What Is the Geometric Mean?
Unlike the arithmetic mean — which treats values as if they can simply be added — the geometric mean reflects the fact that many real-world quantities interact multiplicatively. When something grows by a percentage each period, or when you are comparing ratios, proportions, or index changes, the geometric mean is the mathematically appropriate average. The arithmetic mean is not.
A simple illustration: the numbers 2 and 8. Arithmetic mean = (2 + 8) / 2 = 5. Geometric mean = √(2 × 8) = √16 = 4. Neither answer is wrong — they answer different questions. The arithmetic mean is the additive balance point. The geometric mean is the multiplicative balance point.
The geometric mean is a measure of central tendency for positive values equal to the nth root of their product: GM = (x₁ × x₂ × … × xₙ)^(1/n). It is the correct average for multiplicative data such as growth rates, ratios, and investment returns. For all positive values, GM ≤ arithmetic mean, with equality only when all values are identical.
- Standard formula: GM = (x₁ × x₂ × … × xₙ)^(1/n)
- Log equivalent: GM = exp[(ln x₁ + ln x₂ + … + ln xₙ) / n]
- Domain: All values must be strictly positive (xᵢ > 0)
- AM–GM inequality: Arithmetic mean ≥ geometric mean for all positive values; equal only when all values are identical
- Best for: Growth rates, ratios, investment returns, index numbers, multiplicative processes, log-scale data
- Not suitable when: Any value is zero or negative, or the data combines additively
Geometric Mean Calculator
Geometric Mean Calculator
Enter positive values separated by commas or spaces. Switch tabs for growth-rate averaging or the weighted geometric mean. All calculations run locally in your browser — no data is sent anywhere.
Enter annual percentage growth rates (10 for +10%, −5 for −5%). The calculator converts each to a growth factor, computes the geometric mean, then converts back. A rate of −100% or worse cannot be processed.
Enter values and their corresponding positive weights. Formula: Weighted GM = exp[Σ(wᵢ × ln xᵢ) / Σwᵢ]. Both lists must have the same count.
Your calculations are performed locally in your browser. No personal information is required or transmitted.
The Geometric Mean Formula
The standard formula multiplies all n positive values and takes the nth root of the product:
xᵢ = each positive value
n = number of values
Π = product of all values
An equivalent form uses natural logarithms. Because ln converts multiplication into addition, the formula becomes:
ln = natural logarithm
exp = e raised to a power
Both forms give identical results. Statistical software — and the calculator above — uses the logarithmic version internally, because direct multiplication of many large numbers can overflow standard floating-point arithmetic.
Geometric Mean of Two Numbers
When there are exactly two positive values a and b, the formula reduces to the square root of their product:
Example — 4 and 9: GM = √(4 × 9) = √36 = 6. Arithmetic mean = (4 + 9) / 2 = 6.5. The geometric mean (6) is lower, as the AM–GM inequality requires.
Complete Formula Reference
| Formula | Expression | When to Use |
|---|---|---|
| Standard GM | GM = (Πxᵢ)^(1/n) | Direct calculation with small datasets |
| Logarithmic GM | GM = exp(Σln xᵢ / n) | Large datasets, programming, avoids overflow |
| Two-value GM | GM = √(a × b) | Exactly two positive values |
| Weighted GM | GM = exp[Σ(wᵢ ln xᵢ) / Σwᵢ] | Values with unequal importance |
| CAGR | CAGR = (Vₙ/V₀)^(1/n) − 1 | Compound growth from start to end value |
| Arithmetic mean | AM = Σxᵢ / n | Additive, interval-scale data |
| Harmonic mean | HM = n / Σ(1/xᵢ) | Rates with equal denominators (speed, fuel) |
How to Calculate the Geometric Mean — Step by Step
Confirm Every Value Is Strictly Positive
The standard geometric mean formula requires xᵢ > 0 for all observations. If any value is zero, the entire product becomes zero and the geometric mean is zero regardless of the other values. Negative values produce non-real results with the standard formula. Both cases require treatment beyond the introductory formula — see the domain restrictions section below.
