citest-array: Test for conditional independence in a contingency table

citest-arrayR Documentation

Test for conditional independence in a contingency table

Description

Test for conditional independence in a contingency table represented as an array.

Usage

ciTest_table(
  x,
  set = NULL,
  statistic = "dev",
  method = "chisq",
  adjust.df = TRUE,
  slice.info = TRUE,
  L = 20,
  B = 200,
  ...
)

Arguments

x

An array of counts with named dimnames.

set

A specification of the test to be made. The tests are of the form u and v are independent condionally on S where u and v are variables and S is a set of variables. See 'details' for details about specification of set.

statistic

Possible choices of the test statistic are "dev" for deviance and "X2" for Pearsons X2 statistic.

method

Method of evaluating the test statistic. Possible choices are "chisq", "mc" (for Monte Carlo) and "smc" for sequential Monte Carlo.

adjust.df

Logical. Should degrees of freedom be adjusted for sparsity?

slice.info

Logical. Should slice info be stored in the output?

L

Number of extreme cases as stop criterion if method is "smc" (sequential Monte Carlo test); ignored otherwise.

B

Number (maximum) of simulations to make if method is "mc" or "smc" (Monte Carlo test or sequential Monte Carlo test); ignored otherwise.

...

Additional arguments.

Details

set can be 1) a vector or 2) a right-hand sided formula in which variables are separated by '+'. In either case, it is tested if the first two variables in the set are conditionally independent given the remaining variables in set. (Notice an abuse of the '+' operator in the right-hand sided formula: The order of the variables does matter.)

If set is NULL then it is tested whether the first two variables are conditionally independent given the remaining variables.

Value

An object of class citest (which is a list).

Author(s)

Søren Højsgaard, sorenh@math.aau.dk

See Also

ciTest, ciTest_df, ciTest_mvn, chisq.test

Examples


data(lizard)

## lizard is has named dimnames
names( dimnames( lizard ))
## checked with
is.named.array( lizard )

## Testing for conditional independence:
# the following are all equivalent:
ciTest(lizard, set=~diam + height + species)
# ciTest(lizard, set=c("diam", "height", "species"))
# ciTest(lizard, set=1:3)
# ciTest(lizard)
# (The latter because the names in lizard are as given above.)

## Testing for marginal independence
ciTest(lizard, set=~diam + height)
ciTest(lizard, set=1:2)

## Getting slice information:
ciTest(lizard, set=c("diam", "height", "species"), slice.info=TRUE)$slice

## Do Monte Carlo test instead of usual likelihood ratio test. Different
# options:

# 1) Do B*10 simulations divided equally over each slice: 
ciTest(lizard, set=c("diam", "height", "species"), method="mc", B=400)
# 2) Do at most B*10 simulations divided equally over each slice, but stop
# when at most L extreme values are found
ciTest(lizard, set=c("diam", "height", "species"), method="smc", B=400)

gRim documentation built on Sept. 11, 2024, 7:02 p.m.