The Stata Journal (2005)

5, Number 3, pp. 288–308

Making regression tables from stored estimates

Ben Jann
ETH Zurich, Switzerland

Abstract. Organizing and archiving statistical results and processing a subset

of those results for publication are important and often underestimated issues
in conducting statistical analyses. Because automation of these tasks is often
poor, processing results produced by statistical packages is quite laborious and
vulnerable to error. I will therefore present a new package called estout that
facilitates and automates some of these tasks. This new command can be used
to produce regression tables for use with spreadsheets, LATEX, HTML, or word
processors. For example, the results for multiple models can be organized in
spreadsheets and can thus be archived in an orderly manner. Alternatively, the
results can be directly saved as a publication-ready table for inclusion in, for
example, a LATEX document. estout is implemented as a wrapper for estimates
table but has many additional features, such as support for mfx. However, despite
its flexibility, estout is—I believe—still very straightforward and easy to use.
Furthermore, estout can be customized via so-called defaults files. A tool to
make available supplementary statistics called estadd is also provided.
Keywords: st0085, estout, estoutdef, estadd, estimates, regression table, latex,
html

1 Introduction
Statistical packages are usually very good at fitting all kinds of regression models, but
they are rather poor at keeping the results for those models organized or processing
them for publication. This is a real problem because gathering the relevant figures by
hand from the large amount of statistical output that is usually produced and arrang-
ing the results in clear and presentable tables can be very inefficient and error-prone
processes. Furthermore, results must often be processed repeatedly, for example, when
operationalizations are modified or mistakes are detected. In order to reduce transcrip-
tion errors and avoid having to repeat the laborious tasks by hand, it makes sense to
automate the processing of results as much as possible.
Fortunately, Stata provides the basis for such an automation. One of the great
features in Stata is that, after an estimation command has been carried out, all the
relevant results are not only displayed onscreen but are returned in places where they can
be accessed by the user. This storage of results provides the user with the opportunity
to further process the results in a more-or-less automated manner. Furthermore, Stata 8
saw the introduction of the estimates command (see [R] estimates), which facilitates
the handling of the estimation results for multiple models. More specifically, results from
up to 20 models can be stored at a time. Stata also provides a utility for compiling a
table of the coefficients for all stored models called estimates table. Although the
estimates table command is rather limited and cannot be used to translate the table

c 2005 StataCorp LP st0085
B. Jann 289

to spreadsheet formats or LATEX code, it does a good job assembling a raw matrix of
models and parameters that can be used as a starting point for creating a complex and
well-formatted regression table.
In the remainder of this paper, I will present the new estout package, a program
that makes use of the possibilities provided by Stata and produces regression tables
in what I believe is a very flexible and functional way. Note that there also are other
user programs available to produce tables from regression results. John Luke Gallup’s
outreg is probably the most widely used package of this kind (Gallup 1998, 1999, 2000,
2001). Among the other packages are outtex by Antoine Terracol, est2tex by Marc
Muendler, and mktab by Nicholas Winter. Also see Newson (2003) for a very appealing
approach. However, estout represents a good compromise between functionality and
usability.

2 Description and basic examples

estout assembles a table of coefficients, “significance stars”, summary statistics, stan-
dard errors, t or z statistics, p-values, confidence intervals, and other statistics calculated
for up to twenty models previously fitted and stored by estimates store. It then writes
the table to the Stata log or to a specified text file.
The full syntax of estout is rather complex and is therefore to be found in the
appendix in section 4.1 (also see estout’s online help). However, consider the following
basic syntax, which includes only the most important options:




estout namelist
using filename , cells(array) stats(scalarlist)

style(style) more options

where namelist is a list of the names of stored estimates (the namelist can be entered
as * to refer to all stored estimates). The cells() and stats() options determine the
primary contents of the table. The style() option determines the basic formatting of
the table.

