Imported from haoyu-haoyu/stata-ai-fusion (skill/SKILL.md). Install upstream with npx skills add haoyu-haoyu/stata-ai-fusion --skill skill. Copyright stays with the author.

Stata AI Fusion Skill

You have access to comprehensive Stata reference files. Do not load all files. Read only the 1-3 files relevant to the user's current task using the routing table below.

1. Execution Mode Detection

Detect which execution mode is available and adapt accordingly.

MCP Mode (Stata MCP Server Available)

When the MCP server is connected, you can execute Stata code directly:

• Use stata_run_command to run code and return results
• Use stata_run_do_file to run full .do files
• Use stata_inspect_data to inspect loaded datasets (summary stats, variable info)
• Use stata_codebook to generate a codebook for the current dataset
• Use stata_get_results to retrieve stored estimation results (r(), e(), s())
• Use stata_export_graph to export the current Stata graph to an image
• Use stata_search_log to search through Stata log output
• Use stata_install_package to install community-contributed packages
• Use stata_cancel_command to cancel a running Stata command
• Use stata_list_sessions to list all active Stata sessions
• Use stata_close_session to close a specific Stata session
• Always check execution results before proceeding
• If execution fails, read references/error-codes.md for diagnosis

Offline Mode (No MCP Server)

When MCP is not available, generate code for the user to run manually:

• Write complete, self-contained .do files
• Include version, clear all, set more off preamble
• Add log using for output capture
• Comment every non-obvious line
• Wrap risky operations with capture and check _rc

Detection Logic

IF MCP tools are listed in available tools:
    -> Use MCP execution mode
    -> Run code incrementally, check results between steps
ELSE:
    -> Use offline code generation mode
    -> Produce complete .do file with full preamble

2. Core Coding Standards (Mandatory Rules)

Every piece of Stata code you generate MUST follow these rules. Violations cause silent bugs.

Missing Values Sort to +Infinity

Stata . (and .a-.z) are greater than all numbers.

* WRONG -- includes observations where income is missing!
gen high_income = (income > 50000)

* RIGHT
gen high_income = (income > 50000) if !missing(income)

* WRONG -- missing ages appear in this list
list if age > 60

* RIGHT
list if age > 60 & !missing(age)

= vs ==

= is assignment; == is comparison. Mixing them is a syntax error or silent bug.

* WRONG -- syntax error
gen employed = 1 if status = 1

* RIGHT
gen employed = 1 if status == 1

Local Macro Syntax

Locals use `name' (backtick + single-quote). Globals use $name or ${name}. Forgetting the closing quote is the number-one macro bug.

local controls "age education income"
regress wage `controls'        // correct
regress wage `controls         // WRONG -- missing closing quote
regress wage 'controls'        // WRONG -- wrong quote characters

by Requires Prior Sort (Use bysort)

* WRONG -- error if data not sorted by id
by id: gen first = (_n == 1)

* RIGHT -- bysort sorts automatically
bysort id: gen first = (_n == 1)

Factor Variable Notation (i. and c.)

* WRONG -- treats race as continuous
regress wage race education

* RIGHT -- creates dummies automatically
regress wage i.race education

* Interactions
regress wage i.race##c.education    // full interaction
regress wage i.race#c.education     // interaction only (no main effects)

generate vs replace

generate creates new variables; replace modifies existing ones.

gen x = 1
gen x = 2          // ERROR: x already defined
replace x = 2      // correct

String Comparison Is Case-Sensitive

* May miss "Male", "MALE", etc.
keep if gender == "male"

* Safer
keep if lower(gender) == "male"

merge Always Check _merge

merge 1:1 id using other.dta
tab _merge                      // always inspect
assert _merge == 3              // or handle mismatches
drop _merge

Line Continuation Uses ///

regress y x1 x2 x3 ///
    x4 x5 x6, ///
    vce(robust)

Stored Results: r() vs e() vs s()

