Written and maintained by CASRAI Editorial Board
Published
On this page
The data analysis process is the sequence of work that takes a research question to a reported result: define what is needed, gather and understand the data, analyse or model it, and communicate the findings. This guide is for researchers who need a defensible order of steps, and it adds two things most process explainers omit: where the modelling step sits under different statistical traditions, and what to fix before any data is seen.
What the data analysis process is
Statistics Canada’s data-literacy introduction, An Introduction to the Data Journey (released 2026-07-09, modified 2026-10-02; read 2026-11-08), describes four stages: Define, Find, and Gather; Explore, Clean, and Describe; Analyze and Model; and Tell the Story. It opens the first stage with “Every data journey starts with a need” and closes with “The final step is about communicating findings clearly, accurately, and responsibly.”
The same material is often taught as a list of six steps. The mapping below is an editorial cross-walk, not a quotation from either source.
The steps in order
| Common six-step list | Statistics Canada stage | What the researcher produces | CASRAI guide |
|---|---|---|---|
| 1. Define the question | Define, Find, and Gather | A written question and the data it needs | Pre-analysis plan |
| 2. Collect or obtain the data | Define, Find, and Gather | A dataset with known provenance | Secondary data analysis |
| 3. Clean the data | Explore, Clean, and Describe | A cleaned dataset and a record of changes | Data cleaning steps |
| 4. Explore and describe | Explore, Clean, and Describe | Summaries and plots | Exploratory data analysis in Python |
| 5. Analyse and model | Analyze and Model | Estimates, tests, fitted models | Python for data analysis |
| 6. Interpret and communicate | Tell the Story | A report that states findings accurately | – |
Statistics Canada’s own summary of the middle stages is that after you have the data, “you need to understand it before you use it”, and that analysis and modelling “is where data begins to answer the question.”
Process for data analysis, process of analysis of data
These phrasings name the same thing. In every version the order is a default, not a rule: exploring the data often sends you back to cleaning or to the question itself.
Where the model comes in: classical, EDA and Bayesian
The NIST/SEMATECH e-Handbook of Statistical Methods (section 1.1.2, read 2026-11-08) shows that statistical traditions differ in when a model is introduced. It describes exploratory data analysis (EDA), in section 1.1.1, as “an approach/philosophy for data analysis” and “not a set of techniques, but an attitude/philosophy about how a data analysis should be carried out.” EDA relies mostly on graphical techniques and postpones the usual assumptions about what model the data follow.
| Approach | Sequence (NIST wording) | Where the model step falls |
|---|---|---|
| Classical | Problem => Data => Model => Analysis => Conclusions | Before analysis; analysis, estimation and testing are focused on the parameters of that model |
| EDA | Problem => Data => Analysis => Model => Conclusions | After analysis of the data |
| Bayesian | Problem => Data => Model => Prior Distribution => Analysis => Conclusions | Before analysis, followed by a prior distribution |
Practical reading (editorial interpretation): if your field expects a pre-specified model, your step 4 above is description only; if you are exploring, step 4 is where the model is chosen.
Plan before the data
In confirmatory research, analysis choices are written down before the data are examined. A pre-analysis plan and a statistical analysis plan (SAP) are the usual records; see the statistical analysis plan template for the contents. Exploratory work should be reported as exploratory, so readers can tell which results were planned.
Tools for each stage
Cleaning and exploration are covered in the guides linked in the table above. For choosing a package for the analysis step, see SAS vs SPSS for research analysis; for text and interview data, see how to analyze qualitative data.
FAQ
What are the steps of the data analysis process?
Define the question, obtain the data, clean it, explore and describe it, analyse or model it, and communicate the results. Statistics Canada groups these into four stages: Define, Find, and Gather; Explore, Clean, and Describe; Analyze and Model; Tell the Story.
Is the process linear?
No. The order is a default; findings at one stage commonly send you back to an earlier one. NIST’s EDA sequence, which places modelling after analysis, is itself evidence that the order varies by approach.
When should the model be chosen?
Under NIST’s description, before analysis in classical and Bayesian work, and after analysis in EDA.
Sources
- Statistics Canada, An Introduction to the Data Journey, read 2026-11-08.
- NIST/SEMATECH e-Handbook of Statistical Methods, 1.1.1 What is EDA? and 1.1.2 How does EDA differ from classical and Bayesian analysis?, read 2026-11-08.
