
H.J.Park, K.Choi
1 3
3 Data andmethodology
The HCV datasets were acquired from the Houston Housing Authority (HHA) in January
2016 and were already aggregated at the census block group level for personal information
confidentiality. The LIHTC dataset was available on the HUD website, and it shows spe-
cific individual LIHTC project addresses that were in place by 2015. In Harris County, 316
projects existed, of which 300 projects were able to be accurately located using recorded
geographical information. The 16 other recordsdid not include coordinates as these were
not placed in service yet, although the credits of the project were already allocated.
Public transportation datasets were mainly available on the Houston–Galveston Area
Council (HGAC) and Houston METRO websites. We analyzed the street network to derive
the accessibility to public transportation facilities, which was conducted by using an Arc-
GIS analysis tool that assesses real street routes. This analysis was used to measure the
distance from each census block group centroid to the closest metro rail station and transit
center. The street file was available from 2017 topologically integrated geographic encod-
ing and referencing (TIGER) line shapefiles on the Census Bureau website. Although the
network analysis does not directly mean the level of real public transportation use of the
affordable housing program tenants, it measures the true distance to the closest facility,
which could reflect the shortest time that the tenants have to spend getting to use public
transportation in reality.
We computed the numbers and shares of HCV and LIHTC tenants within preset dis-
tances to both the closest metro rail station and the closest transit center. The centroids
of the census block groups were used to assess the number of tenants for each program.
5
Distances were more specifically categorized within three miles with an accumulated half a
mile (e.g., a half-mile,
6
one mile, one mile and a half, two miles, two and a half miles, and
three miles), and distance thresholds were more widely but intuitively set outside of the
three-mile range (e.g., five miles, 10 miles, 20 miles, and 40 miles).
Employment information was available from the US Census longitudinal employer-
household dynamic (LEHD) origin–destination employment statistics (LODES) datasets.
We used the 2015 employment datasets, including origins of employers, or residential char-
acteristics, and destinations of employers, or work characteristics, with three types of pre-
categorized income levels: jobs with earnings of $1250 per month or less; $1251 to $3333
per month; and $3333 per month or greater. We chose the lowest level of earnings ($1250
per month or less) for both origins and destinations of employers. For job accessibility,
network distances between a pair of centroids for all census block groups in Harris County
were calculated, based on the street connection, in a similar way to the public transporta-
tion accessibility analysis. To assess the accessibility for both renter groups’ residences,
we used HUD’s formula, which computes the accessibility that is assigned to each census
block group (HUD 2020b). More specifically, HUD’s original job proximity index uses a
gravity model, where the job accessibility of each residential block group is a collective
5
We used the centroids of census block groups as locations of the residences of subsidized housing tenants
instead of individual addresses, mainly because the information of the HCV program locations was aggre-
gated at the census block group level for the confidentiality of individual voucher families.
6
Although thresholds that are smaller than a half-mile are empirically important, the thresholds smaller
than a half-mile did not capture a reasonable number of residents from the closest transit facilities because
the average distance from each census block group’s centroid to its boundary was, in general, larger than a
half mile except for a few census block groups in or near Downtown Houston.