accessibility atlas cms medicaid enrollment 2026

Overview

Source: /home/coolhand/datasets/accessibility-atlas/cms_medicaid_enrollment_2026.csv

Saturn profiled 10,302 rows across 44 columns. The stats below are deterministic and machine-readable; the prose is a language-model interpretation of those stats (opt-in, added after the fact, never sees raw rows).

State Abbreviation categorical feature

Two-letter US state abbreviation, with 51 distinct values (the 50 states plus DC) covering all 10,302 rows without nulls. The distribution is perfectly uniform: every state appears exactly 202 times, and entropy_ratio is 1.0, indicating the dataset was constructed as a balanced panel across states rather than sampled from real-world population.

Treatment: One-hot or target-encode for modelling; can also serve as a join key to state-level reference tables.

State Name categorical feature

This column lists US state names, with 51 unique values (the 50 states plus District of Columbia) and zero nulls across 10,302 rows. The distribution is perfectly uniform: every state appears exactly 202 times, yielding a top_rate of 0.0196 and entropy_ratio of 1.0. That balance suggests the dataset was constructed as a state-by-period grid rather than sampled from real-world activity.

Treatment: Use as a categorical grouping key; one-hot or target-encode for modelling.

Reporting Period numeric timestamp

Values like 202510, 202107, 201309 in YYYYMM form indicate this is a reporting period encoded as an integer year-month, not a true numeric measure. The range spans 201309 to 202510 with 102 unique periods across 10,302 rows and no nulls, consistent with monthly snapshots over roughly a decade. Summary stats like mean 202112.5 and std 249.6 are arithmetic artefacts of the YYYYMM encoding and should not be interpreted as a distribution.

Treatment: parse YYYYMM into a proper date and treat as a temporal key, not a numeric feature.

State Expanded Medicaid categorical feature

Binary Y/N flag indicating whether the state expanded Medicaid, fully populated across all 10,302 rows. The distribution is moderately imbalanced, with 'Y' covering 72.6% (7,475) of rows versus 2,827 'N'. With only 2 categories and no nulls, this is a clean feature.

Treatment: Encode as a 0/1 binary indicator for modelling.

Preliminary or Updated categorical metadata

Binary flag distinguishing preliminary versus updated records, taking values 'P' and 'U'. The split is exactly even at 5151 each, yielding maximum entropy (1.0) and a top_rate of 0.5 — this perfect balance suggests every event appears once in each state rather than being a natural distribution.

Treatment: Use to filter to one revision state (likely 'U') before analysis to avoid double-counting.

Final Report categorical label

Binary Y/N flag indicating whether a final report exists, with exactly 10302 rows split evenly into 5151 Y and 5151 N. The perfect 50/50 balance (entropy 1.0, top_rate 0.5) is unusual for organic data and suggests deliberate sampling or stratification.

Treatment: Use directly as a binary target; encode Y/N to 1/0.

New Applications Submitted to Medicaid and CHIP Agencies numeric feature

Counts of new Medicaid/CHIP applications submitted to state agencies, reported across 10,302 rows. The distribution is heavily right-skewed (skew 4.08, kurtosis 23.5) with a median of 14,644 but a max of 733,651, and 11.5% of values flagged as outliers. About 4% of rows are zero and 0.5% are null, suggesting either non-reporting periods or genuinely inactive agencies.

Treatment: log1p-transform before modelling to tame the heavy right tail.

New Applications Submitted to Medicaid and CHIP Agencies - footnotes categorical metadata

This is a footnote/qualifier column annotating the 'New Applications Submitted to Medicaid and CHIP Agencies' metric, explaining caveats like inclusion of renewals, administrative data transfers, or excluded application types. It is sparse with a 76.6% null rate, leaving only 18 distinct footnote strings across 10,302 rows; the most common, 'Includes Renewals and/or Redeterminations', covers 27% of non-null entries. Entropy ratio of 0.84 across just 18 categories indicates the present footnotes are spread fairly evenly, and several values are concatenations of multiple caveats separated by semicolons.