Basic usage

The procedure for using estout is to first store several models using the estimates
store command and then apply estout to save or display a table of the estimates. By
default, estout produces a plain, tab-separated table of the coefficients of the models
indicated by the command:

. sysuse auto
(1978 Automobile Data)
. replace price = price/1000
price was int now float
(74 real changes made)
290 Making regression tables

. replace weight = weight/1000

weight was int now float
(74 real changes made)
. regress price weight mpg
(output omitted )
. estimates store m1, title(Model 1)
. generate forXmpg = foreign * mpg
. regress price weight mpg forXmpg foreign
(output omitted )
. estimates store m2, title(Model 2)
. estout * using example.txt
m1 m2
b b
weight 1.746559 4.613589
mpg -.0495122 .2631875
forXmpg -.3072165
foreign 11.24033
_cons 1.946068 -14.44958

The table produced by the estout command looks messy in the Stata Results window
or the Stata log because the columns are tab-separated (note that tab characters are not
preserved in the Results window or the log). However, the stored example.txt would
look better if it were opened, for example, in a spreadsheet program.

Choosing a style

To align the columns in Stata’s Results window, fixed widths can be specified for the
columns and tab characters can be removed. This is most easily done via the style()
option, which provides a style called fixed:
. estout *, style(fixed)
m1 m2
b b
weight 1.746559 4.613589
mpg -.0495122 .2631875
forXmpg -.3072165
foreign 11.24033
_cons 1.946068 -14.44958

Other predefined styles are tab (the default), tex, and html, but it is also possible to
define one’s own styles (see appendix 4.3). The tex style, for example, modifies the
output table for use with LATEX’s tabular environment:
. estout *, style(tex) varlabels(_cons \_cons)
& m1& m2\\
& b& b\\
weight & 1.746559& 4.613589\\
mpg & -.0495122& .2631875\\
forXmpg & & -.3072165\\
foreign & & 11.24033\\
\_cons & 1.946068& -14.44958\\
B. Jann 291

Note that cons has been replaced by its LATEX equivalent in the example above using
the varlabels() option (since the underscore character produces an error in LATEX
unless it is preceded by a backslash). For more information on the varlabels() option,
consult estout’s online help.

The cells option

Use the cells() option to specify the parameter statistics to be tabulated and how they
are to be arranged. The parameter statistics available are b (coefficients, the default), se
(standard errors), t (t/z statistics), p (p-values), ci (confidence intervals; to display the
lower and upper bounds in separate cells, use ci l and ci u), as well as any additional
parameter statistics included in the e()-returns for the models (also see section 3.7).
For example, cells(b se) reports raw coefficients and standard errors:

. estout *, cells(b se) style(fixed)

m1 m2
b/se b/se
weight 1.746559 4.613589
.6413538 .7254961
mpg -.0495122 .2631875
.086156 .1107961
forXmpg -.3072165
.1085307
foreign 11.24033
2.751681
_cons 1.946068 -14.44958
3.59705 4.42572

Multiple statistics are placed in separate rows beneath one another by default, as in
the example above. However, elements that are listed in quotes are placed beside one
another. For example, specifying cells("b se t p") produces the following table:

. estout m2, cells("b se t p") style(fixed)

m2
b se t p
weight 4.613589 .7254961 6.359219 1.89e-08
mpg .2631875 .1107961 2.375421 .0203122
forXmpg -.3072165 .1085307 -2.830687 .0060799
foreign 11.24033 2.751681 4.084896 .0001171
_cons -14.44958 4.42572 -3.26491 .0017061

The two approaches can be combined. For example, cells("b p" se) would pro-
duce a table with raw coefficients and standard errors beneath one another in the first
column and p-values in the top row of the second column for each model.
Note that for each statistic named in the cells() option, a set of suboptions may
be specified in parentheses. For example, in social sciences, it is common to report
standard errors or t statistics in parentheses beneath the coefficients and to indicate the
significance of individual coefficients with stars. Furthermore, the results are rounded.
Such a table can be created using the following procedure:
292 Making regression tables

. estout *, cells(b(star fmt(%9.3f)) t(par fmt(%9.2f))) style(fixed)

m1 m2
b/t b/t
weight 1.747** 4.614***
(2.72) (6.36)
mpg -0.050 0.263*
(-0.57) (2.38)
forXmpg -0.307**
(-2.83)
foreign 11.240***
(4.08)
_cons 1.946 -14.450**
(0.54) (-3.26)

The estout default is to display * for p < .05, ** for p < .01, and *** for p < .001.
However, note that the significance thresholds and symbols are fully customizable (see
the starlevels option in appendix 4.1).

