Request describe_distribution()

[DominiqueMakowski]

From Andy for his book update:

• When you use group_by() with describe_distribution() (you're going to tell me not to do this .....) it adds a column to the end .group, can the default be to put this column at the start of the tibble, not the end?
• Also, why .group not the name of the variable, and why variable=category rather than the group names? I think it would be really helpufl if the default behaviour was to stick a column at the start of the output, names with the same name as the grouping variable, with rows labelled with the factor levels/categories they represent

Should we add a by argument to that function as we now have in more and more ones?

[etiennebacher]

When you use group_by() with describe_distribution() [...] it adds a column to the end .group

I don't see this?

library(datawizard)

iris |> 
  dplyr::group_by(Species) |> 
  describe_distribution()
#> # Species=setosa
#> 
#> Variable     | Mean |   SD |  IQR |        Range | Skewness | Kurtosis |  n | n_Missing
#> ---------------------------------------------------------------------------------------
#> Sepal.Length | 5.01 | 0.35 | 0.40 | [4.30, 5.80] |     0.12 |    -0.25 | 50 |         0
#> Sepal.Width  | 3.43 | 0.38 | 0.52 | [2.30, 4.40] |     0.04 |     0.95 | 50 |         0
#> Petal.Length | 1.46 | 0.17 | 0.20 | [1.00, 1.90] |     0.11 |     1.02 | 50 |         0
#> Petal.Width  | 0.25 | 0.11 | 0.10 | [0.10, 0.60] |     1.25 |     1.72 | 50 |         0
#> 
#> # Species=versicolor
#> 
#> Variable     | Mean |   SD |  IQR |        Range | Skewness | Kurtosis |  n | n_Missing
#> ---------------------------------------------------------------------------------------
#> Sepal.Length | 5.94 | 0.52 | 0.70 | [4.90, 7.00] |     0.11 |    -0.53 | 50 |         0
#> Sepal.Width  | 2.77 | 0.31 | 0.50 | [2.00, 3.40] |    -0.36 |    -0.37 | 50 |         0
#> Petal.Length | 4.26 | 0.47 | 0.60 | [3.00, 5.10] |    -0.61 |     0.05 | 50 |         0
#> Petal.Width  | 1.33 | 0.20 | 0.30 | [1.00, 1.80] |    -0.03 |    -0.41 | 50 |         0
#> 
#> # Species=virginica
#> 
#> Variable     | Mean |   SD |  IQR |        Range | Skewness | Kurtosis |  n | n_Missing
#> ---------------------------------------------------------------------------------------
#> Sepal.Length | 6.59 | 0.64 | 0.75 | [4.90, 7.90] |     0.12 |     0.03 | 50 |         0
#> Sepal.Width  | 2.97 | 0.32 | 0.40 | [2.20, 3.80] |     0.37 |     0.71 | 50 |         0
#> Petal.Length | 5.55 | 0.55 | 0.80 | [4.50, 6.90] |     0.55 |    -0.15 | 50 |         0
#> Petal.Width  | 2.03 | 0.27 | 0.50 | [1.40, 2.50] |    -0.13 |    -0.60 | 50 |         0

Also, why .group not the name of the variable, and why variable=category rather than the group names?

I don't really understand this, it would be nice if he had an example of what currently happens and what he'd like the output to be.

Should we add a by argument to that function as we now have in more and more ones?

Looks similar to #582, I'll try to tackle this in the next week.

[strengejacke]

tibble

We don't use tibbles.

[strengejacke]

Looks similar to #582, I'll try to tackle this in the next week.

Would probably need some minor refactoring, putting the code for the .data.frame() methods in a small helper function, so we can split by by in the .data.frame() methods and apply to each subset?

[DominiqueMakowski]

I don't see this?

library(datawizard)

iris |> 
  dplyr::group_by(Species) |> 
  describe_distribution() |> 
  knitr::kable()