Multiply All Values Together (or Sum Their Logs)
Calculate P = x₁ × x₂ × … × xₙ. For larger datasets, compute the sum of natural logs instead: L = ln x₁ + ln x₂ + … + ln xₙ. Both routes lead to the same geometric mean.
Take the nth Root (or Exponentiate the Mean Log)
Standard: GM = P^(1/n). Logarithmic: compute L / n, then GM = exp(L / n). Retain full precision throughout — round only the final displayed result to avoid accumulated rounding error.
Interpret the Result
The geometric mean is the single positive value that, if repeated n times in a multiplicative chain, produces the same total product as the original values. For growth-rate applications, it is the constant rate that produces the same total multiplicative change over the same number of periods.
Geometric Mean Examples — 20 Fully Worked
Example 1 — Two Numbers
Find the geometric mean of 2 and 8.
Formula: GM = √(2 × 8)
Product: 2 × 8 = 16
Square root: √16 = 4
✅ Geometric Mean = 4. Arithmetic mean = (2 + 8) / 2 = 5. The geometric mean (4) is lower, as the AM–GM inequality requires. Interpretation: 4 is the value that, squared, gives 16 — the same as 2 × 8.
Example 2 — Three Numbers
Find the geometric mean of 2, 4, and 8.
Formula: GM = (2 × 4 × 8)^(1/3)
Product: 2 × 4 × 8 = 64
Cube root: 64^(1/3) = 4
✅ Geometric Mean = 4. These three values form a geometric sequence (common ratio 2). A key property: the geometric mean of a geometric sequence equals its middle term.
Example 3 — Five Numbers
Find the geometric mean of 1, 2, 4, 8, and 16.
Product: 1 × 2 × 4 × 8 × 16 = 1,024
Fifth root: 1,024^(1/5) = 4
✅ Geometric Mean = 4. Arithmetic mean = (1+2+4+8+16)/5 = 6.2. The geometric mean equals the middle term of this geometric sequence (common ratio 2).
Example 4 — Decimal Values
Find the geometric mean of 1.5, 2.4, and 3.6.
Product: 1.5 × 2.4 × 3.6 = 12.96
Cube root: 12.96^(1/3) ≈ 2.3474
✅ Geometric Mean ≈ 2.35. Arithmetic mean = (1.5 + 2.4 + 3.6) / 3 = 2.50. The geometric mean is lower, as the AM–GM inequality requires.
Example 5 — Values Less Than 1
Find the geometric mean of 0.25, 0.5, and 1.0.
Product: 0.25 × 0.5 × 1.0 = 0.125
Cube root: 0.125^(1/3) = 0.5
✅ Geometric Mean = 0.5. Geometric means of values less than 1 are also less than 1 — this is valid. The only requirement is that all values remain strictly positive.
Example 6 — Logarithmic Method
Use logarithms to find the geometric mean of 3, 12, and 48.
Natural logs: ln(3) ≈ 1.09861 | ln(12) ≈ 2.48491 | ln(48) ≈ 3.87120
Sum: 1.09861 + 2.48491 + 3.87120 = 7.45472
Mean of logs: 7.45472 / 3 = 2.48491
Exponentiate: exp(2.48491) ≈ 12
✅ Geometric Mean = 12. Verify directly: (3 × 12 × 48)^(1/3) = 1,728^(1/3) = 12 ✓. The log method produces the same result while avoiding numerical overflow on large inputs.
Example 7 — GM vs Arithmetic Mean
Compare the arithmetic and geometric means of 1 and 100.
Arithmetic mean: (1 + 100) / 2 = 50.5
Geometric mean: √(1 × 100) = √100 = 10
✅ AM = 50.5, GM = 10. The values span two orders of magnitude, so the arithmetic mean is strongly pulled toward 100. The geometric mean (10) lies at the logarithmic midpoint: 10 is equally distant from 1 and 100 on a multiplicative scale (1 × 10 = 10; 100 / 10 = 10).
