apsrtable: APSR-style latex tables with multiple models

In apsrtable: apsrtable model-output formatter for social science

Description

Produce well-formatted LaTeX tables of multiple models side-by-side.

*** Requires \usepackage{dcolumn} in Latex/Sweave preamble. ***

Usage

 apsrtable (..., 
                   se=c("robust","vcov","both","pval"),
                   model.names=NULL, model.counter=1, digits=2,
                   stars=1,lev=.05,
                   align=c("left","center","right"),
                   order=c("lr","rl","longest"),
                   notes=list(se.note, stars.note),
                   omitcoef=NULL,coef.names=NULL,
                   coef.rows=2,
                   multicolumn.align=c("center","left","right"),
                   col.hspace=NULL,
                   Sweave=FALSE, float="table",
                   Minionfig=FALSE,
                   label=NULL,caption=NULL,
                   caption.position=c("above","below"))

Arguments

...	One or more fitted model objects of a supported class such as lm or glm. The model-object (a list) may also optionally contain an item named se: model$se may be a vector of standard errors, or a variance-covariance matrix, in which case the square root of the diagonal is used as the “robust” standard errors in the output. See the se argument.
se	A character string controlling the quantities in parentheses, can take the following values: “robust” Print (and calculate significance using) user-supplied errors in an $se element appended to the model object. “vcov” Use the standard errors contained in the second column of coef(summary(model)). This behavior may be useful in writing apsrtableSummary methods to customize output or format new model classes. “both” User-supplied errors are printed in (parentheses) and the default are printed in [square brackets.] “pval” Prints the p value contained in the fourth column of coef(summary(model)). Quietly switches se.note to say “(Robust) p values in parentheses.” Robust values are used (and so labeled contextually) whenever an $se element is present in any model. If any model in ... contains an se element and “robust” is chosen (the default), output is labeled as “robust;” if no models have an se element (all use model vcov) but se="robust", labeling is simply “Standard errors in parentheses.” Default = "robust"
model.names	Optional vector of names to use as column headings in the table. If more models than names are supplied, unnamed models are numbered (starting at one more than the number of names).
model.counter	Change the number to start counting from when using automatically numbered models. Default = 1.
digits	Number of decimal places to report. Default = 2
stars	Show statistical significance “stars”, either “1” or “default” where “default” is based on the output of summary.lm, except that a superscript dagger is used instead of a dot for p < .10. Here “default” means “the R default”, not to be confused with the function's (perhaps confusing) Default=1
lev	When stars=1, what level should be used for the test to reject statistical insignificance and bestow the glittering star? Disable decoration entirely by specifying lev=0. Default=.05.
align	How should columns be aligned in the output? Model summaries are always decimal-aligned using dcolumn (and therefore also set in math mode), but dcolumn also provides for decimal-point centering. Model names are set in \multicolumn spans with alignment given here, as are model terms (leftmost column of table). Default = “left”.
order	Determines the order in which terms (rows) are included in the output when more than one model (column) is present. “lr” and “rl” take the order of terms from the first or last (leftmost or rightmost) model and appends new terms as they are encountered. “longest” uses the order of terms in the model with the most terms.Default = “lr”.
notes	A list to be evaluated and placed, one item per full-width (multicolumn) line, in footnote size. The default uses two functions, se.note and stars.note to generate notes about the standard errors and indicators of statistical significance. Other notes can be named function calls or simple character strings.
omitcoef	An optional integer or character vector of coefficient indices, or an expression involving coefnames that evaluates to integer or character, of rows to exclude from the output. See details.
coef.names	An optional vector of names for coefficients. It is recommended to establish the omitcoef and order settings with automatic symbolic naming before supplying a vector of “pretty” variable names. If automatic symbolic naming is used, names are taken from the variables in the models and “sanitized” for latex. If coef.names are supplied, they must be valid latex, with double-backslash escape characters.
coef.rows	The number of rows in the table given to each coefficient: by default each coefficient's standard error is printed in a row beneath it, but setting coef.rows to 1 places it in a new column to the right instead.
multicolumn.align	Alignment for the table's multicolumn spans: typically only the model names at the top, but, in the case of coef.rows=1, the model.info is also aligned beneath both columns. Default=“center”
col.hspace	Optional hspace (number+tex units such as em) to insert between each model column(s). Intended mainly to separate models from each other when coef.rows=1. Default=NULL
Sweave	Toggle whether to include \begin{table}...\end{table}, label, and caption, or only the \begin{tabular} ... \end{tabular}. When called from within an Sweave document one would typically write such elements in the “documentation” (latex-part) rather than inside the code chunk. When called from an Sweave document, make sure to set the code chunk option results=tex. Default = FALSE
float	if Sweave is false – that is, if apsrtable is supposed to wrap the output in the float environment, float allows you to specify an arbitrary custom float environment. Some useful ones include “sidewaystable” (latex package rotating), or “longtable”. In the special case of “longtable”, the header row of model names is included on all pages, and the label and caption arguments are included even when Sweave=TRUE because of the structure of the latex environment.
Minionfig	Include latex command to change the figure style to “tabular” and back to “proportional”, specifically for the MinionPro latex package. Default = FALSE
label	A string to be used as the label in latex for the table. Used only when Sweave=FALSE, unless float="longtable".
caption	A latex string to be used as the caption for the table (remember to use double backslashes for latex commands in R). Used only when Sweave=FALSE, unless float="longtable".
caption.position	Determines the placement of the caption and label with respect to the tabular environment. (Not thoroughly tested with alternative float environments, but should work with the standard table/tabular combination.) Default="above"