Treatment: Treat as qualifier metadata: split semicolon-separated flags into binary indicators and use to caveat or filter the main metric rather than as a feature.

Applications for Financial Assistance Submitted to the State Based Marketplace numeric feature

This column appears to count applications for financial assistance submitted to a State Based Marketplace, recorded as a numeric tally per row. The distribution is dominated by zeros (zero_rate 0.77, median 0, IQR 0), yet the mean is 11228.57 and the max reaches 762069.0, producing extreme skew (8.41) and kurtosis (82.64). Roughly 23% of rows (2357) flag as outliers, so the non-zero tail carries nearly all the signal.

Treatment: Log1p-transform and consider a zero-vs-nonzero indicator before modelling.

Applications for Financial Assistance Submitted to the State Based Marketplace - footnotes categorical metadata

This is a footnotes/qualifier column annotating SBM application counts, with only 3 distinct caveat strings observed. It is 97.43% null, and among the 265 populated rows, 83.0% carry the single note "Includes Renewals and/or Redeterminations". The other two values layer on duplicates and Medicaid-coverage caveats that materially affect comparability of the underlying counts.

Treatment: Keep as a data-quality flag joined to the metric; do not model directly.

Total Applications for Financial Assistance Submitted at State Level numeric feature

This is a numeric count of financial-assistance applications submitted at the state level, with 10,302 records and 5,591 unique values. The distribution is heavily right-skewed (skew 4.40, kurtosis 26.40): the median is 18,257 but the mean is 41,125 and the max reaches 762,069, with 12.2% of rows flagged as outliers. About 2% of values are zero and only 0.5% are null.

Treatment: Log-transform (log1p) before modelling to tame the heavy right skew and outliers.

Total Applications for Financial Assistance Submitted at State Level - footnotes categorical metadata

Free-form footnote annotations qualifying state-level financial assistance application counts, drawn from a small controlled vocabulary of 17 caveat strings (often concatenated with semicolons). 73.44% of rows are null, and among the 2,737 populated rows the single value "Includes Renewals and/or Redeterminations" covers 42.5% (1,163 rows), indicating these are methodology caveats rather than data values. Entropy ratio of 0.73 shows the non-null distribution is moderately spread across the 17 caveat combinations.

Treatment: Keep as a qualifier for interpreting the paired count column; split on ';' into caveat flags rather than modelling as a single category.

Individuals Determined Eligible for Medicaid at Application numeric feature

This is a numeric count of individuals deemed Medicaid-eligible at application, likely aggregated by some reporting unit (state, month, or office) given 10,302 rows and 5,568 unique values. The distribution is heavily right-skewed (skew 2.93, kurtosis 11.06): the median is 11,008 but the mean is 27,437 and the max reaches 435,560, with 10.25% of rows flagged as outliers. About 5.86% of values are zero and 0.5% are null, so empty reporting periods are present but rare.

Treatment: Log-transform (or log1p, since zeros exist) before any regression or distance-based modelling.

Individuals Determined Eligible for Medicaid at Application - footnotes categorical metadata

Footnote/caveat column annotating the Medicaid eligibility count, with 18 distinct semicolon-joined qualifier strings such as 'Includes Renewals and/or Redeterminations' (28.1% of non-nulls) and 'Does Not Include All Medicaid Determinations Made At Application'. 72.8% of rows are null, triggering the null_rate alert, meaning most reporters submitted no caveat. Entropy ratio of 0.82 across only 18 categories shows the qualifiers are spread fairly evenly rather than dominated by one note, and several values are compound flags that would need splitting to analyze cleanly.

Treatment: Split on '; ' into binary caveat flags and use to qualify or filter the paired count column; do not model directly.

Individuals Determined Eligible for CHIP at Application numeric

Individuals Determined Eligible for CHIP at Application - footnotes categorical metadata

Free-text footnote annotations qualifying CHIP eligibility counts, populated only when a state needs to caveat its figures. The column is 90.99% null and only 7 distinct notes appear across 10,302 rows, with 'Includes Renewals and/or Redeterminations' covering 46.55% of the non-null cases. These caveats materially change comparability — some rows count households not individuals, others include presumptive or conditional determinations.

