Chi-square in nursing research helps you examine whether two categorical variables are associated. For example, you might ask whether medication adherence differs between patients who received discharge education and those who did not. The test adds an inferential question to the counts and percentages in your table.
Its importance comes from matching a common nursing question to the right type of data. Many outcomes have categories: infection present or absent, screening completed or missed, and referral accepted or declined. However, a useful test still needs independent observations, suitable expected counts and careful interpretation.
This guide explains why chi-square matters, shows a worked nursing example and identifies situations where another method fits better. For a broader comparison of methods, use our statistical-test selection guide.
Why chi-square in nursing research is useful
Nursing datasets often record events and classifications rather than numerical scores. A researcher may count how many patients developed a pressure injury or how many nurses completed a safety checklist. These questions require an analysis of category frequencies.
Pearson’s chi-square test of independence compares the observed cell counts with those expected under a model of independence. It therefore addresses whether the distribution of one categorical variable differs across categories of another (McHugh, 2013).
Consider an audit of falls in two ward types. The descriptive table tells you the proportion of patients who fell in each setting. Chi-square adds evidence about an association under the model assumptions. It does not explain why that pattern arose.
This distinction matters when writing a nursing dissertation. You can describe an association between ward type and recorded falls. However, differences in patient frailty, staffing or observation periods may still affect the comparison. The statistical test cannot resolve those issues by itself.
Consequently, chi-square is useful when the research question, variables and design support a categorical comparison. It should follow the approved analysis plan rather than become a default choice for every nursing dataset.
Recognise categorical nursing data before choosing a test
A categorical variable places an observation into a defined group. Binary variables have two categories. Other variables may have three or more. The category labels should describe the measurement clearly and consistently.
For example, “screening completed” needs an operational definition. Does it mean a documented assessment within 24 hours, or completion at any point during admission? Different definitions create different datasets, even when researchers use the same statistical test.
| Nursing question | First variable | Second variable |
|---|---|---|
| Is discharge education associated with adherence? | Education received: yes/no | Adherence classification: adherent/non-adherent |
| Does screening completion vary by ward? | Ward: medical/surgical/rehabilitation | Screening: completed/not completed |
| Is training attendance associated with checklist use? | Training attended: yes/no | Checklist use: documented/not documented |
| Is referral uptake associated with discharge pathway? | Pathway: routine/supported | Referral: accepted/declined |
These are hypothetical research questions. They do not imply that any relationship exists in practice. Your own definitions, sampling and authorised records determine what you can analyse.
Also, avoid converting a continuous score into “high” and “low” merely to use chi-square. A threshold needs a substantive justification. Otherwise, you may discard useful information and change the question your study answers.
What a contingency table reveals before testing
A contingency table, also called a crosstab, displays combinations of categories. Rows might represent education exposure and columns might represent adherence. Each cell contains a count of observations meeting both definitions.
Read the table before interpreting the p-value. First, check the group sizes. Next, choose percentages with denominators that answer your question. Finally, inspect empty categories and missing values.
If you want to compare adherence within education groups, row percentages may be appropriate when education occupies the rows. Column percentages answer a different question: how education exposure is distributed within each adherence category.
IBM’s Crosstabs documentation distinguishes observed counts, expected counts and percentages. Its cell-display options also include residuals, which describe differences between observed and expected frequencies (IBM, n.d.-a).
These options help you explain the pattern transparently. Nevertheless, do not treat every available percentage as a separate finding. Select the denominator that readers need and label it clearly.
For instance, 30 patients with an event in a group of 300 is different from 30 in a group of 60. Counts alone can hide that distinction. A useful nursing table reports both the number of events and the relevant group denominator.
A worked example of chi-square in nursing research
Teaching example: The following counts are invented for explanation. They are not patient records, published study findings or SPSS screenshots. The calculation uses Pearson’s chi-square without a continuity correction.
Suppose a cross-sectional dataset contains 200 independent patient observations. Of these, 100 patients received discharge education and 100 did not. Adherence was classified using a predefined measurement rule.
| Education exposure | Adherent | Non-adherent | Total |
|---|---|---|---|
| Received education | 80 (80%) | 20 (20%) | 100 |
| Did not receive education | 60 (60%) | 40 (40%) | 100 |
| Total | 140 | 60 | 200 |
The percentages are within education groups. Therefore, the observed adherence difference is 20 percentage points. That descriptive finding gives readers a concrete comparison before they encounter the inferential result.
Calculate the expected counts
Under independence, the expected count for a cell equals its row total multiplied by its column total, divided by the grand total. For the educated-and-adherent cell, this gives 100 × 140 ÷ 200 = 70.
The four expected counts are 70, 30, 70 and 30. These are model-based expectations, not targets for care and not values you should substitute into the dataset.
Connect the calculation to the nursing question
For this table, χ² = 9.52, with one degree of freedom and p = .002. Cramér’s V is .218, rounded to .22. All expected counts exceed five.
The result provides evidence against independence under the stated model. The table identifies the pattern: adherence was more common among patients who received education. However, this observational example does not establish that education caused the difference.
