ctreeMI: Conditional Inference Trees with Stacked Multiple Imputation
Implements the stacked-imputation workflow for conditional
inference trees ('ctree') described in Sherlock et al. (2026)
<doi:10.1080/00273171.2026.2661244>. When data contain missing values,
multiply imputed datasets (e.g., from 'mice') are stacked vertically
and a single 'ctree' is fit on the combined data. To correct for the
artificially inflated sample size introduced by stacking, every
node-level test statistic is divided by the number of imputations M,
the node-level p-values are recomputed from the chi-squared reference
distribution 'ctree' uses (including its multiplicity adjustment across
candidate splitting variables), and the tree is compressed bottom-up
(the Stack/M correction). Degrees of freedom are derived for each node
and each candidate variable, so univariate, bivariate and
higher-dimensional outcomes are all handled, as are unordered factor
predictors, whose degrees of freedom depend on how many levels remain
in a node. The result is a conservative but interpretable single tree
that incorporates imputation uncertainty without requiring pooling of
structurally different trees. Also exports stack_imputations(),
rescale_statistic(), prune_stackM(), node_table() and
report_ctreeMI() as standalone utilities. The underlying 'ctree'
algorithm is provided by 'partykit' (Hothorn & Zeileis, 2015; Hothorn,
Hornik & Zeileis, 2006 <doi:10.1198/106186006X133933>).
Documentation:
Downloads:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=ctreeMI
to link to this page.
[aut, cre]