How to test normality in SPSS for nursing research affects test selection, model assumptions and reporting. However, one p-value cannot establish normality on its own. Instead, examine the study design and sample size. Then combine graphs, descriptive measures and the assumptions of the planned analysis to reach a defensible decision.
This guide shows nursing students how to examine a continuous variable using histograms, normal Q–Q plots, boxplots, skewness, kurtosis and the Shapiro–Wilk test. It also explains what to do when evidence is mixed and how to report the decision without claiming that data are “perfectly normal”.
What Does Normality Mean in Nursing Research?
Normality describes how closely a distribution follows a symmetrical, bell-shaped normal distribution. The issue often arises before a t-test, analysis of variance, Pearson correlation or linear regression. However, the exact assumption depends on the analysis.
An independent-samples t-test concerns the outcome distribution within each group. A paired t-test concerns the distribution of difference scores. Linear regression focuses on residuals rather than requiring every raw variable to be normally distributed. Therefore, do not run one check on the whole dataset and assume every planned test is covered.
Normality is relevant only to suitable quantitative variables. It is not meaningful to test whether nominal categories such as ward type are normal. A single Likert item is ordinal and should not be treated as a continuous normal variable merely because it is coded 1 to 5.
When Should You Test Normality in SPSS for Nursing Research?
Assess normality after verifying data entry, missing-value codes and impossible values, but before interpreting a method that relies on distributional assumptions. Begin with the nursing SPSS data-cleaning guide if screening is incomplete.
A normality assessment is commonly relevant for a continuous clinical outcome compared between independent groups, a before-and-after outcome, mean differences across several groups, Pearson correlation, linear regression residuals, or justified multi-item scale totals.
Do not test every numeric column automatically. Participant IDs, category codes and dates may look numeric in SPSS but are not continuous outcomes. Match each check to a research question and planned analysis.
Prepare the Dataset Before Checking Normality
Confirm variable type and coding
In Variable View, confirm that the outcome is numeric and its measurement level is appropriate. Check value labels and user-defined missing values. A code such as 99 for “not recorded” can create extreme skewness if SPSS treats it as a genuine score.
Run frequencies and descriptives
Check minimum and maximum values against the questionnaire, measurement protocol or codebook. Investigate unusual observations against original records instead of deleting them because they inconvenience the analysis.
Define groups correctly
If comparing intervention and control groups, examine the outcome within each group. A combined distribution can hide group-specific problems or appear non-normal merely because the groups have different means.
Missing values also change the number of cases used. The guide to missing data in nursing research using SPSS explains why automatic deletion or mean substitution is rarely defensible.
How to Test Normality in SPSS for Nursing Research Using Explore
SPSS Explore brings useful numerical and graphical checks together. IBM documents Explore as a procedure producing summary statistics and graphical displays for distributions and groups.
- Select Analyze.
- Choose Descriptive Statistics, then Explore.
- Move the continuous outcome into Dependent List.
- If comparing groups, move the group variable into Factor List.
- Select Plots.
- Tick Histogram and Normality plots with tests.
- Keep boxplots enabled where unusual observations require examination.
- Select Continue, then OK.
The output includes descriptives, tests of normality, histograms, Q–Q plots and boxplots. Interpret these as connected evidence rather than competing verdicts.
Interpret the Histogram
Look for overall symmetry, a central peak, long tails, gaps, floor or ceiling effects, and multiple clusters. A roughly bell-shaped pattern supports approximate normality. However, appearance changes with bin width and sample size. Small samples often look irregular, while large samples reveal subtle deviations.
Do not write that a histogram “proves” normality. State that it showed an approximately symmetrical distribution without marked tailing or clustering. If a strong floor effect appears, describe it and consider the clinical and measurement context.
Interpret the Normal Q–Q Plot
A Q–Q plot compares observed quantiles with those expected under a normal distribution. Points reasonably close to the diagonal support approximate normality. Systematic curves, one-sided tail departures or distant points require attention.
