Why Data Preparation Comes Before Any Analysis
Most SPSS mistakes do not happen in the test itself. They happen before it, in a data file that was never properly checked. A misplaced decimal, a missing-value code treated as a real score, or a grouping variable coded backwards can quietly produce output that looks perfect and means nothing.
Data preparation is the work you do between entering your data and running your first test: structuring the file, naming and labelling variables, defining missing values, screening for errors, and documenting every decision. It is unglamorous, but it is what makes everything after it trustworthy. Analysts who skip it spend far more time later trying to explain strange results that were never real in the first place.
This checklist gives you nine steps, in order, to take any raw SPSS data file from messy to analysis-ready. Work through them once, save a clean copy, and you will never have to wonder whether your output can be trusted.
Step 1: Back Up Your Original File
Before you change anything, save a copy of the raw file exactly as you received it, and never edit that copy. Name it clearly, for example survey_raw_2026-11-03.sav, and store it somewhere separate from your working file.
This is your safety net. If a recode goes wrong, a merge overwrites the wrong rows, or you later need to prove what the original data looked like, the untouched raw file is your evidence. Every later step in this checklist happens on a working copy, never on the original.
Step 2: Check the Structure — One Row per Case, One Column per Variable
Open Data View and confirm the basic structure: each row should be one case (one respondent, one participant, one observation) and each column should be one variable. If your file came from a spreadsheet, check for common problems: a header row that SPSS read as data, merged cells, blank rows at the top, or several questions crammed into one column.
To fix a file imported with the wrong header row, re-import it properly: File > Import Data > Excel, tick the option that the first row contains variable names, and check the preview before you finish. Getting the structure right first saves you from fixing the same error in every later analysis.
Step 3: Name Your Variables Properly
Switch to Variable View and give every variable a short, clear name. SPSS variable names cannot contain spaces, cannot start with a number, and should avoid special characters. Use names you will still understand in six months: age_years is better than v12, and anxiety_total is better than score2.
Work in Variable View, one row at a time: type the name in the Name column, then move on to its type, width and decimals. For questionnaire items, keep a consistent pattern, such as q01 to q20, so that later steps like computing a total score are easy to write and easy to check.
Step 4: Set the Measurement Level for Every Variable
Still in Variable View, set the Measure column for each variable: Nominal for categories with no order (gender, group, yes/no), Ordinal for ordered categories (a Likert scale from strongly disagree to strongly agree), and Scale for numbers where differences are meaningful (age, test scores, reaction times).
This step matters more than it looks. SPSS uses the measurement level to decide which procedures and charts to offer, and treating a nominal grouping variable as Scale — or a Likert scale as Nominal — can lead you to the wrong test. If you are unsure about a variable, think about what its numbers mean: if the numbers are just labels for groups, the variable is Nominal; if higher means more of something, it is Ordinal or Scale.
Step 5: Add Variable Labels and Value Labels
A variable named q07 tells you nothing in output. A variable label tells you everything. In Variable View, click the Label cell and type the full question or description, for example “I feel nervous in social situations”. Your output tables will then be readable without a codebook open beside you.
Then add value labels for every coded variable: click the Values cell, and enter each code with its meaning — 1 = Male, 2 = Female, or 1 = Strongly disagree through 5 = Strongly agree. Value labels are what turn a table full of 1s and 2s into results a reader can understand, and they are essential when you later split the file by group or run crosstabs.
Step 6: Define Your Missing Values
Real data always has gaps, and SPSS needs to know which numbers mean “missing” rather than a real answer. A code like 99 or 999 entered for “did not answer” will be averaged into your results as if it were a genuine score of 99 unless you tell SPSS otherwise.
In Variable View, click the Missing cell for each variable and define the codes you used for missing data, for example discrete missing values 99 and 999. From then on, SPSS excludes those cases from calculations for that variable while keeping them in the file. Check your questionnaire coding first so that your missing codes never overlap with real answers — if a scale genuinely runs to 99, pick a different missing code.
Step 7: Screen for Errors with Frequencies and Descriptives
Now check the data itself for errors. For categorical variables, run Analyze > Descriptive Statistics > Frequencies, move the variables in, and look at the frequency tables. Any value that should not be there — a gender coded 3, a Likert item showing a 7 on a 1-to-5 scale — is a data-entry error to fix at the source, not a result to report.
For Scale variables, run Analyze > Descriptive Statistics > Descriptives and check the minimum, maximum, mean and standard deviation. An age with a maximum of 250, a test score above its possible range, or a standard deviation of zero all signal a problem. Fix genuine errors against the original questionnaires or raw records; never silently delete a value just because it looks odd.
Step 8: Decide How to Handle Missing Data and Outliers
Screening always finds two things: missing values and unusual values. Decide on a plan for each before you analyse, and write it down. For missing data, the honest options are listwise deletion (analyse only complete cases), pairwise deletion, or multiple imputation; the right choice depends on how much data is missing and whether it is missing at random. A small amount of missing data on a few items is very different from a whole scale missing for a third of your sample, and your method section should say which approach you used and why.
