SPSS Data Cleaning Checklist Before Thesis Analysis

Keep the original data and a decision log

Save an untouched, access-controlled copy of the raw export. Work in a separate version and record each change: date, variable, rule, number of affected records and reason. Use participant codes rather than names wherever your ethics approval and institution permit. Check the approved protocol and data-sharing conditions before transferring a dataset to a consultant.

The goal is not to make results look significant. Cleaning should make the recorded observations understandable and the analysis reproducible. The US National Institutes of Health data-analysis-plan guidance recommends deciding how data will be organised, cleaned and analysed when planning the study.

Eight checks before your first inferential test

1. Confirm the unit of analysis. Is each row one participant, organisation, visit or questionnaire? Flag duplicate IDs and repeated visits; do not delete a second row until you know whether it is a legitimate follow-up.
2. Build a codebook. For each variable, record its question or source, allowed values, units, labels, type and any reverse coding. Keep the questionnaire and approved protocol alongside it.
3. Compare counts. Reconcile invited, consented, submitted, eligible and analysable records. Document exclusions and their reasons instead of silently removing incomplete cases.
4. Inspect ranges and categories. Frequencies and minimum/maximum values can expose impossible dates, out-of-range scale answers, misspelled categories and mixed units. Check suspicious values against the source; never replace them by guesswork.
5. Define missing codes. A blank, “not applicable”, “prefer not to answer” and a skipped question may mean different things. In SPSS, user-defined missing codes must be declared; otherwise a value such as 99 can be analysed as a real answer. See IBM’s missing-values documentation.
6. Check derived scores. Verify which items belong in a scale, which are reversed, the permitted number of answered items and the direction of the final score. Test calculations on a few records by hand.
7. Inspect missingness and outliers. Report how many values are missing for important variables and whether exclusions affect the intended sample. An extreme value may be a genuine observation; investigate before changing it. IBM’s Missing Value Analysis describes pattern checks and method choices.
8. Match the analysis plan. Check that the variables actually collected can answer each research question. Record the planned descriptive summaries, comparisons or models, assumptions and treatment of missing data before looking for a favourable result.

A 200-response UAE university survey

Suppose an approved five-point questionnaire has 200 submissions. A quick frequency table shows that one item contains 99 in seven rows. The questionnaire used 99 for “not applicable”, so those entries are labelled as user-missing rather than treated as a sixth response category. A second check finds three duplicate respondent IDs. The researcher compares timestamps and source records; two are repeat submissions, while one is a legitimate second-wave observation. Only the confirmed duplicates are handled under a written rule. The final record lists the starting count, every exclusion and the sample size for each analysis.

This example illustrates the workflow, not a rule for treating every 99 or duplicate. The codebook, consent terms and study design decide the correct action. If the dataset was collected before an analysis plan existed, write the plan now and identify any decisions made after seeing the data.

Choose a test only after the dataset is interpretable

Data cleaning and statistical test selection are related but separate tasks. Once the codebook, exclusions and analysis dataset are stable, use our statistical-test selection guide to connect the research question with the study design and variable types. For an early sample-size question, use the sample-size resource as a planning aid, with your supervisor or statistician checking the assumptions.

Reporting guidance such as STROBE for observational studies asks authors to explain variables, study size, statistical methods and missing-data handling. Check the reporting guideline appropriate to your own design and the target journal.

What should the analyst receive?

Provide the approved research questions, questionnaire or instrument, variable codebook, study protocol, de-identified working dataset, cleaning log and current analysis plan. Include the software version and any syntax used to recode variables or compute scores. A table that simply says “data were cleaned in SPSS” is not enough for someone else to understand the exclusions.

Before handing over data, compare the number of cases at each stage: submitted, eligible, excluded for a documented reason and included in each analysis. A participant can be present in one descriptive table yet absent from a model if the model requires a missing covariate. Explain these differing denominators rather than forcing all tables to use the same count.

The NIH data-analysis-plan guidance places cleaning and management procedures within the planned research workflow. For observational research, use the relevant STROBE checklist to check how participants, variables, study size and missing data should be reported.