Stata

Stata Skill

Use this skill whenever writing, editing, running, or debugging Stata .do files, or when you need to look up Stata syntax, commands, or options.

1. What is Stata

Stata is a statistical software package for data management, analysis, and graphics. It is widely used in economics, political science, epidemiology, and other social sciences.

Key concepts:

• .do files: Scripts (plain text) containing Stata commands, executed sequentially.
• .dta files: Stata's binary data format.
• .log files: Text output captured during a Stata session or batch run.
• ado files: Stata programs (user-written or official) that extend functionality.
• Macros: local (temporary) and global (persistent within session) named values.
• Factor variables: i.varname notation for categorical regressors.
• Postestimation: Commands run after a model fit (e.g., predict, margins, estat).

2. Running Stata in This Environment

Finding the installation: Stata is installed in C:\Program Files\ on Windows. To auto-detect the path:

STATA_DIR=$(ls -d "/c/Program Files"/Stata* "/c/Program Files"/StataNow* 2>/dev/null | sort -V | tail -1)
echo "$STATA_DIR"

From bash (Claude Code terminal):

stata -b do path/to/script.do        # batch mode, creates .log file
stata path/to/script.do              # interactive window

IMPORTANT: ALWAYS use the stata wrapper command (at ~/bin/stata), NEVER call StataSE-64.exe directly. The wrapper auto-moves batch-mode logs from the current directory to quality_reports/stata_logs/. Calling the exe directly leaves stray .log files in the project root.

From PowerShell (user terminal):

stata -b do path\to\script.do

The stata alias must point to StataSE-64.exe (or StataMP-64.exe depending on edition). See SETUP.md for how to configure this.

Checking for errors after batch run:

grep "^r(" script.log    # Stata error codes start with r(

If the log contains r( lines, the script hit an error at that point.

3. Writing .do Files — Essentials

File structure

version 17
clear all
set more off

* --- 0. Paths ---
do config_local.do          // sets $root

* --- 1. Load data ---
use "$root/data/processed/tweets_processed.dta", clear

* --- 2. Analysis ---
reg vote_share engagement_score, robust

* --- 3. Output ---
esttab using "$root/output/tables/table1.tex", booktabs label replace
graph export "$root/output/figures/fig1.png", replace width(2400)

Key patterns

• version 17 at top for reproducibility
• All paths via $root global macro (set in config_local.do)
• clear all / set more off at the start
• replace on all output commands (allows re-running)
• width(2400) for ~300 DPI figures at 8 inches
• graphregion(color(white)) bgcolor(white) for white backgrounds
• Line continuation with ///

Common pitfalls

• encode assigns codes alphabetically — verify ordering before interpreting
• sort is not stable — add enough keys to uniquely identify rows
• Use vce(cluster var) not just robust when observations are grouped
• destring for numeric-as-string; encode for true categoricals — never confuse them

4. Stata PDF Documentation

Location

Stata bundles its full documentation as PDFs inside the installation directory:

<STATA_INSTALL_DIR>/docs/

To find the docs directory automatically:

STATA_DOCS=$(ls -d "/c/Program Files"/Stata*/docs "/c/Program Files"/StataNow*/docs 2>/dev/null | sort -V | tail -1)
echo "$STATA_DOCS"

Complete manual index (37 PDFs, ~17,000 total pages)

Start with stoc.pdf (59 pages) to find which manual covers a topic.

5. Reading Documentation Efficiently (Token-Saving Strategies)

Problem: 17,000 pages of PDFs. Reading even one manual wastes tokens. Solution: Use targeted extraction — never read a full manual.

Prerequisites: pdftotext (bundled with poppler/mingw) and pip install pdfplumber. See SETUP.md for installation.

Tool 1: pdftotext (fast, plain text, best for prose)

# Auto-detect docs path
STATA_DOCS=$(ls -d "/c/Program Files"/Stata*/docs "/c/Program Files"/StataNow*/docs 2>/dev/null | sort -V | tail -1)

# Extract specific pages (e.g., pages 1200-1220 for regress)
pdftotext -f 1200 -l 1220 "$STATA_DOCS/r.pdf" -

# Search the subject TOC for a command
pdftotext "$STATA_DOCS/stoc.pdf" - | grep -i "regress"

Tool 2: pdfplumber (Python, best for tables and structured content)

import pdfplumber, glob, os

def find_stata_docs():
    """Auto-detect Stata docs directory."""
    for pattern in [r"C:\Program Files\StataNow*\docs",
                    r"C:\Program Files\Stata*\docs"]:
        matches = glob.glob(pattern)
        if matches:
            return sorted(matches)[-1]
    return None

def stata_doc_lookup(manual: str, start_page: int, end_page: int) -> str:
    """Extract text from a Stata manual. Pages are 0-indexed."""
    docs = find_stata_docs()
    path = os.path.join(docs, manual)
    with pdfplumber.open(path) as pdf:
        text = []
        for i in range(start_page, min(end_page, len(pdf.pages))):
            page_text = pdf.pages[i].extract_text()
            if page_text:
                text.append(page_text)
    return "\n".join(text)

# Example: read TOC of r.pdf to find page numbers
print(stata_doc_lookup("r.pdf", 2, 8))

Tool 3: pdftotext + grep (search without reading)

STATA_DOCS=$(ls -d "/c/Program Files"/Stata*/docs "/c/Program Files"/StataNow*/docs 2>/dev/null | sort -V | tail -1)

# Find which page mentions "margins" in the base reference
pdftotext "$STATA_DOCS/r.pdf" - | grep -n "margins"

# Find a command across ALL manuals
for f in "$STATA_DOCS"/*.pdf; do
    if pdftotext "$f" - 2>/dev/null | grep -q "didregress"; then
        echo "Found in: $(basename $f)"
    fi
done

Recommended lookup workflow

1. Start with stoc.pdf — search the subject TOC to identify which manual
2. Read the manual's own TOC (pages 2-8) to find the exact page range
3. Extract only those pages with pdftotext -f START -l END manual.pdf -
4. Never extract more than 20 pages at once — if you need more, narrow the search

Token cost estimates

• 1 PDF page ~ 500-800 tokens
• Full r.pdf ~ 2.1M tokens (NEVER do this)
• stoc.pdf (59 pages) ~ 35K tokens (acceptable for initial lookup)
• Targeted 10-page extract ~ 6K tokens (ideal)

6. Quick Reference: Common Tasks