Stata 15 help for teffects intro

[TE] teffects intro -- Introduction to treatment effects estimation for observational data

Description

This entry provides a nontechnical introduction to treatment-effects estimators and the teffects command in Stata. Advanced users may want to instead read [TE] teffects intro advanced or skip to the individual commands' entries.

The teffects command estimates average treatment effects (ATEs), average treatment effects among treated subjects (ATETs), and potential-outcome means (POMs) using observational data.

Treatment effects can be estimated using regression adjustment (RA), inverse-probability weights (IPW), and "doubly robust" methods, including inverse-probability-weighted regression adjustment (IPWRA) and augmented inverse-probability weights (AIPW), and via matching on the propensity score or nearest neighbors.

The outcome can be continuous, binary, count, fractional, or nonnegative. Treatments can be binary or multivalued.

Remarks: A quick tour of the estimators

The teffects command implements six estimators of treatment effects. We introduce each one by showing the basic syntax one would use to apply them to our birthweight example. See each command's entry for more information.

Regression adjustment

teffects ra implements the RA estimator. We estimate the effect of a mother's smoking behavior (mbsmoke) on the birthweight of her child (bweight), controlling for marital status (mmarried), the mother's age (mage), whether the mother had a prenatal doctor's visit in the baby's first trimester (prenatal1), and whether this baby is the mother's first child (fbaby). We use linear regression (the default) to model bweight:

. webuse cattaneo2 . teffects ra (bweight mmarried mage prenatal1 fbaby) (mbsmoke)

Inverse-probability weighting

teffects ipw implements the IPW estimator. Here we estimate the effect of smoking by using a probit model to predict the mother's smoking behavior as a function of marital status, the mother's age, and indicators for first-trimester doctor's visits and firstborn status:

. teffects ipw (bweight) (mbsmoke mmarried mage prenatal1 fbaby, probit)

Inverse-probability-weighted regression adjustment

teffects ipwra implements the IPWRA estimator. We model the outcome, birthweight, as a linear function of marital status, the mother's age, and indicators for first-trimester doctor's visits and firstborn status. We use a logistic model (the default) to predict the mother's smoking behavior, using the same covariates as explanatory variables:

. teffects ipwra (bweight mmarried mage prenatal1 fbaby) (mbsmoke mmarried mage prenatal1 fbaby)

Augmented inverse-probability weighting

teffects aipw implements the AIPW estimator. Here we use the same outcome- and treatment-model specifications as we did with the IPWRA estimator:

. teffects aipw (bweight mmarried mage prenatal1 fbaby) (mbsmoke mmarried mage prenatal1 fbaby)

Nearest-neighbor matching

teffects nnmatch implements the NNM estimator. In this example, we match treated and untreated subjects based on marital status, the mother's age, the father's age, and indicators for first-trimester doctor's visits and firstborn status. We use the Mahalanobis distance based on the mother's and father's ages to find matches. We use exact matching on the other three variables to enforce the requirement that treated subjects are matched with untreated subjects who have the same marital status and indicators for first-trimester doctor's visits and firstborn statuses. Because we are matching on two continuous covariates, we request that teffects nnmatch include a bias-correction term based on those two covariates:

. teffects nnmatch (bweight mage fage) (mbsmoke), ematch(prenatal1 mmarried fbaby) biasadj(mage fage)

Propensity-score matching

teffects psmatch implements the PSM estimator. Here we model the propensity score using a probit model, incorporating marital status, the mother's age, and indicators for first-trimester doctor's visits and firstborn status as covariates:

. teffects psmatch (bweight) (mbsmoke mmarried mage prenatal1 fbaby, probit)

Video examples

Introduction to treatment effects in Stata, part 1

Introduction to treatment effects in Stata, part 2


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index