Treatment: Retain as a caveat flag joined to the eligibility count column; do not use as a modelling feature.

Total Medicaid and CHIP Determinations numeric feature

Numeric counts of total Medicaid and CHIP determinations per record, likely a state-month or state-period rollup. The distribution is heavily right-skewed (skew 2.92, kurtosis 10.77) with a median of 11,977 sitting far below the mean of 29,811 and a max of 467,780, and 10.5% of values flagged as outliers. About 5.5% of rows are zeros and 0.5% are null, which is worth checking against reporting periods or non-reporting states.

Treatment: Log-transform (after handling zeros) before any regression or distance-based modelling.

Total Medicaid and CHIP Determinations - footnotes categorical metadata

Free-text footnote field qualifying the Medicaid/CHIP determinations metric, with 12 distinct semicolon-delimited caveat strings drawn from a small controlled vocabulary. 81.58% of rows are null, and among the 18% populated, 47.10% carry the single caveat 'Includes Renewals and/or Redeterminations'. Most other values are compound caveats concatenated with semicolons, signalling these are flags about how each state reported the count rather than analytic content.

Treatment: Split on '; ' into boolean caveat flags and use to qualify or filter the paired determinations metric.

Medicaid and CHIP Child Enrollment numeric feature

This column reports Medicaid and CHIP child enrollment counts, ranging from 0 to 5,339,904 with a mean of 740,683 and median of 511,370. The distribution is severely right-skewed (skew 2.80, kurtosis 8.85) with 804 outliers (7.84%) and a 2.13% zero rate, consistent with a few large states dominating while smaller jurisdictions cluster low. Null rate is negligible (0.5%).

Treatment: log-transform before regression to tame the heavy right tail.

Medicaid and CHIP Child Enrollment - footnotes categorical metadata

This is a footnotes/caveat column annotating Medicaid and CHIP child enrollment figures, with only 7 distinct values across 10,302 rows. It is overwhelmingly null (92.19%), and among the populated rows nearly half (49.57%) carry the same caveat about counting individuals enrolled at any time in the month rather than a point-in-time. The second most common note flags states 'Unable to provide data due to system limitations' (218 rows), which is a meaningful data-quality signal worth surfacing alongside any enrollment analysis.

Treatment: Keep as a qualitative flag joined to enrollment rows; do not model directly.

Total Medicaid and CHIP Enrollment numeric feature

State- or region-level Medicaid and CHIP enrollment counts, with a median of 1,031,822 and a max of 14,462,560 stretching the distribution well past the mean of 1,567,204. Skew of 3.67 and kurtosis of 16.57 confirm a long right tail, and 692 rows (6.7%) flag as outliers — consistent with a few very large states dominating. Near-zero null and zero rates mean the column is essentially fully populated.

Treatment: log-transform before regression to tame the heavy right tail.

Total Medicaid and CHIP Enrollment - footnotes categorical metadata

Free-text footnote qualifiers attached to state-month Medicaid/CHIP enrollment figures, explaining methodology caveats. The column is 93.11% null and only 710 rows carry any of 9 distinct notes; among those, 56.19% read 'Includes Individuals Enrolled At Any Time in Month (Not a Point-in-Time Count)', flagging that the headline counts are not point-in-time and so are not directly comparable across rows.

Treatment: Keep as a methodology flag; binarize or one-hot the few recurring notes when filtering comparable enrollment values.

Total Medicaid Enrollment numeric feature

Numeric tally of Medicaid enrollees, almost certainly aggregated by state-month or similar jurisdiction-period grain given 10302 rows and 8221 unique values. The distribution is severely right-skewed (skew 3.61, kurtosis 16.1) with the mean (1,432,519) sitting well above the median (949,244) and a max of 13,160,563 dwarfing Q3 of 1,646,917. About 6.7% of rows (691) flag as outliers, while nulls are negligible (0.5%) and zeros essentially absent.

