Exploratory Analysis
Jana Stoilova, Adrian Waddell and Gabriel Becker
2024-04-20
Source:vignettes/exploratory_analysis.Rmd
exploratory_analysis.RmdIntroduction
In this vignette, we would like to introduce how
qtable() can be used to easily create cross tabulations for
exploratory data analysis. qtable() is an extension of
table() from base R and can do much beyond creating two-way
contingency tables. The function has a simple to use interface while
internally it builds layouts using the rtables
framework.
Getting Started
Load packages used in this vignette:
Let’s start by seeing what table() can do:
table(ex_adsl$ARM)#
# A: Drug X B: Placebo C: Combination
# 134 134 132
table(ex_adsl$SEX, ex_adsl$ARM)#
# A: Drug X B: Placebo C: Combination
# F 79 77 66
# M 51 55 60
# U 3 2 4
# UNDIFFERENTIATED 1 0 2We can easily recreate the cross-tables above with
qtable() by specifying a data.frame with variable(s) to
tabulate. The col_vars and row_vars arguments
control how to split the data across columns and rows respectively.
qtable(ex_adsl, col_vars = "ARM")# A: Drug X B: Placebo C: Combination
# (N=134) (N=134) (N=132)
# ———————————————————————————————————————————————
# count 134 134 132
qtable(ex_adsl, col_vars = "ARM", row_vars = "SEX")# A: Drug X B: Placebo C: Combination
# count (N=134) (N=134) (N=132)
# ——————————————————————————————————————————————————————————
# F 79 77 66
# M 51 55 60
# U 3 2 4
# UNDIFFERENTIATED 1 0 2Aside from the display style, the main difference is that
qtable() will add (N=xx) in the table header by default.
This can be removed with show_colcounts.
qtable(ex_adsl, "ARM", show_colcounts = FALSE)# count all obs
# ————————————————————————
# A: Drug X 134
# B: Placebo 134
# C: Combination 132Any variables used as the row or column facets should not have any empty strings (““). This is because non empty values are required as labels when generating the table. The code below will generate an error.
Nested Tables
Providing more than one variable name for the row or column structure
in qtable() will create a nested table. Arbitrary nesting
is supported in each dimension.
# A: Drug X B: Placebo C: Combination
# S1 S2 S1 S2 S1 S2
# count (N=73) (N=61) (N=67) (N=67) (N=56) (N=76)
# ————————————————————————————————————————————————————————————————————————
# F
# A 12 9 11 13 7 11
# B 14 11 12 15 9 12
# C 17 16 13 13 14 13
# M
# A 5 11 10 9 6 14
# B 13 8 7 10 9 12
# C 8 6 13 6 8 11
# U
# A 1 0 1 0 1 0
# B 1 0 0 1 0 1
# C 1 0 0 0 1 1
# UNDIFFERENTIATED
# A 0 0 0 0 0 1
# C 1 0 0 0 1 0Note that by default, unobserved factor levels within a facet are not
included in the table. This can be modified with
drop_levels. The code below adds a row of 0s for
STRATA1 level “B” nested under the SEX level
“UNDIFFERENTIATED”.
qtable(
ex_adsl,
row_vars = c("SEX", "STRATA1"),
col_vars = c("ARM", "STRATA2"),
drop_levels = FALSE
)# A: Drug X B: Placebo C: Combination
# S1 S2 S1 S2 S1 S2
# count (N=73) (N=61) (N=67) (N=67) (N=56) (N=76)
# ————————————————————————————————————————————————————————————————————————
# F
# A 12 9 11 13 7 11
# B 14 11 12 15 9 12
# C 17 16 13 13 14 13
# M
# A 5 11 10 9 6 14
# B 13 8 7 10 9 12
# C 8 6 13 6 8 11
# U
# A 1 0 1 0 1 0
# B 1 0 0 1 0 1
# C 1 0 0 0 1 1
# UNDIFFERENTIATED
# A 0 0 0 0 0 1
# B 0 0 0 0 0 0
# C 1 0 0 0 1 0In contrast, table() cannot return a nested table.
Rather it produces a list of contingency tables when more than two
variables are used as inputs.