For outliers, first check whether the value is an error (Step 7). If it is a genuine but extreme value, do not delete it by default: consider whether it belongs to your population, run the analysis with and without it, and report what difference it makes. Whatever you decide, document it. A reviewer can accept a judgement call; what they cannot accept is an undocumented one.
Step 9: Recode, Compute, Save a Clean File, and Document Everything
Only now, with a clean and checked file, should you create the variables you will actually analyse. Reverse-score questionnaire items where needed (Transform > Recode into Different Variables), compute scale totals (Transform > Compute Variable), and create any grouping variables your analysis requires. Doing this after screening means your new variables inherit clean inputs, not hidden errors.
Then save the result as a new file — for example survey_clean_2026-11-03.sav — keeping the raw file untouched. Finally, write a short preparation log: what you changed, what you recoded, how you handled missing data and outliers, and the date. That log is the difference between a file only you can use today and a dataset anyone, including future you, can defend tomorrow.
Paste-Ready SPSS Syntax for the Checklist
The syntax below is a reusable template for the screening part of the checklist. Replace the variable names with your own, and run it on a copy of your data. It labels a variable, defines missing values, and produces the frequency and descriptive output you need for Steps 6 and 7:
* Step 5: variable and value labels. VARIABLE LABELS age_years 'Age in years' / group 'Experimental group'. VALUE LABELS group 1 'Control' 2 'Intervention'. * Step 6: define missing values (99 = did not answer). MISSING VALUES age_years group (99). * Step 7: screen categorical and scale variables. FREQUENCIES VARIABLES=group /ORDER=ANALYSIS. DESCRIPTIVES VARIABLES=age_years /STATISTICS=MEAN STDDEV MIN MAX. EXECUTE.
Save your syntax file alongside your clean data file. Syntax is a record of exactly what you did, and re-running it on the raw file should reproduce your clean file — which is the strongest proof that your preparation was careful and repeatable.
How to Interpret the Screening Output
Read the Frequencies output first: each table should show only the values you expect, with counts that add up to your sample size once missing cases are accounted for. A stray value, a category with a single case, or a “Valid Percent” column that does not match your expectations is a flag to investigate, not a number to quote.
Then read the Descriptives output: the minimum and maximum should fall inside the possible range for each variable, and the mean and standard deviation should look plausible for what you measured. If the mean age of your sample is 84 when you surveyed undergraduates, the data — not the theory — is wrong. Fix the file, re-run the screening, and only move on when every table looks the way it should.
Reporting in APA Style (7th ed.)
Data preparation is reported briefly but explicitly in the Method section, so that a reader can judge and replicate your analysis. A typical paragraph reads: “Data were screened for entry errors, missing values, and outliers prior to analysis. Missing values (less than 5% on any variable) were handled with listwise deletion. No cases were removed as outliers; one data-entry error was corrected against the original questionnaire.” Adapt the details to your own decisions — the point is that the decisions are stated, not hidden.
When you report results, give the sample size your analysis actually used, not just the number of questionnaires collected. A sentence such as “Analyses were based on 187 participants with complete data” tells the reader exactly what the tables describe. If you used multiple imputation or another method, name it and report the number of imputations, so that your Method section matches what your syntax actually did.
Key Takeaways
- Never edit your raw data file — work on a copy and keep the original untouched.
- Set variable names, measurement levels, labels and missing values in Variable View before running any test.
- Screen every variable with Frequencies and Descriptives, and fix errors against the source records.
- Decide how to handle missing data and outliers before you analyse, and document every decision.
- Save a clean file plus your syntax and a short preparation log — that package is what makes your results reproducible.
Frequently Asked Questions
How long should data preparation take?
For a small survey it is often an hour or two; for a thesis dataset with dozens of variables it can take a day or more. That time is not wasted — it is usually far less than the time lost chasing strange results caused by an unchecked file.
Do I really need value labels for every variable?
Yes, for every categorical variable at least. Without them, your output is a wall of numbers that only you can decode, and any collaborator, supervisor or reviewer will struggle to follow your results.
What is the difference between a blank cell and a missing-value code?
A blank cell is a system-missing value: SPSS treats it as missing automatically. A code like 99 is user-missing: SPSS treats it as a real number unless you define it as missing in Variable View. Both are legitimate, but only the second needs your explicit instruction.
Should I delete outliers before analysing?
Not by default. First check whether the value is an error; if it is genuine, keep it unless you have a stated, defensible reason to exclude it, and report the effect of that decision on your results.
Can I skip data preparation for a quick exploratory look?
A quick look at raw data is fine for getting a feel for it, but treat anything you see as provisional. Before you report a single number, run the full checklist — exploratory results from an unchecked file have a habit of becoming quoted results.
Need this done on your data? Get a free quote → /services/