Dynamic Panel BMA: Which Factors Truly Drive Economic Growth?
Status: Script reviewed (ACCEPT)
Language: R
Last run: 2026-04-01
Overview
Bayesian Model Averaging for dynamic panel data with weakly exogenous regressors using the bdsm R package. Analyzes which of 9 candidate growth determinants truly drive economic growth across 73 countries over 4 decades (1970–2000), accounting for reverse causality through lagged dependent variables and fixed effects. Improvement over v1: data preparation demeaning by both year AND entity (country).
Pipeline Progress
| # |
File |
Description |
| 1 |
r_bdsm_model_pmp.png |
Prior vs posterior model probabilities (bdsm built-in) |
| 2 |
r_bdsm_model_sizes.png |
Model size distribution (bdsm built-in) |
| 3 |
r_bdsm_coef_hist_pop.png |
Population coefficient histogram (bdsm built-in) |
| 4 |
r_bdsm_sizes_dilution.png |
Model sizes under dilution prior (bdsm built-in) |
| 5 |
r_dynamic_bma2_pip.png |
PIP bar chart – custom ggplot2 dark theme |
| 6 |
r_dynamic_bma2_coef.png |
Coefficient estimates with 95% CIs – custom ggplot2 dark theme |
| 7 |
r_dynamic_bma2_sensitivity.png |
Prior sensitivity dumbbell chart – custom ggplot2 dark theme |
Generated Tables (CSV)
| # |
File |
Description |
| 1 |
economic_growth.csv |
Raw source dataset from bdsm package |
| 2 |
original_economic_growth.csv |
Source dataset with lagged GDP column |
| 3 |
data_prepared.csv |
Final processed dataset (standardized + demeaned by year + entity) |
| 4 |
fe_regression.csv |
Kitchen-sink fixed effects regression coefficients |
| 5 |
bma_binomial.csv |
BMA results with binomial prior (EMS = 4.5) |
| 6 |
bma_binomial_beta.csv |
BMA results with binomial-beta prior |
| 7 |
best_models_inclusion.csv |
Top 8 models inclusion matrix |
| 8 |
best_models_estimates.csv |
Top 8 models coefficient estimates |
| 9 |
prior_sensitivity.csv |
PIPs across all prior specifications |
| 10 |
jointness_hcghm.csv |
HCGHM jointness matrix |
Datasets
| File |
Rows |
Cols |
Description |
economic_growth.csv |
365 |
12 |
Raw panel data (73 countries, 5 periods, 9 regressors) |
original_economic_growth.csv |
292 |
13 |
Panel data with lagged GDP (excludes initial period) |
data_prepared.csv |
365 |
12 |
Standardized + demeaned by year + demeaned by entity |
Pipeline Documents
| File |
Description |
plan.md |
Approved scope and design decisions |
script-review.md |
Code review report (verdict: ACCEPT) |
README.md |
This file – artifact inventory and status |
Packages
bdsm – Bayesian Dynamic Systems Modeling
tidyverse – data manipulation and visualization
parallel – parallel computing for model space estimation
scales – number formatting for ggplot axes