Write your dissertation data analysis plan by linking each question to its variables or themes. Then set out your methods, sampling and data preparation before collecting data, so your questionnaire, interview guide and recruitment target support the analysis.
A strong plan explains what you’ll analyse and why each method fits your questions, design and data type. It also sets out how you’ll handle foreseeable problems, whether your dissertation uses quantitative, qualitative or mixed-methods analysis.
Key Takeaways
- Match each research question to its variables, measurement levels and analysis method.
- Specify data preparation, assumptions, missing-data rules and reporting before recruitment starts.
- For qualitative and mixed-methods studies, explain interpretation, sample adequacy and how different evidence sources will connect.
1. Position the Plan Within Your Methodology
The analysis plan usually belongs near the end of the methodology chapter, often Chapter 3, before its conclusion. Two to five paragraphs can suit a straightforward proposal, although complex studies may need subsections or an analysis table. The methodology chapter sets out proposed analyses; Chapter 4 will report the findings.
Write in the future tense: “Responses will be analysed…” rather than describing work as completed. The proposal is a justified, forward-looking plan, not a guarantee that nothing will change.
Keep it consistent with your research questions, sampling strategy and collection instruments. Our UK dissertation proposal guidance explains how those decisions fit together.
Before data collection, state how relevant ethics approval, informed consent and data-protection arrangements will be addressed.
Name the software, such as R or SPSS for quantitative work, or qualitative software such as NVivo. Explain the analytical procedure too, since software alone isn’t a method.
2. Map Questions to Variables and Design
Define Your Measures
In a quantitative study, define and record each variable before data collection. The dependent variable is the outcome modelled as depending on independent variables, also called predictors. In experimental work, researchers assign an independent variable, such as treatment-group assignment; observational researchers measure exposures without assigning them.
Record each variable’s level of measurement. Nominal variables contain unordered categories, whilst ordinal variables contain ordered categories. Interval measures have equal numerical intervals; ratio measures also have a meaningful zero.
Define scoring, units and coding before collection. An individual Likert item is ordinal; a summed scale needs its own measurement justification.
Establish What the Design Can Answer
Three questions guide inferential test selection: What is the research question? What is the study design? What is the level of measurement?
Relational questions examine associations through correlation or regression. Causal questions often examine differences between intervention groups, but a difference alone doesn’t establish causation.
Next, establish whether observations are independent or related. Measurements collected twice from the same participants require a different approach from measurements collected from separate groups.
3. Choose a Statistical Test That Fits

For independent or paired comparisons, use the t test examples below as a starting point, alongside the University of Utah’s statistical methods selection table.
| Research Task | Common Starting Method |
|---|---|
| Compare two independent groups on a continuous outcome | Independent, or unpaired, t-test |
| Compare the same participants at two time points | Paired t-test |
| Compare three or more independent group means | ANOVA |
| Examine a linear association between continuous variables | Pearson correlation |
| Examine a monotonic association using ranks | Spearman correlation |
| Examine association between categorical variables | Chi-square test |
| Predict an outcome or adjust for confounders | Regression analysis |
These are starting points, not automatic selections. Correlation analysis should match a linear or monotonic association. Sparse categorical tables may require Fisher’s exact test, whilst binary outcomes may require logistic regression.
Parametric statistics commonly suit interval or ratio outcomes when their assumptions are met. Normality usually concerns within-group errors, model residuals or paired differences, which are expected to be normally distributed, not every raw variable.
When assumptions don’t support the proposed procedure, consider nonparametric statistics such as Mann-Whitney U for independent groups, Wilcoxon signed-rank for paired observations or Spearman for monotonic associations. These methods don’t necessarily test the same hypotheses as their parametric counterparts.
4. Prespecify Preparation, Assumptions and Reporting
Set Data Preparation and Quality Rules
Explain how you’ll identify duplicates, correct coding errors, score scales and handle missing responses. Specify exclusion rules before seeing results, and state whether data from the pilot study will enter the final dataset.
Before collection, plan ethics approval, informed consent, privacy protections and secure access to data.
Use descriptive statistics such as frequencies, percentages, means or modes to summarise demographics. Pie charts are possible, although tables often communicate categories more clearly. Demographics may also enter justified models as predictors or confounders.
For scales, Cronbach’s alpha of at least 0.70 is a common reliability convention, not proof of validity. Likewise, absolute skewness below 2 isn’t a universal normality rule. Descriptive statistics can summarise your sample, while plots and diagnostics help check assumptions. Whether residuals or paired differences should be normally distributed depends on the procedure. Plan relevant checks, informed by the UNC test-selection guidance.
Define Inference and Recruitment Requirements
Descriptive statistics summarise primary outcomes. Inferential statistics support conclusions beyond the sample, subject to sampling and design limitations.