Treatment: Log-transform before any regression or distance-based modelling to tame the heavy right tail.

Total Medicaid Enrollment - footnotes categorical metadata

This is a footnote/annotation field qualifying Total Medicaid Enrollment figures, with only 7 distinct caveat strings across 10,302 rows. It is empty 93.21% of the time, and when populated, 57% of values are the single caveat 'Includes Individuals Enrolled At Any Time in Month (Not a Point-in-Time Count)'. The remaining notes flag methodology variations like retroactive enrollments, limited-benefit inclusions, or system limitations preventing reporting.

Treatment: Keep as a qualitative caveat flag; binarize as 'has_footnote' or one-hot the 7 categories rather than treating as a feature.

Total CHIP Enrollment numeric feature

Numeric column capturing total CHIP (Children's Health Insurance Program) enrollment counts, likely per state-month or similar administrative unit. The distribution is heavily right-skewed (skew 3.90, kurtosis 18.21) with the mean (136,663) far above the median (78,933) and a max of 1,317,347 versus a Q3 of 171,960. About 5.3% of rows (542) are flagged outliers, and nulls and zeros are negligible.

Treatment: log-transform before regression to tame the heavy right tail.

Total CHIP Enrollment - footnotes categorical metadata

This is a sparse footnote/qualifier column accompanying CHIP enrollment counts, with only 4 distinct annotation strings across 10,302 rows. It is null 95.92% of the time, and when present is dominated (76.0%) by "Includes Individuals Enrolled At Any Time in Month (Not a Point-in-Time Count)". The remaining notes flag retroactive enrollments, incomplete coverage, or system-limitation gaps — caveats that materially affect comparability of the enrollment figures.

Treatment: Keep as a caveat flag joined to the enrollment metric; do not use as a model feature.

Total Adult Medicaid Enrollment numeric feature

Numeric tally of total adult Medicaid enrollees, almost certainly aggregated by state and reporting period given the multi-million maximum (8,497,290) and median of 551,447. The column is sparsely populated—84.65% null—and heavily right-skewed (skew 4.56, kurtosis 23.4), with 62 outliers and a standard deviation (1,268,819) larger than the mean (810,066). Only 1,415 unique values across 10,302 rows suggests repeated state-level totals rather than per-person records.

Treatment: Impute or filter the 84.65% nulls and log-transform before any regression to tame the skew.

Total Adult Medicaid Enrollment - footnotes categorical metadata

This is a footnote/qualifier column for the Total Adult Medicaid Enrollment metric, carrying free-text caveats about how a state's count was constructed. It is overwhelmingly empty (98.79% null) with only 4 distinct notes across 10,302 rows; the dominant note (56% of non-null) flags that counts include anyone enrolled at any time in the month rather than a point-in-time snapshot. A small but notable 3 rows admit the data could not be provided due to system limitations.

Treatment: Keep as a qualitative caveat flag joined to the enrollment column; do not use as a model feature.

Total Medicaid and CHIP Determinations Processed in Less than 24 Hours numeric feature

Counts of Medicaid/CHIP determinations processed in under 24 hours, reported per state-month or similar reporting unit. The distribution is severely right-skewed (skew 6.44, kurtosis 47.28) with a median of 3,470 but a max of 791,175 and 699 outliers (11.9%). Notably, 43.1% of rows are null, suggesting many reporters did not submit this metric.

Treatment: Log-transform and impute or flag missingness before modelling.

Total Medicaid and CHIP Determinations Processed in Less than 24 Hours - footnotes categorical metadata

Footnote/caveat field annotating the '<24 hour Medicaid/CHIP determinations' metric, capturing reporting caveats reported by states. It is mostly empty (89.21% null) and only 20 distinct strings appear across 10302 rows; when present, the dominant caveat is 'Incorrectly reporting processing time at application level, as opposed to the individual level' at 37.77% of non-nulls. Several values are semicolon-concatenated combinations of base caveats, so the 20 categories are not independent.

