Skip to content

Introduction

sdc-redistribute moves count measures from one set of geographies onto another by areal interpolation — the area-weighted way to push a value recorded for a larger unit (a census tract) down onto smaller units (block groups) that partition it. This is the spatial analogue of the disaggregation example in the R package: a value on a source frame is distributed across a target frame.

Setup

pip install sdc-redistribute

import tempfile, pathlib
import geopandas as gpd
import pandas as pd
from sdc_redistribute import redistribute_direct

Redistributing a count

redistribute_direct takes long-format source data, a GeoJSON for the source geometries, and a {region_type: geojson} mapping for each target geography. Here we use a real census tract in Arlington County, VA (51013100100) and its four 2020 block groups, shipped with this page as a small GeoJSON. We give the tract 1,000 people and redistribute them down to the block groups.

# tract_bgs.geojson ships with this article: the four block groups of tract 51013100100.
bgs = gpd.read_file("tract_bgs.geojson")
bgs["geoid"] = bgs["geoid"].astype(str)
tract_id = bgs["geoid"].str[:11].iloc[0]

# The tract is the union (dissolve) of its block groups.
tract = bgs.dissolve().assign(geoid=tract_id)[["geoid", "geometry"]]

# redistribute_direct reads GeoJSON paths, so write the geometries out.
tmp = pathlib.Path(tempfile.mkdtemp())
tract.to_file(tmp / "tract.geojson", driver="GeoJSON")
bgs[["geoid", "geometry"]].to_file(tmp / "bgs.geojson", driver="GeoJSON")

# 1,000 people recorded for the whole tract in 2020.
source_df = pd.DataFrame(
    {"geoid": [tract_id], "year": [2020], "measure": ["pop"], "value": [1000.0]}
)

out = redistribute_direct(
    source_df,
    source_geo=tmp / "tract.geojson",
    target_geos={"block_group": tmp / "bgs.geojson"},
    count_cols=["pop"],
)
print(out[["geoid", "measure", "value"]].to_string(index=False))

       geoid    measure      value
510131001001 pop_direct 351.635089
510131001002 pop_direct 160.281519
510131001003 pop_direct 309.106123
510131001004 pop_direct 178.977268

A census tract's 1,000 people redistributed to its four block groups by area

The tract's count is split across its block groups in proportion to each one's share of the tract area — the largest block group receives the most people.

The output is long-format, one row per target geoid, the values sum back to the tract's 1,000, and the measure is suffixed _direct to record the method used.

Area-weighting assumes the count is spread evenly across the source geometry. When that assumption is poor — population clusters in part of a tract — use parcel-weighted redistribution instead (see the method comparison).

See also