StataCorp Stata
Version 17.0 · StataCorp · Windows 10 / 11 (64-bit) · ranked #217 in the library
The statistics package that half of applied econometrics and epidemiology is written in, where every analysis is a command with a reproducible log rather than a sequence of clicks nobody can retrace.
Read the installation section below before you start. The order of the steps matters on this release.
About this release
Overview
Stata is a statistical package built around a command language. You can drive it from menus, and the menus write the command into the results window so you can see what they did, but serious work happens in do-files: scripts that read data, clean it, run the models and produce the tables, all reproducible from raw input to final output. That reproducibility is why journals and research groups standardised on it.
The estimation coverage is broad and deep on the inference side rather than the machine learning side. Linear and generalised linear models, panel data with fixed and random effects, instrumental variables, survival analysis, multilevel and mixed models, time series with ARIMA and state space methods, survey data with proper design based standard errors, treatment effects, structural equation modelling and multiple imputation are all first class, with consistent syntax and consistent postestimation commands across them.
Data management is a large part of the daily work and it is handled properly: reshaping between wide and long, merges with match diagnostics, collapsing to summary datasets, labels on variables and values that follow the data into output, and frames for holding several datasets in memory at once. Graphics are command driven and fully specified, so a figure regenerates identically from a script months later, and results export to LaTeX, Word, Excel and PDF through the reporting commands.
What it does
Feature breakdown
Command language with reproducible do-files
Every action is a command that can be scripted, logged and rerun, so an entire analysis regenerates from raw data with one file.
Broad estimation coverage
Panel, survival, multilevel, time series, survey, causal inference and structural equation models with a common syntax and shared postestimation tooling.
Serious data management
Reshape, merge with diagnostics, collapse, encode, label and validate, plus frames for working with several datasets simultaneously.
Publication quality graphics
Fully scriptable graphs with schemes controlling every visual element, exported to vector formats that hold up in print.
Reporting and table generation
Build formatted tables and complete documents straight from results, exported to LaTeX, Word, Excel, HTML or PDF.
Extensible with user packages
A large ecosystem of community written commands installs into the program and behaves like built-in commands, with the source readable.
Changed in this build
Version 17.0
- New table command with a unified syntax for building and exporting result tables.
- Bayesian estimation extended to more model classes with improved sampler diagnostics.
- Difference in differences and treatment effect commands expanded with heterogeneity robust estimators.
- Faster handling of very wide datasets in the multithreaded edition.
- Do-file editor gained better autocompletion and inline help lookup.
Installation
Follow these in order
- Take the machine off the network before installing.
- Extract the archive to a folder with a few gigabytes free.
- Pause real time protection.
- Run setup as administrator and choose the edition you want installed.
- Apply the included activation before launching for the first time.
- Start Stata, confirm the edition line in the startup banner, then restore protection.
Release notes
From whoever packed it
- The edition determines how many processor cores are used. Pick the multiprocessor build if the machine has the cores to feed it.
- Do not run the update command inside the program. It replaces the executable and the activation with it.
- Community packages install to a separate folder and survive a reinstall of the program itself.
System requirements
Minimum and recommended
| Operating system | Windows 10 or Windows 11, 64-bit |
|---|---|
| Processor | Any modern multicore processor, more cores used by the multiprocessor edition |
| Memory | 8 GB minimum, more directly determines the dataset size you can hold |
| Graphics | No specific requirement |
| Storage | 3 GB for the program and documentation |
Questions about this release
Answered before you ask
- Does it read files from other statistics packages?
- Yes, and it exports to them. Delimited text, Excel, SAS and SPSS files all import directly.
- Is the memory limit fixed?
- No. It uses what the machine has, so dataset size is limited by installed memory rather than a setting.
- Can I run it without the interface?
- Yes, batch mode runs a do-file from the command line and writes a log, which is how it ends up on compute servers.
Download
Pick a mirror