Treatment: Keep as a data-quality flag joined to the metric; split on ';' into boolean caveat indicators rather than modelling as a single category.

Total Medicaid and CHIP Determinations Processed Between 24 Hours and 7 Days numeric feature

This is a state/period-level count of Medicaid and CHIP eligibility determinations completed within a 24-hour to 7-day window. The distribution is heavily right-skewed (skew 4.39, kurtosis 29.5) with a median of 2,312 but a max of 133,996, and 10.3% of values flag as outliers. Notably, 43.1% of rows are null, suggesting many reporting periods or jurisdictions did not submit this metric.

Treatment: Log-transform and impute or flag missingness before modelling.

Total Medicaid and CHIP Determinations Processed Between 24 Hours and 7 Days - footnotes categorical metadata

This is a footnotes/caveats column accompanying a Medicaid/CHIP determinations metric, holding free-form annotations about reporting irregularities. It is empty 89.21% of the time, and among the 1,111 populated rows just 21 distinct messages appear, with 'Incorrectly reporting processing time at application level...' covering 37.77% of non-nulls. Notably, several entries are semicolon-joined composites of multiple caveats, suggesting the field is a concatenation rather than a clean code.

Treatment: Keep as a qualitative caveat flag; split on ';' and one-hot the distinct footnote codes if used as features.

Total Medicaid and CHIP Determinations Processed Between 8 Days and 30 Days numeric feature

Numeric count of Medicaid/CHIP determinations processed within an 8-30 day window, likely reported per state-month or similar reporting unit. The distribution is heavily right-skewed (skew 3.98, kurtosis 19.86) with a median of 2,528 but a max of 155,529, and 10.2% of values flagged as outliers. Note that 43.1% of rows are null and 2.6% are exactly zero, so coverage is partial before any modelling.

Treatment: Impute or filter the 43% nulls, then log-transform to tame the heavy right tail before modelling.

Total Medicaid and CHIP Determinations Processed Between 8 Days and 30 Days - footnotes categorical metadata

Free-text footnote field qualifying caveats on the 8–30 day Medicaid/CHIP determination counts; 89.26% of the 10,302 rows are null, so footnotes are the exception not the rule. Among the 1,106 populated rows there are only 16 distinct values (entropy 2.82, ratio 0.70), with 39.60% being the single caveat 'Incorrectly reporting processing time at application level, as opposed to the individual level' — a sizeable data-quality warning on the underlying metric. Several top values are semicolon-concatenated combinations, indicating the field stacks multiple caveats per row.

Treatment: Keep as a caveat flag on the paired metric; split on ';' to derive boolean indicators rather than modelling directly.

Total Medicaid and CHIP Determinations Processed between 31 days and 45 days numeric feature

This is a numeric operational metric counting Medicaid and CHIP determinations processed in a 31-45 day window, likely reported per state-month. The distribution is heavily right-skewed (skew 4.90, kurtosis 31.95) with a median of 693 but a mean of 2917 and max of 81475, indicating a few jurisdictions with very large backlogs. Notable concerns: 43.07% of values are null, 6.63% are zero, and 12.75% are flagged as outliers.

Treatment: Investigate the 43% nulls (likely non-reporters) and log-transform before modelling to tame the heavy right tail.

Total Medicaid and CHIP Determinations Processed between 31 days and 45 days - footnotes categorical metadata

This is a footnote/caveat column annotating data quality issues for the 31-45 day Medicaid/CHIP determinations metric. It's null 89.26% of the time, meaning footnotes apply to only ~11% of rows, but among those the top caveat — 'Incorrectly reporting processing time at application level, as opposed to the individual level' — covers 46.11% (510 records), suggesting a systemic reporting inconsistency. The 15 unique values include compound entries joined by semicolons, indicating multiple caveats can co-occur on a single row.

Treatment: Keep as a qualitative caveat flag; split on '; ' into boolean indicators if you need to filter unreliable rows from the main metric.

