Show how participants move between clusters as k increases

Answers "as clusters are added, is an existing cluster being split, or are participants reshuffled?" For each model family, participants are followed from the solution with k clusters to the solution with k + 1 clusters and shown as an alluvial plot, with a transition table summarizing each step.

Usage

PlotClusterSplits(
  object,
  models = NULL,
  k_range = NULL,
  title = "Where participants go as clusters are added"
)

Arguments

object

An exploratory model from CreateClusterModel_SOM_MClust(), or a data frame whose columns are cluster assignments for the same participants at increasing k (column names such as "k2", "k3"; one row per participant; treated as a single model family).

models

Optional integer vector of model families to show (SOM + Mclust objects only). Defaults to every family with at least two consecutive k.

k_range

Optional integer vector restricting the k shown.

title

Plot title.

Value

A list with plot (a ggplot) and transitions (a tibble with one row per model and step).

Details

Cluster labels are arbitrary at every k, so each solution's clusters are reordered to follow their largest parent in the previous solution; this keeps a cluster that is carried forward in the same vertical position.

For each step from k to k + 1 the transition table reports:

• ParentPurity: the share of participants whose cluster at k + 1 came mostly from their own cluster at k (each cluster at k + 1 is credited to its largest parent). Values near 1 mean new clusters are carved out of existing ones (a hierarchical split); lower values mean participants are reshuffled between clusters.

• SplitCluster: the cluster at k whose members are divided most evenly among clusters at k + 1 (its largest child holds the smallest share of it), with SplitRetained giving that largest share.

• SmallestNewCluster: the smallest cluster at k + 1 as a share of participants.

With SOM + Mclust, assignments are the phenotypes of participants' best-matching SOM nodes.

Examples

# \donttest{
data("SampleData")
data("SampleVariableTypes")
df_Labelled <- RevalueData(SampleData, SampleVariableTypes)$RevaluedData
model <- CreateClusterModel_SOM_MClust(
  df_Labelled, c("age", "AXL", "Adiponectin", "Alpha_1_Antitrypsin"),
  method = "exploratory", k_range = 2:4, models = 1,
  som_xdim = 5, som_ydim = 5, min_nodes_per_cluster = NULL,
  Relabel = FALSE)
#> Warning: Explicit SOM dimensions do not meet one or more requested mixture-support or participant-occupancy targets; diagnostics are retained for review.
splits <- PlotClusterSplits(model)
splits$plot

splits$transitions
#> # A tibble: 2 × 7
#>   Model    From    To ParentPurity SplitCluster SplitRetained SmallestNewCluster
#>   <chr>   <int> <int>        <dbl>        <int>         <dbl>              <dbl>
#> 1 Model …     2     3        0.957            1         0.648              0.239
#> 2 Model …     3     4        0.969            1         0.538              0.208
# }