope (2014) are not necessarily applicable to the stores local Durham area.

Harris Teeter

The three Harris Teeter store openings had significant results in terms of the neighborhoods they were placed in though none of the three showed any significant effects on nearby housing prices after the stores were opened. The Harris Teeter store at 2017 Hillsborough Rd was placed in a location where houses 1 to 2 miles away were worth 9.3% less than homes slightly farther away. The 1501 Horton Rd store was placed where homes 0.5 to 1 mile away were worth 10.2% less and homes 1 to 2 miles away were worth 9.2% less. The 1125 West NC 54 Hwy store was placed where homes 0.5 to 1 mile away were worth 11.8% more and homes 1 to 2 miles away were worth 8.5% more. While the coefficients are similar in magnitude, the differing signs suggest that there was no deliberate effort by Harris Teeter to locate its stores in certain kinds of neighborhoods. The regressions also show that none of the three stores had an impact on surrounding housing prices after they were opened. Homes near the first two Harris Teeter stores were still worth approximately 10% less and homes near the third store were still worth about 10% more.

TROSA Grocery

TROSA, Triangle Residential Options for Substance Abusers, Inc., opened a grocery store in East Durham, at 2104 Angier Ave, in 2010. East Durham is a significantly poorer area of the city, and the regression quantified this economic difference. Before the store opened, homes within 0.5 miles of the location were worth 74.7% less, homes 0.5 to 1 mile away were worth 70.0% less, and homes 1 to 2 miles away were worth 53.6% less as compared to homes slightly farther away. After the store opened, prices of homes within 0.5 miles fell 65.3% further, and homes 0.5 to 1 mile away fell a further 24.3%. Due to low traffic and high operating costs, the store closed in 2012. As expected from the previous regression, before the closing, the homes near the store were worth significantly less than homes farther away. After the store closed, home prices near the store actually increased; homes less than 0.5 miles away experienced a 35.9% increase from the previous baseline, and homes 1 to 2 miles away experienced a 33.3% increase.

These results do not necessarily mean that the grocery store exacerbated the poor economic conditions of the East Durham area. The store was located two miles from downtown Durham, an area that has seen significant investment and growth in recent years. Since the control group for the regression uses home sales two to four miles away, increasing prices in the downtown area would lead to homes within two miles becoming worth relatively less. While the grocery store was the first in fifty years to open in the neighborhood, offering nearby residents a source of fresh produce within walking distance, these results show that any positive effects that this increased food access provided were not significant when comparing East Durham’s poor economic conditions to downtown.


The ALDI store that opened at 7906 NC 751 South showed no significance for any of the distance indicator coefficients. This store is located near Southpoint Mall, and so it is unsurprising that the store opening did not change nearby housing prices given the very large number of shopping options already available in the area.


The regression for the Save-a-lot store opening was unique in that it showed significance for all of the distance coefficients. Before the store opening, homes within 0.5 miles were worth 75.6% less than farther homes, homes 0.5 to 1 mile away were worth 78.8% less, and homes 1 to 2 miles away were worth 8.4% less. Save-a-lot is located about a mile away from downtown Durham in the same East Durham area as the TROSA Grocery store was, and so these results make sense considering the much higher value of homes downtown compared to East Durham. After the store opened, homes within 0.5 miles experienced a 45.4% price increase, homes 0.5 to 1 mile away experienced a 33.2% increase, and homes 1 to 2 miles away experienced an 11.4% increase. It is unlikely that the opening of a single discount grocery store single-handedly led to these dramatic price increases, and so these results need to be interpreted with caution. We should note that the closing of the TROSA Grocery store led to similar increases; both the opening of the Save-a-lot and closing of the TROSA Grocery occurred in 2012, and we can hypothesize that East Durham experienced some sort of revitalization during this time that led to these results.


This paper shows varying results for the impacts of store openings and closings in Durham. While Pope and Pope (2014) found that new Walmart stores slightly increase nearby housing prices, this study did not find conclusive support for their conclusion. As a whole, the results for the different stores were difficult to interpret, and no retailer as a whole showed a definitively positive causal impact. This study shows that while there may be beneficial trends on a larger scale from introducing a new store into a neighborhood, the impacts of individual stores on home prices can be difficult to predict.


Emrath, P (2006). “How Much is a Bathroom Worth?” National Association of Home Builders. Retrieved from http://www.nahb.org/generic.aspx?genericContentID=62422.

Holmes, M. (2011). “The Diffusion of Wal-Mart and Economies of Density.” Econometrica 79: 253-302.

Pope, D. and J. Pope (2014). “When Walmart comes to town: Always low housing prices? Always?” Journal of Urban Economics 87: 1-13.


