Full Factorial Design
Study several factors at once by testing every combination of their levels. Estimate all main effects and interactions, rank factor importance, and find the settings that optimize your response with a complete ANOVA.
Run Factorial Analysis →What is a Full Factorial Design?
A Full Factorial Design tests every possible combination of the levels of two or more factors. With a factors each at two levels, a full factorial requires 2a runs; with mixed levels it is the product of all level counts.
Its defining strength is that it estimates not only each factor's main effect but every interaction among them. Interactions, where one factor's effect depends on another factor's level, are invisible to one-factor-at-a-time testing and are often where the real process knowledge lies.
Because every combination is run, the design is orthogonal and the effect estimates are independent and efficient. The cost is that the run count grows exponentially with the number of factors, which motivates fractional factorials for large screening studies.
In plain terms: Instead of changing one knob at a time, you try every combination of knob settings. That way you learn not just what each knob does on its own, but how they work together, which is where surprises usually hide. The catch: more knobs means a lot more runs.
Design Concepts
Main Effects
The average change in the response when a factor moves from its low to high level, across all settings of the other factors.
Interaction Effects
When the effect of one factor depends on the level of another. A significant interaction means you cannot interpret the main effects in isolation.
2^k Designs
The most common screening layout: k factors each at two levels. A 2³ design has 8 runs and estimates 3 main effects, 3 two-way and 1 three-way interaction.
Key Formulas
Reading a Factorial Analysis
Start with the highest-order interactions. If a two-way interaction AB is significant, interpret A and B together using an interaction plot; do not read their main effects alone.
Rank effects by magnitude, not only by p-value. Effect-size ordering (often shown on a Pareto or normal probability plot of effects) reveals the vital few factors that drive the response, which is the practical goal of a factorial study.
Assumptions & Validation
Independence
Runs are performed and randomized independently to avoid time-order bias.
If violated: Randomize run order; add blocking if runs span batches.
Normality
Residuals from the fitted model are approximately normal.
If violated: Transform the response or use a generalized model.
Constant Variance
Residual spread is the same across the design space.
If violated: Apply a variance-stabilizing transformation such as log.
Correct Model Form
The chosen terms (main effects and interactions) adequately describe the response.
If violated: Add curvature checks (center points) or move to a response-surface design.
⚠️ Check assumptions first
A two-level full factorial assumes the response is linear between the low and high settings; it cannot detect curvature. If you suspect a peak or optimum lies inside the range, add center points to test for curvature, and if curvature is present, extend to a response-surface design. Also remember the run count doubles with each added factor, so screen before you optimize.
When NOT to Use Full Factorial Design
Many Factors to Screen
With six or more factors, a full factorial is expensive. Use a fractional factorial to screen for the vital few first.
Curvature Expected
Two-level factorials assume linearity. If an optimum lies between levels, use a response-surface design (central composite or Box-Behnken).
Uncontrollable Factors
If some factors cannot be freely set (noise factors), a Taguchi robust design may fit better than a standard factorial.
Industry Applications
Process Optimization
Find the temperature, pressure and time combination that maximizes yield, including their interactions.
Product Formulation
Test ingredient levels together to see how they jointly affect strength, taste or stability.
Manufacturing Robustness
Identify which settings most affect a quality characteristic so control effort focuses on the vital few.
Software & Marketing Experiments
Test multiple design elements at once to detect interaction effects a one-at-a-time test would miss.
Frequently Asked Questions
What is the advantage of a full factorial over one-factor-at-a-time testing?
One-factor-at-a-time (OFAT) testing changes a single factor while holding others fixed, so it can never detect interactions and is statistically inefficient. A full factorial tests all combinations, estimating every main effect and interaction with independent, orthogonal contrasts, and typically needs fewer runs than OFAT to achieve the same precision.
How many runs does a full factorial need?
The run count is the product of the number of levels of each factor. For k factors each at two levels it is 2 to the power k, so three factors need 8 runs, four need 16, and five need 32. This exponential growth is why fractional factorials are used to screen many factors.
What is an interaction effect?
An interaction occurs when the effect of one factor on the response depends on the level of another factor. For example, temperature may increase yield strongly at high pressure but weakly at low pressure. When an interaction is significant, the factors must be interpreted together, not through their main effects alone.
When should I use a fractional factorial instead?
When you have many factors and mainly want to identify which few matter, a fractional factorial runs a carefully chosen subset of combinations. It saves runs at the cost of aliasing some higher-order interactions with each other, which is usually an acceptable trade during screening.
Can a full factorial detect a curved response?
A standard two-level full factorial only fits straight lines between the low and high settings, so it cannot detect curvature. Adding center-point runs lets you test for curvature; if it is present, a response-surface design such as central composite or Box-Behnken is needed to model the optimum.
How do I identify the most important factors?
Rank the estimated effects by magnitude, often using a Pareto chart of effects or a normal probability plot. Effects that stand out from the near-zero cluster are the vital few driving the response. Combine effect size with the ANOVA p-values to decide which terms to keep in the final model.
Estimate Every Effect and Interaction
Run a full factorial and uncover the interactions OFAT testing misses. Free during Beta.
Run Factorial Analysis →