The trimmed mean is a measure of center that reduces the influence of unusually small or unusually large values. It is useful when a data set has outliers or long tails that would pull the ordinary mean away from the main cluster. Instead of using every value, a trimmed mean removes the same percentage of data from both ends before averaging.
This makes it a practical compromise between the mean, which uses all values, and the median, which focuses only on the middle.
Understanding Statistics: The Trimmed Mean
Start by putting the observations in order from least to greatest. This step is essential because trimming is based on position, not on which values merely look unusual. Suppose a class records the time needed to finish a short run.
Most times lie between nine and thirteen minutes, but one student stops to help someone and records twenty-five minutes. A usual average moves upward because that one result is included fully.
If the chosen rule trims the lowest and highest positions, the twenty-five minute time may be removed before the calculation. The result describes the typical running time of the main group more closely.
The same amount must be removed from each end. This balance matters because removing values from only one side would deliberately shift the result. A symmetric rule is meant to protect the center from unusually distant values in either direction.
In a small data set, the chosen percentage may not produce a whole number of observations. A common classroom rule rounds down the calculated number to remove. For example, ten percent of seventeen is one point seven, so this rule removes one value at each end.
Different textbooks or software can use different rounding conventions. Students should always state the trimming percentage and the rule used for handling fractional counts.
Choosing the trim amount involves a tradeoff. A very small trim changes little, so a few extreme observations can still affect the result. A large trim gives stronger protection, but it throws away more real information.
Values at the ends are not automatically mistakes. A very high electricity bill could reveal a large family home. A low test score could show that a student missed lessons.
Removing such values may be sensible for describing a typical group, yet it can hide an important pattern if the goal is to study the full range of outcomes. The right choice depends on why the data were collected.
Trimmed means appear in settings where occasional unusual measurements are expected. Judges in some scored events may discard the highest and lowest scores to reduce the effect of one unusually strict or generous judge. Scientists may use robust summaries when instruments sometimes produce faulty readings.
In business data, a trimmed average can describe normal delivery times without letting a few major delays dominate. It is still good practice to inspect the original list or a graph first. Look for gaps, clusters, repeated values, and possible recording errors.
Compare the ordinary mean, the trimmed mean, and the median when possible. If these summaries differ greatly, the data are uneven or contain influential extreme values. That difference is useful evidence, not something to ignore.
Key Facts
- A p percent trimmed mean removes p percent of the data from the low end and p percent from the high end before averaging.
- Mean = (sum of all data values) / n
- Trimmed mean = (sum of remaining values after trimming both tails) / (number of remaining values)
- For n data values and trim proportion p, remove k = floor(pn) values from each tail when using a simple trimming rule.
- A 10% trimmed mean on 20 sorted values removes 2 smallest and 2 largest values, then averages the remaining 16 values.
- The trimmed mean resists outliers better than the ordinary mean because extreme values are excluded before calculating the average.
Vocabulary
- Trimmed mean
- A mean calculated after removing an equal percentage of the smallest and largest data values.
- Outlier
- A data value that is unusually far from the rest of the data.
- Tail
- The low or high end of a sorted data distribution where the most extreme values are found.
- Robust statistic
- A statistic that is not strongly affected by outliers or small departures from ideal conditions.
- Sorted data
- Data values arranged in order from smallest to largest.
Common Mistakes to Avoid
- Trimming only one side of the data is wrong because a trimmed mean removes the same amount from both tails unless a special one-sided method is stated.
- Trimming before sorting the data is wrong because the smallest and largest values cannot be identified reliably unless the data are ordered first.
- Removing p percent total instead of p percent from each tail is wrong because a 10% trimmed mean usually means 10% from the low end and 10% from the high end.
- Averaging the original number of data values after trimming is wrong because the denominator must be the number of values that remain.
Practice Questions
- 1 Find the 10% trimmed mean of the data set 3, 4, 5, 6, 7, 8, 9, 100 using k = floor(0.10n) values removed from each tail.
- 2 A class has sorted quiz scores 40, 72, 75, 76, 78, 80, 82, 84, 86, 88, 91, 100. Find the 10% trimmed mean using k = floor(0.10n) values removed from each tail.
- 3 A data set of home prices includes several extremely expensive mansions. Explain why a trimmed mean might describe a typical home price better than the ordinary mean.