table(ex_adsl$SEX, ex_adsl$STRATA1, ex_adsl$ARM, ex_adsl$STRATA2)# , , = A: Drug X, = S1
#
#
# A B C
# F 12 14 17
# M 5 13 8
# U 1 1 1
# UNDIFFERENTIATED 0 0 1
#
# , , = B: Placebo, = S1
#
#
# A B C
# F 11 12 13
# M 10 7 13
# U 1 0 0
# UNDIFFERENTIATED 0 0 0
#
# , , = C: Combination, = S1
#
#
# A B C
# F 7 9 14
# M 6 9 8
# U 1 0 1
# UNDIFFERENTIATED 0 0 1
#
# , , = A: Drug X, = S2
#
#
# A B C
# F 9 11 16
# M 11 8 6
# U 0 0 0
# UNDIFFERENTIATED 0 0 0
#
# , , = B: Placebo, = S2
#
#
# A B C
# F 13 15 13
# M 9 10 6
# U 0 1 0
# UNDIFFERENTIATED 0 0 0
#
# , , = C: Combination, = S2
#
#
# A B C
# F 11 12 13
# M 14 12 11
# U 0 1 1
# UNDIFFERENTIATED 1 0 0With some help from stats::ftable() the nested structure
can be achieved in two steps.
t1 <- ftable(ex_adsl[, c("SEX", "STRATA1", "ARM", "STRATA2")])
ftable(t1, row.vars = c("SEX", "STRATA1"))# ARM A: Drug X B: Placebo C: Combination
# STRATA2 S1 S2 S1 S2 S1 S2
# SEX STRATA1
# F A 12 9 11 13 7 11
# B 14 11 12 15 9 12
# C 17 16 13 13 14 13
# M A 5 11 10 9 6 14
# B 13 8 7 10 9 12
# C 8 6 13 6 8 11
# U A 1 0 1 0 1 0
# B 1 0 0 1 0 1
# C 1 0 0 0 1 1
# UNDIFFERENTIATED A 0 0 0 0 0 1
# B 0 0 0 0 0 0
# C 1 0 0 0 1 0NA Values
So far in all the examples we have seen, we used counts to summarize
the data in each table cell as this is the default analysis used by
qtable(). Internally, a single analysis variable specified
by avar is used to generate the counts in the table. The
default analysis variable is the first variable in data. In
the case of ex_adsl this is “STUDYID”.
Let’s see what happens when we introduce some NA values
into the analysis variable:
tmp_adsl <- ex_adsl
tmp_adsl[[1]] <- NA_character_
qtable(tmp_adsl, row_vars = "ARM", col_vars = "SEX")# F M U UNDIFFERENTIATED
# count (N=222) (N=166) (N=9) (N=3)
# —————————————————————————————————————————————————————————————
# A: Drug X 0 0 0 0
# B: Placebo 0 0 0 0
# C: Combination 0 0 0 0The resulting table is showing 0’s across all cells because all the
values of the analysis variable are NA.
Keep this behavior in mind when doing quick exploratory analysis
using the default counts aggregate function of qtable.
If this does not suit your purpose, you can either pre-process your
data to re-code the NA values or use another analysis
function. We will see how the latter is done in the Custom Aggregation section.
