What Is the Harmonic Mean?
The harmonic mean is a type of average calculated by dividing the number of observations by the sum of the reciprocals of each value. It gives the greatest weight to the smallest values in a dataset, making it the appropriate average when dealing with rates, ratios, and any quantities defined in terms of a unit denominator. If a delivery driver completes one route at 30 mph and another at 60 mph, the arithmetic mean (45 mph) overstates average speed. The harmonic mean (40 mph) gives the correct answer because speed is a rate, distance per time, and equal distances were covered at different rates. The harmonic mean is always the smallest of the three Pythagorean means (arithmetic, geometric, harmonic).
Why the Harmonic Mean Matters
Rates and ratios show up constantly in research and business metrics, cost per acquisition, responses per hour, pages per visit. Averaging these with the arithmetic mean produces inflated results that misrepresent actual performance. The harmonic mean corrects for this by properly accounting for the denominators. It's less commonly taught than arithmetic and geometric means, which means it's also more commonly misapplied, researchers who don't know about it default to the arithmetic mean and get wrong answers.
How the Harmonic Mean Works
The Formula
HM = n / Σ(1/xᵢ)
Where:
- n = the number of values
- xᵢ = each individual value
- Σ(1/xᵢ) = the sum of the reciprocals
Worked Example: Average Speed
You drive 120 miles to a meeting at 60 mph and 120 miles back at 40 mph. What's your average speed for the round trip?
Arithmetic mean: (60 + 40) / 2 = 50 mph
But let's check: 120 miles at 60 mph = 2 hours. 120 miles at 40 mph = 3 hours. Total: 240 miles in 5 hours = 48 mph.
Harmonic mean: 2 / (1/60 + 1/40) = 2 / (0.01667 + 0.025) = 2 / 0.04167 = 48 mph
The harmonic mean gives the correct answer (48 mph) while the arithmetic mean (50 mph) overestimates.
Worked Example: Cost Per Acquisition
You ran two ad campaigns:
- Campaign A: $5,000 spend, $25 cost per acquisition (CPA) → 200 acquisitions
- Campaign B: $5,000 spend, $50 CPA → 100 acquisitions
Arithmetic mean CPA: ($25 + $50) / 2 = $37.50
Actual blended CPA: $10,000 total spend / 300 total acquisitions = $33.33
Harmonic mean: 2 / (1/25 + 1/50) = 2 / (0.04 + 0.02) = 2 / 0.06 = $33.33
Again, the harmonic mean matches reality. The arithmetic mean overstates the average CPA because it doesn't account for the fact that the cheaper campaign produced more conversions.
When the Harmonic Mean Applies
The harmonic mean is correct when you're averaging rates and the numerator (the "stuff" being measured) is held constant across observations:
- Equal distances at different speeds → harmonic mean of speeds
- Equal budgets at different CPAs → harmonic mean of CPAs
- Equal work at different rates → harmonic mean of rates
If the denominator is held constant instead (e.g., equal time at different speeds), the arithmetic mean is correct. The key question is: "what's being held equal?"
Relationship to Arithmetic and Geometric Means
For any set of positive, unequal numbers:
Harmonic Mean ≤ Geometric Mean ≤ Arithmetic Mean
This is called the AM-GM-HM inequality. The three means are equal only when all values are identical. The more spread out the values, the larger the gaps between the three means.
For two numbers a and b:
- Arithmetic mean = (a + b) / 2
- Geometric mean = √(a × b)
- Harmonic mean = 2ab / (a + b)
The F1 Score Connection
In machine learning and information retrieval, the F1 score is the harmonic mean of precision and recall:
F1 = 2 × (Precision × Recall) / (Precision + Recall)
The harmonic mean is used here because it penalizes extreme imbalances, a classifier that achieves 99% precision but only 1% recall gets an F1 of 0.02, not the arithmetic mean of 50%. This property makes the harmonic mean valuable any time you want an average that's sensitive to low values.
When to Use the Harmonic Mean
- Averaging speeds when equal distances are covered at different rates
- Averaging rates like cost per unit, time per task, or output per hour when the "per" quantity is equal
- Calculating F-scores that balance precision and recall in classification tasks
- Financial ratios like price-to-earnings ratios when comparing across equal investments
- Any scenario where you're averaging denominators and the numerators are equal
Common Mistakes to Avoid
- Using the arithmetic mean for rates when the harmonic mean is correct: this is the most common mistake; whenever you're averaging "something per something," check whether the harmonic mean applies
- Applying the harmonic mean when the arithmetic mean is appropriate: if equal time (not equal distance) is spent at different speeds, the arithmetic mean is correct
- Including zeros in the calculation: the harmonic mean is undefined when any value is zero because you'd be dividing by zero
How Quali-Fi Supports Rate-Based Analysis
Quali-Fi's reporting engine recognizes rate-based metrics (cost per response, time per survey, responses per hour) and applies the appropriate averaging method. When you aggregate rate data across segments or time periods, the platform uses harmonic means where applicable to prevent inflated summaries.
Get accurate rate averages with Quali-Fi
Frequently Asked Questions
Why is the harmonic mean always the smallest?
Because it gives the most weight to the smallest values. The reciprocal of a small number is a large number, which dominates the sum of reciprocals in the denominator. This pulls the result toward the lower values in the dataset. It's a feature, not a bug, in rate-based problems, smaller rates (lower speeds, lower efficiency) have more impact on the overall average because they consume more of the shared resource.
Can the harmonic mean be negative?
If all values are negative, the harmonic mean will be negative. If values have mixed signs, the harmonic mean can produce counterintuitive results. In practice, the harmonic mean is almost exclusively used with positive values (rates and ratios), where negative values don't make conceptual sense.
When would I use harmonic mean in market research?
The most common application is averaging rate-based KPIs across segments or time periods, like cost per response, average revenue per user (ARPU), or survey completion rates. Any time you're combining "per-unit" metrics and the units are equal across what you're averaging, the harmonic mean gives the correct answer.