ACADEMIC WRITING GUIDE
How to Write a Dissertation Data Analysis Plan (With Worked Examples)
Before running a statistical test, coding an interview, or opening an analysis program, you need to know what each research question requires. A data analysis plan is the bridge between the evidence you intend to collect and the claims you will eventually be able to make.
A strong plan does not promise impressive results. It specifies the decisions you will make regardless of whether the findings support your expectations: which records are eligible, how variables or themes will be prepared, which analyses address each question, what checks are necessary, and how uncertainty will be reported. This guide shows you how to build that plan, with complete hypothetical quantitative, qualitative, and mixed-methods examples.
All numerical values and study scenarios in the examples are illustrative teaching material, not results from real participants. Follow your institution’s requirements and obtain relevant ethics approval before collecting or analysing identifiable research data.
What is a dissertation data analysis plan?
A dissertation data analysis plan is a pre-specified, research-question-led description of how collected information will be prepared, examined, interpreted and presented. The US National Center for Advancing Translational Sciences defines such a plan as a roadmap covering data organisation, analytical methods and assumptions, cleaning, transformation and presentation. Planning before collection is especially useful because the analysis you need may affect the questions you ask and the variables you measure.
It is not the same as a finished results chapter: a plan states intended procedures; the results chapter reports what those procedures produced. Nor is a plan simply a list of software commands. “Use SPSS and Excel” names tools but leaves the analytic reasoning unanswered.
What belongs in the plan?
| Component | Question your plan must answer |
|---|---|
| Research objective | What exactly must be answered? |
| Evidence and units | What records, participants, observations, variables or transcripts will be analysed? |
| Preparation | How will eligibility, duplicate entries, missingness, coding and quality be handled? |
| Method | Which analysis can address this question, and why? |
| Assumptions and checks | What must be true for the interpretation to be defensible? |
| Interpretation and reporting | What estimates, themes, displays, uncertainty and limitations will be reported? |
The precise level of detail varies by department and design. Quantitative research often needs an explicit outcome, exposure or predictor, model and assumptions; qualitative work needs a systematic account of interpretation and reflexivity; mixed methods needs a genuine integration strategy rather than two unconnected analyses.
Nine steps to write a defensible data analysis plan
Step 1. Start with your research questions—not with your favourite software
Create one row per research question. Copy the wording exactly from your approved proposal and write the particular evidence needed to answer it. For example, “How are perceived feedback usefulness and engagement associated?” calls for two measured constructs and an association estimate; “How do students experience feedback?” calls for participants’ accounts and a method of interpreting them. A question about causes demands a stronger design than an observational correlation alone can supply.
If the questions remain broad, revise them using our research-question guide before choosing analytical methods. A method chosen for the wrong question cannot rescue a poorly aligned study.
Step 2. Specify the unit of analysis, population and sample
Say whether each row represents a person, class, organisation, clinic visit or repeated observation. State which records will be included and excluded, the planned sample size or sample logic, and whether clustering or repeated measures require special handling. A survey with multiple responses from the same student is not automatically a dataset of independent students.
Also name the analytic population: all recruited participants, all respondents, complete cases, or an eligibility-defined subset. If attrition, low response rates or convenience sampling could restrict generalisability, do not silently treat the sample as representative.
Step 3. Define what each variable or qualitative source means
For quantitative studies, distinguish the concept from the operational measure: a five-item engagement score is not identical to engagement itself. Specify item coding, scoring directions, scale range, handling of reverse-keyed items, source, and which result is the primary outcome. For qualitative studies, identify the types of interviews, documents or observations and what each can reveal.
Use the operational-definitions guide to connect theoretical concepts to reproducible measurements. Unclear constructs should be resolved before analysis.
Step 4. Set the cleaning and missing-data rules in advance
Plan how you will check unique identifiers, allowed values, skip patterns, outliers, missingness and possible duplicate records. A missing response should not automatically be changed to zero; zero may be a legitimate observed value. Differentiate “not asked,” “declined,” “not applicable,” and truly unknown where the instrument supports it.
For the primary analysis, say how many complete cases are required and whether a suitable imputation or other missing-data method is justified. A rule such as “exclude any record with any blank field” can remove useful participants and bias comparisons. Document reasons for exclusions and compare included versus excluded cases when feasible. State how you will protect the raw dataset and keep an auditable log of transformations.
Our research-data cleaning walkthrough provides the hands-on checking and documentation steps. Your analysis plan should record the rules that will govern those actions.
