The data analysis for a thesis does not start when you open SPSS; it starts when you re-read your objectives. The most repeated mistake among doctoral students is jumping straight to the software, running tests to see what comes out significant, and then trying to write something coherent. This guide covers the correct order, step by step, and applies equally to a thesis in psychology, nursing, medicine, education or sport science: what changes between disciplines is the constructs, not the logic of the analysis.
Step 1: go back to your objectives and hypotheses
Every analysis you are going to run must answer a specific objective or test a specific hypothesis. Before touching the data, write the list: for each objective, what question it answers and which variables it involves. If an analysis does not map onto an objective, it is surplus; if an objective has no associated analysis, you are missing one. This map is the backbone of your results section and what prevents the analysis from turning into a significance hunt.
Step 2: classify your variables
The appropriate test is determined, above all, by the nature of your variables. For each one, identify whether it is quantitative (continuous or discrete) or categorical (nominal or ordinal), and which is the dependent variable and which are the independent or predictor variables. This classification, combined with the design (independent groups or repeated measures, number of groups), leads you almost automatically to the right family of tests.
Step 3: start with descriptive statistics
Before any significance test, describe. Means, standard deviations, frequencies, percentages of missing values and a look at the distributions. This step is not filler: here you catch coding errors, impossible values, missing data and distributional features that condition what comes next. A clean descriptive table is also the first thing any reviewer asks for.
Step 4: choose the test or model
With variables classified and design clear, you choose the test. Comparing two independent groups on a continuous variable leads to the t-test (or its nonparametric alternative); comparing three or more groups leads to ANOVA; relating two continuous variables leads to correlation or regression; predicting a categorical variable leads to logistic regression. If you are uncertain, the guide on how to choose the statistical test walks you through the decision.
Step 5: check the assumptions
Each test assumes things about the data (normality of residuals, homogeneity of variances, independence, linearity), and using a test whose assumptions are violated invalidates the result. Checking assumptions is not a formality: it decides whether you use the parametric test, a robust alternative or a transformation. The guide on how to verify statistical assumptions covers what to check and what to do when they fail. One key detail: normality is almost never judged by the Shapiro-Wilk test in large samples, where it almost always rejects regardless of whether it matters.
Step 6: run the analysis in your software
SPSS, R, JASP or jamovi: any of them works if used correctly. SPSS is the most common in health theses; R is the most powerful and reproducible; JASP and jamovi are free and produce output that is nearly APA-ready. If you have not yet decided, the comparison of SPSS, R and JASP will orient you. Always save the syntax or script: an analysis you cannot reproduce is not defensible.
Step 7: interpret in substantive language, not just the p-value
A result is not "significant and done". Report and interpret the effect size and its confidence interval, because an effect can be statistically significant and clinically trivial, or non-significant due to low power despite being relevant. Translate every number into what it means for your research question. The guide on common errors in interpretation and the one on confidence intervals help you avoid the classics.
Step 8: report in APA (or Vancouver) format
The final step is writing the results with the conventions your committee or journal expects. In psychology and social sciences it is APA 7; in health, often Vancouver. The guide on how to report results in APA 7 covers the format for each test.
Common errors
Starting with the software instead of the objectives; choosing the test by habit rather than design; ignoring assumptions; stopping at the p-value without reporting effect size; not reporting missing values and how they were handled; and changing the analysis after seeing the data to get significance. Deciding the analysis before seeing the data (a data analysis plan) prevents almost all of these. And if, after reading this, you prefer to delegate the analysis, compare how much statistical analysis for a thesis costs and how to choose a statistician first.
Stuck on the analysis for your thesis?
I am a PhD in Psychology and I guide thesis data analysis from start to finish, in any health or social-sciences discipline: I choose the right model, run it in SPSS or R, interpret it and deliver the write-up for the committee. Free initial diagnosis.
See the statistical consulting →