Missing Data in a Questionnaire: How to Spot It, What to Do and How to Report It
Why missing data matters, how to define and find it in SPSS, when to drop a participant and when to fill in a value, Listwise vs Pairwise, and how to write it up in the method section.
Part of the series: Statistical Analysis for a Seminar Paper or Thesis: The Complete Guide
Almost every questionnaire has participants who skipped a question, stopped halfway or ticked two answers. What you do with them affects your sample size, your results and sometimes your conclusion - and it has to appear in the paper. Supervisors ask about it.
Step 1: define missing values in SPSS
An empty cell in SPSS is "System Missing" and handled automatically. But if you coded missing answers (e.g. 99 or 999), you must define the code in Variable View → Missing → Discrete missing values. Otherwise SPSS treats 99 as a real answer and a 1-5 scale average jumps to 17.
Step 2: check how much and where
Analyze → Descriptive Statistics → Frequencies shows the missing count per item. Ask:
- How much is missing overall? Under about 5% usually doesn't change much.
- Is one item missing a lot? It may have been unclear or sensitive (income, mental health).
- Did some participants stop halfway? A questionnaire 30% complete is a different problem from one skipped item.
Step 3: understand why
- Missing completely at random (MCAR): skipped by chance. Deleting them doesn't bias results, it only shrinks the sample.
- Missing at random (MAR): related to something measured - younger people skipped the income question more.
- Missing not at random (MNAR): related to the value itself - people feeling worst skipped the depression scale. The problem case: what's left paints too rosy a picture.
A seminar paper doesn't have to prove which type - but should show you thought about it.
Step 4: decide what to do
| Situation | Accepted practice |
|---|---|
| A participant who answered little of the questionnaire | exclude them (and report how many) |
| One item missing from a multi-item scale | the mean of the answered items, if about 80%+ were answered |
| A few scattered missing values (up to ~5%) | Listwise (SPSS default) - fine |
| Lots missing, important study (thesis) | multiple imputation or models that handle missing data - discuss with your supervisor |
Mean of answered items: instead of summing, use MEAN.n in Compute Variable, e.g. MEAN.4(q1,q2,q3,q4,q5) - a mean only if at least 4 items were answered.
Listwise vs Pairwise: Listwise drops a participant from any analysis where one of its variables is missing - more consistent. Pairwise uses everything available for each pair of variables, so each correlation rests on a slightly different sample. Fine for a correlation table, but report the range of N.
What not to do: replace missing values with the sample mean. It's easy, but shrinks variance and inflates significance. Many supervisors reject it.
Reporting it
In the method section or at the start of the results:
"Of 214 questionnaires returned, 11 were excluded because fewer than 50% of items were answered, leaving 203 participants. Missing data on individual items ranged from 0% to 3.4%. Scale scores were computed as the mean of answered items, provided at least 80% of the scale's items were answered."
Common mistakes
- Not defining missing-value codes (99) in SPSS.
- A different N in every table with no explanation.
- Summing items (Sum) when one is missing - the score comes out artificially low.
- Dropping "inconvenient" participants and calling it missing data.
FAQ
Should I check this before Cronbach's alpha? Yes - missing data affects reliability too. See Low Cronbach's alpha.
Where does it fit? Step 3 of 7, preparing the data: Statistical analysis for a seminar paper: the complete guide. A file full of gaps and no idea what to do? Tell us.
Tell us what you need, and we'll come back with an expert in Statistical analysis and a quote.
Last updated: