2. Defining the problem and response variables

Every successful design of experiments (DOE) begins long before the first factor is varied. The most common cause of a failed or inconclusive experiment is not poor statistical analysis — it is a poorly defined problem and an ambiguous response variable. This chapter focuses on the critical translation step: converting a vague engineering objective into measurable outcomes, identifiable factors, and actionable levels.

From Engineering Objective to Measurable Response

An engineering objective is typically stated in qualitative terms: "improve surface finish," "reduce cycle time," "increase yield," or "make the process more robust." These statements are useful for motivating a project, but they cannot be analyzed statistically. The first task is to translate each objective into one or more response variables — quantities that can be measured, recorded, and compared across experimental runs.

A well-defined response variable must satisfy several criteria:

• Measurable: It can be quantified with an instrument, sensor, or inspection method.
• Repeatable: Repeated measurements under identical conditions yield consistent values within a known measurement error.
• Relevant: It directly reflects the engineering objective, not a proxy that is convenient but weakly correlated.
• Unambiguous: The measurement procedure, units, and evaluation criteria are clearly specified.

Consider an injection molding process where the objective is "reduce warpage." A naive response might be "visual appearance," which is subjective and non-repeatable. A better response is "maximum out-of-plane deviation measured in millimeters using a coordinate measuring machine (CMM) at three specified datum points." This definition specifies the instrument, the metric, the units, and the measurement locations — all of which are necessary for reproducibility.

Types of Response Variables

Response variables fall into several categories, each with implications for experimental design and analysis:

Continuous responses are preferred whenever possible because they carry more information per run and are compatible with the widest range of statistical tools. When only pass/fail data can be collected, the experiment must be larger to achieve the same power, and the analysis must use appropriate binary response models.

Identifying Factors and Levels

Once the response is defined, the next step is to identify the factors — the input variables that are believed to influence the response and that can be deliberately set or controlled during the experiment. Factors can be:

• Continuous: Temperature, pressure, speed, concentration — set to specific numeric values.
• Categorical: Supplier, machine, material type, operator — set to discrete categories.
• Held constant: Variables that are not of interest but must be fixed to prevent confounding.
• Nuisance: Variables that cannot be controlled but can be recorded and included as covariates.

Each factor is tested at two or more levels — the specific values or settings used in the experiment. For a continuous factor such as temperature, levels might be 150°C and 180°C for a two-level screening design, or 150°C, 165°C, and 180°C for a response surface study. For a categorical factor such as supplier, levels are the specific suppliers included.

The choice of levels is not arbitrary. Levels should:

1. Span the range of practical interest — wide enough to reveal effects, but not so wide that the process fails or produces unsafe conditions.
2. Be achievable and repeatable with the available equipment.
3. Be chosen based on process knowledge, preliminary data, or a pilot study — not guesswork.

Structuring the Problem Definition

A structured problem definition ensures that the experimental team shares a common understanding before any resources are committed. The following elements should be documented:

• Problem statement: A concise description of the issue or opportunity.
• Objective: What the experiment is intended to achieve (e.g., maximize yield, minimize variability, identify significant factors).
• Response variables: Primary and secondary responses, with measurement methods and units.
• Factors: List of potential inputs, with classification (continuous/categorical) and proposed levels.
• Constraints: Budget, time, material, equipment, safety, or regulatory limits.
• Success criteria: What constitutes a successful outcome (e.g., "identify factors that reduce warpage by at least 20% with 95% confidence").

This structured definition becomes the foundation for selecting an experimental design, determining sample size, and planning the analysis.

Practical Example: Defining a Problem for Injection Molding

To illustrate the translation process, consider a real-world scenario. A manufacturer produces a polycarbonate housing and observes excessive warpage after molding. The engineering objective is "reduce warpage to meet dimensional specifications."

Step 1 — Define the response: The team selects "maximum deviation from flatness (mm)" measured on a CMM at three fixed datum points. The measurement procedure is documented, and a gauge R&R study confirms repeatability within ±0.05 mm.

Step 2 — Identify factors: A brainstorming session with process engineers and operators yields potential factors: melt temperature, mold temperature, packing pressure, packing time, cooling time, injection speed, and gate location. After discussion, the team selects four factors for the initial screening: melt temperature (continuous), packing pressure (continuous), cooling time (continuous), and gate location (categorical, two cavity designs).

Step 3 — Set levels: Based on historical process data and equipment capability, the team sets levels:

Factor              Low Level    High Level
Melt Temperature    220 °C       260 °C
Packing Pressure    40 MPa       80 MPa
Cooling Time        10 s         20 s
Gate Location       Design A     Design B

Step 4 — Define constraints and success criteria: The experiment is limited to 16 runs due to material and time constraints. Success is defined as identifying at least one factor that significantly reduces warpage (p < 0.05) and achieving a predicted warpage reduction of at least 20% at the optimal settings.

This structured definition now provides everything needed to select a 2⁴ factorial design, plan the runs, and proceed to analysis.

Common Mistakes and How to Avoid Them

• Vague responses: "Better quality" is not measurable. Always specify the metric, instrument, and units.
• Too many responses: Measuring everything increases cost and complexity. Focus on one primary response and a few secondary ones.
• Ignoring measurement error: If the measurement system is not capable, the experiment cannot detect real effects. Conduct a gauge R&R study before the DOE.
• Unrealistic levels: Levels that cause defects or unsafe conditions waste runs. Use pilot studies to validate the range.
• Missing constraints: Failing to account for budget, time, or material limits leads to incomplete or underpowered experiments.