Shape, Center, Spread, and Outliers | Malone Math

The Central Idea

A compact description of a quantitative distribution should usually address four questions:

What is the shape?
Where is the center?
How large is the spread?
Are there any unusual observations?

These features work together to describe the distribution as a whole.

What Is Distribution Shape?

The shape of a distribution describes the broad pattern formed by its observations.

Important shape features include:

symmetry;
skewness;
number of peaks;
gaps;
clusters;
tails;
possible outliers.

Symmetric Distributions

A distribution is approximately symmetric when its left and right sides have broadly similar shapes around a central region.

In a perfectly symmetric distribution, reflecting the distribution across its center would leave the shape unchanged.

Real datasets are rarely perfectly symmetric, so the term is usually used approximately.

Center in a Symmetric Distribution

In a roughly symmetric, unimodal distribution, the mean and median are often close.

If the distribution is also concentrated around one dominant value, the mode may be near them as well.

This makes the mean and standard deviation a natural summary pair when strong outliers are absent.

Right-Skewed Distributions

A distribution is right-skewed when its longer tail extends toward larger numerical values.

Most observations may be concentrated at lower values, while a smaller number stretch far to the right.

Examples can include:

household income;
waiting times;
transaction sizes;
insurance claims.

Mean and Median in Right-Skewed Data

Large values in the right tail pull the mean upward more strongly than they affect the median.

A common pattern is

\[ \operatorname{median}<\operatorname{mean}. \]

This is a useful tendency, not a universal rule for every right-skewed dataset.

Left-Skewed Distributions

A distribution is left-skewed when its longer tail extends toward smaller numerical values.

A smaller number of unusually low observations may stretch the distribution to the left.

Mean and Median in Left-Skewed Data

Low values in the left tail pull the mean downward.

A common pattern is

\[ \operatorname{mean}<\operatorname{median}. \]

Again, this is a common tendency rather than an absolute rule.

The Direction of Skewness Is Named for the Tail

A frequent mistake is to name skewness according to where most of the data are concentrated.

Instead:

a long right tail means right-skewed;
a long left tail means left-skewed.

Unimodal Distributions

A distribution with one major concentration or peak is called unimodal.

A single peak often suggests one dominant cluster of observations.

Bimodal Distributions

A distribution with two distinct peaks is called bimodal.

Two peaks can indicate that the data combine two underlying groups.

For example, commute times might be bimodal if one group travels locally while another group commutes from distant locations.

Multimodal Distributions

A distribution with several substantial peaks may be described as multimodal.

Multiple modes can suggest:

mixtures of populations;
different operating regimes;
seasonal effects;
grouping by hidden categorical variables.

Clusters

A cluster is a region where observations are concentrated.

Clusters can reveal structure that a single center statistic may hide.

If two clusters are far apart, reporting only one mean may create a value that represents neither cluster well.

Gaps

A gap is an interval containing few or no observations.

Gaps may separate clusters or indicate:

distinct subgroups;
threshold effects;
measurement practices;
structural limits in the process generating the data.

Tails

A tail is a region where the distribution thins out toward unusually low or high values.

Tail length and direction are central to identifying skewness.

Describing Center

Two common measures of center are:

the mean;
the median.

The best choice depends on distribution shape and sensitivity to extreme observations.

Mean as Center

The sample mean is

\[ \bar{x} = \frac{1}{n}\sum_{i=1}^{n}x_i. \]

Because every value contributes directly, the mean responds strongly to extreme observations.

Median as Center

The median identifies the middle position of the ordered data.

It is relatively resistant to extreme values because changing a value far out in a tail often does not change the middle position.

Describing Spread

Common spread measures include:

range;
interquartile range;
variance;
standard deviation.

These statistics emphasize different aspects of variability.

Standard Deviation as Spread

Standard deviation measures the scale of variation around the mean.

For a sample,

\[ s = \sqrt{ \frac{1}{n-1} \sum_{i=1}^{n}(x_i-\bar{x})^2 }. \]

Like the mean, standard deviation is sensitive to extreme values.

