How to Verify the Source of an Advocacy Statistic

Advocacy statistics can make a campaign memorable, but a striking number is only as reliable as the evidence behind it. This guide explains how to trace a statistic to its original source, evaluate the research, identify misleading presentation, and document what you can and cannot verify.

Start with the exact claim

Before searching, rewrite the statistic as a complete claim. Record every detail that affects its meaning:

• The exact number or percentage.
• The population being described.
• The location or countries included.
• The date or time period.
• The subject being measured.
• The definition of the key term.
• Whether the number is an estimate, survey result, administrative count, or projection.
• Who published or repeated the claim.

For example, “one in four families cannot afford healthy food” is not sufficiently precise. A usable version might be: “In a 2024 survey of adults in a particular country, 25% reported that they sometimes lacked enough money for food.” Those two statements may refer to very different evidence.

Copy the wording exactly if possible. Screenshots, posters, social posts, and speeches often shorten a finding so much that important qualifications disappear. Save the page address, publication date, author, and the date you accessed it. If the claim appeared in a printed report, note the report title, page number, publisher, and edition.

Search for the earliest available source

The organization sharing a statistic is not necessarily the organization that produced it. An advocacy group may have obtained the number from a government dataset, academic paper, polling company, nonprofit report, or another campaign.

Use a progressive search process:

1. Put a distinctive phrase in quotation marks.
2. Search the exact number together with a key phrase.
3. Search the claimed percentage without punctuation, such as 25 percent and 25%.
4. Add likely source terms such as survey, report, dataset, methodology, or table.
5. Search distinctive fragments from the claim separately.
6. Look for citations, footnotes, hyperlinks, and references in the page where you found it.

Do not assume the first search result is the original. Search engines commonly rank later summaries, press releases, and reposts above older primary documents. Open several results and compare their publication dates and citations.

A useful source chain might look like this:

campaign webpage → nonprofit report → commissioned poll → questionnaire and technical appendix

Continue following the chain until you reach the organization that collected, calculated, or officially recorded the data. If the trail stops at an unattributed infographic or a post saying “research shows,” treat the statistic as unverified rather than filling the gap with an assumption.

Identify what kind of evidence supports it

Different source types answer different questions. A national census may provide broad population counts, while a small survey may reveal attitudes among a selected group. Neither should automatically be treated as proof of the other.

Source type	What it can show	Main limitation to check
Census or official administrative data	Recorded events, population characteristics, or service use	Missing records, changing definitions, and reporting delays
Probability sample survey	Estimates of opinions or behaviors in a defined population	Sampling error, nonresponse, and question wording
Convenience or online poll	Responses from people who chose or were able to participate	Selection bias and weak generalizability
Academic study	Results from a stated research design and analysis	Narrow sample, old data, or limited replication
Forecast or model	A projection under specified assumptions	Results change when assumptions or inputs change
Advocacy report	A focused synthesis or campaign analysis	Selective framing, unclear methods, or conflicts of interest

Ask whether the source actually measured the thing the advocacy statistic claims to measure. A survey asking whether respondents “worry about” an issue does not directly measure the number of people who experienced that issue. A budget report may show allocated funds, not money actually spent. A model may estimate future outcomes, not record present conditions.

Find the methodology, not just the headline

A credible report should explain how its number was produced. Look for a methodology section, technical appendix, data notes, codebook, or research protocol. At minimum, try to find:

• The sample size.
• The target population.
• How participants or records were selected.
• The fieldwork dates.
• The response rate, if it was a survey.
• The wording and order of questions.
• The definition of each measured term.
• How missing responses were handled.
• Whether weighting or adjustment was applied.
• The margin of error or uncertainty interval, when appropriate.
• Who funded or commissioned the work.
• The analysis date and version of the data.

A large sample is not automatically representative. Ten thousand self-selected respondents can produce a less reliable population estimate than a carefully designed sample of one thousand. Conversely, a small study may be useful for describing a specific group while being unsuitable for claims about everyone.

For survey statistics, inspect the question itself. Small wording changes can produce substantially different answers. Compare “Have you ever experienced housing insecurity?” with “Have you been homeless in the past 12 months?” They do not measure the same condition.

For administrative statistics, check coverage. Does the database include every relevant person, or only people who used a particular service, reported an event, or met an eligibility rule? A rise in recorded cases might mean the underlying problem increased, reporting improved, or the definition changed.

For modeled estimates, list the assumptions. Ask which inputs are measured directly and which are inferred. Find out whether the estimate is a single central value, a range, or one scenario among several.

Check the date, definitions, and denominator

Many misleading statistics are technically based on real data but presented without necessary context. Verify the following details.

First, check the date. A source may have been published recently using data collected years earlier. A current webpage can therefore contain an old estimate. Compare the collection period, publication date, and any revision date.

Second, check the denominator. “Thirty percent of cases” could mean 30% of all residents, 30% of reported cases, 30% of survey respondents, or 30% of people in a smaller subgroup. These denominators are not interchangeable.

Third, check the unit. Dollars, households, individuals, incidents, percentage points, and percentages describe different quantities. An increase from 10% to 15% is a five-percentage-point increase but a 50% relative increase.

Fourth, check the scope. A result from one city, age group, workplace, or online community should not be silently generalized to a country or the entire population.