Variable	Mean	SD	IQR	Min	Max	Skewness	Kurtosis	n	n_Missing	.group
Sepal.Length	5.006	0.3524897	0.400	4.3	5.8	0.1200870	-0.2526888	50	0	Species=setosa
Sepal.Width	3.428	0.3790644	0.525	2.3	4.4	0.0411665	0.9547033	50	0	Species=setosa
Petal.Length	1.462	0.1736640	0.200	1.0	1.9	0.1063939	1.0215761	50	0	Species=setosa
Petal.Width	0.246	0.1053856	0.100	0.1	0.6	1.2538614	1.7191302	50	0	Species=setosa
Sepal.Length	5.936	0.5161711	0.700	4.9	7.0	0.1053776	-0.5330095	50	0	Species=versicolor
Sepal.Width	2.770	0.3137983	0.500	2.0	3.4	-0.3628448	-0.3662374	50	0	Species=versicolor
Petal.Length	4.260	0.4699110	0.600	3.0	5.1	-0.6065077	0.0479033	50	0	Species=versicolor
Petal.Width	1.326	0.1977527	0.300	1.0	1.8	-0.0311796	-0.4100592	50	0	Species=versicolor
Sepal.Length	6.588	0.6358796	0.750	4.9	7.9	0.1180151	0.0329044	50	0	Species=virginica
Sepal.Width	2.974	0.3224966	0.400	2.2	3.8	0.3659491	0.7060705	50	0	Species=virginica
Petal.Length	5.552	0.5518947	0.800	4.5	6.9	0.5494446	-0.1537786	50	0	Species=virginica
Petal.Width	2.026	0.2746501	0.500	1.4	2.5	-0.1294769	-0.6022645	50	0	Species=virginica

^{Created on 2025-04-01 with reprex v2.1.1}

[strengejacke]

Add format():

library(datawizard)
iris |> 
  dplyr::group_by(Species) |> 
  describe_distribution() |> 
  format() |> 
  insight::export_table(format = "md")

Variable	Mean	SD	IQR	Range	Skewness	Kurtosis	n	n_Missing
Sepal.Length	5.01	0.35	0.40	[4.30, 5.80]	0.12	-0.25	50	0
Sepal.Width	3.43	0.38	0.52	[2.30, 4.40]	0.04	0.95	50	0
Petal.Length	1.46	0.17	0.20	[1.00, 1.90]	0.11	1.02	50	0
Petal.Width	0.25	0.11	0.10	[0.10, 0.60]	1.25	1.72	50	0

# Species=setosa

Variable	Mean	SD	IQR	Range	Skewness	Kurtosis	n	n_Missing
Sepal.Length	5.94	0.52	0.70	[4.90, 7.00]	0.11	-0.53	50	0
Sepal.Width	2.77	0.31	0.50	[2.00, 3.40]	-0.36	-0.37	50	0
Petal.Length	4.26	0.47	0.60	[3.00, 5.10]	-0.61	0.05	50	0
Petal.Width	1.33	0.20	0.30	[1.00, 1.80]	-0.03	-0.41	50	0

# Species=versicolor

Variable	Mean	SD	IQR	Range	Skewness	Kurtosis	n	n_Missing
Sepal.Length	6.59	0.64	0.75	[4.90, 7.90]	0.12	0.03	50	0
Sepal.Width	2.97	0.32	0.40	[2.20, 3.80]	0.37	0.71	50	0
Petal.Length	5.55	0.55	0.80	[4.50, 6.90]	0.55	-0.15	50	0
Petal.Width	2.03	0.27	0.50	[1.40, 2.50]	-0.13	-0.60	50	0

# Species=virginica

^{Created on 2025-04-01 with reprex v2.1.1}

But maybe we should just add a print_md() and print_html() method for internal printing. The .group variable is needed for grouping / layouting the output.

[DominiqueMakowski]

Can't we put this .group into attributes when using format? I reckon many people still use and will use kable() instead of display()

[DominiqueMakowski]

Would it be possible to add kable() methods to our objects that call knitr::kable(format(x)) ?

[strengejacke]

Can't we put this .group into attributes when using format? I reckon many people still use and will use kable() instead of display()

Or we indeed rename it into Group or something, and move that column to the beginning. And then in format(), we can remove that column to create a list of tables for insight::export_table().

[DominiqueMakowski]

That seems unnecessary, the column in the example above should be "Species" with the species levels as values

As then we could have a "group" attribute that just says "Species"

This way we know that the data is grouped by the Species column (and we can keep the same separate printing as we do currently)

[strengejacke]

Yes, sounds good.

[profandyfield]

Can I ask what the likley outcome of this discussion will be? Essentially I have to finish my book by May/June. I'm trying not to oscillate between display() and kable()because it'll be inherently confusing to new R users. I currently favour kable() because working with tibbles gives you a lot of versatility, but I like the fact that display doesn't require a separate package (easier for newbies) ... and also recent changes like overriding column names (thanks for that) make display() more appealing. Perhpas most important @DominiqueMakowski is a persuasive guy:)

However, getting descritpives by groups is a really common task, and kable() handles this better than display() (IMO).

