diff --git a/DESCRIPTION b/DESCRIPTION
index 5db1962..f86befa 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -2,8 +2,8 @@ Package: vcrpart
 Type: Package
 Title: Tree-Based Varying Coefficient Regression for Generalized Linear
         and Ordinal Mixed Models
-Version: 0.2-3
-Date: 2014-11-24
+Version: 0.3-1
+Date: 2015-01-04
 Authors@R: c(
   person("Reto", "Buergin", role = c("aut", "cre", "cph"), email = "rbuergin@gmx.ch"),
   person("Gilbert", "Ritschard", role = c("ctb", "ths"), 
@@ -14,11 +14,12 @@ License: GPL (>= 2)
 Depends: R (>= 3.1.0), parallel, partykit
 Imports: stats, grid, graphics, methods, nlme, rpart, numDeriv, ucminf,
         zoo, sandwich, strucchange
-URL: http://vcrpart.wordpress.com
+URL: http://vcrpart.wordpress.com,
+        https://r-forge.r-project.org/projects/traminer/
 LazyLoad: yes
 NeedsCompilation: yes
-Packaged: 2014-11-25 07:26:37 UTC; reto
+Packaged: 2015-01-04 10:34:57 UTC; reto
 Author: Reto Buergin [aut, cre, cph],
   Gilbert Ritschard [ctb, ths]
 Repository: CRAN
-Date/Publication: 2014-11-25 09:24:41
+Date/Publication: 2015-01-04 12:46:23
diff --git a/MD5 b/MD5
index 119891c..0205f29 100644
--- a/MD5
+++ b/MD5
@@ -1,20 +1,21 @@
-d9422ace5f5ef22d494ef37fbf19f831 *DESCRIPTION
+929438bed344b7ec2fc1864a751ff373 *DESCRIPTION
 adcba7e46ec21c27cfea94b0454d05f9 *NAMESPACE
-dc7c790c64c073850f833738ae0a4f59 *NEWS
+0b853ff955861ffa43fd11c9e4785172 *NEWS
 8b40e51cf24df0eb99503a7070b84a93 *R/AAA.R
 b70bc37834d2f3b8c94d9d1c0020c614 *R/AllGeneric.R
 91d50efd23854dea04fca307a8c9ce97 *R/fvcm.R
 80b84163397dcda539b20d9ab51c56e7 *R/import.R
-0801b64fedea58af67c400f8b0e15bd2 *R/olmm-methods.R
-7e4d28175b9184bf3a0992401f277072 *R/olmm-utils.R
+5658d937bde2aabee0e4493cdc76e85f *R/olmm-methods.R
+8057fd21604296b718055bd3dbb4f370 *R/olmm-utils.R
 ecdcf046dcbb466c4cf29bc7e9efaf8a *R/olmm.R
 f4e45051571cc6994e06109412d540af *R/otsplot.R
 2af1d39305eedf8fc4128b14307e1317 *R/tvcm-cv.R
-32a6d478af156ed579d5909795d5cc98 *R/tvcm-methods.R
+c1191337d41d4dc145e7ec221acd6ae0 *R/tvcm-methods.R
 d7de761ae5514253233d0540ddd85ec5 *R/tvcm-plot.R
-cd08110234d8f5bbcb6833bab79abad8 *R/tvcm-utils.R
-efa265167aaa866b67a1acd78df5a859 *R/tvcm.R
+a19aaa41484d924fb4aeeed7588d851c *R/tvcm-utils.R
+282d16856de0ec13cfbbbac8b57cf091 *R/tvcm.R
 645eb28d9f7e8259dfb5edf34edbb34b *R/utils.R
+dcc3606dae744cc31c166c3895363003 *ToDo
 8562fccdfc9dc113b72f8eb8df66bf57 *data/PL.RData
 a2c9a87cf50549fd5802cbb6a88d281a *data/movie.RData
 72890e1f368da6d42d9cf9414c5396e0 *data/poverty.RData
@@ -23,31 +24,31 @@ ea62be4e5d10df5bc683725a7f7935f6 *data/schizo.RData
 80e2db657049956161e19e3e3c2bdd32 *data/vcrpart_1.RData
 cb803decf95c39567f648e1e5e0ddb45 *data/vcrpart_2.RData
 3ddf5ac69dc37253946b83c5bfd5fca7 *data/vcrpart_3.RData
-3cd9ba9d0476859aa2f24592acd9629b *inst/CITATION
-3bc5b4abcf89156967fbf807a9125ee6 *man/PL.Rd
-7141810b94046f0dbfa66fd3e0477a6a *man/contr-wsum.Rd
-02a3675b72ff9f699d45dab0183780b8 *man/fvcm-methods.Rd
-9b47db74c740a7be156dce5564cf350f *man/fvcm.Rd
-edb080c605c0dacb43ae037d6199abd4 *man/movie.Rd
+c957bcc9e1c02457a3e2c0a3b3434c7c *inst/CITATION
+2d7f2b0ce2324e387983211e19ac55cc *man/PL.Rd
+79349138eb3456f71d2b2f9d1121d572 *man/contr-wsum.Rd
+1e6b881c70bc9464f3f870ebd929be25 *man/fvcm-methods.Rd
+a10e2af8c56ccd80e774c45b625fbbbb *man/fvcm.Rd
+c5210f80a7291c066c5cb99d430c306f *man/movie.Rd
 01062811412189130594c3983d1868ee *man/olmm-control.Rd
-85658aad587f90ada41ff1808c725904 *man/olmm-gefp.Rd
-82d6b519900b895e47ac8bb7fe83971f *man/olmm-methods.Rd
-e58a118afbf3ff05498f9e5bbc73e893 *man/olmm-predict.Rd
+dc2431d8c17e61515d75bd86c800cbda *man/olmm-gefp.Rd
+3ab54e39ac59254729fe58ceac548a8b *man/olmm-methods.Rd
+23e05ccd57e17ad0c231caad41210365 *man/olmm-predict.Rd
 c914ee5c769aacd190558f0b967f1072 *man/olmm-summary.Rd
-99fd9612846185c8b253368ad62640bc *man/olmm.Rd
-79da070b250ee6646ecbda2e4d33fcb3 *man/otsplot.Rd
-b5d023f2e25d796fc013667de0bf0c2b *man/poverty.Rd
-198b4f1a33689dc49700c02f14e42510 *man/schizo.Rd
-830a40f120259f6c1e30cce2f06a224e *man/tvcglm.Rd
-fc4d871d97b9315c31c720307d4b481c *man/tvcm-control.Rd
-5b0157aea2925e83c7325814864fad14 *man/tvcm-cv.Rd
-c2434053aa241bce6f70d969bf385e25 *man/tvcm-methods.Rd
-0fa2459d031af80f1b4dbd7d3d1da797 *man/tvcm-plot.Rd
-395961700e672d7ba7f7901564e83892 *man/tvcm.Rd
-81b709a611fe8b1ac9d3d6ebb806bab2 *man/tvcolmm.Rd
+5a0c35b431a537a976ff45eb197c7be9 *man/olmm.Rd
+1d5c9b3c457d06bf33220267be476307 *man/otsplot.Rd
+fe34790a447d3e5e8245fb0c0669c51c *man/poverty.Rd
+bce49a4bfca2f33530e20dcb8adf800b *man/schizo.Rd
+b1a8ea0cfa6a3e1e62435be8ce9f405e *man/tvcglm.Rd
+7dfdeb548a05ed57530085f9718d6464 *man/tvcm-control.Rd
+7d2b4dc5887bb05c129579f2a28021a3 *man/tvcm-cv.Rd
+0320ab676aaf77e818adc0ff4fa92c37 *man/tvcm-methods.Rd
+41720b3b0902b0340e55b55495d7311d *man/tvcm-plot.Rd
+e7e5e2934a4d2214baa75e89f6002b48 *man/tvcm.Rd
+c9e55e6137afe94e96cbf08c9eda6a69 *man/tvcolmm.Rd
 158d7c7045232f33bd35bf5c39dcedcf *man/vcrpart-demo.Rd
 a14ef1287c994bf09c716cc21d50517f *man/vcrpart-formula.Rd
-1e0af65b11fb043bd18f7982291d076f *src/Makevars
+8d46d69896a1f1d31ed76035a0e49d67 *src/Makevars
 e0ee3aca34161fb42f8fffa717fc6c3e *src/init.c
 62783432fffd5da456a3bb3e61c2be35 *src/olmm.c
 1e5f560c59e4ea73b51fa72c057166ec *src/olmm.h
diff --git a/NEWS b/NEWS
index 58faedd..82bdf57 100644
--- a/NEWS
+++ b/NEWS
@@ -1,12 +1,33 @@
+Changes in Version 0.3-1
+
+    o    Improvements for 'prunepath' method. 
+
+    o    Moved documentation of 'prunepath' to 'tvcm-assessment'.
+
+    o    Added example for 'prune'.
+
+    o    Internal changes in 'estfun.olmm' and new argument 'center' for 
+         the function 'predecor_control'.
+
+    o    New argument 'fast' for 'tvcm_control' to control whether an
+         approximative or the accurate model is used at exhaustive search.
+
+    o    Corrected bugs in 'tvcm_exsearch_nomToOrd' and 'tvcm_getNumSplits' 
+         which appeared in in special cases.
+
+    o    Improved help manual for 'tvcglm' and 'tvcolmm'.
+
+    o    Revised references.
+
 Changes in Version 0.2-3
 
-    o    Added contr.wsum function
+    o    Added contr.wsum function.
 
-    o    Added 'seed' argument to 'otsplot_control'
+    o    Added 'seed' argument to 'otsplot_control'.
 
-    o    Internal changes for exhaustive search of 'tvcm'
+    o    Internal changes for exhaustive search of 'tvcm'.
 
-    o    Improved documentation
+    o    Improved documentation.
 
 
 Changes in Version 0.2-2
@@ -31,7 +52,7 @@ Changes in Version 0.2-2
 
     o    Added 'nimpute' argument for 'tvcm_control'.
 
