Full Factorial Design

Systematic experimental approach that tests all possible combinations of factor levels to identify main effects and interactions.

Design Overview & Methodology

What it accomplishes:

Full factorial designs test every possible combination of factor levels, providing complete information about main effects and all interaction effects. This systematic approach ensures no potential factor relationships are missed.

Advantages:

• • Detects all main effects and interactions
• • Provides complete factor information
• • High statistical power
• • Excellent for confirming hypotheses
• • Foundation for optimization studies

Considerations:

• • Number of runs grows exponentially (2^k)
• • Can be resource-intensive
• • May include unnecessary runs
• • Higher cost than screening designs

Effects Model & Run Count Formula

Number of runs for a 2-level full factorial

= number of factors

= total experimental runs required

Linear model with two-factor interaction

= measured response

= grand mean (intercept)

β1,β2

= main effect coefficients for factors A and B

β12

= interaction effect coefficient

= random error term

Main effect estimate

= average response at the high (+1) level of the factor

= average response at the low (-1) level of the factor

An effect is the change in mean response produced by moving a factor from its low to its high level; the coefficient equals half the effect.

Key Assumptions

• • Errors are independent and normally distributed with constant variance
• • Factor levels are set precisely and held constant during each run
• • Run order is randomized to protect against lurking (time-based) variables
• • Effects are additive unless an interaction term is explicitly modeled

Reading the Full Factorial Design Layout

Example Output & Result Interpretation

Worked Example (2² design):

Using the run matrix below (Factor A: Temperature, Factor B: Pressure), the main effect of A is estimated from the average response at A=+1 minus the average response at A=-1:

Main effect of Factor A

A large effect estimate relative to experimental noise (assessed via ANOVA p-value) indicates Factor A materially shifts the response.

Effect Analysis:

Main Effects:

• • Individual factor impact
• • Average response change per level
• • Statistical significance (p-value)
• • Practical significance (effect size)

Interaction Effects:

• • Factor dependency relationships
• • Non-additive behavior
• • Synergistic or antagonistic effects
• • Optimal factor combinations

Decision Rules:

• p-value ≤ 0.05: Effect is statistically significant — retain in the model

• p-value > 0.05: Effect is not distinguishable from noise — consider dropping

• R² Analysis: Model fit and predictive capability

• Residual Analysis: Check for model adequacy before trusting effect estimates

sixsigmacenter.com Implementation Guide

Design Construction:

1. Factor Identification:

• Define controllable factors (A, B, C, ...)

• Set high (+1) and low (-1) levels

• Ensure levels span practical ranges

2. Run Matrix Generation:

• Create all 2^k combinations

• Randomize run order

• Include center points if desired

3. Response Measurement:

• Collect data for each run

• Include replicates if possible

• Document experimental conditions

Platform Implementation:

The sixsigmacenter.com platform automatically generates optimal factorial designs, calculates effect estimates, performs ANOVA, creates interaction plots, and provides statistical significance testing with proper randomization.

When to Use Full Factorial Designs

Ideal Applications:

Optimization Studies:

• • Process parameter tuning
• • Product formulation optimization
• • Quality characteristic improvement
• • Response surface modeling

Interaction Investigation:

• • When interactions are suspected
• • Complex system behavior
• • Synergistic effects expected
• • Complete factor relationship mapping

Factor Guidelines:

• 2-4 factors: Highly recommended (4-16 runs)

• 5-6 factors: Feasible but consider fractional (32-64 runs)

• 7+ factors: Usually impractical (128+ runs)

• Mixed levels: Use general factorial designs (3^k, 4^k, etc.)

Example Data Layout for 2² Factorial Design

Standard 2² factorial with center points. Factor levels: A (Temperature: Low/High), B (Pressure: Low/High). Sample Size Recommendation: Minimum 2^k runs required, add 3-5 center points for curvature detection.

This is how your spreadsheet or CSV file should be structured before you upload it. Match the column names, keep one row per observation, and leave the analysis to the platform.