Total Medicaid and CHIP Determinations Processed in More than 45 Days numeric feature

Counts of Medicaid/CHIP applications processed beyond the 45-day SLA, reported per state-month or similar reporting unit. The distribution is severely right-skewed (skew 5.26, kurtosis 37.4): the median is 395 but the mean is 3,027 and the max reaches 106,943, with 15.4% flagged as outliers. Note that 43% of rows are null and 10% are exact zeros, so coverage is uneven across reporters.

Treatment: Log1p-transform and impute or flag the 43% nulls before any modelling or aggregation.

Total Medicaid and CHIP Determinations Processed in More than 45 Days - footnotes categorical metadata

Footnote/caveat field annotating data-quality issues with the '>45 days' Medicaid/CHIP determinations metric. Sparse - 89.26% null - and dominated by one caveat ('Incorrectly reporting processing time at application level...') at 46.11% of the 1,104 non-null rows, with only 15 distinct values that include semicolon-concatenated combinations. Entropy ratio of 0.652 confirms a skewed but multi-category caveat vocabulary worth preserving for data-quality filtering.

Treatment: Keep as a data-quality flag; split on '; ' into a multi-hot caveat indicator and exclude or down-weight affected rows in downstream metric analysis.

Total Call Center Volume (Number of Calls) numeric feature

Numeric volume metric capturing total calls handled by a call center, reported per row (likely per period or center). The column is sparsely populated with a 69.49% null rate and only 1,592 unique values across 10,302 rows. Distribution is heavily right-skewed (skew 4.66, kurtosis 25.11) with the mean (172,294.63) more than double the median (73,754) and a max of 2,615,575 against a Q3 of 180,553, plus 223 flagged outliers.

Treatment: Impute or flag the 69% missing rows and log-transform before modelling to tame the heavy right tail.

Total Call Center Volume (Number of Calls) - footnotes categorical metadata

This column holds free-form footnotes qualifying the Total Call Center Volume metric, with 27 distinct semicolon-joined caveats describing what each call count does or doesn't include. 70.96% of rows are null, meaning most volume figures carry no caveat, but among the 2,994 annotated rows the disclaimers are spread broadly (entropy ratio 0.787) with the most common note ('Does not include all calls received after business hours; Includes calls for other benefit programs') covering only 20.09% of non-nulls. The recurring themes — excluded after-hours calls, bundled benefit programs, and live-agent-only counts — signal that the underlying volume metric is not comparable across rows without parsing these flags.

Treatment: Split on '; ' into binary caveat flags and use them to gate or adjust comparisons of the call-volume metric.

Average Call Center Wait Time (Minutes) numeric feature

This column captures the average call-center wait time in minutes per record, but it's only populated for ~30% of rows (null_rate 0.6947). Values range from 0 to 72 minutes with a median of 5 and mean of 9.95, showing a strong right skew (1.78) and 11.6% zeros. About 145 records (4.6%) are flagged as outliers, consistent with a long tail of unusually long waits.

Treatment: Impute or flag missingness, then log-transform to tame the right skew before modelling.

Average Call Center Wait Time (Minutes) - footnotes categorical metadata

This is a footnote/caveat field qualifying the 'Average Call Center Wait Time (Minutes)' metric, holding semicolon-joined methodological notes (e.g., callbacks included, after-hours exclusions, other benefit programs counted). 69.73% of rows are null, yet across the 30.27% populated there are 44 distinct combinations with very high entropy ratio (0.918) and no dominant value—the top combination only covers 7.89%. The notes themselves reveal non-comparable measurement methods across rows, which undermines straight comparisons of the wait-time metric.

Treatment: Split on '; ' into boolean flags per caveat and use to stratify or adjust comparisons of the wait-time metric.

Average Call Center Abandonment Rate numeric feature

This is a numeric operational metric capturing call center abandonment rate, ranging from 0.0 to 0.652 with a median of 0.088 and mean of 0.132. The distribution is right-skewed (skew 1.22) with 54 outliers (1.7%) and a heavy null rate of 69.49%, meaning only about 30% of rows carry a value.

Treatment: Impute or add a missingness flag given 69% nulls, then consider a log or sqrt transform before modelling.