Thesis Data Analysis Support

Who this is for

Most people arrive at one of a few points, and it is worth recognising which is yours, because they need different things.

• You have data and are not sure which analysis it supports. Usually the question needs sharpening before the test becomes obvious — and that is a conversation rather than a service.
• You have run something and do not trust it. Assumptions unchecked, output you cannot interpret, or a result that looks wrong. Common, and usually quick to resolve.
• Your supervisor has questioned the analysis and you need an independent read before responding.
• You are stuck between coded data and findings. The qualitative stall: thorough coding, no themes that say anything. A separate piece of work from the coding itself.
• You need a method nobody around you uses. Structural equation modelling, multilevel models, mixed methods integration — where your department has no one to ask.
• Your results are not what you hoped and you need to know whether the study can still say something defensible. It usually can, with a narrower claim.

What it covers

Quantitative. Choosing tests that suit your design and data, checking assumptions and deciding what to do when they fail, regression and comparison of groups, factor analysis and reliability, structural equation modelling including mediation and moderation, power and sample size, and interpreting output into claims your design actually supports.

Qualitative. Coding approaches and codebook development, moving from codes to themes that make a claim rather than name a topic, thematic and content analysis, framework analysis, and evidencing trustworthiness in a way that convinces rather than asserts.

Mixed methods. Integration — which is what these theses are actually assessed on — joint displays, and handling divergence between strands as a finding rather than a problem.

Across all of it: data preparation, missing data, reporting to your discipline’s standard, and the sentence that turns output into a finding.

Software is not the constraint. Work happens in whatever you are using — SPSS, R, Stata, Python, AMOS, SmartPLS, NVivo — and where a different tool would genuinely be better you will be told, along with whether it is worth switching at your stage.

How it works

1. You describe the problem. Your design, your data, what you are trying to establish, and where it has stalled. In whatever state it is in.
2. A first conversation, at no cost. This is diagnostic. Its purpose is to establish what the actual problem is, which sometimes means being told you do not need the service you asked for, or that the difficulty is one stage upstream of where it feels.
3. You are matched with a PhD in your own field. Methodological judgement does not transfer cleanly between disciplines — what counts as an adequate sample, an acceptable effect or a defensible design differs, and a generalist statistician will miss those conventions.
4. The work is done with you. Method explained and justified, assumptions checked in the open, output walked through until you can read it. You should finish able to answer questions about it.
5. You get it in a form you can use. Results reported to your discipline’s conventions, the reasoning written down, and whatever code or syntax was used so the analysis can be re-run.

If confidentiality matters, an NDA can be signed before any detail is discussed. Your data, your ideas and your authorship remain yours throughout.

What you receive

• The analysis itself, run correctly and documented.
• Results formatted for your discipline, rather than raw software output pasted into a document.
• The reasoning, written out: why this method, what assumptions were checked and what was found, and what the results are and are not entitled to claim.
• The syntax, script or project file, so the analysis is reproducible and you can re-run it if a supervisor asks.
• An explanation you can act on — walked through until you can read the output yourself and answer questions about it.
• Honest limitations, including what your design cannot support, which is generally better heard now than from an examiner.

What you should expect not to receive: written thesis chapters, or an analysis handed over as a finished product with no explanation. Neither serves you in a viva, which is where the analysis has to stand up.

When you probably do not need this

Worth saying, because paying for the wrong thing is a common waste.

If the problem is your research question, no analysis will fix it. A question too vague to imply a method produces an analysis that answers nothing, and sharpening the question is faster and cheaper.

If your design cannot support the claim, the honest answer is to narrow the claim rather than to analyse harder. A cross-sectional survey will not be made to demonstrate causation by any technique.

If you have not yet collected data, what you need is a design review, not analysis support. That is a different and much cheaper conversation, and it prevents most of the problems this page describes.

If your university offers a statistics clinic, use it. Many do, they are free, and for a straightforward question they may be all you need.