HOLC “Redlining” Maps and the Modifiable Area Unit Problem

I’ve been working with PhD Student Shanise Owens to analyze structural racism –historical African American housing descrimination and its association with chronic disease risk in subsequent generations. As part of this work, we are analyzing the Home Owners’ Loan Corporation (HOLC) maps curated and provided by the Mapping Inequality group.

Because the historical maps do not align with more recent census geography boundaries, we run into the Modifiable Area Unit Problem, in which decisions must be made as to how families in the longitudinal cohort that located by census block ID are assigned to HOLC grades. Some potential choices we’ve considered:

  • Majority Area – Assign the grade for intersecting HOLC area that overlaps the most with the census block. If there’s a “tie” — equal areas of overlap between different grades, use one of the following methods to tie-break.
  • Assign the lesser grade (A grade “less than” D grade) – conceptually considers those living at the border between multiple HOLC grades to potentially experience more advantage than those living in the greater intersecting HOLC grade.
  • Assign the greater grade (D grade “greater than” A grade) — conceptually considers those living at the border between HOLC grades to potentially experience more disadvantage than those lving in the lower intersecting HOLC grade.
  • Some other method, e.g., considering those with multiple intersecting HOLC grade areas to be in their own kind of multi-HOLC area.

There are nuances in the spatial analysis code. For example, when considering area overlap by grade, it’s possible (and it occurs often) that a census block intersects multiple D areas, and so it’s useful to “dissolve” the original HOLC areas into combined D-graded, C-graded, etc. polygons before doing the area calculations. These issues are exacerbated when looking at larger census geographies, such as census block groups and tracts.

Consider the following 2 chunks of R code:

area_majority1 <-
joined_block_step3[
order(joined_block_step3$BLOCKID10, joined_block_step3$holc_grade_int), ] %>%
group_by(BLOCKID10) %>%
slice(which.max(area_prop))

area_majority2 <-
joined_block_step3[
order(joined_block_step3$BLOCKID10, -joined_block_step3$holc_grade_int), ] %>%
group_by(BLOCKID10) %>%
slice(which.max(area_prop))

They differ by only a negative sign, and yet the bottom code chunk assigns families with overlapping C and D areas to a D HOLC grade assignment, whereas the top code chunk assigns them to a C HOLC grade.

There are other conceptual considerations in analyzing HOLC data. Does it even make sense to simplify the assignment to simply a HOLC grade? Does it matter if the family lived in a larger HOLC D area vs a smaller HOLC D area? Does it matter if they lived in the center of the HOLC D areas and were surrounded by D areas versus if they lived at the edge of the D area next to a C area?

Moreover, what I appreciate working with Shanise is the retrospective multigenerational analysis she’s conducting, in which she’s analyzing families that experienced living in HOLC redlined neighborhoods when discriminatory housing loan policies were in effect.

Other studies that are cross-sectional in nature or are looking at modern day populations living in historically redlined communities potentially miss the point that this form of structural racism affected not only the neighborhoods, but the families that lived in the neighborhoods at the time of HOLC policy. Subsequent generations of these families may have remained in the redlined areas, but they may have moved out, yet may remain affected by the discrimination their grandparents or great grandparents experienced, and the potential loss of housing equity and weath, which passes down through generations.