Time-Mean vs Space-Mean Speed: Choose the Right Traffic Average

Which observation does each mean summarize?

Time-mean speed is the arithmetic average of observed vehicle speeds at a fixed location over an observation period. A segment travel-speed measure is based on distance and travel time. Confusing those observations can turn a reasonable detector output into an unsupported journey-time estimate.

The FHWA Travel Time Data Collection Handbook distinguishes point-based speeds from travel through a segment. Its basic measurement distinction remains useful; the historical handbook is not a claim about the newest sensing technology.

Define the consumer’s question first. “How fast did vehicles pass this location?” differs from “How long did vehicles take to traverse this block?” A signal near the downstream end can make the second answer very different even when the first is measured accurately.

The harmonic-mean connection

For vehicles traversing the same distance, dividing that distance by their average travel time equals the harmonic mean of their individual journey speeds. For positive speeds v₁ through vₙ, the harmonic mean is n / Σ(1/vᵢ). The arithmetic mean is Σvᵢ / n.

This relationship is exact for that defined equal-distance calculation. Using spot speeds to estimate a segment measure introduces additional assumptions about how speeds behave over the segment.

A two-vehicle example

Suppose two vehicles each traverse a 1 km segment. One maintains 30 km/h and takes 120 seconds; the other maintains 60 km/h and takes 60 seconds. These are hypothetical constant-speed journeys without stops.

The difference is not a sensor fault. The averages summarize the observations differently. Over this equal-distance example, the slower journey contributes more time, which lowers the distance-over-average-time result.

If those same two constant speeds are recorded as the vehicles pass a point, their arithmetic spot mean is 45 km/h. The equivalence with the segment calculation relies on the constant-speed assumption; it should not be transferred silently to a signalized street.

When is a harmonic mean of spot speeds insufficient?

When speed changes with location, a point may not represent the journey. A vehicle passing the detector at 40 km/h may stop at a downstream signal; another may accelerate after leaving a queue. Neither full journey can be reconstructed from that isolated spot speed.

FHWA’s Traffic Flow Theory chapter on traffic-stream characteristics explicitly discusses the difference when speeds vary over a road length. Treat a harmonic conversion of spot measurements as an estimator with stated assumptions, not a substitute for measured segment travel time.

For density calculations, align the definitions, observation windows and spatial coverage of flow, speed and density. The occupancy-versus-density guide explains why those quantities are not interchangeable. A dashboard should expose which speed field is being used.

An aggregation contract for the data feed

Specify the observation unit, location or segment, time window, speed definition, weighting, valid sample count and exclusion rules. Retain enough supporting data to recompute the statistic after a configuration change.

For an arithmetic vehicle mean across intervals, weight each interval mean by its valid vehicle count. Ten observations at a mean of 20 km/h and ninety at 60 km/h combine to 56 km/h, not the unweighted mean of 40 km/h. Again, these are illustrative values.

For equal-distance journey data, retain total travel time and trip count so the distance-over-mean-time calculation can be reproduced. If segment lengths differ, do not apply the equal-distance formula without defining the intended weighting. Combining lanes, vehicle classes or directions likewise needs an explicit purpose.

Keep delay and missing data visible

A stopped vehicle is not the same as an unavailable speed measurement. Do not replace missing values with zero or discard genuine delay merely to make a reciprocal calculation work. For a completed journey over a nonzero segment, include the elapsed time required by the metric’s definition, including stops if travel speed is the objective.

Test the export with unequal interval counts, a delayed journey and a missing record. Retain the expected results with the integration specification so changes to aggregation logic can be checked.

A label such as “average speed” is incomplete. An accepted feed states what was observed, how it was weighted and which real-world question the result can answer.