-    o    Added detail section to the help page of 'tvcm_control'
+    o    Added detail section to the help page of 'tvcm_control'.
 
     o    Removed AIC table from 'print.tvcm' (AIC and BIC seem not
          relevant measures for models fitted by 'tvcm').
@@ -49,7 +70,7 @@ Changes in Version 0.2-1
     o    First CRAN release.
 
     o    'tvcm' and 'fvcm' allow for multiple 'vc' terms, i.e. 
-         coefficient-specific partitions
+         coefficient-specific partitions.
 
     o    Complete revision of syntaxes, argument names and  default 
          parameters. R commands for the former version 0.1-14 are 
@@ -60,7 +81,7 @@ Changes in Version 0.2-1
          (old) coefficient constancy tests combined with greedy loss 
          minimization to (new) exhaustive greedy loss minimization. Splitting 
          based on the (old) requires to set the argument 'sctest = TRUE' in 
-         'tvcm_control'
+         'tvcm_control'.
 
      o	 The greedy loss reduction minimization stage of 'tvcm' does now 
          fixate coefficients which are not splitted at the estimates of 
diff --git a/R/olmm-methods.R b/R/olmm-methods.R
index 0f22c9f..68aed8f 100644
--- a/R/olmm-methods.R
+++ b/R/olmm-methods.R
@@ -1,6 +1,6 @@
 ##' -------------------------------------------------------- #
 ##' Author:          Reto Buergin, rbuergin@gmx.ch
-##' Date:            2014-10-24
+##' Date:            2014-12-08
 ##'
 ##' Description:
 ##' methods for olmm objects.
@@ -40,6 +40,7 @@
 ##' weights:     Weights
 ##'
 ##' Modifications:
+##' 2014-12-07: - add argument 'center' to 'predecor_control'
 ##' 2014-10-24: - improve simulate.olmm
 ##'             - improved 'estfun.olmm' call in 'gefp.olmm'
 ##' 2014-10-23: - fix bug in predict.olmm
@@ -157,20 +158,21 @@ deviance.olmm <- function(object, ...) return(-as.numeric(2.0 * logLik(object)))
 
 
 predecor_control <- function(impute = TRUE, seed = NULL,
-                             symmetric = TRUE,  reltol = 1e-6,
-                             maxit = 250L, minsize = 1L, 
+                             symmetric = TRUE, center = FALSE,
+                             reltol = 1e-6, maxit = 250L, minsize = 1L, 
                              verbose = FALSE, silent = FALSE) {
   stopifnot(is.logical(impute) && length(impute) == 1L)
   stopifnot(is.null(seed) | is.numeric(seed) && length(seed) == 1L)
   stopifnot(is.logical(symmetric) && length(symmetric) == 1L)
+  stopifnot(is.logical(center) && length(center) == 1L)
   stopifnot(is.numeric(reltol) && reltol > 0 && length(reltol) == 1L)
   stopifnot(is.numeric(maxit) && maxit > 0 && length(maxit) == 1L)
   stopifnot(is.numeric(minsize) && minsize > 0 && length(minsize) == 1L)
   stopifnot(is.logical(verbose) && length(verbose) == 1L)
   stopifnot(is.logical(silent) && length(silent) == 1L)
   return(structure(list(impute = impute, seed = seed, 
-                        symmetric = symmetric, reltol = reltol,
-                        maxit = maxit, minsize = minsize, 
+                        symmetric = symmetric, center = center,
+                        reltol = reltol, maxit = maxit, minsize = minsize, 
                         verbose = verbose, silent = silent),
                    class = "predecor_control"))
 }
@@ -214,12 +216,12 @@ estfun.olmm <- function(x, predecor = FALSE, control = predecor_control(),
   attr <- list() # default attributes
 
   scores <- x$score_obs
-  subsImp <- rep.int(FALSE, nrow(scores))
 
   if (control$verbose) cat("OK")
   
   ## impute data
-  
+
+  subsImp <- rep.int(FALSE, nrow(scores))  
   if (predecor && any(Ni != Nmax)) {
     
     Nimpute <- Nmax - Ni
@@ -251,21 +253,21 @@ estfun.olmm <- function(x, predecor = FALSE, control = predecor_control(),
     x$eta <- rbind(x$eta, matrix(0.0, sum(Nimpute), x$dims["nEta"]))
     x$score_obs <- rbind(x$score_obs, matrix(0.0, sum(Nimpute), x$dims["nPar"]))    
 
-    ## simulate responses
-    if (control$impute) {
-      
-      if (control$verbose) cat("\n* impute scores ... ")
-      
-      ## set seed
-      if (!is.null(control$seed)) set.seed(control$seed)
+    ## simulate responses      
+    if (control$verbose) cat("\n* impute scores ... ")
+    
+    ## set seed
+    if (!is.null(control$seed)) set.seed(control$seed)
 
+    if (control$impute) {
+    
       ## impute predictors
       times <- Nimpute[x$subject[!subsImp]]
       rows <- unlist(tapply(1:sum(Ni), x$subject[!subsImp], function(x) sample(x, times[x[1L]], replace = TRUE)))
       x$frame[subsImp,] <- x$frame[rows,,drop=FALSE]
       x$X[subsImp, ] <- x$X[rows,,drop=FALSE]
       x$W[subsImp, ] <- x$W[rows,,drop=FALSE]
-
+      
       ## draw responses
       subsW <- c(rep(which(attr(xOld$W, "merge") == 1L), x$dims["nEta"]),
                  which(attr(xOld$W, "merge") == 2L))
@@ -280,15 +282,16 @@ estfun.olmm <- function(x, predecor = FALSE, control = predecor_control(),
         ordered(apply(probs, 1L, function(x) sample(yLevs, 1L, prob = x)), yLevs)
       
       ## recompute scores
-      .Call("olmm_update_marg", x, x$coefficients, PACKAGE = "vcrpart")
-
-      scores <- 0.0 * x$score_obs
-      scores[!subsImp, ] <- x$score_obs[!subsImp,,drop=FALSE]
-      
-      if (control$verbose) cat("OK")
+      .Call("olmm_update_marg", x, x$coefficients, PACKAGE = "vcrpart")      
     }
-  }
+    
+    scores <- x$score_obs
 
+    if (control$center && max(abs(cSums <- colSums(scores))) > 1e-6)
+      scores <- scores -
+        matrix(cSums / nrow(scores), nrow(scores), ncol(scores), byrow = TRUE)
+  }
+  
   ## drop the nuisance coefficients
   scores <- scores[, parm, drop = FALSE]
   
diff --git a/R/olmm-utils.R b/R/olmm-utils.R
index 4da28b5..97731ff 100644
--- a/R/olmm-utils.R
+++ b/R/olmm-utils.R
@@ -543,9 +543,10 @@ olmm_rename <- function(x, levels, family, etalab = c("int", "char", "eta")) {
 ##' -------------------------------------------------------- #
 
 olmm_decormat <- function(scores, subject, control = predecor_control()) {
-
+  
   stopifnot(inherits(control, "predecor_control"))
   Nmax <- max(table(subject))
+  
   ## estimate variances and covariances
   sVar <- olmm_scoreVar(scores, subject)
   sCovWin <- olmm_scoreCovWin(scores, subject)
diff --git a/R/tvcm-methods.R b/R/tvcm-methods.R
index 79eef14..af16403 100644
--- a/R/tvcm-methods.R
+++ b/R/tvcm-methods.R
@@ -632,6 +632,7 @@ prune.tvcm <- function(tree, cp = NULL, alpha = NULL, maxstep = NULL,
 
 
 prunepath.tvcm <- function(tree, steps = 1L, ...) {
+  steps <- intersect(steps, seq_along(tree$info$prunepath))
   rval <- tree$info$prunepath[steps]
   rval <- lapply(rval, function(x) {
     x$tab <- as.data.frame(x$tab)
diff --git a/R/tvcm-utils.R b/R/tvcm-utils.R
index 08c3575..dcb0569 100644
--- a/R/tvcm-utils.R
+++ b/R/tvcm-utils.R
@@ -49,6 +49,13 @@
 ##' tvcm_grow_splitpath:      creates a 'splitpath.tvcm' object
 ##'
 ##' Last modifications:
+##' 2014-12-10: - added 'drop = FALSE' commands in 'tvcm_exsearch_nomToOrd'
+##'               which produced errors
+##'             - 'tvcm_getNumSplits' yielded sometimes more than 'maxnumsplit'
+##'               values. Now a random selection is applied for these cases
+##' 2014-12-09: implemented accurate search model. Involves changes in
+##'             'tvcm_formula', 'tvcm_grow_exsearch', 'tvcm_exsearch_dev'
+##'             and 'tvcm_control'.
 ##' 2014-11-11: modified transformation of nominal into ordinal variables
 ##'             to accelerate exhaustive search. There is now a function
 ##'             'tvcm_exsearch_nomToOrd'.
@@ -750,6 +757,10 @@ tvcm_getNumSplits <- function(z, w, minsize, maxnumsplit) {
 
     ## delete largest value
     rval <- rval[rval < maxz]
+
+    ## sometimes the while loop yields too many values ...
+    if (length(rval) > maxnumsplit)
+      rval <- sort(sample(rval, 9))
     
   } else {
 
@@ -1060,26 +1071,26 @@ tvcm_setsplits_splitnode <- function(splits, spart, snode, nodeid) {
 
 tvcm_setsplits_rselect <- function(splits, partid, nodeid, varid, control) {
 
-  ## get the node partitions
+  ## get the candidate node
   nodeidC <- nodeid
   for (pid in seq_along(partid))
     for (nid in seq_along(nodeid[[pid]]))
       if (all(sapply(splits[[pid]][[nid]], function(x) length(x) == 0L)))
         nodeidC[[pid]] <- setdiff(nodeidC[[pid]], nid)
 
-  ## get the partition candidates
+  ## get the candidate partitions
   partidC <- partid
   for (pid in seq_along(partid))
-    if (length(nodeidC[[pid]]) == 0L) setdiff(partidC, pid)
+    if (length(nodeidC[[pid]]) == 0L)
+      partidC <- setdiff(partidC, pid)
 
-  ## get variable candidates for each partition
+  ## get candidates variables for each partition
   varidC <- varid
   for (pid in seq_along(partid))
     for (vid in seq_along(varid[[pid]]))
       if (all(sapply(splits[[pid]], length) == 0L))
         varidC[[pid]] <- setdiff(varidC[[pid]], vid)
 
-
   ## random selections
   spart <- sort(sample(partidC, min(length(partidC), control$ptry)))
   svar <- lapply(seq_along(varidC), function(pid) {
@@ -1089,9 +1100,9 @@ tvcm_setsplits_rselect <- function(splits, partid, nodeid, varid, control) {
   snode <- lapply(seq_along(nodeidC), function(pid) {
     s <- sample(length(nodeidC[[pid]]), min(length(nodeidC[[pid]]),control$vtry[pid]))
     return(nodeidC[[pid]][sort(s)])
-  })
+  })  
 
-  ## delete not selected nodes from 'splits'
+  ## delete unselected nodes
   for (pid in seq_along(partid)) 
     for (nid in seq_along(nodeid[[pid]])) 
       for (vid in seq_along(varid[[pid]])) 
@@ -1321,11 +1332,11 @@ tvcm_exsearch_nomToOrd <- function(cp, pid, nid, vid,
     cp <- tvcm_getOrdSplits(z[subs], weights[subs],
                             control$minsize[pid],
                             control$maxordsplit)
-    cp <- cp[, levels(partData[, vid])]
+    cp <- cp[, levels(partData[, vid]),drop=FALSE]
   } else {
 
     ## avoid splitting (better solution?)
-    cp <- cp[-(1:nrow(cp)), ]
+    cp <- cp[-(1:nrow(cp)),,drop=FALSE]
   }
   attr(cp, "type") <- "coef"
 
@@ -1335,7 +1346,9 @@ tvcm_exsearch_nomToOrd <- function(cp, pid, nid, vid,
 
 tvcm_exsearch_dev <- function(cutpoint, 
                               pid, nid, vid, 
-                              model, modelNuis, nuisance,
+                              model, start,
+                              modelNuis, startNuis, 
+                              nuisance,
                               where, partData, 
                               control, loss0, mfName) {
       
