AP Statistics: Blocking and Experimental Design

Designing a valid experiment is one of the most powerful skills in statistics, allowing you to move from passive observation to active discovery. While randomization is the cornerstone of a good experiment, it doesn't always guarantee a fair comparison. Blocking is the strategic technique you use to account for known sources of variability before randomization, creating a more precise and powerful experiment. Mastering this concept is essential for the AP Statistics exam and for any field where rigorous testing is required, from agriculture to engineering to medicine.

What is Blocking in Experimental Design?

Blocking is the process of grouping experimental units together based on a characteristic, or lurking variable, that you suspect will strongly influence the response variable. These groups are called blocks. The key principle is that units within a block are as similar as possible, while units between different blocks can be quite different. After forming blocks, you then randomize the assignment of treatments to the experimental units within each block separately.

Think of it like organizing a race. If you want to test two new running shoe designs, it wouldn't be fair to randomly assign all runners to shoes. A professional sprinter will likely outperform a novice regardless of the shoe. A better design is to create blocks: one block of professional runners and another block of novice runners. Then, within each block, you randomly assign half the runners to Shoe A and half to Shoe B. This ensures each shoe type is tested on comparable athletes.

The structure this creates is called a randomized block design. It is a major category of experimental design, distinct from a completely randomized design. Its primary goal is to account for a major source of variability, thereby "cleaning up" the comparison and making it easier to detect a true treatment effect if one exists.

How Blocking Reduces Variability and Controls Confounding

To understand why blocking works, you must understand variability. All experiments have natural variation in their results. Some of this variation is due to the treatments, and some is due to other factors. A completely randomized design (CRD) spreads all sources of variation—both known and unknown—randomly across treatment groups. This is good for controlling unknown lurking variables through randomization.

However, if there is a known major source of variation (like skill level in the race, soil type in a farm plot, or a patient's age in a medical trial), a CRD might accidentally create a misleading situation. For example, if by chance most professional runners got Shoe A, Shoe A would appear better even if it wasn't. This is a form of confounding, where the effect of the treatment is mixed up with the effect of the lurking variable.

Blocking proactively eliminates this problem. By creating homogeneous groups first, you isolate the variability due to the blocking variable. The statistical analysis can then separate this "block-to-block" variation from the "within-block" variation used to compare treatments. This dramatically increases the precision of your experiment. Essentially, you are comparing treatments under more similar conditions, so any differences you see are more likely to be caused by the treatments themselves.

Designing a Blocked Experiment: A Worked Example

Let's walk through designing a randomized block experiment from start to finish.

Scenario: A gardener wants to test the effect of three new liquid fertilizers (A, B, and C) on tomato plant yield. Their garden has three distinct areas: one gets full sun, one gets partial sun, and one is mostly shaded. Sunlight is a known major factor for plant growth.

1. Identify the Experimental Units, Treatment, and Blocking Variable.

• Experimental Units: Individual tomato plants.
• Treatment: Type of fertilizer (A, B, or C).
• Blocking Variable: Amount of sunlight (Full, Partial, Shade).

1. Create the Blocks.

Create three blocks, one for each sunlight level. All plants within the "Full Sun" block are grown in the full sun area, and so on.

1. Randomize Within Blocks.

Within each sunlight block, randomly assign an equal number of plants to each of the three fertilizers. For instance, if you have 12 plants in the Full Sun block, use a random number generator to assign 4 to Fertilizer A, 4 to B, and 4 to C. Repeat this randomization process independently for the Partial Sun and Shade blocks.

This design ensures each fertilizer is tested across all sunlight conditions equally. When you analyze the results, you can assess the fertilizer effect after accounting for the substantial differences in yield caused by sunlight.

Blocked Design vs. Completely Randomized Design: A Strategic Choice

Choosing between a randomized block design (RBD) and a completely randomized design (CRD) is a critical decision.

The mathematical intuition is that blocking partitions the total variability $SST$ into two components: variability between blocks $SSB$ and variability within blocks $SSE$. In a well-blocked experiment, $SSB$ is large (blocks are very different), which makes the leftover $SSE$ smaller. Since we use $SSE$ to estimate experimental error, a smaller error leads to a more sensitive test. In a CRD, all that block-to-block variability gets lumped into the error term, inflating it.

Common Pitfalls

Confusing Blocks with Factors: A block is not a treatment level. You are not intentionally applying "shade" to plants; you are grouping them based on the shade they naturally receive. The goal is not to study the effect of the blocking variable, but to neutralize its influence. In contrast, if you intentionally applied different levels of sunlight as part of the experiment, sunlight would be a factor in a factorial design, which is a different, more advanced concept.

Blocking on the Wrong Variable or Over-Blocking: Blocking is only beneficial if the variable you choose creates groups that are truly different in terms of the response. Blocking on a variable with little to no effect just complicates the design without benefit. Similarly, creating too many small blocks can leave you with very few units per block for treatment assignment, weakening the experiment. A good blocking variable has a strong, known association with the outcome.

Forgetting to Randomize Within Blocks: Blocking alone is not enough. After forming blocks, you must randomly assign treatments within each block. If you don't, you reintroduce the potential for bias. The full procedure is "block, then randomize."

Misidentifying the Design in an Exam Question: On the AP exam, you must correctly label an experimental design. Remember: if the description mentions grouping similar subjects together first (e.g., "separated by gender," "grouped by field location"), then randomly assigning within those groups, it is a randomized block design. If subjects are randomly assigned to treatments from one big pool with no prior grouping, it is a completely randomized design.

Summary

• Blocking is a pre-randomization step that groups similar experimental units to control for a known source of unwanted variability, increasing the precision of an experiment.
• A randomized block design involves creating homogeneous blocks based on a lurking variable and then randomly assigning treatments to units within each block.
• The primary purpose of blocking is to reduce variability and prevent confounding, making it easier to detect a true treatment effect if one exists.
• Blocking is strategically chosen over a completely randomized design when a major, known source of variation is present among experimental units.
• Effective blocking requires a strong blocking variable and strict adherence to the process: form blocks, then randomize within them. Avoid blocking on trivial variables or treating a block as an experimental factor.