The stats option

Finally, use the stats() option to specify scalar statistics to be displayed in the last
rows of each model’s table. The available scalar statistics are aic (Akaike’s information
criterion), bic (Schwarz’s information criterion), rank (the rank of e(V), i.e., the num-
ber of free parameters in the model), p (the p-value of the model), as well as any scalar
contained in the e()-returns for the models (also see section 3.7). For example, specify
stats(r2 bic N) to add the R-squared, BIC, and the number of cases to the bottom
of the table:

. estout *, stats(r2 bic N) style(fixed)

m1 m2
b b
weight 1.746559 4.613589
mpg -.0495122 .2631875
forXmpg -.3072165
foreign 11.24033
_cons 1.946068 -14.44958
r2 .2933891 .5516277
bic 356.2918 331.2406
N 74 74

3 Advanced applications
The estout package has many features, and it is beyond the scope of this text to provide
examples for all of these options. The following presentation is therefore restricted to a
few selected examples illustrating the spectrum of estout’s capabilities and introducing
some of its less-obvious applications.
B. Jann 293

3.1 Using labels

The labels option will cause estout to use variable labels and model labels, if they are
available. Furthermore, there are options for specifying custom labels for the different
table elements, displaying a legend explaining the significance symbols and thresholds,
and inserting lines of text at various places in the table. The following example is
intended to provide a first impression of these possibilities:
. label variable foreign "Foreign car type"
. label variable forXmpg "Foreign*Mileage"
. estout m1 m2, cells("b(star label(Coef.)) se(label(Std. err.))")
> stats(r2 N, labels(R-squared "N. of cases")) label legend
> varlabels(_cons Constant) posthead("") prefoot("") postfoot("")
> varwidth(16) style(fixed)
Model 1 Model 2
Coef. Std. err. Coef. Std. err.
Weight (lbs.) 1.746559** .6413538 4.613589*** .7254961
Mileage (mpg) -.0495122 .086156 .2631875* .1107961
Foreign*Mileage -.3072165** .1085307
Foreign car type 11.24033*** 2.751681
Constant 1.946068 3.59705 -14.44958** 4.42572
R-squared .2933891 .5516277
N. of cases 74 74
* p<0.05, ** p<0.01, *** p<0.001

3.2 LATEX tables

The highest degree of automation can probably be attained by using estout in combi-
nation with LATEX.

Table 1: The auto data

Model 1 Model 2
Coef. p-value Coef. p-value
Weight (lbs.) 1.747 .008 4.614 .000
Mileage (mpg) –.050 .567 .263 .020
Foreign*Mileage –.307 .006
Foreign car type 11.240 .000
Constant 1.946 .590 –14.450 .002
Adj. R2 .273 .526
No. of cases 74 74
Source: auto.dta

Table 1 was produced by inserting the line

\input{auto.tex}

in the LATEX document for this article after having run the following command:
294 Making regression tables

. estout m1 m2 using auto.tex,

> cells("b(label(Coef.) fmt(%9.3f)) p(label(\$p\$-value))")
> stats(r2_a N, fmt(%9.3f %9.0f) labels("Adj. \$R^2\$" "No. of cases"))
> label msign(--) nolz varwidth(16) modelwidth(13) style(tex)
> title(The auto data\label{auto}) varlabels(_cons Constant)
> mlabels(, span prefix(\multicolumn{@span}{c}{) suffix(}))
> prehead("\begin{table}\caption{@title}" "\begin{center}"
> "\begin{tabular}{l*{@M}{rr}}" "\hline") posthead(\hline)
> prefoot(\hline) postfoot("\hline" "\small\textit{Source:} auto.dta"
> "\end{tabular}" "\end{center}" "\end{table}")
\begin{table}\caption{The auto data\label{auto}}
\begin{center}
\begin{tabular}{l*{2}{rr}}
\hline
&\multicolumn{2}{c}{Model 1}&\multicolumn{2}{c}{Model 2}\\
& Coef.& $p$-value& Coef.& $p$-value\\
\hline
Weight (lbs.) & 1.747& .008& 4.614& .000\\
Mileage (mpg) & --.050& .567& .263& .020\\
Foreign*Mileage & & & --.307& .006\\
Foreign car type& & & 11.240& .000\\
Constant & 1.946& .590& --14.450& .002\\
\hline
Adj. $R^2$ & .273& & .526& \\
No. of cases & 74& & 74& \\
\hline
\small\textit{Source:} auto.dta
\end{tabular}
\end{center}
\end{table}

Note that most of the options in the above command could also have been provided
via defaults files (see appendix 4.3). Working with defaults files can be very efficient if
you want to produce a large number of similar tables.

