Survey methods
Whether your data require a simple weighted adjustment because of differential sampling rates or you have data from a complex multistage survey, Stata's survey features can provide you with correct standard errors and confidence intervals for your inferences. Simply specify the relevant characteristics of your sampling design, such as sampling weights (including weights at multiple stages), clustering (at one, two, or more stages), stratification, and poststratification. After that, most of Stata's estimation commands can adjust their estimates to correct for your sampling design.

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.

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.

Panel data
Take full advantage of the extra information that panel data provide while simultaneously handling the peculiar difficulties that panel data present. Study the time-invariant idiosyncratic 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 correlated random-effects models. Fit linear models with high-dimensional fixed effects. Fit dynamic models, vector autoregressive models, or models with endogeneity. Fit Bayesian panel-data models.

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.

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), or count (number of children), don't worry. Stata has maximum likelihood estimators—probit, ordered probit, Poisson, 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.

Structural equation modeling (SEM)
Estimate mediation effects, analyze the relationship between an unobserved latent concept such as a person's level of conservatism and the observed variables that measure conservatism, 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.

Latent class analysis (LCA)
Discover and understand unobserved groups (latent classes) in your data–whether the groups are healthy and unhealthy individuals or teens with high, medium, and low risk of high school drop out. Use LCA as a model-based method of classification. Or fit SEM path models and test for differences across the unobserved groups. Estimate the proportion of the population in each group, estimate group means, perform latent class model-comparison tests, and more.

Adjusted predictions, interactions, and moderation
Adjusted predictions and marginal means let you analyze the relationships between your outcome variable and your covariates, even when that outcome is binary, count, ordinal, or categorical. Compute adjusted predictions with covariates set to interesting or representative values. Or compute marginal means for each level of a categorical covariate. Make comparisons of the adjusted predictions or marginal means using contrasts. If you have multilevel or panel data and random effects, these effects are automatically integrated out to provide marginal (that is, population-averaged) estimates. After fitting almost any model in Stata, analyze the effect of moderating variables, and easily create interaction plots.

Choice models
Model your discrete choice data. If your outcome is, for instance, high-school graduates' choices to attend college, attend a trade school, or to work, you can fit a conditional logit, multinomial probit, or mixed logit model. Is your outcome instead a ranking of prefered alternatives? Fit a rank-ordered probit or rank-ordered logit model. Regardless of the model fit, you can use the margins to easily interpret the results. Estimate how much distance to the nearest college affects the probability of enrolling in college and even the probability of going to a trade school.

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. Use Bayesian variable selection for linear regression to identify predictors important to your outcome and perform Bayesian inference.

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. Explore treatment-effect heterogeneity across individuals or across groups with conditional average treatment effects (CATEs). 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.

Machine learning
With machine learning via H2O, you can use ensemble decision trees—random forests and gradient boosting machines—for regression and classification. Or use lasso for sparse regression and classification. Or use Bayesian variable selection or Bayesian model averaging to select predictors in a linear model. For causal inference with machine learning, use double-selection lasso, partialing-out lasso, and double machine learning. You can use PCA or kmeans, kmedians, or hierarchical clustering for unsupervised learning. And use search to find community-contributed commands for neural networks, support vector machines, graphical lasso, text mining, and more.

Programming
Want to program your own commands to perform estimation, perform data manipulation, 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.

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.

Reproducibility
Stata is the only software for data science and statistical analysis featuring a comprehensive integrated versioning that ensures your code continues to run, unaltered, even after updates or new versions are released. No need to keep around multiple legacy installations to avoid breaking your system; Stata code from 40 years ago can still be run without modification. Datasets, graphs, scripts, programs, and more are 100% cross-platform and backward compatible.