Creating labels for a variable confuses gtsummary

Question

I have a particular problem with labeling categorical variables using {gtsummary}.
When I label the levels of a variable, {gtsummary} does not work properly.

Assume the following data:

library(survey)
library(gtsummary)
library(cobalt)

# Create data and add a categorical variable named hello with five levels:

data <- lalonde
hello <- sample(1:5, nrow(data), replace = TRUE)
data$hello <- as.factor(hello)

# Labels:
levels(data$hello) <- c("hello_1", "hello_2", "hello_3", "hello_4", "hello_5")

# Now we specify the variables we want to include in our table1:

vars <- c("hello")

# Now we create table 1:

datasvy <- survey::svydesign(~1, data = data, strata = ~treat)

table1 <- tbl_svysummary(
  datasvy,
  by = treat,
  include = all_of(vars),
  statistic = list(
    all_categorical() ~ "{n}({p}%)"
  ),
  percent = "col",
  digits = list(
    all_categorical() ~ c(0, 1)
  )
)
table1 <- as.data.frame(table1)

Now this is the result:

Characteristic	Group 0 (N = 429)	Group 1 (N = 185)
hello	NA	NA
hello_1	91(NA%)	40(NA%)
hello_2	71(NA%)	35(NA%)
hello_3	82(NA%)	31(NA%)
hello_4	86(NA%)	37(NA%)
hello_5	99(NA%)	42(NA%)

Is there a solution to that?
When I do not use the labels, everything works fine.

Reprex:

library(gtsummary)

data <- cobalt::lalonde
hello <- sample(1:5, nrow(data), replace = TRUE)
data$hello <- as.factor(hello)
levels(data$hello) <- c("hello_1", "hello_2", "hello_3", "hello_4", "hello_5")

# Now we specify the variables we want to include in our table1:
vars <- c("hello")

# Now we create table 1:
survey::svydesign(~1, data = data, strata = ~treat) |> 
  tbl_svysummary(
    by = treat,
    include = all_of(vars),
    statistic = list(
      all_categorical() ~ "{n} ({p}%)"
    ),
    percent = "column",
    digits = list(
      all_categorical() ~ c(0, 1)
    )
  ) |> 
  as_kable()
#> Warning in svydesign.default(~1, data = data, strata = ~treat): No weights or
#> probabilities supplied, assuming equal probability
Characteristic  0, N = 429  1, N = 185
hello       
hello_1 86 (NA%)    43 (NA%)
hello_2 86 (NA%)    33 (NA%)
hello_3 96 (NA%)    35 (NA%)
hello_4 87 (NA%)    39 (NA%)
hello_5 74 (NA%)    35 (NA%)

Created on 2024-11-25 with reprex v2.1.1

Answer 1

I'd like to suggest using surveytable instead. It's easy to use and gives you all the estimates that you might be looking for.

# Create data
data <- cobalt::lalonde
hello <- sample(1:5, nrow(data), replace = TRUE)
data$hello <- as.factor(hello)
levels(data$hello) <- c("hello_1", "hello_2", "hello_3", "hello_4", "hello_5")

# For surveytable, treat needs to be a factor
data$treat = as.factor(data$treat)
datasvy = survey::svydesign(~1, data = data, strata = ~treat)

# surveytable
library(surveytable)
set_survey(datasvy)

tab_subset("hello", "treat")

Output:

