Skip to content
PlagiatScanner.de
Research integrity · analysis

Honest data analysis: make analytical decisions traceable

Honest analysis preserves original material, separates planned work from exploration and explains every consequential change. The audit trail below does not record every click. It captures the decisions without which another person could not understand the result, its origins and its limits.

Page ID INT-04Sources checked 4 September 2026

Traceability reaches beyond technical reproducibility

A script may recreate an output perfectly while concealing a poor scientific choice. Conversely, a qualitative interpretation can be traceable even though it cannot be reproduced like a numerical calculation. The shared requirement is a defensible account: Which version of the material was used? What question guided the step? Which rule applied? What alternatives were considered, and how did the decision affect the conclusion?

The DFG Code of Conduct calls for documentation that makes findings and their development understandable. Proportionate form depends on discipline and method. A dissertation will often need no more than a structured journal linked to files, code, memos and the methods chapter. What matters is that retrospective changes remain visible and that cleaned material does not overwrite the only original state.

Audit trail for each consequential analytical step

  1. ID and date: Assign a short decision number, date it and name the responsible person.
  2. Starting state: Identify the immutable input file, dataset version, interview set or document corpus.
  3. Analytical question: State the sub-question this step is meant to answer.
  4. Prior rule: Link to the proposal, preregistration, codebook or analysis plan. If no prior rule exists, label the decision as new.
  5. Observation: Describe the trigger without judgement, such as missingness, ambiguous coding or a failed model assumption.
  6. Options: List realistic alternatives, including making no change.
  7. Decision and rationale: Give a methodological reason. “Produces a nicer result” is not a quality criterion.
  8. Implementation: Point to the code line, transformation file, memo or new data version.
  9. Effect: Where informative, compare the result before and after. Record when the interpretation does not change too.
  10. Reporting location: Note whether the choice will appear in methods, results, limitations or an appendix.

Compact example: “AT-07 · 18 August 2026 · dataset v02: two observations were outside the pre-defined instrument range. Both were coded missing under rule R3; direction and interpretation of analysis A were unchanged. Implementation: clean.R lines 44–51; report: Methods 3.4.”

Qualitative work: retain the path of interpretation

For interviews, observations and document analysis, integrity does not mean pretending that coding is mechanical. Preserve source material separately from transcripts, version the codebook and write memos when categories emerge, edge cases arise or interpretations are rejected. If one passage changes a category, record the revision and review earlier cases. Choosing vivid quotations must not make contrary passages disappear.

Reflexivity belongs in the trail. Record aspects of role, assumptions or relationship to the field that may shape interpretation. This is methodological context rather than personal confession. Where a team codes material, do not reduce every disagreement to a coefficient. Note how concepts were clarified and rules revised. Anonymisation, consent and access controls will properly limit which raw records can be shared or appended.

Quantitative work: control analytical freedom

Keep raw data read-only and conduct cleaning and analysis through readable code or clearly recorded steps. Define variables, exclusions, outlier rules, transformations and the main analysis before seeing the target result whenever the design allows. Data-prompted analysis can be informative when labelled as exploratory. The integrity problem arises when many variants are tried but only a favourable one is reported as though it had been the original plan.

Decisions that should remain visible in an analysis trail
DecisionMinimum recordTransparent report
Case exclusionRule, affected IDs, dateGive number and reason
Missing valuesPattern, treatment and softwareExplain effect and uncertainty
Model changeTrigger, previous and new modelLabel as planned or exploratory
Multiple analysesList of material variantsDo not hide outcome-based selection
Figure processingTransformation and original imageShow axes and processing clearly

What an audit trail should leave out

Do not add unnecessary personal data, passwords, credentials or copies of protected material. A decision number can point to a controlled location without exposing confidential records. Nor is an unannotated screen recording of the entire project helpful. It creates volume without explanation. A good trail is selective about routine operations but complete about choices that can change the scientific meaning.

Apply a retention logic early. Decide which evidence must remain available, who may access it, how versions will be named and when lawful deletion is required. These controls are not barriers to openness. They allow appropriate transparency while respecting participants, rights holders and institutional responsibilities.

Translate the private trail into an honest methods section

The internal record may be longer than the submitted work, but every decision affecting sample, data quality, analysis or scope belongs in the account available to readers. Distinguish a planned primary analysis, a reasoned deviation and an exploratory addition. Report robust and conflicting findings in a way that permits uncertainty to be judged. A precise limitation is not a weakness when it follows from the evidence and design.

Before submission, perform a reverse trace. Take each central result statement back to its output, code or analytical memo, then to the correct data version and recorded decision. If a link is missing, restore it or narrow the statement. The audit trail then becomes the working bridge between day-to-day analysis and a faithful research report.

Important: Retention, data protection, ethics review and consent can constrain documentation. Follow the rules and permissions governing the particular project rather than publishing raw evidence indiscriminately.

Technical foundations

  1. DFG Code of Conduct: Guidelines for Safeguarding Good Research Practice, including documentation.
  2. ALLEA: European Code of Conduct for Research Integrity, Revised Edition 2023, official edition.
  3. UK Data Service: Research data management, institutional guidance.