Skewness and outliers
Studying this course? Keep your progress in one place.
Key facts
- Positive skew has a longer upper tail: and mean median mode.
- Negative skew has a longer lower tail: and mean median mode.
- The rule for outliers is given in the question: commonly beyond from the quartiles, or beyond standard deviations from the mean.
- On a box plot, whiskers stop at the most extreme non-outlier and outliers are marked with crosses.
- Outliers affect the mean and especially the standard deviation, but not the median or IQR much.
Describing skewness
Skewness describes the symmetry of a distribution. The skew is named after the longer tail, and quartile gaps show which tail is longer.
Skewness describes the symmetry of a distribution.
- If the upper tail is longer, the distribution has positive skew.
- If the lower tail is longer, it has negative skew.
- A symmetrical distribution has no skew.
Using quartiles, compare the two gaps and .
Positive skew
- Longer upper tail
Symmetrical
- No skew
Negative skew
- Longer lower tail
Worked example
, , . Describe the skew.
- 1
.
- 2
.
- 3
, so the upper half is more spread out.
, , . What is the skew?
Skewness from mean, median and mode
In a positively skewed distribution the mean is pulled above the median and mode by high values; in a negatively skewed one it is pulled below.
For a positively skewed distribution, extreme high values pull the mean up most. For a negatively skewed distribution, extreme low values pull the mean down most.
For a symmetrical distribution the three averages are equal or very close.
Example: mode 8, median 10 and mean 12 indicates positive skew. This is why the median is often preferred for skewed data such as incomes: it is not distorted by a few very large values.
Positive skew
- mean median mode
Symmetrical
- mean = median = mode (or very close)
Negative skew
- mean median mode
A distribution has mean 35, median 40 and mode 44. What is the skew?
Outliers
An outlier is a value unusually far from the rest. The identification rule is always given in the question.
An outlier is a value that is unusually far from the rest of the data. Outliers may be genuine extreme values or recording errors, so investigate before removing them.
In an exam the rule for identifying outliers is always given in the question. Two common rules:
- more than above or below
- more than standard deviations from the mean, outside
Measure from upwards and from downwards, and test both ends.
- Upper limit
- Lower limit
Worked example
and . Using the rule, are 40 and 19 outliers?
- 1
, so .
- 2
Upper limit: . Lower limit: .
- 3
40 is above 24.5, so it is an outlier. 19 is below 24.5, so it is not.
With and , what is the upper limit using ?
Outliers on a box plot
Each whisker stops at the most extreme value that is not an outlier, and each outlier is marked separately with a cross.
A box plot shows the minimum, , median, and maximum. The box runs from to with a line at the median.
When there are outliers, each whisker stops at the most extreme value that is not an outlier, and each outlier is marked separately, usually with a cross.
If there is no outlier at an end, the whisker goes to the smallest or largest value.
- 1
Work out both limits
for outliers
- 2
Find the extremes
largest and smallest non-outliers
- 3
Draw the box
Q₁ to Q₃ with a line at the median
- 4
Draw whiskers and crosses
whiskers to the extremes, crosses for outliers
Worked example
, median 5, , smallest 0, largest 40, next largest 19, with 40 an outlier. Describe the box plot.
- 1
The box runs from 2 to 11 with a line at 5.
- 2
The lower whisker goes to 0, as there is no lower outlier.
- 3
The upper whisker stops at 19, the largest non-outlier.
- 4
40 is marked separately with a cross.
Where does the upper whisker end when the largest value is an outlier?
The effect of outliers
Outliers have a large effect on the mean and especially the standard deviation, but little effect on the median and IQR.
Outliers have a large effect on the mean and especially on the standard deviation, because deviations are squared. They have little effect on the median and the IQR.
When an outlier is removed, update , and and recalculate.
Whether the mean rises or falls depends on which side the outlier lies.
- Before
- After adding 30
Affected a lot
- Mean
- Standard deviation
Affected little
- Median
- Interquartile range
Worked example
, , (mean 62, standard deviation 10). Remove the marks 38 and 85.
- 1
, .
- 2
.
- 3
Mean .
- 4
Standard deviation .
Which measure is least affected by an outlier?
Try an exam question
A data set has , and . (a) Describe the skewness, giving a reason. (b) An outlier is a value more than above or below . Determine whether the value is an outlier.
[4 marks]
- [1] and .
- [1], so the data are positively skewed.
- [1]Upper limit .
- [1], so is an outlier.
That's the notes covered.
Carry on to the next subtopic.
Try a question
Describe the skewness of the distribution using the quartiles.
1 mark, marked against the mark scheme
Describe the skewness of the distribution using the quartiles.
1 mark, marked against the mark scheme