-
Notifications
You must be signed in to change notification settings - Fork 0
Stats Regression panel
One entry point, PanelRegression, for data that follows the same
entities over several periods: firms by year, patients by visit, regions by month. An entity's
unobserved, time-invariant traits bias least squares when they move with the regressors; the four
estimators here each deal with them differently — projecting them out, averaging them, differencing
them away, or treating them as a random draw — and each reports the table linearmodels does,
under a covariance that lets errors correlate within an entity, within a period, or along time.
No .NET library publishes a panel estimator: Math.NET Numerics, Accord and ML.NET fit pooled least
squares and stop
(decision 0004). The
estimators are closed forms over transformed rows, so they are replayed against linearmodels at
1e-9 from a frozen corpus.
| Type | What it is |
|---|---|
PanelRegression |
Fits fixed effects, between, first-difference and random-effects regressions. |
The data it takes and returns — PanelDesign,
PanelOptions, PanelSummary and
PanelCovarianceType — live in
Lodestar.Stats.Regression.Panel.
- Ordinary least squares — the pooled regression, without the panel's structure.
- Shared regression types — the tests and kernels.
- Python → C# equivalence.