Split-Half Reliability

Measurement is fundamental to almost all forms of research and applied science. To conduct quantitative research, scientists must measure at least one variable. For example, researchers studying the effect of social rejection on self-esteem must measure participants’ self-esteem in some way. Similarly, to apply scientific knowledge, practitioners often rely heavily on measurement. For example, school psychologists measure children's academic and cognitive aptitudes to place them in appropriate classes and to identify potential academic difficulties. Given the importance of measurement, researchers and practitioners must evaluate the quality of the measurement tools that they use. Reliability is a key facet of measurement quality, and split-half reliability is a method of estimating the reliability of a measurement instrument.

Computing and Interpreting Split-Half Reliability

The split-half method of estimating reliability is most directly applicable to instruments that have multiple items. Indeed, many instruments in behavioral research are tests, questionnaires, [Page 1411]inventories, or surveys that include two or more items.

Table 1 Split-Half Reliability Example Data

Note: SD = standard deviation.

Consider the hypothetical set of responses in Table 1. Imagine that a researcher wishes to estimate the reliability of a four-item test of trait self-esteem, in which each item presents a statement relevant to self-esteem (e.g., “I often feel that I am a good person”). People respond to each item using a seven-point scale indicating their level of agreement with the statements (e.g., 1 = strongly disagree, 4 = neutral, and 7 = strongly agree)—thus, larger numbers reflect greater self-esteem. Peoples’ responses are summed to create a total score indicating their level of trait self-esteem. Of course, many good tests include negatively keyed items, for which an endorsement or agreement reflects a low level of the characteristic being measures (e.g., “I rarely feel like I'm a good person”). Such items must be reverse scored before scoring the scale and evaluating reliability. As shown in Table 1, Person 2 has the highest level of self-esteem and Person 4 has the lowest. Being aware that scores on the instrument might be affected by measurement error, the researcher estimates the reliability of these scores.

...