# Recode NA values
tmp_adsl[[1]] <- addNA(tmp_adsl[[1]])
qtable(tmp_adsl, row_vars = "ARM", col_vars = "SEX")# F M U UNDIFFERENTIATED
# count (N=222) (N=166) (N=9) (N=3)
# —————————————————————————————————————————————————————————————
# A: Drug X 79 51 3 1
# B: Placebo 77 55 2 0
# C: Combination 66 60 4 2In addition, row and column variables should have NA
levels explicitly labelled as above. If this is not done, the columns
and/or rows will not reflect the full data.
tmp_adsl$new1 <- factor(NA_character_, levels = c("X", "Y", "Z"))
qtable(tmp_adsl, row_vars = "ARM", col_vars = "new1")# X Y Z
# count (N=0) (N=0) (N=0)
# ——————————————————————————————————————
# A: Drug X 0 0 0
# B: Placebo 0 0 0
# C: Combination 0 0 0Explicitly labeling the NA levels in the column facet
adds a column to the table:
tmp_adsl$new2 <- addNA(tmp_adsl$new1)
levels(tmp_adsl$new2)[4] <- "<NA>" # NA needs to be a recognizible string
qtable(tmp_adsl, row_vars = "ARM", col_vars = "new2")# X Y Z <NA>
# count (N=0) (N=0) (N=0) (N=400)
# ————————————————————————————————————————————————
# A: Drug X 0 0 0 134
# B: Placebo 0 0 0 134
# C: Combination 0 0 0 132Custom Aggregation
A powerful feature of qtable() is that the user can
define the type of function used to summarize the data in each facet. We
can specify the type of analysis summary using the afun
argument:
qtable(ex_adsl, row_vars = "STRATA2", col_vars = "ARM", avar = "AGE", afun = mean)# A: Drug X B: Placebo C: Combination
# AGE - mean (N=134) (N=134) (N=132)
# ————————————————————————————————————————————————————
# S1 34.10 36.46 35.70
# S2 33.38 34.40 35.24Note that the analysis variable AGE and analysis
function name are included in the top right header of the table.
If the analysis function returns a vector of 2 or 3 elements, the result is displayed in multi-valued single cells.
qtable(ex_adsl, row_vars = "STRATA2", col_vars = "ARM", avar = "AGE", afun = range)# A: Drug X B: Placebo C: Combination
# AGE - range (N=134) (N=134) (N=132)
# ————————————————————————————————————————————————————————
# S1 23.0 / 48.0 24.0 / 62.0 20.0 / 69.0
# S2 21.0 / 50.0 21.0 / 58.0 23.0 / 64.0If you want to use an analysis function with more than 3 summary elements, you can use a list. In this case, the values are displayed in the table as multiple stacked cells within each facet. If the list elements are named, the names are used as row labels.
fivenum2 <- function(x) {
setNames(as.list(fivenum(x)), c("min", "Q1", "MED", "Q3", "max"))
}
qtable(ex_adsl, row_vars = "STRATA2", col_vars = "ARM", avar = "AGE", afun = fivenum2)# A: Drug X B: Placebo C: Combination
# AGE - fivenum2 (N=134) (N=134) (N=132)
# ————————————————————————————————————————————————————————
# S1
# min 23.00 24.00 20.00
# Q1 28.00 30.00 30.50
# MED 34.00 36.00 35.00
# Q3 39.00 40.50 40.00
# max 48.00 62.00 69.00
# S2
# min 21.00 21.00 23.00
# Q1 29.00 29.50 30.00
# MED 32.00 32.00 34.50
# Q3 38.00 39.50 38.00
# max 50.00 58.00 64.00More advanced formatting can be controlled with
in_rows(). See function documentation for more details.
