Build the dyad table

Reads the Schedule A and Schedule R dyad extracts, keeps rows where both ends carry a usable EIN, symmetrises to one row per unordered pair, and classifies each pair as substitutable or related.

Usage

pf_build_dyads(
  supported = pf_dyad_sources()$supported,
  related = pf_dyad_sources()$related,
  bmf = NULL,
  min_years = 1L,
  max_fanout = 3L,
  max_degree = 5L,
  cache = NULL,
  verbose = TRUE
)

Arguments

supported

Path to dyad_supported_orgs.csv, or the table itself. NULL to skip Schedule A.

related

Path to dyad_related_exempt_orgs.csv, or the table itself. NULL to skip Schedule R.

bmf

Processed BMF (path or data frame). Supplies the names used for the name-core test and the subsection codes used to veto cross-subsection substitution. Strongly recommended; without it the filed names are used and no subsection veto is applied.

min_years

Drop pairs disclosed in fewer than this many tax years. The default keeps everything; 2 removes 12% of pairs and is a reasonable setting if single-year disclosures look like data entry noise to you.

max_fanout

A Schedule A edge licenses substitution only if the supporter names at most this many organisations. Above it the filer is a grantmaking supporting organisation, not an arm.

max_degree

No EIN may end up with more than this many substitutable partners; edges touching one that does are demoted to related. This is what removes federated networks.

cache

Path to cache the result as .rds. NULL to skip.

verbose

Print progress.

Value

A keyed data.table with ein_a < ein_b and columns dyad_id, dyad_class, dyad_reason, src_a (Schedule A seen), src_r (Schedule R seen), sched_a_dir (a_supports_b / b_supports_a / NA), fan_out, n_years, first_year, last_year, same_state, same_city, name_a, name_b, in_bmf. Deliberately not given an S3 class: it is a table people subset constantly, and a class would hijack the print method of every subset. Use pf_dyad_summary().

See also

pf_dyad_collapse(), pf_dyad_annotate(), pf_cluster_arms()