Is the geometric mean always below the arithmetic mean?

For the same positive values with equal weights, it is no greater; equality holds when all values are equal.

Can a decrease appear in a growth calculation?

Yes. A decrease smaller than 100% still has a positive factor, such as 0.75 for a 25% decline.

What happens with a zero value?

The nonnegative product-root extension gives zero, but log(0) is not finite and the practical interpretation needs care.

AI Glossary / Statistics and probability / Beginner

Geomteric mean

Definition

The geometric mean is an average obtained by multiplying positive values and taking the root corresponding to the number of values. Equivalently, it is the exponential of their average logarithm. It is useful for multiplicative changes, ratios, and quantities naturally interpreted on a logarithmic scale. The supplied heading “Geomteric mean” contains a spelling error; the standard term used in this explanation is geometric mean.

What it means

An arithmetic mean answers an additive question: what equal value preserves the total? A geometric mean answers a multiplicative question: what equal factor preserves the product? The choice depends on the meaning of the data, not simply a preference for a smaller average.

For example, successive percentage changes act by multiplication. An increase of 20% multiplies a quantity by 1.20; a decrease of 25% multiplies it by 0.75. Average growth should be calculated from these positive factors, then translated back into a rate.

The geometric mean is not the geometric median, and it does not describe the center of a multivariate point cloud. It also does not automatically replace the arithmetic mean when totals, additive costs, or average quantities are the relevant question.

How it works

For positive values x1, …, xn, calculate G = (x1 × … × xn)^(1/n). A numerically convenient equivalent is G = exp[(log(x1) + … + log(xn))/n]. Working with logarithms can avoid excessively large or small intermediate products.

A weighted geometric mean uses exp[Σw_i log(x_i)/Σw_i] for nonnegative weights with positive total weight. Explain what the weights represent before applying them.

For period growth rates r_i, use factors 1 + r_i, calculate their geometric mean, and subtract one. Periods must be comparable or handled using an appropriate duration-based formulation. The usual logarithmic calculation requires strictly positive inputs. Zeros and negative values need explicit treatment; adding an arbitrary constant changes the question and is not a neutral repair.

Three practical examples

Operational growth: summarize successive changes in a positive activity measure through the constant per-period factor that reproduces its cumulative change.
AI benchmarking: summarize positive performance ratios across tasks when multiplicative relative performance is the intended comparison, while retaining task-level results and declared weights.
Measurement analysis: describe a typical positive measurement on a log scale when multiplicative variation is meaningful, alongside suitable information about dispersion and the original distribution.

Think of it this way

Think of a chain of adjustable gears. Each gear multiplies the effect of the previous one. The geometric mean is the identical gear ratio that, repeated the same number of times, produces the same overall multiplication. The arithmetic mean does not generally preserve that combined effect.

Worked example

Suppose an original activity index starts at 100 and changes over three equal periods: first +20%, then −25%, then +20%. Its path is 100 × 1.20 = 120, then 120 × 0.75 = 90, then 90 × 1.20 = 108.

The three factors have product 1.20 × 0.75 × 1.20 = 1.08. Their geometric mean is 1.08^(1/3), approximately 1.025985568. Subtracting one gives an equivalent constant growth rate of approximately 2.5986% per period.

Check the interpretation: 100 × 1.025985568³ = 108, apart from rounding. Independently averaging the logarithms and exponentiating gives the same factor. The arithmetic average of the three percentage changes is (20 − 25 + 20)/3 = 5%, which would instead produce 100 × 1.05³ = 115.7625 if repeated. It therefore does not reproduce this cumulative path. The geometric summary preserves the start-to-end multiplication, while hiding the intervening fall to 90.

Illustrative case study

Fictional case: ClearBench compares an AI tool’s speed with a reference tool across several tasks. The team uses positive speed ratios and a geometric average to summarize proportional performance.

Analysts publish the ratio direction, task weights, hardware conditions, and individual task results. They also report total runtime for the actual workload, since a geometric average of ratios does not directly determine total time saved. When a task fails or times out, they use a declared evaluation rule rather than silently dropping it to improve the summary.

Common misconceptions and mistakes

Taking the geometric mean of percentage numbers such as 20, −25, and 20 instead of their growth factors.
Using it for arbitrary negative values in an ordinary real-valued logarithmic calculation.
Dropping zeros, failures, or missing values without explaining the resulting change in the dataset.
Assuming an average ratio directly gives an average absolute saving.
Confusing the geometric mean with the median or with a geometric-median point.
Reporting a smooth equivalent growth rate as though every actual period had that rate.

Business use

Use the geometric mean when the business question concerns compounded factors or proportional comparisons. First establish that the ratios or positive values are comparable and that multiplicative averaging reflects the decision.

Report the number of observations, weights, missing-value rules, and relevant variability. For growth, distinguish the equivalent constant rate from the observed sequence. For benchmarks, keep absolute costs and performance by task visible. A single aggregate can be mathematically correct yet insufficient for choosing a system when workload proportions, failure cases, or the worst-performing tasks matter.

Check your understanding

Answer: 1.

What is the geometric mean of 2, 8, and 32?

Answer: The cube root of 512, which is 8. 2.

What factor represents a 10% decline?

Answer: 0.90. 3.

Does a constant equivalent rate reveal period-to-period volatility?

Answer: No; different paths can have the same cumulative factor.

Related terms

Harmonic mean
Logarithm
Geometric median
Arithmetic mean
compound growth
weighted mean
geometric standard deviation
performance ratio

Learn next

Compare additive and multiplicative averages on small examples. Practice converting rates into factors and back, then examine how task weights, unequal periods, and invalid values affect the quantity being summarized.

Sources

Howell glossary: https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html — supplied title and spelling verified; title retained for row fidelity.

NIST, Geometric Mean: https://www.itl.nist.gov/div898/software/dataplot/refman2/auxillar/geommean.htm — product-root definition read.

SciPy, gmean: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.gmean.html — weighted log formula and implementation checked. Worked values and case are original.

Previous: Geometric transformation Next: Georelational data model