An appropriate results sentence is: “Discharge education exposure was associated with adherence classification, χ²(1, N = 200) = 9.52, p = .002, Cramér’s V = .22; adherence was recorded in 80% of the education group and 60% of the comparison group.”
Check independence and expected frequencies
A meaningful analysis starts with the unit of observation. Ask whether each row represents a different patient, nurse or other independent unit. Repeated records from the same person cannot automatically be treated as unrelated observations.
For example, counting five admissions from one patient as five independent patients can distort the analysis. Likewise, several assessments from the same nurse may reflect within-person dependence. Discuss the structure with your supervisor before selecting a standard independence test.
The categories must also be mutually exclusive within each variable. If a questionnaire permits respondents to select several answers, the total number of responses may exceed the number of people. Such data need a method that respects their structure.
Expected counts require a separate check. A commonly used screening rule for the Pearson approximation is that no expected count is below one and no more than 20% are below five. This is a guideline about expected frequencies, not observed counts or the total sample alone (McHugh, 2013).
In a sparse 2 × 2 table, Fisher’s exact test may be more appropriate. Larger sparse tables may require a suitable exact or Monte Carlo method. Kim (2017) discusses the distinction between chi-squared and Fisher’s exact testing.
For sparse tables, see our dedicated guide to Fisher’s exact test in nursing research, including test selection and reporting.
Do not merge categories simply to obtain significance. Combine them only when the grouping makes clinical and methodological sense, and explain the decision. Similarly, an exact test does not repair dependent observations or poor measurement.
Use the p-value with proportions and effect size
The p-value addresses evidence under a specified statistical model. It does not show the probability that your nursing hypothesis is true. Nor does it measure how important an association is for patients or services.
The American Statistical Association advises against basing scientific conclusions solely on whether a p-value crosses a threshold. Its statement also distinguishes statistical significance from effect magnitude and importance (Wasserstein and Lazar, 2016).
Therefore, report the category pattern alongside the test result. In the teaching example, readers need to know about the 80% and 60% adherence proportions. Reporting only p = .002 would leave the practical comparison unclear.
Phi or Cramér’s V can summarise association strength. However, their interpretation should consider the table structure and research context. Avoid declaring an association clinically important merely because a conventional effect-size label seems favourable.
For a 2 × 2 table, a risk difference, risk ratio or odds ratio with a confidence interval may offer a more directly interpretable comparison. Choose the measure that fits the study design and outcome. These measures are not interchangeable.
For further guidance on magnitude, use our effect-size guide for nursing research. The broader SPSS output interpretation guide explains how to present several types of statistical results.
When a different analysis fits better
Chi-square is important because it answers a particular question well when the assumptions fit. Its usefulness does not mean it should replace methods designed for paired observations, continuous outcomes or adjusted comparisons.
| Data situation | Why standard chi-square may not fit | Possible direction |
|---|---|---|
| The same patients have a binary outcome before and after education | The observations are paired | McNemar’s test or another appropriate paired method |
| Groups differ in a continuous confidence score | The outcome is numerical | A suitable mean, distribution or regression comparison |
| A binary outcome needs adjustment for age and comorbidity | A simple crosstab does not adjust for confounding | Logistic regression, if the design and data support it |
| Patients are clustered within wards | Observations may be dependent within clusters | A survey-adjusted or multilevel approach |
| A small table has sparse expected counts | The asymptotic approximation may be unsuitable | An appropriate exact method |
IBM documents McNemar procedures for paired categorical comparisons (IBM, n.d.-b). For example, the question “Did these patients change their adherence status?” differs from “Do two independent groups have different adherence distributions?”
Also, chi-square treats category membership rather than the spacing between ordered levels. If your question concerns a graded trend across categories, a method that uses that ordering may be more informative.
Choose the procedure from the nursing question and design. Do not run several alternatives and retain only the smallest p-value. If the analysis is exploratory, identify it openly.
How chi-square can support nursing audits and research
In an audit, a crosstab can reveal differences in documentation or screening completion across independent groups. For example, it may help identify where recorded compliance varies between wards. The comparison can then guide further investigation of local processes.
However, differences in recording practices may create an apparent pattern. A ward with more complete documentation can seem to have more incidents simply because staff record them more consistently. Define the outcome before interpreting its association with the setting.
Quality-improvement projects need particular care. The same patients, staff or wards may be measured repeatedly. Changes over time may also reflect seasonal pressure or another service change. A simple pre/post table does not automatically justify an independence test.
In a survey, chi-square may explore whether professional role is associated with a categorical response. Nevertheless, a convenience sample limits generalisation. The p-value cannot turn a local volunteer sample into a representative national nursing survey.
For a dissertation, connect each comparison to a research question. Explain why both variables were measured as categories, how missing responses were handled and what population the sample can reasonably represent.
This creates a stronger link between statistical analysis and nursing reasoning. It also helps your reader distinguish a documented finding from a proposed explanation that would require further evidence.
