data trove us county state boundaries geojson

STATEFP is the U.S. Census Bureau FIPS state code, a two-digit numeric string identifying each U.S. state or territory. With exactly 56 unique values and 3,234 rows it likely represents one record per county or similar sub-state geographic unit. The top value '48' (Texas, 254 rows) alone accounts for 7.85% of all records — consistent with Texas having the most counties of any U.S. state — while values like '13' (Georgia, 159) and '51' (Virginia, 133) also rank high, reflecting those states' large county counts. The entropy ratio of 0.919 indicates a fairly even spread across states despite Texas's dominance. Treatment: Use as a categorical grouping key or left-join with a FIPS lookup table to enrich with state names and region attributes; do not treat as numeric. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

56

top_value

48

top_rate

0.07854

cardinality

56

entropy

5.337

entropy_ratio

0.919

COUNTYFP is a FIPS county code — a standardized 3-digit zero-padded numeric string used in US geographic identifiers. With 330 unique values across 3,234 rows and a high entropy ratio of 0.85, codes are broadly distributed with near-uniform frequency: the most common value ('003') appears only 50 times (~1.5% top_rate). The sequential odd-number pattern in top values (001, 003, 005, 007…) is characteristic of FIPS county numbering conventions, confirming this is a geographic lookup key rather than a raw feature. Treatment: Left-join on COUNTYFP (combined with state FIPS) to enrich with county-level attributes; do not encode ordinally. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

330

top_value

003

top_rate

0.01546

cardinality

330

entropy

7.118

entropy_ratio

0.8508

COUNTYNS is a FIPS-style county National Standard (ANSI/GNIS) code — an 8-character, zero-padded numeric identifier assigned by the U.S. Geological Survey to uniquely identify counties. Every one of the 3,234 rows carries a distinct value (duplicate_rate 0.0, n_unique 3,234) with no nulls, and all values are exactly 8 characters long (len_min = len_max = 8), consistent with the fixed-width GNIS format. The perfect uniqueness and fixed length make this a reliable surrogate key for county-level joins to official geographic reference tables. Treatment: Use as a join key against TIGER/GNIS county reference data; do not encode or embed. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

3,234

len_min

8

len_max

8

len_mean

8

len_median

8

len_p95

8

word_mean

1

word_median

1

n_empty

0

n_duplicates

0

duplicate_rate

0

vocab_size

3,234

readability_flesch_mean

121.2

emoji_rate

0

url_rate

0

one_word_rate

1

allcaps_rate

1

boilerplate_rate

0

GEOID is a US Census geographic identifier column containing 5-digit FIPS county codes (e.g., '06091', '48327'), where the first two digits encode state and the last three encode county. Every value is exactly 5 characters long (len_min=5, len_max=5, len_mean=5.0), perfectly unique across all 3,234 rows with zero nulls or duplicates, confirming this is a primary key for county-level geographic records. The allcaps_rate of 1.0 is a classifier artifact — these are numeric strings, not alphabetic text. The vocab_size of 3,234 matching n_unique=3,234 means this dataset likely covers a near-complete set of US counties (there are ~3,243 counties/equivalents in the US). Treatment: Use as a primary key for joining to Census TIGER shapefiles or other county-level datasets; zero-pad-preserve when merging (already 5 chars). high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

3,234

len_min

5

len_max

5

len_mean

5

len_median

5

len_p95

5

word_mean

1

word_median

1

n_empty

0

n_duplicates

0

duplicate_rate

0

vocab_size

3,234

readability_flesch_mean

121.2

emoji_rate

0

url_rate

0

one_word_rate

1

allcaps_rate

1

boilerplate_rate

0

This column contains names of U.S. counties or county-equivalent administrative divisions, dominated by patriotic/presidential surnames (Washington, Jefferson, Franklin, Lincoln, Jackson, Madison) and geographic terms (Union, Lake, Montgomery, Marion). The duplicate rate of 40.5% (1,311 out of 3,234 rows) is expected given that common county names repeat across states, but analysts should note this column alone cannot serve as a unique identifier. The vocabulary of 1,958 tokens across 3,234 rows and a median length of 7 characters confirms these are short, single-word labels in most cases (93.1% one-word rate). Treatment: Combine with a state column to form a composite key before joining or aggregating. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

1,923

len_min

3

len_max

21

len_mean

7.04

len_median

7

len_p95

11

word_mean

1.074

word_median

1

n_empty

0

n_duplicates

1,311

duplicate_rate

0.4054

vocab_size

1,958

readability_flesch_mean

31.97

emoji_rate

0

url_rate

0

one_word_rate

0.9307

allcaps_rate

0

boilerplate_rate

0

NAMELSAD is a US Census Legal/Statistical Area Description field containing the full human-readable name of county-equivalent geographic units (e.g., 'Washington County', 'Jefferson Parish'). The word 'county' appears in 3,007 of 3,234 rows, with 'municipio' (78) and 'parish' (64) indicating Puerto Rico and Louisiana records respectively. The 39.1% duplicate rate (1,265 duplicates across 1,969 unique values) is fully expected — common county names like 'Washington County' (30 occurrences) repeat across different US states. The multi-word structure (mean 2.08 words) and short length (median 14 chars) are consistent with standardized geographic labels. Treatment: Use as a display label; join on a GEOID/FIPS key for spatial analysis rather than matching on this name alone due to cross-state duplicates. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