Example 8 — Investment Returns
A portfolio returns +10%, −5%, and +20% in three consecutive years. Find the geometric average annual return.
Growth factors: 1 + 0.10 = 1.10 | 1 + (−0.05) = 0.95 | 1 + 0.20 = 1.20
Product of factors: 1.10 × 0.95 × 1.20 = 1.25400
GM of growth factors: 1.25400^(1/3) ≈ 1.07833
Convert back: 1.07833 − 1 = 0.07833 ≈ 7.83% per year
✅ Geometric average annual return ≈ 7.83%. A $1,000 investment earning 7.83% compounded for 3 years reaches $1,000 × 1.07833³ = $1,254 — exactly what the three actual returns produce. The arithmetic average of the raw percentages (10 − 5 + 20) / 3 = 8.33% overstates the true compound effect.
Example 9 — CAGR
Revenue grew from $500,000 to $1,250,000 over 6 years. Calculate the CAGR.
Vₙ = ending value
V₀ = starting value
n = number of periods
Ratio: 1,250,000 / 500,000 = 2.5
Sixth root: 2.5^(1/6) ≈ 1.16472
CAGR: 1.16472 − 1 ≈ 16.47% per year
✅ CAGR ≈ 16.47%. A constant annual rate of 16.47% would take $500,000 to exactly $1,250,000 over 6 years. CAGR uses only the starting and ending values; the geometric mean of annual growth factors uses every year's individual factor — related concepts but not interchangeable.
Example 10 — Population Growth
A city's annual growth rates over four years were 2%, 3%, 1.5%, and 4%. Find the geometric average.
Growth factors: 1.02, 1.03, 1.015, 1.04
Product: 1.02 × 1.03 × 1.015 × 1.04 ≈ 1.10973
Fourth root: 1.10973^(1/4) ≈ 1.02621
Average rate: 1.02621 − 1 ≈ 2.62% per year
✅ Geometric average growth ≈ 2.62% per year. Arithmetic average = (2 + 3 + 1.5 + 4) / 4 = 2.625% — nearly identical here because the rates are small, but the geometric mean is always the correct choice for compounding growth rates.
Examples 11–20 — Summary Reference Table
| # | Values | n | Geometric Mean | Arithmetic Mean | Note |
|---|---|---|---|---|---|
| 11 | 4, 9 | 2 | 6 | 6.5 | √36 = 6 |
| 12 | 1, 10, 100 | 3 | 10 | 37.0 | Large spread; GM much lower than AM |
| 13 | 0.5, 2 | 2 | 1.000 | 1.250 | Values reciprocal; GM = 1 exactly |
| 14 | 6, 24 | 2 | 12.000 | 15.000 | √144 = 12 |
| 15 | 5, 5, 5, 5 | 4 | 5.000 | 5.000 | Identical values: GM = AM exactly |
| 16 | 1.10, 1.05, 1.08 | 3 | 1.07688 | 1.07667 | Growth factors near 1; GM ≈ AM |
| 17 | 10, 40, 160 | 3 | 40.000 | 70.000 | Geometric sequence; GM = middle term |
| 18 | 0.1, 1, 10, 100 | 4 | 3.162 | 27.775 | Log-scale data; GM far lower than AM |
| 19 | 2.5, 3.5, 4.5 | 3 | 3.398 | 3.500 | Close values; GM close to AM |
| 20 | 1.20, 1.30, 1.15, 1.25 | 4 | 1.2243 | 1.2250 | Annual growth factors; GM slightly below AM |
Examples 15 and 16 confirm two properties: when all values are identical, GM equals that common value; when values are very close together (near 1), GM and AM converge. Example 18 shows how sharply GM and AM diverge when data spans orders of magnitude.