3.3 Selective information

estout has a keep() and a drop() option to select the parameters (or equations) to
be tabulated (an example can be found in section 3.6), as does estimates table (see
[R] estimates). However, a useful additional feature of estout is that the information
displayed can be varied by regressors. Sometimes certain statistics are of interest only
for some parameters and not for others. Those statistics can therefore be suppressed for
individual parameters, using the keep() or the drop() suboption within the cells()
option to save space:

. estout *, cells(b(star) t(par keep(mpg))) style(fixed)

m1 m2
b/t b/t
weight 1.746559** 4.613589***
mpg -.0495122 .2631875*
(-.5746806) (2.375421)
forXmpg -.3072165**
foreign 11.24033***
_cons 1.946068 -14.44958**
B. Jann 295

Furthermore, the parameter statistics reported for the various models can be specified
using the pattern() suboption within the cells() option (for example, it is possible
to print the t statistics for, say, the second model only; an example can be found in
section 3.6).

3.4 Summary statistics only

estout can also be used to produce a table displaying only summary statistics:

. estout *, cells(none) stats(r2_a bic N, star) style(fixed)

m1 m2
r2_a .2734846*** .5256351***
bic 356.2918 331.2406
N 74 74

Note that in the example the models’ overall significance is denoted by stars attached
to values of the adjusted R-squared (both models are significant at the 0.001 level).

3.5 Multiple-equation models

The default in estout is to arrange the different equations of multiple-equation models
in vertical order. However, for models like mlogit or sureg, it is sometimes convenient
to arrange the equations horizontally, which can be achieved through the use of the
unstack option:

. sureg (price foreign weight length) (mpg displ = foreign weight)

(output omitted )
. estimates store m4
. estout m4, cells(b t(par)) unstack stats(r2 chi2 p) style(fixed)
m4
price mpg displacement
b/t b/t b/t
foreign 3.57526 -1.650029 -25.6127
(5.749891) (-1.565555) (-2.047999)
weight 5.691462 -6.587886 96.75485
(6.182983) (-10.55641) (13.06594)
length -.0882711
(-2.809689)
_cons 4.506212 41.6797 -87.23547
(1.255897) (19.64914) (-3.46585)
r2 .548808 .6627029 .8115213
chi2 89.73586 145.3912 318.6174
p 2.50e-19 2.68e-32 6.50e-70

In the case of the multiple-equation models reg3, sureg, and mvreg, summary statistics
for all the model’s equations will be printed in separate columns in the same row. For
all other models, the summary statistics will be placed in the first column.
296 Making regression tables

3.6 Marginal effects

estout supports Stata’s mfx command for calculating marginal effects or elasticities
(see [R] mfx). In order to report the mfx results in estout, use the margin option.
However, you must have saved the model after applying mfx, as illustrated by the
following example. Note that the last column of the table in the example below displays
the points around which the marginal effects were estimated (mfx returns these values
in e(Xmfx X)).

. generate record = 0
. replace record = 1 if rep > 3
(34 real changes made)
. logit foreign mpg record
(output omitted )
. estimates store raw
. mfx
(output omitted )
. estimates store mfx
. estout raw mfx, cells("b Xmfx_X(pattern(0 1))" se(par)) margin legend
> style(fixed)
raw mfx
b/se b/se Xmfx_X
mpg .1079219 .0184528 21.2973
(.0565077) (.0101674)
record (d) 2.435068 .4271707 .4594595
(.7128444) (.1043178)
_cons -4.689347
(1.326547)
(d) marginals for discrete change of dummy variable from 0 to 1

With single-equation models, the incorporation of results from mfx in the table
is straightforward. However, matters become more complicated for multiple-equation
models. Marginal effects have nothing to do with the equations per se, so it is not clear
where to report the mfx results if some variables appear in several different equations.
The default in estout is to print the mfx coefficients in each row that relate to the
variable in question. This default can be changed with the meqs() option, which specifies
that the mfx results be printed only in select equations. For example, proceed as follows
to report the marginal effects for the probability of only the main outcome in heckprob:
. set seed 6630
. generate u = uniform() > 0.5
. heckprob u headroom, select(foreign = turn headroom) nolog
(output omitted )
. estimates store raw
. mfx
(output omitted )
. estimates store mfx
B. Jann 297

