StatsPAI_skill

What it does

Guides a user through a full empirical/causal analysis workflow in Python, oriented to producing paper-ready artifacts such as Table 1, Table 2, event-study figures, and robustness panels. Emphasizes an estimand-first DSL, identification graphs, and a pipeline that yields exportable results to Word/Excel/LaTeX. Supports multiple flavors (AER-style econometrics, epi/public health causal inference, ML causal methods, and distributional/gap-decomposition approaches).

How it works

The skill describes importing the package as import statspai as sp and running a start-to-finish sequence that includes:

• Generating Table 1 via sp.mean_comparison(df, ...) and exporting to Word/Excel/LaTeX.
• Defining an estimand-first plan with sp.causal_question(...).identify().
• Creating an identification figure using sp.callaway_santanna(...) and plotting with sp.enhanced_event_study_plot(...).
• Building main regression tables through sp.regress(...) and sp.feols(...), then exporting with rt.to_word(...), rt.to_excel(...), and rt.to_latex().
• Heterogeneity analysis via a metalearner like sp.metalearner(...) and plotting with sp.cate_plot(...).
• Robustness analyses using sp.oster_bounds(...), sp.evalue(...), and sp.sensitivity_plot(...) with a reference from sp.honest_did(...).
• Creating a one-file replication bundle with sp.collect(...) and saving multiple formats.

When to use it

Triggers include keywords like causal inference in python, applied microeconomics pipeline, AER empirical analysis, QJE style robustness, DID IV RD, callaway_santanna, synthetic control, double machine learning, causal forest, event study plot, Oster bound, honest_did, and various decomposition and export tasks. The skill emphasizes an end-to-end pipeline that starts from data contract to replication export.

What it can touch

The workflow relies on the StatSPAI API surface (e.g., sp.mean_comparison, sp.causal_question, sp.callaway_santanna, sp.regress, sp.feols, sp.metalearner, sp.cate_plot, sp.oster_bounds, sp.evalue, sp.sensitivity_plot, sp.collect). It also references optional extras to install such as statspai[fixest], statspai[plotting], statspai[neural], and statspai[text] to enable specific estimator stacks and plotting capabilities. Exports to Word/Excel/LaTeX are performed via methods like .to_word(), .to_excel(), .to_latex(), and sp.collect saves.

Caveats

Install requires extras to enable certain estimators and plotting backends; missing extras raise ImportError (e.g., pyfixest for sp.feols, matplotlib/seaborn for plots, torch for neural causal modules, and text for text-as-treatment features). The description notes that a full pipeline requires the extras to be installed and that the bare pip install statspai is not sufficient for default pipeline functions.