  1. Great paper Li! Very much enjoyed the analysis and specifically the store-by-store breakdown and their unique effects on home prices. It is always hard proving causality but I think your analysis of looking at store openings and closings and their time correspondence with home prices, is sufficient in proving a degree of causality because there are definitely other forces at play. An interesting extension of this paper could be the inclusion of customer volume at a store and its change as one store closes and another opens at the same location. By doing so you can see how people in surrounding neighborhoods directly value one store over another. That data might not exist but it would interesting to see in numbers how much or how little one store brand is valued in comparison to another. In theory, that could then explain with greater certainty, the causal effect of a new store on home prices because you would be able to see the change in popularity/value of a store and compare that to changes in home prices. If a relationship between the two variables exists, then I believe a stronger degree of causality can be proven.

  2. I’ve thoroughly enjoyed your term paper and term presentation. Proximity to grocery stores is one of the things that’s very important to me in my personal search for housing (though not quite a potential home buyer yet), and I imagine many others—especially those with limited access to private transportation—share this affinity. As you mentioned, it’s difficult to disentangle all the different effects given the two way causality of home prices and possibility of grocery stores wanting to locate in neighborhoods with certain levels of purchasing powers, which would in turn be correlated with housing prices in that neighborhood.
    An interesting extension of the paper might be to consider the actual distance between each house sale and grocery store instead of using the by mile benchmark or modeling time of sale as a continuous variable (perhaps with the month of sale being 0) instead of indicator variables for whether the house sale occurred before or after the store opening. Accounting for houses that are close to more than one grocery store may also be worthwhile—perhaps the marginal benefit of additional grocery stores decline, perhaps they counteract the effects of one another. Very interesting analysis overall, I’d be very interested in seeing this analysis contextualized with proximity to other amenities (restaurants, coffee shops, banks) and possible interaction effects.

  3. I really enjoyed this paper, especially because I wrote my Durham paper using similar methods. For my paper, I specifically studied the effects of university on Durham apartment prices. So, your paper illustrated how hedonic housing model can be applied to a different topic. That being said, one suggestion I have about your hedonic housing model is that you could have used different variables as well. For example, you could have used # of bathroom, sqr footage and garage by just using one combination variable since it seems natural that they are very correlated.
    Also, I liked how you used various cases to compare and contrast since many actually yielded different results from each other. The results (about how an increase in sale price in a few years after a new store is built in the area signals that it is significant) could be more convincing if you analyze other economic effects that might have influenced housing prices at the time. For example, an increase in sale price could have been merely due to some good news in the area that the store was built or just stronger economy. Overall, I think this is a very analytical and informative paper!

  4. I think it would be interesting to use a similar framework to test Li’s hypotheses with different kinds of stores. Larger grocery stores (particularly Walmart) tend to be locations that people expect to commute to from longer distances than other kinds of stores, whereas other kinds of retail establishments are expected much closer to home. It would be interesting to add facilities such as quick service restaurants, coffee shops, or gas stations into the mix to see what effect these businesses have on nearby house prices.

    With many new housing developments slated for construction in the Research Triangle over the next 20 years, an analysis such as this could help developers decide which tenants to have in retail space that they may be incorporating into residential development projects. Various multi-family residential developments in Raleigh have retail space on the street level; I imagine the presence of a Starbucks or Chipotle Restaurant in such spaces may reap higher rent rates in the residential units above than if these retail locations were further away.

    It is certainly frustrating that some supermarkets in the analysis didn’t seem to have any effect on housing prices. In an analysis of local residential developments under construction aimed at first-time buyers, I also found developments further from supermarkets at a discount when compared to those with supermarkets closer in proximity. However, these closer developments were also closer to other facilities such as shopping malls and restaurants, so this may explain the lack of consistency with an analysis focussing solely on supermarkets. Newly constructed supermarkets tend to be part of shopping centers that have many retail outlets next to each other, so an extension of this paper would be to add in provision to account for these different kinds of facilities. I suspect that certain types of retail outlets will have a more significant effect on housing prices than larger big-box style supermarkets in some areas.

  5. Great paper and really interesting findings. I would not expect the opening of most stores like WalMart or Harris Teeter to significantly affect nearby housing prices in middle class neighborhoods. Personally, I don’t derive a lot of utility from being within 1 mile of a store versus within 5 miles because I would normally drive to a grocery store anyway. I would imagine that this would be the case for many people with cars.

    However I did find it particularly interesting that the Save-A-Lot caused such a huge difference in home prices. It seems intuitive to me that poorer neighborhoods would be more affected by the opening of stores because more of the residents would be walking to stores rather than driving. I think you did a great job of picking out specific case studies and varying the type of stores that you looked at.



Leave a comment

Your email address will not be published.