. estout raw mfx, cells(b se(par)) margin meqs(u) keep(u: foreign:)

> style(fixed)
raw mfx
b/se b/se
u
headroom -1.003445 -.2843565
(.6077779) (.2326952)
_cons 2.176479
(1.923797)
foreign
turn -.2954961
(.0675027)
headroom -.1261772
(.2919013)
_cons 11.05306
(2.479492)

Taking the additional step of inserting the marginal effects for the selection proba-
bility in the example above is rather involved because the marginal effects for the two
functions must be saved in different models. The solution is to print only the main
equation in a first estout call and then append the rest of the table in a second call:

. mfx, predict(psel)
(output omitted )
. estimates store mfx2
. tempfile foo
. estout raw mfx using "‘foo’", cells(b se(par)) margin keep(u:)
> style(fixed) notype
. estout raw mfx2 using "‘foo’", cells(b se(par)) margin
> keep(foreign:) mlabels(, none) collabels(, none)
> style(fixed) notype append
. type "‘foo’"
raw mfx
b/se b/se
u
headroom -1.003445 -.2843565
(.6077779) (.2326952)
_cons 2.176479
(1.923797)
foreign
turn -.2954961 -.068597
(.0675027) (.0158482)
headroom -.1261772 -.029291
(.2919013) (.0665186)
_cons 11.05306
(2.479492)

3.7 Adding supplementary statistics

Results that are included in the e()-returns for the models can be tabulated by estout.
Thus one approach for, for example, reporting certain transformations of the coefficients
is to add a matrix of the transformed results to the e()-returns and then tabulate the
results using estout. The estadd command is designed to support this approach. It
298 Making regression tables

may be used, for example, to add standardized coefficients or the means and standard
deviations of the regressors to the e()-returns for the stored models. However, estadd’s
basic capabilities can be extended by writing subroutines to allow for additional statis-
tics.
The basic syntax of estadd is


estadd namelist , stats(statslist)

where namelist is again a list of stored estimates (if namelist is empty, estadd will be
applied to the current estimates). Use stats() to specify the statistics to be added to
the e()-returns of the indicated models. For more details, see estadd’s online help.

Table of descriptives

estadd is equipped with a few predefined statistics, such as beta (standardized coef-
ficients), mean (means of regressors), and sd (standard deviations of regressors)1 . The
latter can be used, for example, to produce a table of descriptives for the variables in
the models in our examples:

. quietly generate x = uniform()

. quietly regress x price weight mpg foreign
. estadd, stats(mean sd(nobinary))
. estimates store m3
. estout m3, cells("mean sd") stats(N) mlabels(,none) drop(_cons) style(fixed)
mean sd
price 6.165257 2.949496
weight 3.019459 .7771936
mpg 21.2973 5.785503
foreign .2972973
N 74

Adding user-defined statistics

Writing new estadd subroutines to add user-defined statistics is not overly complicated,
as we will illustrate below. In general, a new subroutine should be called estadd mystat.
mystat will be available to the stats() option of the estadd command after the program
code has been executed or the subroutine file has been saved as estadd mystat.ado
in either the current directory or somewhere else in the ado path ([P] sysdir). The
subroutine will be called once for each model with the model’s estimates restored. The
e()-returns for the model in question may be therefore used to calculate new statistics.
Within a subroutine, use the ereturn command ([P] ereturn) to append new statis-
tics to the existing e()-returns. New summary statistics should be returned as scalars
using the ereturn scalar command, whereas new parameter statistics (e.g., transfor-
mations of the regression coefficients) should be returned as matrices (row vectors, to
1 More functions are provided by the estadd plus package (available from the SSC archive).
B. Jann 299

be precise) using the ereturn matrix command. Note that the columns of the added
matrices should be named according to the row names of the coefficients matrix e(b) in
order to ensure estout’s ability to tabulate the new parameter statistics. Use the ex-
amples below or the estadd beta, estadd mean, and estadd sd subroutines, which are
supplied with the estadd package, as a starting point for programming new routines.
To report the Cox and Snell pseudo-R-squared, for example, define the estadd-
subroutine
program estadd_coxsnell, eclass
ereturn scalar coxsnell = 1 - exp(e(ll_0)-e(ll))^(2/e(N))
end

and then type2

. logit foreign price weight

(output omitted )
. estimates store m5
. logit foreign price weight mpg
(output omitted )
. estimates store m6
. estadd m5 m6, stats(coxsnell)
. estout m5 m6, stats(coxsnell) style(fixed)
m5 m6
b b
price .9295969 .9263907
weight -5.878539 -6.849737
mpg -.1210918
_cons 9.000472 14.42237
coxsnell .518701 .5291797

