Skip to contents

Group a candidate list into operating/arm clusters

Source: R/affiliation.R
pf_cluster_arms.Rd

For each query, finds candidate pairs that share a name core, sit in the same state, and split into one arm and one non-arm. Each such pair gets a joint_id; the disclosure sources, when supplied, are attached as evidence.

Usage

pf_cluster_arms(
  cand,
  id = "id",
  ein = "ein",
  name = "name",
  state = "state",
  links = NULL,
  require_city = NULL
)

Arguments

cand

A candidate frame, one row per query/candidate pair, with columns named by id, ein, name and state.

id, ein, name, state

Column names in cand.

Optional output of pf_build_affiliation_links().

require_city

Optional column name; when given, the pair must also agree on it. Tighter, and disclosure rates are higher on the subset, but it drops arms that use a lockbox or a separate mailing address.

Value

A data.table, one row per detected cluster, with joint_id, the two EINs and names, their ranks in the candidate list, evidence (SCHED_A / SCHED_R / NAME_GEO_ONLY) and direction.

Details

Roughly 13% of MAYBE grants contain such a pair. Collapsing the pair removes the runner-up from the tie band, which converts MAYBE-because-two-candidates-tie into a single-body decision.

See also

pf_arm_order() for choosing between the members.