Confidence intervals in nursing research show the uncertainty around an estimated result. They help you interpret the size and direction of a finding instead of relying only on whether a p-value crosses a threshold. Reading an interval well means identifying the estimate, checking its units and deciding which clinically relevant values remain compatible with the analysis.
This guide explains interpretation and reporting through invented nursing examples. It focuses on uncertainty rather than the full process of choosing a statistical test. For the wider question of magnitude, see our effect-size guide for nursing research.
Confidence intervals in nursing research: quick answer
A confidence interval accompanies an estimate such as a mean difference, proportion, correlation or odds ratio. Its limits describe uncertainty under a specified statistical procedure and assumptions. A wider interval usually signals less precision, while a narrower interval gives a more restricted range of compatible values.
For a conventional frequentist 95% interval, the 95% refers to the long-run coverage of the method across repeated samples. It is not a claim that 95% of individual patients fall between the limits. Nor does it automatically assign a 95% probability to the fixed population parameter lying in the particular interval you obtained.
Always interpret the range in context. A statistically significant estimate can still be too small to matter in practice. Conversely, an interval that includes the null value may also include meaningful benefit or harm, indicating that the study has not resolved the question precisely.
Why an interval improves your findings chapter
A single estimate makes a result easy to summarise, but it hides uncertainty. Suppose a dissertation estimates a two-point reduction on a symptom scale. Without an interval, the reader cannot tell whether the analysis distinguishes a modest reduction from no difference or a much larger change.
Gardner and Altman (1986) argued for greater emphasis on estimation and confidence intervals in medical reporting. Their central concern remains useful for nursing students: a binary significance decision gives less information than an estimate accompanied by uncertainty.
In your findings chapter, report the result on a scale the reader can recognise. State what is being estimated and how the comparison is defined. Then use the discussion chapter to consider whether the compatible effects would be meaningful for patients, staff or service delivery.
Our SPSS output interpretation guide explains where estimates sit within a wider results workflow. Reading the interval should follow checking the design, coding, analytical sample and model.
Read the estimate, lower limit and upper limit together
An interval normally appears beside the point estimate. The lower and upper limits use the same underlying parameter scale, although ratio measures often require calculations on a transformed scale. The displayed numbers should not be interpreted without knowing the measure.
| Measure | Null value | Example interpretation question |
|---|---|---|
| Mean difference | 0 | Which differences in the original units remain compatible? |
| Risk difference | 0 | Which percentage-point differences remain compatible? |
| Risk ratio | 1 | Which relative risks remain compatible? |
| Odds ratio | 1 | Which relative odds remain compatible? |
| Correlation coefficient | 0 | Which strengths and directions of association remain compatible? |
Check the reference category before interpreting direction. An odds ratio below one can represent lower odds of an adverse outcome, but it can also represent lower odds of a desirable outcome. The number does not tell you whether the result is favourable without the variable definition.
For a mean difference, find out which group was subtracted from which. A negative difference in pain scores may favour an intervention when lower scores are better. A negative difference in confidence scores might have another meaning.
Worked nursing example: a mean and its uncertainty
The following calculation uses invented summary data for teaching. Suppose 100 independently sampled nurses have a mean knowledge score of 60 points and a sample standard deviation of 10 points. We want an interval for the population mean under an appropriate sampling model.
The estimated standard error is 10 divided by the square root of 100, which equals one point. Using the two-sided 95% t critical value with 99 degrees of freedom, approximately 1.984, the interval is 60 ± 1.984 × 1. The resulting limits are approximately 58.02 and 61.98 points.
A results sentence could read: “The mean knowledge score was 60.0 points (SD 10.0; n = 100), with a 95% confidence interval for the mean of 58.02 to 61.98.” This clearly identifies the estimated quantity and avoids confusing the standard deviation with the confidence interval.
The interval does not mean that nearly all individual nurses scored between 58.02 and 61.98. Individual scores can vary much more widely. It also does not prove that the sample was representative: the sampling approach and response patterns remain relevant.
What if the sample was recruited differently?
If those 100 nurses were volunteers from one ward, their responses might not represent nurses throughout the country. A narrow statistical interval can coexist with a serious limitation in generalisability. Precision and representativeness answer different questions.
Similarly, if each nurse completed the questionnaire several times, treating all records as independent observations would produce a misleading standard error. The method needs to account for the repeated structure before the displayed interval can be trusted.
Therefore, describe the recruitment and measurement process alongside the result. An interval is one part of the evidence, rather than a replacement for the methodological account.
