Factorial experiments study two or more factors at the same time, so researchers can estimate both individual factor effects and combined effects. This cheat sheet helps students organize treatment combinations, interpret interaction effects, and connect experimental designs to ANOVA models. It is especially useful when reading research studies, planning experiments, or analyzing data from designed studies.
Factorial designs are efficient because one experiment can answer several questions at once.
The core ideas are factors, levels, main effects, interactions, and error variation. A main effect compares the average response across levels of one factor, while an interaction asks whether that effect changes across levels of another factor. In a two-factor ANOVA, the model is often written as .
Evidence for effects is usually judged with and supported by interaction plots or simple effects analysis.
Key Facts
- A factorial design with factor having levels and factor having levels has treatment combinations.
- With replicates per treatment combination, the total sample size in a balanced two-factor design is .
- A common two-factor fixed-effects model is , where represents random error.
- The main effect of factor compares marginal means, such as .
- An interaction effect exists when the difference between levels of one factor changes across levels of another factor, meaning for at least one combination.
- The ANOVA test statistic for any effect is , where estimates within-cell error variance.
- For a balanced two-factor ANOVA, the degrees of freedom are , , , and .
- In a factorial design, factors each have levels, producing treatment combinations before replication.
Vocabulary
- Factor
- A factor is an explanatory variable that the experimenter studies, such as temperature, fertilizer type, or teaching method.
- Level
- A level is a specific value or category of a factor, such as low and high temperature.
- Treatment Combination
- A treatment combination is one specific pairing or grouping of factor levels used in the experiment.
- Main Effect
- A main effect is the average effect of one factor on the response after averaging over the levels of other factors.
- Interaction Effect
- An interaction effect occurs when the effect of one factor on the response depends on the level of another factor.
- Balanced Design
- A balanced design has the same number of observations in every treatment combination.
Common Mistakes to Avoid
- Ignoring a significant interaction, then interpreting main effects alone. This is wrong because an interaction means the effect of one factor changes across levels of another factor.
- Using cell means when marginal means are required. Main effects compare averages over the other factor, so using only one cell can give a misleading conclusion.
- Forgetting replication within treatment combinations. Without replication, the experiment may not provide a reliable estimate of error variance .
- Treating nonparallel lines in an interaction plot as automatic proof of significance. Apparent nonparallelism suggests interaction, but the ANOVA test using determines statistical evidence.
- Confusing factors with levels. A factor is the variable being studied, while levels are the specific settings or categories of that variable.
Practice Questions
- 1 A study has factor with levels, factor with levels, and replicates per cell. Find the number of treatment combinations and the total sample size .
- 2 In a balanced two-factor ANOVA with , , and , compute , , , and .
- 3 An ANOVA table gives and . Compute the interaction test statistic .
- 4 An interaction plot shows that the lines for two levels of factor cross as factor changes. Explain what this suggests about interpreting the main effects.
Understanding Factorial Experiments and Interaction Effects
A factorial study is built from treatment cells. Each cell is one specific combination of factor levels. For example, a greenhouse study might use two fertilizer types at three watering levels.
Plants in every cell should be treated alike except for the assigned fertilizer and water amount. Random assignment is important because it spreads hidden differences, such as soil quality or seed condition, across the cells. Replication means using several plants in each cell.
Without replication, natural plant to plant variation can be mistaken for a treatment pattern. A balanced layout, with the same number of observations in every cell, makes the analysis easier to interpret.
Interaction effects are often the most scientifically useful result. Suppose one fertilizer improves growth under low water but has little benefit under high water. The best fertilizer then depends on watering level.
An interaction plot can show this pattern clearly. Put one factor along the horizontal axis and draw a separate line for each level of the other factor. Parallel lines suggest that the effect is fairly consistent.
Lines that separate, meet, or cross suggest an interaction. Crossing lines are memorable, but they are not required. Even nonparallel lines can show that one factor changes the size of the other factor's effect.
When a meaningful interaction is present, broad averages can hide the real story. An overall average for fertilizer may combine results from low water and high water even though the fertilizer behaves differently in those settings. This is why students should examine simple effects.
A simple effect compares fertilizer types at one particular watering level, or compares watering levels for one fertilizer type. Researchers often follow an interaction test with these focused comparisons. They should use a planned method for multiple comparisons when many cell means are checked.
A common interpretation rule is to discuss the interaction first, then describe the relevant simple effects. Main effects may still be reported, but they need careful wording.
ANOVA separates variation due to the studied factors from variation among observations receiving the same treatment. The error term reflects the ordinary spread within a treatment cell. Large between cell differences are more convincing when within cell variation is small.
This reasoning depends on good design before any calculation begins. Observations should be independent, groups should have roughly similar spread, and the response values within cells should not be extremely distorted by outliers.
In real settings, factors may be teaching method and class schedule, drug dose and exercise plan, or website layout and device type. When reading a study, check which combinations were actually tested, whether enough replicates were used, and whether a graph supports the written conclusion.