Data colophon

The argument in this book rests on public data that anyone can obtain. This page explains where it came from, how it was processed, and how to replicate or extend the analysis — whether you are a journalist, a researcher, a policy advocate, or a curious resident of any American city. The specific files, field names, and code tables described here are specific to King County, Washington. Other jurisdictions publish equivalent data under different names with different structures — the methodology transfers, the field names almost certainly do not.

Data sources

Key methodological decisions

Known limitations

Python environment

The analysis requires Python 3.9 or later. Install dependencies:

pip install geopandas pandas numpy requests

For the web visualizations:

pip install pydeck

All scripts assume the working directory contains the data files. King County CSV extracts require encoding='latin-1' — the default UTF-8 encoding will fail on some records.

Core pipeline

Load and join the assessor CSVs

import pandas as pd

parcel = pd.read_csv('EXTR_Parcel.csv', encoding='latin-1', low_memory=False)
rpacct = pd.read_csv('EXTR_RPAcct_NoName.csv', encoding='latin-1', low_memory=False)

parcel['pin'] = parcel['Major'].astype(str).str.zfill(6) + \
                parcel['Minor'].astype(str).str.zfill(4)
rpacct['pin'] = rpacct['Major'].astype(str).str.zfill(6) + \
                rpacct['Minor'].astype(str).str.zfill(4)

df = parcel.merge(rpacct, on='pin', how='left', suffixes=('_p', '_r'))
seattle = df[df['DistrictName'] == 'SEATTLE'].copy()

for col in ['ApprLandVal', 'ApprImpsVal', 'SqFtLot']:
    seattle[col] = pd.to_numeric(seattle[col], errors='coerce').fillna(0)

Find stub parcels

stubs = seattle[seattle['ApprImpsVal'] <= 1000]
taxable_stubs = stubs[stubs['TaxStat'] == 'T']

print(f"Taxable stubs: {len(taxable_stubs):,}")
print(f"Total land value: ${taxable_stubs['ApprLandVal'].sum():,.0f}")

Find taxable vacant commercial parcels

vacant_commercial = taxable_stubs[taxable_stubs['PresentUse'] == 309].copy()

# Remove split assessments
exclude = ['imps on', 'associated with', 'econ unit', 'imp data on']
mask = vacant_commercial['PropName'].str.lower() \
       .str.contains('|'.join(exclude), na=False)
clean = vacant_commercial[~mask].copy()

print(f"Taxable vacant commercial (clean): {len(clean):,}")
print(f"Land value: ${clean['ApprLandVal'].sum():,.0f}")
print(f"Acres: {(clean['SqFtLot']/43560).sum():,.1f}")

Spatial join to census tracts

import geopandas as gpd
import numpy as np

parcels = gpd.read_file('Parcels_for_King_County_with_Address,...shp')
tracts  = gpd.read_file('2020_Census_Tracts_for_King_County.shp')

# Both use NAD83(HARN) Washington North — no reprojection needed for join
seattle_shp = parcels[parcels['LEVY_JURIS'] == 'SEATTLE'].copy()
for col in ['APPRLNDVAL', 'APPR_IMPR', 'LOTSQFT']:
    seattle_shp[col] = pd.to_numeric(seattle_shp[col], errors='coerce').fillna(0)

seattle_shp['is_stub'] = (seattle_shp['APPR_IMPR'] <= 1000).astype(int)

joined = gpd.sjoin(seattle_shp,
                   tracts[['GEO_ID_TRT', 'NAME20', 'geometry']],
                   how='left', predicate='within')

agg = joined.groupby('GEO_ID_TRT').agg(
    tract_name   = ('NAME20',      'first'),
    parcel_count = ('PIN',         'count'),
    land_value   = ('APPRLNDVAL',  'sum'),
    imps_value   = ('APPR_IMPR',   'sum'),
    stub_count   = ('is_stub',     'sum'),
    lotsqft      = ('LOTSQFT',     'sum'),
).reset_index()

agg['imps_ratio']    = agg['imps_value'] / (agg['land_value'] + agg['imps_value'])
agg['stub_pct']      = agg['stub_count'] / agg['parcel_count']
agg['land_per_acre'] = agg['land_value'] / (agg['lotsqft'] / 43560)

result = tracts.merge(agg, on='GEO_ID_TRT', how='left').to_crs('EPSG:4326')
result.to_file('tracts_enriched.geojson', driver='GeoJSON')

