Projects | Keegan's Time Series

Enterprise Workforce Laboratory

Mon, 02 Nov 2026 00:00:00 +0000

A deliberately separate firm-level model

Economy-wide labor statistics describe the aggregate outcome of decisions, while enterprise systems record the administrative details of employment relationships. This project uses synthetic enterprise records, not confidential employer data, to demonstrate how those details can be modeled.

Enterprise semantic model

Synthetic enterprise Power BI model (development snapshot). Dimensions and event/snapshot facts remain separate to preserve their distinct grains. Some date relationships are inactive and must be addressed deliberately in measures.

The model distinguishes people, employment spells, assignments, time-bound snapshots, compensation events, leave episodes, and applications. These entities represent different analytical units and should not be merged into a single generic employee table.

The diagram includes:

Dimensions for person, job, organization, and date.
Employment spell, assignment history, and workforce month-end facts.
Compensation events, leave episodes, and recruiting applications.
DAX measures for headcount, hires, separations, and workforce reconciliation.

Analytical questions

How sensitive is turnover to the selected separation events and denominator? How do month-end counts differ from exposure-based measures? What changes when a model uses employment spells and time-varying covariates rather than a one-row-per-person extract?

Modeling roadmap

The synthetic dataset is intended to support reproducible data-quality tests, descriptive dashboards, survival modeling, and predictive diagnostics. Any reported model performance will require actual fitted results and documented validation.

Status: Architecture under development. The diagram demonstrates structure; it does not claim deployed forecasting or prediction performance.

Open Workforce Analytics Standard (OWAS)

Mon, 02 Nov 2026 00:00:00 +0000

The question

When two organizations report the same workforce metric, are they measuring the same thing?

OWAS is an independent, evidence-based reference project on workforce metrics, temporal structure, data architecture, and implementation. It distinguishes authoritative external standards, published research, common practices, OWAS recommendations, and open methodological questions. Its recommendations are not presented as ISO or SHRM requirements.

From definition to implementation

The analytical approach specifies, as applicable, metric purpose, numerator, denominator, eligible population, grain, reference period, aggregation, variants, edge cases, data quality, privacy, and external sources.

Technical work developed alongside the reference includes:

R ingestion, transformations, validations, and Parquet exports for public labor-market statistics.
A Power BI semantic model that preserves source metadata and incompatible grains.
A separate synthetic enterprise model covering persons, employment spells, workforce snapshots, and event-level facts.
Dashboard views for employment, labor movement, and compensation.

Examples of the implemented work

See Workforce economics for the public-data dashboards and Enterprise workforce laboratory for the synthetic semantic model. These examples demonstrate technical implementation; they are not themselves formal OWAS requirements.

Why the micro/macro connection matters

Within a firm, headcount, hiring, separation, and compensation begin as administrative transactions. At the level of the economy, related phenomena appear as employment stocks, labor flows, and wage or cost indexes. Questions connecting these levels are useful, but inference must respect their very different data-generating processes.

Status: Independent work in active development. Evidence and methodology should be assessed at the specific metric or study level rather than as a blanket endorsement of the project.

Workforce Economics: Public Data and Semantic Modeling

Mon, 02 Nov 2026 00:00:00 +0000

A labor market, viewed through different instruments

What can employment, worker flows, and compensation tell us about the adjustment of the labor market? The first question is whether the statistical programs involved describe comparable populations and events.

This project uses publicly reported series from BLS CES, CPS, JOLTS, ECI, and Census QWI and J2J, with BED and BDS being incorporated into the broader model. A dashboard is a presentation layer over a source-aware measurement system, not a substitute for one.

Dashboard: employment and participation

Macro overview (Power BI, OWAS v0.4). Payroll employment counts jobs; CPS unemployment and participation describe people. The estimates should not be treated as interchangeable.

The overview pairs CES nonfarm payroll employment with CPS household-based labor-market measures. The distinction between jobs and people is essential: the charts answer related questions with different underlying universes.

Dashboard: worker flows

Workforce movement (Power BI, OWAS v0.4). JOLTS establishment estimates, Census job-to-job transitions, and QWI administrative job flows have different definitions, reference periods, and coverage.

This view juxtaposes establishment-survey flows, job-to-job transitions, and Texas administrative job flows. Putting the series near each other is a prompt for analysis, not a claim that a count in one program can be equated to a count in another.

Dashboard: compensation

Compensation (Power BI, OWAS v0.4). CES hourly earnings, ECI compensation indexes, and QWI stable-job earnings are distinct measures; level comparisons are not warranted without careful harmonization.

The compensation view brings together average hourly earnings, compensation cost indexes, and administrative stable-job earnings. Their different units and populations are precisely why the model retains measurement metadata.

Data engineering and modeling

The workflow uses R for ingest, cleaning, validation, and exports; Parquet for reusable outputs; and Power BI for dimensional relationships and reporting measures. Data-quality checks include missingness, time-period consistency, source metadata, and unit interpretation.

Interpretation and limitations

The screenshots are a development snapshot (v0.4), not a real-time economic release. Metrics may have different latest periods, vintage policies, seasonal-adjustment statuses, and geographic coverage. Apparent correlations do not establish firm-level mechanisms or causal relationships. Reproduction instructions and an updated technical diagram are planned.