Mean

The mean, or arithmetic average, is a measure of central tendency calculated by summing all values in a dataset and dividing by the number of values. It is the most commonly used summary statistic for continuous data and provides a single value that represents the center of the distribution.

The mean serves as the starting point for most statistical analyses of continuous data. It is used in hypothesis tests (t tests, ANOVA), regression models, and meta analyses as the primary summary of group performance or population characteristics. Researchers report means alongside standard deviations to describe the center and spread of their data. Clinicians use mean values to characterize treatment effects (mean difference), and health economists use mean costs and mean QALYs in economic evaluations. Because the mean is so pervasive, understanding its properties, including its sensitivity to outliers and its behavior in skewed distributions, is essential for any researcher working with quantitative data.

The arithmetic mean of a sample is calculated as x bar = (sum of all x values) divided by n, where n is the number of observations. The population mean (mu) is the corresponding parameter for the entire population. The sample mean is an unbiased estimator of the population mean, meaning that if many random samples were drawn, the sample means would average out to the population mean in the long run.

The mean has several important mathematical properties. It is the value that minimizes the sum of squared deviations: the squared distances of the data from the mean add up to less than their squared distances from any other value, which is why it appears in the formulas for variance and standard deviation. It is the balance point of the distribution: if the data were placed on a number line balanced on a fulcrum, the fulcrum would sit at the mean. In a symmetric, unimodal distribution, the mean, the median and the mode coincide. In a skewed distribution, the mean is pulled toward the long tail, so it usually lies above the median in a right skewed distribution and below it in a left skewed one (the rule can fail, particularly for discrete data).

Other types of means exist for specific applications. The weighted mean assigns different weights to different observations, used, for example, to combine results in meta analysis, where each study is weighted by its precision (the inverse of its variance), so larger, more precise studies count for more. The geometric mean, calculated as the nth root of the product of n values, is appropriate for data that are multiplicative or log normally distributed, such as antibody titers or fold changes. The harmonic mean, the reciprocal of the arithmetic mean of reciprocals, is used in certain rate calculations.

Use the arithmetic mean to summarize continuous data that are approximately symmetrically distributed and free of extreme outliers. It is appropriate for normally distributed data, for data that are approximately normal (as assessed by histograms, Q Q plots, or tests of normality), and for any situation where the mathematical properties of the mean (minimizing squared deviations, serving as an unbiased estimator) are desirable. When data are skewed or contain extreme outliers, the median usually describes the typical value better because it is less sensitive to extreme values; the mean remains the relevant summary when totals matter, as with costs in economic evaluations. For ratios, fold changes or growth rates, consider the geometric mean. For rates or speeds, consider the harmonic mean.

A common mistake is reporting the mean for heavily skewed data without noting that the median would provide a more representative summary. For example, reporting the mean household income in a population with a small number of extremely wealthy individuals can give a misleading impression of typical income. Another error is computing the mean of ordinal data (such as individual Likert items) as if the intervals between categories were equal, without justifying that assumption. A third mistake is confusing the standard deviation (a measure of spread in the original data) with the standard error of the mean (a measure of precision of the mean estimate), which are used for different purposes. Finally, presenting means without measures of variability (standard deviation or confidence interval) provides an incomplete picture of the data.

A single arm clinical trial reports the change in LDL levels in 200 patients taking a new cholesterol medication. At baseline, the mean LDL is 162 mg/dL (standard deviation: 28 mg/dL). After 12 weeks of treatment, the mean LDL drops to 98 mg/dL (standard deviation: 22 mg/dL). The mean within patient reduction (64 mg/dL; standard deviation of the individual changes: 25 mg/dL) and its 95% confidence interval (60.5 to 67.5 mg/dL) are reported as the primary efficacy measure, and a histogram confirms that LDL values are approximately normally distributed at both time points.