- An S-shaped pattern may indicate skewness or unusual tail weight.
- Departure at one end can indicate a long tail.
- A few distant points may be influential observations.
- Minor deviations do not automatically require a non-parametric test.
Explain the observed pattern rather than reporting only “normal” or “not normal”.
Use the Boxplot to Investigate Outliers
SPSS boxplots flag potential outliers according to graphical rules; they do not establish that values are errors. For each unusual observation, check whether it was entered correctly, is within the permitted range, represents a genuine clinical case and materially influences the analysis.
Deleting a genuine value solely to obtain a non-significant normality test can bias a dissertation. Retain valid observations unless a defensible analysis rule supports another approach. A sensitivity analysis can show whether conclusions depend on them.
Interpret the Shapiro–Wilk Test
The Shapiro–Wilk test evaluates whether values are compatible with a normal distribution. If p exceeds the chosen alpha, evidence is insufficient to reject normality. If p is at or below alpha, the test indicates departure from normality.
A non-significant result does not prove normality. Small samples may lack power to detect important departures. Large samples may produce significance for a slight deviation with little practical effect on a robust method.
Never decide from Shapiro–Wilk alone. Mishra and colleagues (2019) recommend graphical and numerical evidence together, while Ghasemi and Zahediasl (2012) explain why sample size affects interpretation.
Assess Skewness and Kurtosis
Skewness describes asymmetry; kurtosis describes aspects of tail weight and concentration. SPSS reports both values and their standard errors. Although textbooks propose fixed acceptable ranges, no universal cut-off fits every design and sample.
Use these measures to support the graphs and the planned model. A deviation tolerable in a large balanced comparison may matter in a small group with influential values. Avoid claiming that one threshold guarantees valid inference.
How Sample Size Changes the Decision
With a small sample, individual observations can strongly affect the mean, standard deviation and test statistic. Visual evidence therefore deserves close attention. With a large sample, Shapiro–Wilk may detect a small departure even when a common parametric method remains useful.
Large sample size does not repair biased sampling, incorrect measurement, severe outliers or dependence. For comparisons, inspect each group. A total sample of 200 is not reassuring if one group contains only 12 participants and is strongly skewed.
Normality Requirements by Analysis
| Analysis | What to examine | Qualification |
|---|---|---|
| Independent t-test | Outcome within each group | Also assess independence, influential values and variance assumptions |
| Paired t-test | Paired difference scores | Do not rely on separate checks of pre- and post-scores |
| One-way ANOVA | Outcome or residuals within groups | Unequal groups and unequal variances require care |
| Pearson correlation | Joint pattern, linearity and influential points | Separate univariate checks do not establish bivariate normality |
| Linear regression | Residuals and diagnostics | Raw predictors need not all be normally distributed |
| Repeated measures | Relevant residuals or differences | Sphericity and dependence are separate assumptions |
Use the statistical-test selection guide for nursing dissertations to connect variable type, design and assumptions to the correct analysis.
What to Do When Data Are Not Normal
A significant normality test does not automatically mean “use Mann–Whitney U”. First identify the nature and cause of the departure.
Check coding and data entry
Correct verified errors, retain an untouched raw file and document every change.
Consider whether the pattern is meaningful
Length of stay, symptom counts and cost data can be naturally right-skewed. The distribution may represent the phenomenon rather than poor quality.
Review the intended analysis
Consider the size of the departure, group balance, influential observations and robustness. Check all assumptions, not normality alone.
Choose a defensible alternative when necessary
Depending on the question, options include a justified transformation, robust approach, suitable non-parametric test or model designed for the outcome distribution. Transformations change the scale. Rank-based tests may address a different hypothesis from tests of means.
Use sensitivity analysis
If two approaches are defensible, compare whether conclusions materially differ. Do not select whichever method produces the preferred p-value.
How to Report Normality in a Nursing Dissertation
State what was assessed, which evidence was considered and how the judgement affected analysis.
