Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Biostatistics / Data and description

Summarising data: centre, spread and shape

15 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Calculate and interpret the mean, median and mode, and know when each is the right choice

  • Calculate and interpret the range, interquartile range, variance and standard deviation

  • Recognise skewed and symmetric distributions and see how outliers move the mean

A list of 400 CD4 counts is unreadable. A summary replaces it with a few numbers that capture what matters: where the data are centred (location), how far they spread (dispersion) and what shape they take. A good summary is honest: it never hides that a handful of extreme values are pulling the average.

Measures of centre

MeasureHow to get itStrengthWeakness
MeanAdd all values, divide by nUses every value; basis of most testsPulled by outliers and skew
MedianMiddle value when sorted (average of the two middle values if n is even)Not affected by outliersIgnores the size of extremes
ModeMost frequent valueWorks for categoriesMay not exist or be unique

The mean

x = (1)/(n)Σ_(i=1)^n x_i

For a sample of n values, the sample mean x is the sum divided by n.

Length of hospital stay: one outlier changes the story

Eleven patients stayed in hospital for (days): 2, 3, 3, 4, 4, 5, 5, 6, 7, 9, 30.

  • Sum = 78, so the mean = 78 / 11 = 7.1 days.
  • Sorted, the middle (6th) value is 5, so the median is 5 days.
  • Remove the 30-day patient: the mean drops to 4.8 days, and the median to 4.5 days.

One patient moved the mean by 2.3 days but the median by only 0.5. Which do you report? When data are skewed (a long tail to one side) or contain outliers, the median is the better description of the "typical" patient. Hospital stay, income, viral load and many biological measures are right-skewed.

010203040Length of stay (days)Patients123median456789101112+
A right-skewed distribution: the long tail pulls the mean above the median, so the median describes the typical patient better.

Measures of spread

Two clinics can both have a mean of 120 mmHg systolic blood pressure, yet one has everyone between 115 and 125 and the other between 90 and 160. The mean alone cannot tell them apart.

  • Range: maximum minus minimum. Simple but depends on two values only, so one outlier ruins it.
  • Quartiles and interquartile range (IQR): Q1 is the median of the lower half of the sorted data, Q3 the median of the upper half; IQR = Q3 − Q1 contains the middle 50%. For our stay data (excluding the median from each half): Q1 = 3 and Q3 = 7, so IQR = 4 days. Reported as "median 5 days (IQR 3 to 7)".
  • Variance and standard deviation (SD): the SD is roughly the typical distance of values from the mean.

Sample variance and standard deviation

s^2 = (Σ (x_i - x)^2)/(n-1), s = sqrt(s^2)

We divide by n-1, not n, because the mean was estimated from the same data, which makes the deviations slightly too small (Bessel's correction).

Calculating the SD by hand

Five systolic readings: 110, 120, 125, 130, 140. The mean is 125.

Deviations: −15, −5, 0, 5, 15. Squared: 225, 25, 0, 25, 225. Sum = 500.

Variance = 500 / (5 − 1) = 125, so SD = √125 = 11.2 mmHg.

For our hospital-stay data the SD is 7.9 days: larger than the median itself, a warning sign of a skewed distribution. As a rule of thumb, if the mean minus 2 SD falls below the smallest possible value (here below zero days), the data are not symmetric.

Shape: symmetric, skewed, and how the three centres line up

In a symmetric distribution (for example adult height) the mean, median and mode are about equal. In a right-skewed distribution the tail points to the high values and mean > median > mode; in a left-skewed distribution the order is reversed (age at death in a high-income country is left-skewed).

What to report

Data shapeCentreSpreadExample
Roughly symmetricMeanSDHeight 165.2 cm (SD 8.1)
Skewed or with outliersMedianIQR (give Q1 and Q3)Stay 5 days (IQR 3 to 7)
CategoriesPercentage in eachNot applicable57% female
Wide range of positive valuesGeometric meanRatio-scale SDViral load, antibody titres

Geometric mean for multiplicative data

Antibody titres and viral loads vary by factors of 10, so work on the log scale: take logarithms, calculate the mean and then back-transform (exponentiate). The result is the geometric mean, a sensible "typical" value for data whose spread is multiplicative. Always say which mean you are reporting.

Mistakes to avoid

  • Reporting the mean and SD of a clearly skewed variable, which can give impossible values (a mean minus 2 SD of negative days).
  • Calling the SD the "standard error". They are different: the SD describes variation between individuals, the standard error describes uncertainty in the mean (covered in chapter 3).
  • Reporting a mean to more precision than the data support (a mean of 5.4382 days from whole-day data).

Say what you used

Always state the summary you used and its spread: "mean 125 mmHg (SD 11.2)" or "median 5 days (IQR 3 to 7)". A bare number with no measure of spread is not a summary.

Summary

  • The mean uses every value but is pulled by outliers; the median is robust and best for skewed data.

  • Report spread with the SD for symmetric data, or the IQR (Q1 and Q3) for skewed data.

  • The SD divides by n − 1 because the mean was estimated from the same data.

  • In right-skewed data mean > median; in left-skewed data mean < median.

  • Use the geometric mean for data that vary by factors (titres, viral loads).

Check your understanding

Question 1
1

Eleven hospital stays (days) are 2, 3, 3, 4, 4, 5, 5, 6, 7, 9, 30. What is the median?

Question 2
2

Which summary is best for a clearly right-skewed variable such as length of hospital stay?

Question 3
3

Five readings are 110, 120, 125, 130, 140 (mean 125). What is the sample SD?

Question 4
4

Why does the sample variance divide by n − 1 rather than n?

Question 5
5

In a right-skewed distribution, which ordering is correct?

Question 6
6

Viral loads in a trial range from 50 to 5,000,000 copies/mL. A sensible 'typical' value is:

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Types of data and scales of measurementGraphs that tell the truth