Details

Given one or several model objects of various types, apsrtable() will produce side-by-side output in well-formatted LaTeX using either automated numbering or user-supplied model names and dcolumn decimal-aligned columns. Terms are matched across rows, with options for determining the order of terms. Nuisance terms (e.g. controls, or other quantities not of primary interest) may be omitted. Standard errors of parameter estimates are placed below estimates and in parentheses, with the option for the user to supply a replacement vector of standard errors or a replacement variance-covariance matrix, such as one estimated using the sandwich package. By default a single star denotes statistical significance at the .05 level, with the option to employ further decorations or specify another arbitrary level for the test. Finally, some model diagnostics are included along with a (somewhat) flexible means to program or include different items depending on model object class.

The argument omitcoef suppresses the output of specific rows. It may be either a valid subscript index (integer or logical if opacity is desired, or character for transparency), or an expression, such as a grep expression to be evaluated with respect to coefnames (without a dot). The internal objectcoefnames is the union of all model terms, in the desired order. In the example below, “(Intercept)” is excluded by a regular expression matching the parenthesis.

To exclude multiple regular expressions, or a mix of expressions with other types, you may supply a list, but you must ensure that the result is a valid subscript list: all character, all numeric, or all logical. For example, if you refer to a specific coefficient by its character name, include the argument value=TRUE in any grep expressions in the list.

Model diagnostic information (“model info”) is handled by formal modelInfo methods defined for model summaries. These methods return lists of S3 class model.info, named formatted (character) elements. To include fit (or other) information that is available from fitted model objects but not their summaries, write an apsrtableSummary method to prepare a summary with the items needed for your own modelInfo method.

Included are modelInfo functions for lm, glm, and tobit, coxph, clogit, and a skeleton (incomplete modelInfo) for gee and svyglm objects. Please email the author any modelInfo functions you write for different model objects for inclusion in future releases.

Value

A character vector containing lines of latex code. It can be written out using writeLines for inclusion via \input{} in latex documents.

Author(s)

Michael Malecki <malecki at wustl.edu>

See Also

modelInfo for changing the model diagnostic summary information presented and how to include it for different classes of model objects; notefunctions for functions to produce dynamic “notes” beneath tables; and apsrtableSummary for creating model summaries that produce results compatible with what apsrtable expects.

Examples

 
     ## Use the example from lm() to show both models:
     ## Annette Dobson (1990) "An Introduction to Generalized Linear Models".
     ## Page 9: Plant Weight Data.
     ctl <- c(4.17,5.58,5.18,6.11,4.50,4.61,5.17,4.53,5.33,5.14)
     trt <- c(4.81,4.17,4.41,3.59,5.87,3.83,6.03,4.89,4.32,4.69)
     group <- gl(2,10,20, labels=c("Ctl","Trt"))
     weight <- c(ctl, trt)
     lm.D9 <- lm(weight ~ group)
     glm.D9 <- glm(weight~group)
     lm.D90 <- lm(weight ~ group - 1) # omitting intercept
     apsrtable(lm.D90, lm.D9, glm.D9, digits=1, align="center", 
               stars="default", model.counter=0, order="rl")
     ## Not run: 
apsrtable(lm.D90, lm.D9, glm.D9, digits=1, align="l", 
          stars=1, model.counter=0, order="rl",
          coef.rows=1, col.hspace="3em", float="sidewaystable")

## Omit rows by regular expressions
apsrtable(lm.D9, omitcoef=expression(grep("\\(",coefnames)))
apsrtable(lm.D90,lm.D9,
          omitcoef=list("groupCtl",
            expression(grep("\\(",coefnames,value=TRUE))
            )
          )

## End(Not run)

apsrtable documentation built on May 2, 2019, 5:20 a.m.