Step 5. Match each question to an appropriate method
A question about a single distribution may require counts, percentages or a median. A comparison between groups might call for an appropriate group-comparison model, depending on the measurement scale and sampling structure. A question about the association between two constructs may call for correlation or regression. Interviews about experience might be explored through reflexive thematic analysis, framework analysis or another defensible qualitative approach. The analysis must follow the research question, design and evidence—not whichever test produces the smallest p-value.
Name the primary analysis and justify it. If secondary subgroup analyses are planned, list them separately. Avoid presenting exploratory findings as if they were specified before data were seen.
Step 6. Define assumptions, diagnostics and fallback actions
For a statistical model, describe checks relevant to that model: independence, distribution of residuals where pertinent, functional form, influential observations, sparse categories, multicollinearity or expected cell counts. These checks do not all apply to every method. Decide how you will respond to problems: correct a data error, use a method with more suitable assumptions, transform a variable for a substantive reason, or report a limitation.
Do not write “normality must be satisfied” without saying what would need to be normally distributed and why. Similarly, removing all outliers simply because they are extreme risks changing the research question. For qualitative work, explain attention to coding consistency or analytic transparency as appropriate to the methodological tradition, along with reflexive memos, attention to contrary cases and a record of analytic decisions.
Step 7. Predefine what a meaningful result looks like
For quantitative questions, identify the measure of interest: for example, a correlation coefficient or an adjusted regression slope with a confidence interval. If using a significance threshold, state it and avoid equating a p-value with practical importance or causal proof. For qualitative questions, explain what makes an interpretation convincing: quotations connected to a coherent analytic account, context and consideration of variation and disconfirming evidence.
A study can produce a useful null result or a nuanced finding that contradicts initial expectations. The plan should make it possible to report either honestly.
Step 8. Plan the output and interpretation
List the specific tables and figures you expect: participant flow, variable summary, primary model estimates, theme map or integrated joint display. Distinguish a descriptive table from a model-output table. Explain how you will avoid repeating every number in prose and how you will connect each output back to the corresponding research question.
For accessible, accurately scaled presentation, see our tables-and-figures guide. A chart is useful only when its labels, denominators and uncertainty allow readers to understand the comparison.
Step 9. Specify transparency, deviations and safeguards
Record analysis software and versions when known, access controls and storage arrangements, who performs the analysis, and whether any decisions require supervisor or ethics review. If a method changes after examining the data, document the original approach, the reason for change and how the revision affects interpretation. Do not retroactively call a newly invented hypothesis “pre-planned.”
Worked example 1: Quantitative survey analysis plan
Study question and measures
Suppose a dissertation asks: “What is the association between perceived usefulness of written feedback and self-reported academic engagement among undergraduate students at one university?” The hypothetical design is a cross-sectional survey. Perceived feedback usefulness is a mean score from four 1–5 response items, with higher scores indicating greater usefulness. Engagement is a mean of six properly keyed 1–5 items, with higher scores indicating greater self-reported engagement. Study year is collected as a potential covariate; no causal conclusion is planned.
| Research question | Evidence | Planned method | Defensible output |
|---|---|---|---|
| How useful is feedback perceived to be? | Four-item usefulness score | Frequency/missingness and descriptive summary | Valid n, distribution, mean/SD or median/IQR as appropriate |
| How engaged are respondents? | Six-item engagement score | Descriptive summary and scale checks | Valid n, distribution and scale reliability assessment |
| How are the scores associated? | Both scores for eligible respondents | Correlation; regression if model assumptions and sample permit | Association estimate, interval and transparent limitations |
Complete sample data analysis plan
“Survey responses will be exported to a restricted-access dataset after the collection window closes. Unique participant identifiers, duplicate submissions, response ranges, reverse-keyed engagement items and item-level missingness will be checked against a documented codebook. The primary analysis will include eligible respondents with valid usefulness and engagement scale scores according to predeclared scoring rules. Missingness will be summarised by item and compared across study-year groups; no absent item will be replaced with zero. Scale distributions will be inspected before choosing appropriate descriptive summaries. The primary research question will be evaluated using an association estimate between usefulness and engagement scores, with its confidence interval. Where the sample size and model diagnostics permit, a regression including study year will be reported as a secondary adjusted analysis. Assumptions and influential observations will be checked and any alternative method or exclusion will be documented. Findings will be presented as a sample-description table, score-summary table and association estimate. Because the survey is cross-sectional and observational, results will not be interpreted as proof that feedback usefulness causes engagement.”
