Stata Commands Cheat Sheet for Researchers

Stata is a statistical software package that you control mainly by typing commands, either in the Command window or in a do-file. Nearly all Stata code follows the same structure:

[by varlist:] command [varlist] [if expression] [in range] [weight] [, options]

Learn this pattern once and most commands become predictable. Options always come after a single comma. Commands can be abbreviated, so summarize can be typed as sum and tabulate as tab. Type help followed by any command name to open its manual entry.

Loading and saving data

sysuse auto, clear                          // built-in example dataset
use "mydata.dta", clear                     // open a Stata dataset
import delimited "survey.csv", clear        // import a CSV file
import excel "survey.xlsx", firstrow clear  // import Excel, first row as names
save "mydata_clean.dta", replace            // save (overwrite if it exists)

The clear option discards the data currently in memory, and replace allows Stata to overwrite an existing file. Without them, Stata stops with an error to protect your work.

and, or and other operators in Stata

Conditions in if use == for equals, != for not equal, & for and, | for or, and ! for not.

list make price mpg if foreign == 1 & price > 10000
count if rep78 == 1 | rep78 == 2
summarize price if !missing(rep78)

Watch out: Stata treats missing values (.) as larger than any number, so count if rep78 > 3 also counts the five cars with missing repair records. Write count if rep78 > 3 & !missing(rep78) instead.

Start every analysis by checking what is in the dataset.

sysuse auto, clear
describe                 // variable names, types, labels, number of obs
codebook rep78           // values, missing count, labels for one variable
list make price mpg in 1/5
summarize                // obs, mean, sd, min, max for all variables
summarize price mpg weight

How to find the median in Stata

Plain summarize does not show the median. Add the detail option to get percentiles (the 50th percentile is the median), the four largest and smallest values, variance, skewness and kurtosis:

summarize price, detail
display r(p50)           // prints the median stored by summarize

After any summarize, results are stored in r(); type return list to see them all, including r(mean) and r(sd).

Summary statistics by group

bysort foreign: summarize price mpg
tabstat price mpg weight, statistics(n mean sd median min max)
tabstat price mpg, by(foreign) statistics(mean sd median) columns(statistics)

tabstat is the quickest way to build a compact table of means, medians and standard deviations, and the by() option splits it by group. To get a single Stata mean with a confidence interval, use mean price.

tabulate (or tab) produces frequency tables for categorical variables.

tabulate foreign                     // one-way frequency table
tabulate rep78, missing              // include missing as a category
tabulate rep78 foreign               // two-way cross-tabulation
tabulate rep78 foreign, row chi2     // row percentages and chi-square test
tabulate rep78 foreign, column exact // column percentages, Fisher's exact test
tabulate foreign, summarize(price)   // mean and sd of price by category

To produce every pairwise cross-tab among several variables at once, use tab2. The tab1 command gives separate one-way tables.

tab2 rep78 foreign headroom
tab1 rep78 foreign

The table command

In Stata 17 and later, table builds customizable tables of statistics:

table foreign, statistic(mean price) statistic(sd price)
table rep78 foreign, statistic(frequency)
table foreign, statistic(mean price mpg) nformat(%9.1f)

Earlier versions use a different syntax (for example table foreign, contents(mean price sd price)), so check help table for your version.

generate and replace

generate price_k = price / 1000
generate heavy = (weight > 3000) if !missing(weight)
replace price_k = round(price_k, 0.1)
generate mpg_cat = .
replace mpg_cat = 1 if mpg < 20
replace mpg_cat = 2 if mpg >= 20 & mpg < 30
replace mpg_cat = 3 if mpg >= 30 & !missing(mpg)

generate creates a new variable; replace changes an existing one. A logical expression such as (weight > 3000) returns 1 if true and 0 if false.

egen for group statistics

egen mean_price = mean(price), by(foreign)   // group mean on every row
egen price_q = cut(price), group(4)          // quartile groups 0 to 3
egen n_cars = count(price), by(rep78)
egen id = group(foreign rep78)                // one id per combination

