"We need 5,000 completes across ten countries." I hear some version of this every week, and it's usually the wrong starting point. The right question is: what's the smallest group you need to talk about with confidence, in each market?

Key takeaways

  • Size each country on its own. A global total hides weak markets.
  • The smallest reportable subgroup sets your minimum, not the headline margin of error.
  • Plan for n=100 to 150 in any cell you'll report on.
  • Check incidence and feasibility per market before you promise stakeholders a number.

Start from the questions you need to answer

List the cuts that will appear in the deck. "Gen Z vs. Boomers in Brazil." "IT decision makers at 1,000+ employee companies in Germany." Each of those is a cell. Your sample size is whatever it takes to fill the smallest cell to a readable size.

Margin of error, quickly

At 95% confidence, for a 50/50 split (the worst case):

n Margin of error
100 ±9.8 pts
200 ±6.9 pts
400 ±4.9 pts
500 ±4.4 pts
1,000 ±3.1 pts
2,000 ±2.2 pts

Notice the curve flattens. Going from 1,000 to 2,000 buys you less than a point. That money is usually better spent boosting a small subgroup.

Minimum cell sizes

We treat n=100 as the floor for any cell shown on a chart, and n=150 or more when the client needs to compare cells against each other. Below that, results swing too much between waves to be useful.

Equal or proportional allocation?

If the goal is comparing countries, give each one the same n. If the goal is a single global number, allocate proportionally (or equal n plus weighting to population). Many trackers do both: equal n in field, weighted to market size for the global total. Just remember weighting reduces your effective sample size. We cover that in weighting survey data.

Check feasibility before you commit

A sample plan that works in the US can fall apart in Vietnam or Saudi Arabia, where incidence and online reach differ. Ask for country-by-country feasibility up front. Our team quotes global sample market by market for this reason, and we'll tell you when a cell isn't realistic online.

FAQ

What is a good sample size per country?

For a single national read, n=400 to 500 gives about ±5 points. If you'll cut by age, gender or region inside the country, plan for 800 to 1,000 so each subgroup has at least 100 to 150.

Can I combine countries for a global total?

Yes, but weight each country to its share of your target population, otherwise small markets with large samples will overpower the result.