Guide
Sample Size and Sampling Basics, Without Heavy Maths
You don’t need a statistics degree to choose a sensible sample. You need a few ideas: who you are asking, how you pick them, and roughly how much uncertainty comes with a given number of answers.
Three words to know
- Population: everyone you want to learn about, for example all households within two kilometres of your shop.
- Sample: the people you actually ask.
- Sampling frame: the list or place you draw the sample from, such as a customer register, a WhatsApp group, or the people walking past your stall on a Saturday morning.
Problems usually start with the sampling frame. If your list only includes people who already follow you on social media, your “sample of customers” is really a sample of your fans.
Who you ask matters more than how many
A large sample drawn from the wrong people is still wrong. A thousand responses from your most loyal customers will tell you very little about why other people walk past. Before you worry about numbers, ask: “Does my sample include the kinds of people whose behaviour I need to understand?”
Common ways to choose a sample
| Method | How it works | Good for | Watch out for |
|---|---|---|---|
| Simple random | Everyone in the frame has an equal chance of being picked, e.g. drawing numbers from a full customer list | Results you can generalise to that list | Needs a complete list; can be impractical |
| Stratified | Split the population into groups (by area or age band, say), then pick randomly within each | Making sure small groups are represented | You must know which group each person belongs to |
| Cluster | Randomly pick whole groups, like a few housing societies or colleges, then survey people within them | Spread-out populations; lower travel cost | People inside a cluster can be alike, reducing variety |
| Quota | Keep recruiting until set numbers per group are filled, e.g. 50 men and 50 women | Quick fieldwork with some balance | Not random; it still depends on who is easy to reach |
| Convenience | Ask whoever is easy to reach | Early exploration; testing a questionnaire | Can be heavily biased; treat results as hints |
| Snowball | Respondents refer others like them | Hard-to-reach groups and niche professions | People refer friends who think like them |
How many responses is enough?
When you survey a sample, your result is an estimate. The margin of error tells you roughly how far that estimate might be from the true figure. A handy rule of thumb for a simple random sample, at the commonly used 95% confidence level and in the worst case (an answer split close to 50/50), is:
Margin of error ≈ 1 ÷ √(number of responses)
| Responses | Approximate margin of error |
|---|---|
| 100 | ± 10 percentage points |
| 400 | ± 5 percentage points |
| 1,000 | ± 3 percentage points |
These figures come straight from the rule above. They are approximations, not guarantees, and they only hold for well-drawn random samples. Notice the pattern: to halve the margin of error you need about four times as many responses, so going from 400 to 1,600 costs a lot of extra effort for a smaller gain.
Population size matters less than you think
For large populations, the margin of error depends mainly on the number of responses, not on the size of the population. Four hundred good responses tell you roughly as much about a town of 10 lakh people as about a city of 50 lakh. The exception is small populations: if you have only 300 customers in total, hearing from 150 of them gives you a much clearer picture than the formula suggests.
Plan for subgroups
If you want to compare groups, such as students versus working professionals, each group needs enough responses on its own. A total of 200 split across ten small groups gives you ten very uncertain answers.
Response rate and who stays silent
If you send a link to 1,000 people and 80 reply, the question is not only “Is 80 enough?” but “Are the 80 who replied different from the 920 who didn’t?” People with strong feelings, delighted or annoyed, are more likely to respond. Send one polite reminder, keep the survey short, and compare respondents with what you already know about your customers, such as their area or how long they have been buying, to spot obvious gaps.
Deciding your sample: step by step
- Define the population in one sentence.
- Identify a realistic sampling frame, and note who it leaves out.
- Choose a method from the table that fits your budget and time.
- Decide how precise you need to be. For a rough go/no-go decision, ±10 points may be fine; for choosing between two close prices you may need ±5 or better.
- Work out the responses you need, including enough for each subgroup you plan to compare.
- Estimate your likely response rate from past experience or a small test, and invite enough people to reach your target.
- Write down your method and its limits, so anyone reading the results understands them.