Logarithms in Stata

Both ln() and log() return the natural logarithm; use log10() for base 10. The log of zero or a negative number is missing.

generate ln_price = ln(price)
generate log10_price = log10(price)

Labels and recoding

generate expensive = (price > 10000)
label variable expensive "Price above 10,000 USD"
label define yesno 0 "No" 1 "Yes"
label values expensive yesno
recode rep78 (1 2 = 1 "Poor") (3 = 2 "Average") (4 5 = 3 "Good"), generate(rep3)

Concatenating strings

Join strings with +. Numbers must be converted with string() first, or Stata reports a type mismatch.

generate label_mpg = make + " (" + string(mpg) + " mpg)"
decode foreign, generate(origin)                 // value labels to text
generate make_origin = make + " - " + origin
egen combo = concat(foreign rep78), punct("_")    // egen handles numbers too

To convert between types, use destring (text that holds numbers to numeric), tostring (numeric to text) and encode (text categories to labeled numeric codes).

merge adds variables (columns) from another file by matching on key variables. append adds observations (rows). The examples below split auto into pieces and join them back so you can run them.

merge 1:1

Use 1:1 when the key uniquely identifies observations in both files.

sysuse auto, clear
keep make price mpg
save auto_a, replace

sysuse auto, clear
keep make weight length
save auto_b, replace

use auto_a, clear
merge 1:1 make using auto_b
tabulate _merge

merge m:1

Use m:1 when many observations in memory match one observation in the using file, such as adding group-level data to individuals.

sysuse auto, clear
collapse (mean) avg_price = price, by(foreign)
save origin_means, replace

sysuse auto, clear
merge m:1 foreign using origin_means
generate price_gap = price - avg_price

Understanding _merge in Stata

Always tabulate _merge and investigate unmatched cases before running drop _merge. If you expect every case to match, let Stata check it for you, or keep only matched rows without creating the variable:

use auto_a, clear
merge 1:1 make using auto_b, assert(match) nogenerate

Replace assert(match) with keep(match) to keep only matched rows instead of stopping when some fail to match. If a merge fails with "variable make does not uniquely identify observations in the master data", check for duplicates with duplicates report make.

append

sysuse auto, clear
keep if foreign == 0
save domestic, replace

sysuse auto, clear
keep if foreign == 1
append using domestic
count                      // back to 74 observations

reshape converts data between wide format (one row per unit, repeated measures in separate columns) and long format (one row per unit per time point). Panel and repeated-measures commands usually need long format.

clear
input id inc2020 inc2021 inc2022
1 50 52 55
2 40 41 45
3 61 60 64
end

reshape long inc, i(id) j(year)    // wide to long
list, sepby(id)

reshape wide inc, i(id) j(year)    // long back to wide
list

In reshape long inc, i(id) j(year), inc is the stub shared by the wide variables, i() names the variable that identifies each unit, and j() names the new variable that will hold the suffix (2020, 2021, 2022). If Stata reports that the i() variable does not uniquely identify observations, you have duplicate ids; find them with duplicates list id. If the suffix contains letters, add the string option.

The regress command

The syntax is regress, the dependent variable, then the independent variables.

sysuse auto, clear
regress price mpg weight
regress price mpg weight i.foreign                 // factor variable dummy
regress price mpg weight i.foreign, vce(robust)    // robust standard errors
regress price mpg weight ib3.rep78                 // category 3 as base
regress price c.mpg##i.foreign                     // interaction

The prefix i. tells Stata a variable is categorical and creates the dummy variables for you; c. marks a continuous variable in interactions, and ## includes both main effects and the interaction. vce(robust) (or simply , robust) gives heteroskedasticity-robust standard errors; use vce(cluster id) for clustered data.

