confounding

**Confounding** in DOE occurs when two or more effects **cannot be independently estimated** because they are mathematically mixed together in the experimental design. The effects are said to be "confounded" or "aliased" — a change in the response could be due to either effect, and the data cannot distinguish which one is responsible. **Why Confounding Happens** - In **fractional factorial** designs, the number of runs is reduced by deliberately confounding certain effects with each other. This is the price paid for using fewer runs. - In blocked designs, the block effect is confounded with specific interaction effects. - Confounding is **intentional** and controlled — the experimenter chooses which effects to confound, ideally confounding effects that are expected to be negligible. **Example** In a $2^{3-1}$ design (4 runs for 3 factors instead of 8): - Factor C is defined as $C = A \times B$. - This means the main effect of C is **confounded with** the AB interaction. - If the analysis shows a significant "C + AB" effect, you can't tell whether it's due to factor C, the A×B interaction, or both. **Alias Structure** The complete set of confounded effect pairs is called the **alias structure**. For the $2^{3-1}$ design: - $A$ is aliased with $BC$ - $B$ is aliased with $AC$ - $C$ is aliased with $AB$ **Resolution and Confounding** - **Resolution III**: Main effects confounded with 2-factor interactions — risky if interactions are important. - **Resolution IV**: Main effects clear, but 2-factor interactions confounded with other 2-factor interactions. - **Resolution V**: Main effects and 2-factor interactions clear — confounding only with 3-factor and higher interactions (usually negligible). - **Full Factorial**: No confounding at all — all effects independently estimated. **Managing Confounding** - **Assume Higher-Order Interactions Are Negligible**: Most physical processes have small 3+ factor interactions. If A is aliased with BCD, assume the observed effect is due to A. - **Follow-Up Experiments**: If confounded effects are both plausible, run additional experiments (fold-over designs) to de-alias them. - **Effect Hierarchy**: Prioritize main effects over interactions, and 2-factor interactions over 3-factor interactions. - **Subject Matter Knowledge**: Use process understanding to judge which of two aliased effects is more likely to be real. **Deliberate Confounding for Blocking** - When blocking a full factorial, the highest-order interaction is often confounded with the block effect. Since 3+ factor interactions are rarely important, this is a good trade — you gain the benefit of blocking while losing only negligible information. Confounding is the **fundamental tradeoff** in fractional factorial design — fewer runs in exchange for ambiguity about certain effects. Understanding and managing this tradeoff is essential for efficient experimentation.