@@ -1347,28 +1360,37 @@ tvcm_exsearch_dev <- function(cutpoint,
   } else {
     zs <- z %in% levels(z)[cutpoint > 0L]            
   }
-  model[[mfName]]$Left <- 1 * (subs & zs)
-  model[[mfName]]$Right <- 1 * (subs & !zs)
-  parm <- grep("Left", names(coef(model)), value = TRUE)
-  
-  ## fit the 'update' model
+
+  if (control$fast) {
+    model[[mfName]]$Left <- 1 * (subs & zs)
+    model[[mfName]]$Right <- 1 * (subs & !zs)
+  } else {
+    model[[mfName]][subs & zs, paste("Node", LETTERS[pid], sep = "")] <- "Left"
+    model[[mfName]][subs & !zs, paste("Node", LETTERS[pid], sep = "")] <- "Right"
+    model[[mfName]][!subs, paste("Node", LETTERS[pid], sep = "")] <-
+      as.integer(droplevels(where[[pid]][!subs]))
+  }
+  model$coefficients <- vcrpart_copy(start)
   model <- tvcm_grow_update(model, control)
+  rval <- rep(NA, 2L)
   
   if (!inherits(model, "try-error")) {
-    rval <- c((loss0 - control$lossfun(model)),
-              length(coef(model)[grep("Left", names(coef(model)))]) -
-              length(nuisance))
+    rval[1L] <- loss0 - control$lossfun(model)
+    rval[2L] <- length(coef(model)[grep("Right", names(coef(model)))])
     if (is.null(modelNuis)) {
       return(rval)
     } else {
       modelNuis[[mfName]]$Left <- 1 * (subs & zs)
       modelNuis[[mfName]]$Right <- 1 * (subs & !zs)
+      modelNuis$coefficients <- vcrpart_copy(startNuis)
       modelNuis <- tvcm_grow_update(modelNuis, control)
       rval[1L] <- rval[1L] - (loss0 - control$lossfun(modelNuis))
+      rval[2L] <- rval[2L] -
+        length(coef(modelNuis)[grep("Right", names(coef(modelNuis)))])
       return(rval)
-    }  
+    } 
   } else {
-    return(c(NA, NA))        
+    return(rval)        
   }
 }
 
@@ -1385,40 +1407,66 @@ tvcm_grow_exsearch <- function(splits, partid, nodeid, varid,
 
   mcall$data <- eval(mcall$data, environment(mcall))
   w <- weights(model)
-  mcall$offset <- predict(model, type = "link")
+
+  if (control$fast)
+      mcall$offset <- predict(model, type = "link")
+  
   if (inherits(model, "glm")) {
     mcall$x <- TRUE
     mcall$y <- TRUE
     mcall$model <- TRUE
   } else if (inherits(model, "olmm")) {
-    mcall$restricted <- grep("ranefCholFac", names(coef(model)), value = TRUE)
-    mcall$start <- coef(model)[mcall$restricted]
+    if (control$fast) {
+      mcall$restricted <- grep("ranefCholFac", names(coef(model)), value = TRUE)
+      mcall$start <- coef(model)[mcall$restricted]
+    }
   }
 
-  ff <- tvcm_formula(formList, rep.int(FALSE, length(partid)), 
+  if (control$fast) {
+    root <- rep.int(FALSE, length(partid))
+  } else {
+    root <- sapply(nodes, width) == 1
+  }
+  
+  ff <- tvcm_formula(formList, root, 
                      eval(mcall$family, environment(mcall)),
-                     environment(mcall), full = FALSE, update = TRUE)
+                     environment(mcall), full = FALSE,
+                     update = TRUE, fast = control$fast)
+
   Left <- sample(c(0, 1), nobs(model), replace = TRUE)
   Right <- Left - 1
+  Node <- lapply(where, function(x) {
+    levs <- c("Left", "Right", seq(1, nlevels(x) - 1, length.out = nlevels(x) - 1))
+    return(factor(rep(levs, length.out = length(x)), levels = levs))
+  })
   
   for (pid in seq_along(partid)) { 
     if (length(unlist(splits[[pid]])) > 0L) {
      
       mcall$formula <- ff$update[[pid]][[1L]]
+
       mcall$data$Left <- Left
-      mcall$data$Right <- Right 
+      mcall$data$Right <- Right      
+      mcall$data[, paste("Node", LETTERS[pid], sep = "")] <- Node[[pid]]
+
       sModel <- tvcm_grow_fit(mcall, doFit = FALSE)
-      sModel$coefficients[grepl("Left", names(sModel$coefficients))] <- 0.0
-      sModel$coefficients[grepl("Right", names(sModel$coefficients))] <- 0.0
+      sStart <- vcrpart_copy(sModel$coefficients)
+      sStart[intersect(names(sStart), names(coef(model)))] <-
+        coef(model)[intersect(names(sStart), names(coef(model)))]
+      
       sModel$control <- model$control
       
       if (length(control$nuisance[[pid]]) == 0L) {
         sModelN <- NULL
+        sNStart <- NULL
       } else {
         mcallN <- mcall
+        mcallN$offset <- predict(model, type = "link")
         mcallN$formula <- ff$update[[pid]][[2L]]
         sModelN <- tvcm_grow_fit(mcallN, doFit = FALSE)
-        sModelN$coefficients[] <- 0.0
+        sNStart <- sModelN$coefficients
+        sNStart[intersect(names(sNStart), names(coef(model)))] <-
+          coef(model)[intersect(names(sNStart), names(coef(model)))]
         sModelN$control <- model$control
       } 
       
@@ -1451,11 +1499,13 @@ tvcm_grow_exsearch <- function(splits, partid, nodeid, varid,
                           pid = partid[pid],
                           nid = nodeid[[partid[pid]]][nid],
                           vid = varid[[partid[pid]]][vid],
-                          model = sModel, modelNuis = sModelN,
+                          model = sModel, start = sStart,
+                          modelNuis = sModelN, startNuis = sNStart,
                           nuisance = control$nuisance[[pid]],
                           where = where, partData = partData,
                           control = control, loss0 = loss0,
                           mfName = mfName)
+
               if (is.matrix(st)) st <- t(st) else st <- matrix(st, ncol = 1L)
               splits[[pid]][[nid]][[vid]][[2L]][subs] <- st[, 1L]
               splits[[pid]][[nid]][[vid]][[3L]][subs] <- st[, 2L]
@@ -1649,8 +1699,9 @@ tvcm_grow_splitnode <- function(nodes, where, dev, partData, step, weights) {
 
 tvcm_formula <- function(formList, root, family = cumulative(),
                          env = parent.frame(),
-                         full = TRUE, update = FALSE) {
+                         full = TRUE, update = FALSE, fast = TRUE) {
 
+    
   yName <- rownames(attr(terms(formList$original), "factors"))[1L]
 
   ## puts the predictors for fixed effects and varying effects
@@ -1689,8 +1740,7 @@ tvcm_formula <- function(formList, root, family = cumulative(),
       rval <- paste(rval, paste(feTerms, collapse = "+"), sep = "")
     if (rval != "" && inherits(family, "family.olmm"))
       rval <- paste(effect, "(", rval, ")", sep = "")
-
-    
+   
     return(c(vcTerms, feTerms))
   }
   
@@ -1774,6 +1824,10 @@ tvcm_formula <- function(formList, root, family = cumulative(),
   ## update formulas
   
   fUpdate <- NULL
+
+  feCeTerms <- getTerms(formList, "ce", rep(FALSE, length(root)), family)
+  feGeTerms <- getTerms(formList, "ge", rep(FALSE, length(root)), family)
+  
   if (update) {
 
     ## get nuisance terms
@@ -1786,25 +1840,39 @@ tvcm_formula <- function(formList, root, family = cumulative(),
     for (pid in seq_along(fUpdate)) {
       fUpdate[[pid]] <- vector("list", 2L)
       nLab <- paste("Node", LETTERS[pid], sep = "")
-      
+
+      if (fast) {
+        feIntTmp <- "none"
+        feCeTmp <- feCeTerms[grep(nLab, feCeTerms)]
+        feGeTmp <- feGeTerms[grep(nLab, feGeTerms)]
+        feCeTmp <- c(gsub(nLab, "Left", feCeTmp), gsub(nLab, "Right", feCeTmp))
+        feGeTmp <- c(gsub(nLab, "Left", feGeTmp), gsub(nLab, "Right", feGeTmp))
+      } else {
+        rootTmp <- root
+        rootTmp[pid] <- FALSE
+        feIntTmp <- formList$fe$intercept
+        if (!is.null(vcInt) && any(direct) && rootTmp[direct])
+          feIntTmp <- "ce"
+        feCeTmp <- getTerms(formList, "ce", rootTmp, family)
+        feGeTmp <- getTerms(formList, "ge", rootTmp, family)         
+      }
+
       ## full formula
-      feCeTmp <- feCeTerms[grep(nLab, feCeTerms)]
-      feCeTmp <- c(gsub(nLab, "Left", feCeTmp), gsub(nLab, "Right", feCeTmp))
-      feGeTmp <- feGeTerms[grep(nLab, feGeTerms)]
-      feGeTmp <- c(gsub(nLab, "Left", feGeTmp), gsub(nLab, "Right", feGeTmp))
       fUpdate[[pid]][[1L]] <-
-        getForm(yName,feCeTmp,feGeTmp,"none",reForm, family, env)
+          getForm(yName,feCeTmp,feGeTmp,feIntTmp,reForm, family, env)
       
-      ## null formula
+      ## null formula (always use approximative model, even if fast = FALSE)
       feCeTmp <- feCeTerms[grep(nLab, feCeTerms)]
-      feCeTmp <- feCeTmp[grep(nLab, feCeTmp)]
       feCeTmp <- intersect(feCeTmp, nuisance[[pid]])
-      feCeTmp <- c(gsub(nLab, "Left", feCeTmp), gsub(nLab, "Right", feCeTmp))
+      
       feGeTmp <- feGeTerms[grep(nLab, feGeTerms)]
-      feGeTmp <- feGeTmp[grep(nLab, feGeTmp)]
       feGeTmp <- intersect(feGeTmp, nuisance[[pid]])
+
+      feCeTmp <- c(gsub(nLab, "Left", feCeTmp), gsub(nLab, "Right", feCeTmp))
       feGeTmp <- c(gsub(nLab, "Left", feGeTmp), gsub(nLab, "Right", feGeTmp))
-      fUpdate[[pid]][[2L]] <- getForm(yName,feCeTmp,feGeTmp,"none",reForm,family, env)
+           
+      fUpdate[[pid]][[2L]] <-
+          getForm(yName,feCeTmp,feGeTmp,"none",reForm,family, env)
     }
   }
   return(list(full = fFull, update = fUpdate))
diff --git a/R/tvcm.R b/R/tvcm.R
index f42e0db..c302978 100644
--- a/R/tvcm.R
+++ b/R/tvcm.R
@@ -1,7 +1,7 @@
 ##' -------------------------------------------------------- #
 ##' Author:      Reto Buergin
 ##' E-Mail:      reto.buergin@unige.ch, rbuergin@gmx.ch
-##' Date:        2014-10-23
+##' Date:        2015-01-04
 ##'
 ##' Description:
 ##' The 'tvcm' function
@@ -13,9 +13,11 @@
 ##' tvcm            the main fitting function
 ##' tvcm_control    control function for 'tvcm'
 ##'
-##' all functions are documented as *.Rd files
-##'
 ##' Last modifications:
+##' 2014-12-08: - enable 'sctest = FALSE' in 'tvcolmm_control'
+##'             - remove checks on length of argument list, which is
+##'               not necessary because R assigns the argument names
+##'               automatically
 ##' 2014-11-05: - set seed at start of 'tvcm' and re-establish old seed
 ##'               at the end
 ##' 2014-10-23: - improved extraction of fitting arguments (see 'fitargs')
@@ -72,15 +74,17 @@ tvcolmm <- function(formula, data, family = cumulative(),
 
 
 tvcolmm_control <- function(alpha = 0.05, bonferroni = TRUE, minsize = 50,
-                            maxnomsplit = 5, maxordsplit = 9, maxnumsplit = 9,
+                            maxnomsplit = 5, maxordsplit = 9,
+                            maxnumsplit = 9, fast = TRUE,
                             trim = 0.1, estfun.args = list(), nimpute = 5,
                             seed = NULL, ...) {
 
   mc <- match.call()
   mc[[1L]] <- as.name("tvcm_control")
-  if (!"minsize" %in% names(mc) & length(mc) < 7L)
+  if (!"minsize" %in% names(mc))
     mc$minsize <- formals(tvcolmm_control)$minsize
-  mc$sctest <- TRUE
+  if (!"sctest" %in% names(mc))
+    mc$sctest <- TRUE
   return(eval.parent(mc))
 }
 
