Custom table significance / Count of missing values in total rows

[JB0207]

Custom table significance / Count of missing values in total rows

[gdemin]

Hi!

tab_last_sig_cpct works only on percents. When you apply it on the table with counts, as in your case, it gives incorrect results.

As for count of NA:

library(expss)

data("mtcars")

mtcars$am[1:3] = NA

var_lab(mtcars$am) = 'Transmission'

count_na = function(x, weight = NULL){
    if(NCOL(x)>1){
        # for multiple choice
        vec = rowSums(!is.na(x))==0
        
    } else {
        vec = is.na(x)
    }
    w_sum(vec, weight = weight)
}

mtcars %>% 
    tab_cols(total(), vs) %>% 
    tab_cells(am) %>% 
    tab_stat_cpct() %>% 
    tab_stat_fun(count_na, label = 'NA') %>% 
    tab_pivot()

[JB0207]

Thank you for your reply, clarification on tab_last_sig_cpct and your suggestion for a solution counting NAs. Unfortunately, the function does not work correctly in combination with subtotal. Example code for error replication:

# set digits
expss_digits(digits = 2)

# set output option
expss_output_viewer()

# for replication
set.seed(123)

test_df <- data.frame(
  Y1 = sample(1:3, 100, replace = TRUE),
  Q1 = sample(0:1, 100, replace = TRUE),
  Q3 = sample(c(1:5, NA), 100, replace = TRUE, prob = c(rep(1, 5), 0.1)),
  Q4 = sample(c(1:5, NA), 100, replace = TRUE, prob = c(rep(1, 5), 0.1))
)

var_lab(test_df$Q1) = "Test1"
var_lab(test_df$Q3) = "Test2"
var_lab(test_df$Q4) = "Test3"

test_df %>% 
  tab_cols( 
    total(),
    Y1
  ) %>%
  tab_cells( 
    Q1,
    subtotal(Q3 %to% Q4, "Bottom 2 Box" = 1:2, "Top 2 Box" = 4:5, position = "bottom") 
  ) %>%
  tab_stat_cpct(label="%", total_statistic = "w_cpct", total_label = "#Total cases") %>%
  tab_stat_fun(count_na, label = 'NA') %>% 
  tab_pivot()

But when I specified NA within the subtotal command, it actually worked, but the count of NAs is displayed twice as category 6 (not wanted) and as defined by me with "NA". With hide I could not hide the unwanted category 6 in the output. Is there any way to do this, or is it possible to avoid the automatic setting of a value label?

Code:

test_df %>%
  tab_cols(total(),
           Y1) %>%
  tab_cells(Q1,
            subtotal(Q3 %to% Q4, "Bottom 2 Box" = 1:2, "Top 2 Box" = 4:5, "NA" = NA, position = "bottom")
            ) %>%
  tab_stat_cases(label = "Cases", total_row_position = "below") %>%
  tab_stat_cpct(label = "%", total_statistic = "w_cpct", total_label = "#Total cases") %>%
  tab_pivot(stat_position = "inside_columns")

[gdemin]

Which version of the expss dou you use? hide works for me:

test_df %>%
    tab_cols(total(),
             Y1) %>%
    tab_cells(Q1,
              subtotal(Q3 %to% Q4, "Bottom 2 Box" = 1:2, "Top 2 Box" = 4:5, "NA" = hide(NA), position = "bottom")
    ) %>%
    tab_stat_cases(label = "Cases", total_row_position = "below") %>%
    tab_stat_cpct(label = "%", total_statistic = "w_cpct", total_label = "#Total cases") %>%
    tab_pivot(stat_position = "inside_columns")


