Repeated measures vs mixed ANOVA is a study-design question before it is an SPSS menu choice. Are you measuring the same nurses or patients repeatedly? Are you also comparing separate groups? Those two questions determine the factors your analysis must represent.
In common teaching usage, a one-factor repeated measures ANOVA examines a continuous outcome measured across occasions in one set of participants. A mixed ANOVA includes a within-subject factor, such as time, and a between-subject factor, such as intervention group. A mixed ANOVA therefore also contains repeated measurements. The terms describe related designs rather than entirely separate statistical families.
This guide focuses on that distinction and the group-by-time interaction in nursing intervention studies. For an overview of other options, use the statistical test selection guide. For general data preparation, see the SPSS nursing analysis workflow.
Repeated measures vs mixed ANOVA: the design difference
Imagine a nursing education project measuring confidence before training, immediately afterwards and six weeks later. If every participant receives the same training, time is the within-subject factor. The analysis recognises that the three scores from one nurse are related.
Now imagine an intervention group and a separate comparison group, each measured at those same three occasions. Time remains within subjects, while group is between subjects. The mixed design allows the analysis to ask about time, group and their interaction.
| Question | One-factor repeated measures | Mixed ANOVA |
|---|---|---|
| Same people measured repeatedly? | Yes | Yes |
| Separate participant groups included? | No between-subject group factor | Yes |
| Typical focus | Differences across occasions | Whether change differs by group |
| Example | Confidence at three times in one cohort | Confidence at three times in two cohorts |
IBM’s GLM Repeated Measures procedure can represent within-subject and between-subject factors (IBM, n.d.-a). Consequently, using that procedure does not by itself tell the reader which research design you analysed. State the factors and their levels explicitly.
Do not confuse a mixed ANOVA with mixed-methods research. Mixed methods combines qualitative and quantitative strands. Here, “mixed” refers to the combination of within-subject and between-subject factors. It also does not mean that every feature of a linear mixed-effects model is available in a conventional mixed ANOVA.
Why the group-by-time interaction matters
For an intervention comparison, the central question is often whether the groups changed differently. An overall time effect can occur when both groups improve. An overall group effect can occur when one group starts higher and remains higher. Neither result alone demonstrates a different pattern of change.
The interaction addresses the relationship between group and time. A useful first step is to draw the means at each occasion for each group. The plot helps you understand the pattern, but the visual impression must be considered alongside the appropriate model estimates and uncertainty.
Consider these fictional group means for a confidence scale. The values illustrate the design; they are not findings from a real study.
| Group | Baseline | Immediately after | Six weeks |
|---|---|---|---|
| Training | 40 | 52 | 50 |
| Comparison | 40 | 44 | 43 |
The training group shows a larger descriptive increase. However, these means do not provide sample sizes, variability, within-person relationships or confidence intervals. You cannot calculate a defensible p-value or infer effectiveness from the table alone.
A frequent error is to test change separately within each group and conclude that the intervention worked because one p-value is below .05 and the other is above .05. The question requires a direct comparison of the changes. Different significance labels are not themselves evidence that the changes differ.
Write the analysis question before opening SPSS
Specify the outcome, groups, measurement occasions and primary comparison in ordinary language. For example: “Does the change in confidence from baseline to six weeks differ between the education and comparison groups?” This sentence is more useful than starting with “I will run an ANOVA”.
Then identify whether the primary question concerns a particular contrast or a broader pattern across occasions. A baseline-to-follow-up difference and an omnibus interaction across three time points are related but not identical questions. Your analysis plan should explain which comparison answers the main objective.
Record whether participants were randomised, how groups were formed and whether data are clustered within wards. A mixed ANOVA does not create randomisation or remove the consequences of a weak comparison design. If all intervention participants come from one ward and all controls from another, group and ward may be inseparable.
Also check whether the outcome is suitable for this model. A continuous scale score is a different analytical problem from a yes/no event, a count of incidents or time to discharge. Repeated measurement alone does not make ANOVA appropriate for every outcome.
Prepare the repeated observations in SPSS
For the conventional GLM Repeated Measures workflow, data are commonly arranged in wide format: one row per participant, a group variable and separate columns for each occasion. Keep identifiers consistent and verify that the same instrument and scoring direction were used each time.
A simple layout might contain ID, group, confidence_baseline, confidence_post and confidence_followup. These are suggested variable names, not required software syntax. The column sequence matters because the within-subject levels must be matched to the intended occasions.
