The big conclusion from this is that writing complicated/large packages like this in Stata is a PITA. Even well thought code (estout) gets really complicated as complexity increases. A saner alternative would be to code in in Python, but we would need a package that reads .ster files
This is an experimental package intended for personal use.
cap ado uninstall quipu
net from https://raw.githubusercontent.com/sergiocorreia/quipu/master/package/
net install quipu
What's the proper level of abstraction for this? Ideally we don't want to have to go into the details of VCV footnotes, etc.--and more generally-- we don't care about presentation details, instead just on what we want to report and the program takes care of everything.
However, abstractions leak. A lot. See the -estout- examples. mgroups() should be able to group by e.g. estimation method from e(), and shouldn't need the pattern suboption.
We shouldn't care about point estimates to form beliefs, but if the only thing reported is the beta and the pvalue, then we do. This is particularly true if the Pvalue is close to zero, as it's non-informative.
Why not just report the Tstat so people can at least do a 66% ci (beta+-tstat). I see way too many people trying to extract info. from the betas without checking how wide are the CIs (maybe 95% CIs are too harsh, but at least use 66%)
- Same LHS, but different samples: I need to replace e(depvar)
- Group cols by a category, and label it. Problem: How to set it? (how to map each depvar or regr into a cat?)
- Same LHS but different RHS (e.g. different versions of a control). Instead of having it like an identity matrix (waste of space), just label the difference in the header and rename the vars so they are the same
- What about subheaders?
- In general, what if we just set e.g. headers(method depvar #) -> # is (1), the others are just e(). If the FIRST (besides #) is repeated, group by it (and span by it). If the second is repeated, do the same. But how to set up estout for this?!?
header(group varname method #)
labcol2(+ ? + -, title("" Hypothesis))
indicate(rep dummies = _Irep78*)
Problem: These don't work well with areg
and reghdfe
.
refcat(_Idrug_2 "Placebo", label(1))
refcat(weight "Main effects:" turn "Controls:", nolabel)
rename(altmpg mpg)
There is a nice syntax for saying how to allocate stats within cells:
stats(F p N, layout("@ @" @) fmt(a3 3 a3) labels("F statistic" "Observations"))
Basically, quotes are the row delimiters. By using "@" as placeholders, you can add text/format to the cells.
Some stats have meaningful labels by default. Other complex stats ("weakid", etc.) are also included. Also, there are some sensible defaults depending on the model. If you want to keep them and add others on top, use the keyword default
mgroups(A B, pattern(1 0 1 0)
begin, delimited and end control the delimiters in a row. incelldelimiter is within a row. Also see: hlinechar interaction mlabels mgroups numbers abbrev modelwidth varwidth (span?)
Sergio Correia, Duke University
- Labels
- RHS
- LHS
- Notes
- Footnotes
- VCV
- Significance
- Group columns
- Main groups
- Subgroups
- asd
- For later
- HTML: use rowspan and colspan attributes for the TH and TD