Why Logarithms Work for the Geometric Mean
Logarithms convert multiplication into addition. Because ln(x₁ × x₂ × … × xₙ) = ln x₁ + ln x₂ + … + ln xₙ, the geometric mean can be computed by:
- Taking the natural log of each value
- Computing their arithmetic mean
- Exponentiating the result
This equivalence — GM = exp(arithmetic mean of logs) — is not just a computational convenience. It reveals the deeper statistical meaning: the geometric mean is the arithmetic mean on a logarithmic scale, then back-transformed. When data follows a lognormal distribution, the geometric mean equals the median of that distribution. Many real-world datasets in biology, finance, and environmental science are approximately lognormal, which is why the geometric mean appears so often in those fields.
For 100 values each equal to 10, the direct product is 10¹⁰⁰ — far beyond standard floating-point range. The log method computes (100 × ln 10) / 100 = ln 10 ≈ 2.3026, then exp(2.3026) = 10. No overflow, exact result. This is why Excel's GEOMEAN and all serious statistical software use this approach internally.
Geometric Mean vs Arithmetic Mean
Choosing between these two averages is a question about the data's structure, not personal preference. Using the wrong one produces misleading results.
| Feature | Arithmetic Mean | Geometric Mean |
|---|---|---|
| Formula | AM = Σx / n | GM = (Πx)^(1/n) |
| Data type | Additive (temperature, height, test score) | Multiplicative (ratio, rate, return, index) |
| Domain | Any real numbers | Strictly positive values only |
| AM–GM inequality | AM ≥ GM for all positive values; equality only when all values are identical | |
| Sensitivity to large values | Pulled strongly toward outliers | Less pulled on a multiplicative scale |
| Classic example: 1 and 9 | AM = (1+9)/2 = 5 | GM = √(1×9) = 3 |
| Investment returns | Overstates compound return when rates vary | Gives the true compound average rate |
| Best for | Heights, weights, exam scores, summed totals | Growth rates, CAGR, ratios, index numbers |
The AM–GM inequality follows from the concavity of the logarithm function (Jensen's inequality). In plain terms: when values vary, multiplication compresses the large values more than addition does, pushing the geometric mean below the arithmetic mean. The gap between the two grows with the spread of the data.
Geometric Mean vs Harmonic Mean vs Median
There are four main measures of central tendency. Each genuinely measures something different, so selecting the right one matters.
| Measure | Arithmetic Mean | Geometric Mean | Harmonic Mean |
|---|---|---|---|
| Formula | Σx / n | (Πx)^(1/n) | n / Σ(1/x) |
| Best for | Sums, interval data, additive quantities | Products, ratios, growth, multiplicative data | Rates (speed, fuel efficiency) with equal denominators |
| Example: 2, 4, 8 | (2+4+8)/3 = 4.67 | 64^(1/3) = 4.00 | 3/(½+¼+⅛) ≈ 3.43 |
| Ordering rule | HM ≤ GM ≤ AM for all positive values (equality when all values are identical) | ||
The median — the middle value after sorting — does not follow the HM–GM–AM ordering because it uses position rather than algebraic computation. For the dataset 2, 4, 8 the median is 4, which coincidentally equals the geometric mean here, but that is not generally true. The median is resistant to extreme values in a way the geometric mean is not, because the geometric mean multiplies all values together.
Deciding which measure to report requires understanding what the data represents. Speed averaged over equal time intervals uses the arithmetic mean. Speed averaged over equal distances uses the harmonic mean. Compound growth uses the geometric mean. For a deeper comparison of all three means alongside the median and mode, see the mean vs median vs mode guide on Statistics Fundamentals.
Weighted Geometric Mean
When values do not contribute equally — for instance, when some observations represent more data, or when assets represent different portfolio shares — the weighted geometric mean assigns different importance through a weight wᵢ attached to each value.
xᵢ = each positive value
wᵢ = weight for value i (must be positive)
Σwᵢ = sum of all weights
When all weights are equal, this formula reduces to the standard geometric mean. The weighted version is used in portfolio performance analysis, economic index construction, and any situation where observations carry different levels of importance.
