Introduction to the 'memisc' Package
00memisc.RdThis package collects an assortment of tools that are intended to make
work with R easier for the author of this package
and are submitted to the public in the hope that they will be also be useful to others.
The tools in this package can be grouped into four major categories:
Data preparation and management
Data analysis
Presentation of analysis results
Programming
Data preparation and management
Survey Items
memisc provides facilities to work with what users from other
packages like SPSS, SAS, or Stata know as `variable labels', `value labels'
and `user-defined missing values'. In the context of this package these
aspects of the data are represented by the "description",
"labels", and "missing.values" attributes of a
data vector.
These facilities are useful, for example, if you work with
survey data that contain coded items like vote intention that
may have the following structure:
Question: “If there was a parliamentary election next tuesday, which party would you vote for?”
| 1 | Conservative Party |
| 2 | Labour Party |
| 3 | Liberal Democrat Party |
| 4 | Scottish Nation Party |
| 5 | Plaid Cymru |
| 6 | Green Party |
| 7 | British National Party |
| 8 | Other party |
| 96 | Not allowed to vote |
| 97 | Would not vote |
| 98 | Would vote, do not know yet for which party |
| 99 | No answer |
A statistical package like SPSS allows to
attach labels like `Conservative Party', `Labour Party', etc.
to the codes 1,2,3, etc. and to mark
mark the codes 96, 97, 98, 99
as `missing' and thus to exclude these variables from statistical
analyses. memisc provides similar facilities.
Labels can be attached to codes by calls like labels(x) <- something
and expendanded by calls like labels(x) <- labels(x) + something,
codes can be marked as `missing' by
calls like missing.values(x) <- something and
missing.values(x) <- missing.values(x) + something.
memisc defines a class called "data.set", which is similar to the class "data.frame".
The main difference is that it is especially geared toward containing survey item data.
Transformations of and within "data.set" objects retain the information about
value labels, missing values etc. Using as.data.frame sets the data up for
R's statistical functions, but doing this explicitely is seldom necessary.
See data.set.
More Convenient Import of External Data
Survey data sets are often relative large and contain up to a few thousand variables.
For specific analyses one needs however only a relatively small subset of these variables.
Although modern computers have enough RAM to load such data sets completely into an R session,
this is not very efficient having to drop most of the variables after loading. Also, loading
such a large data set completely can be time-consuming, because R has to allocate space for
each of the many variables. Loading just the subset of variables really needed for an analysis
is more efficient and convenient - it tends to be much quicker. Thus this package provides
facilities to load such subsets of variables, without the need to load a complete data set.
Further, the loading of data from SPSS files is organized in such a way that all informations
about variable labels, value labels, and user-defined missing values are retained.
This is made possible by the definition of importer objects, for which
a subset method exists. importer objects contain only
the information about the variables in the external data set but not the data.
The data itself is loaded into memory when the functions subset or as.data.set
are used.
Recoding
memisc also contains facilities for recoding
survey items. Simple recodings, for example collapsing answer
categories, can be done using the function recode. More
complex recodings, for example the construction of indices from
multiple items, and complex case distinctions, can be done
using the function cases. This function may also
be useful for programming, in so far as it is a generalization of
ifelse.
Code Books
There is a function codebook which produces a code book of an
external data set or an internal "data.set" object. A codebook contains in a
conveniently formatted way concise information about every variable in a data set,
such as which value labels and missing values are defined and some univariate statistics.
An extended example of all these facilities is contained in the vignette "anes48",
and in demo(anes48)
Data Analysis
Tables and Data Frames of Descriptive Statistics
genTable is a generalization of xtabs:
Instead of counts, also descriptive statistics like means or variances
can be reported conditional on levels of factors. Also conditional
percentages of a factor can be obtained using this function.
In addition an Aggregate
function is provided, which has the same syntax as genTable, but
gives a data frame of descriptive statistics instead of a table
object.
Presentation of Results of Statistical Analysis
Publication-Ready Tables of Coefficients
Journals of the Political and Social Sciences usually require that estimates of regression models are presented in the following form:
==================================================
Model 1 Model 2 Model 3
--------------------------------------------------
Coefficients
(Intercept) 30.628*** 6.360*** 28.566***
(7.409) (1.252) (7.355)
pop15 -0.471** -0.461**
(0.147) (0.145)
pop75 -1.934 -1.691
(1.041) (1.084)
dpi 0.001 -0.000
(0.001) (0.001)
ddpi 0.529* 0.410*
(0.210) (0.196)
--------------------------------------------------
Summaries
R-squared 0.262 0.162 0.338
adj. R-squared 0.230 0.126 0.280
N 50 50 50
==================================================
Such tables of coefficient estimates can be produced
by mtable. To see some of the possibilities of
this function, use example(mtable).
LaTeX Representation of R Objects
Output produced by mtable can be transformed into
LaTeX tables by an appropriate method of the generic function
toLatex which is defined in the package
utils. In addition, memisc defines toLatex methods
for matrices and ftable objects. Note that
results produced by genTable can be coerced into
ftable objects. Also, a default method
for the toLatex function is defined which coerces its
argument to a matrix and applies the matrix method of toLatex.
Programming
Looping over Variables
Sometimes users want to contruct loops that run over variables rather than values.
For example, if one wants to set the missing values of a battery of items.
For this purpose, the package contains the function foreach.
To set 8 and 9 as missing values for the items knowledge1,
knowledge2, knowledge3, one can use
foreach(x=c(knowledge1,knowledge2,knowledge3),
missing.values(x) <- 8:9)
Changing Names of Objects and Labels of Factors
R already makes it possible to change the names of an object.
Substituting the names or dimnames
can be done with some programming tricks. This package defines
the function rename,
dimrename, colrename, and rowrename
that implement these tricks in a convenient way, so that programmers
(like the author of this package) need not reinvent the weel in
every instance of changing names of an object.
Dimension-Preserving Versions of lapply and sapply
If a function that is involved in a call to
sapply returns a result an array or a matrix, the
dimensional information gets lost. Also, if a list object to which
lapply or sapply are applied
have a dimension attribute, the result looses this information.
The functions Lapply and
Sapply defined in this package preserve such
dimensional information.
Combining Vectors and Arrays by Names
The generic function collect collects several objects of the
same mode into one object, using their names, rownames,
colnames and/or dimnames. There are methods for
atomic vectors, arrays (including matrices), and data frames.
For example
leads to
x y
a 1 10
b 2 NA
c NA 30
Reordering of Matrices and Arrays
The memisc package includes a reorder
method for arrays and matrices. For example, the matrix
method by default reorders the rows of a matrix according the results
of a function.