• r() -- r-class commands (summarize, tabulate, etc.)
• e() -- e-class commands (estimation: regress, logit, etc.)
• s() -- s-class commands (parsing)

A new estimation command overwrites previous e() results. Store them first:

regress y x1 x2
estimates store model1

Weights Are Not Interchangeable

• fweight -- frequency weights (replication)
• aweight -- analytic/regression weights (inverse variance)
• pweight -- probability/sampling weights (survey data, implies robust SE)
• iweight -- importance weights (rarely used)

preserve / restore for Temporary Changes

preserve
collapse (mean) income, by(state)
* ... do something with collapsed data ...
restore   // original data is back

capture Swallows Errors

capture some_command
if _rc != 0 {
    di as error "Failed with code: " _rc
    exit _rc
}

3. Workflow Templates

Data Cleaning Pipeline

* 1. Load and inspect
import delimited "raw_data.csv", clear varnames(1)
describe
codebook, compact

* 2. Clean
rename *, lower
destring income, replace force
replace income = . if income < 0

* 3. Label
label variable income "Annual household income (USD)"
label define yesno 0 "No" 1 "Yes"
label values employed yesno

* 4. Save
compress
save "clean_data.dta", replace

Regression Table Workflow

eststo clear
eststo: regress y x1 x2, vce(robust)
eststo: regress y x1 x2 x3, vce(robust)
eststo: regress y x1 x2 x3 x4, vce(cluster id)

esttab using "results.tex", replace ///
    se star(* 0.10 ** 0.05 *** 0.01) ///
    label booktabs ///
    title("Main Results") ///
    mtitles("(1)" "(2)" "(3)")

Panel Data Setup

xtset panelid timevar
xtdescribe
xtsum outcome

xtreg y x1 x2, fe vce(cluster panelid)
* Or with reghdfe
reghdfe y x1 x2, absorb(panelid timevar) vce(cluster panelid)

Difference-in-Differences

* Classic 2x2 DiD
gen post = (year >= treatment_year)
gen treat_post = treated * post
regress y treated post treat_post, vce(cluster id)

* Modern staggered DiD (Callaway & Sant'Anna)
csdid y x1 x2, ivar(id) time(year) gvar(first_treat) agg(event)
csdid_plot

Publication-Quality Graph

twoway (scatter y x, mcolor(navy%50) msize(small)) ///
       (lfit y x, lcolor(cranberry) lwidth(medthick)), ///
    title("Title Here") ///
    xtitle("X Label") ytitle("Y Label") ///
    legend(off) scheme(plotplain)
graph export "figure1.pdf", replace as(pdf)
graph export "figure1.png", replace as(png) width(2400)

Multiple Imputation

mi set mlong
mi register imputed income education
mi impute chained (regress) income (ologit) education ///
    = age i.gender, add(20) rseed(12345)
mi estimate: regress wage income education age i.gender

Survival Analysis Quick Start

stset time_var, failure(event_var)
stsum
sts graph, by(group)
sts test group
stcox i.group age, vce(robust)
estat phtest, detail

4. Reference Document Navigation (Routing Table)

Read only the files relevant to the user's task. Paths are relative to this SKILL.md file.

Data Operations

Econometrics & Statistics

Clinical & Domain-Specific

Output & Visualization

Programming & Advanced

Community Packages

5. MCP Tools Usage Guide

Tool Combination Patterns

Pattern A: Iterative Analysis

1. stata_run_command("use mydata.dta, clear")  -> load data
2. stata_run_command("describe")               -> understand structure
3. stata_run_command("regress y x1 x2")        -> run model
4. Read results, decide next step
5. stata_run_command("esttab using ...")        -> export table

Pattern B: Debug Workflow

1. User pastes error from .log file
2. Read references/error-codes.md for the error code
3. Identify the root cause
4. stata_run_command(corrected_code)      -> verify fix
5. Explain what went wrong

