Geometric Mean Calculator Decision guide
Arithmetic mean (AM) is correct for additive values: scores 72+80+88+96 → AM = 84. Geometric mean (GM) is correct for multiplicative values: growth factors 1.06×0.65×1.10 → GM = 0.9144 = −8.56%/yr. Harmonic mean (HM) is correct for equal-denominator rates: 30 mph and 60 mph over equal distance → HM = 40 mph. Median is correct for skewed data: incomes {$30k,$34k,$35k,$38k,$200k} → median = $35k (mean = $67.4k is 93% too high). Mode is correct for categories: survey {A,B,B,C,B,A} → mode = B.
Use AM when values add (scores, counts). Use GM when values multiply (growth factors, ratios). Use HM when rates share an equal denominator (equal-distance speed, equal-budget price-per-unit). Use median when outliers skew the mean. Use mode when values are categorical or when frequency is the question. When observations carry unequal importance, reach for a weighted mean before any of these.
Each example shows the correct average and what the most common wrong choice would produce.
AM is correct because scores add to a class total. AM×4 = 336 = sum ✓. GM is close but algebraically unjustified for additive data.
Check the input type and formula side by side before calculating.
| Method | The correct average depends on how the values combine — add, multiply, rate, rank, or repeat | A useful average must reproduce or predict the real-world outcome |
|---|---|---|
| Arithmetic mean | AM = sum/n. Preserves the sum. | Scores, counts, temperatures. Wrong for returns (AM of 50%, −50% = 0% but actual = −25%). |
| Geometric mean | GM = (product)^(1/n). Preserves the product. | Growth factors, ratios, fold changes. GM of 1.06×0.65×1.10 = 0.9144 → −8.56%/yr ✓. |
| Harmonic mean | HM = n / Σ(1/xᵢ). Preserves equal-denominator rates. | Speeds over equal distance, equal-budget price-per-unit. HM(30,60) = 40 mph ✓. |
| Median | Middle value when sorted. Ignores magnitudes. | Skewed data: income, home prices, response times. {$30k–$200k}: median = $35k ✓. |
| Mode | Most frequent value. Can be multiple or none. | Categories, survey answers, repeated choices. {A,B,B,C,B,A}: mode = B ✓. |
This is the clearest demonstration that no average is a safe default. The same four values support four defensible summaries that disagree by a factor of seven.
Situation: The dataset 2, 8, 32, 128 — each value four times the one before it.
| Step-by-step: how to use this calculator correctly | The calculation | When this is the right answer |
|---|---|---|
| Arithmetic mean | (2 + 8 + 32 + 128) / 4 = 42.50 | Correct if the four values add to a meaningful total — 170 units handled between them. |
| Geometric mean | (2 × 8 × 32 × 128)^(1/4) = 65,536^(1/4) = 16.00 | Correct if each value multiplies the next — a quantity doubling steadily over four periods. |
| Harmonic mean | 4 / (1/2 + 1/8 + 1/32 + 1/128) = 6.02 | Correct if the values are rates measured over a fixed distance, budget, or workload. |
| Median | (8 + 32) / 2 = 20.00 | Correct if one extreme value should not be allowed to dominate the summary. |
| The spread between them | 42.50 vs 16.00 vs 6.02 vs 20.00 | The data does not choose. The question you are asking chooses. |
On this dataset the arithmetic mean is seven times the harmonic mean, and both are correct answers to different questions. Decide what the average has to preserve — the sum, the product, the rate, or the middle — before you reach for a formula.
Three practical constraints that often settle the choice before the logic does.
Geometric mean needs strictly positive values, and harmonic mean cannot accept a zero at all. Arithmetic mean and median will take anything. If your data contains zeros or negatives, that constraint may decide the question before the add-versus-multiply test does.
For the set 98, 99, 100, 101 the four measures come out at 99.50, 99.49, 99.49, and 99.50. The choice only matters when values are widely spread — which is precisely the situation where people stop checking and reach for the default.
