How to Read a Survey Statistic: Sample, Population, Error

A survey statistic is an estimate for a defined population, made from a sample, with a margin of error around it. To read one correctly, check five things: who the population is, how large the sample was, how respondents were recruited, what the question actually asked, and who ran or paid for the survey. A figure such as "48% of knowledge workers say their work feels chaotic" is meaningful only once you know it comes from Microsoft's 2025 Work Trend Index survey across 31 countries, as shown on our employee burnout statistics page.

This guide walks through each check with figures from Statlyst datasets.

Start with the population

Every survey describes a population, and the percentage applies to that group only. Hybrid work is the clearest case on Statlyst.

• Gallup reports that 54% of US employees with remote-capable jobs worked a hybrid schedule in August 2026.
• The BLS American Time Use Survey reports that 35% of all employed Americans did some or all of their work at home on days worked in 2025.
• Statistics Canada reports that 9.8% of employed Canadians had a hybrid arrangement in May 2026.

The Gallup figure is much higher because it excludes everyone whose job cannot be done remotely. The definitions also differ: Gallup counts as hybrid anyone who works remotely 10% to less than 100% of the time, according to its methodology note. Before comparing two survey figures, write down the population of each in one line. If the lines differ, the numbers are not directly comparable.

Check the sample size and the margin of error

A sample of a few thousand people can describe a large population well, provided it was drawn properly. The uncertainty is expressed as a margin of error. Pew Research Center explains that a simple random sample of 1,067 cases has a margin of error of plus or minus 3 percentage points, and that this means the result would fall within 3 points of the true population value in 95 of 100 repeated surveys (Pew Research Center).

Three practical rules follow from that.

• Small differences may be noise. If two figures are 2 points apart and each has a 3 point margin, the survey cannot tell them apart.
• Subgroups are less precise. Pew notes that estimates for subgroups have fewer cases and so larger margins, in some cases much larger. A figure for "workers aged 18 to 24" carries more uncertainty than the headline.
• Gaps need a wider margin. Pew says the margin for the difference between two figures is generally about twice the margin for a single figure.

Official producers often flag this directly. The Health and Safety Executive describes the 3,220 per 100,000 rate of work-related stress among female workers in Great Britain as statistically significantly higher than the all-worker rate. When a source says "statistically significant", it has done this check for you. When it does not, treat small gaps with caution.

Large samples help with random error but not with bias. Resmed's 2026 sleep survey of 30,000 people in 13 countries is big, but its precision tells you nothing about whether the respondents represent the population of those countries.

Ask how respondents were recruited

How people entered the sample matters more than how many there are.

• Probability samples. Respondents are selected at random from a defined frame, so each member of the population has a known chance of selection. Gallup describes its employee data as coming from a random sample of panel members recruited with probability-based methods and weighted to Current Population Survey targets. National statistical offices such as the BLS, ONS and Statistics Canada also use probability samples for their labor surveys.
• Opt-in or convenience samples. Respondents volunteer, often through an online panel or a link. These samples can be large and fast but may overrepresent people who are online, engaged with the topic or customers of the sponsor. Results described as "respondents" to an open e-survey, such as the Eurofound figures on our hybrid work page, describe those respondents and should be read with that in mind.

If a report does not explain recruitment, treat its figures as describing the people who answered, not the wider population.

Read the exact question

Small wording changes produce different answers. Compare these from the same Gallup indicator:

• 33% of remote-capable employees preferred to work exclusively remotely in November 2025.
• 25% actually did so in August 2026.

One figure measures preference, the other behavior. Neither should be cited as the other.

Look for qualifiers in the question:

• "a lot of the previous day" (Gallup's 48% daily stress measure) is narrower than "ever".
• "always or very often" (Gallup's 29% burnout figure) excludes "sometimes".
• "in the past 12 months" sets a time window that a citation should keep.

Where a page on Statlyst carries the verbatim source quote, read it. The quote shows the exact wording the publisher used, which is often more cautious than a headline.

Separate surveys from other kinds of data

Not every percentage comes from asking people. Statlyst datasets mix several types, and each has its own strengths.

Data type	Example on Statlyst	What it measures well
Official household or labor survey	BLS time use survey	Population shares with known methods
Opinion or attitude poll	Gallup preferences	What people say they want or feel
Product telemetry	Microsoft 365 email volume	Activity of that product's users
Administrative records	HSE working days lost	Counts from official reporting systems
Randomized trial	Trip.com hybrid trial	Cause and effect within one firm

Telemetry, for instance, is not a sample of all workers. Microsoft's finding that users receive 117 emails a day describes Microsoft 365 users, and a randomized trial of 1,612 Trip.com employees shows what happened at that company, not across an economy.

Vendor surveys versus official statistics

Many widely shared figures come from companies that sell products related to the topic. That does not make them wrong, but it changes how you should read them.

Official statistics are produced by public bodies under published methods, usually with probability samples, documented definitions and revision policies. Examples on Statlyst include the ONS hybrid work estimates and the HSE stress statistics.

Vendor surveys are commissioned or run by companies. On our workplace communication statistics page, the ClickUp figures are labeled as coming from a vendor survey of more than 5,000 respondents, and the Atlassian estimate that teams lose 25% of their time searching for answers comes from its 2025 survey of 12,000 knowledge workers and 200 executives. Owl Labs, which sells meeting room video equipment, supplies the 77% of workers who say they lost time to meeting technology problems.

Questions to ask of a vendor survey:

• Does the topic of the finding match the sponsor's product?
• Is the sample described (country, size, dates, recruitment)?
• Is the full questionnaire or a methodology appendix available?
• Do official figures on the same topic point in the same direction?

Vendor surveys are often the only source for newer topics, such as meeting habits or AI tools at work. Use them, and label them.

Watch for sudden jumps

A large change between survey waves deserves a second look before you report it as a trend. Resmed's survey shows wearable sleep tracking rising from 16% to 53% of respondents between 2025 and 2026. A jump of that size can reflect real change, but it can also reflect a new question, a different panel or a different country mix. Check whether the publisher explains the change before treating it as fact.

FAQ

Is a bigger sample always better?

A bigger sample reduces random error, but it does not fix bias from how people were recruited or how the question was worded. A well drawn sample of 1,000 can be more reliable than an opt-in sample of 30,000.

What does "weighted" mean in a survey?

Weighting adjusts the sample so that its mix of age, sex, education, region and similar traits matches the population. Gallup, for example, weights its employee panel to Current Population Survey targets. Weighting corrects known imbalances but cannot correct for traits the survey did not measure.

Can I compare a vendor survey with an official statistic?

You can place them side by side, but state the differences in population, method and question. The hybrid work figures from Gallup, the BLS and Owl Labs measure different groups, so a gap between them is expected.

Where can I find the methodology for a Statlyst figure?

Each figure links to its source, and the publisher's methodology is usually on that page or one click away. Our own process for selecting and checking sources is described on the methodology page.