IQR as Spread

The interquartile range is

\[ \operatorname{IQR}=Q_3-Q_1. \]

It measures the width of the middle half of the data and is relatively resistant to outliers.

Natural Summary Pairs

Two common descriptive pairings are:

\[ \boxed{\text{mean and standard deviation}} \]

and

\[ \boxed{\text{median and IQR}}. \]

The first pair is often appropriate for roughly symmetric data without strong outliers. The second pair is often better for skewed data or data containing extreme values.

Why Center and Spread Should Be Reported Together

Consider:

\[ A:\quad 9,\ 10,\ 10,\ 10,\ 11 \] \[ B:\quad 0,\ 5,\ 10,\ 15,\ 20. \]

Both have mean $10$, but Dataset B is much more spread out.

Center without spread gives an incomplete description.

What Is an Outlier?

An outlier is an observation unusually far from the main body of the data.

Outliers may be important because they can:

change the mean substantially;
increase the range;
increase variance and standard deviation;
signal data-quality problems;
identify genuinely unusual cases.

Visual Identification of Possible Outliers

Dot plots, histograms, and other displays may reveal observations separated from the main cluster.

Visual evidence is useful for investigation, but a formal rule can provide a more systematic criterion.

The 1.5-IQR Rule

A common introductory outlier rule uses the interquartile range.

Lower fence:

\[ \boxed{ Q_1-1.5(\operatorname{IQR}) }. \]

Upper fence:

\[ \boxed{ Q_3+1.5(\operatorname{IQR}) }. \]

Values beyond these fences are flagged as potential outliers.

Example: Applying the IQR Rule

Suppose

\[ Q_1=18,\qquad Q_3=30. \]

Then

\[ \operatorname{IQR} = 30-18 = 12. \]

Lower fence:

\[ 18-1.5(12) = 18-18 = \boxed{0}. \]

Upper fence:

\[ 30+1.5(12) = 30+18 = \boxed{48}. \]

Values below $0$ or above $48$ are flagged as potential outliers.

Potential Outlier Does Not Mean Error

A value beyond an outlier fence is not automatically incorrect.

It may represent:

a genuine extreme observation;
a rare but important event;
a different subgroup;
a measurement problem;
a data-entry problem.

The correct response is investigation, not automatic deletion.

Outliers and the Mean

Consider

\[ 5,\ 6,\ 6,\ 7,\ 8. \]

The mean is

\[ \bar{x} = \frac{32}{5} = 6.4. \]

Replace $8$ with $80$:

\[ 5,\ 6,\ 6,\ 7,\ 80. \]

The new mean becomes

\[ \bar{x} = \frac{104}{5} = 20.8. \]

The extreme value dramatically changes the mean.

Outliers and the Median

For both

\[ 5,\ 6,\ 6,\ 7,\ 8 \]

and

\[ 5,\ 6,\ 6,\ 7,\ 80, \]

the median remains

\[ \boxed{6}. \]

This illustrates the median's resistance.

Outliers and Standard Deviation

Standard deviation is based on squared deviations from the mean.

Large deviations contribute disproportionately because

\[ (x_i-\bar{x})^2 \]

grows rapidly as the observation moves farther from the mean.

Extreme values can therefore increase standard deviation substantially.

Outliers and the IQR

The IQR is based on $Q_1$ and $Q_3$, which depend on ordered positions.

A single very large or very small observation usually affects the IQR much less than it affects the range or standard deviation.

Shape Affects the Meaning of "Typical"

In a symmetric distribution, the mean can provide a strong description of a typical observation.

In a strongly skewed distribution, the mean may be pulled toward a tail.

In that setting, the median may better describe the typical central position.

Example: Income Distribution

Suppose most households in a region earn between $\$40{,}000$ and $\$90{,}000$, while a small number earn several million dollars.

The distribution is likely right-skewed.

The high-income tail may pull the mean well above the income of a typical household.

Median income and IQR can therefore provide a more resistant summary.