Why the example works
Every stage corresponds to the question: the constructs are measurable, missing responses have a rule, a primary association is distinguished from a secondary adjusted model, and the final claim is deliberately limited. A student must still choose and justify scoring, sample size and specific statistical methods with their supervisor; “run a correlation” is not a substitute for considering variable distributions and design.
Worked example 2: Qualitative interview analysis plan
Study question and analytic orientation
Consider the question: “How do first-year university students describe deciding whether and how to use written feedback on later assignments?” A hypothetical study uses semi-structured interviews, anonymised transcripts and reflexive thematic analysis. The aim is an interpretive account of student experience, not a percentage estimate for all students.
Complete sample data analysis plan
“Following consent and approved transcription procedures, audio recordings will be transcribed and checked against the original recordings for accuracy. Direct identifiers will be removed from the analysis copy and participants will receive pseudonyms. The researcher will read each transcript in full, record analytic memos and code passages relevant to participants’ understanding, emotional responses and reported actions after receiving feedback. Codes will be iteratively reviewed across cases and developed into candidate themes that explain patterned meanings rather than simply list interview topics. Theme boundaries will be reconsidered against the entire dataset, including accounts that complicate the emerging interpretation. Reflexive notes will document the researcher’s assumptions and changes in analytic focus. Final findings will use carefully selected anonymised excerpts to substantiate each theme while protecting participants’ identities. The analysis will not treat the number of quotations supporting a theme as a population prevalence estimate.”
Why the example works—and what would be weaker
It describes actual analytic operations from transcription to interpretive reporting. Merely stating “interviews will be coded into themes” would omit what is coded, how ideas are refined, how differing accounts are handled and what supports the conclusions. Depending on the chosen qualitative tradition, practices such as independent coder agreement may or may not be appropriate; avoid treating one tradition’s criteria as universal.
If interview design is still unsettled, consult our qualitative interview guide. Questions and analysis must be designed together.
Worked example 3: Mixed-methods plan with meaningful integration
Research question and sequence
A hypothetical campus support study asks whether students’ satisfaction ratings differ between two service formats and how students explain those ratings. An explanatory sequential design begins with a structured survey and then uses interviews to investigate notable survey patterns. The integration question is: “How do interview accounts help explain the quantitative comparison?”
| Phase | Evidence | Planned analysis | Connection |
|---|---|---|---|
| 1. Quantitative | Satisfaction score and format used | Descriptive summaries; justified group comparison and uncertainty | Identify patterns and cases for follow-up |
| 2. Qualitative | Interviews with purposively selected survey respondents | Systematic coding and thematic interpretation | Explore reasons behind high, low or unexpected survey responses |
| 3. Integration | Survey contrasts plus interview themes | Joint display and narrative comparison | Assess convergence, tension and added explanation |
Complete sample data analysis plan
“The survey dataset will be checked for duplicate entries, eligibility and missing responses before group summaries and a prespecified comparison of satisfaction scores by support format. The method of comparison will be selected to match the observed measurement scale and study design, with its assumptions and uncertainty reported. After preliminary survey analysis, consenting respondents representing contrasting satisfaction patterns will be invited to interviews, subject to ethics approval and documented sampling criteria. Interview transcripts will be analysed thematically to explore how access, expectations and staff interactions shape reported satisfaction. A joint display will align the quantitative group findings with the qualitative themes for each service format. The interpretation will identify where the strands converge, where they diverge and where interviews suggest explanations that the survey cannot test. Any post hoc analyses prompted by interview findings will be labelled exploratory. Because formats are not randomly assigned, differences in satisfaction will not be described as treatment effects.”
Why integration matters
This plan specifies what the survey does, what the interviews add, how participants are selected between phases and how the evidence is combined. Conducting a survey and some interviews without explaining their connection is not enough to justify a mixed-methods analysis.
A weak plan versus a stronger plan
Weak: “The data will be analysed using SPSS. Tables, percentages and tests will be used. Interview data will be grouped into themes.”
Why weak: it names software and generic outputs, but identifies no unit, question-to-method mapping, score construction, eligibility, missing-data handling, interpretation rule or integration logic.
Stronger: “Eligible survey responses will first be checked for duplicates, allowable score ranges and missing items. A prespecified association analysis will address the feedback-usefulness research question; the estimate and confidence interval will be reported, along with diagnostics and limitations. Interviews will be examined using a named and justified qualitative approach, with a documented progression from transcript review to themes. An integrated discussion will compare interview explanations with the survey findings, including disagreement.”
