Error bars displayed in incorrect order with color attribute assigned

[saudiwin]

When error_x is used along with a factor variable assigned to the color attribute in plot_ly, the resulting plot displays the error bars around the wrong points. Below is a working example using source code from the plotly for R book. You can see in the plot that the standard errors are being mapped to the wrong coefficients.

m <- lm(Sepal.Length ~ Sepal.Width * Petal.Length * Petal.Width, data = iris)

d <- broom::tidy(m) %>%
arrange(desc(estimate)) %>%
mutate(term = factor(term, levels = term),
one_col=cut(estimate,3,labels=c("Low","Medium","High")))
plot_ly(d, x = ~estimate, y = ~term,color=~one_col) %>%
add_markers(error_x = ~list(array=std.error,type="array")) %>%
layout(margin = list(l = 200))

[harveyl888]

I came across this too. One workaround is to add each sample group via add_trace. The code below demonstrates this. Error bars in the first plot (p1) are incorrectly assigned whereas error bars in the second (p2) are correct.

library(plotly)
library(dplyr)
library(tidyr)

## Raw data
df <- data.frame(sample = rep(paste0('sample ', 1:5), 4),
                 x = rnorm(20),
                 group = rep(paste0('group ', 1:2), each = 10),
                 stringsAsFactors = FALSE
                 )

## Stats table
df2 <- df %>%
  group_by(sample, group) %>%
  summarise(avg = mean(x), sd = sd(x)) %>%
  ungroup()

## Plotly barchart with error bars.  Error bars are incorrectly assigned
p1 <- plot_ly(df2, x = ~sample, y = ~avg, color = ~group, type = 'bar', error_y = list(array = ~df2$sd))
p1

## Create individual columns for group data and errors
df3 <- df2 %>%
  gather(key, value, -c(sample, group)) %>%
  mutate(ref = paste0(group, ifelse(key == 'sd', '_sd', ''))) %>%
  select(-group, -key) %>%
  spread(ref, value)

## Plotly barchart displays error bars correctly
p2 <- plot_ly(df3, type = 'bar')
for (g in unique(df2$group)) {
  p2 <- add_trace(p2, x = df3[['sample']], y = df3[[g]], name = g, error_y = list(array = df3[[paste0(g, '_sd')]]))
}
p2

[erfanv]

Is this still the most viable solution for getting error bars to show correctly still? I have run into a similar situation with grouped time-series data. The y-error bars are not being associated with the correct bar.

Attached is the data and below is the code I'm running:

(will need to convert XLSX to csv)
nymphs <- read.csv("nymphs.csv", header=TRUE")

ptest <- plot_ly(nymphs, x = ~week, y = ~nymphs, type= "bar", color = ~treat,
error_y = ~list(
array= se,
type="array",
color="#000000")) %>%
layout(xaxis= list(title="WAT"),
yaxis = list(title= "Means nymphs per plant"))

Below is a screenshot that I'm getting - the large error bars should be on "WAT 10", where the bars are much larger as well. Any help would be greatly appreciated - can't find documentation addressing this issue anywhere.

nymphs.xlsx

[ho-stil]

I was running into this issue when I wanted to filter production data with crosstalk on manufacturing parameters/dates and label the scatterplot by one main parameter.
To get the position of the error bars fixed I modified the workaround by harveyl888 (thanks for the input). My modified version for crosstalk with model-data (I have chosen ordered instead of randomized data to get an idea of the structure behind the issue):

library(crosstalk)
library(plotly)
TestDF <- data.frame(a = c(1:9), b = c(rep(c(1:3), 3)), c = c(1:9/10), d = LETTERS[rep(c(1:3),3)])
d_list <- unique(TestDF$d)
shared_Test_all <- SharedData$new(TestDF, group = "Test")
shared_Test <- list()
p1 <- plot_ly(type = "scatter", mode = "markers")  
for (i in 1:length(d_list)) {
  shared_Test[[i]] <- SharedData$new(TestDF[TestDF$d == d_list[i],], group = "Test")
  p1 <- add_trace(p1, data = shared_Test[[i]], x = ~a, y = ~b, name = ~d, error_y = ~list(array = c))
}
bscols(list(filter_select("d", "filter by d", shared_Test_all, ~d, multiple = TRUE),
            bscols(plot_ly(data = shared_Test_all, x = ~a, y = ~b, 
                           type = "scatter", mode = "markers",
                           error_y = ~list(array = c)),
                   plot_ly(data = shared_Test_all, x = ~a, y = ~b, name = ~d, 
                           type = "scatter", mode = "markers",
                           error_y = ~list(array = c)),
                   p1)))