New parameter statistics can be added in a similar manner. For example, the fol-
lowing lines of code comprise a subroutine to insert the standardized factor change
coefficients, or exp(βj Sj ), where Sj is the standard deviation of regressor j, which are
sometimes reported for logistic regression (see Long 1997):

2 Also see the eret2 package (available from the SSC Archive). The eret2 command allows you to

add statistics to the e()-returns of a model without having to program subroutines. However, eret2
can be applied only to the currently active estimates.
300 Making regression tables

program estadd_ebsd, eclass

if "‘e(cmd)’" != "logit" | "‘e(wexp)’" != "" exit
tempname results
matrix ‘results’ = e(b)
local vars: colnames ‘results’
local j 0
foreach var of local vars {
local ++j
capture confirm variable ‘var’
if _rc matrix ‘results’[1,‘j’] = .z
else {
quietly summarize ‘var’ if e(sample)
matrix ‘results’[1,‘j’] = exp( ‘results’[1,‘j’] * r(sd) )
}
}
ereturn matrix ebsd = ‘results’
end

If the program is saved in the ado path as estadd ebsd.ado, it can, for example, be
called as follows:

. estadd m5, stats(ebsd sd)

. estout m5, eform drop(_cons)
> cells("b(label(e^b)) ebsd(label(e^(b*sdx))) sd(label(sdx))") style(fixed)
m5
e^b e^(b*sdx) sdx
price 2.533488 15.51554 2.949496
weight .0027989 .0103708 .7771936

4 Appendix
4.1 Full syntax of estout




estout namelist using filename , parameter statistics options
summary statistics option significance stars options layout options

labeling options output options defaults option


where namelist is either all or * or name name ... , and name is the name of
stored estimates. The results estimated last may be indicated by a period (.), even if
they have not yet been stored. For a detailed discussion of estout’s options, see the
online help. A brief list of the options is provided below.

The parameter statistics options are

cells(array | none) specify the contents of the table cells (co-

efficients, standard errors, etc.)
drop(droplist) drop individual parameters or equations
keep(keeplist) keep individual parameters or equations
B. Jann 301

equations(eqmatchlist) match the models’ equations


eform (pattern) | noeform display the results in exponentiated form


margin (u | c | p) | nomargin report marginal effects or elasticities
discrete(string) | nodiscrete identify dummy variables when reporting
marginal effects
meqs(eq list) select equations for marginal effects
level(#) set the level for confidence intervals

where array is


row row ...

row is


 



‘ " element element ... " ’

element is


el (el subopts)

el is one of the following statistics

b raw coefficients
se standard errors
t t statistics
p p-values
ci confidence intervals
ci l lower bounds of confidence intervals
ci u upper bounds of confidence intervals
myel additional statistics included in e()

and the el subopts are


no star attach “significance stars”


fmt(%fmt %fmt ... ) set the display formats
label(string) define a label for el


par (leftstring rightstring)
| nopar place el in parentheses
drop(droplist) drop certain individual statistics
keep(keeplist) keep certain individual statistics
pattern(pattern) report el for selected models only


no abs use absolute t statistics
302 Making regression tables

The summary statistics option is


stats(scalarlist , stats subopts ) specify scalar statistics to be displayed at
the bottom of the table

where the stats subopts are


fmt(%fmt %fmt ... ) set the display formats


labels(stringlist , label subopts ) label the scalar statistics


star (scalarlist) | nostar denote overall model significance

The significance stars options are

starlevels(levelslist) define thresholds and symbols for “sig-

nificance stars”


no stardetach display the stars in their own column

where levelslist is


symbol # symbol # ...

with # ∈ (0, 1] and listed in descending order.

