Standard Errors Tutorials¶
Learning Path
Prerequisites: Fundamentals tutorials, basic econometrics Time: 3--8 hours Level: Beginner -- Advanced
Overview¶
Correct standard errors are essential for valid inference. Panel data frequently violates the classical OLS assumptions: errors may be heteroskedastic, serially correlated within entities, or cross-sectionally dependent across entities. Using the wrong standard errors leads to inflated t-statistics, misleadingly small p-values, and incorrect confidence intervals.
These tutorials cover the complete range of robust inference methods available in PanelBox: heteroskedasticity-consistent (HC0--HC3) standard errors, cluster-robust errors, Newey-West HAC for serial correlation, Conley spatial HAC for geographic correlation, Driscoll-Kraay for cross-sectional dependence, Panel Corrected Standard Errors (PCSE), and bootstrap methods.
The existing Standard Errors Series provides additional depth including a decision tree for method selection.
Which Method Should I Use?¶
| Data Characteristic | Recommended Method | Notebook |
|---|---|---|
| Heteroskedasticity only | HC0--HC3 | 01 |
| Within-entity correlation | Cluster-robust | 02 |
| Serial correlation (known structure) | Newey-West HAC | 03 |
| Geographic/spatial correlation | Conley Spatial HAC | 04 |
| Cross-sectional dependence (large T) | Driscoll-Kraay | 06 |
| Small sample, unknown structure | Bootstrap | 07 |
| MLE models | Sandwich estimator | 05 |
Notebooks¶
| # | Tutorial | Level | Time | Colab |
|---|---|---|---|---|
| 1 | Robust Fundamentals (HC0--HC3) | Beginner | 45 min | |
| 2 | Clustering in Panels | Beginner | 45 min | |
| 3 | HAC (Newey-West) | Intermediate | 45 min | |
| 4 | Spatial HAC (Conley) | Intermediate | 60 min | |
| 5 | MLE Sandwich Inference | Intermediate | 45 min | |
| 6 | Driscoll-Kraay & PCSE | Advanced | 60 min | |
| 7 | Methods Comparison | Advanced | 60 min |
Learning Paths¶
Basic (3 hours)¶
Essential inference for any panel analysis:
Notebooks: 1, 2
Covers robust SE basics and clustering. These two methods handle the vast majority of applied work.
Intermediate (5 hours)¶
Add HAC and sandwich methods:
Notebooks: 1, 2, 3, 4, 5
Includes Newey-West, Conley spatial HAC, and sandwich inference for MLE models.
Advanced (8 hours)¶
Master every inference method:
Notebooks: 1--7
Adds Driscoll-Kraay, PCSE, bootstrap, and a systematic comparison of all methods.
Key Concepts Covered¶
- HC0--HC3: Heteroskedasticity-consistent SE (White, MacKinnon-White)
- Cluster-robust: Accounting for within-cluster correlation
- Two-way clustering: Simultaneous entity and time clustering
- HAC (Newey-West): Heteroskedasticity and autocorrelation consistent
- Spatial HAC (Conley): SE for spatially correlated data
- Driscoll-Kraay: Cross-sectionally dependent panels (large T)
- PCSE: Panel Corrected SE (Beck & Katz)
- MLE sandwich: Robust SE for maximum likelihood estimators
- Bootstrap: Nonparametric, wild, and block bootstrap
Quick Example¶
import panelbox as pb
# FE with cluster-robust SE
fe = pb.FixedEffects(
data=data,
formula="y ~ x1 + x2",
entity_col="id",
time_col="year",
cov_type="clustered"
).fit()
print(fe.summary())
Related Documentation¶
- Standard Errors Series -- Detailed tutorial with decision tree
- Inference -- Standard error theory and methods
- User Guide -- API reference