Three assets return 5%, 10%, and 15%. They represent 20%, 50%, and 30% of a portfolio. Find the weighted geometric mean return.
Convert returns to growth factors: 1.05 (w=0.20) | 1.10 (w=0.50) | 1.15 (w=0.30)
Weighted log sum:
0.20 × ln(1.05) + 0.50 × ln(1.10) + 0.30 × ln(1.15)
= 0.20 × 0.04879 + 0.50 × 0.09531 + 0.30 × 0.13976
= 0.009758 + 0.047655 + 0.041928 = 0.099341
Σwᵢ: 0.20 + 0.50 + 0.30 = 1.00
Mean weighted log: 0.099341 / 1.00 = 0.099341
Exponentiate: exp(0.099341) ≈ 1.10444
Weighted return: 1.10444 − 1 ≈ 10.44% per year
✅ Weighted Geometric Mean Return ≈ 10.44%. The simple unweighted geometric mean of the three growth factors would give ≈ 9.92% — different because it treats each asset as equally important regardless of portfolio share.
Domain Restrictions: Zero and Negative Values
Both GM = (Πxᵢ)^(1/n) and GM = exp(Σln xᵢ / n) require every value to be strictly greater than zero. Zero and negative values cause specific failures explained below.
When a value equals zero: The product of all values becomes zero, forcing GM = 0 regardless of every other value. Additionally, ln(0) approaches −∞, so the logarithmic method is undefined. A single zero in a dataset makes the standard geometric mean zero — a result that rarely reflects any meaningful center of the data.
When values are negative: With an even count of negative values the product is positive (a real result emerges by accident), but with an odd count the product is negative and the nth root is not a real number. Because behavior is inconsistent across cases and loses intuitive meaning, the standard formula should not be applied to datasets containing negative values without explicit mathematical justification beyond the introductory level.
The standard workaround for investment data with losses: Convert percentage returns to growth factors by adding 1. A −30% return becomes 0.70 — a positive growth factor. Calculate the geometric mean of the growth factors (all positive), then subtract 1. This is the mathematically correct approach used in finance, provided no return is −100% or worse (which would give a growth factor of zero or below).
When to Use the Geometric Mean
The geometric mean is the appropriate average whenever quantities interact multiplicatively rather than additively. That covers a broad range of fields:
Investment Returns
Multi-year portfolio performance and compound growth require the geometric mean. Arithmetic averaging of returns overstates performance when returns vary across periods.
Economic and Population Growth
GDP growth over multiple periods, population growth rates, and similar compounding quantities are correctly summarized by the geometric mean.
Scientific Measurements on Log Scales
pH, seismic magnitude, decibels, and other logarithmic-scale measurements span orders of magnitude. The geometric mean is more representative than the arithmetic mean for such data.
Index Numbers
Many price indices and market indices weight constituent performance geometrically to avoid the distortion that a single high-priced asset creates under arithmetic averaging.
Biological Growth Rates
Cell doubling times, bacterial colony growth, and similar exponential biological processes are naturally summarized by the geometric mean.
Ratios and Proportions
When the data itself is a ratio — exchange rates, relative risks, concentration ratios — the geometric mean respects the ratio structure where the arithmetic mean does not.
Mathematical Properties of the Geometric Mean
- Positivity: GM is always positive when all input values are positive.
- Identity: If all values equal c, then GM = c (matches the arithmetic mean in this special case).
- AM–GM inequality: AM ≥ GM for all positive values; equality holds only when all values are identical.
- HM–GM–AM ordering: HM ≤ GM ≤ AM for all positive values.
- Multiplicative scaling: GM(kx₁, kx₂, …, kxₙ) = k × GM(x₁, x₂, …, xₙ) for any positive constant k.
- Log–arithmetic link: GM = exp(mean(ln x)), so the geometric mean is the back-transformed arithmetic mean on a log scale.
