Factorial Design Explained: Testing Multiple Factors

What is a Factorial Design?

A factorial design is an experimental design that simultaneously assesses more than one factor. By evaluating multiple factors at the same time, this design uncovers not only individual effects but also how factors interact. With this technique, each experimental run involves a random combination of factor values in a structured setting.

For example, imagine testing a new recipe for baking cookies. You might wonder if the results depend on baking temperature (325°F or 375°F) and type of sugar (white or brown). A factorial design allows you to compare all combinations—cookies baked at 325°F with white sugar, 325°F with brown sugar, 375°F with white sugar, and 375°F with brown sugar. This setup helps answer questions like: Does the type of sugar matter more at higher temperatures? Does baking temperature affect texture differently depending on the sugar used?

Factorial designs provide a clearer picture than testing one factor at a time. If you only tested baking temperature while keeping the type of sugar constant, you might miss how sugar type influences the results. By considering multiple factors at once, researchers can detect interaction effects, which occur when the impact of one factor depends on another.

In real-world scenarios, individual factors don’t always work independently. Multiple factors frequently interact and influence outcomes simultaneously. One factor can affect how other factors behave. For example, in a product testing experiment, a company might want to see how different packaging materials affect freshness, but the material’s effect could depend on storage temperature. A factorial design helps reveal these interactions, ensuring researchers don’t overlook essential patterns.

Experimenters frequently use this design in scientific research, medicine, psychology, and business. Whether testing a new drug formulation, evaluating how lighting and screen brightness affect reading speed, or optimizing the texture of a food product, factorial designs help researchers make better-informed decisions.

Learn more about Experimental Designs: Definition and Types.

The Limits of Testing One Factor at a Time

A common but flawed approach is the one-factor-at-a-time (OFAT) method, where researchers change and test one variable at a time while keeping everything else fixed. At some point in school, you probably learned that it is a promising experimental approach.

This method feels intuitive, and researchers commonly use it in business and industry. It seems straightforward—until you need to study multiple factors at once. Testing each factor separately is not only slow and expensive, but it also has significant drawbacks relative to a factorial design:

• It requires too many tests to study multiple factors.
• It misses critical interactions between factors.

Fewer Experimental Runs

Instead of testing one factor at a time, a factorial design allows researchers to change multiple factors simultaneously, following a structured plan. This approach reduces the number of experimental runs needed while providing deeper insights into how factors influence each other.

For example, if you have three factors, each with two levels, a one-at-a-time approach requires 16 runs while a factorial design only needs 8. The relative efficiency of factorial designs increases from there as you add more factors. Imagine you’re improving a product and have identified 5 possible factors. Assessing all 5 factors in a single factorial design is much quicker than performing five separate experiments that require many more total runs.

Interaction Effects

Factorial designs assess potential interactions. Many real-world processes involve interactions between factors. Some factors work together, making their combined effect stronger than expected. Others work against each other, canceling out or reducing their individual effects. We can’t fully understand how each factor impacts the outcome without studying these interactions.

Ignoring interactions can lead to misleading conclusions. A treatment might seem ineffective if its effect depends on another condition. A marketing strategy might fail in one condition but thrive in another. By detecting interactions, factorial designs help researchers avoid oversimplifications and make better decisions.

Learn more about Understanding Interaction Effects.

Performing a Factorial Design Experiment

When conducting a factorial experiment, a researcher follows a structured plan. Each experimental run involves setting the tested factors to specific values, collecting results, and then moving on to the next combination. For quantitative factors, each variable typically has a high and low setting, such as a low and high temperature. For qualitative factors, there are usually two categories, such as using two different materials.

The key to a well-run factorial experiment is that the experimenters randomize the sequence of runs to prevent order effects or hidden biases from influencing the results. Instead of testing all low settings first and all high settings later, the researcher follows a randomized list that mixes conditions, ensuring that external factors, known as confounders, don’t skew the findings. This randomized approach helps maintain the experiment’s integrity while efficiently gathering data on how multiple factors interact.

The partial worksheet below contains the experimental runs for a factorial design with four factors that each have two levels. It has one qualitative variable: Material – testing two types. The experiment also includes three quantitative variables: Pressure, Temperature, and Cooling Temperature. Researchers picked a high and low value for each. The outcome variable is the measured strength of the product.

