Causal inference
Estimate experimental-style causal effects from observational data. With Stata's treatment-effects estimators, you can use a potential-outcomes (counterfactuals) framework to estimate, for instance, the effect of family structure on child development or the effect of unemployment on anxiety. Fit models for continuous, binary, count, fractional, and survival outcomes with binary or multivalued treatments using inverse-probability weighting (IPW), propensity-score matching, nearest-neighbor matching, regression adjustment, or doubly robust estimators. If the assignment to a treatment is not independent of the outcome, you can use an endogenous treatment-effects estimator. In the presence of group and time effects, you can use difference-in-differences (DID) and triple-differences (DDD) estimators. In the presence of high-dimensional covariates, you can use lasso. If causal effects are mediated through another variable, use causal mediation with mediate to disentangle direct and indirect effects.

Panel data
Take full advantage of the extra information that panel data provide while simultaneously handling the peculiarities of panel data. Study the time-invariant features within each panel, the relationships across panels, and how outcomes of interest change over time. Fit linear models or nonlinear models for binary, count, ordinal, censored, or survival outcomes with fixed-effects, random-effects, or population-averaged estimators. Fit dynamic models or models with endogeneity. Fit Bayesian panel-data models.

Multilevel mixed-effects models
Whether the groupings in your data arise in a nested fashion (students nested in schools and schools nested in districts) or in a nonnested fashion (regions crossed with occupations), you can fit a multilevel model to account for the lack of independence within these groups. Fit models for continuous, binary, count, ordinal, and survival outcomes. Estimate variances of random intercepts and random coefficients. Compute intraclass correlations. Predict random effects. Estimate relationships that are population averaged over the random effects.

Structural equation modeling (SEM)
Estimate mediation effects, analyze the relationship between an unobserved latent concept such as consumer confidence and the observed variables that measure consumer confidence, model a system with many endogenous variables and correlated errors, or fit a model with complex relationships among both latent and observed variables. Fit models with continuous, binary, count, ordinal, fractional, and survival outcomes. Even fit multilevel models with groups of correlated observations such as children within the same schools. Evaluate model fit. Compute indirect and total effects. Fit models by drawing a path diagram or using the straightforward command syntax.

Linear, binary, and count regressions
Fit classical linear models of the relationship between a continuous outcome, such as wage, and the determinants of wage, such as education level, age, experience, and economic sector. If your response is binary (for example, employed or unemployed), ordinal (education level), count (number of children), or censored (ticket sales in an existing venue), don't worry. Stata has maximum likelihood estimators—probit, ordered probit, Poisson, tobit, and many others—that estimate the relationship between such outcomes and their determinants. A vast array of tools is available to analyze such models. Predict outcomes and their confidence intervals. Test equality of parameters or any linear or nonlinear combination of parameters.

Survival analysis
Analyze duration outcomes—outcomes measuring the time to an event such as failure or death—using Stata's specialized tools for survival analysis. Account for the complications inherent in survival data, such as sometimes not observing the event (right-, left-, and interval-censoring), individuals entering the study at differing times (delayed entry), and individuals who are not continuously observed throughout the study (gaps). You can estimate and plot the probability of survival over time. Or model survival as a function of covariates using Cox, Weibull, lognormal, and other regression models. Predict hazard ratios, mean survival time, and survival probabilities. Do you have groups of individuals in your study? Adjust for within-group correlation with a random-effects or shared-frailty model. If you have many potential covariates, use lasso cox and elasticnet cox for model selection and prediction.

Meta-analysis
Combine results of multiple studies to estimate an overall effect. Use forest plots to visualize results. Use subgroup analysis and meta-regression to explore study heterogeneity. Use funnel plots and formal tests to explore publication bias and small-study effects. Use trim-and-fill analysis to assess the impact of publication bias on results. Perform cumulative and leave-one-out meta-analysis. Perform univariate, multilevel, and multivariate meta-analysis. Use the meta suite, or let the Control Panel interface guide you through your entire meta-analysis.

