Stata 16 announced

[Bill Gould (StataCorp)]

We just announced Stata 16. The webpages are up at stata.com that describe the release. In just a moment, I'll post my full comments about the release at the Stata Blog.

There are 21 highlights:

1. Lasso, both for prediction and for inference
2. Reproducible and automatically updating reports
3. New meta-analysis suite
4. Revamped and expanded choice modeling (margins works everywhere)
5. Integration of Python with Stata
6. Bayesian predictions, multiple chains, and more
7. Extended regression models (ERMs) for panel data
8. Importing of SAS and SPSS datasets
9. Flexible nonparametric series regression
10. Multiple datasets in memory, meaning frames
11. Sample-size analysis for confidence intervals
12. Nonlinear DSGE models
13. Multiple-group IRT
14. Panel-data Heckman-selection models
15. NLMEs with lags: multiple-dose pharmacokinetic models and more
16. Heteroskedastic ordered probit
17. Graph sizes in inches, centimeters, and printer points
18. Numerical integration in Mata
19. Linear programming in Mata
20. Do-file Editor: Autocompletion, syntax highlighting, and more
21. Stata for Mac: Dark Mode and tabbed windows

Certainly there must be something in this list that will interest you.

Pay attention to number 10. I don't mention this about frames in the blog, but frames are programmable and I expect that frames will launch a new category of community-contributed software. Try not to take the good names. We have our own plans, and I remind you that we have a long history of appropriating good names.

We also have a long history of letting users lead the way. Along those lines, I direct your attention to items 1 and 3. We not only appropriate command names, we appropriate good ideas. Be aware, however, that these are our efforts and our code, and if there any shortcomings, we are the responsible party.

Getting a release out is misery. There is little, however, that is as gratifying as announcing it.

[Philip Jones]

Sounds like a great release. Looking forward to reading your detailed comments!

[Sergiy Radyakin]

Congratulations on the new release!

Many new and welcomed features (for me, humbly: 5, 8, 10, 20, but I am sure there are tons more behind the scenes). Thank you!!

Has the data file format changed again due to #10 to store multiple frames together? Or each frame is saved in the same dataset format compatible with v15 as it seems to be indicated by https://www.stata.com/help.cgi?dta ?

Thank you, and best regards!

[Ariel Karlinsky]

Looks like a great release, with many wanted features such as working with multiple datasets, better do-file editor with auto-complete and more. Kudos!

[Richard Williams]

I am curious about the new maintenance option:

https://www.stata.com/order/maintenance/

It says "Support. You will enjoy uninterrupted access to technical support from a dedicated staff of expert Stata programmers and statisticians." It seems like I have that now. Does this mean that Stata Tech are going to start ignoring me if I don't get a maintenance agreement?

Otherwise the benefits seem pretty minor to me. Unless prices of future editions are going to skyrocket, it seems like you are mostly paying to get new versions a little earlier. I'm a fanatic but not that much of a fanatic.

I plan to upgrade soon (in fact I planned to do it already!) but since maintenance agreements can only be bought at the time of purchase I want to get a better feel for how much it is really worth it.

[Marc Kaulisch]

Richard Williams I think this is more an option for businesses that they can deduct their expenses from taxes yearly.

I am curious to know if handling/importing JSON-formatted files getting easier as well - or will this depend on the Python integration...

[Dick Campbell]

I'd guess this is primarily a means of smoothing out StataCorp's cash flow and its tax costs. It implicitly assumes a new version every two years; otherwise buying the agreement to get "free" updates makes no sense. At current cost levels, the two year cost of the maintenance agreement for plain old Stata MP is slightly less than if one were to ignore it and buy the next upgrade. The discount increases as one moves up to more powerful levels, e.g. from 2 cores to 4. As Richard Williams notes the real question is access to tech support. It would be unfortunate if no maintenance agreement means no tech support. And if that is the case then the more powerful level (flavor?, I never get that straight) of Stata you have the more you are paying for support because the maintenance cost increases as you go up the ladder.

[Richard Williams]

Marc Kaulisch you may be right. But, it concerns me when it says that, with the maintenance agreement, I will get support. It also says you will get updates and security patches. But, I already get those things, at no charge. Am I now only going to get those if I pay for a maintenance agreement? If so, that is a major reduction in what Stata currently offers. But if, in fact, these things are staying the same, the maintenance agreement is claiming as benefits things you would get anyway.

if my fears are unwarranted it seems like I am committing myself to buying Stata 17 two years in advance and getting only a small cost reduction in return.

