RML: Constrained maximum likelihood estimation for a random...
In nshi-stat/netiim3: Improved Inference Methods for Network Meta-Analysis

Description Usage Arguments Details Value References Examples

Constrained maximum likelihood estimation for a random effects model for multivariate meta-analysis when the first component of the grand mean vector is fixed.

1	RML(y, S, ml0, mu0, maxitr = 200)

`y`	N x p matrix of outcome variables.
`S`	Series of within-study covariance matrices of the outcome variables. A matrix or data frame with N rows and p(p+1)/2 columns.
`ml0`	Initial value of the grand mean vector except for the first component.
`mu0`	The value of the first component of the grand mean vector.
`maxitr`	The maximum iteration number of the Newton-Raphson algorithm.

The correlation matrix of the between-studies covariance matrix is set to compound symmetry with the correlation coefficient 0.50. It can be changed by modifying the source code. Please see Noma et al. (2017) for details.

The loglikelihood at the converged point.

Noma, H., Nagashima, K., Maruo, K., Gosho, M., Furukawa, T. A. (2017). Bartlett-type corrections and bootstrap adjustments of likelihood-based inference methods for network meta-analysis. ISM Research Memorandum 1205.

# dae <- data.aug.edit(smoking)
# y <- dae$y
# S <- dae$S

# ml1 <- ML(y, S)

# a1 <- ml1$Coefficients[,1]
# a2 <- (ml1$`Between-studies_SD`)^2
# a3 <- a2*(ml1$`Between-studies_COR`)
# a4 <- c(a1, a2, a3)

# mlike0 <- .5*ml1$Loglikelihood
# mlike1 <- .5*RML(y, S, ml0 = a4, mu0 = 0)
# LR0 <- -2*(mlike1 - mlike0)  # the likelihood ratio statistic