Start with the question and the reader
Every analysis begins with a business question, and every report ends with an answer. Before you open the data, write one sentence for each of these: what decision will this support, who will read the report and what would change their mind. A report that opens with here is the data I have rarely ends well. A report that opens with can we cut delivery cost without losing customers? tells you what to look for and what to ignore.
Write for a manager who is intelligent but not a statistician. Methods belong in the report, but the headline findings should be readable without them.
A standard structure
| Section | What it contains | Share of the report |
|---|---|---|
| Title and executive summary | The question, the main findings with numbers and the recommendation, on one page | 5 to 10 percent |
| Background and objectives | Business context, the decision at stake and the specific questions | 5 to 10 percent |
| Data | Source, period, variables, sample size and how the data were cleaned | 10 to 15 percent |
| Method | The techniques used and why they fit the question | 5 to 10 percent |
| Results | Tables and charts with short interpretation, one finding per section | 30 to 40 percent |
| Discussion and limitations | What the results mean, what could be wrong, what was not tested | 10 to 15 percent |
| Recommendations | Specific actions linked to findings | 5 to 10 percent |
| Appendices | Full output, code or formulas, data dictionary | Not counted |
See our guide to executive summaries and business report format for the front and layout.
Describe and clean the data honestly
Readers need to know what you analyzed. Describe where the data came from, the time period, how many records you started with and how many you used. Then explain what you changed.
| Issue | What to do | What to report |
|---|---|---|
| Missing values | Decide whether to remove, fill with a sensible value or analyze separately | How many were missing, and which method you chose and why |
| Duplicates | Remove exact duplicates | How many were removed |
| Outliers | Check whether they are errors or genuine extremes before removing them | Which values, and the reason you kept or removed them |
| Inconsistent formats | Standardize dates, units, spellings and categories | The rules applied |
| Impossible values | Such as negative quantities or future dates; investigate and correct or remove | The count and treatment |
A short sentence such as of 5,214 order records, 112 duplicates and 38 records with missing delivery dates were removed, leaving 5,064 for analysis, builds trust. Never silently drop data. Keep a data dictionary in the appendix describing each variable, its unit and its meaning.
Match the method to the question
| If the question is... | Typical method | Reported as |
|---|---|---|
| What does the data look like? | Descriptive statistics and charts | Means, medians, spreads, distribution plots |
| Is there a difference between groups? | Hypothesis tests | Test statistic, p-value and effect size |
| Are two things related? | Correlation or cross-tabulation | Correlation coefficient or chi-square result |
| What drives an outcome? | Regression | Coefficients, R-squared and significance |
| What will happen next? | Forecasting | Forecast values and accuracy measures |
| How can we group customers? | Segmentation or clustering | Segment profiles and sizes |
State why you chose the method, and state its main assumptions. See our guides on hypothesis testing and regression.
Choose charts that answer the question
| You want to show | Best chart | Avoid |
|---|---|---|
| Change over time | Line chart | Pie charts or 3D bars |
| Comparison across categories | Bar chart, sorted by size | Too many categories on one pie |
| Share of a whole (few parts) | Stacked bar or a simple pie with three to five slices | Pies with many slices |
| Distribution of one variable | Histogram or box plot | Bar charts with arbitrary bins |
| Relationship between two variables | Scatter plot, with a trend line if useful | Lines connecting unrelated points |
| Exact values | A table | A chart with unreadable labels |
- One message per chart Write the message as the title, such as Online orders grew 18 percent while store sales fell 4 percent.
- Label everything Axes, units, legend and source.
- Start bar axes at zero Truncated axes exaggerate differences.
- Cut decoration Remove gridlines, shadows and 3D effects that do not help.
- Refer to every chart in the text Say what to notice before the chart appears.
Working on this assignment now? Get a price for help with your paper.
Get an instant quoteTranslate statistics into plain English
| Statistical statement | Business wording |
|---|---|
| The difference is significant at the 5 percent level (p = 0.026) | It is unlikely that the difference is just luck, and we can reasonably act on it |
| The slope coefficient is 4.11 (p < .001) | Each extra $1,000 of advertising was associated with about $4,100 more in sales |
| R-squared is 0.62 | The model explains about 62 percent of the variation in sales; the rest comes from other factors |
| The confidence interval is $0.40 to $6.00 | The true improvement is likely to be somewhere between 40 cents and six dollars per customer |
| We fail to reject the null hypothesis | The data do not give strong evidence of a difference |
Always add what it means for the decision. The number is not the finding. The finding is what the number says about the question.
A worked results paragraph
Results write-up (hypothetical)
Finding 1: The new layout increases spending. Customers in stores with the new layout spent an average of $48.20 (n = 40), compared with $45.00 in stores with the old layout (n = 45). A two-sample t-test indicates the difference of $3.20 is statistically significant (t = 2.27, p = .026). A 95 percent confidence interval suggests the true increase lies between about $0.40 and $6.00 per visit. The effect is modest but, at current traffic, would be worth roughly $70,000 a year across the chain.
Limitation: stores were not randomly assigned, so differences in location or customer mix could account for part of the effect.
Recommendation: extend the new layout to ten more stores selected at random, and repeat the test before a chain-wide rollout.
The figures are invented. Notice that the paragraph leads with the finding, supports it with the statistic, quantifies the likely range, says what it is worth, notes a limitation and ends with an action.
A model findings paragraph
Reporting a result in plain language (hypothetical)
Customers who used the mobile app spent more per order than those who did not: $64 against $52 on average, a difference of $12 (23 percent). The difference is unlikely to be chance (p < .01), but app users also order more often and skew toward higher-income postal codes, so the app alone may not explain it. A test that matches users with similar non-users is the next step.
The paragraph gives the result, the size, the evidence, the caveat and the next step in four sentences.
Stating assumptions and limits
- Data source and period Where the data came from and what dates they cover.
- Cleaning decisions What you removed or fixed and why.
- Missing data How much and how it was handled.
- Method choice Why this test or model suits the question.
- What the data cannot show Causation, other groups, other periods.
Limitations, recommendations and the appendix
Showing the limits of your analysis builds credibility. Typical limitations include a small or unrepresentative sample, a short time period, missing variables, data quality problems and the fact that observational data cannot prove causation. Say which apply and how they could affect the result. Do not hide behind them, though. After describing limits, say what you would do about each, such as collecting more data.
Recommendations should follow from findings, be specific and say who should do what. Put detailed output, code or spreadsheet formulas in an appendix so a reader can check your work. Our guides on Excel formulas and pivot tables and dashboards help with the analysis behind the report.
- A clear question and a clear answer The first page should tell the reader both.
- Honest data description Sources, size, cleaning steps and a data dictionary.
- A method that fits State why, and state its assumptions.
- Readable charts One message each, labeled, and referred to in the text.
- Plain-English interpretation Say what the numbers mean for the decision.
- Limits and next steps Say what could be wrong and what to do about it.
If you want help with the analysis or the report, you can order a data analysis report and upload your data and brief.