Quotas and balancing

Overview

Quotas and balancing are how you control the composition of your sample, so it matches the population you're studying, or the specific comparison your study is designed to make.

What quotas are

A quota is a numeric target that tells you how many participants to recruit from a specific subgroup, such as age, gender, or region. Quotas can be set in a few different ways:

• Minimum — a floor, at least this many participants from a group
• Maximum — a cap, no more than this many participants from a group
• Exact target — precisely this many participants from a group

Proportional vs. non-proportional quotas

• Proportional quotas match your sample's composition to real-world population data, so if a group makes up 30% of your target population, it makes up roughly 30% of your sample too
• Non-proportional quotas (also called oversampling) set a minimum for smaller groups specifically so you have enough participants in that group to analyse it on its own, even if it's a small slice of the overall population

Use proportional quotas when you want your overall results to reflect the real population. Use non-proportional quotas when a smaller group matters to your analysis and needs its own reliable sample, even if that means it's over-represented relative to its real-world size.

What balancing does

Balancing adjusts your sample so no single group skews your overall results. This matters not just for individual traits on their own, but for how traits intersect, for example, making sure you have enough participants in "women aged 18-34" specifically, not just enough women and enough 18-34-year-olds separately.

Balancing also protects against a common problem in non-random recruitment: some participants are simply easier to reach than others, and without balancing, your sample can end up over-representing whoever was easiest to recruit, rather than reflecting who you actually intended to study.

Why use quotas and balancing

• Faster and cheaper than strict random probability sampling
• Guarantees enough participants in niche segments or key demographics, so subgroup or cross-group comparisons are actually reliable
• Prevents skewed results caused by over-representing easy-to-reach participants

When should I use quotas?

Use quotas when specific participant characteristics matter to your research or analysis. Common quota variables include age, gender, location, customer type, brand usage, usage frequency, or audience segment.

Example: If you want to compare current users and competitor users, set quotas to ensure both groups are represented in the proportions your research needs. This prevents recruitment from concentrating too heavily in one group, leaving too few participants in the other for a meaningful comparison.

Set quotas based on your analysis plan

Before setting quotas, decide which groups need to appear separately in your final analysis. If a group needs its own results, make sure you're recruiting enough participants into it specifically.

Interlocking quotas

Sometimes the combination of characteristics matters more than any single one. For example, if you need balanced representation across both age and gender, you may need separate quotas for groups like Women 18-34, Women 35-54, Men 18-34, and Men 35-54, rather than balancing age and gender independently.

Qualitative quotas

For qualitative research, quotas are often better expressed as participant counts rather than population percentages, for example, "at least 5 heavy users" and "at least 5 light users." This ensures the groups you want to understand are represented, without treating a small qualitative sample as if it were statistically representative of the wider population.

Don't overuse quotas

Not every demographic variable needs a quota. Use quotas only where a variable is important to your research objective, your target audience, or a comparison you actually plan to make. Too many quotas can make recruitment unnecessarily difficult.

What to do next

See Statistical confidence in Decode reports to understand how to interpret results once responses come in.