Outlier: A data point that differs significantly from other observations in a dataset.
Sensitivity: The degree to which a measure changes when outliers are introduced.
Robustness: The resistance of a measure to changes caused by outliers.
- Calculate mean and median without outlier
- Add outlier and recalculate both measures
- Calculate the absolute and percentage change for each measure
- Compare the sensitivity of mean vs median
Dataset: 10, 12, 15, 18, 20 (already sorted)
Mean: (10 + 12 + 15 + 18 + 20) ÷ 5 = 75 ÷ 5 = 15
Median: Middle value (3rd position) = 15
New dataset: 10, 12, 15, 18, 20, 100 (sorted)
Mean: (75 + 100) ÷ 6 = 175 ÷ 6 = 29.17
Median: Average of 3rd and 4th values = (15 + 18) ÷ 2 = 16.5
Mean change: 29.17 - 15 = 14.17
Mean % change: (14.17 ÷ 15) × 100% = 94.44%
Median change: 16.5 - 15 = 1.5
Median % change: (1.5 ÷ 15) × 100% = 10%
Mean sensitivity: Increased by 94.44% (highly sensitive)
Median robustness: Increased by only 10% (resistant to outliers)
Mean: 15
Median: 15
Mean: 29.17
Median: 16.5
Mean: Increased from 15 to 29.17 (94.44% increase)
Median: Increased from 15 to 16.5 (10% increase)
Conclusion: Mean is highly sensitive to outliers, median is robust.
• Mean sensitivity: Affected by every value in dataset
• Median robustness: Only affected by middle values
• Outlier impact: Mean is pulled toward the outlier
Cumulative effect: Multiple outliers can have a stronger combined impact on measures of central tendency.
Directional bias: Outliers on one side of the distribution pull measures in that direction.
Dataset: 5, 8, 10, 12, 15, 18, 20 (already sorted)
Mean: (5 + 8 + 10 + 12 + 15 + 18 + 20) ÷ 7 = 88 ÷ 7 = 12.57
Median: Middle value (4th position) = 12
New dataset: 5, 8, 10, 12, 15, 18, 20, 50, 75 (sorted)
Mean: (88 + 50 + 75) ÷ 9 = 213 ÷ 9 = 23.67
Median: Middle value (5th position) = 15
Mean change: 23.67 - 12.57 = 11.10
Mean % change: (11.10 ÷ 12.57) × 100% = 88.31%
Median change: 15 - 12 = 3
Median % change: (3 ÷ 12) × 100% = 25%
Mean: Significantly affected by multiple high outliers (88.31% increase)
Median: Moderately affected (25% increase) but still robust
Cumulative impact: Multiple outliers amplify the mean's sensitivity
Single outlier (Exercise 1): Mean increased 94.44%
Two outliers (this exercise): Mean increased 88.31%
The relative impact depends on the outlier's distance from the original mean
Mean Sensitivity: 88.31% increase
Median Robustness: 25% increase
Without outliers: Mean = 12.57, Median = 12
With outliers: Mean = 23.67, Median = 15
Effect: Mean increased 88.31%, Median increased 25%. Even with multiple outliers, median remains more robust.
• Cumulative effect: Multiple outliers have additive impact on mean
• Median resilience: Still resistant even with multiple outliers
• Directional bias: High outliers pull mean upward
Business applications: Understanding how extreme values affect reported averages in practical scenarios.
Reporting considerations: Choosing appropriate measures for different audiences.
- Calculate initial mean and median
- Add the CEO's salary and recalculate
- Determine which measure better represents typical employee salary
- Consider implications for reporting
Dataset: $3000, $3200, $3400, $3500, $3600, $3800, $4000, $4200, $4500, $5000
Mean: (3000 + 3200 + 3400 + 3500 + 3600 + 3800 + 4000 + 4200 + 4500 + 5000) ÷ 10
Sum = $38,200, Mean = $38,200 ÷ 10 = $3,820
Median: Average of 5th and 6th values = ($3600 + $3800) ÷ 2 = $3,700
New dataset: $3000, $3200, $3400, $3500, $3600, $3800, $4000, $4200, $4500, $5000, $50,000
Mean: ($38,200 + $50,000) ÷ 11 = $88,200 ÷ 11 = $8,018.18
Median: Middle value (6th position) = $3,800
Mean change: $8,018.18 - $3,820 = $4,198.18
Mean % change: ($4,198.18 ÷ $3,820) × 100% = 109.90%
Median change: $3,800 - $3,700 = $100
Median % change: ($100 ÷ $3,700) × 100% = 2.70%
Mean distortion: Mean jumps from $3,820 to $8,018 due to CEO's salary
Median stability: Median only increases by $100
Typical employee: Median ($3,800) better represents typical salary than mean ($8,018)
For reporting typical employee compensation, median is more appropriate
Mean would mislead stakeholders about typical employee earnings
Median reflects reality for 10 out of 11 employees
Mean: $3,820
Median: $3,700
Mean: $8,018
Median: $3,800
Regular employees: Mean = $3,820, Median = $3,700
With CEO: Mean = $8,018, Median = $3,800
Effect: Mean increased 109.90%, Median increased only 2.70%. The median better represents typical employee salary.
