xtabs Cross Tabulation

Description

Create a contingency table (optionally a sparse matrix) from cross-classifying factors, usually contained in a data frame, using a formula interface.

Usage

xtabs(formula = ~., data = parent.frame(), subset, sparse = FALSE,
      na.action, addNA = FALSE, exclude = if(!addNA) c(NA, NaN),
      drop.unused.levels = FALSE)

## S3 method for class 'xtabs'
print(x, na.print = "", ...)

Arguments

formula	a formula object with the cross-classifying variables (separated by +) on the right hand side (or an object which can be coerced to a formula). Interactions are not allowed. On the left hand side, one may optionally give a vector or a matrix of counts; in the latter case, the columns are interpreted as corresponding to the levels of a variable. This is useful if the data have already been tabulated, see the examples below.
data	an optional matrix or data frame (or similar: see model.frame) containing the variables in the formula formula. By default the variables are taken from environment(formula).
subset	an optional vector specifying a subset of observations to be used.
sparse	logical specifying if the result should be a sparse matrix, i.e., inheriting from sparseMatrix Only works for two factors (since there are no higher-order sparse array classes yet).
na.action	a function which indicates what should happen when the data contain NAs. If unspecified, and addNA is true, this is set to na.pass. When it is na.omit and formula has a left hand side (with counts), sum(*, na.rm = TRUE) is used instead of sum(*) for the counts.
addNA	logical indicating if NAs should get a separate level and be counted, using addNA(*, ifany=TRUE) and setting the default for na.action to na.pass.
exclude	a vector of values to be excluded when forming the set of levels of the classifying factors.
drop.unused.levels	a logical indicating whether to drop unused levels in the classifying factors. If this is FALSE and there are unused levels, the table will contain zero marginals, and a subsequent chi-squared test for independence of the factors will not work.
x	an object of class "xtabs".
na.print	character string (or NULL) indicating how NA are printed. The default ("") does not show NAs clearly, and na.print = "NA" maybe advisable instead.
...	further arguments passed to or from other methods.

Details

There is a summary method for contingency table objects created by table or xtabs(*, sparse = FALSE), which gives basic information and performs a chi-squared test for independence of factors (note that the function chisq.test currently only handles 2-d tables).

If a left hand side is given in formula, its entries are simply summed over the cells corresponding to the right hand side; this also works if the lhs does not give counts.

For variables in formula which are factors, exclude must be specified explicitly; the default exclusions will not be used.

In R versions before 3.4.0, e.g., when na.action = na.pass, sometimes zeroes (0) were returned instead of NAs.

Note that when addNA is false as by default, and na.action is not specified (or set to NULL), in effect na.action = getOption("na.action", default=na.omit) is used; see also the examples.

Value

By default, when sparse = FALSE, a contingency table in array representation of S3 class c("xtabs", "table"), with a "call" attribute storing the matched call.

When sparse = TRUE, a sparse numeric matrix, specifically an object of S4 class dgTMatrix from package Matrix.

See Also

table for traditional cross-tabulation, and as.data.frame.table which is the inverse operation of xtabs (see the DF example below).

sparseMatrix on sparse matrices in package Matrix.

Examples

## 'esoph' has the frequencies of cases and controls for all levels of
## the variables 'agegp', 'alcgp', and 'tobgp'.
xtabs(cbind(ncases, ncontrols) ~ ., data = esoph)
## Output is not really helpful ... flat tables are better:
ftable(xtabs(cbind(ncases, ncontrols) ~ ., data = esoph))
## In particular if we have fewer factors ...
ftable(xtabs(cbind(ncases, ncontrols) ~ agegp, data = esoph))

## This is already a contingency table in array form.
DF <- as.data.frame(UCBAdmissions)
## Now 'DF' is a data frame with a grid of the factors and the counts
## in variable 'Freq'.
DF
## Nice for taking margins ...
xtabs(Freq ~ Gender + Admit, DF)
## And for testing independence ...
summary(xtabs(Freq ~ ., DF))

## with NA's
DN <- DF; DN[cbind(6:9, c(1:2,4,1))] <- NA
DN # 'Freq' is missing only for (Rejected, Female, B)
tools::assertError(# 'na.fail' should fail :
     xtabs(Freq ~ Gender + Admit, DN, na.action=na.fail), verbose=TRUE)
op <- options(na.action = "na.omit") # the "factory" default
(xtabs(Freq ~ Gender + Admit, DN) -> xtD)
noC <- function(O) `attr<-`(O, "call", NULL)
ident_noC <- function(x,y) identical(noC(x), noC(y))
stopifnot(exprs = {
  ident_noC(xtD, xtabs(Freq ~ Gender + Admit, DN, na.action = na.omit))
  ident_noC(xtD, xtabs(Freq ~ Gender + Admit, DN, na.action = NULL))
})

xtabs(Freq ~ Gender + Admit, DN, na.action = na.pass)
## The Female:Rejected combination has NA 'Freq' (and NA prints 'invisibly' as "")
(xtNA <- xtabs(Freq ~ Gender + Admit, DN, addNA = TRUE)) # ==> count NAs
## show NA's better via  na.print = ".." :
print(xtNA, na.print= "NA")


## Create a nice display for the warp break data.
warpbreaks$replicate <- rep_len(1:9, 54)
ftable(xtabs(breaks ~ wool + tension + replicate, data = warpbreaks))

### ---- Sparse Examples ----

if(require("Matrix")) withAutoprint({
 ## similar to "nlme"s  'ergoStool' :
 d.ergo <- data.frame(Type = paste0("T", rep(1:4, 9*4)),
                      Subj = gl(9, 4, 36*4))
 xtabs(~ Type + Subj, data = d.ergo) # 4 replicates each
 set.seed(15) # a subset of cases:
 xtabs(~ Type + Subj, data = d.ergo[sample(36, 10), ], sparse = TRUE)

 ## Hypothetical two-level setup:
 inner <- factor(sample(letters[1:25], 100, replace = TRUE))
 inout <- factor(sample(LETTERS[1:5], 25, replace = TRUE))
 fr <- data.frame(inner = inner, outer = inout[as.integer(inner)])
 xtabs(~ inner + outer, fr, sparse = TRUE)
})