Universal and Mass
Customization of Tables in Stata
Roy Wada
University of Illinois at Chicago
Purpose of this presentation:
• Strong demand for universal approach to
systematic table-making in Stata
• Strangest advice seems to be coming from
people with no background in empirical
• Latest vaporware features in outreg2
Recent complaints about table-making
• “With eight dimensions it is difficult to see how any program
could cope with line lengths ~ 100 (rather than say ~ 1000)
and not produce an awful mess one way or the other.” - Tue,
21 Apr 2009
• “But there is a point where it no longer makes sense for
official Stata or ssc contributions to support complicated
structures that are only needed every once in a while...” Wed, 11 Nov 2009
• “Stata has *no* ability to do this (though I assume one could
program it, but I doubt it would be easy)” - Thu, 13 Jan 2011
Universal table-making
• There is no important difference between various
types of tables
– Regression tables & summary statistics are merged on
conditions (happens to be variable names)
– Cross-tabulation (Stub-and-Banner)
• Tabulation is merely conditional counting
• Cross-tabulation is a conditional counting merged on
• Stub-and-Banner happens to be a particular type of
conditional counting
Mass customization
• “Mass customization … is the use of flexible
computer-aided manufacturing systems to
produce custom output” (Wikipedia 07/14/2011)
• A solution was to do it column by column (the
original outreg by John Gallup), which had been
criticized as somehow nontechnical or outdated
Some issues with previous efforts
• Many good programs exist (outreg, estout,
parmest, xml_tab, tabout, etc)
• General commentary:
– Rube-Goldberg syndrome – over-engineered nonsolutions for performing “simple” tasks
– Mata-based programs are expensive and generally not
extensive nor easily upgraded
– Wrapper-based programs suffer from the existential
question, i.e. then why didn’t Stata Corporation do it
that way
– Some programs were clearly designed by people with
no background in empirical research
Desirable functionality
• What researchers wanted was quantity of tables,
not quality (hundreds of regressions per day)
• Rarely for publication purposes (they don’t get
• Exact formatting often gets destroyed by the
journal type-setters anyway
• Excel is a fact of life – virtually every researcher
uses it
• LaTeX is not that popular – just look at the
working papers floating around, clearly
something else
Latest vaporware in outreg2
• Already does cross-tabulation
• Extending it to including summary stats is
• Sideway-tranpose operation doubles the type
of table format
• Some minor tweaks and bug fixes
• Probably by the beginning of August 2011
Terms of use (fair use)
• outreg2 implements tasks previously unknown
in Stata or described nearly impossible. It is
provided as a professional courtesy. I strongly
object to re-publication this work of under
false pretense
• Plagiarism is unethical, unprofessional, and
academically dishonest, and egregious cases
should be publicized as a deterrent.

similar documents