Should Stratified Sampling Use Proportional or Disproportionate Allocation?

Should Each Stratum Receive a Sample Proportional to Its Population Size?

You have divided your target population into meaningful strata and are ready to determine how many participants to select from each group.

Suppose 80% of the population belongs to public institutions and 20% belongs to private institutions. Should your sample follow the same 80:20 distribution? Or should you select equal numbers from both sectors to support comparisons?

Both approaches can be methodologically defensible, but they serve different purposes. Proportional allocation reflects population composition in the selected sample, whereas disproportionate allocation deliberately uses another distribution.

The choice affects subgroup precision, overall estimation, sampling weights, and potentially the total sample size required. It should therefore be determined by the study's analytical objectives rather than by an assumption that one allocation method is inherently more representative or statistically rigorous.

Use Proportional Allocation for Population Shares and Disproportionate Allocation for Specific Estimation Needs

How Proportional and Disproportionate Allocation Affect Your Sampling Design

What Is Sample Allocation in Stratified Sampling?

Sample allocation refers to how the total intended sample size is distributed among predefined population strata.

In stratified random sampling, researchers divide the population into mutually exclusive strata and select probability samples separately within each stratum.

For example, a population of 10,000 teachers might be divided into 8,000 public-school teachers and 2,000 private-school teachers. If the total intended sample is 1,000, the researcher must determine how many teachers to select from each sector.

The allocation determines the sampling fraction within each stratum. Under simple random sampling without replacement within strata, this fraction also determines each unit's inclusion probability.

Allocation is therefore not merely an administrative decision. It influences the statistical properties of the resulting estimators.

What Is Proportional Allocation?

Proportional allocation assigns each stratum a sample size corresponding to its share of the total population.

If a stratum contains 30% of the population, it receives approximately 30% of the total selected sample.

For a total sample size n, the proportional allocation is calculated as follows:

When exact proportional allocation produces fractional values, researchers must use a consistent rounding or integer-allocation procedure while preserving the intended total sample size.

Proportional allocation is particularly attractive when overall population estimation is central and no stratum requires a deliberately larger sample for separate analysis.

What Is Disproportionate Allocation?

Disproportionate allocation assigns stratum sample sizes according to a rule other than their population shares.

For example, a researcher might select 500 public-school teachers and 500 private-school teachers even though the population consists of 80% public-school teachers and 20% private-school teachers.

This approach deliberately samples private-school teachers at a higher rate.

Disproportionate allocation is not a single formula or technique. It is a category of allocation strategies that may serve different objectives, including subgroup comparisons, oversampling rare populations, and improving the efficiency of particular estimates.

The justification matters because different objectives may produce very different allocations.

When Is Proportional Allocation Appropriate?

Proportional allocation is often suitable when the primary objective is to estimate characteristics of the overall population and the resulting stratum sample sizes are adequate for the intended analyses.

It can be particularly convenient when:

• The research emphasizes population-wide means, proportions, or totals.
• The population distribution across strata is known reliably.
• Smaller strata do not require substantially larger samples for separate estimation.
• Within-stratum variability and data collection costs do not justify another allocation.
• Equal base inclusion probabilities are desirable for operational or analytical simplicity.

Under exact proportional allocation with simple random sampling within each stratum, the sampling fraction is the same across strata. This produces equal first-order inclusion probabilities and constant base design weights.

However, proportional allocation is not automatically the minimum-variance allocation for an overall estimate. If outcome variability differs substantially between strata, another allocation may be more statistically efficient.

When Is Disproportionate Allocation Appropriate?

Disproportionate allocation is particularly useful when proportional sampling would leave too few observations in analytically important groups.

Suppose a study requires a comparison between two population groups, one comprising 95% of the population and the other 5%. A proportional sample of 400 would allocate only 20 observations to the smaller group.

If the study requires a reasonably precise estimate for that smaller group, 20 observations may be inadequate.

The researcher could deliberately allocate more observations to the smaller stratum, even though its population share is limited.

This is one reason for oversampling rare population groups.

Disproportionate allocation may also be appropriate when one stratum has substantially greater outcome variability or when the cost of collecting an observation differs between strata.

Importantly, oversampling should follow a documented statistical or substantive justification. It should not be used simply to make the selected sample appear more balanced.

Does Equal Allocation Mean Equal Statistical Precision?

Not necessarily. Equal allocation gives the same number of selected units to each stratum, but equal sample sizes do not guarantee equal standard errors.

Precision also depends on outcome variability, sampling fractions, response patterns, and the estimator.

For example, two strata with 200 observations each may produce different standard errors if one has much greater variation in the outcome.

Likewise, equal allocation may improve the precision of estimates for a small stratum while reducing the number of observations available for a much larger stratum.

If a study prioritizes a comparison between two groups, the optimal allocation depends on the expected variance of the comparison and potentially the cost of obtaining observations from each group.

Equal allocation can be a reasonable starting point under similar variances and costs, but it should not be treated as a universal optimum.

What Is Neyman Allocation?

Neyman allocation is a form of disproportionate allocation designed to minimize the variance of an overall stratified mean for a fixed total sample size under specified assumptions, including equal per-unit data collection costs.

It allocates more observations to strata that are larger and have greater outcome variability.

Neyman allocation requires information about within-stratum variability, which may come from prior surveys, administrative data, or pilot studies.