For example:

cloak_tib <- tibble::tibble(
    id = c("Kiara", "Anupama", "Collin", "Ian", "Vanessa", "Darrell", "Tyler", "Steven", "Katheryn", "Kirkpatrick", "Melissa", "Kinaana", "Shajee'a", "Sage", "Jorge", "Conan","Tamara","Man", "Joseph", "Israa", "Kathryn", "Luis", "Devante", "Jerry"),
    cloak = gl(2, 12, labels = c("No cloak", "Cloak")),
    mischief = c(3, 1, 5, 4, 6, 4, 6, 2, 0, 5, 4, 5, 4, 3, 6, 6, 8, 5, 5, 4, 2, 5, 7, 5)
  )

cloak_tib |> 
  group_by(cloak) |> 
  describe_distribution(ci = 0.95) |> 
  display()

renders with these horrible labels and isn't space efficient in a rendered document:

I could live with the labels if the cloak= weren't there, or they could be over-ridden, but as is they aren't really usable for students producing reports.

This code

cloak_tib |> 
  group_by(cloak) |> 
  describe_distribution(ci = 0.95) |> 
  format_table() |> 
  display()

renders more kable-like (but see @DominiqueMakowski original summary of the issue), but I have to explain why and extra function is required and it's confusing for newbies to remember when they do and do not need the extra function in the pipe:

I'm currently of a mind to fudge this with a 'use kable() for grouped data', but I'd really prefer to stick to one function in the book (and indeed my teaching) to render tables. I realise that this is a much more pressing issue for me than the easystats team, but it's be helpful to know whether the behaviour of display() is likley to change anytime soon. Needless to say, fine if the plan is not to change it!

[strengejacke]

Just to be clear, the desired result would be something like

library(easystats)
#> # Attaching packages: easystats 0.7.4
#> ✔ bayestestR  0.15.2.11   ✔ correlation 0.8.7    
#> ✔ datawizard  1.0.2.9     ✔ effectsize  1.0.0    
#> ✔ insight     1.1.0.32    ✔ modelbased  0.10.0.23
#> ✔ performance 0.13.0.6    ✔ parameters  0.24.2.11
#> ✔ report      0.6.1       ✔ see         0.11.0.1
out <- iris |> 
  data_group("Species") |> 
  describe_distribution(ci = 0.95) |> 
  format_table()

out <- cbind(Species = out$.group, out)
out$Species <- gsub("Species=", "", out$Species, fixed = TRUE)
out$.group <- NULL

display(out)

Species	Variable	Mean	SD	IQR	95% CI	Min	Max	Skewness	Kurtosis	n	n_Missing
setosa	Sepal.Length	5.01	0.35	0.40	[4.93, 5.08]	4.30	5.80	0.12	-0.25	50.00	0
setosa	Sepal.Width	3.43	0.38	0.52	[3.34, 3.54]	2.30	4.40	0.04	0.95	50.00	0
setosa	Petal.Length	1.46	0.17	0.20	[1.41, 1.50]	1.00	1.90	0.11	1.02	50.00	0
setosa	Petal.Width	0.25	0.11	0.10	[0.22, 0.27]	0.10	0.60	1.25	1.72	50.00	0
versicolor	Sepal.Length	5.94	0.52	0.70	[5.80, 6.05]	4.90	7.00	0.11	-0.53	50.00	0
versicolor	Sepal.Width	2.77	0.31	0.50	[2.67, 2.84]	2.00	3.40	-0.36	-0.37	50.00	0
versicolor	Petal.Length	4.26	0.47	0.60	[4.17, 4.41]	3.00	5.10	-0.61	0.05	50.00	0
versicolor	Petal.Width	1.33	0.20	0.30	[1.27, 1.38]	1.00	1.80	-0.03	-0.41	50.00	0
virginica	Sepal.Length	6.59	0.64	0.75	[6.39, 6.75]	4.90	7.90	0.12	0.03	50.00	0
virginica	Sepal.Width	2.97	0.32	0.40	[2.91, 3.06]	2.20	3.80	0.37	0.71	50.00	0
virginica	Petal.Length	5.55	0.55	0.80	[5.43, 5.70]	4.50	6.90	0.55	-0.15	50.00	0
virginica	Petal.Width	2.03	0.27	0.50	[1.96, 2.10]	1.40	2.50	-0.13	-0.60	50.00	0

^{Created on 2025-04-14 with reprex v2.1.1}

[profandyfield]

exactly that!

[profandyfield]

In tidy data format you also might group by 2 or more variables and have a single outcome variable (e.g. group_by(species, variable).

For example, this currently won't work:

palmerpenguins::penguins |> 
  group_by(species, island) |> 
  describe_distribution(select = "bill_length_mm", ci = 0.95)