The worksheet displays a run order that indicates the settings for each experimental run. Note that the researchers randomized the run order so each run has a random combination of settings, but the experiment as a whole will cover all possible combinations (not shown). A full two-level factorial design with four factors, like this one, requires 16 runs.

Factorial Designs Are Generally Used for Shorter Runs

Factorial designs are powerful tools for studying multiple factors simultaneously, but researchers use them most frequently for experiments where they can complete each run relatively quickly. This experimental design requires setting multiple test conditions for each run and observing the outcome. The study can become impractically slow and expensive if each experimental run takes a long time.

Factorial designs work best when researchers can precisely control all factors throughout the experiment. They are much more manageable for short-term experiments, where conditions can be kept stable. Hence, experimenters are more likely to use a factorial design in laboratory testing, manufacturing optimization, and other controlled experiments where researchers can rapidly test different combinations. Each trial might take only minutes or hours.

For example, in a laboratory study testing how temperature and humidity affect battery performance, researchers can set exact conditions for each run.

Additional Strengths and Considerations of Factorial Designs

Factorial designs offer an efficient way to study how multiple factors influence an outcome. Unlike traditional experiments that test one factor at a time, factorial designs allow researchers to gather more insights with fewer experimental runs. However, this efficiency comes with some considerations.

Efficiency Through Linearity Assumption

2-level factorial designs assume that each factor’s effect on the response is linear—meaning that a straight-line relationship exists between input and output changes. Because it requires only two points to define a line, a 2-level factorial design can estimate these effects using just two levels per factor. This assumption makes the experiment cost-effective and faster compared to testing more levels for every factor.

However, if the relationship is curved, the design requires additional experimental runs to capture that complexity. Researchers can extend an initial factorial design to test for curvature if needed.

Independent Estimation of Effects Through Orthogonality

A key strength of factorial designs is their orthogonality, meaning that the design evaluates factors in a balanced way across all levels of other factors. This process ensures that the analysis estimates each factor’s effect independently, without being distorted by other variables.

In contrast, when analyzing observational data, factor changes often occur together in uncontrolled ways. For example, a company might increase marketing spending while changing its product pricing. If these shifts happen together, it’s difficult to separate which factor truly caused a change in sales. Factorial designs prevent this issue by structuring the experiment so that effects are isolated and unbiased.

Wider Applicability of Results

Another advantage of factorial designs is that they estimate each factor’s effect across a broader range of conditions. Because multiple factors are studied together, the results are applicable to a wider set of real-world scenarios rather than being tied to a specific, narrow set of conditions.

For example, if a researcher studies how temperature and pressure affect a chemical reaction, a factorial design assesses temperature’s effect across a range of pressure levels rather than at only a single fixed pressure. This design makes the findings more generalizable and valuable than one factor at a time.

Types of Factorial Designs

Factorial designs come in different forms, depending on the number of factors, the number of levels within each factor, and whether every combination is tested. The right design depends on the experiment’s goals and practical limitations.

General Factorial Design

A general factorial design tests every possible combination of factors and their levels. Factors can have as many levels as researchers want to evaluate. This approach provides the most complete picture but can become complex as the number of factors and levels increase. Factors do not need to have the same number of levels.

Example: A beverage company wants to test different formulations of a new sports drink using:

• Sweetener type (sugar, artificial, or stevia) → 3 levels
• Flavor (lemon, berry) → 2 levels

A full factorial design would test all six combinations:

1. Sugar + Lemon
2. Sugar + Berry
3. Artificial + Lemon
4. Artificial + Berry
5. Stevia + Lemon
6. Stevia + Berry

Because the design assesses all possible combinations, researchers can determine the following:

• Whether one sweetener type is preferred overall (main effect of sweetener).
• Whether one flavor is preferred overall (main effect of flavor).
• Whether the best sweetener depends on the flavor (interaction effect).

General factorial designs are great when identifying interactions is critical, but they become impractical if too many factors or levels are involved. For example, a 3-factor design with 4 levels per factor would require 64 combinations (4 × 4 × 4), which may be too expensive or time-consuming to test fully.