The layout options are

varwidth(#) set the width of the table’s left stub

modelwidth(#) set the width of the results columns


no abbrev abbreviate long names and labels


no unstack place individual equations from multiple-
equation models in separate columns
begin(string) specify the beginning of the table rows
delimiter(string) specify the column delimiter
end(string) specify the ending of the table rows
dmarker(string) determine the decimal marker
msign(string) determine the minus sign


no lz print the leading zero of fixed format
numbers in (−1, 1)
substitute(subst list) apply end-of-pipe substitutions

where subst list is


from to from to ...
B. Jann 303

The labeling options are

title(string) specify a title for the table


no legend add a legend explaining the significance
symbols
prehead(stringlist) add text lines before the table heading
posthead(stringlist) add text lines after the table heading
prefoot(stringlist) add text lines before the table footer
postfoot(stringlist) add text lines after the table footer


no label use variable labels


varlabels(matchlist , varl subopts ) relabel the parameters


mlabels(stringlist , mlabels subopts ) label the models


collabels(stringlist , label subopts ) label the columns within models


eqlabels(stringlist , label subopts ) label the equations


mgroups(stringlist , mgroups subopts ) define and label groups of models

where the varl subopts are

blist(matchlist) assign specific prefixes to certain rows

elist(matchlist) assign specific suffixes to certain rows
label subopts

the mlabels subopts are


no numbers number the models


no depvars use dependent variables as models’ labels
label subopts

the mgroups subopts are

pattern(pattern) establish the grouping of the models

label subopts

and the label subopts are


no none suppress the labels
prefix(string) add a common prefix
suffix(string) add a common suffix
begin(string) add an overall prefix
304 Making regression tables

end(string) add an overall suffix


no last print the last occurrence of end


no span span columns if appropriate
erepeat(string) add a “span” suffix
lhs(string) insert string into the left stub of the table

The output options are


no replace overwrite an existing file


no append append the output to an existing file


no type print the table in the Results window


no showtabs display tabs as <T>s

The defaults option is

style(style) specify a style for the output table

where style is one of the following

tab tab delimited table (the default)
fixed fixed format table
tex table for use with LATEX
html table for use with HTML
mystyle user-defined addition

4.2 Using @-variables

estout features several variables that can be used within string specifications. The
following list provides an overview of these variables (also see the example in section 3.2):

@span Returns the value of a count variable for the total number of physical
columns of the table if used in the labels in the blist() and elist()
suboptions of varlabels() or in the text specified in prehead(),
posthead(), prefoot(), or postfoot().

@span Returns the number of spanned columns if used in the text specified
in the prefix() and suffix() suboptions of mgroups(), mlabels(),
eqlabels(), or collabels(), or in the labels specified in these options.

@span Returns the range of spanned columns (e.g., 2-4 if columns 2, 3, and 4
are spanned) if used in the text specified in the erepeat() suboption
of mgroups(), mlabels(), eqlabels(), or collabels().
B. Jann 305

@M Returns the number of models in the table if used in the text specified
in prehead(), posthead(), prefoot(), or postfoot().

@title Returns the title specified with the title() option if used in the text
specified in prehead(), posthead(), prefoot(), or postfoot().

@discrete Returns the explanations provided by the discrete() option (if the
margin option is activated) if used in the text specified in prehead(),
posthead(), prefoot(), or postfoot().

@starlegend Returns a legend explaining the significance symbols if used in the text
specified in prehead(), posthead(), prefoot(), or postfoot().

4.3 Working with defaults files

estout’s style() option may be used to specify a style for the output table. A style is a
named combination of options that is saved in an auxiliary file called estout style.def.
estout is already equipped with four such files. The four styles and their particulars
are

settings styles
tab fixed tex html
begin <tr><td>
delimiter tab " " & </td><td>
end \\ </td></tr>
varwidth 0 12 12 12
modelwidth 0 12 12 12
abbrev off on off off

It is very easy to generate your own set of default options. Type

. estoutdef style, edit

to open one of the existing defaults files (where style is the name of the defaults set,
e.g., tab; the estoutdef command is provided with the estout package), make the
desired modifications, and save the file as estout newstyle.def in the current directory
or elsewhere in the ado path (see [P] sysdir). To use the new option settings, type