Before analysis, inspect valid ranges and missing codes. A code such as 99 must not enter the model as a real confidence score. Check duplicate identifiers, group labels and whether the intended baseline occurred before the intervention. Correct documented data errors while preserving a record of changes.
Use the site’s missing-data guide for detailed handling decisions. For this article, the key issue is how incomplete measurement occasions affect the participants included in the conventional analysis. Report the analysed sample rather than assuming it equals everyone recruited.
Set up the model deliberately
IBM documents the repeated-measures procedure through within-subject factor definitions and the model specification (IBM, n.d.-b; n.d.-c). In commonly used desktop versions, the route is Analyse, General Linear Model, Repeated Measures. Labels and available options can vary by version and licence.
- Define the within-subject factor, such as time, and the correct number of levels.
- Assign the repeated outcome columns in their intended chronological order.
- For the mixed design, place the grouping variable in the between-subject factors field.
- Inspect the model so the effects match the analysis question.
- Request appropriate descriptive information, estimates and plots.
- Save the analysis syntax and record relevant options for reproducibility.
Do not add variables simply because they are available. A covariate should have a defensible role in the planned analysis. Adjusting for a measure taken after intervention can change the question and introduce interpretive problems. Discuss design-specific adjustment with a suitable statistical supervisor.
Before running the final model, review the factor labels with someone who understands the study. A technically valid model with reversed time columns can answer the wrong question while producing plausible output. This simple mapping check is often more valuable than requesting every available table.
Assumptions and sphericity need careful interpretation
Assumptions concern the model and study design, not just one preliminary test. Consider independence between participants, the distribution of model errors, influential observations and relevant variance-covariance requirements. Observations within one participant are intentionally related; treating that relationship as a data error misunderstands the design.
For within-subject effects involving more than two levels, sphericity can matter in a univariate repeated-measures analysis. It concerns the variances of differences between measurement levels. Where an appropriate correction is used, report the correction and the degrees of freedom associated with the result.
With only two levels of a within-subject factor, that factor does not present the same sphericity issue. Avoid copying a three-occasion reporting sentence into a two-occasion analysis without checking its relevance. Equally, a favourable sphericity test does not establish that every other model requirement is satisfied.
Inspect unusual values before removing them. A very low score may reflect a genuine participant experience. Exclusion rules should be justified and documented, with sensitivity analysis where appropriate. Removing cases because the p-value becomes more attractive undermines the interpretation.
When another approach may fit better
A conventional repeated-measures analysis may be restrictive when follow-up times vary, observations are incomplete or participants are clustered. A linear mixed-effects model can offer additional flexibility, but it requires explicit choices about fixed effects, random effects and covariance structure. It is not an automatic remedy for all missing data or design limitations.
For a randomised study with one follow-up, a baseline-adjusted comparison may be a relevant alternative to a change-score approach. The choice should follow the estimand, design and analysis plan. Do not switch methods after viewing results merely to obtain significance.
For ordinal, binary or count outcomes, consider methods suited to those measurement structures. A total scale score may sometimes support a continuous-outcome analysis, but one individual Likert item should not be treated as equivalent without justification.
The practical lesson is to seek statistical input when the study departs from the simple design. Bring the research question, measurement schedule, dataset structure and missingness summary. Those details are more useful than a screenshot containing an unexplained significant result.
Report repeated measures vs mixed ANOVA accurately
A clear results account identifies the design before interpreting an effect. State the outcome, sample, group levels, occasions and analysis method. Provide descriptive results so the reader can understand the direction and size of the observed pattern.
Use a reporting scaffold rather than inventing numbers: “A mixed ANOVA examined [outcome] across [occasions] in [groups]. The group-by-time interaction was [result], F([degrees of freedom]) = [verified statistic], p = [verified value], with [appropriate effect size]. [Planned follow-up comparison and uncertainty] indicated [restrained interpretation].” Replace every bracket from your actual analysis.
If a correction was necessary, identify it consistently in the text and table. Explain which comparisons were planned, which were exploratory and how multiplicity was addressed where relevant. An effect size should be named correctly rather than described as a generic percentage improvement.
Keep statistical and practical importance separate. A detectable difference on a confidence scale does not automatically establish better patient outcomes. Discuss the outcome’s meaning, uncertainty, design limitations and follow-up duration before claiming value for nursing practice.
Common errors and their corrections
- Error: using an independent-groups ANOVA for three scores from each nurse. Correction: account for the repeated structure.
- Error: interpreting the time effect as proof of intervention benefit. Correction: examine the comparison relevant to differential change.
- Error: comparing two within-group significance labels. Correction: test the difference directly.