Prespecify the significance threshold, usually 0.05. This permits a 1-in-20 long-run risk of rejecting a true null hypothesis under the test’s assumptions. It doesn’t mean a significant finding has a 5% probability of being wrong.
Also plan effect sizes and confidence intervals. If you test several hypotheses, explain how you’ll address multiple testing.
For a quantitative study, base recruitment on the primary analysis, meaningful effect size, variability and desired power. Our dissertation sample size guide explains how to distinguish the required analysable sample from recruitment numbers.
5. Plan Qualitative Analysis and Sample Adequacy

Explain Transcription and Coding
Choose and justify an approach to qualitative data analysis that fits your questions and data. Reflexive thematic analysis develops interpretive patterns across accounts; qualitative content analysis may suit systematic categorisation. Explain whether coding will draw mainly on the data, existing theory or both. The approach should fit the aims of your qualitative study.
For Braun and Clarke’s thematic analysis, plan familiarisation, coding, theme development, theme review, definition and writing. Describe the process as recursive, with dated analytical memos.
Specify transcription accuracy checks, anonymisation and secure storage. Require a confidentiality agreement for an external transcriptionist. You can organise responses separately by interview question initially, but analyse across whole transcripts and the dataset. Interview questions don’t automatically become themes.
Define an Appropriate Stopping Rationale
Where data saturation fits your methodology, begin analysis during recruitment. One possible schedule is a formal review after about 12 interviews, followed by batches of three or four.
Record whether further interviews add relevant codes, themes or interpretive depth. However, don’t treat those numbers as universal thresholds. Fusch and Ness’s discussion of qualitative saturation explains why saturation varies between studies.
For reflexive thematic analysis, justify adequacy through the study’s scope, participant diversity and richness of accounts rather than imposing a mechanical “no new themes” rule.
6. Explain How Mixed-Methods Evidence Will Connect
A mixed-methods plan needs more than separate quantitative and qualitative paragraphs. State which strand comes first, how participants connect across strands and where integration will happen.
In an explanatory sequential design, survey findings can inform interview sampling and questions. Plan a joint display that places numerical findings beside related qualitative interpretations, including disagreement.
Also distinguish triangulation across methods or sources from participant feedback on your interpretation. Re-contacting respondents can help assess whether an interpretation fits their accounts, but doesn’t establish statistical reliability by itself.
Choose follow-up numbers for a methodological reason. Seventeen respondents out of 830 is about 2%; a 5% follow-up would involve approximately 42. Neither percentage is a universal validation requirement.
7. Turn Decisions Into Proposal Wording
Adapt the following wording to the design you’ve justified. These are drafting models, not evidence of supervisor approval.
Paired Quantitative Analysis
The primary analysis will compare continuous outcome scores measured before and after the intervention in the same participants. A paired t-test will estimate the mean change, subject to assessment of the paired differences and influential outliers. Results will include the mean change, a 95% confidence interval and the p value. Two-sided testing will use a significance threshold of 0.05.
Follow this with your missing-data rules, sample-size justification, software and defensible contingency analysis. A single-group before-and-after comparison cannot isolate the intervention’s causal effect.
Qualitative Interview Analysis
Interview recordings will be transcribed, anonymised and checked against the audio. Analysis will follow Braun and Clarke’s reflexive thematic approach, with repeated movement between coding, theme development and review. Analytical memos will document interpretive decisions and contradictory accounts. Reporting will connect themes to contextualised quotations.
Add the recruitment rationale and explain how your position may shape interpretation.
Frequently Asked Questions About Dissertation Data Analysis Plans
Do I Always Need Inferential Statistics?
No. A descriptive question may require frequencies, percentages or summary measures without hypothesis testing. Qualitative research requires an appropriate interpretive method rather than statistical tests.
However, descriptive work still needs decisions about inclusion, coding, missing information and uncertainty. Explain what your analysis can establish and where its claims stop.
Can I Change the Plan After Collecting Data?
Yes, when data quality, access or justified methodological concerns require changes. Keep dated versions and explain the reason for each revision.
Distinguish prespecified analyses from exploratory ones. Don’t switch procedures merely to obtain statistical significance. If a change affects approved recruitment, consent or data handling, an ethics amendment may be necessary before further work.
Build the Analysis Before the Dataset
A defensible dissertation data analysis plan connects your question, design and evidence before recruitment begins. Its strength lies in justified decisions, including how you’ll prepare data, assess assumptions and report uncertainty.
Draft the plan alongside your instruments and sample-size rationale. When those pieces agree, you can collect data that answer the question you’ve asked.
Use this checklist to confirm your data analysis plan:
- Map each question to variables or themes.
- Match analysis methods to the design and data type, and justify sampling choices.
- Specify data preparation, assumption checks, software and ethics.
- Set out how you’ll report the findings.