The resulting plots:

The 1st plot is without labels for reference - the error bars are at the right position.
The 2nd plot shows the bug for the model-data: the error bar increases with the parameter "a" within the group first, instead of being "a/10" for all points. Looks like the order of the errorbar values remains like without groups while they are matched to the scatterplot folowing the order within the groups...
My workaround (3rd plot): I used a list of SharedData-environments - one list element for each group of labels. I grouped the environments for common filtering.

[bd-cameron-willden]

Bump. Please fix this

[RichGordon90]

Bump again - this is v cumbersome workaround.

[syeddans]

Bump again - Please for the love of god fix this

[ArthurAndrews]

I've also been bitten by this issue. I did my best to diagnose it from a user-perspective and provide a reprex.

I don't think this issue is necessarily linked to the color attribute.

I propose it's simply that R plotly removes some missing observations in the R data it sends to plotly js. It may also reorder the observations in the case of factor variable x-axis. However, it doesn't do the same treatment to data sent to error_y (and I assume also error_x) marker attributes.

This reprex shows the JSON data of plotly plots with and without error bars. You can see that the x and y marker attributes are missing the NAs that were in the R data frame, but the error_y array attributes still have the NAs.

This causes a dangerous misalignment of the data such that error bars are plotted on the wrong observations.

A possible workaround is to carefully remove NAs in your own data and possibly avoid factor variables and instead rely on layout to order categorical axes.

library(tidyverse)
library(plotly)
library(jsonlite)

# function to show the pertinent json data of the plotly plot
show_plotly_data <- \(plt) plt |> 
  plotly_json(jsonedit = FALSE) |> 
  fromJSON(simplifyDataFrame = FALSE) |> 
  pluck("data") |> 
  pluck(1) |> 
  with(lst(x, y, error_y$array, error_y$arrayminus))

# sample data
dat <- tibble(
  x = c("thin", "medium", "missing", "wide"),
  y = c(1, 2, NA, 3), 
  upper = c(1, 2, NA, 3),
  lower = c(1, 2, NA, 3)
) |> 
  print()
#> # A tibble: 4 × 4
#>   x           y upper lower
#>   <chr>   <dbl> <dbl> <dbl>
#> 1 thin        1     1     1
#> 2 medium      2     2     2
#> 3 missing    NA    NA    NA
#> 4 wide        3     3     3

# plot with no error bars
ok_plt <- dat |> 
  plot_ly() |> 
  add_markers(x = ~x, y = ~y) |> 
  print()
#> Warning: Ignoring 1 observations

# the R plotly package seems to remove the observations with missing y values
ok_plt |> 
  show_plotly_data()
#> Warning: Ignoring 1 observations
#> $x
#> [1] "thin"   "medium" "wide"  
#> 
#> $y
#> [1] 1 2 3
#> 
#> $`error_y$array`
#> NULL
#> 
#> $`error_y$arrayminus`
#> NULL

# plot with error bars is wrong
buggy_plt <- dat |> 
  plot_ly() |> 
  add_markers(x = ~x, y = ~y, error_y = list(type = "data", array = ~upper, arrayminus = ~lower)) |> 
  layout(title = list(text = "buggy plot")) |> 
  print()
#> Warning: Ignoring 1 observations

# R plotly doesn't do the same NA treatment for error_y "array" values, and this makes the plot and data wrong
buggy_plt |> 
  show_plotly_data()
#> Warning: Ignoring 1 observations
#> $x
#> [1] "thin"   "medium" "wide"  
#> 
#> $y
#> [1] 1 2 3
#> 
#> $`error_y$array`
#> [1]  1  2 NA  3
#> 
#> $`error_y$arrayminus`
#> [1]  1  2 NA  3

# attempt to work around with a drop_na()
workaround_plt <- dat |> 
  drop_na() |> 
  plot_ly() |> 
  add_markers(x = ~x, y = ~y, error_y = list(type = "data", array = ~upper, arrayminus = ~lower)) |> 
  layout(title = list(text = "workaround plot")) |> 
  print()

^{Created on 2025-06-20 with reprex v2.1.1}