Descriptive Statistics — Part IV: Reading Charts Critically and a Table 1

Before you begin, take a few minutes to check what you already know and how confident you feel. You will see the same questions again at the end of the module — this helps both you and us measure what you have learned. Click an option for every question and confidence rating, then click Next to continue.

In Part III you drew the standard figures — histograms, density curves, boxplots, scatters, and facets. Drawing a chart is only half the skill. The same data, plotted differently, can tell opposite stories, and the table that opens a clinical paper can be honest or quietly misleading. This part shows you how to read a figure with suspicion and how to build the baseline table reviewers expect.

This part of the module covers two skills, from critique to a finished table:

• Reading a chart critically — watch where the y-axis begins, remember that a histogram's shape is an artefact of the binwidth you chose, and ask what a summary hides.
• The publication-ready Table 1 — tbl_summary(by = arm) with add_overall(), add_p(), a missing-value row, and a caption — plus how to read the result: why continuous variables default to median (IQR), when mean (SD) is safe, and why a baseline p-value is rarely informative.

By the end of this part you will be able to spot a truncated axis, an unstable binwidth, and an over-collapsed summary in someone else's figure; produce a stratified, publication-ready Table 1 with gtsummary using tbl_summary(by = ...), add_overall(), and add_p(); and read a published one with a sceptical eye.

Try every snippet in the R Scratchpad on the right. This part needs no data file — each block first builds the same simulated 200-patient cohort with set.seed() and tibble(), so you can run any block on its own and watch the table appear.

You can now draw the standard figures. Just as important is reading them with suspicion — because the same data, plotted differently, can tell opposite stories. Three habits protect you.

First, watch the y-axis. A bar chart whose y-axis starts at 90 instead of 0 turns a trivial difference into a dramatic-looking cliff. A truncated axis exaggerates differences — check where the axis begins before you believe the gap.

Second, remember the binwidth. You already saw that a histogram's shape is an artefact of the slice width you chose. A reader who only sees one binwidth is seeing one of many possible pictures.

Third, ask what a summary hides. A boxplot or a bar of means collapses every patient into a few numbers. The honest move on a modest sample is to show the points — geom_jitter() over a boxplot reveals gaps, clumps, and a tiny-n group masquerading as a confident summary.

Every clinical paper opens with Table 1: the baseline characteristics of the patients, usually split by treatment arm. It reports each variable's centre and spread (for continuous data) or count and percent (for categories), so a reader can judge whether the groups started out comparable.

You build it with tbl_summary from the gtsummary package. Pass the tibble and a by = argument naming the grouping column, and gtsummary picks a sensible summary for each variable type automatically.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(gtsummary)
cohort |>
  tbl_summary(by = arm)

By default gtsummary reports a continuous variable as median (IQR) — the median and the interquartile range. That default is deliberate and clinically right, because reporting mean (SD) for a skewed lab like CRP misleads: a few extreme values drag the mean up to a number no typical patient has. Median (IQR) is robust to that long tail. Reach for mean (SD) only when you have checked the variable is roughly symmetric.

Three additions turn the bare table into a publishable one. add_overall puts a whole-cohort column beside the per-arm columns. add_p adds a column of p-values comparing the arms. And a caption names the table. gtsummary also adds an Unknown row for any variable with missing values, so the denominators are honest.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(gtsummary)
cohort |>
  tbl_summary(by = arm) |>
  add_overall() |>
  add_p() |>
  modify_caption("**Table 1. Baseline characteristics by treatment arm**")

The worked example below fades its support: first you study a full Table 1, then you fill the gap, then you build a fresh table on your own. Run each stage in the scratchpad.

library(dplyr)
set.seed(2024)
n <- 200
cohort <- tibble(
  arm = factor(sample(c("Standard", "Intensive"), n, replace = TRUE)),
  sex = factor(sample(c("Female", "Male"), n, replace = TRUE)),
  age = round(rnorm(n, mean = 62, sd = 9)),
  bmi = round(rnorm(n, mean = 28, sd = 4), 1),
  crp = round(rexp(n, rate = 1/5), 1),
  hba1c = round(rnorm(n, mean = 7.8, sd = 1.1), 1),
  responder = factor(sample(c("Yes", "No"), n, replace = TRUE))
)
library(gtsummary)
tbl <- tbl_summary(cohort, by = arm, missing = "ifany")
tbl <- add_overall(tbl)
tbl <- add_p(tbl)
tbl

You have reached the end of the module. Try the same questions again — your answers here, paired with your pre-test answers, are how we measure what the module taught you. Answer every question and confidence rating, then click Submit and see results to view your score.

Here is how your post-test answers compare with your pre-test answers. The pre/post pairing is the most reliable way to see what this module actually taught you.