Missing data in nursing research can change sample sizes, statistical precision and the interpretation of dissertation findings. The correct response is not to hide blank cells or automatically replace them with averages. Nursing students using SPSS should first establish what is missing, why it may be missing, how much information is affected and whether the planned analysis remains defensible.
This guide focuses specifically on missing-data decisions in quantitative nursing dissertations. It complements the site’s broader guides to SPSS data analysis for nursing dissertations, statistical-test selection and SPSS results chapters.
Missing-data workflow
- Preserve the original dataset before making changes.
- Distinguish true missing values from valid values such as zero.
- Quantify missingness by variable and, where useful, by participant.
- Examine whether missingness is concentrated in particular groups or measures.
- Choose a handling method that fits the design, amount and plausible missing-data mechanism.
- Report exclusions, assumptions and sensitivity analyses transparently.
- Never invent participant responses to make a dataset complete.
What counts as missing data in a nursing dissertation?
Missing data occur when a value that should have been observed is unavailable. Examples include an unanswered questionnaire item, an unrecorded follow-up blood-pressure measurement, a laboratory value unavailable at a study time point or a participant who did not complete a later assessment.
Not every blank-looking value means the same thing. A code such as 999 may have been used for “not recorded”, while zero may be a genuine clinical value. Before analysis, check the questionnaire, codebook and data-collection procedure so that valid observations are not accidentally treated as missing.
Why missing data matter
Missingness can reduce the effective sample size and therefore widen uncertainty. More importantly, results may become biased when the people or observations with missing information differ systematically from those with complete data.
For example, suppose a nursing study measures treatment adherence at baseline and three months. If participants experiencing the greatest treatment burden are also more likely to miss follow-up, analysing only complete cases may produce a more favourable picture than the original cohort supports. The issue is therefore not simply the percentage of blank cells; the reason and pattern matter.
Start with a protected raw dataset
Keep an untouched copy of the original data. Perform cleaning and recoding in a working copy and document every substantive change. This protects the audit trail and makes it possible to reproduce the analysis.
Do not fill genuine missing observations merely because a statistical procedure requires complete values. Any imputation or exclusion rule should be an explicit analytical decision rather than an invisible spreadsheet correction.
Define missing values correctly in SPSS
SPSS can recognise system-missing numeric values and user-defined missing codes. If a questionnaire uses values such as 99 or 999 for “not applicable” or “missing”, define these consistently before producing descriptive statistics. Otherwise, SPSS may treat the code as a genuine high score and distort means, ranges and models.
Check frequencies and minimum/maximum values for each variable. Implausible values can reveal coding errors that initially resemble genuine observations.
Quantify the amount of missingness
Begin with simple descriptive checks. For each important variable, record the number of valid observations and the number and percentage missing. For longitudinal studies, also examine retention at each time point.
| Variable | Expected n | Valid n | Missing n | Missing % |
|---|---|---|---|---|
| Age | 180 | 180 | 0 | 0.0% |
| Baseline symptom score | 180 | 176 | 4 | 2.2% |
| Three-month symptom score | 180 | 153 | 27 | 15.0% |
This example is illustrative only. Dissertation results must be calculated from the student’s genuine dataset.
Examine the pattern, not only the percentage
Two datasets can both contain 10% missing values but have very different analytical implications. Missing observations may be scattered across participants and variables, concentrated in one questionnaire, or associated with follow-up loss in a particular clinical group.
Compare relevant observed characteristics between participants with and without key outcome data where the design supports such comparisons. For example, baseline age, clinical severity or treatment group may help identify whether follow-up missingness is patterned. These comparisons do not prove why data are missing, but they can inform the plausibility of assumptions.
Understand MCAR, MAR and MNAR carefully
Missing-data discussions commonly distinguish three mechanisms. Missing completely at random (MCAR) means the probability of missingness does not depend on observed or unobserved values. Missing at random (MAR) means missingness may depend on observed information after conditioning on variables in the model. Missing not at random (MNAR) means the probability of missingness still depends on the unobserved value after considering observed information.
These are assumptions about the data-generating process, not labels that SPSS can establish with certainty. A statistical test alone should not be presented as proof that data are MAR or MNAR. Clinical knowledge, study procedures, observed patterns and sensitivity analysis all contribute to the judgement.
When is complete-case analysis reasonable?
Complete-case analysis uses observations with all values required for a particular model. It is simple and may be defensible in some circumstances, especially when missingness is limited and the assumptions are plausible. However, automatically deleting every participant with any missing value can waste information and introduce bias.
Also distinguish listwise from pairwise deletion. Listwise deletion removes a case from an analysis when a required variable is missing. Pairwise approaches can use different subsets for different calculations, which may produce changing sample sizes that need careful interpretation.
Why mean substitution is usually a weak default
Replacing every missing value with the sample mean may preserve the row count, but it treats uncertain values as though they were known and can distort variability and relationships between variables. It should not be used merely because it is easy to perform.
Similarly, carrying forward an earlier clinical measurement is not automatically justified. The method assumes something about the participant’s later state and can be misleading when outcomes genuinely change over time.
Multiple imputation in nursing research
Multiple imputation creates several plausible completed datasets using an imputation model, analyses each dataset and then combines the estimates while accounting for imputation uncertainty. It can be more defensible than single-value substitution when its assumptions and model are appropriate.
