How to Calculate Sample Size for a Cross-Sectional Survey

If you are running a household survey, a knowledge attitudes and practice (KAP) study, a coverage survey, or almost any one off data collection exercise, you are running a cross-sectional study. This guide first explains the general formula behind the calculation, then walks through how to apply it in AnalyZ Solutions.

01The general sample size formula

Before opening any tool, it helps to understand what is actually being calculated. For a survey estimating a proportion, the standard formula, often attributed to Cochran, is:

Each part of the formula plays a distinct role.

• Z (confidence level). This reflects how sure you want to be that your estimate falls within the margin of error, across repeated samples. A 95% confidence level, the most common choice in social research, corresponds to Z = 1.96.
• p (expected proportion). This is your best guess at the true value in the population, expressed as a decimal. If you truly do not know, 0.5 is used because it produces the largest possible sample size for any given precision, which makes it the safest assumption.
• e (margin of error). This is how far off your estimate is allowed to be from the true population value. A margin of error of 0.05 on an estimate of 60% means the true value plausibly lies between 55% and 65%.

If your outcome is a numeric average rather than a percentage, for example average household expenditure or average test score, the formula changes shape slightly, because it needs to account for the underlying variability of the measure rather than the variability of a proportion:

Why is 384 such a common number?

If you plug the most conservative, most common assumptions into the proportion formula, 95% confidence, p = 0.5, and a 5% margin of error, the result is:

Rounded up, this becomes 385, though it is frequently quoted as 384. It shows up so often in published research and survey guidance because it represents the worst case, most defensible default when nothing specific is known about the population. It is not a magic number. It is simply what falls out of the most cautious set of standard assumptions.

02Two further adjustments the basic formula does not include

The formula above assumes a simple random sample from an effectively infinite population, with perfect response. Two common adjustments extend it to more realistic conditions.

Finite population correction

If your population is small, meaning you plan to sample a meaningful share of everyone who exists, the basic formula overestimates the sample you need. The finite population correction adjusts for this:

As N grows large relative to n, this correction has almost no effect, which is why it is usually skipped for large or unknown populations and applied only when the population is small, commonly under a few thousand.

Design effect for cluster and multi-stage sampling

If you plan to sample by group, for example by village, school, or clinic, rather than selecting individuals completely at random, people within the same group tend to be more alike than people picked at random from the whole population. The design effect (DEFF) accounts for this loss of independence:

A DEFF of 1.8, for example, means the sample needs to be 80% larger than a simple random sample of equivalent precision would require. This is covered in detail in a separate guide, linked at the end of this article.

AnalyZ Solutions applies all of the formulas above automatically, so there is no need to calculate them by hand. The steps below map directly onto the Cross-sectional calculator in the Sampling module.

1. Choose what you are estimating. Select Proportion if your outcome is a percentage or rate, or Mean if it is a numeric average. This determines which formula is applied.
2. Set your confidence level and margin of error. Enter these as numbers, not from a fixed list, so any value is possible. For a proportion, also enter the expected proportion, leaving it at 50% if you have no prior estimate. For a mean, enter the expected standard deviation instead.
3. Choose how you will select participants. Pick Simple random, Stratified, or Cluster and multi-stage. If you choose Cluster, you can enter a design effect directly, or calculate one automatically from an intraclass correlation and your average cluster size.
4. Enter a population size, if relevant. If your population is small enough that the finite population correction should apply, enter it here. Otherwise, leave this field blank. AnalyZ Solutions applies the correction automatically whenever a population size is entered, and skips it otherwise.
5. Set your expected non-response rate. This inflates the final recruitment target so that, after accounting for people who do not respond, you still end up with enough completed responses.
6. Calculate, and compare confidence levels if useful. After calculating, the interpretation panel explains the result in plain language, and a Compare confidence levels option shows the required sample at several confidence levels side by side.

03Worked example

Suppose you are planning a household survey to estimate the percentage of households with an improved water source in a district where you expect coverage to be around 55%, based on a prior assessment. You will sample by village, visiting about 15 households per village.

A simple random sample at the same precision would need only about 380 households. Clustering by village nearly doubles the requirement, which illustrates why the design effect is worth entering carefully rather than left at its default.

04Common mistakes to avoid

• Leaving the design effect at 1 when sampling by cluster. If clusters are used but DEFF is not adjusted, the survey will likely be underpowered.
• Choosing an unnecessarily precise margin of error. A margin of error of 2% sounds better than 5%, but can roughly quadruple the required sample for a level of precision the decision at hand may not need.
• Forgetting non-response until fieldwork begins. Build the inflation into the target from the start, rather than as an emergency addition mid-survey.
• Assuming a large national total covers every subgroup evenly. A proportionate stratified allocation can still leave a small stratum underpowered. If subgroup reporting is needed, use the option for precision within each stratum instead.