What to do next
Show the working, step by step
Where I used this
I ran a year-long campus harm-reduction campaign that was segmented by audience and evaluated from start to finish.
Read the case: Campus harm-reduction campaign: segmented, evaluated prevention →Everything is calculated in your browser. Nothing you type is sent or stored.
Choosing the inputs
- Margin of error. ±5 points is common for program surveys. Use ±3 if you need to detect small changes between rounds.
- Expected proportion. If you already know roughly what share of people have the behaviour or outcome, use it. If not, 50% gives the largest, safest sample.
- Design effect. Household and cluster surveys, common in public health, are less efficient than simple random samples. A design effect of 1.5 to 2 is a typical allowance; use the value from a previous survey if you have one.
- Subgroups. If you need results for each district, age group or gender, each subgroup needs its own sample. Multiply accordingly.
Get new free tools by email. One short note when a new tool or essay goes live. No spam, unsubscribe any time.
Planning a survey you can defend?
Bring your design to a free 30-minute call, and I will check your sample, design effect and non-response before you go to the field.
Questions
How many people do I need to survey?
For a large population, 95% confidence and a ±5 point margin of error, about 385 completed responses with simple random sampling. Cluster designs and non-response push the number up.
Does population size matter?
Only when it is small. Above a few tens of thousands, the sample needed barely changes. The finite population correction shrinks the sample for small populations.
What is a design effect?
A multiplier that accounts for cluster or multi-stage sampling, where people in the same cluster tend to answer alike. It inflates the sample needed for the same precision.