• Extreme outlier effect: Single extreme value can dramatically change mean
• Median robustness: Resistant to extreme values in practical contexts
• Reporting choice: Use median for typical values when outliers present
Outlier: A data point that significantly differs from other observations, typically defined as values beyond Q1 - 1.5×IQR or Q3 + 1.5×IQR.
Sensitivity: The degree to which a statistical measure changes in response to changes in the dataset.
Robustness: The resistance of a statistical measure to changes caused by outliers or violations of assumptions.
Mean Sensitivity: The mean is highly sensitive to outliers because it incorporates every value in its calculation.
Median Robustness: The median is robust because it only depends on the middle value(s) of an ordered dataset.
- Baseline calculation: Calculate mean and median for original dataset
- Outlier introduction: Add outlier(s) to the dataset
- Recalculation: Calculate new mean and median
- Change quantification: Calculate absolute and percentage changes
- Comparison: Compare sensitivity levels
- Interpretation: Determine appropriate measure for context
• Mean sensitivity: Mean = Σx/n, so every value affects the result
• Median robustness: Only middle position matters, not actual values
• Outlier definition: Values beyond Q1 - 1.5×IQR or Q3 + 1.5×IQR
• Measure choice: Use median for skewed data with outliers
Outlier distance: The farther an outlier is from the main cluster of data, the greater its impact on the mean.
Proportional effect: Mean sensitivity increases proportionally with outlier distance.
Dataset: 1, 2, 3, 4, 5
Mean: (1 + 2 + 3 + 4 + 5) ÷ 5 = 15 ÷ 5 = 3
Median: Middle value = 3
Dataset: 1, 2, 3, 4, 5, 10
Mean: (15 + 10) ÷ 6 = 25 ÷ 6 = 4.17
Median: Average of 3rd and 4th = (3 + 4) ÷ 2 = 3.5
Mean change: 4.17 - 3 = 1.17 (39% increase)
Dataset: 1, 2, 3, 4, 5, 20
Mean: (15 + 20) ÷ 6 = 35 ÷ 6 = 5.83
Median: Average of 3rd and 4th = 3.5
Mean change: 5.83 - 3 = 2.83 (94.33% increase)
Dataset: 1, 2, 3, 4, 5, 50
Mean: (15 + 50) ÷ 6 = 65 ÷ 6 = 10.83
Median: Average of 3rd and 4th = 3.5
Mean change: 10.83 - 3 = 7.83 (261% increase)
Outlier at 10: Mean increases 39%
Outlier at 20: Mean increases 94.33%
Outlier at 50: Mean increases 261%
Median: Remains stable at 3.5 in all cases
Key Insight: As outlier distance increases, mean sensitivity increases dramatically while median remains unchanged.
Outlier at 10: Mean = 4.17 (39% increase)
Outlier at 20: Mean = 5.83 (94.33% increase)
Outlier at 50: Mean = 10.83 (261% increase)
Median: Always 3.5 (0% change). Mean sensitivity increases with outlier distance.
• Distance effect: Farther outliers have greater impact on mean
• Median stability: Distance doesn't affect median
• Proportional sensitivity: Mean change is proportional to outlier distance
Left-sided outliers: Extremely low values that pull the mean downward.
Right-sided outliers: Extremely high values that pull the mean upward.
Directional bias: Outliers pull the mean in their direction regardless of position.
Dataset: 10, 12, 15, 18, 20
Mean: (10 + 12 + 15 + 18 + 20) ÷ 5 = 75 ÷ 5 = 15
Median: Middle value = 15
Dataset: 1, 10, 12, 15, 18, 20
Mean: (75 + 1) ÷ 6 = 76 ÷ 6 = 12.67
Median: Average of 3rd and 4th = (12 + 15) ÷ 2 = 13.5
Mean change: 12.67 - 15 = -2.33 (-15.56% change)
Dataset: 10, 12, 15, 18, 20, 50
Mean: (75 + 50) ÷ 6 = 125 ÷ 6 = 20.83
Median: Average of 3rd and 4th = (15 + 18) ÷ 2 = 16.5
Mean change: 20.83 - 15 = 5.83 (+38.89% change)
Low outlier (1): Mean decreases 15.56%, Median decreases 10%
High outlier (50): Mean increases 38.89%, Median increases 10%
Mean direction: Always pulled toward the outlier's direction
Low outlier: Median decreases from 15 to 13.5 (10% decrease)
High outlier: Median increases from 15 to 16.5 (10% increase)
Median pattern: Moves toward the middle of the new range
Mean: 12.67
Median: 13.5
Mean: 20.83
Median: 16.5
Left outlier (1): Mean = 12.67 (-15.56%), Median = 13.5 (-10%)
Right outlier (50): Mean = 20.83 (+38.89%), Median = 16.5 (+10%)
Directional effect: Mean is pulled toward the outlier's direction; Median shows consistent 10% change.
