saturn

This dataset contains 2,327 New York City census tracts with median household income, geographic identifiers (state, county, tract), and tract names. The headline issue is median_household_income: it has a minimum of -666,666,666 and a mean of about -36 million, indicating sentinel/missing-value codes that must be filtered before any analysis — the median of $76,833 is the more trustworthy central value. County coverage is uneven, with Brooklyn (Kings) holding 34.6% of tracts and Staten Island only 126, so per-borough comparisons should be normalized. The state column is constant (36 = New York) and can be dropped.

Likely U.S. median household income in dollars, with median 76833 and IQR spanning 53242.5 to 102359.5. The minimum of -666666666 is a sentinel null code that is poisoning the mean (-36017397.46) and standard deviation (150923371.88), and 208 rows (8.94%) flag as outliers. Skew of -3.94 and kurtosis of 13.53 are entirely artifacts of that sentinel.

This column holds fully-qualified Census tract names for New York City, with every one of the 2327 rows unique and non-null. Lengths cluster tightly between 38 and 46 characters and every record contains the words 'new', 'york', 'census', 'tract', and 'county;', confirming a rigid template; the borough token (Kings 805, Queens 725, Bronx 361, Richmond 126) is the only meaningful variation. It is effectively a row identifier, not a feature.

The column 'state' is numeric but holds the single value 36 across all 2327 rows, with zero variance and no nulls. It carries no information for modelling and likely encodes a fixed jurisdiction or processing-state code captured during a single-state extract.

Despite being typed as numeric, `county` has only 5 unique values across 2327 rows (5, ?, 47, 81, 85 implied by the quartiles) with no nulls — these are almost certainly FIPS-style county codes rather than a measured quantity. The distribution is left-skewed (skew -0.72) with the median at 47 and Q1 also at 47, meaning at least a quarter of rows share that single code. Treating mean (55.0) or std (25.97) as meaningful would be misleading given the categorical nature.

Almost certainly U.S. Census tract codes stored as integers, with 1530 distinct values across 2327 rows and no nulls. The distribution is severely right-skewed (skew 10.14, kurtosis 189.82) with a max of 990100 sitting far above the q3 of 57900.5 and median of 30100, producing 63 outliers (2.7%); this is an artifact of tract numbering conventions, not a true numeric magnitude.

This column lists the NYC borough/county for each record, with all 5 expected values present across 2327 rows and no nulls. Distribution roughly tracks borough population: Brooklyn (Kings) leads at 805 (top_rate 0.346), followed by Queens (725), Bronx (361), Manhattan (310), and Staten Island (126). Entropy ratio of 0.898 indicates the categories are fairly evenly spread rather than dominated by one value.