 # |       |              | #Total |     |    Y1 |        |       |        |       |        |
 # |       |              |  Cases |   % |     1 |        |     2 |        |     3 |        |
 # |       |              |        |     | Cases |      % | Cases |      % | Cases |      % |
 # | ----- | ------------ | ------ | --- | ----- | ------ | ----- | ------ | ----- | ------ |
 # | Test1 |            0 |     44 |  44 |    16 |  48.48 |    11 |  34.38 |    17 |  48.57 |
 # |       |            1 |     56 |  56 |    17 |  51.52 |    21 |  65.62 |    18 |  51.43 |
 # |       | #Total cases |    100 | 100 |    33 | 100.00 |    32 | 100.00 |    35 | 100.00 |
 # | Test2 |            1 |     16 |  16 |     5 |  15.15 |     9 |  28.12 |     2 |   5.71 |
 # |       |            2 |     23 |  23 |     7 |  21.21 |     7 |  21.88 |     9 |  25.71 |
 # |       |            3 |     20 |  20 |     5 |  15.15 |     7 |  21.88 |     8 |  22.86 |
 # |       |            4 |     21 |  21 |     7 |  21.21 |     5 |  15.62 |     9 |  25.71 |
 # |       |            5 |     19 |  19 |     9 |  27.27 |     4 |  12.50 |     6 |  17.14 |
 # |       | Bottom 2 Box |     39 |  39 |    12 |  36.36 |    16 |  50.00 |    11 |  31.43 |
 # |       |    Top 2 Box |     40 |  40 |    16 |  48.48 |     9 |  28.12 |    15 |  42.86 |
 # |       |           NA |      1 |   1 |       |        |       |        |     1 |   2.86 |
 # |       | #Total cases |    100 | 100 |    33 | 100.00 |    32 | 100.00 |    35 | 100.00 |
 # | Test3 |            1 |     25 |  25 |     5 |  15.15 |    11 |  34.38 |     9 |  25.71 |
 # |       |            2 |     19 |  19 |     6 |  18.18 |     6 |  18.75 |     7 |  20.00 |
 # |       |            3 |     15 |  15 |     9 |  27.27 |     5 |  15.62 |     1 |   2.86 |
 # |       |            4 |     19 |  19 |     8 |  24.24 |     5 |  15.62 |     6 |  17.14 |
 # |       |            5 |     17 |  17 |     4 |  12.12 |     4 |  12.50 |     9 |  25.71 |
 # |       | Bottom 2 Box |     44 |  44 |    11 |  33.33 |    17 |  53.12 |    16 |  45.71 |
 # |       |    Top 2 Box |     36 |  36 |    12 |  36.36 |     9 |  28.12 |    15 |  42.86 |
 # |       |           NA |      5 |   5 |     1 |   3.03 |     1 |   3.12 |     3 |   8.57 |
 # |       | #Total cases |    100 | 100 |    33 | 100.00 |    32 | 100.00 |    35 | 100.00 |

[JB0207]

Ah, I was using hide incorrectly. Now it works! Is there a way to remove the NA count from the total number of cases and the percentage column so that for Variable Q3 ("Test2") n=99 is used as the base and for Variable Q4 ("Test3") n=95 is used as the base?

[gdemin]

Sorry for delay. Below the code - it is not very nice but works. The idea is to reorder rows according to variable labels.

# set digits
library(expss)
expss_digits(digits = 2)

# set output option
expss_output_default()

count_na = function(x, weight = NULL){
    if(NCOL(x)>1){
        # for multiple choice
        vec = rowSums(!is.na(x))==0
        
    } else {
        vec = is.na(x)
    }
    w_sum(vec, weight = weight)
}

tab_reorder = function(data){
    to_split = gsub('^([^|]+).+$', "\\1", data$row_labels)
    keep_order = unique(to_split)
    res = split(data, to_split) 
    do.call(add_rows, res[keep_order])   
}

# for replication
set.seed(123)

test_df <- data.frame(
    Y1 = sample(1:3, 100, replace = TRUE),
    Q1 = sample(0:1, 100, replace = TRUE),
    Q3 = sample(c(1:5, NA), 100, replace = TRUE, prob = c(rep(1, 5), 0.1)),
    Q4 = sample(c(1:5, NA), 100, replace = TRUE, prob = c(rep(1, 5), 0.1))
)

var_lab(test_df$Q1) = "Test1"
var_lab(test_df$Q3) = "Test2"
var_lab(test_df$Q4) = "Test3"

test_df %>% 
    tab_cols( 
        total(),
        Y1
    ) %>%
    tab_cells( 
        Q1,
        subtotal(Q3 %to% Q4, "Bottom 2 Box" = 1:2, "Top 2 Box" = 4:5, position = "bottom") 
    ) %>%
    tab_stat_cpct(label="%", total_statistic = "w_cpct", total_label = "#Total cases") %>%
    tab_cells(Q1, Q3 %to% Q4) %>%  
    tab_stat_fun(count_na, label = 'NA') %>% 
    tab_pivot() %>% 
    tab_reorder()