[Richard Williams]

I am also confused about the pricing. With Educational pricing, I can get Stata 16/MP2 for $545. The maintenance agreement is $260 a year. So, if I pay $260 now, $260 a year from now, and $260 two years from now, at which point I get Stata 17, haven't I paid $780 for it?

A straight subscription plan for $260 a year would make more sense to me. Or, maybe say a 1 year subscription is included for free.

I'm afraid not to get the agreement because it sounds like I may get cut off from support! But, if that isn't the case, the benefits of the agreement seem pretty trivial, and may even cost a lot more in the long run.

[John Mullahy]

Richard: From https://www.stata.com/support/tech-support/policy/

Eligibility for Technical Services

All registered users of the current release of Stata (Stata 16) and the previous release (Stata 15) are eligible for free technical support. If you have not registered your copy of Stata, please fill out the online registration form.

[Karen Strope (StataCorp)]

I am curious about the new maintenance option:

https://www.stata.com/order/maintenance/

It says "Support. You will enjoy uninterrupted access to technical support from a dedicated staff of expert Stata programmers and statisticians." It seems like I have that now. Does this mean that Stata Tech are going to start ignoring me if I don't get a maintenance agreement?

Otherwise the benefits seem pretty minor to me. Unless prices of future editions are going to skyrocket, it seems like you are mostly paying to get new versions a little earlier. I'm a fanatic but not that much of a fanatic.

I plan to upgrade soon (in fact I planned to do it already!) but since maintenance agreements can only be bought at the time of purchase I want to get a better feel for how much it is really worth it.

I wanted to reassure you that our technical support policy has not changed. Current users and users of the previous version will continue to have access to free technical support whether you have a maintenance agreement or not. The current version of Stata will also continue to receive updates and patches.

While we certainly have many users that maintain staying on the current version on their own, we also have users that fall behind the current and previous release, often without even realizing it. We introduced maintenance a few releases ago. It allows users to stay on the latest release without having to plan for upgrades. Staying on the latest release ensures the users will maintain access to technical support, the latest features, and updates. As others have mentioned, it is also used as a budgeting tool to plan for upgrade expenses.

Whether you prefer to select the upgrade when it is available or prefer to go with the maintenance plan that you don't have to think about is up to you. We certainly didn't mean to alarm you. I hope you enjoy Stata 16!

[Richard Williams]

Thanks Karen. I don't think the management agreement is right for me personally and I appreciate you confirming that not having it won't hurt me in the ways I feared.

[Rasool Baloch]

Is there any new feature/enhancement in Stata's Project Manager?

Regards,
Rasool Bux

[Nicholas Winter]

It will be a bit before I can upgrade, but I'm looking forward to the Python integration in particular (having requested it on the Stata 16 wishlist thread...). One question I have from looking at the online documentation for it: is it the case that Stata .ado code that includes Python will be less-fully portable and shareable, compared with straight .ado code, and compared with compiled Java programs? It sounds like Stata relies on the Python that users have installed on their systems, so if I write a program that uses Python I can't be guaranteed it will run on another Stata user's system. Is this, in fact, the case?

[Richard Williams]

I'm glad to see hetoprobit added. My oglm program (available from SSC) has been basically doing the same thing since 2009, but it won't surprise me if Stata's code is quicker plus StataCorp is more likely to keep its programs updated 20 years from now than I am.

Here is the oglm code that replicates the Stata manual's examples for hetoprobit:

use https://www.stata-press.com/data/r16/eathealth15
oglm health age bmi i.exercise, het(age i.exercise) link(p)
predict pr1, pr outcome(1)
twoway scatter pr1 bmi, ytitle("Pr(Poor Health)") title("Effect of BMI on Pr(Poor Health)")
margins, at(age = (30(10)70)) predict(outcome(1))
marginsplot
margins, at(age = (30(10)70))
marginsplot
margins, dydx(bmi) at(bmi = (20(5)35)) predict(outcome(1))
margins, predict(s) at(age = (15,85)) noatlegend

StataCorp will be relieved to know that most results are identical to 6 decimal places. However, I did notice that, for the last confidence interval given, oglm says 1.629891 whereas hetoprobit says 1.629884. I will write Stata tech support and let them know there must be a bug in their program. ;-)