Dissertation Data Analysis: Step-by-Step Guide

Dissertation data analysis is where your raw research finally turns into evidence and argument. Analysing your data correctly can make or break your dissertation findings chapter. This step-by-step guide walks through the process from raw data to a clear, defensible set of results. If the statistics feel overwhelming, our dissertation writing services include expert data-analysis support for UK students.

The Short Answer

Dissertation data analysis is a sequence, not a single task: clean and document your raw data, choose a method that answers your research question rather than one you happen to know, run the analysis while recording every decision, then present the findings so that a reader can follow how you got from data to claim. The most common failure is not statistical incompetence but missing justification — a test run without checking its assumptions, a theme reported without evidence of how it was derived, or a result presented without any statement of what it does not show.

Dissertation Data Analysis Step 1: Clean and Organise Your Raw Data

Remove duplicate entries, check for missing values, and standardise formatting before running any analysis. A clean dataset prevents errors from appearing later in your results. Before any test, screen your dataset for missing values, duplicates, out-of-range entries and inconsistent coding. Label variables clearly, decide how you will treat missing data, and keep an untouched master copy so you can always trace back. Clean data is the foundation of every credible result — rushing this step is the most common cause of flawed findings.

Step 2: Choose the Right Analysis Method

Quantitative data usually calls for statistical tests such as regression or ANOVA, while qualitative data often uses thematic or content analysis. Your method should match your research question, not the other way around. Match the method to your research question and data type: descriptive statistics to summarise, t-tests or ANOVA to compare groups, regression to model relationships, or thematic analysis for qualitative interviews. Justify your choice explicitly in the methodology so the examiner sees the logic. Guidance from the Royal Statistical Society is a reliable reference when defending your approach.

Step 3: Run the Analysis and Record Everything

Keep a clear record of every test, software version, and setting you use. This makes your methodology reproducible and helps you answer supervisor questions about your process with confidence. Whether you use SPSS, R, Excel or NVivo, document every step — the test used, assumptions checked, outputs produced and decisions made. Screenshots and syntax files make your work reproducible and protect you if a result is questioned in your viva. Reproducibility is a mark of rigour that examiners actively reward.

Step 4: Present Findings Clearly

Use tables and charts only where they add clarity, and always explain what each result means in plain language before linking it back to your research question. Report results in plain language first, then support them with well-labelled tables and figures rather than dumping raw output. Every table should have a purpose and be referenced in the text. Separate the reporting of results from their interpretation, saving deeper meaning for your discussion chapter.

Common Data Analysis Mistakes to Avoid

• Choosing a test that doesn’t fit the data — always check assumptions first.
• Confusing correlation with causation — be precise about what results actually show.
• Over-reporting raw output — summarise, don’t paste every SPSS table.
• Ignoring effect size — statistical significance alone rarely tells the full story.
• Mixing results and interpretation — keep findings and discussion clearly separate.

Matching Your Analysis to Your Research Question

Before you open any software, write your research question at the top of a blank page and underneath it write what kind of answer would count. If the answer is a difference between groups, you need a comparison test. If it is a relationship between variables, you need correlation or regression. If it is a prediction, you need a model with a stated outcome variable. If it is a description of how people understand something, you need a qualitative approach and no significance test will help you.

This sounds obvious, and it is the single most common thing missing from weak dissertations. Students frequently choose an analysis because it was taught in a module or because a previous student used it, then reverse-engineer a question to suit. Markers spot the mismatch immediately, because the discussion chapter ends up answering a question the introduction never asked.

Work backwards instead. Draft the table or figure you would need in order to answer your question, with fake numbers in it. That single sketch tells you what variables you need, at what level of measurement, from how many participants, and therefore which analysis is available to you. It also reveals impossible designs before you collect anything, which is far cheaper than discovering them afterwards. Our overview of the four primary types of research methodology is a useful cross-check at this stage.

Levels of Measurement and Why They Decide Everything

Every quantitative decision downstream depends on how your variables are measured, so classify them explicitly and write the classification into your methodology. Nominal variables are unordered categories such as department or degree programme. Ordinal variables have order but unequal intervals, which is where Likert items sit. Interval and ratio variables have equal intervals, with ratio having a meaningful zero — age, income, reaction time, tensile strength.

The reason this matters is that it constrains your options. You cannot calculate a mean of a nominal variable. Whether you can legitimately average Likert items is a genuine methodological debate, and the defensible position is to treat a single item as ordinal and a validated multi-item scale as approximately interval, saying so and citing the literature that supports the choice. Simply computing a mean of one Likert question and reporting it to two decimal places invites a straightforward criticism.

