Stacked bar charts

library(litReview)
library(ggplot2)
data(studies)

reviewStackedBar() cross-tabulates two columns: col defines the primary category (one bar each) and group is split within every bar as the fill. It answers questions like “how does risk of bias break down by study design?”. This article walks through every argument of

reviewStackedBar(data, col, group, position = c("fill", "stack"),
                 fill = PALETTE, width = 0.7, sep = "\r\n",
                 study_id = StudyID, base_size = 12, na.rm = TRUE,
                 na_label = "Not reported", labels = TRUE)

Default

Pass the data frame, the primary column, and the grouping column. Column names may be bare or quoted. Each col category becomes one bar, split by group; by default (position = "fill") every bar is scaled to 100% so you read proportions, and each segment is annotated with its count and percentage.

reviewStackedBar(studies, Design, RiskOfBias)

col: the primary category

col sets one bar per level. Any categorical column works; multi-value cells are split first — see sep below. Here we make one bar per intervention type:

reviewStackedBar(studies, InterventionType, OpenAccess)

group: the splitting (fill) variable

group is stacked within each bar and drives the legend. Swapping col and group re-frames the same cross-tabulation — now one bar per risk-of-bias level, split by design:

reviewStackedBar(studies, RiskOfBias, Design)

position: proportions vs counts

position = "fill" (the default) scales each bar to 100%, so bars are comparable regardless of how many studies fall in each col category:

reviewStackedBar(studies, Design, RiskOfBias, position = "fill")

position = "stack" keeps the raw counts, so bar height reflects the number of studies and the labels report frequencies:

reviewStackedBar(studies, Design, RiskOfBias, position = "stack")

fill: segment colors

The default is the package PALETTE, applied across the group levels:

reviewStackedBar(studies, Design, RiskOfBias, fill = PALETTE)

Pass a vector of colors to map one color per group level (recycled if shorter than the number of levels):

reviewStackedBar(studies, Design, RiskOfBias,
                 fill = c("#59a14f", "#edc948", "#e15759"))

A named vector pins specific colors to specific group levels, independent of their order — useful to keep, say, “High” red across figures:

reviewStackedBar(studies, Design, RiskOfBias,
                 fill = c("Low" = "#59a14f",
                          "Moderate" = "#edc948",
                          "High" = "#e15759"))

width: bar thickness

width runs from 0 to 1 (fraction of the available band). Thin bars:

reviewStackedBar(studies, Design, RiskOfBias, width = 0.3)

Full-width bars:

reviewStackedBar(studies, Design, RiskOfBias, width = 1)

sep: multi-value separator

Either col or group may hold several values per cell. In studies, Outcome uses newline separators ("\r\n", the default), so a study reporting multiple outcomes contributes to each. Here Outcome is the primary category, split by design:

reviewStackedBar(studies, Outcome, Design)

If your data uses a different delimiter, set sep. Here we rebuild a semicolon-separated column to demonstrate:

studies_semi <- studies
studies_semi$Outcome <- gsub("\r\n", "; ", studies_semi$Outcome)
reviewStackedBar(studies_semi, Outcome, Design, sep = "; ")

study_id: which column identifies a study

study_id names the column used to identify studies when de-duplicating and counting (default StudyID). It matters for multi-value columns, where the same study can appear in several segments. Pointing it at Author uses that column as the identifier instead:

reviewStackedBar(studies, Outcome, Design, study_id = Author)

base_size: overall text and element scaling

A single knob scales all text and spacing proportionally. Smaller, for multi-panel figures:

reviewStackedBar(studies, Design, RiskOfBias, base_size = 9)

Larger, for slides or posters:

reviewStackedBar(studies, Design, RiskOfBias, base_size = 18)

na.rm and na_label: missing data in the group

This function exposes two missing-data controls, both acting on the group column. By default na.rm = TRUE drops rows with a missing group value. OpenAccess has some unreported entries, which vanish here:

reviewStackedBar(studies, InterventionType, OpenAccess)

Set na.rm = FALSE to keep those rows as their own segment; na_label sets its name:

reviewStackedBar(studies, InterventionType, OpenAccess,
                 na.rm = FALSE, na_label = "Not reported")

The same pair works for any group with gaps, such as RiskOfBias:

reviewStackedBar(studies, Design, RiskOfBias,
                 na.rm = FALSE, na_label = "Unclear")

labels: in-segment annotations

labels = TRUE (the default) prints the count/percentage inside each segment. Turn it off for a cleaner look, leaving the legend to carry the meaning:

reviewStackedBar(studies, Design, RiskOfBias, labels = FALSE)

Composing with ggplot2

Every review*() function returns a plain ggplot, so you can keep adding layers, scales, and labels with +:

reviewStackedBar(studies, Design, RiskOfBias) +
  labs(title = "Risk of bias by study design",
       subtitle = "n = 50 studies", fill = "Risk of bias",
       caption = "Source: example dataset") +
  theme(plot.title.position = "plot")