- Error: copying the wrong corrected row. Correction: match assumptions, correction, statistic and degrees of freedom.
- Error: reporting recruitment numbers as analysis numbers. Correction: verify the actual included cases.
For general table interpretation, consult the nursing SPSS output guide. Use this article for the particular relationship between repeated observations and group comparisons. Keeping those purposes separate makes the analysis easier to plan and the dissertation easier to navigate.
Check the design against the data dictionary
Before finalising your methods, create a short mapping sheet. Write the research term in one column and the actual dataset variable in another. “Time” should map to identifiable occasions, while “group” should map to an unambiguous coding rule. Confirm that labels such as intervention and control match the allocation records rather than an assumption made during cleaning.
Check that participant IDs represent people rather than visits. If each row is a visit, converting the file to a wide arrangement requires careful matching, not simply copying consecutive values into new columns. A mismatch can pair one person’s baseline with someone else’s follow-up. Review several records against the authorised source before analysis.
Finally, preserve the original file and save a separate analysis version. Record recoding, exclusions and any restructuring. This documentation lets a supervisor understand how the model sample was constructed and helps you reproduce tables after corrections. It also makes the methods paragraph more precise: instead of saying that data were “prepared appropriately”, you can state the relevant checks and explain their consequences for the analysis.
Frequently asked questions
Is mixed ANOVA a type of repeated-measures analysis?
Yes. It includes repeated observations and at least one between-subject factor. State the factors explicitly because terminology varies across teaching materials.
Can I use mixed ANOVA for two groups measured twice?
It can represent that design when the outcome and assumptions fit. The analysis plan should also consider the study’s primary comparison and whether an alternative baseline-adjusted model is more appropriate.
Does a significant time effect prove the intervention worked?
No. Both groups might change similarly. A causal claim also depends on allocation, confounding, measurement and other design features.
Do I need a control group for repeated measures ANOVA?
A one-factor within-subject analysis does not require a control group. However, change in a single cohort provides limited evidence about what caused that change.
Is mixed ANOVA the same as a mixed-effects model?
No. The names overlap, but mixed-effects modelling offers different structural choices. Specify the actual procedure and model instead of using the labels interchangeably.
What should I send for statistical feedback?
Provide the research question, design, data dictionary, measurement schedule, missingness summary and saved syntax. Remove identifying data before sharing through an authorised route.
Plan the follow-up interpretation before writing
An omnibus interaction can indicate that the pattern across occasions differs without identifying the specific comparison that matters clinically. Return to the planned question before selecting follow-up tests. If the objective concerned sustained improvement at six weeks, do not make the immediate post-training result the main conclusion simply because it appears more striking.
Prepare a small interpretation table for yourself. Include the effect being tested, the relevant output, the direction of the descriptive pattern and the limitation that matters most. A time effect belongs in a different row from a group effect or interaction. This prevents an attractive descriptive result from being attached to an unrelated p-value.
When findings are uncertain, say what the interval and design allow you to conclude. An inconclusive result does not demonstrate equivalence between groups. Equivalence and non-inferiority questions require their own planning and methods. Similarly, a statistically detectable interaction does not tell you that every time-specific comparison is important or that the improvement will persist beyond the observed follow-up.
Finally, discuss the analysis with the methods chapter open. The results should answer the objectives promised there. If the final analysis differs from the planned approach, explain the reason transparently and distinguish a justified correction from exploratory work. Consistency across question, model and interpretation is more valuable than presenting a large collection of disconnected output tables.
Final analysis check
Choose the model from the nursing question and data structure. Then verify the factor mapping, assumptions, included sample and interpretation. If you need project-specific guidance, the data-analysis support page explains how to prepare a focused request.
Request guidance on your repeated-measures design
Statistical feedback is most useful when it starts with the question and data structure. Prepare an anonymised data dictionary, measurement schedule and analysis plan before requesting help. The reviewer should explain the model and output so you can defend the final interpretation yourself. Do not share identifiable clinical records or ask anyone to invent missing observations.
Request a review of your nursing analysis plan and check the support process before sending material. Our approach to academic guidance, client feedback and nursing dissertation support overview help you assess whether that guidance matches your project.
References
- IBM (n.d.-a) GLM Repeated Measures. Available at: IBM SPSS Statistics documentation (Accessed: 16 October 2026).
- IBM (n.d.-b) GLM Repeated Measures Model. Available at: Model specification documentation (Accessed: 16 October 2026).
- IBM (n.d.-c) Overview (GLM: Repeated Measures command). Available at: GLM command documentation (Accessed: 16 October 2026).
