Stata 15 is here

[Bill Gould (StataCorp)]

With this message, we announce Stata 15.

Downloads are available now.

Shipped copies begin shipping today.

And of course, you can obtain your copy at our website.

Stata 15 is remarkable for its sheer size. I'll list some highlights here and make some snappy comments, but I make about 20 pages of comments on the blog entry I am about to post, which will be a record. The most thorough coverage, of course, is at the Stata 15 features page at the stata.com website.

Here are sixteen of the highlights:

1. Extended regression. It fits continuous, binary, ordered responses with (1) endogenous covariates, (2) endogenous (Heckman-style) selection, and (3) endogenous treatment effects. This command does a lot and is easy to use.
    
2. Latent class analysis. Binary outcomes. Ordinal outcomes. Continuous outcomes (latent profile analysis). Even incorporate latent classes in path models.
    
3. Bayesian prefix command. You can now type -bayes:- in front of estimation commands to fit Bayesian regression models for over fifty likelihoods. And multilevel models are among them!
    
4. Dynamic stochastic general equilibrium (DSGE) models. Fit structural macro models based on theory with lots of equations. Graph impulse-response functions afterward.
    
5. Markdown for Stata. You can now combine Markdown text with executable Stata code to produce web containing prose, Stata output, parts of Stata output, and graphs. By parts, we mean you can write a sentence such as "The F statistic is 52.4 and ...", where the 52.4 is obtained from -e(F)-.
    
6. Nonlinear mixed-effects models. Because some problems are just not linear in the parameters.
    
7. Spatial autoregressive models (SAR). Download shapefiles from the web, merge with your geographic-based data, and fit your SAR models with cross-sectional or panel data.
    
8. Interval-censored parametric survival-time models. For when you only know a time when subjects have not yet failed and another when they have already failed.
    
9. Finite mixture models (FMMs). The new -fmm:- command is a prefix command. You mix one or more regression models using any of 17 existing estimation commands.
    
10. Mixed logit models. Discrete choice models with random coefficients, meaning you can relax the IIA assumption.
     
11. Nonparametric regression. Our implementation uses kernel methods. Bootstrap standard errors. Graph slices after fitting. Predict derivatives or levels.
     
12. Power analysis for cluster-randomized designs (CRDs) and for regression models. You can now add your own power and sample-size methods too.
     
13. Word and PDF documents. It's now just as easy to produce Word and PDF documents as it is to produce Excel worksheets.
     
14. Graph color transparency/opacity. Stata's colors used to be fully opaque, which is to say, not-at-all transparent. Now you can change them so that you can see what's underneath.
     
15. ICD-10-CM/PCS support. These are the codes from NCHS and CMS that are mandated for all medical billing in the U.S. Support means a database system that can even manage codes from different versions in one file!
     
16. Federal Reserve Economic Data (FRED) support. 470,000 U.S. and international economic and financial time series are available to registered users courtesy of the St. Louis Federal Reserve Bank. You should register. It's free and easy. If you do, Stata can browse, search, and assemble datasets for you.
     

I'll stop, but the following are most certainly highlights too:

• Bayesian multilevel models
• Threshold regression
• Panel-data tobit with random coefficients
• Multilevel regression for interval-measured outcomes
• Multilevel tobit regression for censored outcomes
• Panel data cointegration tests
• Tests for multiple breaks in time series
• Multiple-group generalized SEM
• Heteroskedastic linear regression
• Poisson models with Heckman-style sample selection
• Panel-data nonlinear models with random coefficients
• Bayesian panel-data models
• Panel-data interval regression with random coefficients
• SVG export
• Bayesian survival models
• Zero-inflated ordered probit
• Add your own power and sample-size methods
• Bayesian sample-selection models
• And even more in statistics
• Stata in Swedish
• Improvements to the Do-file Editor
• Stream random-number generator
• Improvements for Java plugins
• More parallelization in Stata/MP

By the way, Stata 15's dataset format is the same as Stata 14's. You can upgrade and still share with those who haven't.

Visit our blog and our features page to learn more.

Stata 15 is in distribution now.

[Sergio Correia]

Still going through the changelog, but just wanted to add that it's also available through the online help file with a bit more detail, for those who want it all in one screen

[Richard Williams]

From now on, the answer to every query on this list will be "Upgrade to Stata 15." This is an impressive list of improvements. Not that I understand all of them, but the ones I do look intriguing.

[Marcos Almeida]

Time to go (back, again and again) to the (new) Stata Manuals!

Hopefully newly-edited books from StataCorp are coming soon!

[Richard Williams]

Is Stat/Transfer 14 also on the way? I didn't see an option to order it.

[Richard Williams]

Is Stat/Transfer 14 also on the way? I didn't see an option to order it.

To answer my own question, Stat/Transfer wrote "Stat/Transfer version 14 is available in the upcoming weeks. Version 13 will work for now with Stata 15 but no prior versions of Stat/Transfer will."

[Steve Dubnoff]

We have just been informed by StataCorp (thank you), that there are changes in the file format to accommodate the larger number of variables (> 32K) that are permitted in Stata/MP version 15. Stat/Transfer will support this change in the next release. In the meantime, those with more moderately sized datasets will have no problems using version 13 of Stat/Transfer with Version 15 of Stata.