Common mistakes that weaken the analysis
One frequent mistake is entering percentages instead of the underlying counts. A table showing 40% and 60% does not reveal whether the comparison involved ten people or a thousand. The sample size affects the inference.
Another problem is selecting the wrong denominator. For example, reporting the percentage of all respondents in each cell may hide the proportion within each ward. Label every percentage and check that it answers the stated question.
Missing values also need transparent handling. A blank adherence record is not automatically non-adherence. Separate missing information from a measured category and report how many observations contributed to the analysis.
Do not describe a non-significant result as proof that groups are identical. The study may contain too little information to detect a meaningful association. Report the proportions, sample and uncertainty, while avoiding claims beyond the analysis.
Finally, a significant omnibus result in a larger table does not identify every category difference. Inspect the pattern carefully. If you use post hoc cell comparisons, justify them and address multiple testing.
These errors are preventable when you review the dataset and table before testing. Keep a brief analysis log so that decisions about coding, exclusions and follow-up comparisons remain visible during supervisor revisions.
Plan and report the test in your nursing dissertation
Start by writing the categorical research question in one sentence. For example: “Is discharge education exposure associated with adherence classification at follow-up?” Next, name the unit of observation and define the categories.
Before collection, consider whether the planned sample is large enough for the expected table structure. A large overall sample may still contain rare categories. Our sample-size planning guide explains why recruitment should follow the intended analysis.
For reporting, include the analysed sample, cell counts and meaningful percentages. Then give the named test, statistic, degrees of freedom, p-value and relevant association measure. Explain any exact method or correction used.
Keep the statistical result consistent across the table and narrative. In particular, avoid rounding a p-value to zero. If software displays .000, report a suitable threshold such as p < .001, following your university’s reporting style.
A reporting scaffold is: “[Variable A] was associated with [Variable B], χ²([df], N = [sample]) = [statistic], p = [value], [effect-size measure] = [value]. The proportion of [outcome] was [percentage] in [group] and [percentage] in [comparison group].”
Replace every placeholder with verified output. For chapter organisation, use the nursing SPSS results-chapter guide. For literature comparison and practice implications, continue to discussing SPSS results.
Frequently asked questions
Why is chi-square important in analysing nursing data?
It provides an inferential way to examine associations between categorical variables. Nursing examples include screening status, referral uptake and recorded incidents. Its value depends on suitable measurement, independent observations and the expected-count structure.
Can I use chi-square for a small nursing sample?
Check the expected frequencies rather than relying on a single total-sample rule. A sparse 2 × 2 table may need Fisher’s exact test. The design still matters: an exact method does not make repeated observations independent.
Does a significant chi-square result prove an intervention worked?
No. It supports an association under the model assumptions. A causal interpretation additionally depends on the study design, allocation, measurement and competing explanations. In observational work, education exposure may be linked with other patient characteristics.
Can chi-square compare the same nurses before and after training?
A standard independence test is generally unsuitable when the same people contribute both measurements. For paired binary outcomes, consider McNemar’s test. Other repeated categorical structures require methods suited to their design.
Do I need an SPSS screenshot in the main findings chapter?
Follow the university brief. Usually, a concise table and accurate written result communicate the finding more clearly than a raw output screenshot. If screenshots are required, use genuine anonymised output from your authorised analysis.
Get support with a categorical nursing analysis
Before requesting help, prepare your research question, category definitions, study design and anonymised output. Also, include the university instructions and any supervisor feedback. These materials make it easier to check whether the chosen method fits the intended comparison.
Our nursing dissertation data-analysis support can help explain test selection, output and reporting. Keep responsibility for understanding and submitting your own work, and follow university rules on external assistance.
Use the website’s support route to describe the analysis question. Share only data that you are authorised to disclose. Never change observations or classifications to achieve a preferred result.
Conclusion
Chi-square in nursing research is useful because many nursing questions concern categories and proportions. It connects a contingency table to an inferential assessment of association. However, independent observations, expected counts and clear definitions remain essential.
A strong dissertation reports the pattern, test result and effect size together. It then explains limitations without turning association into causation. When the design calls for a paired, adjusted or clustered analysis, choose a method that respects that structure.
References
- IBM (n.d.-a) Crosstabs cell display. IBM SPSS Statistics documentation. Available at: IBM Crosstabs cell display (Accessed: 30 September 2026).
- IBM (n.d.-b) MCNEMAR subcommand (NPAR TESTS command). IBM SPSS Statistics documentation. Available at: IBM McNemar documentation (Accessed: 30 September 2026).
- Kim, H.-Y. (2017) ‘Statistical notes for clinical researchers: Chi-squared test and Fisher’s exact test’, Restorative Dentistry & Endodontics, 42(2), pp. 152–155. Available at: https://doi.org/10.5395/rde.2017.42.2.152.
- McHugh, M.L. (2013) ‘The chi-square test of independence’, Biochemia Medica, 23(2), pp. 143–149. Available at: https://doi.org/10.11613/BM.2013.018.
- 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: https://doi.org/10.1080/00031305.2016.1154108.