However, multiple imputation is not a button that repairs every dataset. The imputation model should reflect the analysis, include relevant predictors of missingness and outcomes where appropriate, respect variable types and be described transparently. Small samples, unusual distributions, structural missingness and poorly specified models require particular caution.
Students should agree advanced missing-data methods with their supervisor or statistical adviser when these methods go beyond the approved analysis plan.
Missing items within multi-item nursing scales
Questionnaires often have their own scoring rules for incomplete items. A scale developer may specify how many items must be answered before a total or mean score can be calculated. Follow the validated instrument’s scoring guidance rather than inventing a threshold.
This is especially important for Likert-scale data in SPSS. Item-level missingness, reverse scoring and composite-score construction should be handled in the correct order.
Structural missingness is different
Some values are absent by design. For example, a questionnaire may instruct participants who answer “no” to skip several follow-up items. Those cells are not equivalent to accidentally unanswered questions. Code and describe skip-pattern or not-applicable values separately when necessary so the analysis respects the instrument design.
Missing outcome data in longitudinal nursing studies
Attrition deserves explicit reporting. State how many participants entered the study, how many contributed data at each important time point and why losses occurred when that information is genuinely known.
Do not invent reasons for withdrawal. If the reason was not recorded, report it as unknown rather than assuming that the participant deteriorated, recovered or lost interest.
Check how missing data affect each statistical test
Different SPSS procedures may use different effective sample sizes. A descriptive table might contain 170 observations for one variable, a correlation 158 complete pairs and a regression model only 142 complete cases. Always check the n actually analysed rather than assuming it equals the recruited sample.
This becomes particularly important in nursing regression analysis, where adding predictors can substantially reduce the complete-case sample if several variables contain missing values.
Use sensitivity analysis when uncertainty matters
A sensitivity analysis asks whether a substantive conclusion changes under another reasonable analytical decision. For example, a dissertation might compare a justified complete-case model with an approved imputed analysis. If estimates are similar, confidence in the broad interpretation may increase; if they differ substantially, the missing-data assumption becomes an important limitation.
Sensitivity analysis should be planned and reported honestly. Do not run many alternatives simply to select the version producing statistical significance.
How to report missing data in a nursing dissertation
The methodology should explain how missing values were identified, coded and handled. The results should report relevant sample losses and effective analysis sizes. The discussion should consider whether missingness could affect precision, bias or generalisability.
A concise reporting structure might state:
Follow-up outcome data were available for 153 of 180 participants. The primary analysis used the prespecified missing-data approach, and the reduced follow-up sample was considered when interpreting precision and potential attrition bias.
Adapt the wording to the actual design and analysis. Never report an imputation, sensitivity analysis or attrition reason that was not genuinely performed or recorded.
Common missing-data mistakes
- Treating 0, 99 or 999 as missing without checking the codebook.
- Deleting cases before preserving the raw dataset.
- Reporting only the overall percentage missing while ignoring patterns.
- Assuming a formal test proves the missing-data mechanism.
- Using mean substitution as an automatic solution.
- Ignoring validated questionnaire scoring rules.
- Failing to report changing sample sizes across analyses.
- Inventing reasons for participant attrition.
- Choosing a missing-data method because it produces a preferred p-value.
Missing-data checklist for nursing students
- I have preserved the original dataset.
- I know which values represent genuine missingness.
- I have checked frequencies and implausible codes.
- I have quantified missingness for important variables.
- I have examined relevant patterns and attrition.
- My handling method fits the design and approved analysis.
- I followed instrument-specific scoring rules.
- I know the actual sample size used in each major analysis.
- I have not fabricated responses or withdrawal reasons.
- My methods, results and limitations describe missing data consistently.
Frequently asked questions
Should I delete participants with missing data in SPSS?
Not automatically. The appropriate decision depends on which variables are missing, the amount and pattern of missingness, the planned analysis and the assumptions you can defend.
Can I replace missing values with the mean?
Simple mean substitution is generally a weak default because it does not properly represent uncertainty and can distort variability and associations. Use a method justified by the study design and analysis plan.
Does SPSS tell me whether data are MAR?
No statistical output can establish the missing-data mechanism with certainty. The judgement depends on assumptions, observed information, study procedures and subject-matter knowledge.
Should missing data be reported in the results chapter?
Yes when they affect the sample, outcomes or analyses. Readers should be able to understand how many observations contributed to important results and how missingness was handled.
Can I impute questionnaire items?
Follow the validated instrument’s scoring instructions. Some scales permit scoring with a specified number of completed items; others do not. Do not invent an item-replacement rule.
Related Nursing Guides
- Effect Size for Nursing Research: SPSS and G*Power Guide
- G*Power for Nursing Research: Step-by-Step Guide
- How to Discuss SPSS Results in a Nursing Dissertation: Step-by-Step Guide
Conclusion
Missing data in nursing research require transparent decisions rather than automatic deletion or invented replacement values. Define missing codes accurately, preserve the raw data, quantify and examine patterns, choose a method consistent with the study design and report the effective sample used in each important analysis. When uncertainty remains, acknowledge it and use proportionate sensitivity analysis rather than overstating what the dataset can establish.
For project-specific quantitative support, see the nursing SPSS and data-analysis service or send the approved research question, anonymised dataset and analysis requirements for review.
