Checking Your Numbers and Writing Them Up

The analysis is done and the chart looks good. Now comes the part that decides whether anyone trusts you next time: checking every number, then writing them so a tired reader on a phone gets the point.

The number audit

Print the draft or open it beside your spreadsheet. Go through it one figure at a time. For each number, write where it came from: a cell (Pivot!C4), a page of a PDF, a person's quote. Then recompute or re-read it.

Number in story	Source	Recomputed	OK?
"spending rose 8 percent"	Budget sheet, row 6, column E	(6,459,000 minus 5,975,000) / 5,975,000 = 8.1%	Yes
"Acme Paving received 70 percent"	Vendor pivot, C2 / C7	182,350 / 260,505 = 70.0%	Yes
"three of four payments"	Sheriff filter	Blue Ridge Fuel: 3 of 4 Sheriff payments	Yes
"Roads, 76 percent of the money"	Department pivot	197,770 / 260,505 = 75.9%	Yes, rounds to 76

A few habits catch most mistakes:

• Recompute from the raw file. Not from your cleaned copy, not from your notes. If the raw file gives a different total, find out why.
• Match the agency's own totals. If your department totals don't match the audited report, you or they made a choice. Know which.
• Have someone else redo the key number without seeing your spreadsheet. If two people get the same answer independently, you can sleep.
• Read the numbers to the agency before publication. "My analysis shows Roads accounted for 76 percent of payments in this period. Does that match your records?" This isn't asking permission. It is giving them the chance to catch an error you missed, and if they don't respond, you can say you asked.
• Sanity check the size. If your per capita figure says every resident of a county of 9,000 spends $40,000 a year on roads, a decimal moved.

Rounding

Round at the end, never in the middle. Keep full precision in the spreadsheet and round only what goes into the copy.

Then round to what the data supports and what readers can hold:

• Money: $2,347,912 becomes "$2.3 million" in a sentence. Keep exact figures in a table if they matter.
• Percents: whole numbers in most copy. One decimal when differences are small, like unemployment rates. Never 23.4782 percent from a survey of 600 people with a 4-point margin. That is false precision.
• Totals that don't add: shares of 33.4, 33.4 and 33.2 percent round to 33, 33 and 33. Do not bump one to 34. Add "Figures may not total 100 because of rounding."
• Plain fractions: 24.8 percent can be "about one in four." 12.5 percent is "one in eight." Readers picture fractions better than percents.

How many numbers per paragraph

A good working rule: one number per sentence where you can, two at most, and no more than three or four in a paragraph. If you need more, build a table or a chart and let the prose carry the point.

Other writing habits that help:

• Lead with what the number means, then the number. "Most of the new money goes to roads" before "$360,000."
• Give a comparison readers can feel. "About $40 for every resident" lands harder than "$360,000" in a county of 9,000.
• Use the same unit throughout. Don't switch between per 1,000 and per 100,000 in one story.
• Attribute the data in the sentence. "according to county payment records obtained by" or "Census Bureau estimates show."
• Carry uncertainty in plain words. "About," "roughly," "statistically tied." Lesson 6 has the language.

Show your method

A method note tells readers exactly what you did, so they can trust it or redo it. Put a short one at the end of the story or in a box. Three to six sentences is usually enough.

That last sentence is the one that matters. It is also the one most newsrooms skip.

Publish the data

When you can, give readers the data. It shows you have nothing to hide, it lets other reporters and residents find stories you missed, and it invites corrections, which are better found by a reader before the next story than by an opponent after it.

What to publish:

• The cleaned data as CSV. CSV opens anywhere and outlives every spreadsheet format.
• A data dictionary. One line per column: name, meaning, units, source.
• Your data diary or a summary of cleaning steps.
• A link to the original source, or the raw file if it is not posted anywhere.

Where: a shared Google Sheet set to view-only, a file on your own site, or a public code repository. Datawrapper charts can show a "Get the data" link so readers download exactly what the chart shows.

Before publishing, strip what the story does not need. Public records can contain home addresses, birth dates, phone numbers and the names of private people, crime victims or minors. A record being public does not oblige you to amplify every field. Publish what supports the story and leaves private people alone.