Pattern C: Full Pipeline

1. stata_run_command("import delimited ...")  -> load raw data
2. stata_run_command("describe \n codebook")  -> inspect
3. Generate cleaning code based on inspection
4. stata_run_command(cleaning_code)           -> clean
5. stata_run_command(analysis_code)           -> analyze
6. stata_run_command(export_code)             -> export results

Error Recovery in MCP Mode

When an MCP execution returns an error:

1. Parse the error code (e.g., r(111), r(198), r(601))
2. Look up in references/error-codes.md
3. Apply the documented fix
4. Re-execute the corrected code
5. If still failing, try alternative approaches

Code Generation for Offline Mode

When generating .do files for users to run manually:

/* ============================================================
   Project:  [Project Name]
   Purpose:  [What this do-file does]
   Author:   AI-generated via stata-ai-fusion
   Date:     [Current Date]
   Input:    [Input files]
   Output:   [Output files]
   ============================================================ */

version 17
clear all
set more off
set maxvar 10000

* -- Set paths --
global root "[USER_PATH]"
global data "$root/data"
global output "$root/output"

log using "$output/logs/analysis_log.txt", replace text

* -- Body --
// ... analysis code ...

log close

6. Common Code Snippets

Quick Summary Statistics

summarize price mpg weight, detail
tabstat price mpg, by(foreign) stat(n mean sd p25 p50 p75)

Winsorize at 1%/99%

foreach var of varlist income wealth {
    quietly summarize `var', detail
    replace `var' = r(p1) if `var' < r(p1) & !missing(`var')
    replace `var' = r(p99) if `var' > r(p99) & !missing(`var')
}

Create Balanced Panel

xtset id year
bysort id: gen n_years = _N
tab n_years
keep if n_years == [expected_T]

Standardize Variables (Z-score)

foreach var of varlist x1 x2 x3 {
    quietly summarize `var'
    gen `var'_z = (`var' - r(mean)) / r(sd)
}

Lag and Lead Variables in Panel

xtset id year
gen y_lag1 = L.y
gen y_lag2 = L2.y
gen y_lead1 = F.y
gen y_diff = D.y

Quick Coefficient Plot

regress y x1 x2 x3 x4, vce(robust)
coefplot, drop(_cons) xline(0, lcolor(red) lpattern(dash)) ///
    title("Coefficient Estimates") scheme(plotplain)
graph export "coefplot.png", replace width(2000)

Export Summary Table to Excel

putexcel set "summary.xlsx", replace sheet("Summary")
putexcel A1 = "Variable" B1 = "N" C1 = "Mean" D1 = "SD"
local row = 2
foreach var of varlist income age education {
    quietly summarize `var'
    putexcel A`row' = "`var'" B`row' = r(N) ///
        C`row' = r(mean) D`row' = r(sd)
    local ++row
}

Propensity Score Matching (Quick)

logit treatment x1 x2 x3
predict pscore, pr
psmatch2 treatment, pscore(pscore) outcome(y) n(1) common

Event Study Plot (Quick)

gen rel_time = year - treatment_year
forvalues k = -5/5 {
    if `k' < 0 local j = abs(`k')
    else       local j = `k'
    gen D`j'_`=cond(`k'<0,"pre","post")' = (rel_time == `k')
}
* Omit period -1 as reference
reghdfe y D*_pre D*_post, absorb(id year) vce(cluster id)
coefplot, keep(D*) vertical yline(0) xline(5.5, lpattern(dash))

Check for Multicollinearity

regress y x1 x2 x3 x4
estat vif
* VIF > 10 indicates concern

Robust Hausman Test (with clustered SE)

* Standard Hausman fails with robust/cluster SE. Use this instead:
xtreg y x1 x2, fe vce(cluster id)
estimates store fe
xtreg y x1 x2, re vce(cluster id)
estimates store re
hausman fe re, sigmamore