1,969

len_min

4

len_max

33

len_mean

14.12

len_median

14

len_p95

18

word_mean

2.079

word_median

2

n_empty

0

n_duplicates

1,265

duplicate_rate

0.3912

vocab_size

1,965

readability_flesch_mean

32.29

emoji_rate

0

url_rate

0

one_word_rate

0.0003092

allcaps_rate

0

boilerplate_rate

0

LSAD (Legal/Statistical Area Description) is a Census Bureau code that classifies geographic entities by type — values like '06' (county), '13', '15', '25' are standard LSAD codes. The distribution is severely dominated by code '06', which accounts for 3,007 of 3,234 rows (92.98%), indicating this dataset is overwhelmingly composed of one entity type (most likely counties). With only 11 unique values and near-zero entropy ratio (0.156), this column carries very little discriminative information despite being semantically meaningful. Treatment: Use as a stratification or filter variable; consider one-hot encoding if modelling across entity types, but note extreme class imbalance driven by code '06'. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

11

top_value

06

top_rate

0.9298

cardinality

11

entropy

0.5394

entropy_ratio

0.1559

CLASSFP is the FIPS functional classification code for geographic/administrative entities, almost certainly places in a US Census dataset. The distribution is severely imbalanced: 'H1' (incorporated places) accounts for 96.3% of the 3,234 rows, with the remaining four codes (C7, H6, H4, H5) collectively covering only 119 records. The entropy ratio of 0.128 confirms near-minimal informational content, meaning this column carries little discriminatory power as a feature in its current form. Treatment: Treat as a near-constant; consider dropping or collapsing into a binary flag (H1 vs. other) if used in modelling, given 96.3% dominance of a single class. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

5

top_value

H1

top_rate

0.9632

cardinality

5

entropy

0.2962

entropy_ratio

0.1276

MTFCC is a MAF/TIGER Feature Class Code, a U.S. Census Bureau classification code for geographic features. Every single one of the 3,234 rows carries the identical value 'G4020' (which corresponds to a local road/street segment), with zero nulls and an entropy of 0.0 — this column is entirely constant across the dataset. It carries no discriminatory signal whatsoever. Treatment: Drop before modelling; zero-variance constant column adds no information. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

1

top_value

G4020

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

CSAFP is likely a Combined Statistical Area FIPS (or similar geographic area code), given the numeric-string values in the hundreds range and cardinality of 175 — consistent with a US metropolitan/micropolitan area classification code. The most alarming signal is a 61.16% null rate, meaning nearly two-thirds of the 3,234 rows carry no value, which likely indicates records that do not belong to any defined statistical area. The distribution is nearly flat across all 175 codes (entropy ratio 0.936, top value '490' appears only 3.8% of the time), suggesting no single area dominates. Treatment: Impute nulls with a sentinel 'none/rural' category before use; treat as nominal categorical and one-hot or target-encode given 175 levels. medium · anthropic:default

n

3,234

nulls

1,978 (61.2%)

unique

175

top_value

490

top_rate

0.03822

cardinality

175

entropy

6.977

entropy_ratio

0.9364

CBSAFP is a Core Based Statistical Area (CBSA) FIPS code, a U.S. Census geographic identifier linking records to metropolitan or micropolitan statistical areas. With 939 unique codes across 3,234 rows, the distribution is notably flat (entropy_ratio 0.94), meaning records are spread thinly across many areas rather than concentrated. The null rate of 40.75% is a significant concern — likely representing locations outside any defined CBSA (rural areas), which is a meaningful geographic signal rather than simple missingness. The most frequent value '41980' (San Jose-Sunnyvale-Santa Clara, CA) appears only 40 times (~2.1%), confirming no single area dominates. Treatment: Treat nulls as a distinct 'non-CBSA/rural' category; left-join to CBSA reference table for region labels, then encode as categorical feature. high · anthropic:default

n

3,234

nulls

1,318 (40.8%)

unique

939

top_value

41980

top_rate

0.02088

cardinality

939

entropy

9.278

entropy_ratio

0.9395

n

3,234

nulls

3,124 (96.6%)

unique

31

top_value

47894

top_rate

0.2091

cardinality

31

entropy

4.361

entropy_ratio

0.8802

FUNCSTAT is a U.S. Census functional status code, classifying geographic or administrative entities by their operational state (e.g., 'A' = active, 'F' = fictitious, 'C' = consolidated). The distribution is severely imbalanced: 'A' accounts for 96.35% of the 3,234 records, while the remaining 6 categories together cover only 118 rows — with 'G' appearing just once. Entropy ratio of 0.107 confirms near-minimal informational diversity. Minority classes may warrant special handling but will be extremely difficult to model as targets. Treatment: One-hot encode with caution; collapse rare categories (F, C, N, S, B, G — totalling 118 rows) into an 'other' bucket or treat as a filter/stratification variable rather than a model feature. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