2-Level Factorial Design (The Most Common Type)

A 2-level factorial design is by far the most widely used type because they’re simpler versions of the general designs mentioned above, making them more manageable. It’s an efficient design that researchers use to screen factors and determine which ones are most important. For quantitative variables, experimenters strategically select two values representing meaningful extremes or practical choices. This approach makes the experiment easier to run while still capturing key effects.

The number of runs for a 2-level factorial design is 2^k, where k is the number of factors. For example, a design with 3 factors involves 2³ = 8 runs.

Example: A researcher tests how temperature, lighting, and background noise affect productivity in an office setting.

• Temperature: Cool (68°F) vs. Warm (75°F)
• Lighting: Dim vs. Bright
• Background Noise: Quiet vs. Moderate Noise

Researchers chose these levels because they represent realistic office conditions. The researcher could have tested additional temperatures or noise levels, but selecting two for each factor makes the experiment more manageable.

Cube plots commonly represent factorial designs with up to three factors. The dimensions of length, width, and height each represent a factor with two possible values. Below, you can see the eight total combinations. Each dot represents a measurement. For example, the bottom-right dot in the front of the cube depicts an experimental run with temperature at 75, noise at quiet, and lighting at dim.

A full 2-level factorial design is one where you take measurements at all possible combinations. You can see that in the cube plot because there is a measurement dot for all possible factor combinations.

Because this setup is still relatively small, it’s often used in screening experiments, where researchers identify which factors have the most impact before testing more complex designs. After the analysis identifies key factors, later studies might expand to test additional levels.

Fractional Factorial Design (Testing a Subset)

Unlike the previous examples, a fractional factorial design tests only a subset of all possible combinations. This approach is beneficial when testing every combination isn’t practical due to time, cost, or complexity. For example, a 2-level factorial design that assesses 8 factors requires 2⁸ = 256 runs. However, using a half-fractional design reduces the number of runs to only 128.

Example: A company testing three factors—ad format (image vs. video), ad placement (homepage vs. search results), and call-to-action (discount vs. free trial)—would have 8 combinations in a full factorial design. If budget constraints make full testing impossible, they might test only 4 carefully selected combinations to still capture key insights.

In the cube plot, you can see all the combinations, but notice that dots indicating measurements occur for only 4 of the 8 possible combinations.

This method balances efficiency with meaningful results but risks missing subtle interactions.

Choosing the Right Factorial Design

• Use general factorial designs when interactions are critical and the number of factors is manageable.
• Use 2-level factorial designs when exploring new factors efficiently.
• Use fractional factorial designs when the number of combinations is too large. These designs can help determine which factors to eliminate from consideration.

Extending Factorial Designs for Additional Insights

One of the biggest advantages of factorial designs is that they can be extended and refined without starting over using sequential experimentation. Instead of running entirely new experiments, researchers can build on the data they’ve already collected, incorporating the original factorial design into a more complex design as needed.

This structured, sequential experimentation approach saves both time and resources while allowing for deeper insights. Some common ways to extend an initial factorial design include:

• Folding over a fractional factorial design – If an initial experiment used a fractional factorial design (which tests only a subset of factor combinations), researchers can add more runs to increase the resolution and detect smaller effects. Instead of conducting an entire experiment from scratch, this method fills in missing data while preserving the original structure.
• Adding axial points to create a response surface design – A 2-level factorial design assumes that relationships are linear. But what if the real relationship is curved? By adding additional test points beyond the original factor levels, researchers can estimate curvature and fine-tune optimization without repeating unnecessary trials.
• Using a second factorial design in a refined space – A follow-up experiment can zoom in on the most promising conditions after an initial factorial experiment identifies key factors. Instead of testing the entire design space again, this approach focuses only on the high-potential regions, improving precision and efficiency.

By extending an existing factorial design instead of starting from scratch, researchers maximize the value of their initial data while minimizing additional cost and effort. This structured, step-by-step approach allows for efficient learning and continuous improvement, making factorial designs a powerful tool for experimentation.

In a future post, I’ll show you how to analyze and interpret a factorial design!