@@ -90,13 +94,11 @@ tvcglm <- function(formula, data, family,
                    control = tvcglm_control(), ...) { 
     mc <- match.call()
     mc[[1L]] <- as.name("tvcm")
-    if (!"control" %in% names(mc) &
-        (length(mc) < 9L |
-         length(mc) >= 9L && !inherits(eval.parent(mc[[4L]]), "tvcm_control")))
-        mc$control <- formals(tvcglm)$control
+    if (!"control" %in% names(mc))
+      mc$control <- formals(tvcglm)$control
     mc$fit <- "glm"
     return(eval.parent(mc))
-}
+  }
 
 
 tvcglm_control <- function(minsize = 30, mindev = 2.0,
@@ -339,11 +341,11 @@ tvcm_control <- function(minsize = 30, mindev = ifelse(sctest, 0.0, 2.0),
                          trim = 0.1, estfun.args = list(), nimpute = 5, 
                          maxnomsplit = 5, maxordsplit = 9, maxnumsplit = 9,
                          maxstep = 1e3, maxwidth = 1e9, maxdepth = 1e9,
-                         lossfun = neglogLik2, ooblossfun = NULL,
+                         lossfun = neglogLik2, ooblossfun = NULL, fast = TRUE,
                          cp = 0.0, dfpar = 0.0, dfsplit = 1.0, 
                          cv = !sctest, folds = folds_control("kfold", 5),
                          prune = cv, papply = mclapply, papply.args = list(),
-                         center = TRUE, seed = NULL, verbose = FALSE, ...) {
+                         center = fast, seed = NULL, verbose = FALSE, ...) {
   mc <- match.call()
   
   ## check available arguments
@@ -370,6 +372,8 @@ tvcm_control <- function(minsize = 30, mindev = ifelse(sctest, 0.0, 2.0),
   stopifnot(is.function(lossfun))
   stopifnot(is.null(ooblossfun) | is.function(ooblossfun))
 
+  stopifnot(is.logical(fast) && length(fast) == 1L)
+  
   stopifnot(is.numeric(cp) && length(cp) == 1L)
   stopifnot(is.numeric(dfpar) && length(dfpar) == 1L)
   stopifnot(is.numeric(dfsplit) && length(dfsplit) == 1L)
@@ -439,6 +443,7 @@ tvcm_control <- function(minsize = 30, mindev = ifelse(sctest, 0.0, 2.0),
                 maxdepth = as.integer(maxdepth),
                 lossfun = lossfun,
                 ooblossfun = ooblossfun,
+                fast = fast,
                 cp = cp,
                 dfpar = dfpar,
                 dfsplit = dfsplit,
diff --git a/ToDo b/ToDo
new file mode 100644
index 0000000..5780b28
--- /dev/null
+++ b/ToDo
@@ -0,0 +1,4 @@
+- Check Gauss-Hermite integration points of the 'lme4' package
+- Check if 'ucminf' works again
+- Adopt prediction function for 'olmm' for prediction for in-sample individuals
+- Check nloptr package
\ No newline at end of file
diff --git a/inst/CITATION b/inst/CITATION
index b7b52e6..c465eed 100644
--- a/inst/CITATION
+++ b/inst/CITATION
@@ -5,29 +5,29 @@ p2 <- person("Gilbert", "Ritschard", email = "gilbert.ritschard@unige.ch")
 
 citEntry(entry="unpublished",
   author = personList(p1, p2),
-  title = "Coefficient-wise tree-based varying coefficient regression with vcrpart",
+  title = "Coefficient-Wise Tree-Based Varying Coefficient Regression with vcrpart",
   year = "2014",
   note = "Article in progress",
-  textVersion = "B\u00FCrgin R. and Ritschard G. (2014b), Coefficient-wise tree-based varying coefficient regression with vcrpart. Article in progress."
+  textVersion = "B\u00FCrgin, R. and G. Ritschard (2014c), Coefficient-Wise Tree-Based Varying Coefficient Regression with vcrpart. Article in progress."
 )
 
 citEntry(entry="unpublished",
   author = personList(p1, p2),
-  title = "Tree-based varying-coefficient regression for longitudinal ordinal responses",
+  title = "Tree-Based Varying-Coefficient Regression for Longitudinal Ordinal Responses",
   year = "2014",
   note = "Article in progress",
-  textVersion = "B\u00FCrgin R. and Ritschard G. (2014a), Tree-based varying coefficient regression for longitudinal ordinal responses. Article in progress."
+  textVersion = "B\u00FCrgin, R. and G. Ritschard (2014b), Tree-Based Varying Coefficient Regression for Longitudinal Ordinal Responses. Article in progress."
 )
 
 citEntry(entry="article",
   author = personList(p1, p2),
-  title = "A decorated parallel coordinate plot for categorical longitudinal data",
+  title = "A Decorated Parallel Coordinate Plot for Categorical Longitudinal Data",
   journal = "The American Statistician",
   year = "2014",
   volume = "68",
-  pages = "98-103",
+  pages = "98--103",
   number = "2",
-  textVersion="B\u00FCrgin, R. & Ritschard, G. (2014), A decorated parallel coordinate plot for categorical longitudinal data, The American Statistician 68(2), 98-103.",
+  textVersion="B\u00FCrgin, R. and G. Ritschard (2014a), A Decorated Parallel Coordinate Plot for Categorical Longitudinal Data, The American Statistician 68(2), 98-103.",
 )
 
 citFooter("See",sQuote("citation()"),"for citing R itself.")
diff --git a/man/PL.Rd b/man/PL.Rd
index f786124..5363b6e 100644
--- a/man/PL.Rd
+++ b/man/PL.Rd
@@ -58,9 +58,10 @@
 \source{Austrian Social Security Database (ASSD). The data set is also available 
   from \url{https://sites.google.com/site/rafaellalive/research}}  
 
-\references{Lalive, R. and Zweimueller, J. (2009), How does parental leave affect 
-  fertility and return to work? Evidence from two natural experiments. 
-  \emph{The Quarterly Journal of Economics}.
+\references{Lalive, R. and J. Zweimueller (2009), Does Parental Leave
+  Affect Fertility and Return-to-Work? Evidence from Two Natural
+  Experiments. \emph{The Quarterly Journal of Economics} \bold{124}(3),
+  1363--1402.
 } 
 
 \keyword{datasets}
diff --git a/man/contr-wsum.Rd b/man/contr-wsum.Rd
index 2c30535..5cd23ae 100644
--- a/man/contr-wsum.Rd
+++ b/man/contr-wsum.Rd
@@ -19,8 +19,8 @@ contr.wsum(x, weights = rep.int(1.0, length(x)))
 
 \details{
   Computes a contrast matrix similar to \command{\link{contr.sum}}. The
-  contrast for the last category are however weighted by the sum of
-  weights of each category.
+  reference category is however weighted by the sum of weights of the
+  other categories.
 }
 
 \value{A matrix with \code{nlevels(x)} rows and \code{nlevels(x)- 1}
diff --git a/man/fvcm-methods.Rd b/man/fvcm-methods.Rd
index e07da3a..5bd53b3 100644
--- a/man/fvcm-methods.Rd
+++ b/man/fvcm-methods.Rd
@@ -72,14 +72,14 @@
 \author{Reto Buergin}
 
 \references{
-  Leo Breiman (1996). Bagging Predictors. \emph{Machine Learning},
-  123--140
+  Breiman, L. (1996). Bagging Predictors. \emph{Machine Learning},
+  \bold{24}(2), 123--140.
   
-  Leo Breiman (2001). Random Forests. \emph{Machine Learning},
+  Breiman, L. (2001). Random Forests. \emph{Machine Learning},
   \bold{45}(1), 5--32.
-
-  T. Hastie, R. Tibshirani, J. Friedman (2001), The elements of
-  statistical learning, Springer.
+  
+  Hastie, T., R. Tibshirani and J. Friedman (2001), The Elements of 
+  Statistical Learning (2 ed.), Springer-Verlag.
 }
 
 \seealso{\command{\link{fvcm}}, \command{\link{tvcm-methods}}}
diff --git a/man/fvcm.Rd b/man/fvcm.Rd
index 54477ba..d53f215 100644
--- a/man/fvcm.Rd
+++ b/man/fvcm.Rd
@@ -7,10 +7,10 @@
 \alias{fvcglm}
 \alias{fvcglm_control}
 
-\title{Bagging and random forests based on \command{\link{tvcm}}}
+\title{Bagging and Random Forests based on \command{\link{tvcm}}}
 
 \description{
-  Bagging (Breiman, 1996) and random forest (Breiman, 2001) ensemble
+  Bagging (Breiman, 1996) and Random Forest (Breiman, 2001) ensemble
   algorithms for \command{\link{tvcm}}.
 }
 
@@ -70,8 +70,8 @@ fvcglm_control(maxstep = 10, folds = folds_control("subsampling", 5),
 }
 
 \details{
-  Implements the \emph{bagging} (Breiman, 1996) and \emph{random
-  forests} (Breiman, 2001) ensemble algorithms for
+  Implements the \emph{Bagging} (Breiman, 1996) and \emph{Random
+  Forests} (Breiman, 2001) ensemble algorithms for
   \command{\link{tvcm}}. The method consist in growing multiple trees by
   using \command{\link{tvcm}} and aggregating the fitted coefficient
   functions in the scale of the predictor function. To enable bagging,
@@ -102,14 +102,14 @@ fvcglm_control(maxstep = 10, folds = folds_control("subsampling", 5),
 \author{Reto Buergin}
 
 \references{
-  Leo Breiman (1996). Bagging Predictors. \emph{Machine Learning},
-  123--140
+  Breiman, L. (1996). Bagging Predictors. \emph{Machine Learning},
+  \bold{24}(2), 123--140.
   
-  Leo Breiman (2001). Random Forests. \emph{Machine Learning},
+  Breiman, L. (2001). Random Forests. \emph{Machine Learning},
   \bold{45}(1), 5--32.
 
