contingency tables in R

Patrick Ball writes:

Dear List:

Most of the analysis I do involves contingency tables.  I am migrating
to R from Stata and I have a number of questions about using
contingency tables in R.  I suspect that most of the things I want to
do are very short R scripts that people on this list probably have.  I
wonder if you would be willing to share them.

First, the presentation of tables by table() is not analysis-ready.
Is there a way to output the table with the marginals, by cell, row or
column proportions, with the test statistics (especially the chi^2 and
the log-likelihood chi^2), residuals, cross product, and odds ratio?

Not in one monolithic function, I think, and I am not sure I would like
to have such a thing, see below.  But the pieces are all there:

* Use margin.table() and prop.table() to obtain margins and proportions,
 respectively.

* Use chisq.test() [in package ctest] for the chisq analysis (test
 statistic, p-value, chisq residuals)

* Use loglin() for the LR chisq and residuals.

* Not sure about which odds ratios you want.  Function mantelhaen.test()
 in package ctest does exact conditional ones for 2 by 2 tables.

It really depends on how your data is set up.  If you have the raw
values in a data frame, I would actually recommend using xtabs() rather
than table().  Try e.g.

    data(esoph)
    x <- xtabs(cbind(ncases, ncontrols) ~ ., data = esoph)
    x
    summary(x)

the last one prints ``useful'' summary information.

To obtain pretty-printed output from multi-way tables, use ftable().

I also like to make tables that have summary statistics of a given
variable in the columns (mean, s.d., etc.)  with each row being the
value for a sub group of the data.  How do you do this in R?

Use aggregate().

The most complicated piece of this is contingency tables done with
sample data.  The sampling involves several strata with different
sampling weights.  Calculating the cell (or row or column)
probabilities is relatively easy, but the other statistics can be
complicated (the design effect, the finite population correction, the
various chi^2s, and the standard errors and confidence
intervals). Also, I sometimes make these tables with summary
statistics in place of counts or population proportions.

Is there any way to do this stuff in R without hacking it all myself?

The pieces are all there, I think, and it should be fairly simple to
combine them to reflect your personal preferences for displaying
categorical information etc.