Written for university students in the UK and internationally. Conventions and terms vary between countries and institutions, so your own course handbook and your tutor’s guidance are always the final authority.
For many students, collecting the data is the comfortable part. The anxiety starts when a spreadsheet of responses needs to become findings, and software like SPSS sits there waiting. The good news is that the software is the easy bit; it will run whatever you ask. The real skill is knowing what to ask it, and what the answer means. This guide walks through the thinking behind a sound quantitative analysis, so the numbers you report are ones you can actually defend.
Prepare the data before you analyse anything
Good analysis starts with clean data. Before any test, you code your variables consistently, decide how you will represent missing responses, and check for obvious errors — an age of 350, a scale answer outside its range, a duplicated row. Decide in advance how to handle missing data rather than letting the software make silent choices for you. Time spent cleaning is never wasted; an impressive-looking test run on messy data produces confident nonsense, and that is far harder to spot after the fact than before.
Describe before you test
Always look at your data descriptively before running any inferential test. Means, medians, ranges, frequencies and a few simple charts tell you the shape of what you have collected, flag anomalies, and often answer part of your question on their own. Descriptive statistics also tell you whether your data looks the way a given test expects. Jumping straight to a p-value without first understanding your distributions is how students end up applying the wrong test and never realising it.
Match the test to the question and the data
The test you choose depends on what you are asking and what kind of data you have. Are you comparing groups or looking at a relationship? How many groups? Are those groups different people (an independent design) or the same people measured more than once (a repeated-measures design)? Is your outcome a number or a category? These questions, not habit or guesswork, point to the right test. The table below shows some common pairings as a starting map.
| What you are asking | Design / data | Common test |
| Compare the mean of two groups | Two independent groups | Independent-samples t-test |
| Compare means of three or more groups | Independent groups | One-way ANOVA |
| Compare the same people on two occasions | Repeated measures | Paired-samples t-test |
| Compare the same people over three or more occasions | Repeated measures | Repeated-measures ANOVA |
| Relationship between two numeric variables | Two continuous variables | Pearson correlation |
| Association between two categories | Two categorical variables | Chi-square test |
Illustrative pairings only, and note the split between independent-group and repeated-measures designs. The final choice also depends on whether your data meets each test’s assumptions — for example, Welch’s t-test is often better when two groups have unequal variances — so check these before you run anything.
Assumptions are not optional
Most common tests assume certain things about your data, and these are not academic fine print; if they are badly violated, the result can mislead. Know which assumptions your chosen test makes and how to check them in SPSS. A t-test, for instance, assumes independent observations and reasonably similar variances between groups; where the variances differ, Welch’s t-test is often the better choice. A Pearson correlation assumes a roughly linear relationship between two continuous variables and is sensitive to outliers, so plot the data first. A chi-square test needs adequately sized expected counts in each cell. When an assumption fails, the right response depends on which one — a non-parametric test is a common alternative, but it is not an automatic fix for every problem, so match the remedy to the specific violation.
Read the output honestly: what a p-value is
The p-value is the number students most often misread, so it is worth stating carefully. It is the probability of seeing a result at least as extreme as the one you observed if the null model — including the assumption of no real effect — were true. Put another way, a small p-value (commonly below 0.05) indicates that your result is relatively incompatible with that null model, assuming the model’s assumptions hold. It is not the probability that chance produced your result, nor the probability that your hypothesis is true, and it says nothing about how large or important any effect is. Treating “p < 0.05” as the whole story is the single most common error in student analysis.
This framing follows the American Statistical Association’s statement on statistical significance and p-values (Wasserstein & Lazar, 2016, The American Statistician), which sets out these cautions in full.
Report effect size and a confidence interval, not just significance
Because a p-value says nothing about magnitude, report an effect size alongside it — a measure of how big the difference or relationship actually is — and, where you can, a confidence interval, which shows the range of values your data is consistent with. With a large enough sample, a trivially small effect can be statistically significant; with a small sample, a real and important effect can miss significance. Reporting the p-value, the effect size and a confidence interval together, and commenting on practical as well as statistical significance, shows an examiner that you understand what your results mean rather than just whether a threshold was crossed.
Correlation is not causation
The most important sentence to keep in mind during analysis is that a relationship between two variables does not prove one causes the other. A correlation can arise because A affects B, because B affects A, or because some third factor drives both. A survey showing that students who use the library more also get higher marks does not establish that the library raised their marks; more motivated students may simply do both. Unless your design was built to test causation — typically an experiment with a controlled comparison — report relationships as associations, and be careful with the language of cause in your discussion.
Think about sample size before, not after
How many responses you need is a question to settle at the design stage, not once the data is in. A sample that is too small may miss a real effect; an unnecessarily large one wastes effort. The right number depends on the size of effect you expect, how much variation there is, and the analysis you plan — a formal power calculation can estimate it. Whatever you decide, record your reasoning, because an examiner may ask why you aimed for the number you did, and a stated rationale is stronger than “it was what I could get”.
Turn output into words
Finally, your results chapter is not a gallery of SPSS tables. Each analysis should be reported in a sentence or two a reader can follow: what you tested, what you found, and what it means for your research question. Lead with the finding in plain language, then give the supporting statistics. For example (all figures here are hypothetical): students who used the app scored higher on average (M = 68, SD = 9) than those who did not (M = 61, SD = 11); an independent-samples t-test showed the difference was statistically significant, t(98) = 3.21, p = .002, with a medium effect size (d = 0.52, 95% CI [0.19, 0.85]). A results section that explains itself in words, with the tables as evidence, is far stronger than one that drops raw output on the reader.
A second opinion on your analysis
A wrong turn in analysis is easy to make and hard to notice alone, which is why statisticians themselves ask colleagues to check their work. For a similar check, students can turn to assignment and dissertation support for help choosing appropriate tests, verifying assumptions and interpreting output, with the scope agreed in advance and their university’s rules respected. The aim is always numbers you understand well enough to explain and defend in your own words.
