loyoladatamining/usajobs_html_metadata

USAJOBS HTML Metadata Dataset (2017-01 to 2026-03)

Dataset Description

The USAJOBS HTML Metadata Dataset is a companion to usajobs_coded, which contains data extracted by the Job Ad Analysis Toolkit, and the raw USAJOBS HTML files at usajobs.

In this metadata version, we present the temporal anchors, canonical agency / department codes, and raw HTML banner audit-trail strings for each posting (keyed by usajobsControlNumber). Canonical codes use the OPM 2-character department + 4-character sub-element scheme, resolved via a cascade of crosswalk lookups (Resh INSIGHT+ posting-table coverage, OPM Personnel Office Identifier matches, exact agency-name matches against the OPM taxonomy, and a cabinet-token regex fallback). Every row carries an agency_resolved_via provenance label so downstream analyses can filter by confidence.

Date Quality Note

The start_date and end_date fields are extracted from monthly JAAT snapshots and USAJOBS HTML-derived posting metadata. A small number of rows contain an apparent source-date inversion where end_date is earlier than start_date (17,348 rows, or about 0.56% of the 3,089,366 posting records). Most inversions are short, but users constructing Monthly Active Jobs or active-month panels should handle these rows explicitly. In the companion dashboard build, these rows are excluded from active-month expansion because they cannot define a valid active interval without an additional correction rule.

Dataset Structure

Splits

This dataset is organized by month to allow for easy time-series analysis without requiring the user to download the entire multi-year corpus, and to mirror usajobs_coded so the two can be joined split-by-split on usajobsControlNumber.

• Monthly Splits: Format YYYY_MM (e.g., 2017_01, 2025_12).

Data Fields

Each split contains the following fields:

• usajobsControlNumber: Unique identifier code for the USAJOBS job posting. Same key as in usajobs_coded.
• start_date: Date the posting first appears active in the JAAT monthly snapshots.
• end_date: Date the posting last appears active.
• source_month: YYYY-MM of the JAAT monthly partition this posting was extracted from; matches the split filename.
• dept_code: OPM 2-character department code (e.g. VA, AR, TR). NULL when no signal resolved.
• dept_name: Canonical department name in title case (e.g. Department of Veterans Affairs).
• agency_code: OPM 4-character sub-element code (e.g. VATA, ARTC). Set to <DD>OTH (e.g. VAOTH) for rows resolved only at the cabinet level.
• agency_name: Canonical sub-agency name in title case. For cabinet-only rows, set to Other (<dept_name>).
• agency_resolved_via: Provenance of the code assignment. One of: cleaned_jobid (overlap with detailed Resh et al. data), resh_jobid (direct overlap with the full Resh et al. INSIGHT+ posting table), name_exact (exact (dept_name, agency_name) match into the OPM POI taxonomy - https://data.opm.gov/data-standards/personnel-office-identifier), name_exact_taxonomy_agency (exact match of the raw agency string against any taxonomy agency_name), name_exact_resh_bureau (exact match against Resh et al's full bureau-string set), name_exact_resh_dept (cabinet-only string match), cabinet_token (regex on cabinet substring, dept-only), title_unique (vacancy title appears in Resh et al. under exactly one sub-agency), manual (hand-edited override), or NULL (unresolved).
• html_dept_raw: The raw department string extracted from the USAJOBS HTML banner via regex. These are noisy strings.
• html_agency_raw: Same, for the agency banner field.

Citation

If you find this dataset useful or utilize it for your work, please consider citing our working paper:

@article{meisenbacher2025extracting,
  title={Extracting O* NET Features from the NLx Corpus to Build Public Use Aggregate Labor Market Data},
  author={Meisenbacher, Stephen and Nestorov, Svetlozar and Norlander, Peter},
  year={2025}
}

The html_dept_raw and html_agency_raw fields build off of work completed by Resh et al. (2025); if you make use of these fields, please also cite the original data curators:

@article{Resh2025,
  author = "William Resh and Keunyoung Lee and Yi Ming",
  title = "{U.S. Federal Civil Position Job Postings (2018-2023)}",
  year = "2025",
  month = "2",
  url = "https://figshare.com/articles/dataset/U_S_Federal_Civil_Position_Job_Postings_2018-2023_/28509314",
  doi = "10.6084/m9.figshare.28509314.v5"
}