Searching for Help

Everyone searches for help

Professional programmers and applied economists constantly look things up.
You are not expected to memorize every command and option in Stata.
You are expected to learn how to:
- Search effectively
- Read help pages and online answers
- Adapt examples safely to your own work

Today we’ll focus on three sources:

Built-in Stata help and documentation
Online help (Stata docs, Statalist, Stack Overflow, blogs)
Large Language Models (LLMs) like ChatGPT

Built-in Stata help

Before you open a browser, try Stata’s own tools.

`help` for a specific command

* open the help file for summarize
    help            summarize

Shows syntax
Lists options
Often includes examples at the bottom (pay attention to these)

If you know roughly what you want but not the exact syntax (e.g., xtile, merge, collapse), help is your friend.

`search` for related commands

* search for commands related to "percentile"
    search          percentile

This looks through:

Command names and descriptions
Installed help files
Sometimes user-written packages

You can click on entries in the Viewer to open help pages.

Habit to build: when you see a new command in lecture or sample code, run help commandname and skim the help file.

Good targets for Stata help

Places where help/search are especially useful:

Basic data management commands
- help generate, help replace, help egen
- help merge, help append, help collapse
Graphics
- help graph, help histogram, help twoway
Common analysis commands
- help regress, help ttest

When debugging, also consider help r(###) for error codes.

Do Exercise 5 - Stata Help

Online Stata resources

Some high-value websites:

Official Stata documentation
- Mirrors the help system
- Searchable on the web
- Very precise, includes lots of examples
Statalist
- Long-running forum where people ask detailed Stata questions
- Often your exact error or task has been discussed
Stack Overflow / Cross Validated
- More general programming/stats Q&A
- Stata content mixed in with R, Python, etc.

How to search effectively

When you go to Google (or DuckDuckGo, etc.), your goal is to express your problem in search-engine language. One thing I’ve learned in 10 years of teaching is that students often struggle with developing useful prompts or key-words for searches. The returns you get from a search are only as good as what you ask it to search for.

Get the vocabulary right

Include:

stata
The command (if you know it)
A short phrase about what you want
Or the error code

Good examples:

stata collapse mean by group
stata merge 1:m not uniquely identified
stata graph bar mean by group
stata r(111) variable not found

Less good:

collapse not working
missing unique identifier
how do I write bar graph code?
stata help please

Avoid extra words

Start simple:

stata reshape wide long example
stata -999 missing value

You can always add words if results are off.

Reading results: help sites, forums, blogs

When you click a search result:

Help sites / Q&A (Statalist, Stack Overflow)

Read the question carefully
- Does it sound like your situation?
- Same command? Same type of data?
Look at the setup
- Are they reshaping wide-to-long or long-to-wide?
- Are their variable names similar?
Check the age
- Stata syntax is fairly stable, but a 2003 post may reference ancient versions
Check the top answers
- On Stack Overflow, higher-voted answers are usually better
- On Statalist, look for answers from known experts
Glance at the comments
- Sometimes corrections or caveats are discussed there
Test and modify the example
- Never copy-paste blindly
- Adapt variable names and paths to your data
- Confirm that the code does what you think it does

Blog posts and tutorials

Scan for the section you need
- Use your browser’s Find (Ctrl+F / Cmd+F)
- Search within the page for “merge”, “collapse”, etc.
Focus on code blocks
- They usually contain the key ideas
- Try to understand them, then rewrite into house style before using

Testing and integrating code you find

When you grab code from online (or from an LLM):

Try it on a small subset
- Add in 1/50 or a restrictive if to avoid damaging your whole dataset
Watch for paths and filenames
- Change "C:\Users\someone\Downloads\mydata.dta" to your project paths and globals
Check for version issues
- Some commands/options only exist in newer versions of Stata
- If you see version lines in example code, pay attention
Check that it respects course rules
- Doesn’t overwrite raw data
- Keeps data and code separated
- Fits into your project structure (globals, logs, etc.)
Rewrite in your own words and style
- Rename variables to match the rest of your project
- Add comments in your own language
- Use house-style alignment and headings

Do Exercise 6 - Online Help

Using LLMs as a help resource

LLMs (ChatGPT, Copilot, etc.) are like a very fast, sometimes-wrong search engine that can write code.

They are good at:
- Explaining error messages
- Suggesting possible approaches
- Pointing out obvious bugs in short code snippets
They are not always:
- Correct about Stata syntax
- Efficient in their approach to coding
- Aligned with our house style
- Using the same workflow (projects, project.do, etc.)

How to ask good questions to an LLM:

Include:

A short description of your goal
A minimal code snippet (10–20 lines)
The exact error message or unexpected output
Any constraints (e.g., “Stata 17, cannot use user-written packages”)

Example prompt:

“I’m using Stata 17. I’m trying to merge household-level data into plot-level data. Here’s my code and the error not uniquely identified. Can you help me understand what’s wrong?”

Then:

Check its explanation against your understanding of merge (1:m vs m:1, keys, etc.)
Test any suggested fix on a small subset
Rewrite any accepted code into your own .do file with house style

Remember the forklift at the gym analogy: don’t have the LLM do all your “lifting” for you.

Do Exercise 7 - LLM Help

When and how to ask humans

Sometimes the fastest path is to ask a person:

Classmates / study groups
- Great for “am I doing this in a reasonable way?” type questions
Slack
- Include a short description and a small, well-formatted code block
- Mention the exact error message
Office hours
- Bring your .do file
- Be ready to walk through what you’ve already tried

A good question includes:

What you’re trying to do
What you expected to happen
What actually happened (including errors)
What you’ve already tried based on help files / Google / LLMs
The file name and exact lines of code that are a problem
Before you ask for help, push your commits so the person can actually help you!