About

Averaging percentages is not a trivial operation. A naive arithmetic mean of 20% and 50% yields 35%, but if the first percentage applies to 1000 items and the second to 10 items, the true weighted result is approximately 20.3%. Ignoring sample sizes or weights produces misleading KPIs, incorrect financial projections, and flawed scientific conclusions. This tool computes four distinct averages - arithmetic, weighted, geometric, and harmonic - so you can select the method appropriate to your data structure. It assumes all input values represent ratio-scale percentages and that weights are strictly positive real numbers.

Limitations: the geometric mean requires all values to be positive (it is undefined for zero or negative percentages). The harmonic mean is undefined when any value equals zero. For compounding growth rates, convert percentages to growth factors before averaging; this calculator handles that conversion internally when computing the geometric mean. Pro tip: if your percentages represent conversion rates across segments of unequal size, always use the weighted average - never the simple mean.

Formulas

The arithmetic mean of n percentage values is defined as:

p = 1n n∑i=1 p_i

When each percentage p_i is associated with a weight w_i, the weighted average is:

p_w = n∑i=1 p_i ⋅ w_in∑i=1 w_i

The geometric mean, suited for compounding growth rates, is computed as:

p_g = ( n∏i=1 p_i )¹ⁿ

The harmonic mean, used when averaging rates, is:

p_h = nn∑i=1 1p_i

Where p_i = the i-th percentage value, w_i = the weight associated with the i-th value, and n = total number of entries.

Reference Data

Method	Formula	Best Use Case	Restriction
Arithmetic Mean	∑ p_in	Equal-weight samples, quick estimates	None
Weighted Mean	∑ p_i ⋅ w_i∑ w_i	Unequal sample sizes, survey data, portfolio returns	w_i > 0
Geometric Mean	√ⁿ ∏ p_i	Compounding growth rates, investment returns	p_i > 0
Harmonic Mean	n∑ 1p_i	Averaging rates (speed, efficiency, P/E ratios)	p_i ≠ 0
Median	Middle value when sorted	Skewed distributions, outlier resistance	n ≥ 1
Range	p_max − p_min	Spread assessment	n ≥ 2
Standard Deviation	√∑(p_i − p)²n	Variability measurement	n ≥ 2
Coefficient of Variation	σp × 100%	Comparing variability across datasets	p ≠ 0
Sum of Percentages	∑ p_i	Total cumulative percentage	None
Count	n	Number of entries	None
Minimum	min(p_i)	Lowest percentage in set	None
Maximum	max(p_i)	Highest percentage in set	None

Frequently Asked Questions

Use the weighted average whenever your percentages represent rates derived from groups of unequal size. For example, if Department A has a 90% pass rate across 200 employees and Department B has a 70% pass rate across 800 employees, the simple mean gives 80%, but the correct weighted average is (90×200 + 70×800) / (200+800) = 74%. Failing to weight by group size produces Simpson's Paradox - a statistically misleading result.

The geometric mean involves computing the product of all values and then taking the n-th root. If any value is zero, the entire product becomes zero, yielding a meaningless result. If any value is negative, the product may be negative, and even roots of negative numbers are undefined in real number arithmetic. For growth-rate analysis, convert negative returns to growth factors first: a −12% return becomes a factor of 0.88.

The harmonic mean gives greater weight to smaller values. It is the correct average when dealing with rates defined as ratios with a common numerator. For example, averaging speeds over equal distances (not equal times) requires the harmonic mean. If you drive 60 km/h for 100 km and 40 km/h for 100 km, the harmonic mean speed is 2/(1/60 + 1/40) = 48 km/h, not the arithmetic mean of 50 km/h.

Yes, the arithmetic and weighted means handle any real number. Values above 100% occur in year-over-year growth metrics (e.g., 150% growth). Negative percentages appear in loss scenarios. However, the geometric mean is restricted to positive values, and the harmonic mean is undefined at zero. This tool flags these constraints and skips the invalid method rather than producing an error.

The standard deviation (population σ) quantifies how spread out your percentage values are from the arithmetic mean. A low σ indicates the values cluster tightly, meaning the average is a reliable summary. A high σ signals significant dispersion - the average alone may be misleading. The coefficient of variation (CV = σ/mean × 100%) normalises this spread, enabling comparison across datasets with different scales.

When every weight equals 1, the weighted average formula reduces algebraically to the simple arithmetic mean: Σ(pᵢ × 1) / Σ(1) = Σ(pᵢ) / n. The weights represent relative importance or sample size. They do not need to sum to 100 or any specific number - only their ratios matter. Weights of 2, 4, 6 produce the same result as 1, 2, 3.