Stata 18 Jun 2026

The classic two-group, two-period DiD is insufficient for modern staggered treatment designs. Stata 18’s new did command implements the estimator, which is robust to treatment effect heterogeneity across time and groups. It automatically handles "not-yet-treated" vs. "never-treated" control groups.

Perhaps the most anticipated addition in Stata 18 is . In many research scenarios, you face "model uncertainty"—not knowing which predictors truly belong in your model. Instead of picking one "best" model, BMA accounts for this uncertainty by averaging over many potential models. This results in more stable predictions and a more nuanced understanding of variable importance. Causal Inference: Heterogeneous DID

Automatically split summary data across a specific treatment or demographic indicator.

Explore the model space efficiently for linear regression models. Stata 18

Researchers can now use Bayesian methods to select relevant predictors in complex linear models, providing more robust inferences [5.1].

For , do not purchase Stata 17. Stata 18 is the more future-proof, feature-rich, and well-optimized version. StataCorp has clearly listened to its user community—especially regarding Python and modern DiD estimators.

Stata/MP remains the fastest option, especially for mi impute , bootstrap , and xtmixed . All licenses include free updates for the Stata 18.x cycle. The classic two-group, two-period DiD is insufficient for

estimates the effects of endogenous covariates on the quantiles of an outcome‘s conditional distribution. This method allows researchers to examine how treatments affect not just the mean but the entire distribution of outcomes, while accounting for endogeneity. Coefficients can be plotted across quantiles for intuitive visualization.

Provides more reliable inference for models with a small number of clusters. Visual and Workflow Improvements Issue with xthdidregress command on STATA 18 - Statalist

| Feature Category | Stata 17 Highlights | Stata 18 Highlights | | :--- | :--- | :--- | | | Basic Treatment Effects, Lasso | Panel Data Treatment Effects ( eteffects ) , Heterogeneous DID ( hetdid ) | | Survival Analysis | Multilevel Survival Models | Interval-Censored Cox ( stintcox ) | | Output | collect command introduced | Robust tables command for publication tables | | Meta-Analysis | Network Meta-Analysis | Dose-Response Meta-Analysis , Trim-and-Fill | | Visualization | Customizable tables (collect) | Streamplots , Graph Editor macro recording | | Hardware | Apple Silicon via Rosetta | Native Apple Silicon Support | "never-treated" control groups

A significant addition for handling model uncertainty by considering a range of potential models rather than a single "best" one.

New tools allow researchers to disentangle the mechanisms through which an exposure affects an outcome by identifying mediating variables.