AutoStata-Insight

Language: English | 中文

AutoStata-Insight is an empirical-research automation tool. Starting from an Excel dataset, it can run data reading/cleaning, variable mapping, Stata analysis, chart/table export, and Word report generation.

It is suitable for panel data, cross-sectional survey data, econometrics teaching demos, prototypes, and first-pass empirical analysis. Formal papers still require human review.

Architecture

flowchart LR
  Researcher["Researcher"] --> Excel["Excel input"]
  Excel --> Metadata["Metadata agent"]
  Metadata --> Mapping["Variable mapping JSON"]
  Mapping --> DoFile["Generated Stata .do file"]
  DoFile --> Stata["PyStata / Stata engine"]
  Stata --> Outputs["Tables, logs, charts"]
  Outputs --> Reporting["Reporting agent"]
  Reporting --> Word["Word empirical report"]

Loading

Portfolio Metrics

Local report-pipeline baseline for portfolio review. Re-benchmark with the target dataset, Stata edition, and model before claiming production performance.

Metric	Current portfolio baseline	Measurement note
Latency	First analysis step target < 10s	Excel scan + schema preview on local machine
RAG hit rate	N/A	Structured data analysis, not vector retrieval
Agent success rate	Target >= 90%	Variable-role mapping accepted without manual correction on benchmark sheets
Report generation time	Target < 120s	Excel to .do + Stata outputs + Word report
Cost	~$0.01-$0.08 / report	Depends on Qwen text/VL calls and chart-commentary volume

Features

• Read the latest Excel file from input/.
• Recognize Chinese headers and map them to Stata-compatible variable names.
• Infer dependent variables, independent variables, controls, panel IDs, and time variables.
• Run descriptive statistics, correlation, VIF, OLS/Logit/Probit, FE/RE panel regression, Hausman tests, robustness checks, heterogeneity analysis, and 2SLS when applicable.
• Generate reproducible .do files and Word reports.

Requirements

• Python 3.12+
• Local Stata installation with working PyStata
• DashScope/Qwen API key

Run

pip install -r requirements.txt
python main.py

or:

uv sync
uv run python main.py

Outputs

Each run writes artifacts under output/<timestamp>/:

• variable_mapping.json
• Stata logs
• charts and tables
• generated .do file
• Word empirical report