Cleaning your data in SPSS means preparing your raw dissertation dataset so it is accurate, consistent, and ready for analysis. It covers checking each variable for impossible values, fixing how categories are coded, deciding what to do with missing data, and screening for outliers before you run a single test. Done first, it protects every result that follows in your thesis.
Why cleaning comes before any analysis in your thesis
A statistical test cannot tell the difference between a real value and a typing error, so it will happily run on dirty data and hand you a confident but wrong answer. That is why data cleaning is the first stage of your results chapter, not an afterthought. When you clean your dissertation dataset carefully, you can defend every number to your supervisor, and you avoid the painful situation of discovering an error after the whole analysis is written up. The goal is a dataset where every cell means exactly what you think it means.
A practical cleaning order for your SPSS dataset
Work through your dissertation dataset in a fixed order so nothing is missed. First, open Variable View and set each variable's type, measure (nominal, ordinal, or scale), and value labels so the software treats each column correctly. Second, run frequencies on every categorical variable to catch stray codes, such as a 7 appearing in a question that only allows 1 to 5. Third, run descriptives on every continuous variable and read the minimum and maximum: an age of 200 or a negative score is an impossible value you can trace back and correct. Fourth, screen for outliers using boxplots and standardised scores, deciding case by case whether each is a genuine extreme or a data-entry slip.
Throughout, keep a written log of every change you make. Examiners expect a clear audit trail, and a clean log also lets you rebuild the dataset if you ever need to start the analysis again. When the shape of a continuous variable looks unusual after cleaning, that feeds straight into testing for normality, which decides whether parametric tests are appropriate.
Recoding, computing, and clearing variables the right way
Cleaning is not only about removing errors; it is also about reshaping variables so they answer your research questions. Use Recode into Different Variables to collapse categories or reverse-score items, always keeping the original column intact so nothing is lost. Use Compute Variable to build totals and subscale scores from individual items. If a variable was entered by mistake or is no longer needed, clear it cleanly: select its column in Data View or its row in Variable View and delete it, rather than blanking individual cells, which leaves the column structure confusing. Document each recode so a reader can follow how your final variables were built from the raw responses.
Finding and treating outliers in SPSS
An outlier is a value that sits far from the rest of a variable's distribution, and left unexamined it can drag a mean, inflate a standard deviation, or single-handedly distort a correlation. In SPSS the quickest screen is a boxplot from Explore, which flags points beyond the whiskers, paired with standardised (z) scores from Descriptives; a case beyond roughly plus or minus 3.29 standard deviations is worth a second look. The decision is never automatic. A genuine extreme that belongs in your population usually stays, while a clear data-entry slip is corrected or removed with the reason logged.
Treat outliers transparently rather than quietly deleting whatever looks awkward. Report how many you found, what you did with each, and run the key analysis with and without them so the reader sees whether your conclusion depends on a handful of cases. Because extreme values also bend the shape of a distribution, screening them feeds directly into checking whether a variable is normally distributed and into the wider question of choosing between parametric and nonparametric tests.
Transforming variables in SPSS to meet test assumptions
Beyond fixing errors, cleaning often involves transforming variables so they satisfy the assumptions of the tests you plan to run. When a continuous variable is strongly skewed, a log, square root, or inverse transformation through Compute Variable can pull a long tail back toward symmetry, which matters for procedures that expect roughly normal residuals. Other common moves include binning a continuous measure into ordered groups with Visual Binning, or creating dummy variables from a categorical predictor so it can enter a regression model.
Always keep the untransformed column alongside the new one, and state in your write-up exactly which transformation you applied and why, because a reader needs to map your final variables back to the raw responses. If the transformed variable is headed for a regression, confirm it against the wider assumptions behind linear regression before you commit to the model.
Handling the gaps and reading the output
Almost every dissertation dataset has gaps, and how you treat them changes your results. SPSS marks blanks as system-missing, but you should also define user-missing values so codes like 99 for "prefer not to say" are never analysed as real numbers. The strategy you choose, whether listwise deletion, pairwise deletion, or an imputation approach, deserves a deliberate decision rather than a default, which is covered in handling missing data. Once the file is clean, your first real output is a set of descriptive statistics, and the difference between describing your sample and generalising from it is set out in descriptive versus inferential statistics. Reading those tables correctly is the next skill, explained in decoding each box of SPSS output.
Treat cleaning as the foundation everything else rests on. A tidy, well-documented file makes your analysis faster, your write-up clearer, and your defence far easier, because every value in the table can be traced back to a decision you made on purpose.