A randomized block design is an experimental design in which the experimental units are placed in blocks. Randomly, the treatments are assigned to each block’s experimental units. If all treatments are present in every block at least once, then we have a fully randomized block design. We have an incomplete random block design.
This design minimizes the impact of systematic error. The effects of variations between blocks should not be ignored if the experimenter is focusing on only the differences in treatments.
A randomized block design consists of participants with similar characteristics being grouped together to create blocks. Then, the treatment or intervention is randomly assigned within each block.
The randomized block design aims to create groups of participants that are similar and can then be compared.
Randomized block design is the equivalent of stratified randomly sampling. Randomized block designs, like stratified sampling are designed to reduce noise and variance in the data (see Classifying Experimental Designs). How does it work? The researcher must divide the sample into homogeneous blocks or subgroups (analogous with “strata” in stratified samples). Next, you will implement the experimental design that you have in mind within each block or homogenous subgroup. It is important to remember that each block has a lower variability than the whole sample. Each block’s treatment effect estimate is therefore more efficient than the total sample estimates. These more efficient estimates should be combined across blocks to produce a more efficient overall estimate than without blocking.
This is a simple example. Let’s say that we initially intended to do a simple posttest-only randomized experimental design. We recognize that there are many subgroups within our sample. In a college student study, for example, it might be expected that students would be relatively homogeneous in terms of year or class. We decide to divide the sample into four groups: sophomore, junior and senior. If our intuition is right, and the class variability is lower than that of the whole sample, then we can probably estimate the treatment effect in each block more accurately (see the discussion Statistical Power). We would use the simple post-only random experiment in each block.
These are some things you should know about this strategy. To begin with, an outside observer may not see that you are blocking. Each block would have the same design. There is no reason why people from different blocks should be separated or segregated. Blocking doesn’t affect the work you do with research participants. Blocking is a strategy to group people in your data analysis in an effort to reduce noise. It is an analysis strategy. A blocking design will only be beneficial if your intuition is correct that the blocks are more homogeneous then the whole sample. You will be hurt if you are wrong. If different college-level classes aren’t fairly homogeneous in relation to your measurements, you will not benefit from blocking. How can you tell if blocking is a good idea or not? It is important to assess whether the groups are homogeneous. Is it reasonable to assume that freshmen have more in common with sophomores and juniors when you measure political attitudes? Are they more similar in terms of measures related to drug abuse? The researcher must ultimately make the final decision about whether to block the research.
