Statistical analysis for a thesis: a reproducible workflow

Choose an analysis from the research question, design and measurement process rather than from a preferred software package. Define the population, outcome, comparison and estimand before examining results. Identify which conclusions the sampling and study design can support.

Prepare a data dictionary and retain the original dataset securely. Record units, coding conventions, missing-value rules and exclusion criteria. Separate identifiers from analysis data and follow the approved ethics and data-management arrangements. Never place participant information in public examples or an open repository.

Describe the data before applying a model. Check distributions, unusual values and the amount and pattern of missingness. A software warning should lead to investigation; it is not a reason to delete inconvenient observations. Document justified corrections and keep a reproducible trail.

Select methods and diagnostics appropriate to the outcome and design. State assumptions and assess them using relevant evidence. Report effect estimates and uncertainty, alongside any hypothesis tests. A small p-value alone does not establish practical importance, causation or a well-designed study.

In R, keep scripts and package versions; in SPSS, save syntax as well as output. Link tables and figures to the documented analysis rather than editing results manually. Distinguish planned analyses from exploratory checks and describe multiple comparisons and sensitivity analyses honestly.

The thesis should explain what was done well enough for an informed reader to assess it. Seek qualified methodological review when the design is complex or consequences are important. Academic support should strengthen your own understanding and research integrity, not fabricate data, results or authorship.