Statistics Glossary → Blocking (block variable)
Blocking (block variable)concept
A design principle that groups experimental units into similar sets (blocks) according to a characteristic that affects the outcome but is not of interest, and compares treatments within each block, removing that characteristic's variability from the error term.
Sources such as soil fertility in a field trial, the working day or instrument in a laboratory, or the centre in a clinical trial change the outcome. If they are not blocked, the variability they create inflates the error variance and makes treatment differences harder to see. In a randomised complete block design each block contains all treatments, which are assigned at random within the block. In the analysis the block enters the model as a factor, separating the between-block differences from the error term and making treatment comparisons more precise.
Blocking is summed up by the design principle 'block what you can, randomise what you cannot'. If the blocks do not really differ, the block term uses up a few degrees of freedom and costs a little power. The block variable is usually not of interest and is assumed not to interact with treatment; if an interaction is plausible, it needs to be examined separately. Blocking is done at the design stage; a variable added after the experiment is handled as a covariate or by post-stratification.
An everyday example
A chemical engineer tests three catalysts, but the reactor behaves slightly differently from day to day. Each day she tests all three catalysts in random order and uses 'day' as the block. Day-to-day differences then do not contaminate the catalyst comparison, and the same number of runs gives more precise results.
In MerQur
In response surface regression, the 'Block (optional)' field enters a categorical block variable into the model as dummy variables before the surface terms.
Analyses where this concept appears
Response Surface Regression (RSREG)