Data¶
The project turns employment histories and external firm data into model outputs and analysis-ready datasets. This page shows only the main path; each dataset page documents its exact grain, inputs, builder, location, and consumers.
flowchart TB
revelio[(Revelio employment data)]
linked[(Compustat, CRSP, patents,<br/>executives, and governance data)]
employment[(Employment data)]
firmdata[(Reusable firm-linked data)]
periods[(Model-ready relationship periods)]
hct[HCT training and scoring]
scores[(Model outputs)]
analysis[(Analysis-ready datasets)]
revelio --> employment --> periods --> hct --> scores --> analysis
linked --> firmdata --> analysis
employment --> firmdata
classDef source fill:#e8eef7,stroke:#52739e,stroke-width:2px
classDef data fill:#e2f2e7,stroke:#3b7a4b,stroke-width:2px
classDef model fill:#fff0cc,stroke:#a66b00,stroke-width:2px
classDef output fill:#eee5f5,stroke:#76528c,stroke-width:2px
class revelio,linked source
class employment,firmdata,periods data
class hct model
class scores,analysis output
How the documentation is organized¶
Source data¶
Files obtained from vendors or external providers. These pages describe how the data were obtained, the source files available, and the identifiers used by the project. Source files are treated as immutable.
Processed datasets¶
Project-built datasets that clean, reshape, sample, or link source data. Each page begins with a small input–transformation–output diagram and identifies the dataset's grain and keys. The pages are grouped into employment, accounting and market, patent, and executive and governance data.
Model outputs¶
Datasets written by an HCT scorer. The training outputs preserve an
out-of-sample worker group; full_training outputs provide the final MQ measure
used by most downstream analyses. Training itself is documented under
Model.
Analysis datasets¶
Datasets built for statistical analyses, grouped by their unit and research
use. LaTeX tables, fitted models, metrics, and figures are not data products;
they are stored under /artifacts and documented with their analysis.
Storage conventions¶
/data/<coverage>/processed/ canonical processed and model-ready data
/data/<coverage>/analyses/ coverage-level analysis support data
/data/<coverage>/analyses/<mode>/ model outputs and mode-specific analysis data
/artifacts/<coverage>/ checkpoints and analysis results
<coverage> is samples, one_percent, or full. <mode> is training or
full_training. These are the intended conventions, but some legacy products
remain under raw or analyses; those exceptions are stated on their dataset
pages. Some firm-linked products exist only at full coverage.