Example supporting approximate normality
Knowledge scores were examined within each group using histograms, normal Q–Q plots, boxplots and the Shapiro–Wilk test. The plots showed no substantial asymmetry or influential observations. Minor tail departures were not judged sufficient to invalidate the planned independent-samples analysis.
Example showing a marked departure
Length-of-stay values were positively skewed in both groups. Shapiro–Wilk tests were significant and Q–Q plots showed systematic right-tail departures. Because the pattern was substantial and influential observations were present, a method appropriate to the distribution and research question was selected.
Example for paired data
Normality was assessed for the difference between pre-intervention and post-intervention confidence scores. The histogram and Q–Q plot of the differences showed approximate symmetry without influential extreme values.
Replace these models with verified study details. The nursing SPSS results-chapter guide shows where assumption checks fit within Chapter 4.
Common Normality-Testing Mistakes
- Testing IDs or category codes.
- Using Shapiro–Wilk as the only evidence.
- Claiming p > .05 proves normality.
- Automatically rejecting a parametric method when p < .05.
- Checking combined data instead of relevant groups.
- Testing pre- and post-scores rather than paired differences.
- Assuming every raw variable must be normal for regression.
- Deleting valid values to improve the result.
- Selecting a non-parametric method without checking its hypothesis.
Normality Assessment Checklist
- Is the variable genuinely continuous?
- Are missing codes and impossible values handled?
- Were relevant groups examined separately?
- Were histograms, Q–Q plots and boxplots reviewed?
- Was Shapiro–Wilk interpreted with sample size in mind?
- Were skewness, kurtosis and influential values considered?
- Was the correct quantity assessed, such as differences or residuals?
- Does the final method answer the research question?
- Are decisions and sensitivity checks reported transparently?
Frequently Asked Questions
Is Shapiro–Wilk enough?
No. Combine it with graphs, unusual-observation checks, descriptive measures, sample size and the exact assumptions of the intended analysis.
Should I use Kolmogorov–Smirnov or Shapiro–Wilk?
SPSS may show both, but applied researchers commonly prefer Shapiro–Wilk. Either way, do not interpret the p-value without graphs and context.
Does significance mean I must use a non-parametric test?
No. Evaluate the magnitude of departure, outliers, group balance and robustness before deciding.
Do Likert responses need normality testing?
A single Likert item is ordinal. A justified multi-item score requires decisions based on its construction, measurement properties and analysis. See the nursing Likert-scale SPSS guide.
Where should normality output appear?
Present evidence needed to justify the analysis in the main text and follow university guidance for supplementary output. Do not paste every raw table into the chapter.
Conclusion
Knowing how to test normality in SPSS for nursing research requires more than reading one significance value. Verify the dataset, assess the correct outcome, differences or residuals, examine relevant groups, combine graphical and numerical evidence, and connect the judgement to the planned analysis. Transparent reasoning is more defensible than labelling data simply “normal” or “non-normal”.
References
- Ghasemi, A. and Zahediasl, S. (2012) ‘Normality tests for statistical analysis: a guide for non-statisticians’, International Journal of Endocrinology and Metabolism, 10(2), pp. 486–489. https://doi.org/10.5812/ijem.3505
- IBM (n.d.) ‘Explore’, IBM SPSS Statistics Documentation. Available at: IBM SPSS Explore documentation (Accessed: 26 August 2026).
- Mishra, P., Pandey, C.M., Singh, U., Gupta, A., Sahu, C. and Keshri, A. (2019) ‘Descriptive statistics and normality tests for statistical data’, Annals of Cardiac Anaesthesia, 22(1), pp. 67–72. https://doi.org/10.4103/aca.ACA_157_18
Need Help Interpreting Nursing SPSS Output?
If your normality evidence is mixed, our nursing dissertation data analysis and SPSS support can help review the verified dataset, assumptions, output and reporting.
