Your online questionnaire collected 480 responses in a week. How many of them should you trust? Careless and duplicate responses are common in online data, and they can weaken correlations, lower reliability and create spurious factors (Meade & Craig, 2012; Huang et al., 2012). Survey data cleaning is the stage where you find these responses with stated rules, remove or flag them, and document what you did. This guide covers seven checks in a sensible order.
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Survey & Questionnaire Design service. Sources last checked: November 2026.
What is survey data cleaning? Survey data cleaning is the systematic screening of questionnaire responses before analysis to detect and handle invalid records: duplicate submissions, ineligible respondents, careless or insufficient-effort responding (straightlining, speeding, failed attention checks), out-of-range values and extreme outliers. Each rule is defined in advance, applied consistently, and the number of excluded cases is reported.
Why clean before you analyze?
Insufficient-effort responses are not random noise. A respondent who answers 4 to every item adds the same value to positively and negatively worded items, which can inflate or deflate correlations depending on the scale design. Cleaning after you have seen the results invites the suspicion that cases were removed to reach significance, so write your survey data cleaning rules before you open the main analysis, ideally in the proposal or a pre-analysis plan.
The seven checks at a glance
| Check | What it detects | How to run it |
|---|---|---|
| 1. Eligibility | Respondents outside the target population | Screening questions and consent items |
| 2. Duplicates | The same person submitting twice | Data, Identify Duplicate Cases in SPSS; Remove Duplicates in Excel |
| 3. Speeders | Completion too fast to read the items | Duration variable from the survey platform |
| 4. Straightlining | Identical answers across a grid | Within-person SD or longest string |
| 5. Attention checks | Not reading instructions | Instructed-response items |
| 6. Inconsistency | Contradictory answers | Reverse-worded item pairs |
| 7. Outliers and range errors | Impossible values, multivariate extremes | Frequencies, boxplots, Mahalanobis distance |
📱 Get your questionnaire reviewed before fielding — Send the draft on WhatsApp; we point out wording and scale issues before you collect data.
1. Eligibility and consent
Survey data cleaning starts with the simplest rule. Remove respondents who did not consent, who answered a screening question outside the target (wrong profession, wrong country, under 18), and test submissions you made yourself. These exclusions are part of the sampling frame, not quality judgments, and they come first.
2. Duplicates
Duplicates arise from double clicks, reminders answered twice, or deliberate repeat submissions. In SPSS, use Data, then Identify Duplicate Cases, matching on a respondent identifier, an email field if you collected one, or a combination of demographics and all item responses. In Excel, Data, then Remove Duplicates does the same on selected columns. Keep the first complete submission and delete the rest. Many platforms also offer settings that limit repeat responses, which is better than cleaning afterwards.
3. Speeders
Survey platforms record start and end times. Compute the duration in seconds and look at its distribution. Huang et al. (2012) suggested that a pace faster than about 2 seconds per item indicates insufficient effort. Others flag completion times below a fixed fraction of the median duration. Whatever cut-off you choose, justify it, set it before the analysis, and remember that very long durations usually mean an interrupted session, not poor quality.
4. Straightlining
Straightlining means choosing the same response option for every item in a block. Two simple indices:
- Within-person standard deviation across a grid of items: compute it in SPSS with
SD(item1 TO item20). A value of zero means identical answers throughout. - Longest string: the longest run of identical consecutive answers, as described by Curran (2016). It needs a short syntax or a spreadsheet formula.
Compute the index separately for each grid or scale rather than across the whole questionnaire, because a respondent may answer carefully in one section and straightline in a later, longer one. Interpret straightlining carefully. A respondent who genuinely agrees with every item on a short, positively worded scale can produce a zero SD honestly. The evidence is stronger when the grid includes reverse-worded items, or when straightlining coincides with speeding or a failed attention check.
5. Attention checks
Instructed-response items such as “Please select Strongly disagree for this statement” are a direct check that respondents read the items (Meade & Craig, 2012). Use one or two in a long questionnaire, not many, and decide in advance whether failing one leads to exclusion or only a flag. Report the wording of each check in the methodology chapter or an appendix.
6. Inconsistency
If the questionnaire contains pairs of items with opposite wording (“I enjoy my work” and “My work bores me”), a respondent who strongly agrees with both is responding inconsistently. Compute the difference after reverse-coding and flag large discrepancies. This check depends on your instrument design, so plan the pairs when you build the questionnaire.
7. Range errors and outliers
Run Frequencies on all variables to catch impossible values, especially in data typed by hand. For multivariate outliers in a set of scale scores, compute Mahalanobis distance through the Save option in Linear Regression and compare it with a chi-square critical value at p < .001 with degrees of freedom equal to the number of variables (Tabachnick & Fidell, 2019). Inspect flagged cases before removing them; an extreme but valid respondent is still data. A case flagged by several checks at once (fast, straightlined and failing the attention item) is a much clearer candidate for exclusion than a case flagged by one.
How to document survey data cleaning in the thesis
Keep a cleaning log: the rule, the date applied, and the number of cases affected. In the methodology chapter, present it as a short flow from responses received to the analytic sample, for example (illustrative numbers): 480 responses received, 12 test or non-consenting entries removed, 9 duplicates removed, 21 speeders removed, 14 cases failing the attention check removed, leaving 424 for analysis. Report the reliability and main results with and without the removed cases if the exclusions are large.
The survey data cleaning steps should also be reproducible. Save the cleaning syntax in SPSS, keep the original export untouched, and work on a copy.
📱 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
What percentage of online survey responses is usually removed?
It varies widely with the population, the platform and the rules. Rather than aiming for a typical percentage, apply rules you set in advance and report exactly how many cases each rule removed.
Should I delete speeders or keep them with a flag?
If the threshold is well justified, deletion is common. A cautious alternative is to run the main analysis with and without them and report whether the conclusions change.
Do attention checks annoy respondents?
One or two clearly worded items rarely cause problems. Many checks, or trick questions, can irritate respondents and may affect later answers.
Can I clean data in Excel instead of SPSS?
Yes, for duplicates, durations and simple indices such as the within-person standard deviation. Keep a record of each step so that the survey data cleaning can be repeated.
Is removing outliers acceptable to examiners?
Removing entry errors is expected. Removing valid extreme cases needs a stated rule and justification, and reporting results with and without them is the safest approach.
Sources
- Meade, A. W., & Craig, S. B. (2012). Identifying careless responses in survey data. Psychological Methods, 17(3), 437–455. https://doi.org/10.1037/a0028085
- Huang, J. L., Curran, P. G., Keeney, J., Poposki, E. M., & DeShon, R. P. (2012). Detecting and deterring insufficient effort responding to surveys. Journal of Business and Psychology, 27(1), 99–114. https://doi.org/10.1007/s10869-011-9231-8
- Curran, P. G. (2016). Methods for the detection of carelessly invalid responses in survey data. Journal of Experimental Social Psychology, 66, 4–19. https://doi.org/10.1016/j.jesp.2015.07.006
- Greszki, R., Meyer, M., & Schoen, H. (2015). Exploring the effects of removing “too fast” responses and respondents from web surveys. Public Opinion Quarterly, 79(2), 471–503. https://doi.org/10.1093/poq/nfu058
- Tabachnick, B. G., & Fidell, L. S. (2019). Using Multivariate Statistics (7th ed.). Pearson.
- IBM. IBM SPSS Statistics documentation: Identify Duplicate Cases. 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.