-  T. Hastie, R. Tibshirani, J. Friedman (2001), The elements of
-  statistical learning, Springer.
+  Hastie, T., R. Tibshirani and J. Friedman (2001), The Elements of 
+  Statistical Learning (2 ed.), Springer-Verlag.
 }
 
 \seealso{\command{\link{fvcm-methods}}, \command{\link{tvcm}}, 
diff --git a/man/movie.Rd b/man/movie.Rd
index bfdef77..9d5d453 100644
--- a/man/movie.Rd
+++ b/man/movie.Rd
@@ -26,8 +26,9 @@
 
 \source{The data are tabulated in Hartzel et al. (2001).}
 
-\references{Hartzel, J., Agresti A. and Caffo, B. (2001). Multinomial
-  Logit Random Effect Models, \emph{Statistical Modelling} \bold{1}:
-  81--102} 
+\references{
+  Hartzel, J., A. Agresti and B. Caffo (2001). Multinomial Logit Random
+  Effect Models, \emph{Statistical Modelling} \bold{1}(2), 81--102.
+}  
 
 \keyword{datasets}
diff --git a/man/olmm-gefp.Rd b/man/olmm-gefp.Rd
index 6e59ce8..3e91041 100644
--- a/man/olmm-gefp.Rd
+++ b/man/olmm-gefp.Rd
@@ -18,7 +18,8 @@ estfun.olmm(x, predecor = FALSE, control = predecor_control(),
             nuisance = NULL, ...)
 
 predecor_control(impute = TRUE, seed = NULL, 
-                 symmetric = TRUE, reltol = 1e-6,
+                 symmetric = TRUE, center = FALSE,
+                 reltol = 1e-6,
                  maxit = 250L, minsize = 1L,
                  verbose = FALSE, silent = FALSE)
 
@@ -111,11 +112,12 @@ gefp.olmm(object, scores = NULL, order.by = NULL, subset = NULL,
 }
 
 \references{
-Zeileis A., Hornik K. (2007), Generalized M-Fluctuation Tests for Parameter
-Instability, \emph{Statistica Neerlandica}, \bold{61}, 488--508.
-
-Buergin R. and Ritschard G. (2014a), Tree-based varying coefficient 
-regression for longitudinal ordinal responses. Article in progress.
+  Zeileis A., Hornik K. (2007), Generalized M-Fluctuation Tests for
+  Parameter Instability, \emph{Statistica Neerlandica}, \bold{61}(4),
+  488--508. 
+  
+  Buergin R. and Ritschard G. (2014b), Tree-Based Varying Coefficient 
+  Regression for Longitudinal Ordinal Responses. Article in progress.
 }
 
 \author{Reto Buergin}
diff --git a/man/olmm-methods.Rd b/man/olmm-methods.Rd
index 9a945f7..1743c52 100644
--- a/man/olmm-methods.Rd
+++ b/man/olmm-methods.Rd
@@ -125,18 +125,18 @@
 \author{Reto Buergin}
 
 \references{
-  Agresti, A. (2010). \emph{Analysis of Ordinal Categorical Data}, 10
-  edn, Wiley.
+  Agresti, A. (2010). \emph{Analysis of Ordinal Categorical Data} (2
+  ed.), Wiley.
 
   Tutz, G. (2012). \emph{Regression for Categorical Data}, Cambridge
   Series in Statistical and Probabilistic Mathematics.
   
-  Li, C. and Sheperd, B. E. (2012). A new residual for ordinal
-  outcomes, \emph{Biometrika} \bold{99} (2): 437-480
+  Li, C. and B. E. Sheperd (2012). A New Residual for Ordinal
+  Outcomes, \emph{Biometrika}, \bold{99}(2), 437-480.
 
-  Bates D, Maechler M, Bolker BM and Walker S (2014). lme4: 
-  Linear mixed-effects models using Eigen and S4. Submitted to 
-  \emph{Journal of Statistical Software}
+  Bates, D., M. Maechler, B. M. Bolker and S. Walker (2014). lme4: 
+  Linear Mixed-Effects Models Using Eigen and S4. Submitted to 
+  \emph{Journal of Statistical Software}.
 }
 
 \seealso{\command{\link{olmm}}, \command{\link{predict.olmm}},
diff --git a/man/olmm-predict.Rd b/man/olmm-predict.Rd
index 443e299..e72e2d2 100644
--- a/man/olmm-predict.Rd
+++ b/man/olmm-predict.Rd
@@ -61,6 +61,12 @@
 
 \author{Reto Buergin}
 
+\references{
+  Skrondal, A., S. Rabe-Hesketh (2009). Prediction in multilevel
+  generalized linear models. \emph{Journal of the Royal Statistical
+  Society: Series A (Statistics in Society)}, \bold{172}(3), 659--687.  
+}
+
 \seealso{\command{\link{olmm}}, \command{\link{olmm-methods}}}
 
 \examples{
diff --git a/man/olmm.Rd b/man/olmm.Rd
index d81c28b..9ca7819 100644
--- a/man/olmm.Rd
+++ b/man/olmm.Rd
@@ -89,12 +89,10 @@ olmm(formula, data, family = cumulative(),
   
   For alternative fitting functions, see for example the 
   functions \code{clmm} of \pkg{ordinal},
-  \code{cumlogitRE} of package \pkg{glmmAK},
   \code{nplmt} of package \pkg{mixcat},
   \code{DPolmm} of package \pkg{DPpackage}, 
   \code{lcmm} of package \pkg{lcmm},
-  \code{MCMCglmm} of package \pkg{MCMCglmm},
-  \code{sabre} of package \pkg{sabreR} or
+  \code{MCMCglmm} of package \pkg{MCMCglmm} or
   \code{OrdinalBoost} of package \pkg{GMMBoost}. 
 
   The implementation adopts functions of the packages \pkg{statmod}
@@ -178,30 +176,30 @@ olmm(formula, data, family = cumulative(),
 \author{Reto Buergin}
 
 \references{
-  Agresti, A. (2010). \emph{Analysis of Ordinal Categorical Data}, 10
-  edn, Wiley.
+  Agresti, A. (2010). \emph{Analysis of Ordinal Categorical Data} (2
+  ed.), Wiley.
   
-  Hartzel, J., Agresti A. and Caffo, B. (2001). Multinomial Logit Random
-  Effect Models, \emph{Statistical Modelling} \bold{1}: 81--102
+  Hartzel, J., A. Agresti and B. Caffo (2001). Multinomial Logit Random
+  Effect Models, \emph{Statistical Modelling} \bold{1}(2), 81--102.
   
-  Hedeker, D. and Gibbons, R. (1994). A random-effects ordinal
-  regression model for multilevel analysis, \emph{Biometrics} \bold{20}
-  (4): 933--944
+  Hedeker, D. and R. Gibbons (1994). A Random-Effects Ordinal
+  Regression Model for Multilevel Analysis, \emph{Biometrics}
+  \bold{20}(4), 933--944. 
 
-  Tutz, G. and Hennevogl W. (1996). Random effects in ordinal regression 
-  models, \emph{Computational Statistics & Data Analysis} \bold{22} (5):
-  537--557
+  Tutz, G. and W. Hennevogl (1996). Random Effects in Ordinal Regression 
+  Models, \emph{Computational Statistics & Data Analysis} \bold{22}(5),
+  537--557. 
 
   Tutz, G. (2012). \emph{Regression for Categorical Data}, Cambridge
   Series in Statistical and Probabilistic Mathematics.
 
-  Frederick Novomestky (2012). matrixcalc: Collection of functions for 
-  matrix calculations. R package version 1.0-3. URL 
+  Novomestky, F. (2012). matrixcalc: Collection of Functions for 
+  Matrix Calculations. R package version 1.0-3. URL 
   \url{http://CRAN.R-project.org/package=matrixcalc}
 
-  Gordon Smyth, Yifang Hu, Peter Dunn, Belinda Phipson and Yunshun Chen 
-  (2014). statmod: Statistical Modeling. R package version 1.4.20. 
-  URL \url{http://CRAN.R-project.org/package=statmod}
+  Smyth, G., Y. Hu, P. Dunn, B. Phipson and Y. Chen (2014). statmod:
+  Statistical Modeling. R package version 1.4.20. URL
+  \url{http://CRAN.R-project.org/package=statmod} 
 }
 
 \seealso{\command{\link{olmm-methods}}, \command{\link{olmm_control}},
diff --git a/man/otsplot.Rd b/man/otsplot.Rd
index 2c72bd0..bc1fabe 100644
--- a/man/otsplot.Rd
+++ b/man/otsplot.Rd
@@ -116,13 +116,13 @@ otsplot_filter(method = c("minfreq", "cumfreq", "linear"), level = NULL)
 \author{Reto Buergin and Gilbert Ritschard}
 
 \references{
-  Reto Buergin and Gilbert Ritschard, G. (2014). A decorated parallel
-  coordinate plot for categorical longitudinal data, \emph{The American
-    Statistician} \bold{68}: 98-103
+  Buergin, R. and G. Ritschard (2014a). A Decorated Parallel Coordinate 
+  Plot for Categorical Longitudinal Data, \emph{The American
+  Statistician} \bold{68}(2): 98--103. 
 
-  Ross Ihaka, Paul Murrell, Kurt Hornik, Jason C. Fisher, Achim Zeileis (2013). 
+  Ihaka, R., P. Murrell, K. Hornik, J. C. Fisher and A. Zeileis (2013).
   colorspace: Color Space Manipulation. R package version 1.2-4. URL
-  \url{http://CRAN.R-project.org/package=colorspace}
+  \url{http://CRAN.R-project.org/package=colorspace}.
 }
 
 \examples{
diff --git a/man/poverty.Rd b/man/poverty.Rd
index 0022bde..4886a05 100644
--- a/man/poverty.Rd
+++ b/man/poverty.Rd
@@ -72,13 +72,13 @@
 
 \source{VLV survey, see also \url{http://cigev.unige.ch/recherches/vlv.html}}
 
-\references{Ludwig, C., Cavalli, S. and Oris,
-  M. \sQuote{Vivre/Leben/Vivere}: An interdisciplinary survey addressing 
+\references{Ludwig, C., S. Cavalli and M. Oris
+  \sQuote{Vivre/Leben/Vivere}: An interdisciplinary survey addressing
   progress and inequalities of ageing over the past 30 years in
-  Switzerland. \emph{Archives of Gerontology and Geriatrics}. 
+  Switzerland. \emph{Archives of Gerontology and Geriatrics}.  
   
