Đại học Kinh tế Quốc dân – National Economics UniversityMICRO4ASIA
🏆 Arena
Sign inSign up
MICRO4ASIAThe six micro-credentials

Courses are delivered on the Limio learning platform. Limio

← Data Cleaning and Preprocessing with Python

You are viewing a public preview

Sign in to save your progress, do assignments, join discussions and get feedback.

Sign in
Module 3 — Prepare, Report and Reproduce

Period 14 · Interpreting and Reporting Results

Print / Save as PDF
Lecture video — Period 14: Interpreting and Reporting Results

Structuring a short report: the question and data covered, cleaning decisions that affect the answer, findings with evidence, limitations, and a final package check.

Lesson overview

In this period, we explain what its results mean. A useful report answers a business question, gives evidence, and states what the data cannot tell us. We will use a small sales example to practise this. By the end, you should be ready to complete a one-to-two-page report and check that all submitted files agree.

Study notes

By the end of this lesson you can:

  • Structure a short data report around a business question.
  • Turn a summary table into a finding with a period, measure, value and comparison.
  • Separate what the data show from possible explanations and unsupported claims.
  • Explain how cleaning changed the data, with counts that add up.
  • Write limitations and next steps that are specific and useful, and check that all files agree.

Key idea: A useful report answers a business question, gives evidence, and states clearly what the data cannot tell us. Every number must be traceable to the notebook.

Structure of a short report

The lecture suggests this order (a guide, not an extra grading rule):

  1. Question and data covered, for example: which region has the highest recorded revenue in January?
  2. Cleaning decisions that could affect the answer.
  3. Main findings, using a small table or chart.
  4. Limitations and a useful next step.

The report should be readable without running Python, but the reader must be able to find each supporting calculation in the notebook.

From a table to a finding

The lecture's teaching table (one row in the cleaned data = one transaction for one product category; revenue after discount):

RegionTransactionsRevenue (million VND)Share
North406050%
Central303025%
South303025%
Total100120100%

The share uses total revenue across all three regions as the denominator. Defining revenue as "after discount" means every reader uses the same measure.

A good finding: "North records the highest January revenue in this dataset, 60 million dong, or 50% of the total." It has a period (January), a measure (revenue after discount), a value (60 million, 50%) and a comparison (highest of three regions).

What we cannot say: that North has the highest profit (there are no cost data), or that North's staff are better (the data say nothing about causes). Keep observed results separate from possible explanations. A reasonable next step: compare costs and product mix before changing the budget.

Reporting the effect of cleaning

Explain how cleaning changed the data:

  • start with 108 rows;
  • remove 5 confirmed extra duplicate rows, leaving 103;
  • set aside 3 rows with invalid prices, leaving 100 valid transactions.

The two groups do not overlap. Keep the excluded records and reasons so someone can review the decisions, and report the final checks. A lower row count does not automatically mean better data, and the revenue total may omit real sales whose prices still need correction.

Useful limitations

A limitation is useful when it tells the reader how to interpret the result:

  • Only January data: the regional ranking may not hold for the whole year.
  • Three excluded records may be real sales: revenue may be incomplete.
  • No cost information: profit cannot be compared.
  • A regional difference does not prove its cause.

Avoid vague phrases such as "the data may have problems". Limits do not make the analysis useless; they show where its conclusion applies and what to collect next.

Worked example

Scenario. Write one report paragraph from the table above.

  1. Finding (period, measure, value, comparison): "In January, North recorded 60 million dong of revenue after discount, 50% of the 120 million total, twice the share of Central or South."
  2. Coverage: "The table uses 100 valid transactions; 5 duplicate rows were removed and 3 rows with invalid prices were set aside for review."
  3. Limitation: "The excluded transactions may affect the total, and cost data are unavailable, so we cannot compare profit."
  4. Next step: "Check the three excluded records and compare costs and product mix before changing the budget."

You can confirm the numbers in a few lines, which is also how the report's numbers should be traced to the notebook:

import pandas as pd

by_region = pd.DataFrame({"transactions": [40, 30, 30],
                          "revenue_m": [60, 30, 30]},
                         index=["North", "Central", "South"])
by_region["share_pct"] = by_region["revenue_m"] / by_region["revenue_m"].sum() * 100
assert by_region["share_pct"].sum() == 100
assert 108 - 5 - 3 == by_region["transactions"].sum()
print(by_region)

Expected output: shares of 50.0, 25.0 and 25.0, and both checks pass (100 transactions, matching 108 - 5 - 3). In your own report, replace the example values with your notebook results.

Checking the package before submission

Include the notebook, cleaned data, quality table and short report. Run the notebook from the beginning and compare its final values with the report; check dates and units, not only totals. State your sources, any AI assistance, and each member's contribution for group work. Keep the raw data unchanged.

Common mistakes

  • Claiming more than the data show. Revenue share is not profit, and a difference is not a cause.
  • Writing findings without a comparison or period. "North did well" is not a finding; "North had 50% of January revenue" is.
  • Hiding the cleaning. Readers need to know how many rows were removed or set aside, and why.
  • Vague limitations. Name the specific gap and how it affects the conclusion.

Summary

  • A report states the question, data coverage, cleaning decisions, findings, limitations and a next step.
  • A finding gives period, measure, value and comparison, with a clear denominator.
  • Separate observations from explanations; do not infer profit or causes without data.
  • Report cleaning counts that add up, and keep excluded records available for review.
  • Rerun the notebook and check that every number in the report matches the files.

Flashcards · Interpreting and Reporting Results

8 words
WordMeaningExample
  • Finding

    A statement of what the data show, including period, measure, value and comparison.

    North had 50% of January revenue after discount.

  • Denominator

    The total used to calculate a share or rate; it must be stated so readers interpret the percentage correctly.

    60 / 120 million = 50% of all-region revenue.

  • Data coverage

    The period, records and definitions that a result is based on.

    100 valid January transactions; revenue after discount.

  • Limitation

    A specific gap in the data or method that affects how a finding should be interpreted.

    No cost data, so profit cannot be compared.

  • Next step

    A concrete action that would address a limitation or test an explanation.

    Check 3 excluded records and compare costs.

  • Observation vs explanation

    Keeping what the data show separate from possible reasons that still need evidence.

    North sold more (observed); better staff (unproven).

  • Cleaning summary

    Counts showing how the data changed from input rows to analysed rows, with reasons.

    108 - 5 duplicates - 3 invalid prices = 100

  • Traceability

    Every number in a report can be found and reproduced in the notebook.

    Rerun the notebook and compare totals with the report.