Hybrid Design MLE and likelihood

Share:

Description

hybdes() computes the MLE for a Hybrid Design. hyblik() computes the likelihood for a Hybrid Design at a specified parameter vector

Usage

1
2
3
hybdes(MM, NN, cc, ntrue = 0, aprx='binom', start.mle=NA, group.int=FALSE, betafct =
 function(x){return(x[1] + c(0,x[-1]) )}, print.level = 0, iterlim = 100)
hyblik(beta.matrix, MM, NN, cc, aprx = 'binom', ntrue = 0, group.int=FALSE)

Arguments

MM

MM is a matrix of margin totals. Rows are groups, columns are margin totals

NN

NN is a matrix of outcome margin totals. Rows are groups, columns are margin totals. NN is always a K x 2 matrix, where K is the number of groups

cc

cc is a list of case-control data. Each element is a table with exposure (rows) and outcome (columns)

ntrue

The number of groups that should be calculated using the true hybrid likelihood, rather than an approximation.

aprx

Type of approximation to use when calculating the hybrid likelihood. Default is the binomial approximation.

start.mle

Starting value for the Newton-Raphson algorithm used to determine Hybrid Design MLE.

group.int

A logical indicator of whether or not groups should be treated as having different intercept parameters.

betafct

A function used to specify the model of interest by reparamterizing the hybrid likelihood. betafct() takes in group-specific parameters associated with each level of the exposure variable. The default function corresponds to a model with an intercept parameter and log-odds-ratio parameters relating levels of X to the baseline level, X = 0 (i.e. column 1 of MM).

print.level

Argument passed into nlm()

iterlim

Argument passed into nlm()

beta.matrix

Parameter values for likelihood calculation; used only in hyblik(). This should be entered in the form of a matrix, with one row per group and one column per parameter.

Value

mle

MLE of the hybrid design

start.mle

Result of clogit function (stratified case-control MLE)

Author(s)

E. Smoot

References

Smoot, E., and S. Haneuse. "On the Analysis of Hybrid Designs that Combine Group- and Individual-Level Data." Biometrics (in press, 2014).

Examples

1
#hybdes(MM, NN, cc, approx='NA')