R/boot_het.R

In puniform: Meta-Analysis Methods Correcting for Publication Bias

### Function to compute p-value with bootstrapping for test of homogeneity
boot_het <- function(k, est0, vi, ycv, method, con) 
{ 
  
  yi.boot <- rnorm(k, mean = est0, sd = sqrt(vi)) # Sample k effect sizes given est0 and vi
  
  ### Estimate effect size with tau=0
  est.boot <- suppressWarnings(try(uniroot(pdist_nsig, interval = c(-5, 5), tol = con$tol, tau = 0, 
                          yi = yi.boot, vi = vi, param = "est", ycv = ycv, 
                          method = method, val = "es", cv_P = 0)$root, silent = TRUE))
  
  if (inherits(est.boot, what = "try-error")) 
  { # If effect size cannot be estimated, return NA
    stat <- NA
  } else 
  {
    ### Compute conditional probabilities at d.boot
    tr.q <- trq(est = est.boot, tau = 0, yi = yi.boot, vi = vi, ycv = ycv, param = "est")
    
    het.q.boot <- 2*abs(tr.q-0.5) # Compute heterogeneity statistic
    stat <- sum(-log(1-het.q.boot))
  }
  return(stat = stat)
}