fix for negative variances in network.predictability

hfgolino · Mar 26, 2024 · 2c0ed02 · 2c0ed02
1 parent 3783f37
commit 2c0ed02
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 3 deletions.
diff --git a/NEWS b/NEWS
@@ -8,6 +8,8 @@ o FIX: update to revised network loadings signs
 
 o FIX: bug in argument 'returnAllResults' for `EBICglasso.qgraph`
 
+o UPDATE: `network.predictability` uses empirical inverse variances (rather than network-implied)
+
 
 Changes in version 2.0.5
 

diff --git a/R/network.predictability.R b/R/network.predictability.R
@@ -121,7 +121,7 @@
 #' @export
 #'
 # Predict new data based on network ----
-# Updated 02.03.2024
+# Updated 26.03.2024
 network.predictability <- function(network, original.data, newdata, ordinal.categories = 7)
 {
 
@@ -179,8 +179,12 @@ network.predictability <- function(network, original.data, newdata, ordinal.cate
 
   }
 
-  # Get the inverse variances (use absolute for less than ideal matrices)
-  inverse_variances <- abs(diag(pcor2inv(network)))
+  # Get the inverse variances of original data
+  inverse_variances <- diag(solve(auto.correlate(original.data)))
+  # Although it's probably more appropriate to use the network-implied
+  # invariance (co)variances, there are some network estimation methods
+  # that result in non-positive definite or near singular matrices with
+  # negative inverse variances
 
   # Get betas
   betas <- network * sqrt(outer(inverse_variances, inverse_variances, FUN = "/"))