[Kimin Eom]

Has STATA15 been improved in its estimation speed by chance? (general SEM in particular) I tried to run a multilevel mediation model using gsem and it took more than a week and ended up giving up...

[Ariel Karlinsky]

I'm sorry to say that on the face of it, this looks like a very disappointing release.
Most of these additions would better be charcaterized as simmilar to user addons than what I would expect from stata itself. It also seems to ignore most if not all of the suggestions and requests made by avid stata users in this very forum. I understand of course that statacorp can't do everything and please everyone, but It seems to me like there's ignorance of what users want and wish.
I would name a few, broad issues that I and others have mentioned and that should've been dealt with on the software level:

1. Interface and results window: The coercive abbreviation of output (of long variable names etc.) - At least give users the ability to decide whether or not they wish to abbreviate (the infamous ~) output.

2. No multi-core support in non-MP versions: Other stat software support multi-core (which is standard in computers now days, and has been for sometime) natively. The price differentiation between MP and non-MP flavors prevents users from utilizing the speed benefits of multi cores. A single "flavour" that utilizes multi cores, I think, is a long time coming.

3. Limited number of variables - while this has increased, there have been several discussions on this very forum how today's "big data" (or even "medium data") sets can have hundreds of thousands of variables. The current limit is not big enough for 2017 and many users (not in my field, btw) would not use stata due to this reason.

4. Better debugging - Being unable to even set a breakpoint in a do file can be extremly frustrating. debugging programs in stata is more art than sciense, with the user writing nonsense code where I want to program to stop (as it will exit due to error) just to "break" at a given time.

5. Incorporating general-use addons into vanilla stata - User addons are great, but I would have expected statacorp to work with package authors to get their packages into native stata. Packages that a large percentage of users use daily, and that even appear on the stata FAQ. such as outreg/estout, spmap, ivreg2 etc.

6. Working with several databases at the same time - I understand that this will mean a major shift in stata-philosophy, but since other stat software to this at ease, I see no reason for stata not to have this pretty basic feature - Instead the user has to juggle with multiple instances of stata, or keep clearing and using each dataset separately.

7. Speed improvements - I see very little mentioning of "under the hood" improvements, for example - are there not still built-in stata commands which have not yet been mata-ized?

[Sebastian Kripfganz]

Ariel Karlinsky:

1. Isn't it already the default setting that variable names are not abbreviated but instead the regression output tables are stretched? See option nolstretch in help estimation options##display_options.

4. You can put an exit in your do-files to force a break.

5. The new sp commands are such an example where user-written addons now became part of native Stata. As you mentioned the spmap command, the new command grmap steps into its place in native Stata if I interpret the documentation correctly.

[Ariel Karlinsky]

Sebastian,

1. I was referring to other commands, such as summarize or table

4. Many workarounds exist. Yet typing commands (and thus changing the file, even if only temporarliy) instead of specifying break-points is bad practice in all programming languages i'm familiar with.

5. your'e correct regarding the new sp commands. when I looked into the spatial models on the stata 15 page, I thought they used examples with old user packages such as shp2dta - I didn't understand that sp is a new host of stata-vanilla commands. I applaud this suite of commands indeed!

[Friedrich Huebler]

Stata 15's dataset format is the same as Stata 14's.

We have just been informed by StataCorp (thank you), that there are changes in the file format to accommodate the larger number of variables (> 32K) that are permitted in Stata/MP version 15.

Is the format of Stata datasets the same as before or has it changed?

[Nick Cox]

Friedrich: I read this as Steve Dubnoff needing to change his code to accommodate the larger number of variables allowed. So, the dataset format is the same, but that doesn't mean that other programs can read the larger datasets. Clearly this is a guess.

[Richard Williams]

From a note I got from Stata:

In general, the dataset format for Stata 15 remains the same as for Stata 14, and I believe Stat/Transfer 13 will still work. The exception is that Stata/MP now allows for up to 120,000 variables, so a new dataset format is used for those sets between 32,768 and 120,000 variables.

[Alan Riley (StataCorp)]

Is the format of Stata datasets the same as before or has it changed?

Both.

Stata 14's dataset format is what we call internally '118' format, as that is the version number in its header. Format 118 is also used by Stata/IC 15, Stata/SE 15, and, by default, in Stata/MP 15.

Format 118 uses two bytes to represent variable numbers, and as such, Stata/MP 15's new maximum of 120,000 variables is impossible to save in a format 118 dataset. If a dataset in Stata/MP 15 has anywhere between 32,768 variables and 120,000 variables, then format 119 is used. Format 119 is almost identical to format 118, but allows for the larger variable numbers. Stata 14 and earlier cannot load datasets with 32,768 or more variables, so it doesn't really matter whether the format is 118 or 119 in that case -- there is no way they can load a dataset of that size.

In any case, the dataset format has not changed in Stata/IC 15, has not changed in Stata/SE 15, has not changed by default in Stata/MP 15, and only changes to allow larger variable numbers in Stata/MP when absolutely necessary.

Finally, saveold in Stata 15 allows datasets to be saved back to Stata 11 format.