Statistical Methods

Choose a Harmonic Mean for Equal Amounts of Work

Define the task

First identify what is held equal. Averaging rates over equal amounts of work differs from averaging them over equal times. If two tasks involve the same number of units, each task duration is units divided by its rate; the combined rate is total units divided by total duration. This reasoning produces a harmonic mean. Do not choose a mean solely because the values are positive or skewed. Record the numerator and denominator units and the basis for equal weighting.

Carry out the check

For positive rates r, calculate n divided by the sum of 1/r. The tool accepts positive values only and uses equal weights. If work amounts differ, reconstruct the total amount and total time instead or use explicitly justified weights. Zero rates, downtime and missing records need a domain-specific rule; silently deleting them can inflate apparent throughput. Inspect the smallest rates because their reciprocals contribute the most. The tool reports a descriptive rate and does not estimate uncertainty or prove that a process is stable.

Worked hypothetical example

Imagine two fictional tasks of 120 units each, processed at 30 and 60 units per hour. Durations are 4 and 2 hours; 240 units over 6 hours gives 40 units per hour. The arithmetic mean, 45, instead corresponds to equal time spent at each rate. Recompute totals as an independent check, and retain the original rates rather than replacing all records with their summary. State explicitly that equal work, not equal observation time, motivated the calculation.

Method checked 16 November 2026 against SciPy documentation: hmean.

Open the equal-work harmonic rate summary.

Download the rate denominator reconciliation worksheet.

Related reading

Reading path: Analyze data. Guide 154 of 162. Choose any guide to suit your project.

Statistical Methods

Translate lognormal parameters before reporting an arithmetic mean

Translate logarithmic parameters into arithmetic moments and a median. Includes a worked hypothetical example and checks for parameter-transfer errors.

Statistical Methods

Distinguish Uncertainty in a Mean from Variation in a Future Observation

Choose an interval that matches whether the target is a population mean or a new individual measurement.

Browse this reading pathTools for this stageGuides and downloads for this stage