K-Means Clustering

I used a K-means cluster analysis to identify similar homes across our six study area towns. Clusters were based on the similarity of homes’ acreage, year of sale, year built, liveable square footage, and story height.

With this analysis, I divided the clusters into highway and non-highway towns in order to find the average difference between the two towns.

The logic was:

Control group: non-highway towns

• Control group house price = acreage + year of sale + year built + liveable square footage + story height

Treatment group: highway towns, where the treatment is access to the NC-540 highway.

• Treatment group house price = acreage + year of sale + year built + liveable square footage + story height + treatment

Since houses in each cluster grouping are otherwise similar, the difference in their mean house values can be attributed to the effect of access to NC-540.

The K-means algorithm sorted our 104,058 homes into five groups.

A random sample of 5000 homes separated into five clusters is below:

Where were these clusters based?

The map below shows the houses by cluster grouping. As we can see, the cluster types are spread out across our study area, indicating that there are many similar houses in both highway and non-highway towns.

Results

The table below shows the mean sale price per cluster group split by town type. The average value gain by houses in highway towns was about $40,000, or an increase of 18%, compared to houses in non-highway towns.