The geometric mean is the constant factor that would produce the same product as a set of positive factors. Instead of adding the values and sharing a total, it multiplies them and takes the corresponding root. This makes it the natural average for ratios, proportional changes, compound growth, scale factors, and data that vary multiplicatively.
A product-preserving average is different from a balance point
For two positive numbers (a) and (b), the geometric mean is the side length of a square with the same area as a rectangle of sides (a) and (b). The arithmetic mean, by contrast, gives the side of a square with the same perimeter. The two questions preserve different quantities, so the answers generally differ.
Use the controls to change two values. The arithmetic mean remains at or above the geometric mean, and the two coincide only when the values are equal.
Compare two means
Both means agree only when the two values are equal.
Worked example with three measurements
For 2, 8, and 32, multiply the values to obtain 512 and take the cube root because there are three factors. The result is 8. It is easy to verify: replacing all three factors by 8 preserves the product, since 8 times 8 times 8 is also 512.
This answer describes a typical multiplicative scale. The arithmetic mean of the same values is 14, which preserves their sum instead. Neither result is universally “better”; the model determines which total or product must remain unchanged.
Compound growth requires factors, not signed rates
Suppose a quantity rises by 20% in one period and falls by 10% in the next. Convert the changes to factors 1.20 and 0.90. Their product is 1.08, and the geometric mean factor is the square root of 1.08. Subtract one only after taking the root to obtain the equivalent constant rate per period.
The equivalent rate is about 3.92% per period. The arithmetic average of +20% and −10% is 5%, but applying 5% twice produces a factor of 1.1025, not 1.08. Equal period lengths are assumed. If periods differ, use exponents or weights that represent their durations.
Domain restrictions matter
In the standard real-valued statistical definition, every input must be positive. A zero makes the entire product zero and destroys the logarithmic interpretation; negative inputs can make an even root non-real and give inconsistent signs for odd sample sizes. Do not silently take absolute values, because that changes the data's meaning.
For positive values, the geometric mean never exceeds the arithmetic mean. A large gap indicates substantial dispersion on a multiplicative scale. In software, long products may overflow or underflow, so the stable calculation averages logarithms and exponentiates the result. That computational technique still requires positive inputs.
Choosing a defensible application
Use the geometric mean for compound returns over equal periods, repeated scale changes, normalized ratios, and some indices. Use the arithmetic mean when observations contribute additively to a total. For an investment, work with growth factors after fees and losses, not raw percentage labels. A 100% loss creates a zero factor and cannot be recovered by a later finite gain, so no positive constant growth factor represents that sequence.
