binnest {repeated}

Binary Random Effects Models with Two Levels of Nesting

Description

binnest is designed to handle binary and binomial data with two levels of nesting. The first level is the individual and the second will consist of clusters within individuals.

Usage

binnest(
  response,
  totals = NULL,
  nest = NULL,
  ccov = NULL,
  tvcov = NULL,
  mu = ~1,
  re1 = ~1,
  re2 = ~1,
  preg = NULL,
  pre1 = NULL,
  pre2 = NULL,
  binom.mix = c(10, 10),
  binom.prob = c(0.5, 0.5),
  fcalls = 900,
  eps = 0.01,
  print.level = 0
)

Arguments

response

A list of three column matrices with counts, corresponding totals (not necessary if the response is binary), and (second-level) nesting indicator for each individual, one matrix or dataframe of such counts, or an object of class, response (created by restovec) or repeated (created by rmna).

totals

If response is a matrix or dataframe, a corresponding matrix or dataframe of totals (not necessary if the response is binary). Ignored otherwise.

nest

If response is a matrix or dataframe, a corresponding matrix or dataframe of nesting indicators. Ignored otherwise.

ccov

If response is a matrix, dataframe, list, or object of class, response, a matrix of time-constant covariates or an object of class, tccov (created by tcctomat). All of these covariates are used in the fixed effects part of the model. Ignored if response has class, repeated.

tvcov

If response is a matrix, dataframe, list, or object of class, response, an object of class, tvcov (created by tvctomat). All of these covariates are used in the fixed effects part of the model. Ignored if response has class, repeated.

mu

If response has class, repeated, a formula beginning with ~, specifying a linear regression function for the fixed effects, in the Wilkinson and Rogers notation, containing selected covariates in the response object. (A logit link is assumed.)

re1

If response has class, repeated, a formula beginning with ~, specifying a linear regression function for the variance of the first level of nesting, in the Wilkinson and Rogers notation, containing selected covariates in the response object. If NULL, a random effect is not fitted at this level. (A log link is assumed.)

re2

If response has class, repeated, a formula beginning with ~, specifying a linear regression function for the variance of the second level of nesting, in the Wilkinson and Rogers notation, containing selected covariates in the response object. If NULL, a random effect is not fitted at this level. (A log link is assumed.)

preg

Initial parameter estimates for the fixed effect regression model: either the model specified by mu or else the intercept plus one for each covariate in ccov and tvcov.

pre1

Initial parameter estimates for the first level of nesting variance model: either the model specified by re1 or just the intercept. If NULL, a random effect is not fitted at this level.

pre2

Initial parameter estimates for the second level of nesting variance model: either the model specified by re1 or just the intercept. If NULL, a random effect is not fitted at this level.

binom.mix

A vector of two values giving the totals for the binomial distributions used as the mixing distributions at the two levels of nesting.

binom.prob

A vector of two values giving the probabilities in the binomial distributions used as the mixing distributions at the two levels of nesting. If they are 0.5, the mixing distributions approximate normal mixing distributions; otherwise, they are skewed.

fcalls

Number of function calls allowed.

eps

Convergence criterion.

print.level

If 1, the iterations are printed out.

Details

The variance components at the two levels can only depend on the covariates if response has class, repeated.

Value

A list of classes binnest is returned.

Author(s)

T.R. Ten Have and J.K. Lindsey

References

Ten Have, T.R., Kunselman, A.R., and Tran, L. (1999) Statistics in Medicine 18, 947-960.

See Also

gar, read.list, restovec, rmna, tcctomat, tvctomat.

Examples

#y <- rbind(matrix(rbinom(20,1,0.6), ncol=4),
#	matrix(rbinom(20,1,0.4), ncol=4))
y <- matrix(c(1,1,0,1,1,1,1,0,1,1,1,1,1,1,1,1,0,1,1,0,0,1,0,1,1,0,1,0,
              1,1,1,1,1,1,1,1,0,1,1,0),nrow=10,ncol=4,byrow=TRUE)
resp <- restovec(y, nest=1:4, times=FALSE)
ccov <- tcctomat(c(rep(0,5),rep(1,5)), name="treatment")
reps <- rmna(resp, ccov=ccov)

# two random effects
binnest(reps, mu=~treatment, preg=c(1,1), pre1=2, pre2=2)

# first level random effect only
binnest(reps, mu=~treatment, preg=c(1,-1), pre1=1)

# second level random effect only
binnest(reps, mu=~treatment, preg=c(1,-1), pre2=1)