Also record the shape of your data before you test anything. Look at histograms, check skewness, identify outliers and decide in advance what you will do with them. Parametric tests carry assumptions about distribution, variance and independence, and the honest route is to check each one, report what you found, and either proceed or switch to a non-parametric alternative with a stated reason. A dissertation that says a Shapiro–Wilk test indicated non-normality and therefore used a Mann–Whitney U test demonstrates competence; one that runs a t-test on obviously skewed data does the opposite.

Sample Size, Power and What You Can Honestly Claim

Sample size determines the scope of your conclusions, and the sensible time to think about it is before recruitment. A power analysis based on the effect size reported in comparable published studies tells you roughly how many responses you need to detect an effect of that magnitude. Doing this and reporting it, even briefly, distinguishes a planned study from an opportunistic one.

In reality most undergraduate and master’s projects end up with fewer participants than they hoped for, and that is not fatal provided you handle it honestly. A small sample means wide confidence intervals, low power, and a real chance of missing a genuine effect. It does not mean your work is worthless. What damages a dissertation is claiming generalisability the design cannot support — concluding that a finding from thirty-two students at one university applies to UK students in general.

Write your limitations section in terms of specific consequences rather than generic apology. Explain that your sample was self-selected through a single channel, that respondents were concentrated in one discipline and age band, and that the direction of any resulting bias is therefore predictable. Then say what a better-resourced study would do differently. This is credited as methodological awareness, whereas a vague sentence about a small sample size is not. If you are struggling with recruitment, our guide to research interviews covers practical access strategies.

Handling Missing Data, Outliers and Messy Responses

Real datasets arrive incomplete, and how you deal with that is examinable. Start by quantifying it: how many cases have missing values, on which variables, and is the pattern random or systematic? Missingness concentrated in one question often tells you something interesting about the question itself. Report the extent of missing data rather than silently deleting rows, because a reader needs to know how many cases each analysis actually used.

Then choose an approach and justify it. Listwise deletion is transparent and acceptable when missingness is small and random, at the cost of statistical power. Pairwise deletion keeps more data but means different analyses rest on different samples, which must be reported. Mean substitution is widely criticised because it artificially reduces variance. More sophisticated imputation is available but should only be used if you can explain it. Whatever you choose, state it once in the methodology and apply it consistently.

Outliers need a decision rule set in advance rather than case by case. Distinguish impossible values, which are data entry errors and should be corrected or removed, from extreme but genuine values, which usually should be kept because they are part of the phenomenon. If you remove a genuine outlier, report the analysis both with and without it so a reader can see the effect of your choice. Keep a dated log of every cleaning decision — it becomes an appendix, and it is what allows you to answer questions in a viva with confidence.

Get Expert Help With Your Dissertation Data Analysis

Data analysis is where many strong dissertations either shine or stumble. Projectsdeal has supported UK students since 2001 with statistical analysis, interpretation and write-up. Explore our dissertation writing services or read genuine Projectsdeal reviews to see how we help students turn data into distinction-level chapters.

Common Quantitative Analysis Techniques

Choosing the right statistical test depends on your data and your research question. Descriptive statistics (means, medians, standard deviations) summarise your sample. A t-test compares the means of two groups, while ANOVA compares three or more. Correlation measures how strongly two variables move together, and regression goes further by predicting one variable from others. A chi-square test examines relationships between categorical variables. Always check that your data meet each test’s assumptions — for example, a roughly normal distribution for parametric tests — before running it.

Analysing Qualitative Data: Coding and Themes

Qualitative analysis turns words into insight through systematic coding. The most popular approach in UK dissertations is thematic analysis, often following Braun and Clarke’s six phases: familiarising yourself with the data, generating initial codes, searching for themes, reviewing them, defining them, and writing up. Work through your transcripts line by line, group similar codes into themes, and support each theme with representative quotations. The goal is not just to describe what participants said but to interpret what it means for your research question.

Choosing Your Analysis Software

The right tool depends on your method and confidence. Excel handles basic descriptive statistics and charts and is fine for small quantitative datasets. SPSS is the standard for statistical testing in the social sciences and has a menu-driven interface that suits beginners. R is free, powerful and increasingly expected in quantitative fields, though it has a steeper learning curve. For qualitative work, NVivo helps you organise codes and themes across large amounts of text. Whatever you choose, keep a clear record of every step so your analysis is reproducible. If you need support running or interpreting your analysis, ProjectsDeal’s data specialists can help.

