How to prepare your data for SPSS analysis

Start from your research questions, not the software

Before opening SPSS, write each research question or hypothesis on a single line, and beside it note which variables would answer it. This short exercise usually reveals two things: some questions have no matching variable in the questionnaire, and some collected variables answer nothing. Both are much cheaper to discover now than after analysis has started.

For every question, also note what kind of answer you expect — a comparison between groups, a relationship between two measures, a description of a sample, or a model predicting an outcome. You do not need to name a statistical test yet. The appropriate test depends on your measurement levels, your sample and your study design, and choosing it is a decision to make with your supervisor or a statistician rather than in advance.

• Each research question or hypothesis written out in full
• The variables that answer each question listed beside it
• Any question with no supporting variable flagged for discussion
• The study design noted: cross-sectional, repeated measures, experimental, or other

Write a codebook before you enter anything

A codebook is a plain table that describes every column in your dataset. It is the single most useful document you can hand to anyone who works on your data, and it is what lets you return to your own file six months later and still understand it.

• Variable name — short, no spaces, consistent (for example job_satisfaction_1).
• Full question wording — exactly as respondents saw it.
• Measurement level — nominal, ordinal, or scale.
• Value codes — for example 1 = Strongly disagree through 5 = Strongly agree.
• Missing value code — a value that cannot occur naturally, such as 999 or -99.
• Notes — reverse-scored items, filter questions, or items belonging to a named scale.

Mark reverse-scored items clearly in the notes column. Reversed items that are never recoded are one of the most common causes of a reliability figure that looks inexplicably low, and the problem is invisible in the data file itself.

De-identify the file you share

Whether you are sending data to a supervisor, a statistician or a support service, share a de-identified working copy rather than the original collection file. Remove names, email addresses, phone numbers, student or employee numbers, exact addresses and any free-text answer that identifies a person or a workplace.

Replacing identifiers with a sequential participant ID is pseudonymisation, not full anonymisation: as long as a mapping file exists, the data can still be linked back to individuals, so keep that mapping in a separate, access-restricted location. Indirect identifiers matter too. If only one participant belongs to a given department, age band and job grade combination, that person can often be recognised, so consider grouping those categories more broadly or removing them from the shared file.

Before sharing anything, check what your participants consented to and what your ethics approval, institution or funder requires: some approvals allow sharing only with named people, only within a country, or only after a formal data-sharing agreement. Where those documents set conditions, they decide what you may send, not general practice.

Clean consistently, and record what you changed

Cleaning is not about making the data look tidier. It is about making sure every value in the file means what the codebook says it means. Work through the file column by column rather than row by row.

• Run a frequency table for every variable and look for values outside the valid range
• Check that blank cells are genuinely missing and not accidentally recorded as zero
• Decide for each variable whether missing values stay as system missing (empty cells) or use declared codes, and record the choice in the codebook
• Where you use codes such as 99 for no response and 98 for not applicable, declare them as missing values in SPSS and keep them outside the valid range of the scale
• Keep meaningful distinctions separate — not applicable, no response and do not know are different answers and should not be merged
• Never leave a missing code undeclared, or it will be counted as a valid score in means, totals and reliability analysis
• Standardise text categories — Male, male and M should not be three groups
• Check dates and numeric fields imported as text, which is common in spreadsheet exports
• Look for duplicate participant IDs and identical response patterns from the same timestamp
• Confirm the row count matches the number of responses you expect to analyse

Keep a short cleaning log: the date, what you changed, how many cases were affected and why. If you exclude cases — incomplete responses, failed attention checks, respondents outside your sampling frame — record the rule you applied and how many cases each rule removed. Examiners and reviewers frequently ask exactly this, and a log written as you go is far more reliable than one reconstructed later.

Never overwrite the raw data

Save the original export under a name such as raw_export_2026-09-20.xlsx, set it to read-only, and do every piece of cleaning in a copy. If a cleaning step turns out to be wrong, you can go back; if you have overwritten the original, you cannot.

Keep at least two copies in different places, and name files so their order is obvious: 01_raw, 02_cleaned, 03_analysis. Where your software allows it, record cleaning and recoding as syntax rather than manual edits — an SPSS syntax file makes every step repeatable and reviewable, which is difficult to demonstrate when changes were typed directly into cells.

What to send when you ask for analysis help

If you are asking someone to analyse your data, sending these five items together usually removes an entire round of back-and-forth.

• The anonymised dataset, in Excel or SPSS format.
• The codebook, including value labels and reverse-scored items.
• The questionnaire or instrument as it was administered.
• Your research questions and hypotheses.
• Any analysis requirements from your university or supervisor, plus your deadline.

It also helps to say what stage you are at and what you want from the work — output tables only, output plus written interpretation, or a walkthrough so you can explain the analysis yourself in a viva. Those are different pieces of work, and naming the one you need makes the quote and the timeline accurate.