. estout ..., style(newstyle)

estout has two main types of options, which are treated differentially in defaults files.
On the one hand, there are simple on/off options without arguments, such as legend
or showtabs. To turn such an option on, enter the option followed by the options name
as an argument; i.e., add the line
option option
to the defaults file. For example,
306 Making regression tables

legend legend
specifies that a legend be printed in the table footer. Otherwise, if you want to turn
the option off, just delete or comment out the line that contains it (or specify option
without an argument).
To temporarily turn off an option that has been activated in a defaults file, specify
nooption in the command line (do not, however, use nooption in defaults files). For
example, if the legend has been turned on in the defaults file, but you want to suppress
it in a specific call of estout, type
. estout ..., nolegend

On the other hand, options that take arguments, such as prehead(args),

delimiter(args), or stats(args, ...) are specified as
option args
in the defaults file (where args must not include suboptions; see below). Specifying an
option in the command line overwrites the settings from the defaults file. However, note
that a no form, which exists for the first options type, is not available here.
Last but not least,

there are two options

that reflect a combination of the first and
second types: eform (args) and margin (args) . These options can be specified as
either
option option
or
option args
in the defaults file; the no form is allowed.
Many estout options have suboptions; i.e., an option might take the form op-
tion(..., suboption) or option(..., suboption(args)). In the defaults file, the sub-
options cannot be included in the definition of a higher-level option. Instead, they must
be specified in their own lines as either
optionsuboption suboption
or
optionsuboption args
In the case of a two-level nesting of options, the name used to refer to the suboption
is a concatenation of the option’s name and the suboption’s name, i.e., "optionsub-
option"="option"+"suboption". For example, the labels() suboption of the stats()
option would be set by the term statslabels. Analogously, the three level nesting in
the stats() option yields suboption names composed of three names. For instance, the
suboption called by the command
. estout ..., stats(..., labels(..., prefix( args)))
B. Jann 307

would be referred to as
statslabelsprefix args
in the defaults file. The cells() option represents an exception to this rule. It may be
defined in the defaults file using only a simple array of cells elements without suboptions,
e.g.
cells "b se" p
However, the suboptions of the cells elements may be referred to as el suboption, for
example
b star star
or
se par [ ]
Be aware that the support for comments in defaults files is limited. In particular,
the /* and */ comment indicators cannot be used. The other comment indicators work
(more or less) as usual; that is,

• Empty lines and lines beginning with * (with or without preceding blanks) will
be ignored.

• // preceded by one or more blanks indicates that the rest of the line should
be ignored. Lines beginning with // (with or without preceding blanks) will be
ignored.

• /// preceded by one or more blanks indicates that the rest of the line should be
ignored and the part of the line preceding it should be added to the next line. In
other words, /// can be used to split commands into two or more lines of code.

5 Acknowledgments
Some of the estout code has been adapted from the official est table.ado. I would
like to thank Kit Baum, Elisabeth Coutts, Henriette Engelhardt, Jonathan Gardnerand,
Friedrich Huebler, Maren Kandulla, Clive Nicholas, Fredrik Wallenberg, Ian Watson,
and Vince Wiggins for their comments and suggestions.

6 References
Gallup, J. L. 1998. sg97: Formatting regression output for published tables. Stata
Technical Bulletin 46: 28–30. In Stata Technical Bulletin Reprints, vol. 8, 200–202.
College Station, TX: Stata Press.

—. 1999. sg97.1: Revision of outreg. Stata Technical Bulletin 49: 23. In Stata Technical
Bulletin Reprints, vol. 9, 170–171. College Station, TX: Stata Press.
308 Making regression tables

—. 2000. sg97.2: Update to formatting regression output. Stata Technical Bulletin 58:
9–13. In Stata Technical Bulletin Reprints, vol. 10, 137–143. College Station, TX:
Stata Press.

—. 2001. sg97.3: Update to formatting regression output. Stata Technical Bulletin 59:
23. In Stata Technical Bulletin Reprints, vol. 10, 143. College Station, TX: Stata
Press.

Long, J. S. 1997. Regression Models for Categorical and Limited Dependent Variables.
Thousand Oaks, CA: Sage.

Newson, R. 2003. Confidence intervals and p-values for delivery to the end user. Stata
Journal 3(3): 245–269.

About the Author

Ben Jann ([email protected]) is research assistant at the Department of Sociology of the
Swiss Federal Institute of Technology Zurich (ETH) and a Ph.D. candidate at the University
of Bern in Switzerland.