DataTerminology.comNEW
Average computed after removing a specified proportion of the lowest and highest values.
DESCRIPTION
The trimmed mean reduces the influence of outliers by discarding equal proportions from both tails before averaging. It offers a balance between the sensitivity of the mean and robustness of the median. Common trims include 5% or 10%. It is useful in noisy datasets, heavy-tailed distributions, and performance benchmarking. Compared to winsorizing, trimming removes values rather than capping them.
ALTERNATIVE DEFINITIONS
Wikipedia
In statistics, the trimmed mean (or truncated mean) is a method of averaging that removes a small designated percentage of the largest and smallest values before calculating the mean.
Defines trimming and clarifies robustness to outliers.
NIST/SEMATECH e-Handbook of Statistical Methods
A 10% trimmed mean is the mean of the observations remaining after the smallest 10% and the largest 10% of the data values have been removed.
Gives a concrete example of trimming percentage and procedure.
RELATED TERMS
Robust Statistics Outliers Interquartile Range Variance Standard Deviation Percentile Bootstrap
COMPETING TERMS
RELATIONS
- Broader: Mean
KNOWLEDGE AREAS
Mathematics Data Science Testing & Validation
TARGET AUDIENCE
Data Scientist Data Analyst Information Analyst BI Specialist
EXTERNAL SOURCES
- Wikipedia - Truncated mean
- NIST - Robust estimates overview including trimmed means
- UCLA Statistical Consulting - Practical explanation and examples
TOP QUESTIONS
What is a Trimmed Mean?
It’s an average after removing equal percentages from both tails of the distribution to reduce outlier influence.
Category: Definition
When should I use a Trimmed Mean?
Use it when data contain outliers or heavy tails and you want a measure more robust than the mean but more informative than the median.
Category: UseCase
How does Trimmed Mean compare to Median?
The median is fully robust to extreme values; trimmed mean is partially robust and uses more information than the median.
Category: Comparison
What is a common trimming percentage?
5% and 10% per tail are common, but choose based on domain knowledge and data distribution.
Category: BestPractices
How does trimming differ from winsorizing?
Trimming removes extreme values; winsorizing replaces them with boundary values before averaging.
Category: Comparison
What risks exist when using Trimmed Mean?
Over-trimming can discard important data and bias results; under-trimming leaves outlier influence.
Category: Risks
Is Trimmed Mean suitable for small samples?
It can be, but the effective sample size decreases; consider bootstrap confidence intervals.
Category: Opinion
How do I compute a Trimmed Mean in practice?
Sort values, remove the specified fraction from both ends, then compute the mean of remaining values.
Category: HowTo
What is a hot topic around Trimmed Means?
Robust aggregation in benchmarking and ML evaluation to mitigate outlier impacts is increasingly adopted.
Category: Topic