Why stronger: a reader can trace each analysis from a question through evidence, technique and conclusion. A dissertation-ready version would add the exact measures, sampling rationale, detailed analysis rules and ethics arrangements.
Adaptable data analysis plan template
“To address research question [number and exact wording], this study will analyse [unit, population and eligible dataset]. The primary outcome/phenomenon is [definition and measure/source]. Before analysis, the dataset will be checked for [specified quality checks], and missing observations will be handled according to [defensible rule and justification]. The primary analysis will use [method] because [fit with question and design]. Its main assumptions or quality considerations are [list] and any departures will be documented. Findings will be reported through [estimates/themes/tables/figures] with [uncertainty, supporting excerpts or integration details]. The results will support conclusions about [appropriate inference], but not [claims the design cannot establish].”
For mixed methods, add how the quantitative and qualitative strands are sampled, connected, combined and interpreted. For qualitative work, replace statistical assumptions with the analytic logic and quality practices justified by that approach.
Common mistakes—and how to avoid them
-
Choosing a test before refining the research question. Begin with the required inference, then choose an appropriate analysis.
-
Promising to “prove” causation using cross-sectional, convenience-sample data. State the inference that the design supports.
-
Treating software as methodology. Specify the rationale, not only the program.
-
Ignoring the level of observation. Repeated visits from one person cannot automatically be analysed as independent people.
-
Deleting missing or unusual data without predeclared reasons. Describe the decision and its consequences.
-
Using dozens of exploratory tests while presenting only favourable findings. Separate primary, secondary and exploratory analyses.
-
Stating every statistical assumption for every test. Specify only those relevant to the proposed method.
-
Confusing themes with interview headings. Describe how interpretations arise from coded evidence.
-
Calling parallel surveys and interviews “mixed methods” without a documented integration point.
-
Writing the final results in the future tense, or inventing findings in the proposal. Keep plans distinct from completed evidence.
Frequently asked questions
How long should a dissertation data analysis plan be?
There is no universal word count. A proposal may need a few clear paragraphs; a complex mixed-methods dissertation may need multiple subsections or an analysis matrix. Follow your department’s guidance, and prioritise enough detail for an examiner to understand and evaluate the planned work.
Do I have to name SPSS, R, NVivo or another software package?
Name the tool if your institution requires it or if software capabilities matter, but explain analytical procedures independently. Different programs can implement the same defensible method.
Can my analysis plan change after data collection?
Yes, when a justified issue arises. Preserve the original plan, document the reason and distinguish planned from unplanned analysis. Ethics approval may be required for material changes to data use or participant procedures.
What if my data do not meet a statistical test’s assumptions?
Check whether the problem is a coding error, small sample, model misspecification or a substantive feature of the data. Choose and justify a suitable response rather than automatically deleting cases or mechanically transforming values.
Is an analysis plan required for qualitative dissertations?
Yes in substance, even if the label differs. Explain the chosen analytic tradition, how materials are prepared and interpreted, how decisions are recorded, and how claims will be supported.
What is the difference between a methodology section and an analysis plan?
Methodology justifies the overall research design and approach. The analysis plan specifies what you will do with the evidence to answer each research question. In some dissertations, the plan appears as a subsection of the methodology chapter.
Final data analysis plan checklist
-
Every research question maps to one or more analysis steps.
-
The unit of analysis and eligible sample are explicit.
-
Variables, scoring rules or qualitative source types are defined.
-
Primary and secondary analyses are distinguished.
-
Cleaning, missingness and exclusions have defensible rules.
-
Methods match the design and measurement level.
-
Model assumptions or qualitative analytic-quality practices are specified.
-
Confounders, clustering or repeated measures are addressed where relevant.
-
Tables, figures, quotations or integrated displays are planned.
-
Claims are limited to what the design can support.
-
Exploratory changes and analysis deviations will be disclosed.
-
Consent, confidentiality and secure handling remain protected.
Final takeaway
A strong data analysis plan is not a catalogue of tests. It is a series of defensible decisions connecting research questions, evidence, analytic methods and carefully limited conclusions. Write it early enough to improve your study design, and clearly enough that readers can see what was planned, what changed, and why.
If you would value help clarifying your dissertation structure, analysis-plan wording or research-method explanation, explore Acadelyra’s student-focused academic support. Use any guidance responsibly and follow your institution’s academic-integrity requirements.