- Sensitivity to small values: A very small value pulls the geometric mean down strongly, because all values multiply together.
- Middle-term property: For a geometric sequence, the geometric mean of all terms equals the middle term.
Common Mistakes with the Geometric Mean
| Mistake | What Goes Wrong | Correct Approach |
|---|---|---|
| Adding values instead of multiplying | Computing (Σx)^(1/n) instead of (Πx)^(1/n) | Multiply all values: (x₁ × x₂ × … × xₙ)^(1/n) |
| Dividing the product by n | Product / n gives the wrong answer every time | Take the nth root: Product^(1/n) |
| Averaging raw percentage returns | (10 − 5 + 20) / 3 = 8.33% overstates compound return | Convert to growth factors first, then apply geometric mean |
| Using GM with zero | Result is always 0; log method undefined | Standard GM does not apply to datasets containing zero |
| Using GM with negative values | Inconsistent or non-real results | Convert returns to growth factors; ensure all exceed zero |
| Confusing GM with CAGR | CAGR uses start/end values only; geometric mean of factors uses every period | CAGR = (Vₙ/V₀)^(1/n) − 1; GM of growth factors − 1 when all individual factors are known |
| Confusing GM with median | The median is positional; GM uses all values multiplicatively | Use median for positional center; GM for multiplicative center |
| Rounding intermediate values | Rounding ln(x) values introduces compounding error | Keep full precision throughout; round only the final displayed result |
| Applying GM to additive data | GM of test scores distorts the central tendency | Use arithmetic mean for interval-scale additive quantities |
Practice Problems — 25 with Solutions
Work through each problem before revealing the answer. Solutions show full working.
Find the geometric mean of 3 and 27.
Find the geometric mean of 4, 16, and 64.
Use the logarithmic method to find the geometric mean of 5, 10, and 50.
A stock returned +25%, −20%, and +10% over three years. Find the geometric average annual return.
An investment grew from $2,000 to $3,500 over 5 years. What is the CAGR?
For the values 1, 4, and 16, compute both the arithmetic mean and geometric mean. Which is larger?
Two assets return 8% and 12%. Asset A has weight 0.6, Asset B has weight 0.4. Find the weighted geometric mean return.
Can you compute the standard geometric mean of 5, 0, 10, 15? Explain why or why not.
What is the geometric mean of 7, 7, 7, 7, 7?
GDP growth rates over three years: 3.2%, 5.5%, 2.1%. Find the geometric average.
Find the geometric mean of 9 and 25.
Find the geometric mean of 0.1, 0.5, and 2.5.
Verify the AM–GM inequality for 2, 6, 18.
Find the geometric mean of 1/4 and 4.
Find the geometric mean of 1,000 and 1,000,000 using the log method.
Why must you convert percentage returns to growth factors before applying the geometric mean?
A dataset has GM = 8 and n = 3. What does this mean in plain language?
Values 2, 8, and 32 have weights 3, 2, and 1 respectively. Find the weighted geometric mean.
For 4, 8, and 16, compute AM, GM, and HM and confirm the ordering.
A company's price-to-earnings ratios across three quarters were 18, 22, and 15. Which mean is most appropriate and what is its value?
What Excel formula calculates the geometric mean of cells A1 through A10?
Write a Python one-liner to compute the geometric mean of a list of positive values.
For which dataset does the AM–GM inequality become an equality?
The geometric mean of 3, 6, and 12 is 6. What is the geometric mean of 6, 12, and 24?
Explain the difference between the geometric mean of annual growth factors and CAGR.
Frequently Asked Questions
The geometric mean is the multiplicative middle value of a set of positive numbers. You calculate it by multiplying all the numbers together and taking the nth root of the result, where n is how many numbers you have. For two numbers it is their square root. For three numbers it is the cube root of their product. It gives you the "central" value in proportional terms rather than additive terms.
