Categorical factor
The independent variable answers, “Which group?” and has two or more distinct, non-overlapping levels.
ANCOVA · Lesson 05.02 · Pages 569–572
ANCOVA Pre-Flight
Before describing variables or running a test, confirm that the design truly fits ANCOVA. A fair adjusted comparison begins with one categorical grouping factor, one continuous outcome, and a relevant continuous covariate.
By the end, you can
- Confirm whether a research design fits ANCOVA.
- Separate the roles of the factor, outcome, and covariate.
- Check the participant-level data structure.
- Recognize covariate mistakes before analysis.
- Write an aligned ANCOVA research question.
Quick fit check
Make sure the variables tell an ANCOVA story
No statistical adjustment can rescue a design whose variables do not align. Use these three checks before writing the problem, purpose, or research question.
Continuous outcome
The dependent variable answers, “How much?” and is measured on an interval or ratio scale.
Continuous covariate
The control variable represents a background difference that could influence the outcome but is not the main focus.
Boundary check
Two categorical grouping variables point toward a Two-Way ANOVA. A categorical covariate also requires a different model. A bounded rate or proportion may need a generalized model rather than standard ANCOVA.
Clarify each role
Factor, outcome, and covariate do different jobs
The classroom example keeps the design concrete: compare Lecture and Flipped instruction on final exam scores while controlling for pretest scores.
Swipe horizontally to view the full table.
| Variable | Role | Measurement | Classroom example | Key point |
|---|---|---|---|---|
| Independent variable | Grouping factor | Nominal, 2+ groups | Teaching method: Lecture or Flipped | Groups are distinct and non-overlapping. |
| Dependent variable | Outcome | Interval or ratio | Final exam score | The outcome must be continuous. |
| Covariate | Continuous control | Interval or ratio | Pretest score | It should relate to the outcome and not be caused by the factor. |
The classroom story
Compare like with like
Flipped students may finish with higher scores, but they may also have started with stronger pretest performance. ANCOVA asks whether the teaching-method difference remains after accounting for that starting advantage.
Teaching method
Each student belongs to either Lecture or Flipped instruction.
Pretest score
Every student has a continuous baseline score measured before instruction.
Final exam score
Every student has one continuous outcome value after instruction.
Check the participant-level data
- One row per participant.
- One categorical grouping variable.
- One continuous dependent variable.
- One continuous covariate measured before or independently of the factor.
- Each participant belongs to only one group.
- Each participant has one outcome and one covariate value.
Missing covariate data
Plan for missing baseline scores before analysis. The source recommends considering multiple imputation rather than automatically deleting every incomplete case.
Covariate check
Use three rules before calling a variable a control
A covariate should improve the adjusted comparison, not merely add another column to the model.
Related to the outcome
If the covariate does not relate to the dependent variable, there is little to adjust for.
Not caused by the factor
A variable changed by the intervention cannot represent an independent starting difference.
Similar relationship across groups
The covariate-outcome slope should be reasonably similar for each group.
Picture the slopes before adjustment
Reasonably parallel
The covariate relates to the outcome similarly across groups, supporting a common adjustment.
Different or crossing slopes
The relationship changes by group, so a single adjusted comparison may not be appropriate.
Swipe horizontally to view the full table.
| Common mistake | Why it is a problem | Better next step |
|---|---|---|
| Categorical covariate | It does not fit the continuous-covariate role. | Use a factorial design such as Two-Way ANOVA when appropriate. |
| Covariate changed by the factor | It no longer represents a prior or independent difference. | Drop it or rethink the causal role. |
| Covariate unrelated to the outcome | It adds complexity without improving adjustment. | Check the relationship before modeling. |
| Several overlapping covariates | Redundancy can weaken clarity and precision. | Begin with one meaningful covariate and add others deliberately. |
Put the setup into words
Write one aligned ANCOVA question
A clear research question names the covariate, the grouping factor, and the continuous outcome in one sentence.
Pre-flight complete
The design is ready when the group factor is categorical, the outcome and covariate are continuous, the covariate represents a meaningful prior difference, and the data contain one complete participant-level record per case.
Source boundary: printed workbook pages 569–572. The next lesson begins with “Structured Examples — Writing the ANCOVA Story.”