Reliability

Consistency

Measurements of underlying constructs related to oral and written communication typically involve multiple observations that are summed or averaged to form a score or set of subtest scores. A test’s internal consistency refers to the relationships between the individual test items and the overall test score. For example, the average MLU for children between the ages of 3 years, 0 months and 3 years, 6 months is between 3.0 and 3.75 morphemes per utterance. Typical utterances of this length include talking about actions (e.g., “He walks” or “He is walking”). So an adult who asks the child questions about actions during an adult–child interaction would be providing test items like “What is he doing?” that would be consistent with the child’s ability and would probably result in relatively consistent responses by the child. Test items that ask the child to identify objects in the room (e.g., “What is this?” and “Who is this doll?”) would likely result in shorter utterances (e.g., “a ball” and “the mommy”) with scores lower than the typical child’s test average. Thus, some of the adult questions would result in child responses that are more consistent with the child’s overall score than others.

Clinicians and experimenters who construct tests use measures of internal consistency to improve their test procedures. Items not consistent with the total score are, at best, an inefficient part of the measurement. At worst, a lack of relationship between a test item and overall score indicates that the item is not measuring the right underlying ability and thus is not a valid measure. Responses to the unrelated items vary randomly relative to the person’s targeted ability and thus add unwanted variability to the test scores, making the scores less repeatable.

...