Blocking vs stratifying: experiments vs sampling

Blocking vs stratifying: the short answer

Both start the same way, by dividing units into groups that are internally similar, and that shared start is exactly why students confuse them. The difference is where each one lives and what happens after the split.

Blocking belongs to experiments. You form blocks of similar units, then randomly assign the treatments within each block, so every treatment appears in every block. Stratifying belongs to sampling. You form strata of similar individuals, then take a simple random sample within each stratum and combine them into one sample.

A one-line test settles most cases. If treatments are being handed out, you are blocking in an experiment; if respondents are being selected, you are stratifying in a sample. Assign versus select is the whole distinction.

What blocking is

Blocking is part of experimental design. A blocking variable is a source of extraneous variation in the response, such as age, soil type, or baseline severity. In a randomized block design, you first group the experimental units by similar values of that variable into blocks, so units within a block are alike with respect to it. Then, inside each block, you randomly assign the treatments so that every treatment occurs within every block.

The purpose is precision. Blocking separates the variation in the response caused by the blocking variable from the rest of the extraneous variation, which lets you compare treatments without that nuisance variable getting in the way. Within a block, the units start out similar, so a difference in their responses points more cleanly at the treatment.

A matched pairs design is the smallest case: a randomized block design with only two treatments. Units are matched into pairs, and within each pair the two treatments are randomly assigned, one to each member; alternatively a single unit receives both treatments with the order randomized. Blocking never removes the need for random assignment; it organizes the units first, then randomizes inside each group.

What stratifying is

Stratifying is part of sampling. A stratified random sample divides all individuals in the population into non-overlapping groups, called strata, based on one or more shared attributes, so each stratum is homogeneous. Within each stratum you take a simple random sample, then combine the selected individuals into one overall sample.

The purpose is representation and, often, a more precise estimate. Because every stratum is forced to appear, stratifying guarantees that each group shows up in the sample, and when the strata really are internally similar the estimate tends to vary less than a simple random sample of the same total size. No treatment is involved; you are only choosing who gets measured.

Stratifying sits alongside the other sampling methods, cluster and systematic, covered in simple random vs stratified sampling. The through-line is that stratifying decides which individuals enter an observational study or survey, never what is done to them.

The differences side by side

Read the setting row first. Everything else follows from whether you are running an experiment (blocking) or drawing a sample (stratifying).

When to use which

Block when you are running an experiment and you know an extraneous variable is likely to affect the response. Grouping units by that variable first, then randomizing treatments inside each block, keeps its influence from muddying the treatment comparison and gives you more precise results. Age in a drug trial or field position in an agriculture trial are natural blocking variables.

Stratify when you are drawing a sample and you want every subgroup represented, or you expect the subgroups to differ on what you are measuring. Splitting the population into strata and sampling within each guarantees coverage and can tighten the estimate. Income band in an opinion poll or grade level in a school survey are natural strata. The guiding slogan is block in experiments, stratify in samples: same tactic of grouping similar units, matched to the design you are actually running.

The classic mix-up and how to avoid it

The mix-up is using one word for the other, because both create homogeneous groups. The fix is to look at what happens after the grouping. If treatments are randomly assigned within the groups, the study is an experiment and the groups are blocks. If a random sample is drawn within the groups and nothing is done to anyone, the study is a sample and the groups are strata.

A second cue is the goal. Blocking aims to reduce extraneous variation so treatment effects stand out; stratifying aims to represent every subgroup and lower the variability of an estimate. Ask assign or select, and experiment or sample, and the right term follows every time. Both also differ from clustering, where you sample a few whole groups instead of sampling within every group, a contrast drawn out in simple random vs stratified sampling.