Marginal effects and marginal means
Marginal effects and marginal means let you analyze and visualize the relationships between your outcome variable and your covariates, even when that outcome is binary, count, ordinal, categorical, or censored (tobit). Estimate population-averaged marginal effects, or evaluate marginal effects at interesting or representative values of the covariates. Analyze the effect of interactions. You can even trace out the marginal effect over a range of interesting covariate values or covariate interactions. You can do all of this with marginal means, sometimes called potential-outcome means, too—even when your "mean" is a probability of a positive outcome or a count from a Poisson model. If you have panel data and random effects, these effects are automatically integrated out to provide marginal (that is, population-averaged) effects.

Lasso
Use lasso and elastic net for model selection and prediction. And when you want to estimate effects and test coefficients for a few variables of interest, inferential methods provide estimates for these variables while using lassos to select from among a potentially large number of control variables. You can even account for endogenous covariates. Whether your goal is model selection, prediction, or inference, you can use Stata's lasso features with your continuous, binary, count, or time-to-event outcomes.

Bayesian analysis
Fit Bayesian regression models using one of the Markov chain Monte Carlo (MCMC) methods. You can choose from various supported models or even program your own. Extensive tools are available to check convergence, including multiple chains. Compute posterior mean estimates and credible intervals for model parameters and functions of model parameters. You can perform both interval- and model-based hypothesis testing. Compare models using Bayes factors. Compute model fit using posterior predictive values and generate predictions. If you want to account for model uncertainty in your regression model, use Bayesian model averaging.

Endogeneity and selection
When explanatory variables are related to omitted observable variables, or when they are related to unobservable variables, or when there is selection bias, causal relationships are confounded, and parameter estimates from standard estimators produce inconsistent estimates of the true relationships. Stata can fit consistent models when there is such endogeneity or selection—whether your outcome variable is continuous, binary, count, or ordinal and whether your data are cross-sectional or panel. Stata can even combine endogenous covariates, selection, and treatment effects in the same model.

Programming
Want to program your own commands to perform estimation, perform data management, or implement other new features? Stata is programmable, and thousands of Stata users have implemented and published thousands of community-contributed commands. These commands look and act just like official Stata commands and are easily installed for free over the Internet from within Stata. A unique feature of Stata's programming environment is Mata, a fast and compiled language with support for matrix types. Of course, it has all the advanced matrix operations you need. It also has access to the power of LAPACK. What's more, it has built-in solvers and optimizers to make implementing your own maximum likelihood, GMM, or other estimators easier. And you can leverage all of Stata's estimation and other features from within Mata. Many of Stata's official commands are themselves implemented in Mata.

Multiple imputation
Account for missing data in your sample using multiple imputation. Choose from univariate and multivariate methods to impute missing values in continuous, censored, truncated, binary, ordinal, categorical, and count variables. Then, in a single step, estimate parameters using the imputed datasets, and combine results. Fit a linear model, logit model, Poisson model, multilevel model, survival model, or one of the many other supported models. Use the mi command, or let the Control Panel interface guide you through your entire MI analysis.

Automated reporting and customizable tables
Stata is designed for reproducible research, including the ability to create dynamic documents incorporating your analysis results. Create Word or PDF files, populate Excel worksheets with results and format them to your liking, and mix Markdown, HTML, Stata results, and Stata graphs, all from within Stata. Create tables that compare regression results or summary statistics, use default styles or apply your own, and export your tables to Word, PDF, HTML, LaTeX, Excel, or Markdown and include them in your reports.

Jupyter Notebook with Stata
Jupyter Notebook is widely used by researchers and scientists to share their ideas and results for collaboration and innovation. It is an easy-to-use web application that allows you to combine code, visualizations, mathematical formulas, narrative text, and other rich media in a single document (a "notebook") for interactive computing and developing. You can invoke Stata and Mata from Jupyter Notebook with the IPython (interactive Python) kernel. This means you can combine the capabilities of both Python and Stata in a single environment to make your work easily reproducible and shareable with others.