Internal toolkit

EconAI

The causal inference toolkit that runs behind every BDPolicy Lab brief that estimates a treatment effect. Twelve estimators, the figures and tables they produce, an OpenAlex literature search, and a research gap finder.

12
Estimators
5
Figure types
3
Table types
2
Literature APIs

Estimators

Twelve causal designs

Each estimator returns a standardized result object that the figure and table builders consume directly. Robust standard errors and clustered inference where the design supports it.

OLS
Heteroskedasticity-robust standard errors, optional cluster bootstrap.
IV / 2SLS
Just-identified and over-identified, with first-stage F and Anderson-Rubin tests.
Panel FE
Two-way fixed effects with clustered errors at the panel dimension.
Difference-in-Differences
Two-period DiD with parallel-trends placebo support.
Staggered DiD
Callaway-Sant'Anna group-time average treatment effects.
Synthetic DiD
Arkhangelsky et al. (2021) unit and time weights with placebo inference.
RDD
Sharp and fuzzy designs, local-linear with bias correction.
Double / Debiased ML
Chernozhukov et al. (2018) cross-fitting for high-dimensional controls.
Causal Forest
Heterogeneous treatment effects with honest splits.
Shift-Share IV
Bartik instruments with Adao-Kolesar-Morales standard errors.
Manski Bounds
Partial identification under monotone treatment response.
Randomization Inference
Permutation-based exact p-values for sharp nulls.

Figures

Plots

  • Binscatter with linear or LOWESS overlays.
  • Coefficient plot for cross-spec comparisons.
  • Distribution with kernel density and histogram.
  • Event study with dynamic treatment effects and confidence bands.

Tables

LaTeX-ready

  • Regression with stars, clustered SEs and fit statistics.
  • Balance across treatment and control with normalized differences.
  • Summary statistics by group with N, mean, SD, min, max.

Literature

Search the canon

OpenAlex and Semantic Scholar wrappers retrieve papers, abstracts and citation graphs. BibTeX export for citation pipelines. Used to ground every brief in the existing literature before drafting.

Research gap finder

Where the holes are

A topic-design matrix that flags Bangladesh-relevant questions where the data exists, the design is feasible, and the literature is thin. Drives the editorial calendar.

EconAI lives in the codebase, not on the public API surface. Treat it as the engine room behind the briefs you read on this site, not as a self-serve tool. If you want to verify a specific number or run a robustness check, get in touch.