Lesson 8 of 8 · 15 min

Choosing statistics and knowing their limits

Decide which measures of centre and spread suit a data set, explain how outliers and skew affect each statistic, and recognise what summary statistics can't tell you.

What you will be able to do

  • Explain why the median and IQR are resistant to outliers while the mean, range and standard deviation are not.
  • Choose measures of centre and spread to suit the shape of a distribution and justify the choice.
  • Describe the limitations of each summary statistic.

A Gold Coast surf shop advertises for staff: "Our team earns an average of $80,200 a year." Here are the annual salaries of all nine people who work there, in thousands of dollars: 52, 55, 58, 60, 61, 63, 65, 68 and 240. The 240 is the owner–manager.

The mean really is about $80,200. But eight of the nine staff earn less than $70,000. Is the ad honest?

Interactive

One salary, two stories

Drag the owner's salary (240) down to 70. The mean drops by about $19,000 and the standard deviation falls from about 60 to about 6, but the median and IQR don't change at all.

The surf shop salaries, with and without the owner–manager.
StatisticAll nine staffWithout the owner
Mean$80,200$60,250
Median$61,000$60,500
Range$188,000$16,000
IQR$10,000$7,500
Standard deviation sxs_x$60,100$5,200

Key idea

Resistant and sensitive statistics

The median and the IQR are resistant: an extreme value barely moves them, because they depend only on the middle of the ordered data. The mean, the range and the standard deviation are sensitive: they use the actual size of the extreme values, so a single outlier can change them a lot.

So the ad's "average" is the mean, dragged up by one salary. The median, $61,000, is much closer to what a new employee could expect. That leads to a simple rule for choosing statistics: match them to the shape of the data.

Choosing measures of centre and spread.
The dataCentreSpread
Symmetric, no outliersMeanStandard deviation
Skewed, or with outliersMedianIQR
CategoricalModal categoryNone: report the percentage in each category

For which data set is the median a better measure of centre than the mean?

Choose one answer

Every statistic throws some information away. Knowing what each one ignores is how you choose well, and how you spot a statistic being used to mislead.

What each statistic is good at, and what it misses.
StatisticStrengthLimitation
MeanUses every valuePulled towards outliers and the tail of skewed data; may not be a possible value (1.8 children)
MedianResistant to outliers and skewIgnores how large or small the other values are
ModeThe only centre for categorical dataThere may be no mode or several, and it can sit far from the centre
RangeQuick to findDepends only on the two most extreme values
IQRResistant; describes the middle 50%Ignores the top and bottom quarters of the data
Standard deviationUses every valueSensitive to outliers; hard to interpret for skewed data

Note

Statistics hide the shape

One limitation applies to every summary statistic: it hides the shape of the data. A bimodal distribution and a symmetric one can have the same mean and the same standard deviation. Always look at a display before you summarise, and report the shape alongside the numbers.

The data set 3, 4, 4, 5, 6, 20 has a mean of 7. The value 20 turns out to be a recording error and is removed. By how much does the mean decrease?

Worked example

Is the advertised "average rent" fair?

  • Complex unfamiliar
  • Technology active

A real estate agent's flyer says "Average weekly rent for two-bedroom units in this Townsville street: $471". The weekly rents of the 10 two-bedroom units in the street are:

$380, $395, $400, $410, $420, $420, $430, $445, $460, $950

The $950 unit is a renovated beachfront apartment. Evaluate the flyer's claim, and recommend the statistics the agent should report for the centre and spread of these rents.

5 steps

Try it yourself first, then check the working one step at a time.

Method

  1. Check the claimed statistic.

  2. Look at the shape and any outliers.

  3. Calculate the resistant statistics (median and IQR) and compare them with the sensitive ones.

  4. Make a judgement, justify it with values from the data, and recommend.

Watch out for

  • Saying the claim is false. The mean is calculated correctly; it is the choice of statistic that misleads.

  • Recommending the mean and standard deviation for skewed data with an outlier.

Common mistake

"The mean is the average, so it's always best"

The mean, median and mode are all averages. Choose the one that suits the shape of the data, and say why. And never remove an outlier just because it changes the answer: remove it only if it is an error, and say that you did.

Exam tip

Justifying a choice

Name the statistic, give the reason from this data set, and quote the values. For example: "The median ($420) is more appropriate than the mean ($471), because the $950 rent makes the distribution positively skewed and pulls the mean above 9 of the 10 rents."

Summary

  • The median and IQR are resistant to outliers and skew. The mean, range and standard deviation are not.

  • Symmetric data with no outliers: report the mean and standard deviation. Skewed data or outliers: report the median and IQR.

  • Categorical data: report the modal category and percentages.

  • Every statistic has limitations, and all of them hide the shape, so look at a display first.

  • Justify a choice with the shape or outlier in the data, and quote the values.

Finished the lesson?

Try the 2 quick checks above and the lesson completes itself, or mark it complete now.

Lesson 8 of 80% read

Save your progress with a free account