Hot Spots vs Cold Spots

Hot spots show clusters where the model over-predicted value and cold spots show clusters where the model undervalued. NA clusters are areas where the models predictions were accurate or where over-prediction and under-prediction were spatially random.

Our theory is that hop spots are areas that are undervalued by the market and cold spots are over valued by the market.

Assignment of Clusters

Blocks were spatially joined to the to the cluster analysis, taking values the largest collection of cells in a given block. The map below displays the assignment. Cells and Block Groups can be turned off to see the correlation.

Analysis

Block Group level data was taken from the ACS and LODES for 2018.

Housing Value and Income

Median housing value and median household income are summarized below both taken from the ACS.

Cluster Housing_Price_Med Housing_Price_Mean Housing_Price_sd Income_Med Income_Mean Income_SD
Overvalued 200950 225717.9 145449.30 58185.0 63800.88 29671.65
Undervalued 120100 133700.6 67458.46 45101.0 47859.58 18152.31
NA 140500 151374.1 75658.77 45026.5 48695.95 20485.19

The overall distributions are similar except that cold spots have a more significant positive tails. With significantly more blocks having a median housing value of 500,000 dollars and having a median income higher than 150,000.

These differences are statistically significant as shown in the T-Tests Below.

T-Test for Differences of Means, Housing Value

Welch Two Sample t-test: hh_median_price and ll_median_price (continued below)
Test statistic df P value
-14.75 1028 0.00000000000000000000000000000000000000000000788 *
Alternative hypothesis mean of x mean of y
two.sided 133701 225718

T-Test for Differences of Means, Income

Welch Two Sample t-test: hh_median_income and ll_median_income (continued below)
Test statistic df P value
-11.77 1200 0.000000000000000000000000000002304 *
Alternative hypothesis mean of x mean of y
two.sided 47860 63801

Housing Density

There appears to be a minor, but significant difference between the clusters in Housing Density. With under valued areas having more housing per acre than over value areas.

Cluster Housing_Units_Per_Dev_Acre
Overvalued 1.217210
Undervalued 1.361382
NA 1.269104
Welch Two Sample t-test: over_housing_density and under_housing_density
Test statistic df P value Alternative hypothesis mean of x mean of y
-2.434 1140 0.0151 * two.sided 1.217 1.361

Developable Acerage

There appears to be no significant difference between develop able acreage in a given block for the clusters.

Cluster Dev_Acre
Overvalued 2272136
Undervalued 2498401
NA 2180237
Welch Two Sample t-test: overvalued and undervalued
Test statistic df P value Alternative hypothesis mean of x mean of y
-1.349 806.8 0.1776 two.sided 3160 4353

Salary

The main difference is the proportion of low in come jobs in hot spots, which significantly differs from the expected value if these variables were unrelated. See Chi Square Results below:

Pearson’s Chi-squared test: .
Test statistic df P value
75.06 2 0.00000000000000005035 * * *

Demographics

The distribution of most races is consistent across the clusters, except for white and black. In hot spots, there are significantly less White residents and significantly more black residents. The Chi Square Results back up this assumption.

Pearson’s Chi-squared test: .
Test statistic df P value
141420 6 0 * * *

Jobs

The variation in jobs is also significant, though a clear pattern is hard to determine due to the number of categories. It appears that Food Service jobs are more prevalent in cold spots, as are jobs in the arts. In hot spots professional service jobs seem to be less prevalent, as are other high paying industries such as healthcare, financial services, IT, and management.

The Chi Squared Test indicates these differences are significant from expected values.

Pearson’s Chi-squared test: .
Test statistic df P value
58284 19 0 * * *

Jobs Housing Balance

Overall there doesn’t appear to a be a significant difference between the clusters for JH balance. This could be because of a the wide variation in JH balance. If we pulled out urban vs rural we might see a more significant difference.

Cluster Jobs_Housing_Balance n
Overvalued 1.2318176 716
Undervalued 0.9957563 571
NA 1.6256132 648
Welch Two Sample t-test: over_jh and under_jh
Test statistic df P value Alternative hypothesis mean of x mean of y
1.184 1051 0.2366 two.sided 1.232 0.9958