F statistic and F tests in Stata

The regression header reports the overall F statistic, its p value (Prob > F), R-squared and the number of observations. To test specific restrictions after estimation, use test:

regress price mpg weight i.foreign
test weight                 // H0: coefficient on weight = 0
test mpg weight             // joint F test: both coefficients = 0
test 1.foreign              // the foreign dummy
test mpg = weight           // equality of two coefficients
regress price mpg weight i.rep78
testparm i.rep78            // joint test of all rep78 dummies

predict after regression

regress price mpg weight
predict yhat                // fitted values (linear prediction)
predict resid, residuals    // residuals

Logit model in Stata

For a binary outcome coded 0/1, use logit. Add or to report odds ratios, or use logistic, which reports odds ratios by default.

logit foreign price mpg
logit foreign price mpg, or
logistic foreign price mpg
predict phat, pr            // predicted probability of foreign = 1
margins, dydx(*)            // average marginal effects

Ordered logistic regression and its graph

For ordered outcomes such as ratings, use ologit, then graph predicted probabilities with margins and marginsplot:

ologit rep78 price mpg
margins, at(mpg = (15(5)35)) predict(outcome(5))
marginsplot

Panel data with xtreg

Panel models need an id and a time variable declared with xtset. The auto data are cross-sectional, so this example uses another official dataset:

webuse nlswork, clear
xtset idcode year
xtreg ln_wage age tenure, fe        // fixed effects
xtreg ln_wage age tenure, re        // random effects

Multiple imputation in Stata

The mi commands impute missing values and combine results across imputed datasets. Here the five missing values of rep78 are imputed:

sysuse auto, clear
mi set mlong
mi register imputed rep78
mi impute ologit rep78 price mpg weight, add(20) rseed(1234)
mi estimate: regress price mpg weight i.rep78

Scatter plot in Stata

sysuse auto, clear
twoway scatter price mpg
twoway (scatter price mpg) (lfit price mpg)       // add fitted line
twoway (scatter price mpg if foreign == 0) (scatter price mpg if foreign == 1), legend(order(1 "Domestic" 2 "Foreign"))
scatter price mpg, mlabel(make)                   // label points

Line plot

Line plots connect points in the order of the x variable, so they suit time series. This example uses the built-in U.S. life expectancy data:

sysuse uslifeexp, clear
twoway line le year
twoway line le_male le_female year, legend(order(1 "Male" 2 "Female"))

Bar graph

sysuse auto, clear
graph bar (mean) price, over(foreign)
graph bar (mean) mpg, over(rep78) over(foreign)
graph hbar (count) mpg, over(rep78)

Histogram with a kdensity curve

histogram mpg
histogram mpg, frequency
histogram mpg, kdensity                    // overlay kernel density
histogram mpg, kdensity normal             // add a normal curve too
histogram mpg, kdensity kdenopts(bwidth(2))
histogram price, by(foreign)

The kdensity option overlays a smoothed density estimate; kdenopts() passes options such as bandwidth to it. To save any graph for a paper, run graph export "mpg_hist.png", replace immediately after drawing it.

A do-file is a plain text file of Stata commands. Open the editor with doedit, write your commands, and run them all with do. Rerunning one file from raw data to final tables is the most reliable way to keep an analysis correct and to document your methods.

* analysis.do  -  price regressions on the auto data
version 18
clear all
capture log close
log using "analysis.log", replace text

sysuse auto, clear
summarize price mpg weight, detail
tabulate rep78 foreign, row

regress price mpg weight i.foreign, vce(robust)
test mpg weight

log close

Notes on this template:

• Lines starting with * are comments; // adds a comment at the end of a line, and /* ... */ comments out a block. Use /// at the end of a line to continue a long command on the next line.
• version tells Stata which version's behavior to use; set it to the version you actually run.
• capture log close closes any log left open from a previous run without stopping if none is open.
• log using ..., replace text saves all output as a readable text file. Close it at the end with log close.

Run a do-file from the Command window with do analysis.do, or highlight lines in the editor and run only those.

A pattern that avoids most "already defined" errors when rerunning a do-file:

capture drop ln_price
generate ln_price = ln(price)

If your output still looks wrong rather than producing an error, the most common culprit is missing values being included in a > condition. Our Stata help service can review do-files and explain output; you stay the author of your analysis.