OLS regression and post-estimation — Stata Module 7

regress is Stata's OLS command. The output table is one of the cleanest in any statistical software, and the ecosystem of post-estimation commands (predict, test, lincom, margins) turns a single regression into a structured analytical product.

regress

stata

regress lending_rate deposit_rate
regress lending_rate deposit_rate i.year      // with year fixed effects
regress lending_rate c.deposit_rate##i.year   // interactions

Robust standard errors

stata

regress lending_rate deposit_rate, robust                          // HC3
regress lending_rate deposit_rate, vce(cluster bank_id)            // clustered
regress lending_rate deposit_rate, vce(robust) cluster(bank_id)    // both

Reading the output

Source / SS / df / MS — the ANOVA decomposition
Number of obs — sample size
F(p, n-p-1) — joint test of all coefficients = 0
R-squared — fraction of variance explained
Root MSE — sqrt of mean squared error
Coef. / Std. Err. / t / P>|t| / [95% CI] — per coefficient

Post-estimation: predict, test, lincom

stata

regress lending_rate deposit_rate i.year

predict yhat                          // fitted values
predict residual, residuals           // residuals
predict se, stdp                      // standard error of fitted value

test deposit_rate                     // test single coefficient
test 2024.year = 2023.year            // test equality

lincom deposit_rate + 2024.year       // linear combination with SE

store and esttab — publication tables

stata

regress lending_rate deposit_rate
estimates store m1
regress lending_rate deposit_rate i.year
estimates store m2

estout m1 m2, cells(b(star fmt(3)) se(par fmt(3))) ///
    stats(N r2 r2_a, fmt(0 3 3))

regress is one line. The post-estimation is the analysis.

Anyone can run a regression. The discipline of an applied econometrician shows in what they do next: testing constraints, computing margins, comparing specifications, presenting tables. That's where Stata earns its keep.

Exercise

Run a regression of lending_rate on deposit_rate with robust standard errors.