Package index • synthid

Panel linkage

The end-to-end entry point: assign a stable cross-year EMP_ID to every record in a compensation panel, and report on the run.

link_panel(): Assign stable cross-year person identifiers to a compensation panel

synthid_report(): Print a summary of a linkage run

synthid_cols(): Standard column names for a titleclassifier/peopleparser panel

Synthetic identifiers

Deterministic, reproducible identifiers. PERSON_YEAR_ID links a record back to its source Part VII row; EMP_ID identifies a person across years.

person_year_id(): Build deterministic per-record person-year identifiers from source keys

create_emp_ids(): Hash a vector of strings into deterministic EMP identifiers

build_id_string(): Build a canonical, human-readable per-record identifier string

Match scoring and model

The Fellegi-Sunter machinery: per-field comparison vectors, the additive hand-weighted score, and the unsupervised EM latent-class model.

candidate_comparisons(): All candidate pairs with their per-field comparison vectors

candidate_scores(): Score all candidate cross-year pairs (for evaluation and threshold tuning)

default_weights(): Default component weights for the person-match score

fit_match_model(): Fit a match model to candidate comparison vectors

predict_match(): Predict a calibrated match probability for comparison vectors

fs_weights(): Learned Fellegi-Sunter agreement / disagreement weights (EM models)

Name comparison

The comparators that make linkage robust: a compound surname comparator, a nickname/phonetic-aware first-name comparator, and surname-rarity weighting.

compare_two_names(): Similarity between two (possibly compound) surnames

compare_first_names(): Compare two first names, with nickname, initial, and phonetic handling

compare_last_names(): Vectorised compound-name comparator

preprocess_name(): Split a (possibly compound) name into comparable tokens

nickname_table(): Nickname / diminutive dictionary

surname_weight(): Within-organization surname-rarity weight, per record

population_surname_weight(): Population surname-rarity weight

Cross-organization linkage

Link the same person across different organizations (interlocking directorates) via person profiles and hash-lookup blocking.

link_cross_org(): Cross-organization person linkage (interlock) pipeline

build_person_profile(): Reduce an EMP_ID cluster to a canonical person profile

blocking_passes(): Blocking passes for cross-organization candidate generation

person_blocking_keys(): Explode person profiles into a long blocking-key table

candidate_pairs(): Generate cross-organization candidate pairs by hash join

score_candidate_pairs(): Score cross-organization candidate pairs from their profiles

filter_edges_by_role(): Filter cross-organization edges by the role composition of their endpoints

board_only_edges(): Board-governance interlock edges

non_board_edges(): Non-board (paid-employee) edges

stamp_xorg(): Stamp cross-org person ids back onto the record-level panel

Title roles

A coarse role taxonomy (BOARD / OFFICER / STAFF / OTHER) over standardized titles, used to reason about the composition of a match.

title_role: Coarse role taxonomy over standardized titles

classify_title_role(): Assign a coarse role to standardized titles

title_role_rules(): Regex rules behind classify_title_role()

title_role_coverage(): Audit the role assignment for a title vocabulary

Link review and parser triage

Post-linkage quality control: flag links for human review and surface upstream name-parser defects for handoff.

flag_links(): Flag questionable cross-year person links for review

link_review_queue(): The residual human-review queue

parse_fail_log(): Record-level parser-defect log for peopleparser

parse_fail_tokens(): Roll a parse-fail log up into candidate title/credential tokens

honorific_tokens() credential_tokens(): Token sets used by the link-review flaggers