Standard form: GM = (x₁ × x₂ × … × xₙ)^(1/n). Logarithmic form: GM = exp[(ln x₁ + ln x₂ + … + ln xₙ) / n]. Both require all values to be strictly positive. The logarithmic form is preferred for large datasets and programming because it avoids numerical overflow.
GM = √(2 × 8) = √16 = 4. Arithmetic mean = (2 + 8) / 2 = 5. The geometric mean (4) is lower, consistent with the AM–GM inequality. Interpretation: 4 is the number that, when squared, equals 16, the same as 2 × 8.
GM = √(4 × 9) = √36 = 6. Arithmetic mean = (4 + 9) / 2 = 6.5.
GM = √(1 × 9) = √9 = 3. Arithmetic mean = (1 + 9) / 2 = 5. The gap between GM (3) and AM (5) is large here because the values 1 and 9 are nine times apart, so the proportional spread is large even though the absolute spread seems modest.
This follows from Jensen's inequality applied to the concave logarithm function. In plain terms: when values vary, multiplication "compresses" larger values more than addition does, so the geometric mean ends up below the arithmetic mean. The two are equal only when all values are identical, when there is no variation, there is nothing to compress.
No. For all positive values, AM ≥ GM without exception (the AM–GM inequality). The geometric mean can only equal the arithmetic mean, and that happens only when every value in the dataset is identical.
Not with the standard real-valued formula. When the count of negative values is odd, the product is negative and the nth root is complex (non-real). When the count is even, the product is positive by coincidence, but the result loses interpretable meaning. For investment returns that include losses, convert each percentage to a growth factor (add 1) so all inputs are positive, then apply the geometric mean.
The product-based formula gives GM = 0 whenever any value is zero, regardless of the other values. The log-based formula is undefined because ln(0) is not a finite number. The standard geometric mean does not apply to datasets containing zero. Some contexts define GM = 0 by convention when zero is present, but this must be stated explicitly rather than assumed.
Because ln(a × b) = ln(a) + ln(b), the logarithm converts multiplication into addition. So: ln(x₁ × x₂ × … × xₙ) = ln x₁ + ln x₂ + … + ln xₙ. Taking the arithmetic mean of these log values gives (ln x₁ + … + ln xₙ) / n. Exponentiating reverses the log: GM = exp[(Σ ln xᵢ) / n]. This is mathematically equivalent to the direct formula but avoids overflow when values are very large.
Use the geometric mean when values interact multiplicatively, when the next period's quantity is the current quantity times a factor rather than plus a constant. Growth rates, percentage changes, financial returns, ratios, and log-scale measurements are all multiplicative. Use the arithmetic mean for additive data such as heights, test scores, or temperatures.
The geometric mean multiplies all values and takes the nth root, so it uses every value. The median sorts values and picks the middle one, so it uses only position. The median is resistant to extreme values; the geometric mean is affected by all values through multiplication. For lognormal data the two are sometimes close, but they are conceptually and numerically distinct.
The geometric mean of annual growth factors is the constant factor that produces the same total multiplicative change over the same number of periods. CAGR = (Vₙ/V₀)^(1/n) − 1 uses the same logic with only the start and end values. When every annual factor is known, GM of factors − 1 equals the geometric average return. When only start and end values are known, CAGR gives the equivalent constant rate.
The weighted geometric mean assigns different importance to each value: Weighted GM = exp[Σ(wᵢ × ln xᵢ) / Σwᵢ]. It is used in portfolio analysis (weights = portfolio shares), economic index construction (weights = market capitalizations), and any setting where observations have different levels of importance. When all weights are equal it reduces to the standard geometric mean.
Yes. The geometric mean is not restricted to integers. GM of 1.5 and 2.4 = √(1.5 × 2.4) = √3.6 ≈ 1.897, not a whole number. The result is a decimal whenever the nth root of the product is not an integer, which is the typical case for real-world data.
