Basic
$199/moBelow DA/DR targetNot a fitExcluded from recommendation
ORIGINAL RESEARCH & NEWSWORTHY CAMPAIGNS
SITETRAIL CERTIFIED
Your learning path: 12 lessons.
Module 02 / Lesson 05 · Read, apply, check
Create a reproducible cleaning record and verify counts, denominators and comparisons before drafting claims.
Preserve the authorized original export and create a versioned working copy. Record where it came from, when it was extracted and what each field means. Use a cleaning log to document duplicate handling, date standardization, exclusions and corrections. Avoid editing cells until the result looks plausible without retaining the reason for each change. If a value seems impossible, investigate its source rather than automatically deleting it. A timestamp may use another time zone or a different event definition. The aim is a traceable path from the source to the analysis, with enough information for a second reviewer to understand and reproduce the important decisions.
Count the starting rows, each mutually exclusive exclusion step and the final eligible set. Apply the documented order so that one row is not counted as both a duplicate and a missing-value exclusion. In the fictional RouteDesk exercise, two hundred and forty rows become two hundred eligible requests after twenty duplicates, ten out-of-period rows and ten remaining incomplete records are removed. Check that the counts reconcile and retain the exclusion reason. The denominator should match the question actually answered. A percentage of eligible requests cannot quietly be described as a percentage of all exported rows, unique buyers or every supplier in the industry.
Apply the agreed late-reply rule to the eligible records, then check a sample of classifications manually and reconcile totals. In this synthetic case, seventy-two of two hundred eligible requests are late, producing thirty-six percent. The remaining one hundred and twenty-eight are not late under that rule. A separate opt-in survey finds forty-eight of one hundred and twenty respondents reporting unclear expectations, or forty percent. Verify the division, rounding and labels, but also ask whether the calculation describes the intended concept. These figures are not a four-percentage-point increase because they use different measures and populations, not the same measure observed at two comparable times.
Check whether reasonable alternative handling choices materially change the story. For example, ten excluded records with missing timing information cannot simply be assumed on time. If those are otherwise eligible, the late proportion across the two hundred and ten records would range from about thirty-four to thirty-nine percent depending on their unknown classification. That descriptive range is not a confidence interval. It illustrates one missing-data uncertainty and does not address all selection or measurement problems. Explain the actual chosen analysis and relevant limitations, and have another reviewer check both the arithmetic and the wording before the figures move into a headline, chart or quotation.
Synthetic RouteDesk teaching data. Every exclusion remains visible.
On a narrow screen, scroll the chart sideways to inspect its labels. The explanation and supporting data follow below.
| Reconciliation step | Records |
|---|---|
| Raw export | 240 |
| Duplicate rows | −20 |
| Out-of-period rows | −10 |
| Remaining incomplete records | −10 |
| Final classifiable records | 200 |
72 ÷ 200 = 36%. If the ten missing classifications were added, the possible late proportion would range from 72/210 = 34.3% to 82/210 = 39.0%. This is a sensitivity range, not a confidence interval.
Separate perception survey: 48/120 = 40%. It measures something different; the four-point gap is not evidence of deterioration.
Keep the original export and field definitions.
Account for each exclusion and final denominator.
Check classifications, counts and labels.
Test sensitivity and review the wording independently.
RouteDesk’s analyst saves the raw export, applies the planned exclusion order and produces a reconciliation table showing 240 minus 20 minus 10 minus 10 equals 200. A second reviewer checks the late definition against selected records and confirms that 72 divided by 200 is 36 percent. They keep the separate survey’s 48 out of 120 result in its own table. During review, the account manager suggests calling the difference a four-point deterioration, but the analyst explains that the measures are not comparable time points. The report also states that ten records lacked classification data and shows how their possible outcomes could affect the broader eligible-record proportion. The agency now has a calculation record that answers likely journalist questions without implying precision or causation that the dataset cannot provide.
Another analyst calculates percentages directly from a mixed export, removes suspicious values without documenting them and counts the same exclusion twice. A designer then combines the audit and survey figures into a trend chart. The numbers look tidy, but the denominator changes between slides and the apparent trend never existed. These errors could have been caught by a reconciliation table and a second review of units and definitions before the campaign materials were designed.
Data cleaning is not permission to remove inconvenient observations. It is the application of defensible rules to make the material interpretable, with unresolved problems remaining visible. Nor does correct arithmetic make a conclusion valid. The denominator, unit, definition and comparison must all match the claim. A spreadsheet can calculate a misleading percentage perfectly when the analyst asks it the wrong question.
Do not replace missing observations with invented values merely to complete a report. Label any modeled or imputed analysis appropriately and involve qualified support where necessary. Keep synthetic practice figures separate from real findings, and preserve the authorized source, calculation version and review notes without distributing confidential records beyond the agreed access.
Do this today · 30 minutes
Reproduce the audit and survey percentages and write the cleaning reconciliation.
Use a document or spreadsheet and the fictional RouteDesk case. All course figures are synthetic teaching data. No purchase, confidential data, real survey recruitment or contact with journalists is required.
Before: A desired headline with an unclear evidence base, audience and release process.
After: A bounded, reproducible finding connected to a useful audience question and an accountable campaign plan.
Notes stay in this browser. Download a copy to keep them.
Ask a colleague to challenge the claim, denominator, permissions and audience fit. Record the revision, owner and review date before using the approach with a real client.
Before you move on
The cleaned audit contains seventy-two late replies among two hundred eligible requests, while a separate survey has forty-eight of one hundred and twenty respondents reporting unclear expectations. The designer wants a chart showing deterioration from thirty-six to forty percent. What should the analyst correct before release?
Next lesson →PRSA: professional ethics and disclosure
Lesson edition 1.0 · References reviewed 17 October 2026 · Practical exercise reviewed 17 October 2026. Examples are illustrative teaching scenarios.
Sitetrail service · NewsPass
Put your SEO, PR and generative engine optimization strategy into practice. NewsPass supports search visibility, media presence, brand credibility and discoverability in AI answers.
Compare the delivery costs for your campaign using the same calculator as the Sitetrail homepage. Choose a target authority range, your typical placement cost and expected monthly volume.
Illustrative cost comparison using October 2026 plan prices; check current terms before purchase. Assumes one eligible placement per credit and comparable selected opportunities, not a prediction of rankings, revenue, coverage or AI citations. Actual suitability depends on audience fit, editorial requirements and the current plan terms. DA/DR are third-party metrics, not Google scores.
Choose publications available on your plan, prepare useful content for the audience and submit it for editorial review. Track accepted placements in your dashboard and evaluate them against your campaign goals.
Useful questions. Defensible findings. Relevant stories.
Use the worksheets to turn each lesson into a real decision. Build a research-led campaign from a useful question, checked evidence, clear limits and relevant audience needs.
Course outline · Free assessment · Privacy
No account is required. Progress, first-attempt lesson answers, worksheets and assessment results are stored in this browser when storage is available. They do not sync across devices. Download copies to retain them; clearing browser data removes local records.
Your free trial starts automatically. No license key is needed yet.