                      hello (treat = 0) {datasvy}                      
┌─────────┬────┬────────┬────┬────┬─────┬─────────┬─────┬──────┬──────┐
│ Level   │  n │ Number │ SE │ LL │  UL │ Percent │  SE │   LL │   UL │
├─────────┼────┼────────┼────┼────┼─────┼─────────┼─────┼──────┼──────┤
│ hello_1 │ 85 │     85 │  8 │ 70 │ 103 │    19.8 │ 1.9 │ 16.1 │ 23.9 │
├─────────┼────┼────────┼────┼────┼─────┼─────────┼─────┼──────┼──────┤
│ hello_2 │ 99 │     99 │  9 │ 83 │ 118 │    23.1 │ 2   │ 19.2 │ 27.4 │
├─────────┼────┼────────┼────┼────┼─────┼─────────┼─────┼──────┼──────┤
│ hello_3 │ 90 │     90 │  8 │ 75 │ 108 │    21   │ 2   │ 17.2 │ 25.1 │
├─────────┼────┼────────┼────┼────┼─────┼─────────┼─────┼──────┼──────┤
│ hello_4 │ 73 │     73 │  8 │ 59 │  90 │    17   │ 1.8 │ 13.6 │ 20.9 │
├─────────┼────┼────────┼────┼────┼─────┼─────────┼─────┼──────┼──────┤
│ hello_5 │ 82 │     82 │  8 │ 67 │ 100 │    19.1 │ 1.9 │ 15.5 │ 23.2 │
└─────────┴────┴────────┴────┴────┴─────┴─────────┴─────┴──────┴──────┘
  N = 429.                                                             

                     hello (treat = 1) {datasvy}                      
┌─────────┬────┬────────┬────┬────┬────┬─────────┬─────┬──────┬──────┐
│ Level   │  n │ Number │ SE │ LL │ UL │ Percent │  SE │   LL │   UL │
├─────────┼────┼────────┼────┼────┼────┼─────────┼─────┼──────┼──────┤
│ hello_1 │ 50 │     50 │  6 │ 39 │ 64 │    27   │ 3.3 │ 20.8 │ 34.1 │
├─────────┼────┼────────┼────┼────┼────┼─────────┼─────┼──────┼──────┤
│ hello_2 │ 33 │     33 │  5 │ 24 │ 46 │    17.8 │ 2.8 │ 12.6 │ 24.1 │
├─────────┼────┼────────┼────┼────┼────┼─────────┼─────┼──────┼──────┤
│ hello_3 │ 36 │     36 │  5 │ 27 │ 49 │    19.5 │ 2.9 │ 14   │ 25.9 │
├─────────┼────┼────────┼────┼────┼────┼─────────┼─────┼──────┼──────┤
│ hello_4 │ 33 │     33 │  5 │ 24 │ 46 │    17.8 │ 2.8 │ 12.6 │ 24.1 │
├─────────┼────┼────────┼────┼────┼────┼─────────┼─────┼──────┼──────┤
│ hello_5 │ 33 │     33 │  5 │ 24 │ 46 │    17.8 │ 2.8 │ 12.6 │ 24.1 │
└─────────┴────┴────────┴────┴────┴────┴─────────┴─────┴──────┴──────┘
  N = 185.                                                            

Answer 2

I can't replicate your issue. Can you use a proper reprex (reprex.tidyverse.com), which will evaluate in a fresh environment?

library(gtsummary)
packageVersion("gtsummary")
#> [1] '2.0.3'

# Create data and add a categorical variable named hello with five levels:
data <- cobalt::lalonde
hello <- sample(1:5, nrow(data), replace = TRUE)
data$hello <- as.factor(hello)
levels(data$hello) <- c("hello_1", "hello_2", "hello_3", "hello_4", "hello_5")
# Now we specify the variables we want to include in our table1:
vars <- c("hello")

# Now we create table 1:
survey::svydesign(~1, data = data, strata = ~treat) |> 
  tbl_svysummary(
    by = treat,
    include = all_of(vars),
    statistic = list(
      all_categorical() ~ "{n} ({p}%)"
    ),
    percent = "column",
    digits = list(
      all_categorical() ~ c(0, 1)
    )
  ) |> 
  as_kable() # convert to kable to display on SO
#> Warning in svydesign.default(~1, data = data, strata = ~treat): No weights or
#> probabilities supplied, assuming equal probability

Characteristic	0 N = 429	1 N = 185
hello
hello_1	88 (20.5%)	37 (20.0%)
hello_2	93 (21.7%)	35 (18.9%)
hello_3	88 (20.5%)	37 (20.0%)
hello_4	85 (19.8%)	36 (19.5%)
hello_5	75 (17.5%)	40 (21.6%)

^{Created on 2024-11-25 with reprex v2.1.1}