Fix down-weighting

* fixing the weights parameterization

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: RoBMA
 Title: Robust Bayesian Meta-Analyses
-Version: 2.3.0
+Version: 2.3.1
 Maintainer: František Bartoš <[email protected]>
 Authors@R: c( 
     person("František", "Bartoš",     role = c("aut", "cre"),

NEWS.md

@@ -1,6 +1,10 @@
+## version 2.3.1
+### Fixes
+- fixing weighted meta-analysis parameterization 
+
 ## version 2.3
 ### Features
-- weighted meta-analysis by specifying `study_ids` argument in `RoBMA` and setting `weighted = TRUE`. The likelihood contribution of estimates from each study is down-weighted proportionally to the number of estimates in that study. Note that this experimental feature is supposed to provide a conservative alternative for estimating RoBMA in cases with multiple estimates from a study where the multivariate option is not computationally feasible.
+- weighted meta-analysis by specifying `study_ids` argument in `RoBMA()` and setting `weighted = TRUE`. The likelihood contribution of estimates from each study is down-weighted proportionally to the number of estimates in that study. Note that this experimental feature is supposed to provide a conservative alternative for estimating RoBMA in cases with multiple estimates from a study where the multivariate option is not computationally feasible.
 
 ## version 2.2.3
 ### Fixes

R/fit-and-marglik.R

@@ -133,7 +133,7 @@
     marglik                 <- list()
 
     if(attr(model, "weighted")){
-      marglik$logml           <- sum(stats::dnorm(fit_data[["y"]], priors$mu$parameters[["location"]], fit_data[["se"]], log = TRUE) + log(fit_data[["weight"]]))
+      marglik$logml           <- sum(stats::dnorm(fit_data[["y"]], priors$mu$parameters[["location"]], fit_data[["se"]], log = TRUE) * fit_data[["weight"]])
     }else{
       marglik$logml           <- sum(stats::dnorm(fit_data[["y"]], priors$mu$parameters[["location"]], fit_data[["se"]], log = TRUE))
     }
@@ -445,11 +445,11 @@
   # the individual studies
   if(!is.null(data[["weight"]])){
     if(is.null(priors[["omega"]])){
-      log_lik <- log_lik + sum(stats::dnorm(data[["y"]], mean = eff, sd = pop_sd, log = TRUE) + log(data[["weight"]]))
+      log_lik <- log_lik + sum(stats::dnorm(data[["y"]], mean = eff, sd = pop_sd, log = TRUE) * data[["weight"]])
     }else if(priors[["omega"]]$distribution == "one.sided"){
-      log_lik <- log_lik + sum(.dwnorm_fast(data[["y"]], mean = eff, sd = pop_sd, omega = omega, crit_x = t(data[["crit_y"]]), type = "one.sided", log = TRUE) + log(data[["weight"]]))
+      log_lik <- log_lik + sum(.dwnorm_fast(data[["y"]], mean = eff, sd = pop_sd, omega = omega, crit_x = t(data[["crit_y"]]), type = "one.sided", log = TRUE) * data[["weight"]])
     }else if(priors[["omega"]]$distribution == "two.sided"){
-      log_lik <- log_lik + sum(.dwnorm_fast(data[["y"]], mean = eff, sd = pop_sd, omega = omega, crit_x = t(data[["crit_y"]]), type = "two.sided", log = TRUE) + log(data[["weight"]]))
+      log_lik <- log_lik + sum(.dwnorm_fast(data[["y"]], mean = eff, sd = pop_sd, omega = omega, crit_x = t(data[["crit_y"]]), type = "two.sided", log = TRUE) * data[["weight"]])
     }
   }else{
     if(is.null(priors[["omega"]])){
@@ -654,7 +654,7 @@
 
   # fill their weights
   for(i in seq_along(ids_weights$id)){
-    weights[!is.na(data[,"study_ids"]) & data[,"study_ids"] == ids_weights$id[i]] <- 1/ids_weights$weight[ids_weights$id == ids_weights$id[i]]
+    weights[!is.na(data[,"study_ids"]) & data[,"study_ids"] == ids_weights$id[i]] <- ids_weights$weight[ids_weights$id == ids_weights$id[i]]
   }
 
   # assign all remaining studies weight 1

R/utilities.R

@@ -85,6 +85,7 @@ assign("max_cores",       parallel::detectCores(logical = TRUE) - 1,  envir = Ro
     "2.2.2" = c("0.2.3", "999.999.999"),
     "2.2.3" = c("0.2.3", "999.999.999"),
     "2.3.0" = c("0.2.3", "999.999.999"),
+    "2.3.1" = c("0.2.3", "999.999.999"),
     stop("New RoBMA version needs to be defined in '.check_BayesTools' function!")
   )
 

src/distributions/DWN.cc

@@ -47,7 +47,7 @@ double DWN::logDensity(double const *x, unsigned int length, PDFType type,
   double weight  = *par[2];
 
   // compute the nominator
-  double log_lik = dnorm(*x, mu, std::sqrt(var), true) + std::log(weight);
+  double log_lik = dnorm(*x, mu, std::sqrt(var), true) * weight;
 
   return log_lik;
 }

src/distributions/DWWN1.cc

@@ -108,7 +108,7 @@ double DWWN1::logDensity(double const *x, unsigned int length, PDFType type,
   }
 
   // compute the log likelihood
-  double log_lik = nom-std::log(denom)+std::log(weight);
+  double log_lik = (nom-std::log(denom)) * weight;
 
   return log_lik;
 }

src/distributions/DWWN2.cc

@@ -120,7 +120,7 @@ double DWWN2::logDensity(double const *x, unsigned int length, PDFType type,
   }
 
   // compute the log likelihood
-  double log_lik = nom-std::log(denom)+std::log(weight);
+  double log_lik = (nom-std::log(denom)) * weight;
 
   return log_lik;
 }