AI Glossary / Statistics and probability / Intermediate

Factor range

Definition

Factor range is the difference between the highest and lowest values of a quantitative factor used in an experimental design. If an experiment varies temperature from 60°C to 80°C, its temperature factor range is 20°C. The term describes the span investigated for an input, not the variation observed in the outcome or a guarantee that all intermediate settings were tested.

What it means

Choosing a range determines how far apart the experimental settings are. A narrow span studies a local region; a wider span may reveal larger changes or curvature but can involve different operating conditions and constraints.
It is useful to report both the endpoints and their difference. “A range of 20°C” alone does not say whether the experiment ran from 20°C to 40°C or from 60°C to 80°C. Those regions can have very different effects.

How it works

Identify the quantitative input and its measurement scale. Define the low and high settings, L and H, and calculate R = H − L. Specify units and whether the values are intended settings or actually measured inputs.
Choose endpoints using the experiment's objective, process knowledge, feasibility, and measurement precision. A screening experiment may use a broader region than a local refinement study. The largest physically possible range is not automatically the most informative or appropriate choice.
Check joint constraints. Two factors may each have feasible low and high settings, yet a particular combination of those settings may be impossible or unacceptable. The experimental region is therefore more than a list of separate ranges.
Consider the response shape. Two endpoints can estimate a contrast across the selected span, but cannot by themselves distinguish a straight relationship from many curved relationships that pass through the same points. Intermediate or center settings can provide additional information, depending on the design.
Use coding when helpful. For a two-level quantitative factor, center C = (H + L)/2 and half-range D = (H − L)/2. The coded value x = (X − C)/D maps the endpoints to −1 and +1. A zero range cannot be used in this formula and provides no planned variation for estimating that factor's effect.
Interpret effects together with the range. A larger high-minus-low response difference may simply reflect a larger input change. In a local linear model, divide that difference by H − L to obtain the corresponding slope in original units. This conversion does not justify assuming linearity beyond the studied region.
If a factor is transformed, distinguish the original-unit span from the transformed span. For example, equally spaced logarithmic levels are multiplicatively spaced in original units. Categorical factors generally have a set of levels rather than a meaningful numerical range.

Three practical examples

• A process team selects a pressure interval for an experiment and records both endpoints and the resulting span.
• An AI tuning study compares learning rates on a logarithmic scale, documenting that its coded spacing is not equal spacing in original units.
• A manufacturing team restricts combinations of speed and temperature because individually feasible extremes are not feasible together.

Think of it this way

Think of the stretch of a road you choose to inspect. Its length matters, but so does its location. Inspecting a short flat stretch cannot establish what happens beyond the next bend.

Worked example

An invented temperature experiment uses L = 50°C and H = 70°C. The range is 20°C, the center is 60°C, and the half-range is 10°C. Therefore x = (T − 60)/10.
Suppose a fitted local model is ŷ = 100 + 6x. At 50°C, x = −1 and ŷ = 94. At 70°C, x = +1 and ŷ = 106. The high-minus-low predicted difference is 12 response units.
The coded coefficient is six because the coded input changes by two units. In original units the slope is 12/20 = 0.6 response units per degree Celsius.
If a later experiment uses 55°C to 65°C and the same linear relationship holds locally, the predicted endpoint difference is six, not twelve. This is a conditional calculation, not evidence that the relationship remains linear or that either endpoint is optimal.

Illustrative case study

Fictional case: Elm Processing compares two experimental reports and finds different temperature effects. One study used a 10-degree span and the other a 30-degree span. Analysts first align the units and intervals, then examine whether curvature or different operating conditions explain the remaining difference. They avoid concluding that the process changed solely from the raw effect contrast. This invented case illustrates why a factor effect needs its range.

Common misconceptions and mistakes

Do not confuse factor range with response range, tolerance limits, or a confidence interval. Each describes something different.
Do not assume every value between the endpoints was observed. A two-level design may use only the endpoints.
Do not divide an effect by range and assume a universal slope when the response is nonlinear or interacts with other factors. A wider span can change the scientific question as well as the apparent effect size.

Business use

Factor ranges help teams design feasible experiments and interpret what their results support. Recording endpoints, units, and constraints makes comparisons across sites or studies more meaningful. For optimization, the studied region defines where evidence is strongest; decisions outside it require additional assumptions or new experiments.

People also ask

Is factor range the same as its set of levels?

No. Several different sets of levels can share the same endpoints and range.

Can a categorical factor have a numerical range?

Arbitrary numerical labels do not create a meaningful range.

Does a wider range always improve an experiment?

No. It can reveal more signal while also introducing curvature, constraints, or different operating regimes.

Check your understanding

What is the half-range in the example?

Answer: Ten degrees Celsius.

Why is the high-minus-low effect twice the coded coefficient?

Answer: The coded input changes from −1 to +1.

What information is missing from “range equals twenty”?

Answer: The endpoints, units, and relevant experimental conditions.

Related terms

• Factor level
• experimental region
• coded variable
• centre point
• response surface
• slope
• interaction
• extrapolation

Learn next

Practice converting between coded and original units. Then study center points, curvature detection, and constrained experimental regions before designing optimization experiments.

More in Statistics and probability

• F distribution - The F distribution is a continuous probability distribution obtained by taking the ratio of two independent chi-square random variables, each divided by its...
• F-test of equality of variances - The F-test of equality of variances is a classical hypothesis test comparing the population variances underlying two independent normal samples.
• Face-to-face survey - A face-to-face survey collects responses through an interviewer and respondent meeting in person.
• Factorial design - A factorial design is an experimental arrangement that studies two or more factors together by combining their levels.
• Failure bias - Failure bias is an informal label for drawing general conclusions from visible failures while overlooking relevant successes or other cases that did not fail.
• Failure rate - Failure rate, in lifetime analysis, is the instantaneous rate at which units fail conditional on having survived to a given time.

Sources

NIST/SEMATECH glossary, “factor range”: https://www.itl.nist.gov/div898/handbook/glossary.htm NIST/SEMATECH, selecting and scaling process variables: https://itl.nist.gov/div898/handbook/pri/section3/pri32.htm

Previous: Factor level Next: Factored frontier