Interpret a difference on its clinical scale
Consider an invented comparison where the estimated mean difference in a patient-reported score is −2.0 points, with a 95% interval from −3.5 to −0.5. Define the contrast as intervention minus comparison and suppose lower scores are better.
The estimate favours the intervention. The interval excludes zero and spans reductions from half a point to three and a half points. However, clinical importance requires a defensible threshold or other contextual judgement. The statistical output does not supply that judgement automatically.
If a prespecified meaningful difference were three points, the range would include both smaller and larger reductions than that benchmark. It would be inaccurate to write that the analysis had established a clinically important reduction merely because zero was excluded.
The threshold itself needs support. Use an appropriate measurement source, population and purpose. Avoid choosing a convenient benchmark after seeing the result, particularly when the scale has no accepted minimal important difference for your setting.
Read ratio intervals around one
Ratio measures use one as the no-association value. An invented odds ratio of 1.8 with a 95% interval from 1.1 to 2.9 suggests increased odds under the model and defined comparison. It does not mean the outcome probability is necessarily 80% higher.
A ratio interval is often asymmetrical around the point estimate on the displayed scale. This is not automatically a calculation error. Many methods work with logarithms and transform the limits back to the ratio scale.
Write the outcome and reference group in the sentence. “The odds were higher” is incomplete unless the reader knows the odds of what and relative to whom. Where possible, include absolute outcome frequencies to give the comparison a practical context.
Distinguish adjusted from unadjusted estimates. An interval from a regression model refers to the specified model and included variables. Adjustment does not automatically eliminate residual confounding or correct inaccurate measurement.
What determines the width of an interval?
Sample size is one influence, but it is not the only one. Variability, event frequency, study design, model specification and the confidence level also matter. More participants will not necessarily produce a reliable estimate if the underlying data collection is biased.
In the simple mean example, the standard error scales with the square root of the sample size. Doubling the sample does not halve the interval width. A substantial increase in precision can require a larger increase in independent information than students expect.
Rare outcomes can create wide intervals even when the total recruitment number sounds substantial. If only a few participants experience an event, the information available about its association with a predictor may be limited.
Clustering also changes effective information. Two hundred patients from a small number of wards do not necessarily provide the same precision as 200 independently sampled patients. Use a procedure suited to the design.
For planning rather than interpreting completed results, see our sample-size guide for nursing dissertations. Choose a justification linked to the study objective, rather than promising that any convenient sample will give a narrow interval.
Why 90%, 95% and 99% intervals differ
Keeping the data and method fixed, a higher confidence level normally produces a wider interval. The procedure must cover the parameter more often across repeated sampling, so its limits become less restrictive.
Choose the confidence level through the analysis plan and relevant disciplinary expectations. A 95% interval is common, but it is not a universal requirement for every research purpose. State the level clearly in table headings and prose.
Do not switch to a 90% interval after noticing that the 95% interval includes the null value. That choice changes the evidence threshold in response to the results. If alternative levels are reported for a justified reason, explain the decision transparently.
How confidence intervals relate to p-values
For a matched two-sided hypothesis test and confidence-interval procedure, a 95% interval excluding the null commonly corresponds to p < 0.05. The word “matched” matters: different procedures, rounding, discrete exact methods and multiplicity adjustments can complicate this relationship.
The interval provides more than a significance check. It shows which magnitudes the analysis can distinguish. Reading only whether zero or one falls inside the range discards much of the information that makes intervals useful.
A p-value also should not be interpreted as the probability that the hypothesis is true. The American Statistical Association’s statement cautions against treating a threshold as a complete scientific conclusion (Wasserstein and Lazar, 2016). Combine the estimate, interval, design and substantive question.
An interval including the null does not prove no effect
Suppose an invented mean difference is −1.0 points, with a 95% interval from −4.0 to 2.0. The range includes no difference, a potentially useful reduction and an increase. The study is compatible with several interpretations and has not estimated the comparison precisely.
It would be misleading to write “the intervention had no effect”. A better statement is that the estimate was imprecise and the interval included zero alongside differences in both directions. Then explain which differences would matter in your nursing context.
Altman and Bland (1995) warn against confusing a lack of evidence with evidence of absence. Demonstrating equivalence or non-inferiority requires an appropriate design, margin and analysis; an ordinary non-significant comparison does not establish either.
This distinction is especially useful for small dissertations. Students sometimes feel obliged to convert every result into a decisive recommendation. A carefully explained uncertain result is more credible than an unsupported claim of similarity.
Do not confuse confidence, prediction and reference intervals
A confidence interval addresses uncertainty about a parameter, such as a population mean. A prediction interval concerns a future observation under a model and incorporates individual variability as well as estimation uncertainty.