-  Gabriel, R., Oris, M. Studer, M. and Baeriswyl, M. The persistance of
-  social stratification? Submitted.
+  Gabriel, R., M. Oris, M. Studer and M. Baeriswyl (2014). The
+  Persistance of Social Stratification? Submitted.
 } 
 
 \keyword{datasets}
diff --git a/man/schizo.Rd b/man/schizo.Rd
index 0fb9de2..b342428 100644
--- a/man/schizo.Rd
+++ b/man/schizo.Rd
@@ -34,7 +34,7 @@
 
 \source{\url{http://tigger.uic.edu/~hedeker/ml.html}}
 
-\references{Hedeker, D. and Gibbons, R. (2006). Longitudinal Data
+\references{Hedeker, D. and R. Gibbons (2006). Longitudinal Data
   Analysis. Wiley, Palo Alto, CA.}
 
 \keyword{datasets}
diff --git a/man/tvcglm.Rd b/man/tvcglm.Rd
index 3d2f1a3..c02224f 100644
--- a/man/tvcglm.Rd
+++ b/man/tvcglm.Rd
@@ -8,14 +8,14 @@
 
 \description{The \command{\link{tvcglm}} function implements the
   tree-based varying coefficient regression algorithm for generalized
-  linear models introduced by Buergin and Ritschard (2014b). The
+  linear models introduced by Buergin and Ritschard (2014c). The
   algorithm approximates varying coefficients by piecewise constant
   functions using recursive partitioning, i.e., it estimates the
-  coefficients of the model separately for strata of the value space of
-  partitioning variables. The special feature of the algorithm is to
-  assign each varying coefficient a partition, which enhances the
-  possibilities for model specification and to select moderator
-  variables individually by coefficient
+  selected coefficients individually by strata of the value space of
+  partitioning variables. The special feature of the provided algorithm
+  is that it allows building for each varying coefficient an individual
+  partition, which enhances the possibilities for model specification
+  and to select partitioning variables individually by coefficient.
 }
 
 \usage{
@@ -32,11 +32,19 @@ tvcglm_control(minsize = 30, mindev = 2.0,
 \arguments{
   \item{formula}{a symbolic description of the model to fit, e.g.,
 
-    \code{y ~ vc(z1, \ldots, zL, by = x1 + \ldots + xP) + re(1|id)}
+    \code{y ~ vc(z1, z2, z3) + vc(z1, z2, by = x1) + vc(z2, z3, by = x2)}
     
-    where \code{vc} term specifies the varying fixed coefficients. Only
-    one such \code{vc} term is allowed. For details,
-    see \command{\link{olmm}} and \command{\link{vcrpart-formula}}.}
+    where the \code{vc} terms specify the varying fixed
+    coefficients. The unnamed arguments within \code{vc} terms are
+    interpreted as partitioning variables (i.e., moderators). The
+    \code{by} argument specifies the associated predictor variable. If
+    no such predictor variable is specified (e.g., see the first term in
+    the above example formula), the \code{vc} term is interpreted as a
+    varying intercept, i.e., an nonparametric estimate of the direct
+    effect of the partitioning variables. For details, see
+    \command{\link{vcrpart-formula}}. Note that the global intercept may
+    be removed by a \code{-1} term, according to the desired
+    interpretation of the model.} 
   \item{family}{the model family. An object of class
     \command{\link{family.olmm}}.}   
   \item{data}{a data frame containing the variables in the model.}
@@ -75,29 +83,30 @@ tvcglm_control(minsize = 30, mindev = 2.0,
 }
 
 \details{
-  The TVCGLM algorithm uses two stages. The first stage (partitioning)
-  builds too overly fine partitions and the second stage (pruning)
-  selects the best-sized partitions by collapsing inner nodes. For the
-  second stage, which is automatically processed, we refer to
-  \command{\link{tvcm-assessment}}. The partitioning stage iterates the
-  following steps:
+  \command{\link{tvcglm}} processes two stages. The first stage, called 
+  partitioning stage, builds overly fine partitions for each \code{vc}
+  term; the second stage, called pruning stage, selects the best-sized
+  partitions by collapsing inner nodes. For details on the pruning
+  stage, see \command{\link{tvcm-assessment}}. The partitioning stage
+  iterates the following steps: 
 
   \enumerate{
     \item Fit the current generalized linear model
 
     \code{y ~ NodeA:x1 + \ldots + NodeK:xK}
     
-    with \command{\link{glm}}, where \code{NodeK} is a categorical
-    variable with terminal node labels \code{1}, \ldots for the
-    \eqn{K}-th varying coefficient. 
+    with \command{\link{glm}}, where \code{Nodek} is a categorical
+    variable with terminal node labels for the \eqn{k}-th varying
+    coefficient. 
 
-    \item Search and globally optimal split among the candidate
-    splits by exhaustive -2 likelihood training error grid search,
-    by cycling through all partitions, nodes and moderator variables. 
+    \item Search the globally best split among the candidate splits by
+    an exhaustive -2 likelihood training error search that cycles
+    through all possible splits.
 
-    \item If the -2 likelihood training error reduction through the best
+    \item If the -2 likelihood training error reduction of the best
     split is smaller than \code{mindev} or there is no candidate split
-    satisfying the minimum node size \code{minsize}, stop the algorithm.
+    satisfying the minimum node size \code{minsize}, stop the
+    algorithm. 
 
     \item Else incorporate the best split and repeat the procedure.
   }
@@ -108,6 +117,10 @@ tvcglm_control(minsize = 30, mindev = 2.0,
   (a minimum node size of 30) and \code{mindev = 2} (the training error
   reduction of the best split must be larger than two to continue).
 
+  The algorithm implements a number of split point reduction methods to
+  decrease the computational complexity. See the arguments
+  \code{maxnomsplit}, \code{maxordsplit} and \code{maxnumsplit}.
+
   The algorithm can be seen as an extension of CART (Breiman et. al.,
   1984) and PartReg (Wang and Hastie, 2014), with the new feature that
   partitioning can be processed coefficient-wise.
@@ -117,20 +130,22 @@ tvcglm_control(minsize = 30, mindev = 2.0,
 }
 
 \references{
-  Breiman, L., Friedman, J.H., Olshen, R.A. and Stone, C.J. (1984) 
-  \emph{Classification and Regression Trees}. Wadsworth.
+  Breiman, L., J. H. Friedman, R. A. Olshen and C.J. Stone (1984)  
+  \emph{Classification and Regression Trees}. Volume 19. The Wadsworth
+  Statistics / Probability Series.
   
   Wang, J. C., Hastie, T. (2014), Boosted Varying-Coefficient
   Regression Models for Product Demand Prediction, \emph{Journal of
-    Computational and Graphical Statistics}, \bold{23}, 361--382.
+    Computational and Graphical Statistics}, \bold{23}(2), 361--382. 
   
-  Buergin R. and Ritschard G. (2014b). Coefficient-wise tree-based
-  varying coefficient regression with vcrpart. Article in progress.
+  Buergin R. and G. Ritschard (2014c). Coefficient-Wise Tree-Based
+  Varying Coefficient Regression with vcrpart. Article in progress.
 }
 
 \seealso{\command{\link{tvcm_control}}, \command{\link{tvcm-methods}},
   \command{\link{tvcm-plot}}, \command{\link{tvcm-plot}},
-  \command{\link{tvcm-assessment}}, \command{\link{glm}}}
+  \command{\link{tvcm-assessment}}, \command{\link{fvcglm}},
+  \command{\link{glm}}} 
 
 \examples{
 ## ------------------------------------------------------------------- #  
diff --git a/man/tvcm-control.Rd b/man/tvcm-control.Rd
index 8bf6af5..dc81caa 100644
--- a/man/tvcm-control.Rd
+++ b/man/tvcm-control.Rd
@@ -14,11 +14,11 @@ tvcm_control(minsize = 30, mindev = ifelse(sctest, 0.0, 2.0),
              trim = 0.1, estfun.args = list(), nimpute = 5, 
              maxnomsplit = 5, maxordsplit = 9, maxnumsplit = 9,
              maxstep = 1e3, maxwidth = 1e9, maxdepth = 1e9,
-             lossfun = neglogLik2, ooblossfun = NULL,
+             lossfun = neglogLik2, ooblossfun = NULL, fast = TRUE,
              cp = 0.0, dfpar = 0.0, dfsplit = 1.0,
              cv = !sctest, folds = folds_control("kfold", 5),
              prune = cv, papply = mclapply, papply.args = list(),
-             center = TRUE, seed = NULL, verbose = FALSE, ...)
+             center = fast, seed = NULL, verbose = FALSE, ...)
 }
 
 \arguments{
@@ -48,6 +48,11 @@ tvcm_control(minsize = 30, mindev = ifelse(sctest, 0.0, 2.0),
   \item{ooblossfun}{a loss function that defines how to compute the
     validation error during cross-validation. The function will be
     assigned to the \code{fun} argument of \command{\link{oobloss}}.}
+  \item{fast}{logical scalar. Whether the approximative model should be
+    used to search for the next split. The approximative search model
+    uses the fitted values of the current model as offsets and estimates
+    only the coefficients of the added split. If \code{FALSE}, the
+    accurate search model is used.}
   \item{cp}{numeric scalar. The penalty to be multiplied with the
     complexity of the model during partitioning. The complexity of the
     model is defined as the number of coefficients times \code{dfpar}
diff --git a/man/tvcm-cv.Rd b/man/tvcm-cv.Rd
index 7887ec7..36d15f9 100644
--- a/man/tvcm-cv.Rd
+++ b/man/tvcm-cv.Rd
@@ -3,14 +3,13 @@
 \alias{tvcm-assessment}
 \alias{prune}
 \alias{prune.tvcm}
-\alias{folds_control}
+\alias{prunepath}
+\alias{prunepath.tvcm}
 \alias{cvloss}
 \alias{cvloss.tvcm}
-\alias{loss}
-\alias{loss.tvcm}
+\alias{folds_control}
 \alias{oobloss}
 \alias{oobloss.tvcm}
-\alias{print.cvloss.tvcm}
 \alias{plot.cvloss.tvcm}
 
 \title{Model selection utility functions for \command{\link{tvcm}} objects.} 
@@ -25,15 +24,15 @@
 \method{prune}{tvcm}(tree, cp = NULL, alpha = NULL, maxstep = NULL,
       terminal = NULL, original = FALSE, ...)
 
+\method{prunepath}{tvcm}(tree, steps = 1L, ...)
+
+\method{cvloss}{tvcm}(object, folds = folds_control(), ...)
+
 folds_control(type = c("kfold", "subsampling", "bootstrap"),
       K = ifelse(type == "kfold", 5, 30),
       prob = 0.5, weights = c("case", "freq"),
       seed = NULL)
 
-\method{cvloss}{tvcm}(object, folds = folds_control(), ...)
-
-\method{print}{cvloss.tvcm}(x, ...)
-
 \method{plot}{cvloss.tvcm}(x, legend = TRUE, details = TRUE, ...)
 
