[jerome falken]

A few cosmetic issues still present in 15.1 on OSX:

* When in "mata mode", the "Data Browser" icon should stata("browse") instead of "browse" (which isn't a mata command)
* When in "mata mode", upon defining a view, the data isn't updated in real-time in the "Data Browser", while it is in stata mode upon "replace" (a workaround would be a "refresh" button).

[Daniel Feenberg]

Several users here have asked for the ability to return a non-zero return code to the operating system from the -exit- statement Currently there is the ability to have a non-zero return code from a do-file to the console Stata instance, but not from the console back to the OS. That would be valuable for batch jobs.

[John Mullahy]

Might there be a way to locate Stata graph axes other than left/right or top/bottom? For instance, if my objective is to depict (x,y) data where all x's and y's are in (-1,+1), it would be nice to be able to locate the (0,0) origin at the point where the axes intersect. (I thought I had seen this issue raised elsewhere but couldn't find it; apologies if repetitive)

[Sergiy Radyakin]

Another wish for Stata 16 is to add an option noextmiss for outsheet/export delimited commands to suppress extended missing values .a, .b, ... .z and output them as system missing values.
Thank you, Sergiy

[ericmelse]

Stata 16 wishlist:
The possibility to run Stata in some configuration (server mode), or compile Stata syntax into some DLL, to make real time data processing possible on an Internet server.
What I am looking for is to facilitate real time data entry by website visitors, i.e. record their responses to Likert type questions, and, once the questionnaire is completed, automatically run an IRT-model, compute theta-scores etc., generate png files, and finally create a PDF (using Stata's putpdf syntax).
I suppose this will require (new) controls to trigger Stata, or the DLL, to compute etc.
So, it is not to interact as a user with Stata through a web interface but to be able to run Stata commands automatically, like if it was syntax in a do file.

My suggestion above is actually already possible using Numerics by Stata, which I just noticed should be able to deliver what I am looking for.

[ericmelse]

Stata 16 wishlist:
Graded Opacity
The possibility to control the opacity of colors using a variable, very much like how it is now possible to control the size of markers.
E.g., using the example on page 240 from A Visual Guide to Stata Graphics that uses the file allstates.dta where the size of markers is set by [aweight=propval100]:

use allstates , clear
twoway scatter ownhome borninstate [aweight=propval100], msymbol(o) mfcolor(cyan%20)

I added the parameter mfcolor(cyan%20) to set the symbol fill color to cyan and control the opacity. This renders the graph like:

As such, the above functionality is very nice but now a next improvement could be made to include the option to control opacity by means of its own weight as set by a variable.
I suppose the most simple syntax would be to enable the reference to a variable name after the opacity percentage indicator, like: mfcolor(magenta%ownhome).
To have an idea what this would look like, here is the same graph but with color graded opacity by the data of ownhome:

In this example the 'base color' against which the color is graded is gray (but it could be white or its opposite color).
More sophistication could be possible, but I suppose the mere option to control using data of a variable would be a great visualization enhancement of the opacity functionality.

[daniel klein]

I am still a bit skeptical about Allison's statement (from the quoted paper in #94) that

Under identical assumptions, both methods [i.e., MI and FIML] produce estimates that are consistent, asymptotically efficient and asymptotically normal.

What are these identical assumptions?

The Stata manual for sem implies that the assumption of multivariate normality is to be taken pretty seriously with MLMV (or FIML if you prefer):

Method MLMV formally requires the assumption of joint normality of all variables, both observed and latent. If your observed variables do not follow a joint normal distribution, you may be better off using ML, QML, or ADF and simply omitting observations with missing values.

So, yes, both multiple imputation and FIML make the same assumption of the data being MAR (or MCAR), and, yes, multiple imputation via the mutlivariate normal model (mi mvn) , as the name implies, and FIML assume joint normality. However the much more often used chained equations approach of multiple imputation does not require joint normality. So I wonder (a) if the assumptions are really the same and (b) if the FIML method can be generalized to, simultaneously, take into account different distributions, as used by Stata's gsem routine. Perhaps this is the wrong place to discuss this further.

Best
Daniel

[Jonathan Horowitz]

I will repeat my long-standing wish: That Stata would implement FIML (full information maximum likelihood, called MLMV in Stata) for all routines.

As Paul Allison states, this is easily the best way to handle missing data: http://www.statisticalhorizons.com/w...ngDataByML.pdf

I have no idea if this is possible, and maybe the solution is to speed up -sem- and the mlmv option to the point where you can just use it instead of -regress-, but this is probably the only area (that matters to me) where Stata lags far, far behind SAS.

[ericmelse]

Stata 16 wishlist:
The possibility to run Stata in some configuration (server mode), or compile Stata syntax into some DLL, to make real time data processing possible on an Internet server.
What I am looking for is to facilitate real time data entry by website visitors, i.e. record their responses to Likert type questions, and, once the questionnaire is completed, automatically run an IRT-model, compute theta-scores etc., generate png files, and finally create a PDF (using Stata's putpdf syntax).
I suppose this will require (new) controls to trigger Stata, or the DLL, to compute etc.
So, it is not to interact as a user with Stata through a web interface but to be able to run Stata commands automatically, like if it was syntax in a do file.

[jerome falken]

Stata 16 wishlist:
* I would like to see the DCC-MIDAS model with mixed data sampling (MIDAS) added to mgarch.
* faster contour plot

[Chris Larkin]

When you have to enclose snippets in parentheses, brackets, quotation marks, etc. it would be neat to be able to highlight section and hit that symbol.

E.g. if i had the code bysort ip othervar: gen ipseq = _n, and i just want the running _n to be for ip but to sort on othervar then i could just double click on othervar and hit shift + 9 for parentheses to appear around it.

[daniel klein]

Nick: Nice!

[Nick Cox]

daniel klein
Clyde Schechter


That's part of my "not always" in #87.

It's a little tricky but long since possible to concatenate macro names to make a longer readable version of variable names.

Consider

local patient p_
local history h_

so that you could have

`patient``history'*

as a more informative variant on

p_h_*

assuming a bundle of variables on patient history. Clearly you can invent examples of your own close to home.

[daniel klein]

Long variable names will often (not always, agreed) be fairly useless whenever you can't see them.

True, but you can always see long variable names when you read code. This is how I understand the example given by Clyde in #83. Form this perspective, any output of variable names is indeed irrelevant.

Best
Daniel

[Nick Cox]

Andrea Discacciati Those properties will not be orthogonal, much though we all like making such quips.

Long variable names will often (not always, agreed) be fairly useless whenever you can't see them.

I don't know how you'd calculate the correlation, but my prior on it isn't centred at zero.