[John Mullahy]

Different people reasonably want to see different summary statistics in Stata's reports of regression results. Yet I can't imagine I'm alone in thinking it would be valuable to report routinely the dependent variable's estimation-sample mean (alongside the other summary statistics) in summaries where this makes sense (eg. probit but not oprobit).

I'd be willing to wager a dollar (or pound or euro or...) that if you surveyed Stata users they would choose this over other parameters like adjusted R-squared, etc. But there's no need to eliminate the statistics that are currently reported, instead simply add this one.

[David Roodman]

While I appreciate how the python functionality gives us access to matplotlib and ggplot2, it still seems that Stata graphics are falling behind the competition. Compared to them, there are missing graph types, and I think the whole graphics system is slow.

I'd like to see Mata functions that speed up certain operations on matrices. E.g., if I want to get the sum of a submatrix, I'd guess that sum(X[|S|]) causes Mata to copy the relevant part of X to somewhere else in memory, sum it, and then trash it. Inefficient. Other examples would include filling a given set of rows/cols of a matrix with copies of a given row/colvector; computing the norms of all the rows or cols of a matrix with respect to a quadratic form given by a single square matrix; and other operations on submatrices, such as adding a fixed amount or zeroing out.

Support for higher-dimensional arrays in Mata! Sometimes dealing with this gap forces complicated work-arounds.

[Jeff Wooldridge]

I'd like to see HAC options built into regress. For beginners of regression with time series, it's confusing to have to use the newey command to compute Newey-West standard errors because it seems to imply the estimation might be other than OLS. A HAC option is built into ivregress, and that should be migrated to regress, too.

Requiring the newey command to compute standard errors and statistics robust to serial correlation and heteroskedasticity is tantamount to dropping the vce(robust) option for regress and requiring something like

eicker y x1 x2 ... xK

or

ehw y x1 x2 ... xK

to compute the Eicker-Huber-White heteroskedasticity-robust standard errors and test statistics. Fortunately, Stata did not take this route, as I suspect some beginners would think that weighted least squares or some other method is being used.

This is nothing against Newey and West; obviously their method has been very important for time series regression. It's a call for logical consistency and for helping researchers understand that OLS is one estimator and there are many different ways to compute standard errors. (Occasionally I'm asked about "Newey estimation" or "Newey regression." There's no such thing. There's OLS with adjusted standard errors.)

In my mind, it makes sense to allow the following, after the appropriate tsset:

reg y x1 ... xK, vce(hac kernel #)

where kernel can be any of the three kernels allowed in ivregress and # is the (optional) truncation lag.

I know regress is a special case of ivregress, but, again, this puts an undue burden on beginners who may not even be learning instrumental variables.

[Clyde Schechter]

Re #398: good idea, but I think some other syntax would be needed because otherwise -reg y x0 ##(x1 x2 x3 x4)- would be ambiguous as among a 5-way interaction of all x's (and all subinteractions), vs simple two-way interactions of x0 with x1 through x4 separately vs uninteracted x0 along with the four-way interaction of x1 through x4 (plus all subinteractions).. One might let the presence or absence of a blank space between x0 and ## disambiguate that, but generally in Stata spaces are not syntactically relevant. And that is true of most programming languages because it is very difficult for people to maintain consistency about their use of spaces. Maybe ###(x1 x2 x3 x4) would work?

[Julian Reif]

I would like to see Stata incorporate the user-written command -gtools- into its built-in commands.

I would also like Stata to develop a fast regression command (written in C) that can accommodate multiple levels of high-dimensional fixed effects. Currently the only option is -reghdfe-, a user-written command. While -reghdfe- is fantastic--indeed, I cannot do much of my research in Stata without it--its speed now lags behind R commands such as -lfe- and -fixest- by a factor of 10. These types of regressions are now very common in applied economics, and user demand for software that can run them will only increase going forward.

[John Mullahy]

When using factor variables it would sometimes be helpful to have shorthand in addition to what is currently offered. For instance, instead of

reg y x1##x2##x3##x4

one might have something like

reg y ##(x1 x2 x3 x4)

[William Lisowski]

Ronald Anguzu #396 -

This topic is a wishlist of features wanted in the next release of Stata. Your question is out of place here and not likely be answered in this location. Instead, return to the General Forum page, click on the "+ New Topic" button near the top, and pose your question again in a new topic - giving it a helpful name.

