Virata manual

regress

Explains a number using other columns, by fitting a straight line.

What we have

  • factor notation: i., c., # and ##
  • aw and pw weights
  • robust, checked against R's HC1
  • e() results after the model

Coming later

  • vce(cluster ...), which is refused rather than quietly ignored
  • frequency weights (fw), which are refused
  • predict, test, lincom and margins afterwards

You may shorten it to reg.

How you write it

regress depvar indepvars [if] [in] [weight]

Square brackets mean optional. A vertical bar means choose one.

What it does

regress mpce age education asks: as age and education change, how does spending change? The answer comes back as one number per column (a coefficient), each meaning "the change in spending associated with a one-unit rise in this, holding the others fixed".

Beside each coefficient is its margin of error (standard error and confidence interval) and a p-value saying how likely a result this far from zero would be if the column really had no effect. The R-squared at the top is the share of the variation in spending the model accounts for.

Categories go in with i. in front: i.sector fits a separate effect for each sector, measured against whichever one comes first (the base level). c.age forces a column to be treated as a number, and i.a#i.b lets two categories interact.

Watch out. "Holding the others fixed" is arithmetic, not an experiment. A coefficient is an association; calling it an effect requires an argument the regression itself cannot supply.

Try it

Every example below was run after sysuse vira.sample02, clear.

. regress mpce age education i.sector
Linear regression                            Number of obs      =        6,000
                                             F statistic        =       423.40
                                             Prob > F           =       0.0000
                                             R-squared          =       0.1748
                                             Root MSE           =     1348.945

------------------------------------------------------------------------------
        mpce | Coefficient  Std. err.      t   P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
sector=Urban |    944.3234   36.88319   25.60   0.000     872.0191    1016.628
         age |    6.209437   .9287289    6.69   0.000     4.388795     8.03008
   education |    90.83084   3.779603   24.03   0.000     83.42146    98.24022
       _cons |    2425.591   57.07151   42.50   0.000     2313.711    2537.472
------------------------------------------------------------------------------
The same thing in pandas and in R

pandas

import statsmodels.formula.api as smf
model = smf.ols("mpce ~ age + education + C(sector)", data=df).fit()

R

model <- lm(mpce ~ age + education + factor(sector), data = df)
summary(model)

# This R was run in a real R (4.6.1) and gave the same numbers.

Nearby

logit probit poisson nbreg ologit oprobit