 \method{oobloss}{tvcm}(object, newdata = NULL, weights = NULL,
@@ -42,29 +41,6 @@ folds_control(type = c("kfold", "subsampling", "bootstrap"),
 
 \arguments{
   \item{object, tree}{an object of class \command{\link{tvcm}}.}
-  \item{x}{an object of class \code{cvloss.tvcm} as produced by
-    \command{\link{cvloss}}.} 
-  \item{type}{character string. The type of sampling scheme to be used
-    to divide the data of the input model in a learning and a validation
-    set.}    
-  \item{K}{integer scalar. The number of folds.} 
-  \item{prob}{numeric between 0 and 1. The probability for the
-    \code{"subsampling"} cross-validation scheme.}
-  \item{weights}{for \command{\link{folds_control}}, a character that
-    defines whether the weights of \code{object} are case weights or
-    frequencies of cases; for \command{\link{oobloss}}, a numeric vector
-    of weights corresponding to the rows of \code{newdata}.} 
-  \item{seed}{an numeric scalar that defines the seed.}
-  \item{folds}{a list with control arguments as produced by
-    \command{\link{folds_control}}.}
-  \item{legend}{logical scalar. Whether a legend should be added.}
-  \item{details}{logical scalar. Whether the foldwise validation errors 
-    should be shown.} 
-  \item{fun}{the loss function for the validation sets. By default, the
-    (possibly weighted) mean of the deviance residuals as defined by the
-    \command{\link{family}} of the fitted \code{object} is applied.}
-  \item{newdata}{a data.frame of out-of-bag data (including the response
-    variable). See also \command{\link{predict.tvcm}}.}
   \item{cp}{numeric scalar. The complexity parameter to be cross-validated 
     resp. the penalty with which the model should be pruned.}
   \item{alpha}{numeric significance level. Represents the stopping
@@ -78,30 +54,55 @@ folds_control(type = c("kfold", "subsampling", "bootstrap"),
     be equal the number of partitions.}
   \item{original}{logical scalar. Whether pruning should be based on the
     trees from partitioning rather than on the current trees.}
+  \item{steps}{integer vector. The iteration steps from which
+    information should be extracted.}
+  \item{folds}{a list with control arguments as produced by
+    \command{\link{folds_control}}.}
+  \item{type}{character string. The type of sampling scheme to be used
+    to divide the data of the input model in a learning and a validation
+    set.}
+  \item{K}{integer scalar. The number of folds.} 
+  \item{weights}{for \command{\link{folds_control}}, a character that
+    defines whether the weights of \code{object} are case weights or
+    frequencies of cases; for \command{\link{oobloss}}, a numeric vector
+    of weights corresponding to the rows of \code{newdata}.}
+  \item{prob}{numeric between 0 and 1. The probability for the
+    \code{"subsampling"} cross-validation scheme.}
+  \item{seed}{an numeric scalar that defines the seed.}  
+  \item{x}{an object of class \code{cvloss.tvcm} as produced by
+    \command{\link{cvloss}}.}
+  \item{legend}{logical scalar. Whether a legend should be added.}
+  \item{details}{logical scalar. Whether the foldwise validation errors 
+    should be shown.}
+  \item{newdata}{a data.frame of out-of-bag data (including the response
+    variable). See also \command{\link{predict.tvcm}}.}
+  \item{fun}{the loss function for the validation sets. By default, the
+    (possibly weighted) mean of the deviance residuals as defined by the
+    \command{\link{family}} of the fitted \code{object} is applied.}
   \item{...}{other arguments to be passed.} 
 }
 
-\details{By default, \command{\link{tvcm}} is a two stage procedure that 
-  first grows overly large trees and second selects the best-sized trees by 
-  pruning. The here presented functions may be interesting for advanced users 
-  who want to process the model selection stage separately.
+\details{\command{\link{tvcglm}} and \command{\link{tvcm}} processe
+  tree-size selection by default. The functions could be interesting for
+  advanced users. 
 
-  In normal practice, the \command{\link{prune}} function is used to collapse 
-  inner nodes of the tree structures by the tuning parameter \code{cp}. The 
-  aim of pruning by \code{cp} is to collapse inner nodes to minimize the 
-  cost-complexity criterion
+  The \command{\link{prune}} function is used to collapse inner nodes of
+  the tree structures by the tuning parameter \code{cp}. The aim of
+  pruning by \code{cp} is to collapse inner nodes to minimize the
+  cost-complexity criterion 
 
   \deqn{error(cp) = error(tree) + cp * complexity(tree)}
  
-  whereby, the training error \eqn{error(tree)} is defined by \code{lossfun} 
-  and \eqn{complexity(tree)} is defined as the total number of coefficients times
-  \code{dfpar} plus the total number of splits times \code{dfsplit}. The function
-  \code{lossfun} and the parameters \code{dfpar} and \code{dfsplit} are defined 
-  by the \code{control} argument of \command{\link{tvcm}}, see also 
-  \command{\link{tvcm_control}}. By default, \eqn{error(tree)} is minus two 
-  times the total likelihood of the model and \eqn{complexity(tree)} the number 
-  of splits. The minimization of \eqn{error(cp)} is implemented by the 
-  following iterative backward-stepwise algorithm 
+  where the training error \eqn{error(tree)} is defined by
+  \code{lossfun} and \eqn{complexity(tree)} is defined as the total
+  number of coefficients times \code{dfpar} plus the total number of
+  splits times \code{dfsplit}. The function \code{lossfun} and the
+  parameters \code{dfpar} and \code{dfsplit} are defined  by the
+  \code{control} argument of \command{\link{tvcm}}, see also
+  \command{\link{tvcm_control}}. By default, \eqn{error(tree)} is minus
+  two times the total likelihood of the model and \eqn{complexity(tree)}
+  the number of splits. The minimization of \eqn{error(cp)} is
+  implemented by the following iterative backward-stepwise algorithm 
 
   \enumerate{
   \item fit all \code{subtree} models that collapse one inner node of the 
@@ -116,7 +117,7 @@ folds_control(type = c("kfold", "subsampling", "bootstrap"),
   }
 
   The penalty \code{cp} is generally unknown and is estimated adaptively from 
-  the data. \command{\link{cvloss}} implements the cross-validation 
+  the data. The \command{\link{cvloss}} function implements the cross-validation 
   method to do this. \command{\link{cvloss}} repeats for each fold the following 
   steps
  
@@ -127,18 +128,26 @@ folds_control(type = c("kfold", "subsampling", "bootstrap"),
     \code{cp} the average validation error.
   }
 
-  Doing so yields for each fold a  sequence of values for \code{cp} and a 
-  sequence of average validation errors. The obtained sequences for \code{cp} 
-  are combined to a fine grid and the average validation error is averaged 
-  correspondingly. From these two sequences we choose the \code{cp} that 
-  minimizes the validation error. Notice that the average validation error
-  is computed as the total prediction error of the validation set divided  
-  by the sum of validation set weights. See also the argument \code{ooblossfun} in 
-  \command{\link{tvcm_control}} and the function \command{\link{oobloss}}.
-
+  Doing so yields for each fold a sequence of values for \code{cp} and
+  a sequence of average validation errors. These sequences are then
+  combined to a finer grid and the average validation error is averaged 
+  correspondingly. From these two sequences we choose the \code{cp}
+  value that minimizes the validation error. Notice that the average
+  validation error is computed as the total prediction error of the
+  validation set divided  by the sum of validation set weights. See also
+  the argument \code{ooblossfun} in \command{\link{tvcm_control}} and
+  the function \command{\link{oobloss}}.
+
+  The \command{\link{prunepath}} function can be used to backtrack the
+  pruning algorithm. By default, it shows the results from collapsing
+  inner nodes in the first iteration. The interesting iteration(s) can
+  be selected by the \code{steps} argument. The output shows several
+  information on the performances when collapsing inner nodes. The node
+  labels shown in the output refer to the initial tree.
+  
   The function \command{\link{folds_control}} is used to specify the 
   cross-validation scheme, where a random 5-fold cross-validation scheme 
-  is set as the default. Alternatives are \code{type = "subsampling"} 
+  is used by default. Alternatives are \code{type = "subsampling"} 
   (random draws without replacement) and \code{type = "bootstrap"} (random 
   draws with replacement). For 2-stage models (with random-effects) 
   fitted by \command{\link{olmm}}, the subsets are based on subject-wise 
@@ -183,11 +192,12 @@ folds_control(type = c("kfold", "subsampling", "bootstrap"),
 }
 
 \references{
-  Breiman, L., Friedman, J.H., Olshen, R.A. and Stone, C.J. (1984) 
-  \emph{Classification and Regression Trees}. Wadsworth.
+  Breiman, L., J. H. Friedman, R. A. Olshen and C.J. Stone (1984)  
+  \emph{Classification and Regression Trees}. Volume 19. The Wadsworth
+  Statistics / Probability Series.
 
-  T. Hastie, R. Tibshirani, J. Friedman (2001), The elements of
-  statistical learning, Springer.
+  Hastie, T., R. Tibshirani and J. Friedman (2001), The Elements of 
+  Statistical Learning (2 ed.), Springer-Verlag.
 }
 
 \seealso{
@@ -216,6 +226,13 @@ cv <- cvloss(model, folds = folds_control(type = "kfold", K = 2, seed = 1))
 cv
 plot(cv)
 
+## prune model with estimated 'cp'
+model.p <- prune(model, cp = cv$cp.hat)
+
+
+## backtrack pruning
+prunepath(model.p, steps = 1:3)
+
 ## out-of-bag error
 oobloss(model, newdata = vcrpart_2[76:100,])
 
diff --git a/man/tvcm-methods.Rd b/man/tvcm-methods.Rd
index ed7df9d..5ce3242 100644
--- a/man/tvcm-methods.Rd
+++ b/man/tvcm-methods.Rd
@@ -14,8 +14,6 @@
 \alias{neglogLik2.tvcm}
 \alias{nobs.tvcm}
 \alias{predict.tvcm}
-\alias{prunepath}
-\alias{prunepath.tvcm}
 \alias{print.tvcm}
 \alias{ranef.tvcm}
 \alias{resid.tvcm}
@@ -53,8 +51,6 @@
 \method{splitpath}{tvcm}(tree, steps = 1L,
          details = FALSE, ...)
 
-\method{prunepath}{tvcm}(tree, steps = 1L, ...)
-
 \method{summary}{tvcm}(object, ...)
 
 \method{width}{tvcm}(x, ...)
@@ -91,11 +87,12 @@
   observation. In cases of multiple \command{\link{vc}} terms for the same
   predictor, the coefficients are summed up.
 
-  \command{\link{splitpath}} and \command{\link{prunepath}} are new
-  methods to trace the splitting resp. pruning procedures. They shows
-  several information, such as the loss reduction of new splits during
-  partitioning or the loss reduction of collapsing an inner node when
-  pruning. 
+  The \command{\link{splitpath}} function allows to backtrack the
+  partitioning procedure. By default, it shows which split was chosen in
+  the first iteration. The interesting iteration(s) can be selected by
+  the \code{steps} argument. With \code{details = TRUE} it is also
+  possible to backtrack the coefficient constancy tests and/or the loss
+  reduction statistics.  
 
   \command{\link{summary}} computes summary statistics of the fitted model, 
   including the estimated coefficients. The varying coefficient are printed
diff --git a/man/tvcm-plot.Rd b/man/tvcm-plot.Rd
index b7cd474..c66e0d2 100644
--- a/man/tvcm-plot.Rd
+++ b/man/tvcm-plot.Rd
@@ -125,8 +125,8 @@ panel_coef(object, parm = NULL,
 }
 
 \references{
-  T. Hastie, R. Tibshirani, J. Friedman (2001), The elements of
-  statistical learning, Springer.
+  Hastie, T., R. Tibshirani and J. Friedman (2001), The Elements of 
+  Statistical Learning (2 ed.), Springer-Verlag.
 }
 
 \examples{
diff --git a/man/tvcm.Rd b/man/tvcm.Rd
index 29daf83..f7cecea 100644
--- a/man/tvcm.Rd
+++ b/man/tvcm.Rd
@@ -81,25 +81,24 @@ tvcm(formula, data, fit, family,
 }
 
 \references{
-  Zeileis, A., Hothorn, T., and Hornik, K. (2008). Model-Based
+  Zeileis, A., T. Hothorn, and K. Hornik (2008). Model-Based
   Recursive Partitioning. \emph{Journal of Computational and Graphical
     Statistics}, \bold{17}(2), 492--514.
 
