Understand the feedback before you act on it.

A chart becomes useful when the reader can see what changed, how large the change was, why it matters, and what the evidence cannot establish. Use this sequence: Describe → quantify → explain → qualify.

1. Describe the visible pattern

Name the variables, period, groups and direction before offering a reason. “Sales increased” is too vague. “Monthly online sales increased from January to April while store sales remained broadly stable” gives the reader an observable pattern.

2. Quantify the important difference

Use one or two numbers that establish scale. Report units and a meaningful comparison: change from baseline, difference between groups, peak, trough or rate. Do not copy every value from the chart into prose.

3. Explain the decision relevance

Connect the pattern to the assignment question. A useful interpretation states what the result changes for a manager, investor or policy maker. If the chart does not identify a cause, use language such as “is consistent with”, “may reflect” or “suggests”, then name the evidence needed to test the explanation.

4. Qualify the conclusion

State the main limit that affects interpretation: a short sample, missing comparison group, unusual period, aggregation, self-reported data or an uncontrolled factor.

Common mistakes

• Repeating the title without adding scale or meaning.
• Calling correlation a cause.
• Listing values without selecting the decision-relevant comparison.
• Adding a generic limitation that does not affect the conclusion.

A recommendation is defensible when the reader can trace it through Finding → stakeholder → action → expected effect. Skipping one link produces an action that may sound reasonable but is not supported by the report.

Start with the finding

Select a result that is material to the decision. State its magnitude, affected group and uncertainty. A recommendation based only on background literature or a general industry trend may not answer the case you analysed.

Name who must act

Identify the role or team with authority and capability. “The company should improve service” hides ownership. “The customer operations manager should revise the first-response process” makes implementation testable.

Specify the action and mechanism

Explain what changes and why it should affect the measured outcome. The mechanism should follow from your analysis rather than appearing for the first time in the conclusion.

Define the expected effect and review point

State a measurable indicator, a reasonable review period and a condition that would change the recommendation. Avoid inventing a precise benefit when the analysis cannot estimate one.

Common mistakes

• Offering a standard textbook action unrelated to the reported findings.
• Using “increase”, “improve” or “optimise” without specifying how.
• Ignoring cost, feasibility, risk or stakeholder constraints.
• Claiming a guaranteed result from observational evidence.

Decision relevance beats output volume. A report should provide enough evidence to understand the model, evaluate its credibility and answer the question. It should not reproduce the entire software console.

Report the model context first

Define the outcome, explanatory variables, sample and model form. State reference categories, transformations and interaction terms where they affect interpretation. Readers cannot interpret a coefficient without knowing what is held constant and what comparison it represents.

Select coefficients that answer the question

For each important coefficient, report the estimate, unit, uncertainty and practical meaning. A p-value alone does not communicate magnitude. If the coefficient is logged, standardised or part of an interaction, translate it carefully rather than treating it as a simple one-unit effect.

Show model performance and diagnostics

Include fit measures appropriate to the model and evidence relevant to its assumptions. Residual patterns, influential observations, multicollinearity and out-of-sample performance can matter more than an additional decimal place in the coefficient table.

Separate association from causation

Regression adjustment does not automatically remove omitted variables, reverse causality or selection bias. State whether the model is explanatory, predictive or causal, and keep the recommendation within that boundary.

Usually leave out

• Every candidate model tried during exploration.
• Unformatted console output and unexplained variable codes.
• Long significance discussions with no effect-size interpretation.
• Claims that a high R-squared proves the model is correct.

Task coverage and final alignment should be checked separately from proofreading. A polished report can still lose substantial marks when it omits a subquestion, required output or rubric criterion.

1. Build a requirement map

List every numbered task, subquestion and required deliverable. Record where each one is answered. Do not assume a section title proves coverage; identify the paragraph, table, figure or appendix item that supplies the evidence.

2. Reconcile evidence across the report

Check that numbers in prose match tables and figures, figure titles match the discussion, and recommendations follow from reported findings. Confirm that the final document uses the same sample period, units, group names and scenario definitions throughout.

3. Check limitations and uncertainty

Include limitations that change how the result should be used. A generic sentence about “limited data” is weaker than explaining how the sample period, missing variables or validation design affects confidence in the decision.

4. Check citations and presentation rules

Match every in-text citation to a reference entry and review every reference entry for use in the text. Then check word count, file type, naming convention, headings, table and figure labels, appendix rules and any AI-use statement required by the brief.

Do not rely on

• A spelling checker to detect unanswered tasks.
• Automatic citation formatting to verify source identity.
• A high word count as evidence of analytical depth.
• The editable document alone; inspect the submitted format.

Use a simple mapping: Criterion → visible evidence → location. This prevents the rubric from becoming a list you read once and forget while drafting.

Translate the criterion into an observable test

Words such as “critical”, “appropriate”, “integrated” and “insightful” are not tasks by themselves. Ask what a marker would need to see. “Critical analysis” may require comparison, limitations and implications. “Appropriate method” may require assumptions, alternatives and validation evidence.

Identify the evidence form

Decide whether the criterion needs prose, a calculation, table, figure, citation, diagnostic or recommendation. A method criterion cannot be satisfied only by naming the method; the report needs a rationale and evidence that it was applied correctly.

Assign a location before drafting

Place each item in a section and avoid relying on one paragraph to satisfy unrelated criteria. Weighted criteria deserve proportionate space and evidence, but word allocation should still follow the actual assignment tasks.

Common mistakes

• Copying rubric words into headings without supplying evidence.
• Treating every criterion as equally weighted when marks differ.
• Leaving evaluation until the conclusion.
• Using one citation as the only evidence for a major analytical claim.