Desirability
How do POPS vary in value?
definition
We investigated the correlation between residential sales data and accessibility to POPS and parks within a fixed time period. This allows us to suggest whether residential areas of a certain range of values are more desirable for building POPS.
​
hypothesis
We hypothesized that accessibility to POPS would be positive correlated to residential sales values, and that this correlation would be much stronger than for accessibility to parks.
​
method
We used the gravity accessibility metric in the UNA toolkit to measure access to 343 POPS and 2001 Parks from 7826 residential location points extracted from a 2012 New York City real estate dataset (assuming 1000m walkshed radius, and a 0.04 beta factor).
We controlled for 19 other factors, including apartment size and access to transit, restaurants, shopping areas, and other attractive destinations, to isolate the effects of access to POPS and parks on residential sales values.
We ran two models in our analyses:
1. We first ran an OLS regression model to test the general correlation between access to POPS and parks and residential sales values.
2. We then ran a Spatial Lag Model to potentially improve fit adjusting for spatial clustering dependency.
Findings
-
Proximity to POPS is a good indicator for housing values. Our regression models show a positive but insignificant correlation between access to parks and residential sales values (Spatial Lag Model: Coefficient > 0; p = 0.32798). In contrast, access to POPS is positively and very significantly correlated to residential sales values (Spatial Lag Model: Coefficient > 0; p = 0.00012).
-
Only a very small number of apartments are located reasonably close to existing POPS. Only 3.6% of all residential sales are for apartments within 1000m of at least one POPS. Among those represented in the 3.6%, residential sales values in areas of high accessibility to POPS (75th percentile among the 3.6%) are on average $38,000 higher than those in areas of lower accessibility (25th percentile among the 3.6%).
-
Residential sales values are spatially dependent on neighboring dependent variable. The spatial lag model provides a better fit for our dataset compared to the OLS regression model (OLS model: R-squared = 0.651 vs Spatial lag model: R-squared = 0.660). This indicates that residential sales values are driven by dependent factors that are spatially clustered.
​
Access our results in spreadsheet format here.