E-filed 990s are up 1.5x since 2015, and 60k+ new public charities have grown into the full Form 990

The ask

1. How many 990s were e-filed in tax year 2015 versus tax year 2023?
2. How many new public charities, eligible for filing the full Form 990, were founded between 2015 and 2023?

(Requester identity and the verbatim ask are in the gitignored _private.md.)

Data & method

Two canonical artifacts, joined on EIN2:

• E-file Form 990 header tables (s3://nccs-efile/public/efile_v2_1/, F9-P00-T00-HEADER-{year}.CSV, all 16 published tax years 2009-2024). One row per e-filed 990 or 990-EZ. Two fields matter beyond identity: RETURN_TYPE distinguishes the full 990 from the 990-EZ, and F9_00_YEAR_FORMATION is the organization’s founding year as self-reported on the return, a field the BMF does not carry (the BMF has only the IRS ruling date, which lags founding).
• Geocoded unified BMF (via nccsdata::nccs_read), which supplies public-charity status: org_type = "public_charity" is the package’s canonical cut (501(c)(3), foundation code not private-foundation).

Conventions this analysis follows:

• Everything is by tax year. NCCS deliberately re-partitions IRS calendar-year files by tax year during harmonization; all counts below are tax-year counts.
• Form type and tax year are taken from the RETURN_TYPE and TAX_YEAR columns on every row, never inferred from a file’s name. The read helper carries a file_year column so the render itself verifies the partitioning (see the check below).
• “Eligible for filing the full 990” is proxied by demonstrated filing: an EIN counts if it filed at least one full Form 990 (not EZ) in tax years 2015-2024. A threshold-based alternative (receipts >= $200k or assets >= $500k in the current BMF) is reported as a sensitivity check.

Partitioning check on this render: 5,675,704 header rows scanned, 0 rows where the in-column TAX_YEAR disagrees with the file-name year; RETURN_TYPE values present: 990, 990EZ.

Findings

Q1: E-filed 990s by tax year

q1 <- hdr |>
  count(TAX_YEAR, RETURN_TYPE) |>
  group_by(TAX_YEAR) |>
  summarise(
    `Full 990` = sum(n[RETURN_TYPE == "990"]),
    `990-EZ`   = sum(n[RETURN_TYPE == "990EZ"]),
    Total      = sum(n),
    .groups = "drop"
  ) |>
  arrange(TAX_YEAR)
knitr::kable(q1 |> mutate(across(-TAX_YEAR, fmt)), align = "lrrr")

Tax year 2015: 358,413 e-filed returns. Tax year 2023: 527,764, about 1.5x. Two structural notes: e-filing was optional before tax year 2021 (the Taxpayer First Act mandate), so early years reflect e-file adoption rather than all filings, while 2023 is near-complete; and tax year 2024 is still accruing. This table covers the 990 and 990-EZ; the 990-PF and 990-N are separate products.

Q2: New public charities that grew into the full 990

founders <- hdr |>
  filter(
    RETURN_TYPE == "990",
    !is.na(F9_00_YEAR_FORMATION),
    F9_00_YEAR_FORMATION >= 2015, F9_00_YEAR_FORMATION <= 2023,
    F9_00_YEAR_FORMATION <= TAX_YEAR   # formation cannot postdate the tax year
  ) |>
  group_by(EIN2) |>
  summarise(formation_year = min(F9_00_YEAR_FORMATION), .groups = "drop")

new_pc <- founders |> semi_join(pc, by = "EIN2")

knitr::kable(
  new_pc |> count(formation_year, name = "Public charities") |>
    mutate(`Public charities` = fmt(`Public charities`)),
  align = "lr"
)

61,826 public charities founded 2015-2023 have filed at least one full Form 990 (of 70,668 full-990 filers reporting those formation years across all subsections).

Sensitivity and cross-checks:

• Restricting to explicit 170(b)(1)(A) / 509(a) foundation codes (dropping orgs whose foundation code is missing or in suspense) gives 61,074.
• The BMF-only alternative: 657,426 public charities have IRS ruling years 2015-2023, of which 57,888 currently meet the statutory full-990 thresholds (receipts >= $200k or assets >= $500k). Two independent definitions land in the same range, which is reassuring.

How much of the e-file growth is new organizations?

The anticipated follow-up: did full-990 e-filers grow because new charities formed, or because existing charities were not e-filing in 2015?

new_by_ty <- hdr |>
  filter(
    RETURN_TYPE == "990",
    !is.na(F9_00_YEAR_FORMATION),
    F9_00_YEAR_FORMATION >= 2015, F9_00_YEAR_FORMATION <= 2023,
    F9_00_YEAR_FORMATION <= TAX_YEAR
  ) |>
  distinct(TAX_YEAR, EIN2)

decomp <- hdr |>
  filter(RETURN_TYPE == "990") |>
  count(TAX_YEAR, name = "full_990_filers") |>
  left_join(
    new_by_ty |> count(TAX_YEAR, name = "founded_2015_plus"),
    by = "TAX_YEAR"
  ) |>
  left_join(
    new_by_ty |> semi_join(pc, by = "EIN2") |>
      count(TAX_YEAR, name = "of_which_public_charities"),
    by = "TAX_YEAR"
  ) |>
  mutate(founded_pre_2015 = full_990_filers - founded_2015_plus) |>
  filter(TAX_YEAR >= 2015, TAX_YEAR <= 2023)

knitr::kable(decomp |> mutate(across(-TAX_YEAR, fmt)), align = "lrrrr")

Between TY2015 and TY2023 the annual count of full-990 e-filers rose by 95,517. Organizations founded in 2015 or later account for 48,589 of that increase (50,216 of TY2023’s filers, versus 1,627 in TY2015); organizations founded before 2015 account for the remaining 46,928. Roughly half and half.

The coverage context reframes both halves: per the published e-file catalog, only about 68% of 990/990-EZ filings were electronic in tax year 2015, rising to essentially 100% by 2023 (see the paper-vs-electronic table in catalog-efile-v2_1). Adjusting TY2015’s 358,413 e-filed returns for the missing paper share implies a total filing volume close to TY2023’s 527,764. Total 990/990-EZ filing volume was therefore roughly flat: the pre-2015 organizations in the table above are mostly newly visible rather than newly filing, and the new organizations are largely replacing organizations that dissolved.

Takeaways

• E-filed 990/990-EZ volume grew ~1.5x from tax year 2015 to 2023, with the mandate visible as a step in 2020-2021. Coverage-adjusted (68% electronic in 2015 vs ~100% in 2023), total filing volume was roughly flat: the growth in visible filers is predominantly the e-file mandate.
• Roughly 60-62k public charities founded since 2015 have already grown into the full Form 990. The count tapers for recent cohorts (2022-2023) purely from filing lag: young orgs have not yet filed their first 990.
• Of the growth in annual full-990 e-filers, about half is organizations founded 2015+ entering the data and half is pre-2015 organizations becoming newly visible; with flat total volume, new entrants largely replace dissolved organizations rather than expanding the total.
• Formation year is self-reported on the return; rows where it postdates the tax year are dropped as misreports.
• Orgs that only ever paper-filed (possible before tax year 2021) or only filed the EZ/N despite full-990 size are not counted, so the e-file-based number is a floor.

Reproducibility: this story reads the vintages pinned in _pins.csv; rerun from those pins to reproduce. First render downloads ~3.7 GB of e-file headers into the gitignored data/ cache; later renders are fast. See ADR 0024 / 0025 in nccs-contracts.