Problem:

I have a hierarchical set of data, where I want to generate a grouped boxplot. Colour by one variable and fill by another. This can easily be done in ggplot2 (example provided). But I am unable to achieve the same using plotly(). Note: I do not want to use ggplotly() as it is considerably slower with large datasets compared to plotly().

Bonus: how can I define 'group_c' colours manually with the plotly() solution, this can be done with scale_colour_manual() in ggplot2.

Data:

group_a <- 900:904
group_b <- paste0(rep(group_a, each = 3), '_', rep(c('01', '02', '03'), times = length(group_a)))
group_c <- rep(c("A", "B"), times = length(group_b))

group_date <- seq.Date(from = as.Date("2025-01-01"), by = "day", length.out = length(lot_ids))
group_date <- rep(group_date, each = 3)

x <- rep(runif(14), times = length(group_a) * 3 * 2)

# Create the dataframe
df <- data.frame(
  group_a = rep(group_a, each = 3 * 2 * 14),
  group_b = rep(group_b, each = 14 * 2),
  group_date = rep(group_date, each = 2 * 14),
  group_c = group_c,
  x = x
)

ggplot method:

df %>%
  ggplot(aes(
    x = group_date,
    y = x,
    group = interaction(group_c, group_b),
    fill = factor(group_a),
    colour = group_c
  )) + 
  geom_boxplot()

You mean like this? Adjust the group_c colors in transforms to answer your bonus question. I could not figure out, how to make the boxplots wider.

Code

library(plotly)
library(dplyr)

group_a <- 900:904
group_b <- paste0(rep(group_a, each = 3), '_', rep(c('01', '02', '03'), times = length(group_a)))
group_c <- rep(c("A", "B"), times = length(group_b))
group_date <- seq.Date(from = as.Date("2025-01-01"), by = "day", length.out = length(group_a))
group_date <- rep(group_date, each = 3)
x <- rep(runif(14), times = length(group_a) * 3 * 2)

df <- data.frame(
  group_a = rep(group_a, each = 3 * 2 * 14),
  group_b = rep(group_b, each = 14 * 2),
  group_date = rep(group_date, each = 2 * 14),
  group_c = group_c,
  x = x
)


plot_ly(data = df) %>%
  add_boxplot(
    x = ~group_date,
    y = ~x,
    type = "box",
    name = ~group_b,
    color = ~factor(group_a),
    transforms = list(
      list(
        type = 'groupby',
        groups = ~group_c,
        styles = list(
          list(target = "A", value = list(line = list(color = "red"))),
          list(target = "B", value = list(line = list(color = "blue")))
        )
      )
    )
  ) %>%
  layout(
    boxmode = "group",
    boxgap = 0,
    boxgroupgap = 0
  )

Add

You can use a color ramp to adjust the outline colors on unique group_c values:

library(plotly)
library(dplyr)
library(grDevices) # For colorRampPalette

group_a <- 900:904
group_b <- paste0(rep(group_a, each = 3), '_', rep(c('01', '02', '03'), times = length(group_a)))
group_c <- rep(c("A", "B", "C"), times = length(group_b) / 3)
group_date <- seq.Date(from = as.Date("2025-01-01"), by = "day", length.out = length(group_a))
group_date <- rep(group_date, each = 3)
x <- rep(runif(14), times = length(group_a) * 3 * 3)

df <- data.frame(
  group_a = rep(group_a, each = 3 * 3 * 14),
  group_b = rep(group_b, each = 14 * 3),
  group_date = rep(group_date, each = 3 * 14),
  group_c = group_c,
  x = x
)

# Extract unique values of group_c and define colors dynamically
unique_group_c <- unique(df$group_c)
color_palette <- colorRampPalette(c("red", "blue"))(length(unique_group_c))

# Create styles list
styles_list <- lapply(seq_along(unique_group_c), function(i) {
  list(
    target = unique_group_c[i],
    value = list(line = list(color = color_palette[i]))
  )
})

# Create the plot
plot_ly(data = df) %>%
  add_boxplot(
    x = ~group_date,
    y = ~x,
    type = "box",
    name = ~group_b,
    color = ~factor(group_a),
    transforms = list(
      list(
        type = 'groupby',
        groups = ~group_c,
        styles = styles_list
      )
    )
  ) %>%
  layout(
    boxmode = "group",
    boxgap = 0,
    boxgroupgap = 0
  )