meansd_range <- function(x) {
in_rows(
"Mean (sd)" = rcell(c(mean(x), sd(x)), format = "xx.xx (xx.xx)"),
"Range" = rcell(range(x), format = "xx - xx")
)
}
qtable(ex_adsl, row_vars = "STRATA2", col_vars = "ARM", avar = "AGE", afun = meansd_range)# A: Drug X B: Placebo C: Combination
# AGE - meansd_range (N=134) (N=134) (N=132)
# —————————————————————————————————————————————————————————————————
# S1
# Mean (sd) 34.10 (6.71) 36.46 (7.72) 35.70 (8.22)
# Range 23 - 48 24 - 62 20 - 69
# S2
# Mean (sd) 33.38 (6.40) 34.40 (7.99) 35.24 (7.39)
# Range 21 - 50 21 - 58 23 - 64Marginal Summaries
Another feature of qtable() is the ability to quickly
add marginal summary rows with the summarize_groups
argument. This summary will add to the table the count of non-NA records
of the analysis variable at each level of nesting. For example, compare
these two tables:
# A: Drug X B: Placebo C: Combination
# AGE - mean (N=134) (N=134) (N=132)
# ————————————————————————————————————————————————————
# A
# S1 31.61 36.68 34.00
# S2 34.40 33.55 34.35
# B
# S1 34.57 37.68 35.83
# S2 32.79 34.77 36.68
# C
# S1 35.26 35.38 36.58
# S2 32.95 34.89 34.72
qtable(
ex_adsl,
row_vars = c("STRATA1", "STRATA2"), col_vars = "ARM",
summarize_groups = TRUE, avar = "AGE", afun = mean
)# A: Drug X B: Placebo C: Combination
# AGE - mean (N=134) (N=134) (N=132)
# —————————————————————————————————————————————————————————
# A 38 (28.4%) 44 (32.8%) 40 (30.3%)
# S1 18 (13.4%) 22 (16.4%) 14 (10.6%)
# AGE - mean 31.61 36.68 34.00
# S2 20 (14.9%) 22 (16.4%) 26 (19.7%)
# AGE - mean 34.40 33.55 34.35
# B 47 (35.1%) 45 (33.6%) 43 (32.6%)
# S1 28 (20.9%) 19 (14.2%) 18 (13.6%)
# AGE - mean 34.57 37.68 35.83
# S2 19 (14.2%) 26 (19.4%) 25 (18.9%)
# AGE - mean 32.79 34.77 36.68
# C 49 (36.6%) 45 (33.6%) 49 (37.1%)
# S1 27 (20.1%) 26 (19.4%) 24 (18.2%)
# AGE - mean 35.26 35.38 36.58
# S2 22 (16.4%) 19 (14.2%) 25 (18.9%)
# AGE - mean 32.95 34.89 34.72In the second table, there are marginal summary rows for each level
of the two row facet variables: STRATA1 and
STRATA2. The number 18 in the second row gives the count of
observations part of ARM level “A: Drug X”,
STRATA1 level “A”, and STRATA2 level “S1”. The
percent is calculated as the cell count divided by the column count
given in the table header. So we can see that the mean AGE
of 31.61 in that subgroup is based on 18 subjects which correspond to
13.4% of the subjects in arm “A: Drug X”.
See ?summarize_row_groups for how to add marginal
summary rows when using the core rtables framework.
Table Decorations
Tables generated with qtable() can include annotations
such as titles, subtitles and footnotes like so:
qtable(
ex_adsl,
row_vars = "STRATA2", col_vars = "ARM",
title = "Strata 2 Summary",
subtitle = paste0("STUDY ", ex_adsl$STUDYID[1]),
main_footer = paste0("Date: ", as.character(Sys.Date()))
)# Strata 2 Summary
# STUDY AB12345
#
# ———————————————————————————————————————————————
# A: Drug X B: Placebo C: Combination
# count (N=134) (N=134) (N=132)
# ———————————————————————————————————————————————
# S1 73 67 56
# S2 61 67 76
# ———————————————————————————————————————————————
#
# Date: 2024-04-20Summary
Here is what we have learned in this vignette:
qtable()can replace and extend uses oftable()andstats::ftable()qtable()is useful for exploratory data analysis
As the intended use of qtable() is for exploratory data
analysis, there is limited functionality for building very complex
tables. For details on how to get started with the core
rtables layout functionality see the introduction
vignette.