Qualitative Analysis: Making Your Coding Defensible

Qualitative analysis is often assumed to be the easier route and is in fact harder to do well, because rigour has to be demonstrated rather than assumed from a test statistic. Whichever approach you use — thematic analysis, framework analysis, grounded theory, interpretative phenomenological analysis, discourse analysis — name it, cite the methodological source you followed, and describe the steps you actually took rather than paraphrasing the textbook.

A defensible account includes how many times you read the data, whether you coded inductively from the data or deductively against a framework, how initial codes were grouped into candidate themes, how themes were reviewed against the full dataset, and how they were named and defined. Include your coding frame as an appendix with code definitions and illustrative extracts, and state the point at which you stopped generating new codes. Where a second coder was involved, report how disagreements were resolved rather than only quoting an agreement statistic.

Evidence every theme with data. A theme supported by one quotation from one participant is a quotation, not a theme. Indicate how widely a theme appeared without pretending to count, using careful language such as most participants or a minority. Present disconfirming cases too: material that does not fit your themes strengthens your analysis when you engage with it and undermines it when a reader suspects it was omitted. Finally, say something about your own position, because in qualitative work the analyst is part of the instrument.

Reporting Results: Numbers, Tables and Precision

Report enough for a reader to evaluate the claim. For a comparison test that means the descriptive statistics for each group, the test statistic, the degrees of freedom, the exact probability value rather than a threshold, and an effect size with a confidence interval. Effect size is the part most often omitted and the part that answers the question a marker actually cares about: not whether there is a difference, but whether it is big enough to matter.

Keep precision honest. Two decimal places is conventional for most statistics, three for probability values, and no more decimal places than your measurement justifies. Report probabilities as exact values rather than only as above or below a threshold, and never describe a result as approaching significance. Percentages from small samples should be accompanied by the raw counts, because a percentage from a base of eleven people is misleading on its own.

Split the labour between text, tables and figures so nothing is duplicated. Tables carry the full numerical detail; figures show patterns and relationships; the text points the reader to what matters and says what it means. Every table and figure needs a number, a caption, units, and a mention in the text before it appears. And keep interpretation out of the results chapter if your structure separates results from discussion — many UK rubrics penalise mixing them, and our guide to dissertation chapter structure explains where the boundary usually sits.

From Findings to Discussion: Interpretation Without Overclaiming

The discussion is where marks are won, and its job is to explain what your results mean in the context of the literature you reviewed. Take your findings in the order of your objectives and, for each one, say what you found, whether it agrees with previous studies, why it might differ if it does, and what follows practically or theoretically. A discussion that simply restates the numbers in words adds nothing.

Guard against three specific overclaims. Correlation is not causation, however strong the relationship and however plausible the mechanism; if your design is cross-sectional, say so and speak of association. Statistical significance is not practical importance; a tiny effect can be significant in a large sample and irrelevant in the world. And a non-significant result is not evidence of no effect, particularly in a small sample — it is an absence of evidence, and the honest phrasing reflects that.

Engage with your own weaknesses before a marker has to. Name the threats to validity that apply to your design, explain which direction each would push your results, and identify the one change that would most improve a repeat study. Confidence that survives an honest account of limitations reads as expertise; confidence that depends on ignoring them reads as inexperience. For the reasoning moves involved, see our guide to critical writing.

A Reproducibility and Pre-Submission Checklist

Before you write up, make the analysis reproducible by someone else. Keep the raw data file untouched and work only on copies. Give files dated, versioned names. Record every transformation — recoded variables, reversed items, computed scales, excluded cases — either in a syntax or script file or in a written analysis log. If your software allows saved syntax, use it rather than menu clicks, because syntax is a record and clicks are not.

Then check the write-up against the data. Does every number in your text match the output? Do totals in tables add up, and do percentages sum to a hundred where they should? Does the number of participants stated in the methodology match the number in each analysis, and if not, is the discrepancy explained? Are variable names in the text the same as in your tables? Inconsistencies here are common and they undermine confidence in everything else.

Finally, sanity-check the story. Read your abstract, your results and your conclusions in sequence and ask whether they describe the same study. Confirm that each research objective is answered somewhere explicit. Check that appendices contain what the main text says they contain and that each one is referenced. Then leave time to have the whole thing read by someone who has not seen it, because the errors you cannot find are the ones your familiarity is hiding. Our guide to writing an abstract is a good final step once your findings are settled.

Frequently Asked Questions