multiple_correspondence_analysis: Multiple Correspondence Analysis

In ggfacto: Graphs for Correspondence Analysis

Multiple Correspondence Analysis

Description

A user-friendly wrapper around MCA, made to work with ggfacto functions like ggmca, interpret and hierarchical_clust. Variables are selected the way of the 'tidyverse', as in tabxplor::tab(). Supplementary variables are not given here: they are added afterwards, in ggmca.

'MCA2()' keeps the fit of ggfacto 0.3.2, on the individuals: '$ind' has one row per analysed row, so that 'FactoMineR::HCPC()' of it classifies the individuals, in their order. It gives the same graphs, tables and clusters as 'multiple_correspondence_analysis()', more slowly on large data, and will be deprecated.

Usage

multiple_correspondence_analysis(
  data,
  active_vars,
  wt,
  excl = NA,
  ncp = Inf,
  graph = FALSE,
  filter,
  ...
)

MCA2(data, active_vars, wt, excl = NA, ncp = Inf, graph = FALSE, filter, ...)

Arguments

data	The data frame. To analyse a subset of the population, give the whole data frame and 'filter', or filter it inside the call with the native pipe, 'data |> dplyr::filter(...) |> multiple_correspondence_analysis(...)': the analysis then remembers which rows it used, so that ggmca, hierarchical_clust or is_in_analysis can be given the whole data frame afterwards.
active_vars	<tidy-select> The active variables.
wt	<tidy-select> The weight variable, if any.
excl	The levels to exclude from the calculation of the axes (specific multiple correspondence analysis), matched exactly by name. The missing values of each active variable become a level named '<VAR>.NA', and 'NA', the default, excludes all of them: 'excl = NA' for missing values only, 'excl = c(NA, "Other")' to exclude a level too, 'excl = "DIPLOMA.NA"' for the missing values of one variable only, 'excl = NULL' to keep every level.
ncp	The number of axes to keep. All of them by default: the eigenvalue table is how one chooses how many axes to interpret, and a truncated one cannot show the drop — it also renormalises Benzecri's modified rate over the axes it kept, so the same axis gets a different rate. To cluster on the first axes, give hierarchical_clust its own 'ncp'.
graph	By default no graph is made, since the result can be plotted with ggfacto.
filter	A condition on the rows of 'data', as in dplyr::filter(): only the rows where it is 'TRUE' are analysed ('filter = AGE >= 18').
...	Additional arguments to pass to MCA, except those that index its rows or columns ('ind.sup', 'quali.sup', 'quanti.sup', 'tab.disj').

Value

A 'MCA' object from FactoMineR, fitted on the distinct answer profiles (the combinations of active answers), each weighted by its individuals: the eigenvalues and every result on the levels are the individuals', and '$ind' has one row per profile (per individual with 'MCA2()'). Use axis_coord and hierarchical_clust to write coordinates and clusters into the data frame ('FactoMineR::HCPC()' would cluster the profiles). One more element, 'source', records for each row of 'data' its row of '$ind' ('NA' if it was not analysed) and its weight.

Examples

data(tea, package = "FactoMineR")
res.mca <- multiple_correspondence_analysis(tea, 1:18)
interpret(res.mca)                           # the eigenvalues, then the axes

ggfacto(res.mca, tea, sup_vars = c(sex, SPC))  # the graph, with supplementary variables
ggfacto(res.mca, tea, sup_vars = c(sex, SPC), interactive = TRUE)  # hover: the crosstables


# A subset of the population: the analysis remembers which rows it used
res.mca_young <- tea |>
  dplyr::filter(age < 30) |>
  multiple_correspondence_analysis(1:18)
# the same analysis
res.mca_young <- multiple_correspondence_analysis(tea, 1:18, filter = age < 30)

ggfacto documentation built on Sept. 23, 2026, 1:08 a.m.