If those estimates are unreliable, the apparent precision advantage may not materialize.

Where data collection costs differ across strata, cost-sensitive optimal allocation may be more appropriate. Under common assumptions, the allocation is proportional to population size multiplied by the stratum standard deviation and divided by the square root of per-unit cost.

These methods should be evaluated in relation to the study's estimands rather than applied mechanically.

How Do the Allocation Strategies Compare?

Several objectives may compete. An allocation optimized for the overall population mean may leave an important subgroup with too few observations for a separate estimate.

Researchers may therefore impose minimum sample sizes for reporting domains and optimize the remaining allocation subject to those constraints.

How Does Disproportionate Allocation Affect Sampling Weights?

Under independent simple random sampling without replacement within each stratum, the inclusion probability of a unit is the stratum sample size divided by the stratum population size.

When the sampling fractions differ, the corresponding base design weights differ.

For an overall population estimate, failing to account for unequal selection probabilities can overrepresent the deliberately oversampled group.

However, the correct analytical approach depends on the estimand. Some model-based analyses may incorporate the design differently, and weights alone do not resolve every inferential issue.

The broader principles of unequal selection probabilities and sampling weights should be incorporated into the analysis plan before data collection.

Can Disproportionate Allocation Improve Subgroup Precision but Reduce Overall Precision?

Yes. This is an important trade-off.

Allocating more observations to a small stratum can improve the precision of that stratum's estimate. However, if the total sample size remains fixed, those additional observations must generally come from other strata.

As a result, the precision of estimates for larger strata may decline. The variance of the overall population estimator may also increase if the allocation moves away from an efficient distribution for that outcome.

This does not make disproportionate allocation inappropriate. It means the design should be evaluated against the estimates the study actually prioritizes.

A study whose primary objective is comparing two population groups may reasonably accept some loss in precision for the overall population mean in exchange for better subgroup estimates.

Conversely, a survey designed primarily to estimate a national prevalence may favor an allocation that prioritizes the precision of that national estimate.

Should Allocation Be Based on the Selected Sample or Expected Respondents?

Allocation is usually specified in terms of the number of units to select, but the number of usable responses may be smaller.

If response rates are expected to differ between strata, the researcher may need to select different numbers of units to achieve the intended number of completed observations.

For example, if a stratum requires 200 completed responses and the anticipated response rate is 80%, the researcher may initially select approximately 250 eligible units.

Such adjustments should be based on defensible assumptions and documented recruitment procedures.

Importantly, the achieved sample may not preserve the original allocation. Differential nonresponse can introduce bias even when the initial selection was properly stratified.

Researchers should monitor response patterns within strata and avoid treating unplanned replacement of nonrespondents as a substitute for appropriate nonresponse management.

When Might Proportional Allocation Be Preferable to Oversampling?

Proportional allocation may be preferable when the smaller strata are not central to the research questions and the study's principal objective is overall population estimation.

It may also be attractive when the total sample size is sufficiently large to provide adequate subgroup observations without disproportionate allocation.

Researchers should not oversample a group merely because its population share is small. The additional observations should serve a defined analytical or substantive purpose.

Before choosing an allocation, establish which strata are necessary for the intended estimates and whether the planned sample sizes support the required precision.

In some studies, a modest disproportionate allocation may provide a useful compromise between subgroup precision and overall estimation efficiency.

Allocating a Sample Between Public- and Private-School Teachers

Under proportional allocation, the researcher would select 800 public-school teachers and 200 private-school teachers.

Under equal allocation, the researcher would select 500 teachers from each sector.

Suppose the observed adoption rates are 40% among public-school teachers and 60% among private-school teachers. For illustration, assume complete response and no additional weighting adjustments.

Under either valid allocation, the overall population estimate should reflect the actual population shares.

The researcher then considers precision. Equal allocation provides 500 selected private-school teachers instead of 200, potentially improving the precision of the private-sector estimate.

However, it reduces the public-school sample from 800 to 500. For the overall population estimate, this trade-off may or may not be worthwhile depending on the outcome variances and the relative importance of the two research objectives.

If the main objective is estimating overall adoption, proportional allocation may be reasonable. If the public-private comparison is central and the smaller sector requires substantially more observations, disproportionate allocation may be preferable.

A defensible final allocation would use precision or power calculations for the planned estimates rather than choosing equal or proportional sample sizes solely by convention.

Common Misunderstandings About Stratified Sample Allocation

How to Choose an Allocation Strategy for Your Research

Begin by identifying whether the study primarily requires an overall population estimate, separate subgroup estimates, or comparisons between strata.

Then assess the population sizes, expected outcome variability, anticipated response rates, and available resources. Allocation should be based on the estimates the study needs to produce, not simply on the appearance of the final sample distribution.

For the methodology section, report the population size of each stratum, the allocation formula or decision rule, the selected sample size within each stratum, and the rationale for the chosen distribution.

Where disproportionate allocation is used, explain how unequal inclusion probabilities are reflected in the analysis. Also document how anticipated nonresponse influenced the number of units initially selected.

The allocation should be defensible before the data are collected. Revising it solely because the observed results appear more favorable under a different distribution would undermine the transparency of the research design.

Before Finalizing Proportional or Disproportionate Allocation, Verify These Decisions

Frequently Asked Questions About Stratified Sample Allocation

Choose Allocation According to the Estimates Your Study Needs

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.