7

top_value

A

top_rate

0.9635

cardinality

7

entropy

0.3005

entropy_ratio

0.107

ALAND is a US Census land area field (measured in square metres), representing the land area of each geographic entity — likely counties or census tracts given n=3,234. The distribution is extremely right-skewed (skew=27.13, kurtosis=976.66): while the median is ~1.56 billion m², the max reaches 377 billion m², roughly 241× the median, indicating a small number of very large geographic units (e.g., western US counties). 362 values (11.2%) are flagged as outliers, consistent with the well-known size disparity between densely subdivided eastern counties and sprawling western ones. Treatment: Log-transform before use in any distance-based or linear model to reduce skew from 27.13; consider using as a normalisation denominator for density features. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

3,234

min

82,093

max

3.77e+11

mean

2.833e+09

median

1.563e+09

std

9.186e+09

q1

1.079e+09

q3

2.368e+09

iqr

1.29e+09

skew

27.13

kurtosis

976.7

n_outliers

362

outlier_rate

0.1119

zero_rate

0

AWATER is almost certainly a US Census TIGER/Line water area field, representing the total water surface area (in square meters) of a geographic unit such as a county or census tract. All 3,234 rows are unique and non-null, consistent with one record per geographic entity. The distribution is extremely right-skewed (skew 13.33, kurtosis 215.85): the median is ~19.5 million m² while the mean balloons to ~220 million m², and the maximum reaches ~26 billion m² — about 14× the mean — with 456 outliers (14.1% of rows) driven by large water-heavy units like coastal counties or Great Lakes-adjacent areas. Only 1 record has a zero value, which is plausible for fully land-locked units. Treatment: Log-transform (log1p to handle the single zero) before regression or clustering to reduce extreme skew. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

3,234

min

0

max

2.599e+10

mean

2.202e+08

median

1.951e+07

std

1.226e+09

q1

7.044e+06

q3

6.12e+07

iqr

5.416e+07

skew

13.33

kurtosis

215.9

n_outliers

456

outlier_rate

0.141

zero_rate

0.0003092

INTPTLAT is the internal point latitude coordinate for geographic entities (a standard Census Bureau field name), stored as a fixed-width text string rather than a numeric type. Every one of the 3,234 rows is unique, all values are exactly 11 characters long (e.g. '+41.9158651'), and the duplicate rate is 0.0, confirming these are precise geographic identifiers. The surprising signal is that a coordinate stored as text with 'allcaps' and 'one_word' alerts was profiled as a string column — it should be numeric but the leading '+' sign likely forced text treatment. Treatment: Strip leading '+', cast to float64, and use as a numeric geographic coordinate in modelling or spatial joins. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

3,234

len_min

11

len_max

11

len_mean

11

len_median

11

len_p95

11

word_mean

1

word_median

1

n_empty

0

n_duplicates

0

duplicate_rate

0

vocab_size

3,234

readability_flesch_mean

121.2

emoji_rate

0

url_rate

0

one_word_rate

1

allcaps_rate

1

boilerplate_rate

0

INTPTLON is the internal point longitude field, a standard Census Bureau coordinate column storing the longitude of a representative point within each geographic entity. Every one of the 3,234 rows is unique, all values are exactly 12 characters long (mean, median, min, and max all equal 12), and the duplicate rate is 0.0 — consistent with precise decimal-degree coordinates stored as fixed-format strings. All values appear to be negative (Western Hemisphere), ranging roughly from -065 to -123 degrees, aligning with US continental and territory coverage. Treatment: Parse to float64 for geospatial use; pair with INTPTLAT to form a coordinate pair for mapping or spatial joins. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

3,234

len_min

12

len_max

12

len_mean

12

len_median

12

len_p95

12

word_mean

1

word_median

1

n_empty

0

n_duplicates

0

duplicate_rate

0

vocab_size

3,234

readability_flesch_mean

121.2

emoji_rate

0

url_rate

0

one_word_rate

1

allcaps_rate

1

boilerplate_rate

0

This column classifies the geometric representation type of spatial features, distinguishing between 'Polygon' and 'MultiPolygon' geometries across 3,234 records. The severe class imbalance is the standout signal: 'Polygon' dominates at 98.36% (3,181 records), leaving 'MultiPolygon' as a rare minority at only 53 occurrences (1.64%). The near-zero entropy (0.121) confirms this column carries almost no information variance, which limits its predictive utility. Treatment: Flag MultiPolygon rows for geometry normalisation (explode to single polygons) before spatial analysis; otherwise low entropy makes this near-useless as a model feature. high · anthropic:default

n

3,234

nulls

0 (0.0%)

unique

2

top_value

Polygon

top_rate

0.9836

cardinality

2

entropy

0.1207

entropy_ratio

0.1207