Blocking Variable

Careful, systematic planning for the design of a research study is as important as careful, systematic planning for the data analysis. Experimental research is generally considered to be the gold standard of scientific research design due to the rigorous nature of the experimental process. For example, an experimental research design manipulates an independent variable to observe changes in the dependent variable. Specifically, participants are randomly assigned to a treatment group (e.g., stimulus group) or condition group (e.g., no treatment, or control, group) of an independent variable. Random assignment increases controls for extraneous variables by ensuring that all subjects have an equal chance of being in any condition or treatment. However, it is possible for the effects of the independent variable to be obscured by extraneous or confounding variables, even after randomization. If the variance of a sample set is very high for a particular study variable, the statistical power of analysis of that variable is greatly reduced. One way to reduce the variance within sample sets is to design a research study such that sample data are collected independently for groups, or blocks, that are expected to vary significantly with respect to study variables of interest. These blocks are determined based on the value of one or more characteristics of the study population. Characteristics chosen to define the blocks are called blocking variables. The remainder of this entry further defines blocking variables and explains their importance, how to select them prior to data collection, and how to analyze the data within each block.

Each sample block is homogeneous relative to each blocking variable because all individuals within the block share the same value for each blocking variable. For example, women and men, on average, vary significantly with respect to weight. If weight is a variable of interest for a particular research study, sample data for that study can be collected independently for a “female” block and a “male” block based on the value of the blocking variable sex. After data collection, each block will be homogeneous relative to value of the sex variable (e.g., all individuals in the “female” block might share the characteristic sex = 1). Consequently, the variance of each sample set relative to weight is expected to be significantly less than if both women and men were included in the same sample set. Furthermore, the statistical power of any analysis involving weight is expected to be greater for the homogeneous sample blocks (i.e., female-only and male-only) than for a heterogeneous sample that includes both women and men.

A blocking variable may be any continuous variable (e.g., age, weight), ordinal category (e.g., college-level, high-school ranking), or nominal level data (e.g., sex, occupation, major). For a continuous variable to be used, it must first be transformed into categorical data (e.g., the continuous variable age can be grouped into the following categories: “child” for ages 17 years and younger, “adult” for [Page 99]ages 18 through 64 years, and “senior” for ages 65 years and older). Blocking variables may also be defined by a combination of characteristics (e.g., sex, age, education level) or a score based on a composite index of baseline characteristics. For example, a research study may be designed using the blocking variables sex (“female” and “male”) and high-school class (“freshman,” “sophomore,” “junior,” and “senior”), resulting in eight blocks for which data will be collected independently (“female freshman,” “male freshman,” “female sophomore,” etc.).

...