[Leonardo Guizzetti]

Similar to the -merge- command, -frget- could use a -replace- or -update- option which would overwrite/update values for existing variables when the user requests it. A strong warning should be issued to the user, but I think the behaviour deserves implementation.

[Niels Henrik Bruun]

Regarding System variables (_variables) we can easily retrieve estimates by _b[] and standard error by _se[].
However, in many cases it is the confidence interval that needs to be reported.
So, I propose two new functions/variables _lb[] and _ub[] for lower and upper bound for the confidence interval.
I know, it can't be seen what the level is directly and hence can lead to wrong reporting.
But still I think that _lb[] and _ub[] are more than nice-to-have

[Clyde Schechter]

I agree with #498--it would be an enormous convenience. But I will just point out that you can pull the confidence bounds from r(table). It's more typographically cumbersome and requires two commands instead of 1, but it's doable.

[wbuchanan]

Clyde Schechter
I’m not sure if all estimation commands return results in r(table), but do know that community contributed commands are not likely to report results in r(table).

[FernandoRios]

wbuchanan
Actually, as long as the community-contributed command produces e(b) and e(V), It is possible to get r(table) output if you use "ereturn display".

[Bjarte Aagnes]

#499 Two commands should not be required:

r(table)["ll","varname"]

[Niels Henrik Bruun]

Bjarte Aagnes in #501: But does not seem to be working in

statsby b=r(table)["b", "died00:1.aar2020"], by(icd10chap) clear: poisson died00 i.aar2020, vce(robust) nolog

Or am I missing something

[William Lisowski]

Niels Henrik Bruun in #502 -

I suspect you are being stung by statsby.ado, which specifies version control for version 9, which does not allow subscripting r(table). Consider the following maximally inelegant workaround.

. sysuse auto, clear
(1978 Automobile Data)

. statsby b=el(r(table),rownumb(r(table),"b"),colnumb(r(table),"weight")), by(foreign): regress 
> price weight
(running regress on estimation sample)

      command:  regress price weight
            b:  el(r(table),rownumb(r(table),"b"),colnumb(r(table),"weight"))
           by:  foreign

Statsby groups
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
..

. list

     +---------------------+
     |  foreign          b |
     |---------------------|
  1. | Domestic   2.994813 |
  2. |  Foreign    5.36204 |
     +---------------------+

[Bjarte Aagnes]

ref #502 #503

. statsby ll=(r(table)["ll","weight"]) , by(foreign) : regress price weight
(running regress on estimation sample)

      command:  regress price weight
           ll:  r(table)["ll","weight"]
           by:  foreign

[Niels Henrik Bruun]

Bjarte Aagnes Thank you

[Leonardo Guizzetti]

I think small change in consistent to frame commands would be reasonable. StataCorp has taken the approach (where possible) to use verbs to make commands more recognizable and improve readability. To this end, we have

frames dir
frame create
frame change
frame copy
frame put
frame post
frame drop
frame reset
frame rename

But then we have

frget
frlink

I'm not sure why the need to make the get/link commands slightly obfuscated. Especially as -frlink- is intended as a (partial) replacement for -merge- when working with frames, it would be nice to have the same consistency in naming conventions of these commands.

[Leonardo Guizzetti]

Another idea. Set operations for vectors within Mata, analogous to the macro list directives. In particular, intersection, union, difference , ===, and -in-.

[Joao Santos Silva]

Motivated by recent posts, I would like to suggest two changes related to FE estimators:

- Disable the margins command after clogit and after xtlogit and xtpoisson when the fe option is used. The results of the margins command in these cases are meaningless and misleading.

- After clogit and after xtlogit and xtpoisson when the fe option is used, Stata should report the full sample size rather than the sample size after dropping uninformative observations. The current practice leads to much confusion and to unfounded fears of sample selection bias.

[daniel klein]

- Disable the margins command after clogit and after xtlogit and xtpoisson when the fe option is used. The results of the margins command in these cases are meaningless and misleading.

margins produces a variety of results. Could you elaborate on which results are misleading and why? Or, could you point me to the respected posts that might elaborate further?

[Joao Santos Silva]

Dear daniel klein,

The threads I had in mind were this one and this earlier one.

Best wishes,

Joao