Add Census population

import urllib.request, json

url = ('https://api.census.gov/data/2020/dec/pl'
       '?get=P1_001N,NAME&for=tract:*&in=state:53+county:033')
with urllib.request.urlopen(url) as r:
    raw = json.loads(r.read())

pop_df = pd.DataFrame([
    {'GEO_ID_TRT': row[2]+row[3]+row[4], 'population': int(row[0])}
    for row in raw[1:]
])

# Join population to enriched tracts and rewrite
import geopandas as gpd
enriched = gpd.read_file('tracts_enriched.geojson')
enriched = enriched.merge(pop_df, on='GEO_ID_TRT', how='left')
enriched['pop_per_acre'] = enriched['population'] / \
    (enriched['ALAND20'].fillna(0) / 4046.86)
enriched.to_file('tracts_enriched.geojson', driver='GeoJSON')

Replicating for another city

The methodology is not specific to King County. Any jurisdiction that publishes parcel-level assessment data with land and improvement values separated can be analyzed the same way. The general steps:

1. Obtain the data. Most county assessors publish annual extracts. Search for "[county name] assessor parcel data download." Key fields needed: parcel ID, land value, improvement value, present use or property class code, lot square footage, and coordinates or a shapefile.

2. Identify the stub threshold. $1,000 is specific to King County's conventions. Other jurisdictions may use $0, $100, or $500 as their placeholder. Examine the distribution of improvement values — the placeholder will appear as a spike at a round number.

3. Filter to taxable parcels. Most datasets include an exemption or tax status flag. Remove exempt parcels before counting — parks, schools, government buildings, and qualifying nonprofits will otherwise inflate the numbers.

4. Cross-reference with sales data. The most powerful analysis compares assessed land value to sworn sale prices from transfer tax (REET) filings. A parcel that sold for $48.75 million and is assessed at $13.3 million tells the story without any theory attached.

5. Check the methodology against known cases. Before publishing any numbers, pull the records for two or three properties you can verify independently — a recent commercial sale, a named building, a surface parking lot. If the data matches what you can confirm from other sources, the methodology is sound.

How this analysis was conducted

The findings in this book did not emerge from a pre-designed study. They emerged from a process of iterative questioning — pulling a thread, following it, finding something unexpected, and pulling that thread too. The methodology is worth describing because it is reproducible, and because the messiness is part of the point.

The starting question was simple: how many parcels in Seattle carry improvement values of $1,000 or less? The answer — more than six thousand, once public parcels are included — led immediately to a second question: what is actually on those parcels? Which led to a third: why does the same $1,000 figure appear on a surface parking lot, a James Beard award-winning restaurant, a philanthropic campus whose improvements peaked at more than $400 million in assessed value before being written down to $1,000 over a decade, and three large transit-oriented-development zoned parcels adjacent to a light rail station?

Each answer required cleaning a layer of data problems first. The taxable filter removes most public parcels but not all — parks and utilities appear in the taxable population because historical records were never corrected. Split assessments make some commercial parcels look vacant when the improvement value is recorded on a companion parcel. The "PORTION TAXABLE" notation in legal descriptions signals partial exemptions that suppress the visible land value. None of these problems invalidate the analysis. They require acknowledging that the published numbers are upper bounds, not precise counts.

Where the data could be checked against a known case — a named restaurant, a well-documented campus, a specific parcel — it was. The aggregate numbers rest on the same methodology as the individual cases that passed that check.

This is the standard for data journalism, not academic research. The bar is not a peer-reviewed methodology section — it is reproducibility and transparency. The data is public. The code is on this page. The known limitations are documented. A reporter, a researcher, or a policy advocate who disagrees with the findings can download the same files, run the same queries, and check the work. That accountability is the argument for doing it this way.

For those who want to conduct a similar analysis — in King County or elsewhere — the practical advice is this: start with a question you can answer with a single filter, then let the anomalies guide you. The most important findings in this analysis were not planned. They surfaced because the data was asked a question, gave a surprising answer, and that answer was taken seriously enough to follow.

Further reading

McMillen, Daniel and Ruchi Singh. "Land Value Estimation using Teardowns." Lincoln Institute of Land Policy, 2022. Establishes the teardown sale methodology for estimating land values in built-up markets where vacant land sales are scarce — directly relevant to the reproduction cost arguments in this book.

Missemer, Antoine and Gauthier Pottier. "The Neoclassical Domestication of Land." Land Economics, Vol. 101 No. 4, 2025. Documents the ideological capture of land economics by neoclassical theory and the deliberate erasure of land as a distinct factor of production.

Washington State Department of Revenue. Property Tax Guide for Washington State. The statutory framework including RCW 84.40.030 (true and fair value), RCW 84.40.340 (assessor's authority to require financial records), and the 1930 Uniformity Clause history.