Randomized block vs completely randomized design

By Jude Wallis · Published

A completely randomized design assigns all experimental units to treatments by chance in one step. A randomized block design first sorts units into blocks alike on a variable that affects the response, then randomly assigns treatments within each block. Blocking gives a more precise comparison.

AP Statistics: Unit 1 (topics 1.13 Experimental Design). The completely randomized design and the randomized block design are both part of topic 1.13, Experimental Design, in Unit 1 of the Fall 2026 AP Statistics course. A typical task is to name the design from a description and to justify a choice of blocking variable in context.

Randomized block vs completely randomized: the short answer

Both designs assign treatments at random. The difference is whether the units are grouped first.

A completely randomized design puts every experimental unit into one pool and lets chance decide the treatment for each. A randomized block design begins by sorting the units into blocks, groups that are alike on some variable expected to affect the response. The random assignment is then carried out separately inside each block, so every treatment appears in every block.

A completely randomized design leaves it to chance whether a variable such as sunlight or age ends up evenly spread across the treatments. A randomized block design guarantees the spread for the blocking variable, and that removes the variable's effect from the comparison of treatments.

What a completely randomized design is

In a completely randomized design, one random process assigns all units. With 24 garden plots and two fertilizers, you could number the plots 1 to 24, have a random number generator pick 12 distinct numbers for fertilizer A, and give fertilizer B to the rest.

This design is valid whether or not the plots differ. Random assignment gives every plot the same chance of each fertilizer, so no fertilizer is favored. What it does not do is promise balance in any one run. If half the plots are sunny, a particular assignment may put 8 of the 12 sunny plots on fertilizer A. That is the chance variation the design accepts, and it makes the difference between fertilizers harder to pin down.

What a randomized block design is

A randomized block design has two stages.

  1. Form blocks. Group the units by a variable you can observe before the experiment and that you expect to affect the response. For the garden, put the 12 sunny plots in one block and the 12 shady plots in another.
  2. Randomize within each block. Inside the sunny block, randomly choose 6 plots for fertilizer A and give B to the other 6. Do the same, separately, inside the shady block.

Each fertilizer now has exactly 6 sunny and 6 shady plots. Sunlight cannot favor either one. Fertilizers are compared within each block, sunny against sunny and shady against shady, and the results are then combined.

Blocks are formed on purpose, never at random. The randomness is in the assignment of treatments inside them. Blocking in an experiment plays a role similar to stratifying in a sample, and the two are compared in blocking vs stratifying.

The differences side by side

FeatureCompletely randomizedRandomized block
Grouping before randomizationNoneUnits sorted into blocks of similar units
Number of randomizationsOne, over all unitsOne inside each block
Balance on the blocking variableLeft to chanceGuaranteed
Variation from the blocking variableStays in the treatment comparisonSeparated out of the treatment comparison
RequiresNothing known about the unitsA variable known before assignment and related to the response
Best whenUnits are similar, or no useful blocking variable is knownUnits differ on a known variable that affects the response

Both designs allow cause-and-effect conclusions, because both use random assignment. Blocking changes how precisely the treatment effect is measured.

When to use which

Block when you can name a variable, measurable before the experiment, that you expect to change the response. Soil type for crops, age group for a medical treatment, and class period for a teaching method are typical. The more strongly that variable is related to the response, the more blocking helps.

Use a completely randomized design when the units are already similar, or when you have no reason to think any measurable variable matters much. Blocking on a variable unrelated to the response adds steps and gains no precision.

A block must be large enough to hold every treatment. When there are two treatments and each block has two units, the design is called matched pairs, covered in matched pairs design. Writing either design out step by step is covered in how to design an experiment.

The classic mix-up and how to avoid it

The most common error is treating blocks as treatment groups. "Give fertilizer A to the sunny plots and fertilizer B to the shady plots" is not blocking. In that plan sunlight and fertilizer change together, so their effects cannot be separated, which is confounding. In a block design, every block receives every treatment.

A related error is saying the units were randomly assigned to blocks. Units are placed in blocks according to a characteristic they already have. A plot is sunny or shady before the experiment starts. Only the treatments are assigned at random.

When you describe a block design on the exam, say three things: what the blocks are and why that variable matters for the response, that treatments are randomly assigned within each block, and that the treatments are compared within blocks.