A reference interval describes a range of values in a defined reference population. It serves a different purpose from estimating the precision of a mean. The terminology matters when discussing clinical measurements.
| Term | Main target | Nursing reporting question |
|---|---|---|
| Confidence interval | An estimated population or model parameter | How uncertain is the estimated mean or association? |
| Prediction interval | A future observation under a model | What range could a future individual value occupy? |
| Reference interval | Values in a defined reference population | What range characterises that population? |
Check what your source actually reports before copying its limits into an evidence table. Labelling a reference range as a confidence interval changes its meaning and can distort the appraisal.
Find and verify the interval in SPSS output
Identify the procedure and parameter first. Output may contain intervals for a mean, a difference, a coefficient or a ratio. These are separate quantities, even when they appear in nearby tables.
Read the column headings and footnotes. Confirm the confidence level, analytical sample and adjustment status. If several models were run, ensure the interval belongs to the model described in your methods chapter.
Do not calculate an interval by adding and subtracting a standard deviation from the estimate. The standard deviation describes variation among observations; it is not generally the standard error needed for inference about a mean.
Retain the syntax and output. A reproducible analysis record should show the procedure, options, exclusions and data version. If an interval is missing from the output, request it through the appropriate procedure or obtain advice rather than inventing a plausible-looking range.
Reporting confidence intervals in nursing research
Use a consistent structure: estimate, confidence level, lower limit, upper limit and units. For example, a mean difference might be reported as “−2.0 points, 95% CI −3.5 to −0.5”. A ratio needs its measure label and defined reference category.
Use enough decimal places to support interpretation without implying excessive precision. If rounding makes a limit appear equal to the null value, check the unrounded output before making a threshold statement.
Table headings can reduce repetition. A column labelled “Mean difference (95% CI), points” lets each row present its estimate and limits concisely. Define all group contrasts in the notes so the direction remains clear.
In the discussion, separate what the sample suggests from what remains uncertain. Explain whether the interval spans effects that would lead to different practice decisions. A useful interpretation considers the whole range, rather than celebrating the point estimate alone.
Final interpretation checklist
- Identify the estimated parameter and units.
- Confirm the reference group or subtraction order.
- Read both limits and the confidence level.
- Check the null value appropriate to the measure.
- Consider meaningful benefit and harm across the range.
- Verify the method accounts for the study design.
- Distinguish precision from generalisability.
- Match every interval to its model and sample.
- Report uncertainty consistently in tables and prose.
- Avoid claiming no effect from a non-significant result.
Ask a reader to explain your main interval in ordinary language. Their explanation should preserve the units, direction and uncertainty. If it reduces the finding to “significant” or “not significant”, revise the discussion to make the magnitude clearer.
Support for interpreting your own findings
Our nursing data analysis and SPSS support can help you review estimates, intervals and results wording. Bring anonymised output and your analysis plan. The aim is to strengthen your understanding and reporting while meeting your university’s requirements.
Frequently asked questions
Does a 95% interval contain 95% of patients?
No. It concerns uncertainty about an estimated parameter. Individual patients can show much greater variation than the interval around a population mean.
Is a narrow interval always better evidence?
It usually indicates greater precision under the method. However, biased sampling, poor measurement or an unsuitable model can still undermine the result.
Why does an odds-ratio interval use one?
An odds ratio of one represents equal odds under the defined comparison. Difference measures instead use zero as their no-difference value.
What should I write when the interval includes zero?
Describe the compatible range and uncertainty. Do not automatically conclude that there is no effect, especially when meaningful differences remain inside the interval.
Can I choose a different confidence level?
Yes, when justified by the analysis plan and research purpose. State the level and avoid changing it merely to obtain a preferred threshold conclusion.
Interpret the whole range
An informative nursing dissertation reports an estimate with uncertainty and explains the values that matter for its question. Check the parameter, direction, units and method before judging the interval. Then connect its full range to the practical decisions your study can support.
References
- Altman, D.G. and Bland, J.M. (1995) ‘Absence of evidence is not evidence of absence’, BMJ, 311, p. 485. Available at: doi:10.1136/bmj.311.7003.485.
- Gardner, M.J. and Altman, D.G. (1986) ‘Confidence intervals rather than P values: estimation rather than hypothesis testing’, British Medical Journal, 292(6522), pp. 746–750. Available at: Full article.
- Wasserstein, R.L. and Lazar, N.A. (2016) ‘The ASA’s statement on p-values: Context, process, and purpose’, The American Statistician, 70(2), pp. 129–133. Available at: ASA statement.
