Technical methodology

The compact, field-level version: for the plain-language explanation, see the methodology page.

Source data

Enrollment figures come from Arkansas's SIS (Student Information System), not NCES CCD. data/districts.json and data/typology_output.csv are produced by an upstream data pipeline (maintained separately, not part of this site) that builds the district panel from the state's published SIS enrollment reports and classifies each district's trajectory.

Coverage: fall 2013 through fall 2025 (13 years), 235 current-boundary districts.

Why SIS instead of CCD

CCD's fall headcount runs a stable ~3.5–4% higher than SIS across every year 2013-14 through 2024-25 (except an unexplained 2023-24 convergence). This is a real measurement difference, not an error, though SIS's own exact day-to-day methodology isn't fully pinned down. What's confirmed: SIS is not a CCD-style single-day headcount. A differently-named ADE report ("Enrollment by District on October 1") was tested and found byte-identical to this same SIS figure across every year checked, which a true single-day snapshot should not be (it should diverge from SIS by roughly the same ~3.5–4% CCD does). SIS also matches ADE's own published figures and independent media reporting to the exact student (e.g., Springdale 21,097 and Little Rock 18,964, both 2025-26, matched exactly). All 235 districts were individually cross-checked against ADE's current "Enrollment Count by District" report for 2025-26, joined by state LEA code rather than by name (via the crosswalk pipeline's leaid/district_lea mapping, since the two sources use different code schemes), and every one matched exactly. Earlier years have not been cross-checked district-by-district this way. What's confirmed only approximately: SIS's total resembles the state's first-quarter average-daily-membership figure (adm_q1) within about 0.82% mean deviation, from a limited sample comparison: a resemblance, not a proven equivalence. What's not confirmed: the exact technical definition of what SIS's total column measures day-to-day; ADE's own "October 1" naming does not appear to describe this number's real methodology, given the byte-identical result above. SIS was chosen for the reason that's actually confirmed: it's the number ADE itself publishes and treats as authoritative for each district, matching what's independently reported — not because its internal methodology was independently verified. It also resolves two real CCD coverage gaps (Cross County, Imboden Charter) that would otherwise require workaround logic.

The full investigation (the year-by-year offset table and two abandoned CCD/SIS blend designs) is documented in the upstream pipeline's working notes (available on request).

Eras

• Baseline: fall 2013 – fall 2019 ("2013-14" through "2019-20")
• COVID shock: fall 2020 – fall 2021 ("2020-21", "2021-22")
• EFA era: fall 2022 onward ("2022-23"–), named for the Arkansas LEARNS Education Freedom Account program

baseline_cagr and efa_cagr use only the first and last available year within each district's own era window. typology classifies each district's baseline-era vs. EFA-era trajectory shape (10 categories, e.g. sustained_decline, reversal_upward).

Known limitations

• 7 districts have a current-boundary history shorter than 2013, because their boundary itself postdates 2013 (a crosswalk merger absorbed a predecessor district): Smackover Norphlet, Camden Fairview, Magnolia, Nevada (all start 2014), Hackett, West Memphis (both start 2015), and Pine Bluff (starts 2021). A related case, Jacksonville North Pulaski, defaults to a 2013 current_boundary_since but has no real SIS rows before 2016-17 (the district's actual creation date), its thin baseline is caught by baseline_years_thin/n_baseline_years, not by current_boundary_since.
• Pine Bluff is one of the 7 boundary-change districts above (2021), not a special case. An earlier version of this doc said Pine Bluff had zero baseline-era years and typology: null. That was true before a fix pulled its own pre-merger enrollment into a continuous record (context-only, not fed into classification, same treatment as the other 6). Pine Bluff now has a full 2013-2025 series, data_status: "ok", and a real typology like every other district.
• districts.json's own geo field is unused, not missing. It's null for all 235 districts and was never populated. The map instead joins district polygons from a separate boundary file, data/arkansas_districts.topojson (Census TIGER 2024, simplified for the web), to each district's districts.json record by GEOID at render time, not a blocker for anything; every view, including the map, is fully functional.
• The District Rankings row encodes several things at once, by design. Each row shows a rank number, a gradient-colored dot keyed to pct_change_efa_era (the same magnitude scale the map's fill uses), a 2013-2025 trajectory sparkline, the district name, up to three caveat icons (thin baseline / boundary changed / reversal magnitude), and the EFA-era percentage change itself. Rows filter into a Declining/Stable/Growing bucket by that same percentage (±2% thresholds) and sort by it, default most-declining-first.

Crosswalk-edge-level detail, confidence tiers, and the full audit trail live in the upstream pipeline's documentation (available on request).