• Directional bias: Mean always moves toward outlier direction
• Median stability: Similar robustness regardless of outlier direction
• Proportional effect: Greater distance creates greater mean impact
Outlier: A data point that significantly differs from other observations in a dataset. Commonly defined as values beyond Q1 - 1.5×IQR or Q3 + 1.5×IQR.
Mean Sensitivity: The degree to which the mean changes when outliers are introduced. The mean is highly sensitive because it incorporates every value in its calculation.
Median Robustness: The resistance of the median to changes caused by outliers. The median only depends on the middle value(s) of an ordered dataset.
Directional Bias: The tendency of outliers to pull the mean in their direction, regardless of whether they are high or low values.
- Baseline establishment: Calculate mean and median for the original dataset
- Outlier identification: Determine if values qualify as outliers using standard criteria
- Impact assessment: Add outliers and recalculate both measures
- Change quantification: Calculate absolute and percentage changes
- Comparison analysis: Compare sensitivity levels between measures
- Context evaluation: Determine appropriate measure for specific applications
- Reporting recommendation: Choose measure based on analysis goals
• Mean formula: x̄ = Σx/n (sensitive to every value)
• Outlier boundaries: Below Q1 - 1.5×IQR or Above Q3 + 1.5×IQR
• Median calculation: Middle value of ordered dataset (robust)
• Distance effect: Mean sensitivity increases with outlier distance
• Directional bias: Mean always pulled toward outlier direction
Outlier Effects: The differential impact that extreme values have on measures of central tendency, particularly how the mean and median respond differently to the same outlier.
Mean Sensitivity: The characteristic of the mean to be affected by every single value in the dataset, making it highly susceptible to outliers. The mean incorporates all values equally in its calculation.
Median Robustness: The characteristic of the median to be resistant to outliers because it only depends on the middle value(s) of an ordered dataset, ignoring the actual values of the extreme points.
Statistical Robustness: The property of a statistical measure to maintain its value or interpretation despite violations of assumptions or the presence of extreme values.
- Data preparation: Organize and sort the dataset
- Baseline calculation: Calculate mean and median without outliers
- Outlier identification: Use 1.5×IQR rule or other criteria
- Impact simulation: Add outliers and recalculate both measures
- Change quantification: Calculate absolute and percentage changes
- Comparison analysis: Compare sensitivity levels between measures
- Decision making: Choose appropriate measure for context
• Always compare both measures: Calculate mean and median together to understand outlier impact
• Consider the context: Choose measure based on analysis goals and audience
• Identify outliers systematically: Use 1.5×IQR rule for consistent identification
• Understand directional bias: Mean always moves toward outlier direction
• Recognize robustness: Median provides stable measure regardless of extreme values
Questions & Answers
Question: Why is the median considered "robust" while the mean is "sensitive"? What makes them behave so differently?
Answer: The difference comes from how each measure is calculated:
- Mean: Uses every value in the dataset (sum divided by count), so outliers directly affect the sum
- Median: Only depends on the middle position when data is ordered, not the actual values of extremes
Think of the mean as a "democracy" where every value votes equally, so extreme values have outsized influence. The median is like a "dictatorship" of the middle position - it only cares about who sits in the center, not how extreme the edges are.
This is why the median is called "robust" - it resists changes from outliers because it's based on position rather than value.
Question: Are there any situations where I should still use the mean even when outliers are present?
Answer: Yes, there are several valid reasons to use the mean even with outliers:
- When you need the total: For budgeting, resource allocation, or when all values matter (like total costs)
- For further calculations: Many statistical formulas require the mean (variance, standard deviation)
- When outliers are valid: Sometimes extreme values are real and important (like rare events in risk assessment)
- Population parameters: When studying population totals rather than typical individuals
The key is understanding why you're using the measure. If you want to know what's "typical," use median. If you need to account for everything, use mean. Often, reporting both provides the most complete picture.
Question: How do I decide when a value is truly an outlier versus just a high/low value that belongs in the dataset?
Answer: Use both statistical and contextual criteria:
- Statistical tests: Values beyond Q1 - 1.5×IQR or Q3 + 1.5×IQR are commonly considered outliers
- Context consideration: Does the value make sense given the nature of the data?
- Data collection: Was there an error in measurement or recording?
- Domain knowledge: Consult subject experts about what's reasonable
For example, in household income data, a $1 million salary might be an outlier statistically but could be perfectly reasonable for a CEO. Always consider both the mathematical definition and real-world context.
Sometimes it's worth conducting analysis both ways (with and without suspected outliers) to see how much they affect your conclusions.