Full Factorial Experimental Design
Also known as: factorial experiment, 2^k factorial, full factorial, Faktöriyel Deneme Deseni (Full Factorial, 2^k)
A full factorial design is a parametric experimental method in which every combination of factor levels is tested simultaneously, enabling the estimation of all main effects and all interaction effects in a single study. Rooted in R. A. Fisher's foundational work on designed experiments (1926) and systematically developed by Box, Hunter, and Hunter (2005) and Montgomery (2017), the 2^k form tests k two-level factors across 2^k experimental runs and is the benchmark against which all other factorial designs are measured.
Key highlights
- Estimates all main effects and every order of interaction without confounding.
- More statistically efficient than running one-factor-at-a-time experiments because all observations contribute to every effect estimate.
- Reveals factor synergies and antagonisms that would be invisible in single-factor studies.
- Backed by a rich theoretical framework (Box et al. 2005; Montgomery 2017) with well-established software support.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use a full factorial design when you need to study the simultaneous effects of two or more categorical or binary factors on a continuous response, and when you cannot assume that interactions among factors are negligible. Core assumptions are: all factor-level combinations are actually run (balance), responses are approximately normally distributed within each cell (or residuals are), variances across cells are homogeneous, and at least two replicates per cell are collected so that pure error can be estimated. With more than five or six factors the number of runs becomes prohibitive and a fractional factorial design should be considered instead.
Strengths & limitations
- Estimates all main effects and every order of interaction without confounding.
- More statistically efficient than running one-factor-at-a-time experiments because all observations contribute to every effect estimate.
- Reveals factor synergies and antagonisms that would be invisible in single-factor studies.
- Backed by a rich theoretical framework (Box et al. 2005; Montgomery 2017) with well-established software support.
- Run count grows exponentially: k = 6 factors at 2 levels already requires 64 runs per replicate.
- High-order interactions (three-way and above) are rarely interpretable and can obscure lower-order effects.
- Requires the assumption that all factor-level combinations are feasible and can be run in any order.
- Not suitable when sample size per cell falls below approximately 10 observations; interaction effects become unreliable.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How many runs does a full factorial design require?
For k factors each at 2 levels, exactly 2^k runs are needed per replicate. Two factors require 4 runs, three factors 8, four factors 16, and five factors 32. Adding replication multiplies these counts accordingly. When the number of required runs becomes too large, a fractional factorial design retains most of the information at a fraction of the cost.
What is an interaction effect and why does it matter?
An interaction exists when the effect of one factor on the response depends on the level of another factor. For example, a drug may be beneficial at low temperature but harmful at high temperature. Full factorial designs estimate all such interactions; designs that vary one factor at a time completely miss them, which can lead to incorrect conclusions or failed optimisation.
Must all cells have the same sample size?
Balanced designs — equal replication in every cell — are strongly preferred because they ensure orthogonality among effects, making each estimate independent and maximally precise. Unbalanced cells are not prohibited but require more complex analysis and can make main effects and interactions confounded.
When should I use a normal probability plot of effects?
When there is only one replicate per cell and therefore no degrees of freedom for pure error, a normal probability plot of all estimated effects is used as a graphical significance test. Effects that fall off the straight line formed by the majority of (presumably inert) effects are identified as active. This technique, described by Box et al. (2005), lets you identify important effects without a formal F-test.
Sources
- 1.Box, G. E. P., Hunter, J. S., & Hunter, W. G. (2005). Statistics for Experimenters: Design, Innovation, and Discovery (2nd ed.). Wiley.ISBN 978-0471718130
- 2.Montgomery, D. C. (2017). Design and Analysis of Experiments (9th ed.). Wiley.ISBN 978-1119113478
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Full Factorial Design. ScholarGate. https://scholargate.app/experimental-design/factorial-design