Geometric Mean
The geometric mean is the central value of a set of n positive numbers defined as the nth root of their product — multiply all the values together, then take the root whose index is the number of values — and it is the appropriate average when quantities combine by multiplication rather than addition. The geometric mean of 2 and 8 is the square root of 16, which is 4, whereas their arithmetic mean is 5. The name reflects an old geometric interpretation: the geometric mean of two lengths is the side of the square whose area equals the rectangle formed by them.
The decisive use of the geometric mean is averaging growth rates, where the arithmetic mean gives plainly wrong answers: an investment that gains 50 percent in one year and loses 50 percent the next has not broken even, because 1.5 × 0.5 equals 0.75, a 25 percent loss overall, yet the arithmetic mean of the two returns is 0, suggesting no change. The geometric mean of the growth factors, the square root of 0.75, is about 0.866, correctly reporting an average decline of about 13.4 percent per year. This is the logic behind the compound annual growth rate: take the ratio of the final value to the initial value, then its root over the number of periods, and subtract one. The calculation always works on growth factors such as 1.05 rather than on percentages such as 5, which is the usual source of errors.
A clean way to understand the geometric mean is through logarithms: it is the exponential of the arithmetic mean of the logarithms of the values, which converts multiplication into addition so that the ordinary mean can do its work. This is why the geometric mean is the natural centre for data that are roughly log-normal, such as incomes, concentrations, bacterial counts, and particle sizes, and why it is the correct average of ratios: the geometric mean of the ratios A over B and B over A is exactly 1, while the arithmetic mean exceeds 1, a bias that matters in benchmarking, which is why performance suites normalise and then report a geometric mean of speed-up ratios. Francis Galton drew attention to the geometric mean in vital and social statistics in a short 1879 paper, arguing it suited quantities, such as wealth and many biological measurements, in which proportional rather than absolute differences matter.
The geometric mean is never larger than the arithmetic mean of the same positive numbers, a result known as the AM-GM inequality, and the two are equal only when all the values are identical. The gap between them widens as the values become more dispersed, which makes it a rough measure of spread: for returns, the penalty of volatility on compounded growth is exactly this gap, often called volatility drag. The geometric mean also sits between the harmonic mean and the arithmetic mean in the classical ordering of the three Pythagorean means. In index construction it appears in several price indices, including the Jevons index of 1863, and the United Nations Human Development Index has used a geometric mean of its component indices since 2010, so that a deficit in one dimension cannot be fully offset by a surplus in another.
The limits are strict and are the usual pitfalls. The geometric mean is defined only for positive numbers: a single zero makes the product zero, and a negative value makes the root undefined or misleading, so datasets with zeros need a shift or a different average. Percentage changes must be converted to growth factors before averaging. It is also not suited to quantities that add, such as heights or lengths across a sample, where the arithmetic mean is the correct choice. And it can hide large swings, since a smooth average growth rate of 8 percent can result from very different year-by-year paths, so the path and its variability still matter. The geometric mean also differs from the weighted mean, which changes the importance of values rather than the way they combine.
Read the geometric mean as the arithmetic mean on a logarithmic scale: the right average whenever quantities multiply, such as growth rates, ratios, and log-normal measurements, always at or below the arithmetic mean, and useless without strictly positive data.