Example: Controlled Manufacturing Process

Suppose component lengths cluster tightly and symmetrically around the target dimension, with no substantial outliers.

The mean and standard deviation can provide a useful summary because:

the center is naturally represented by the mean;
variation around that center is important;
the distribution is approximately symmetric.

Describe Shape Before Choosing Summaries

A reliable descriptive analysis often proceeds in this order:

inspect a graph;
describe shape;
identify possible outliers;
choose appropriate measures of center and spread;
interpret the numerical summaries.

A Useful Description Framework

One practical framework is:

Shape: symmetric, skewed, unimodal, bimodal, or otherwise structured;
Center: mean or median;
Spread: standard deviation, IQR, or range;
Unusual features: gaps, clusters, or outliers.

Example: Full Distribution Description

Suppose a histogram shows one major peak near $50$, roughly balanced tails, and no isolated observations. The sample statistics are

\[ \bar{x}=51.2,\qquad s=6.4. \]

A reasonable description is:

The distribution is approximately symmetric and unimodal, centered near $51$, with a standard deviation of about $6.4$, and no obvious outliers.

Example: Skewed Distribution Description

Suppose a distribution has a long right tail and several large values. Its median and quartiles are

\[ Q_1=20,\qquad \operatorname{median}=27,\qquad Q_3=36. \]

Then

\[ \operatorname{IQR} = 36-20 = 16. \]

A reasonable description is:

The distribution is right-skewed, centered around a median of $27$, with the middle half spanning $16$ units. Large upper-tail observations should be inspected for possible outliers.

Comparing Two Distributions

When comparing two groups, discuss the same structural features for both.

Compare:

shape;
center;
spread;
outliers or other unusual features.

Example: Comparing Centers

Suppose Group A has median $42$ and Group B has median $55$.

Group B has the higher central position according to the medians.

This statement does not imply that every value in Group B exceeds every value in Group A.

Example: Comparing Spread

Suppose Group A has

\[ \operatorname{IQR}=8 \]

while Group B has

\[ \operatorname{IQR}=20. \]

The middle half of Group B is more widely dispersed.

Overlapping Distributions Can Still Have Different Centers

Two distributions can overlap substantially while still having different means or medians.

Statistical comparison concerns overall distribution structure, not whether every observation from one group lies above every observation from the other.

Similar Centers Can Hide Different Shapes

Two datasets can have the same mean and median while having very different:

spread;
skewness;
number of peaks;
outlier structure.

Numerical summaries should therefore be paired with visual inspection.

Common Error: Describing Only the Mean

Saying "the mean is $25$" does not describe the full distribution.

The distribution may be narrow or wide, symmetric or skewed, unimodal or multimodal, and may or may not contain outliers.

Common Error: Calling Every Extreme Value an Outlier

The minimum and maximum are always extreme in the literal sense of being endpoints.

They are not automatically statistical outliers.

Outlier status depends on the distribution and the rule or context being used.

Common Error: Choosing Mean and Standard Deviation for Strongly Skewed Data

Mean and standard deviation can be heavily influenced by a long tail or extreme values.

For strongly skewed data, median and IQR often provide a more resistant descriptive summary.

Common Error: Choosing Median and IQR Automatically

Resistant measures are useful, but they are not automatically superior in every context.

If arithmetic totals, expected values, or variation around the mean are central to the problem, mean and standard deviation may be operationally important.

Common Error: Ignoring Multimodality

A single center can be misleading when the distribution contains two or more distinct clusters.

In that situation, investigate whether the data combine different populations or processes.

Common Error: Deleting Outliers Without Investigation

Removing an unusual observation simply because it changes the mean or standard deviation can distort the analysis.

Outliers should be investigated and handled according to defensible analytical rules.

Common Error: Comparing Spread Without Considering Units

A standard deviation of $5$ kilograms and one of $5$ grams do not represent the same amount of variability.

Spread must be interpreted in the scale and units of the variable.