When observations carry unequal importance — different sample sizes, course credits, or capital allocations — every mean here has a weighted form. Consider a weighted mean before concluding that the plain average is simply wrong.
Answer these questions in order about your dataset. The first 'yes' determines the correct average.
Scores, counts, dollars, temperatures → use AM. Verify: AM × n should equal the total.
Growth factors, ratios, fold changes → use GM. Convert % returns to factors first (6% → 1.06).
Equal-distance speeds, equal-budget prices → use HM. If time is equal instead of distance, use AM.
Skewed income, home prices → use median. Compare AM with median: if AM > median by >20%, report both.
Survey answers, product choices → use mode. Report frequency distribution alongside mode.
Annual returns: +50%, −50%. AM of factors: (1.50+0.50)/2 = 1.00 → 0%/yr return (implies breakeven). Actual: $1,000×1.5×0.5 = $750. Loss = 25% over 2 years. GM: √(1.5×0.5) = √0.75 = 0.866 → −13.4%/yr. Only GM reproduces the actual ending value: $1,000×0.866² = $750 ✓.
Drive 100 km at 30 mph (takes 3.33 hrs), then 100 km at 60 mph (takes 1.67 hrs). Total: 200 km in 5 hrs → 40 mph = HM(30,60) ✓. AM = 45 mph implies 200/45 = 4.44 hrs → 33 min faster than reality. The error exists because the slower leg takes disproportionately more time.
A board report lists revenue growth percentages alongside headcount for five years. These need different averages in the same table: geometric mean down the growth column, arithmetic mean down the headcount column. Applying one formula to every column because it is the same spreadsheet is the most common way this goes wrong.
Customer spend of $20, $22, $25, $19, and $4,800 has an arithmetic mean of $977. The single enterprise order is not an error and should not be deleted, but it makes the mean describe nobody in the list. Report the median of $22 and disclose the outlier separately.
A correct calculation can still mislead when the method or input type is wrong.
Returns 50%, −50%: AM implies breakeven. Actual = 25% loss. Convert to factors and use GM: √(1.5×0.5)−1 = −13.4%/yr. AM overstates return by 13.4 percentage points.
A single zero makes the product = 0, so GM = 0 regardless of other values. If zero represents a complete wipeout (−100% return → factor = 0.0), GM correctly reports total loss. If zero is missing data, remove it first.
Incomes {$30k,$34k,$35k,$38k,$200k}: AM = $67.4k, median = $35k. AM overstates the typical salary by 93%. Always compare AM with median — if the gap exceeds ~20%, report both and explain the outlier.
There is no universally safe average. AM is the most common but is wrong for compounding (overstates returns by ignoring base changes) and for skewed data (overstates typical value). The safe approach: ask 'do values add, multiply, rate, rank, or repeat?' before choosing. Each question maps to exactly one correct formula.
Use GM when you have period-by-period growth factors: convert returns to factors (6% → 1.06, −35% → 0.65), take GM, subtract 1. Use CAGR when you have beginning value, ending value, and years: CAGR = (EV/BV)^(1/n) − 1. Both are geometric-mean operations on different input formats.
Use HM when distances are equal (each leg covers the same distance). Use AM when times are equal (each leg lasts the same duration). 30 mph and 60 mph: equal distance → HM = 40 mph ✓; equal time → AM = 45 mph ✓. They are different questions and different correct answers.
No. Arithmetic mean is the most common default and is wrong for compounding data, for rates sharing a denominator, and for skewed distributions. What the average must answer is what picks the formula — start from what has to be preserved: the sum, the product, the rate, or the middle.
Split it. A dataset mixing growth factors with absolute counts is really two datasets, and averaging them together produces a number with no interpretation. Compute each part with its own formula and report them side by side rather than forcing one summary.
State what the average preserves. "The geometric mean is 1.038, so applying 3.8% each year reproduces the actual ending balance" is a claim anyone can check. "We took the average" is not. Naming the preserved quantity turns a formula choice into a verifiable statement.