Use =GEOMEAN(A1:A10) for a range. Excel returns an error for zero or negative values. The log equivalent =EXP(AVERAGE(LN(A1:A10))) works the same way; in older Excel it requires Ctrl+Shift+Enter as an array formula, while Excel 365 handles it normally.
For any set of positive real numbers, the arithmetic mean is always greater than or equal to the geometric mean: AM ≥ GM. Equality holds if and only if all values are equal. This is one of the most frequently used inequalities in mathematics, with applications ranging from optimization to information theory.
In descriptive statistics, the geometric mean is a measure of central tendency for positive, multiplicative, or ratio-scale data. It is appropriate when the data follows (or is suspected to follow) a lognormal distribution, when averaging growth rates, when combining index ratios, or when the measurement scale is logarithmic.
They are closely related but not identical. CAGR = (Vₙ/V₀)^(1/n) − 1 uses only starting and ending values. The geometric mean of growth factors uses every individual period's factor. For a single investment with no external cash flows, they give the same numerical result. For a portfolio where individual period returns are recorded, the geometric mean of the factors is the more detailed calculation.
Geometric mean: GM = (Πx)^(1/n), appropriate for multiplicative data such as growth and ratios. Harmonic mean: HM = n / Σ(1/x), appropriate for rates where the denominators add, such as average speed over equal distances. For positive values, HM ≤ GM ≤ AM. The harmonic mean is smaller than the geometric mean except when all values are equal.
You cannot apply the geometric mean directly to percentage growth rates because they can be negative and the formula requires positive values. The correct procedure is: convert each percentage to a growth factor (add 1: a 15% rate becomes 1.15, a −5% rate becomes 0.95), compute the geometric mean of the growth factors, then subtract 1 to convert back to a rate. This gives the geometric average growth rate.
The geometric mean is the constant value that, if used as each element in a multiplicative sequence of length n, produces the same overall product as the original values. For growth rates, it is the constant periodic rate that produces the same cumulative change. For ratios or index numbers, it is the typical ratio. It is not a "middle" in the additive sense, it is a middle in the proportional, logarithmic sense.
Do not use the geometric mean when: the data contains zero or negative values (without converting to growth factors); the data is purely additive (exam scores, temperatures, lengths); the measurement scale is ordinal rather than ratio; or when you need a measure that is resistant to extreme values in the positional sense (use the median instead). The geometric mean is the right tool for multiplicative, ratio-scale, positive data.
Because all values are multiplied together, a very small positive value pulls the geometric mean down substantially. For example, adding a value of 0.001 to the dataset {10, 100, 1000} reduces the geometric mean from 100 to (0.001 × 10 × 100 × 1000)^(1/4) = 1,000^(1/4) ≈ 5.62. This sensitivity to small values is a characteristic of the geometric mean to be aware of when interpreting results.
Yes, indirectly. Convert each percentage return r to a growth factor (1 + r/100). A −20% return becomes 0.80, which is positive. Compute the geometric mean of the growth factors. Subtract 1 to convert back to a rate. This works as long as no return is −100% or worse, because a −100% return gives a growth factor of 0, which is invalid for the geometric mean.
The Geometric Mean in Descriptive Statistics
In descriptive statistics, the geometric mean belongs to the family of measures of central tendency alongside the arithmetic mean, median, and mode. All of these summarize where a dataset is centered, but the choice of measure determines whether the summary is meaningful or misleading.
The geometric mean has a precise statistical connection to the lognormal distribution. When data is lognormally distributed — that is, when the logarithms of the values follow a normal distribution — the geometric mean of the data equals the median of the distribution and the back-transformed arithmetic mean of the log values. Many biological measurements, financial returns, particle sizes, and environmental concentrations follow approximately lognormal distributions, which explains why the geometric mean appears so naturally in those fields.
For computing these measures side by side, the mean calculator, median calculator, mean median mode calculator, and descriptive statistics calculator are available on Statistics Fundamentals. For the full context of how growth rates, compound interest, and the geometric mean connect to probability, the statistics and probability section covers the foundational theory.