You collected 312 questionnaires, but only 241 are complete. Do you delete the other 71, fill the gaps, or analyze what you have? The answer depends on why the data are missing and how much is missing. Handling missing data in SPSS badly can shrink your sample, bias your results and draw a pointed question at the defense. This guide covers how to define, diagnose and treat missing data in SPSS, and what to report.
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.
What is missing data? Missing data are values that were intended to be collected but are absent for some cases: skipped questionnaire items, participants who dropped out, or unreadable entries. Rubin (1976) classified the mechanisms behind them as missing completely at random (MCAR), missing at random (MAR) and missing not at random (MNAR), and the appropriate handling method depends on which mechanism is plausible.
The three mechanisms: MCAR, MAR and MNAR
| Mechanism | Meaning | Example |
|---|---|---|
| MCAR | Missingness unrelated to any variable, observed or not | A page of some questionnaires was lost in printing |
| MAR | Missingness related to observed variables, not to the missing value itself | Older respondents skipped online items more often, and age is recorded |
| MNAR | Missingness related to the missing value itself | Respondents with high debt skip the debt question |
The mechanism matters because each method for missing data in SPSS makes an assumption about it: listwise deletion is unbiased only under MCAR, while multiple imputation and maximum likelihood remain valid under MAR. Only MCAR can be tested to some extent from the data. MAR and MNAR cannot be distinguished statistically; you argue for them from the design and the variables you collected (Schafer & Graham, 2002).
Step 1: Define missing values correctly in SPSS
A blank numeric cell in SPSS is a system-missing value, shown as a dot. If you used codes such as 99 for “no answer”, declare them in Variable View, in the Missing column, as discrete missing values. An undeclared 99 is treated as a real score and will distort every mean and correlation. Run Analyze, then Descriptive Statistics, then Frequencies on all items to check that no out-of-range codes remain.
📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.
Step 2: Diagnose how much and where data are missing
- Per variable: the Frequencies output shows the number of missing values for each item.
- Per case: compute a count of missing items with Transform, Compute Variable and the NMISS function, for example
NMISS(item1 TO item30). - Patterns: Analyze, then Multiple Imputation, then Analyze Patterns gives charts of the overall percentage of missing values and the most frequent patterns.
- Little’s MCAR test: Analyze, then Missing Value Analysis, ticking EM. A non-significant chi-square (p > .05) is consistent with MCAR (Little, 1988). A significant result suggests the data are not MCAR.
Missing Value Analysis and Multiple Imputation belong to the IBM SPSS Missing Values option, so check that your license includes it before you plan this route. Without it, you can still describe missing data in SPSS with Frequencies and NMISS counts, and compare respondents with and without gaps on key variables using t-tests or chi-square.
Step 3: Choose a method for missing data in SPSS
| Method | How in SPSS | When it is acceptable | Main risk |
|---|---|---|---|
| 1. Listwise deletion | Default in most procedures | Small amount of MCAR data | Loses cases; biased under MAR |
| 2. Pairwise deletion | Options, Exclude cases pairwise (correlation, some procedures) | Correlations with scattered gaps | Different n per statistic; inconsistent matrices |
| 3. Mean substitution | Transform, Replace Missing Values, Series mean | Rarely; not recommended | Shrinks variance and correlations |
| 4. Prorated scale scores | Compute with MEAN.n, e.g., MEAN.4(sat1 TO sat5) | A few items missing within a scale | Assumes skipped items resemble answered ones |
| 5. EM estimation | Missing Value Analysis, EM | Estimating means, correlations, alpha | Standard errors too small if used as single imputation |
| 6. Multiple imputation | Multiple Imputation, Impute Missing Data Values | MAR data, moderate amounts | Needs a good imputation model and pooled reporting |
Methodologists recommend maximum likelihood or multiple imputation over deletion and mean substitution whenever data are plausibly MAR (Schafer & Graham, 2002; Enders, 2022).
How does multiple imputation work in SPSS?
- Include in the imputation model all the variables of your analysis, plus auxiliary variables that predict missingness (age, education, mode of completion).
- Run Analyze, then Multiple Imputation, then Impute Missing Data Values. SPSS creates 5 imputed datasets by default; more imputations give more stable results when the percentage missing is higher (van Buuren, 2018).
- Set a random seed under Transform, Random Number Generators so the imputations can be reproduced.
- Split the file by the variable Imputation_ (SPSS does this automatically with the marked dataset) and run your analyses. Procedures that support pooling print a “Pooled” row, which is the result you report.
- Check that the imputed values are plausible, for example within the 1 to 5 range for Likert items.
Multiple imputation is the most demanding way to treat missing data in SPSS, and not every SPSS procedure pools results. If yours does not, you need to combine estimates using Rubin’s rules or choose another approach, and say so.
A practical decision path for questionnaire data
- Remove cases that are almost empty (for example, more than half of the items missing) and record how many.
- If a case misses only one or two items in a scale, use a prorated mean score with
MEAN.n, setting n to the minimum number of answered items you accept. - If missing data in SPSS remain on key variables after this, check Little’s test and the patterns.
- Under small amounts of MCAR data, listwise deletion is defensible. Otherwise, use multiple imputation.
- Compare results with and without imputation as a sensitivity check and mention the comparison.
What to write in the methodology chapter
Report the number and percentage of missing values per variable (or the range), the overall pattern, Little’s test result if run, the rule for excluding cases, the handling method and its justification, and for multiple imputation the number of imputations, the variables in the model and the software. A short paragraph on missing data in SPSS, written this way, pre-empts most examiner questions.
📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.
Frequently asked questions
How much missing data is too much?
There is no universal cut-off. A frequently cited rule of thumb treats less than about 5% as unlikely to matter much, but the mechanism matters more than the percentage: 3% MNAR on a key variable can bias results more than 15% MCAR.
Is Little’s MCAR test enough to justify deletion?
It helps but is not decisive. A non-significant result is consistent with MCAR, yet the test has limited power. Combine it with your knowledge of how the data were collected.
Can I replace missing values with the scale midpoint?
No. Replacing gaps with 3 on a 1-to-5 scale invents neutral answers and shrinks variability. Use prorated scores or multiple imputation instead.
Should I impute demographic variables?
Usually not for reporting the sample profile; report them as “not stated”. If a demographic variable is a predictor in your model, it can be included in multiple imputation.
Does missing data in SPSS affect Cronbach’s alpha?
Yes. The Reliability procedure uses listwise deletion, so alpha is computed only on complete cases. Report the *n* used, or compute alpha from EM estimates and state that you did so.
Sources
- Rubin, D. B. (1976). Inference and missing data. Biometrika, 63(3), 581–592. https://doi.org/10.1093/biomet/63.3.581
- Little, R. J. A. (1988). A test of missing completely at random for multivariate data with missing values. Journal of the American Statistical Association, 83(404), 1198–1202. https://doi.org/10.1080/01621459.1988.10478722
- Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the state of the art. Psychological Methods, 7(2), 147–177. https://doi.org/10.1037/1082-989X.7.2.147
- Enders, C. K. (2022). Applied Missing Data Analysis (2nd ed.). Guilford Press.
- van Buuren, S. (2018). Flexible Imputation of Missing Data (2nd ed.). Chapman & Hall/CRC. https://stefvanbuuren.name/fimd/
- IBM. IBM SPSS Missing Values documentation. https://www.ibm.com/docs/en/spss-statistics
Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.