A Reliable Procedure for Describing One Distribution

Identify the quantitative variable and its units.
Inspect an appropriate graph.
Describe the overall shape.
Identify peaks, gaps, clusters, and tails.
Inspect for possible outliers.
Choose a center measure appropriate to the shape.
Choose a spread measure appropriate to the shape.
Report the numerical values in context.
Combine the features into one coherent description.

A Reliable Procedure for Comparing Distributions

Use comparable graphical scales.
Compare shape.
Compare centers.
Compare spreads.
Compare outliers and other unusual features.
Use relative language such as higher, lower, more variable, or more skewed.
Avoid claims stronger than the data support.

Practice 1: Identify the Shape

A distribution has one main peak near $40$, with roughly equal tails on both sides.

How would you describe its shape?

Practice 2: Direction of Skewness

Most observations lie between $10$ and $30$, but a few values extend to $80$.

Which direction is the distribution skewed?

Practice 3: Choose Center and Spread

A dataset is approximately symmetric with no strong outliers.

Which descriptive pair is generally natural?

Practice 4: Choose Resistant Summaries

A distribution is strongly right-skewed and contains several extreme high values.

Which descriptive pair is generally more resistant?

Practice 5: IQR

Suppose

\[ Q_1=22,\qquad Q_3=34. \]

Find the IQR.

Practice 6: Outlier Fences

Using

\[ Q_1=22,\qquad Q_3=34, \]

find the lower and upper $1.5$-IQR fences.

Practice 7: Identify the Better Summary

A city reports household income data with a small number of extremely high incomes.

Would mean income or median income usually better describe a typical household?

Practice 8: Compare Spread

Group A has $s=4$, and Group B has $s=11$, measured in the same units.

Which group has greater variability around its mean?

Practice 9: Identify the Structural Concern

A histogram has two strong peaks separated by a clear gap.

What feature should the analyst investigate?

Practice 10: Full Description

A distribution is right-skewed, has median $18$, IQR $7$, and one value far above the upper cluster.

Write a concise description using shape, center, spread, and unusual features.

Practice Answers

Practice 1

The distribution is approximately symmetric and unimodal.

Practice 2

The longer tail extends toward larger values, so the distribution is right-skewed.

Practice 3

The mean and standard deviation are generally a natural pair for an approximately symmetric distribution without strong outliers.

Practice 4

The median and IQR are generally more resistant to strong skewness and extreme values.

Practice 5

\[ \operatorname{IQR} = 34-22 = \boxed{12}. \]

Practice 6

Since

\[ \operatorname{IQR}=12, \]

the lower fence is

\[ 22-1.5(12) = 22-18 = \boxed{4}. \]

The upper fence is

\[ 34+1.5(12) = 34+18 = \boxed{52}. \]

Practice 7

Median income usually provides the more resistant description of a typical household when a small number of extremely high incomes create strong right skewness.

Practice 8

Group B has the larger standard deviation, so it has greater variability around its mean.

Practice 9

The analyst should investigate possible bimodality and determine whether the data combine two underlying groups or processes.

Practice 10

The distribution is right-skewed, centered around a median of $18$, with an IQR of $7$. One unusually large observation lies beyond the main upper cluster and should be investigated as a possible outlier.

What to Remember

A complete distribution description considers shape, center, spread, and unusual features.
Symmetric distributions often support mean and standard deviation as a natural summary pair.
Skewed distributions often support median and IQR as a resistant summary pair.
Right skew means a long right tail; left skew means a long left tail.
Multiple peaks can indicate mixtures of different groups or processes.
Outliers can strongly affect the mean, range, variance, and standard deviation.
The median and IQR are relatively resistant to extreme values.
The $1.5$-IQR rule flags potential outliers but does not prove that they are errors.
Graphs and numerical summaries should be interpreted together.
Statistical descriptions should be stated in the context and units of the variable.

Where We Go Next

Once a distribution's shape, center, spread, and unusual observations are understood, individual values can be compared more meaningfully with the distribution as a whole.

The next lesson, Standardization and Z-Scores, develops a unit-free measure of how far an observation lies above or below the mean in standard deviation units.