Loop De Loop"
In ggsankeyfier: Create Sankey and Alluvial Diagrams Using 'ggplot2'

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.ext = "png",
  dev = "png"
)

To cycle or not to cycle

The ggsankeyfier package requires you to specify from which node, to which node an edge flows. See vignette("data_management") for more technical details on how this works. Consequently, this allows you to let edges flow to any node in any stage. In most cases, Sankey or alluvial diagrams are essentially stage-structured directed acyclic graphs. However, the way that ggsankeyfier organises data allows you to create stage-structured directed graphs. In other words, you can create feedback loops in your diagram. This is demonstrated in the present vignette.

For the examples we first need to prepare a basic plot:

library(ggsankeyfier)
library(ggplot2)
theme_set(theme_light())
set.seed(0)

pos <- position_sankey(v_space = "auto")

## Let's subset the data, to make the plot less cluttered:
es_subset <- pivot_stages_longer(
  subset(ecosystem_services, RCSES > 0.02),
  c("activity_realm", "biotic_realm", "service_section"),
  "RCSES")

# And prepare a basis for the plot
p <-
  ggplot(es_subset, aes(x = stage, y = RCSES, group = node,
                        connector = connector, edge_id = edge_id)) +
  geom_sankeynode(position = pos) +
  geom_sankeyedge(position = pos)

Feedback

In order to demonstrate a feedback loop we rbind a fictional edge to the demonstration data. Note that this feedback loop does not make any sense in the context of the data. It only serves as proof of principle. With the %+% operator we update the plot with the data that includes the feedback loop.

es_subset_feedback <-
  es_subset |>
  rbind(
    data.frame(
      RCSES     = 0.05,
      edge_id   = max(es_subset$edge_id) + 1,
      connector = c("from", "to"),
      node      = c("Cultural", "Fish & Cephalopods"),
      stage     = c("service_section", "biotic_realm")
    )
  )

p %+% es_subset_feedback

Self reference

Edges don't even have to flow from one stage to another regardless of its direction. Instead, it is also possible to let an edge flow from and to the same stage. In fact, you can even let an edge flow from and to the same node. This is illustrated in the example below, we we rbind a fictional self-referencing node. Again, in the context of the data, this self-reference does not make sense, but it serves as proof of principle.

es_subset_selfref <-
  es_subset |>
  rbind(
    data.frame(
      RCSES     = 0.05,
      edge_id   = max(es_subset$edge_id) + 1,
      connector = c("from", "to"),
      node      = c("Cultural", "Cultural"),
      stage     = c("service_section", "service_section")
    )
  )

p %+% es_subset_selfref