-  Wang, J. C., Hastie, T. (2014), Boosted Varying-Coefficient
+  Wang, J. C. and T. Hastie (2014), Boosted Varying-Coefficient
   Regression Models for Product Demand Prediction, \emph{Journal of
-    Computational and Graphical Statistics}, \bold{23}, 361--382.
+    Computational and Graphical Statistics}, \bold{23}(2), 361--382.
   
-  Torsten Hothorn, Achim Zeileis (2014). partykit: A Modular Toolkit
-  for Recursive Partytioning in R. Working Paper 2014-10. \emph{Working 
-  Papers in Economics and Statistics, Research Platform Empirical and
-  Experimental Economics, Universitaet Innsbruck}. URL
-  \url{http://EconPapers.RePEc.org/RePEc:inn:wpaper:2014-10} 
+  Hothorn, T. and A. Zeileis (2014). partykit: A Modular Toolkit
+  for Recursive Partytioning in R. In \emph{Working Papers in Economics
+  and Statistics, Research Platform Empirical and Experimental
+  Economics}, Number 2014-10. Universitaet Innsbruck.
   
-  Buergin R. and Ritschard G. (2014a), Tree-based varying coefficient
-  regression for longitudinal ordinal responses. Article in progress. 
+  Buergin R. and G. Ritschard (2014b), Tree-Based Varying Coefficient
+  Regression for Longitudinal Ordinal Responses. Article in progress. 
   
-  Buergin R. and Ritschard G. (2014b), Coefficient-wise tree-based
-  varying coefficient regression with vcrpart. Article in progress.
+  Buergin R. and G. Ritschard (2014c), Coefficient-Wise Tree-Based
+  Varying Coefficient Regression with vcrpart. Article in progress.
 }
 
 \seealso{\command{\link{tvcolmm}}, \command{\link{tvcglm}},
diff --git a/man/tvcolmm.Rd b/man/tvcolmm.Rd
index b4d4b7b..7512fad 100644
--- a/man/tvcolmm.Rd
+++ b/man/tvcolmm.Rd
@@ -8,9 +8,9 @@
 
 \description{The \command{\link{tvcolmm}} function implements the
   tree-based longitudinal varying coefficient regression algorithm
-  proposed in Buergin and Ritschard (2014a). The algorithm approximates
+  proposed in Buergin and Ritschard (2014b). The algorithm approximates
   varying fixed coefficients in the cumulative logit mixed model by a
-  (multivariate) piecewise constant functions using recursive
+  (multivariate) piecewise constant function using recursive
   partitioning, i.e., it estimates the fixed effect component of the
   model separately for strata of the value space of partitioning
   variables. 
@@ -22,7 +22,8 @@ tvcolmm(formula, data, family = cumulative(),
         control = tvcolmm_control(), ...)
 
 tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
-                maxnomsplit = 5, maxordsplit = 9, maxnumsplit = 9,
+                maxnomsplit = 5, maxordsplit = 9,
+                maxnumsplit = 9, fast = TRUE,
                 trim = 0.1, estfun.args = list(), nimpute = 5,
                 seed = NULL, ...)
 }
@@ -30,15 +31,18 @@ tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
 \arguments{
   \item{formula}{a symbolic description of the model to fit, e.g.,
     
-    \code{y ~ -1 + vc(z1, \ldots, zL, by = x1 + \ldots + xP, intercept =
-      TRUE) + re(1|id)}
+    \code{y ~ -1 + vc(z1, \ldots, zL, by = x1 + \ldots + xP, intercept = TRUE) + re(1|id)}
     
     where \code{vc} term specifies the varying fixed coefficients. Only
     one such \code{vc} term is allowed with
-    \command{\link{tvcolmm}}. The example formula removes the global
-    intercept and adds a locally varying intercept by setting
-    \code{intercept = TRUE} in the \code{vc} term. For details, see
-    \command{\link{olmm}} and \command{\link{vcrpart-formula}}.} 
+    \command{\link{tvcolmm}} (in contrast to command{\link{tvcglm}}
+    where multiple \code{vc} terms can be specified). The above example 
+    formula removes the global intercepts and adds locally varying
+    intercepts, by adding a \code{-1} term and specfiying \code{intercept
+    = TRUE} in the \code{vc} term. If varying intercepts are desired, we
+    recommend to always remove the global intercepts. For more details on
+    the formula specification, see \command{\link{olmm}} and
+    \command{\link{vcrpart-formula}}.}   
   \item{family}{the model family. An object of class
     \command{\link{family.olmm}}.}    
   \item{data}{a data frame containing the variables in the model.}
@@ -62,7 +66,10 @@ tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
   \item{minsize}{numeric scalar. The minimum sum of weights in terminal
     nodes.}
   \item{maxnomsplit, maxordsplit, maxnumsplit}{integer scalars for split
-    candidate reduction. See \command{\link{tvcm_control}}}
+    candidate reduction. See \command{\link{tvcm_control}}.}
+  \item{fast}{logical scalar. Whether the approximative model should be
+    used to search for the next split. See
+    \command{\link{tvcm_control}}.}
   \item{trim}{numeric between 0 and 1. Specifies the trimming parameter
     in coefficient constancy tests for continuous partitioning
     variables. See also the argument \code{from} of function
@@ -79,7 +86,7 @@ tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
 }
 
 \details{
-  The TVCOLMM algorithm iterates the following steps:
+  The \command{\link{tvcolmm}} function iterates the following steps:
 
   \enumerate{
     \item Fit the current mixed model
@@ -89,8 +96,8 @@ tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
     with \command{\link{olmm}}, where \code{Node} is a categorical
     variable with terminal node labels \code{1}, \ldots, \code{M}. 
 
-    \item Test for the constancy of the fixed effects \code{Node:x1,
-    \ldots} separately for each moderator \code{z1}, \ldots, \code{zL}
+    \item Test the constancy of the fixed effects \code{Node:x1,
+    \ldots}, separately for each moderator \code{z1}, \ldots, \code{zL}
     in each node \code{1}, \ldots, \code{M}. This yields \code{L} times
     \code{M} (possibly Bonferroni corrected) \eqn{p}-values for
     rejecting coefficient constancy.
@@ -99,7 +106,7 @@ tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
     then select the node and the variable corresponding to the minimum
     \eqn{p}-value. Search and incorporate the optimal
     among the candidate splits in the selected node and variable by
-    exhaustive likelihood maximization grid search.
+    exhaustive likelihood search.
     
     \item Else if minimum \eqn{p}-value is larger than \code{alpha},
     stop the algorithm and return the current model.
@@ -108,7 +115,7 @@ tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
   The implemented coefficient constancy tests used for node and variable
   selection (step 2) are based on the M-fluctuation tests of Zeileis and
   Hornik (2007), using the observation scores of the fitted mixed
-  model. These observation scores can be extracted by
+  model. The observation scores can be extracted by
   \command{\link{estfun.olmm}} for models fitted with
   \command{\link{olmm}}. To deal with intra-individual correlations
   between such observation scores, the \command{\link{estfun.olmm}}
@@ -122,42 +129,52 @@ tvcolmm_control(alpha = 0.05, bonferroni = TRUE, minsize = 50,
   the technique of Hajjem et. al (2011) and Sela and Simonoff (2012) to
   incorporate regression trees into mixed models.
 
+  For the exhaustive search, the algorithm implements a number of split
+  point reduction methods to decrease the computational complexity. See
+  the arguments \code{maxnomsplit}, \code{maxordsplit} and
+  \code{maxnumsplit}. By default, the algorithm also uses the
+  approximative search model approach proposed in Buergin and Ritschard
+  (2014c). To disable this option to use the original algorithm, set
+  \code{fast = FALSE} in \command{\link{tvcolmm_control}}.
+
   Special attention is given to varying intercepts, i.e. the terms that
   account for the direct effects of the moderators. A common
   specification is 
 
-  \code{y ~ -1 + vc(z1, \ldots, zL, by = x1 + \ldots + xP, intercept =
-    TRUE) + re(1 + w1 + \ldots |id)}
+  \code{y ~ -1 + vc(z1, \ldots, zL, by = x1 + \ldots + xP, intercept = TRUE) + re(1 + w1 + \ldots |id)}
 
-  Doing so replaces the globale intercept by local intercepts.
+  Doing so replaces the globale intercept by local intercepts. As
+  mentioned, if a varying intercepts are desired, we recommend to always
+  remove the global intercept. 
  }
 
 \value{An object of class \command{\link{tvcm}}
 }
 
 \references{
-   Zeileis, A., Hothorn, T., and Hornik, K. (2008). Model-Based
-   Recursive Partitioning. \emph{Journal of Computational and Graphical
-   Statistics}, \bold{17}(2), 492--514. 
-
-   Zeileis, A., Hornik, K. (2007), Generalized M-Fluctuation Tests for
-   Parameter Instability, \emph{Statistica Neerlandica}, \bold{61},
-   488--508. 
-
-   Buergin R. and Ritschard G. (2014a), Tree-based varying coefficient
-   regression for longitudinal ordinal responses. Article in progress.
+  Zeileis, A., T. Hothorn, and K. Hornik (2008). Model-Based
+  Recursive Partitioning. \emph{Journal of Computational and Graphical
+    Statistics}, \bold{17}(2), 492--514. 
+
+  Zeileis A., Hornik K. (2007), Generalized M-Fluctuation Tests for
+  Parameter Instability, \emph{Statistica Neerlandica}, \bold{61}(4),
+  488--508.
+  
+  Buergin R. and Ritschard G. (2014b), Tree-Based Varying Coefficient 
+  Regression for Longitudinal Ordinal Responses. Article in progress.
       
-   R. Sela and J. S. Simonoff (2012). RE-EM trees: a data mining
-   approach for longitudinal and clustered data, \emph{Machine Learning}
-   \bold{86}, 169--207. 
-
-   A. Hajjem, F. Bellavance and D. Larocque (2011), Mixed effects
-   regression trees for clustered data, \emph{Statistics & Probability
-   Letters} \bold{81}, 451--459.
+  Sela R. and J. S. Simonoff (2012). RE-EM trees: A Data Mining
+  Approach for Longitudinal and Clustered data, \emph{Machine Learning}
+  \bold{86}(2), 169--207. 
+  
+  A. Hajjem, F. Bellavance and D. Larocque (2011), Mixed Effects
+  Regression Trees for Clustered Data, \emph{Statistics & Probability
+    Letters} \bold{81}(4), 451--459.
 }
 
 \seealso{\command{\link{tvcm_control}}, \command{\link{tvcm-methods}},
-  \command{\link{tvcm-plot}}, \command{\link{glm}}}
+  \command{\link{tvcm-plot}}, \command{\link{fvcolmm}},
+  \command{\link{olmm}}} 
 
 \examples{
 ## ------------------------------------------------------------------- # 
diff --git a/src/Makevars b/src/Makevars
index 8d28f91..9542baf 100644
--- a/src/Makevars
+++ b/src/Makevars
@@ -1 +1,2 @@
-PKG_LIBS = ${LAPACK_LIBS} ${BLAS_LIBS} $(FLIBS)
\ No newline at end of file
+
+PKG_LIBS = $(LAPACK_LIBS) $(BLAS_LIBS) $(FLIBS)