# Representative sample: meaning and how to assess one

> Assess population coverage, selection and nonresponse through a new-versus-returning customer example. Large samples can still leave groups out.


**A representative sample supports reasonable descriptions of a defined population for the question being studied.** Assessing it requires knowing whom you want to describe, how cases were selected and who ultimately responded. Sample size alone cannot establish coverage of the people who matter.

If you want to understand all buyers' experiences during a month, a survey answered only by active Instagram followers leaves an important coverage question. Their answers can reveal useful problems and ideas, while generalization requires examining how they were selected.

## Define the population first

Write down the group and period before recruitment. “People who made at least one purchase in October” is more precise than “our community.” Then identify a sampling frame, such as a customer list, and examine whom it omits or cannot reach.

Someone with three orders is still one person when the study unit is the customer. Sampling orders without accounting for the design may give frequent buyers more chances to appear. Statistics Canada's explanation of [sample selection](https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch13/sample-echantillon/5214900-eng.htm) connects the design with the question you want to answer about the population.

## Example: responding changes the composition

Suppose a store has 1,000 monthly buyers: 800 made their first purchase and 200 were returning customers. For this example, its list is complete and both groups can be contacted. The team randomly selects 80 from the first group and 20 from the second, using proportionate stratified sampling. The numbers and outcomes below are hypothetical.

| Stage | First purchase | Returning customer |
|---|---|---|
| Monthly population | 800 | 200 |
| Selected sample | 80 | 20 |
| People responding | 15 | 15 |

The original selection preserves an **80% and 20%** split. After responses arrive, 30 people remain, split **50% and 50%**. Participation was **15 ÷ 80 = 18.75%** among new buyers and **15 ÷ 20 = 75%** among returning customers.

This reveals a relevant participation difference. If the study concerns first-purchase difficulties, reading the combined result without examining groups can obscure the lower presence of new customers. The table cannot reveal what the missing customers would have said.

## Random samples and representativeness

Probability sampling uses random selection and allows inclusion probabilities to be calculated. It provides a basis for population estimates and uncertainty under the design. It does not force every small sample to match every characteristic exactly or eliminate nonresponse. Statistics Canada explains [probability sampling methods](https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch13/prob/5214899-eng.htm).

In volunteer or convenience samples, inclusion probabilities may be unknown. Quotas matching age or customer type do not establish matching preferences or experiences as well. Its discussion of [non-probability sampling](https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch13/nonprob/5214898-eng.htm) explains that limitation.

## Check before generalizing

Return to the frame and contact records. Was a group absent? Were people duplicated? Did the channel exclude people who do not use it? Compare respondent composition with the population using relevant, authorized data when available. In this example, investigating lower participation among new buyers is more specific than collecting additional answers from anyone willing to respond.

Weighting can adjust observed imbalances under particular assumptions. It cannot automatically recover unknown opinions or correct differences that were never measured. For population estimates with stated precision, plan the design and analysis with methodological support. Do not attach a conventional sampling margin of error to an open survey merely because it collected many answers.

## How large should the sample be?

The count depends on what is being estimated, variability, design, desired precision and expected nonresponse. A fixed number such as 385 cannot certify representativeness for every survey. Establish the population and selection method that will support the conclusion first.

You can limit the description to “among the 30 people who responded” and show their groups. Document collection outcomes with the [response rate and its bases](/en/guides/survey-response-rate/). If you need to explore why something happens before estimating how many people share it, [qualitative and quantitative research](/en/guides/qualitative-vs-quantitative-research/) can help choose the next method.

## Plan your next post in HeyMark.

Keep the idea, review the draft with your team, and see how it performed in the accounts you connected.

[Start free](https://app.heymark.ai) · [See how it works](https://heymark.ai/en/#product)

## Developer and agent resources

- [HeyMark MCP documentation](https://heymark.ai/en/mcp/)
- [llms.txt](https://heymark.ai/llms.txt)
- [Full site content for language models](https://heymark.ai/llms-full.txt)
- MCP protocol endpoint: `POST https://mcp.heymark.ai`
- OAuth protected-resource metadata: [/.well-known/oauth-protected-resource](https://mcp.heymark.ai/.well-known/oauth-protected-resource)
- MCP server card: [/server-card](https://mcp.heymark.ai/server-card)