How often a completely randomized design is unbalanced

A garden has 24 plots, 12 sunny and 12 shady. Two fertilizers will each go on 12 plots. Count the possible assignments under a completely randomized design and under a randomized block design with sunlight as the blocking variable. Then find the probability that the completely randomized design happens to give each fertilizer exactly 6 sunny plots.

  1. Completely randomized: choose 12 of the 24 plots for fertilizer A. The count is (2412)=2,704,156\binom{24}{12} = 2{,}704{,}156.

  2. Randomized block: within the sunny block choose 6 of 12 plots for A, which can be done in (126)=924\binom{12}{6} = 924 ways. The shady block also has 924 ways.

  3. The blocks are randomized separately, so multiply: 924×924=853,776924 \times 924 = 853{,}776 assignments.

  4. Every one of those 853,776 assignments gives each fertilizer exactly 6 sunny and 6 shady plots, and they are the only assignments that do.

  5. Under the completely randomized design all 2,704,156 assignments are equally likely, so the probability of exact balance is 853,7762,704,156≈0.316\frac{853{,}776}{2{,}704{,}156} \approx 0.316.

  6. The probability of some imbalance in sunlight is 1−0.316=0.6841 - 0.316 = 0.684.

The completely randomized design has 2,704,156 possible assignments and the block design has 853,776. A completely randomized assignment balances sunlight exactly only about 31.6% of the time, so about 68.4% of the time one fertilizer gets more sunny plots than the other. The block design is balanced on sunlight every time.

Comparing treatments within blocks

A small block experiment uses 8 plots, 4 sunny and 4 shady, with each fertilizer randomly assigned to 2 plots in each block. Yields in kilograms: sunny plots with A: 52 and 54; sunny with B: 48 and 50; shady with A: 41 and 43; shady with B: 38 and 40. Estimate the effect of fertilizer A compared with B, and show how much variation sunlight adds to the yields.

  1. Find the mean yield for each fertilizer in the sunny block: A is 52+542=53\frac{52 + 54}{2} = 53 and B is 48+502=49\frac{48 + 50}{2} = 49. The difference is 53−49=453 - 49 = 4 kilograms.

  2. Do the same in the shady block: A is 41+432=42\frac{41 + 43}{2} = 42 and B is 38+402=39\frac{38 + 40}{2} = 39. The difference is 42−39=342 - 39 = 3 kilograms.

  3. Combine the two blocks, which are the same size, by averaging: 4+32=3.5\frac{4 + 3}{2} = 3.5 kilograms.

  4. Check against the overall means: A averages 52+54+41+434=47.5\frac{52 + 54 + 41 + 43}{4} = 47.5 and B averages 48+50+38+404=44\frac{48 + 50 + 38 + 40}{4} = 44, and 47.5−44=3.547.5 - 44 = 3.5. The two routes agree because each fertilizer has the same number of plots in each block.

  5. Measure the variation sunlight adds. The four A yields (52, 54, 41, 43) have standard deviation about 6.45 kilograms. Within one block, the two A yields (52 and 54) have standard deviation about 1.41 kilograms.

  6. Compare the effect to each spread. The 3.5 kilogram fertilizer effect is small next to 6.45 and large next to 1.41.

Fertilizer A yields about 3.5 kilograms more than B: 4 more in the sunny block and 3 more in the shady block. Yields vary by about 6.45 kilograms when sunny and shady plots are mixed and by about 1.41 kilograms within a block. Comparing within blocks sets the 3.5 kilogram effect against the smaller spread, which is why blocking makes the comparison more precise.

Frequently asked questions

Does a randomized block design still use random assignment?

Yes. The blocks are formed from a characteristic the units already have, and then treatments are randomly assigned to the units inside each block. Without that random step it would not be an experiment that supports a cause-and-effect conclusion.

Is a completely randomized design wrong when the units differ?

No. It is still a valid design, because random assignment gives every unit the same chance of each treatment. The differences among units stay in the data as extra variation, so the treatment effect is estimated less precisely than it would be with a well-chosen blocking variable.

How do I choose a blocking variable?

Pick a variable that you can observe before treatments are assigned and that you expect to be related to the response. Units inside a block should be as alike as possible on it. A variable with no connection to the response makes a poor blocking variable, because removing its variation changes nothing.