Fifth, check the definition. Terms such as poverty, violence, access, disability, unemployment, discrimination, and food insecurity may have formal definitions that differ from ordinary language. Use the source’s definition when describing the result.

Compare the statistic with independent evidence

Verification does not require finding an identical number elsewhere. It means assessing whether the claim is consistent with other relevant evidence and whether differences can be explained.

Search for independent sources from different institutions. Prefer sources that do not simply cite one another. For example, compare an advocacy report with official statistics, a peer-reviewed study, a transparent survey, or a reputable research data archive.

When numbers differ, investigate instead of choosing the most dramatic figure. Differences may result from:

• Different years or seasons.
• Different geographic areas.
• Different age or income groups.
• Different definitions.
• Different sampling methods.
• Different denominators.
• Revised or corrected data.
• A count compared with an estimate.

Check whether a source is genuinely independent. Five articles may repeat the same press release and still represent one underlying claim. Trace citations backward and identify the distinct dataset or study behind each statement.

If independent evidence broadly agrees, your confidence can increase, but agreement alone does not prove correctness. Several sources may share the same error. If evidence conflicts, report the disagreement and explain the methodological reason where you can.

Test the citation and the calculation

Open the cited source rather than relying on the citation label. Confirm that it contains the claimed number, supports the same population and period, and has not been quoted from a different section or table.

If the number is calculated from a table, reproduce the calculation. Write down the numerator and denominator, then check the arithmetic. For example:

percentage = relevant cases ÷ total cases × 100

Look for rounding. A report may state 26% even though the underlying calculation is 25.6%. If several subgroup percentages are added together, check whether categories overlap. If categories are mutually exclusive, their totals may approach 100%; if people can belong to multiple categories, they may exceed 100%.

Check whether the author reports a raw count, a rate, or a weighted estimate. Rates often use a population base such as “per 100,000 people.” Comparing raw counts between places with different population sizes can create a false impression.

For charts, inspect the axis. A truncated vertical axis can make a modest difference look enormous. Also check whether the chart combines values from different years or uses a moving average. If the underlying data are downloadable, save a copy and record the file version or access date.

Evaluate the publisher and possible incentives

Source credibility is not determined by reputation alone, but the publisher’s role and incentives matter. Ask:

• Does the organization describe its methods openly?
• Does it identify authors, funders, and data sources?
• Does it correct errors publicly?
• Does it distinguish findings from recommendations?
• Does it publish inconvenient results as well as favorable ones?
• Is the statistic being used to raise money, influence policy, sell a service, or support a legal position?

An advocacy organization can produce careful research. Its advocacy role is a reason to inspect the evidence, not automatic proof that the statistic is false. Apply the same standards to government agencies, companies, universities, and campaign groups.

Look for conflicts of interest, but do not treat funding as a substitute for methodological evaluation. A funded study may still be rigorous, while an unfunded claim may still be unsupported.

Use an evidence log

Create a short record for every statistic you plan to repeat. This prevents link loss and makes later updates easier.

Record:

• The exact claim as presented.
• The wording you will use.
• The original source and stable link.
• Page, table, figure, or dataset location.
• Data collection period.
• Population and denominator.
• Method and sample size.
• Important limitations.
• Independent sources checked.
• Your verification status.
• The next review date.

Use simple statuses such as “verified as stated,” “verified with qualification,” “source located but method unclear,” and “not verified.” These labels are more informative than a binary true-or-false judgment.

When publishing, cite the primary source as close as possible to the statistic. If you rely on a secondary source because the original data are inaccessible, say so. Preserve the qualifier: “In a survey of 1,200 adults conducted in May 2024, 25% reported…” is more useful than “25% of adults…”

Troubleshoot common verification problems

If the original link is dead, search the exact title, author, and distinctive phrases. Check an institutional archive, a government document repository, or a web archive. Confirm that an archived copy has not been altered and record its archive date.

If the source is behind a paywall, search for a preprint, author manuscript, data appendix, institutional summary, or later report that describes the same study. Do not assume a summary contains every limitation from the original.

If a PDF is scanned or difficult to search, inspect the table of contents, use page numbers, and look for an accessible HTML version. Manually compare the relevant table and surrounding notes.

If the organization will not provide the data, distinguish between “the organization reports” and “the data show.” You can assess the transparency of the claim, but you may not be able to independently reproduce it.

If the claim changes over time, date-stamp your verification. A statistic can be accurate for one period and outdated later. Set a reminder to check recurring indicators, annual reports, and policy-related numbers.

Recognize the limits of verification

A verified source does not make a statistic universally applicable. It only establishes what a particular source measured, under particular conditions, using a particular method. Even well-designed research can contain sampling error, measurement error, missing data, or uncertainty.

Avoid claiming more than the evidence supports. If the source reports an association, do not describe it as causation unless the design supports a causal conclusion. If the estimate has an uncertainty interval, include it when the precision matters. If the evidence is preliminary, label it preliminary.

The safest practical rule is simple: preserve the claim’s population, time period, definition, and uncertainty when you repeat it. When those details cannot be verified, narrow the wording or leave the statistic out. A transparent limitation is more credible—and more useful—than a precise number whose origin cannot be traced.