With that said, you may be able to solve the problem yourself by becoming better acquainted with how Stata handles date and time data. Stata's "date and time" variables are complicated and there is a lot to learn. If you have not already read the very detailed Chapter 24 (Working with dates and times) of the Stata User's Guide PDF, do so now. If you have, it's time for a refresher. After that, the help datetime documentation will usually be enough to point the way. You can't remember everything; even the most experienced users end up referring to the help datetime documentation or back to the manual for details. But at least you will get a good understanding of the basics and the underlying principles. An investment of time that will be amply repaid.

[Ronald Anguzu]

I am trying to compute inter-pregnancy intervals (in months) and have the variables to compute this interval between two dates. However, the "day of birth" variable is missing. Dataset only has (mm) and years (yy).

My goal: To calculate the pregnancy interval between most recent birth and previous live birth in months.

The numeric variables are listed below;

Most recent birth
1. Month of most recent birth (mm_dob)
2. Year of most recent birth (yy4_dob)

Previous live birth
3. Month of last live birth (mm_llb)
4. Year of last live birth (yy4_llb)

I will be most grateful for your help.

[Erik Ruzek]

I would love to see Stata offer the option to use alternative estimators for sem/gsem. Specifically, with categorical data, diagonal weighted least squares (DWLS; WLSMV in Mplus) is often preferable to maximum likelihood (the only option in gsem). When the number of response options is low, maximum likelihood can be problematic, as shown in a number of studies. I am aware that sem offers the ADF estimator, but this is a band aid for categorical items. There are workarounds for EFA but it would be nice to have this baked into sem/gsem.

[Leonardo Guizzetti]

Hi Orvalho, I understood you correctly the first time. It does require you to change the elements once, then save your preference set. If you work on multiple computers as I do, I set this up once per computer and then it's stored. It's a minor inconvenience, but technically feasible at the moment.

[Orvalho Augusto]

You can do this already. Set up the look of all your windows as you like them, then go to Edit > Preference Set > Save preference set > New ... and define the name of your preferences. Later, if you change colours or themes, you can restore your saved preference set by loading it once again.

Thank you for the response about the Dark Theme.
For the Viewer and the Output window we have something like "classic" our "Dark" color scheme.
For the do-editor, there is not such a thing. One has to change the colour of every element. Working with different laptops my settings get messed all the time (and I imagine this happening to other a couple times). Is that clear?

[Lars Pete]

Dear all,

I have a suggestion. STATA faq pages such as these https://www.stata.com/support/faqs/data-management/creating-group-identifiers/
and UCLA modules such as these https://stats.idre.ucla.edu/stata/seminars/stata-programming/
have been more helpful for me than STATA manuals, since the former address specific questions/tasks and address them right at the heart, which a STATA manual is not designed to do. I understand that STATA manuals are much more comprehensive and formal. However, they are often not helpful when you are set to do a particular task and you have to turn to STATALIST to do little things, which though might be present in the manuals but hard to find. The jargon in STATA manuals often stands in the way. In such cases, modules such as this are helpful since they are easy to find as per your question/task at hand.

It would be great if STATA could update the faq pages to incorporate the discussions and questions answered on STATALIST. The only difference would be the faq page would have all the important excerpts of STATALIST discussions and questions answered, related to a particular topic/task at one place. I think, this would be really helpful in consolidating all the scattered knowledge and induce great ease in searching commands/troubleshooting related to that particular task. For more comprehensive knowledge, we can always turn to STATA manuals.

[Leonardo Guizzetti]

Please add a dark theme on the doeditor. We can change each feature (number, strings, comments etc...) but it would be nice to have something that we just click and voila we get a dark theme.

You can do this already. Set up the look of all your windows as you like them, then go to Edit > Preference Set > Save preference set > New ... and define the name of your preferences. Later, if you change colours or themes, you can restore your saved preference set by loading it once again.

[Orvalho Augusto]

Please add a dark theme on the doeditor. We can change each feature (number, strings, comments etc...) but it would be nice to have something that we just click and voila we get a dark theme.

[Felix Stips]

Generalized Additive Models for Location Scale and Shape could be helpful for some people, see gamlss in R.