AP Statistics: Stratified Random Sampling

When you need to understand a diverse population—be it voters, consumers, or manufactured parts—a simple random sample can sometimes miss crucial subgroups. Stratified random sampling is a powerful design that ensures every segment of your population is properly represented, leading to more precise and reliable estimates. By intelligently grouping the population before sampling, you gain control over variability and enhance the quality of your statistical inference, a skill directly tested on the AP exam and applied in fields from engineering to public policy.

What is Stratified Sampling?

Stratified random sampling is a probability sampling method where the researcher first divides the entire population into distinct, non-overlapping subgroups called strata (singular: stratum). The key is that every member of the population belongs to exactly one stratum. Once the strata are defined, a simple random sample (SRS) is taken independently from within each stratum. These individual stratum samples are then combined to form one complete stratified sample.

The fundamental purpose of stratification is to create homogeneous groups. This means the units within each stratum are as similar as possible with respect to the characteristic being studied. For example, if you are surveying income, you might stratify by profession. If you are testing battery life, you might stratify by production batch. The stratification variable is chosen because it is believed to be related to the response variable of interest. The opposite of homogeneity is heterogeneity, which describes variation across the entire population; a well-stratified design places heterogeneous groups into different strata, minimizing variation within each one.

Why Stratify? Reducing Sampling Variability

The primary advantage of stratified sampling over a simple random sample (SRS) of the same total size is the reduction in sampling variability for your estimates. In simpler terms, your sample statistics (like the mean $\bar{x}$) will tend to be closer to the true population parameters (like the mean $\mu$).

This happens because stratification eliminates variability between strata from the sampling error. Imagine estimating the average height of a high school. The population is heterogeneous—it includes freshmen and seniors. If you take an SRS, you might, by chance, get a few too many freshmen, pulling your estimate down. By stratifying by grade level and sampling proportionally from each grade, you guarantee that each grade's representation in the sample matches its representation in the population. This control over the sample's composition removes the "luck of the draw" related to stratum membership, leading to less variable and more precise estimates.

Mathematically, the standard error of the stratified sample mean is generally smaller than the standard error of the mean from an SRS. The more homogeneous the strata are internally (and the more different they are from each other), the greater the reduction in variability. This principle is crucial for the AP exam: you must be able to explain that stratification yields estimates with lower margin of error for the same sample size, or it can achieve the same margin of error with a smaller, more cost-effective sample.

Selecting Stratification Variables

Choosing the right variable to define your strata is a critical design decision. An effective stratification variable has two key properties:

1. Strong Correlation with the Response: The variable should be closely related to the primary measurement you are collecting. If you are studying standardized test scores, stratifying by prior GPA or course level is effective. If you are studying political opinion, stratifying by geographic region or party registration is wise. Stratifying by an unrelated variable (e.g., stratifying by birth month for a study on commuting time) offers no benefit and adds unnecessary complexity.
2. Practical Measurability: You must be able to identify which stratum every population member belongs to before sampling. This requires a sampling frame that includes the stratification variable. For instance, a university registrar's list can stratify by major and class year. A manufacturing log can stratify by production shift or machine ID.

In engineering and quality control contexts, common stratification variables include time (production shift, hour, day), source (supplier, batch of raw material), or machine (assembly line number). This allows engineers to isolate and analyze potential sources of variation in a process systematically.

Designing and Analyzing a Stratified Sample

Once you have defined strata, you must decide how many individuals to sample from each one. The two primary allocation methods are:

• Proportional Allocation: The sample size for each stratum is proportional to the stratum's size in the population. If Stratum A makes up 20% of the population, it makes up 20% of the total sample. This method is straightforward, ensures the sample is a miniature version of the population, and is highly efficient when the variability within strata is roughly equal.
• Optimal (Neyman) Allocation: This method allocates more samples to larger strata and to strata with greater internal variability. It minimizes the overall standard error of the estimate. The formula for allocating sample size $n_h$ to stratum $h$ considers both the stratum's proportion of the population ($N_h/N$) and its standard deviation ($s_h$): $n_h=n\cdot \frac{N_h{s}_h}{\sum (N_i{s}_i)}$. You should be familiar with this concept for the AP exam, though you typically won't perform the full calculation.

The analysis of data from a stratified sample requires careful weighting. You cannot simply average all the collected data. You must calculate the statistic (like a mean) for each stratum separately and then combine them using weights based on the stratum's proportion of the population.

For example, the stratified estimate of the population mean is a weighted average: $${\bar{x}}_{strat}=\sum _{h=1}^H\left(\frac{N_h}{N}\right){\bar{x}}_h$$ where $H$ is the number of strata, $N_h$ is the population size of stratum $h$, $N$ is the total population size, and ${\bar{x}}_h$ is the sample mean from stratum $h$.

Common Pitfalls

1. Creating Heterogeneous Strata: The most common conceptual error is failing to create internally homogeneous groups. If the units within a stratum are as different from each other as they are from units in other strata, stratification provides no benefit. Correction: Always choose a stratification variable that directly relates to the measurement of interest to maximize within-stratum similarity.

1. Ignoring Proportional Analysis: A major calculation error is treating a stratified sample as if it were an SRS by taking a simple average of all data points. This gives unequal weight to individuals from different strata and biases your estimate. Correction: Always use the weighted average formula, where the weights are the stratum proportions in the population.

1. Confusing Stratification with Clustering: These are opposite designs but are often mixed up. In stratified sampling, you sample from every stratum. In cluster sampling, you randomly select a few clusters (which are heterogeneous mini-populations) and then sample everyone or a subset within those selected clusters. Correction: Remember: Stratify when groups are homogeneous; cluster when groups are heterogeneous and representative of the whole. On the AP exam, read carefully to see if the design samples from all groups (stratified) or only some groups (cluster).

1. Over-Stratifying: Creating too many strata for a given sample size can be counterproductive. If your sample only includes one or two individuals per stratum, you cannot reliably estimate the variability within that stratum. Correction: Balance the desire for homogeneity with practical sample size constraints. Often, 3-6 well-chosen strata are more effective than 20 very narrow ones.

Summary

• Stratified random sampling involves dividing the population into homogeneous subgroups (strata) and then taking a separate simple random sample from each stratum.
• The core benefit is a reduction in sampling variability compared to a simple random sample of the same size, leading to more precise estimates with a smaller margin of error.
• An effective stratification variable must be related to the response variable of interest and measurable for the entire population before sampling.
• Analysis requires calculating weighted averages (e.g., ${\bar{x}}_{strat}=\sum (N_h/N){\bar{x}}_h$) to correctly combine results from each stratum, reflecting their population proportions.
• On the AP exam, be prepared to justify the use of stratification, identify appropriate stratification variables, and distinguish this method from cluster sampling.