Reliability

Types of Reliability

As noted, we can establish reliability for both scores on measurement scales (attitudinal or general knowledge tests) and content analysis coding. All statistical reliability tests provide indices that range from 0 to 100, either as a correlation or as a percentage-of-agreement score. Further, the type of data acquired dictates which specific tests are employed. In general, data can be defined as being either categorical (i.e., the data represent frequency counts and percentages of specified categories, such as yes or no, male or female, good or bad) or continuous (i.e., the data represent responses on a continuum where the distance between one unit and the other is equal, such as age or income).

Measurement Reliability

In traditional measurement a measure's reliability in general has been established by examining how people respond to the statement in a measure. If the measure represents an attempt to assess attitudes or beliefs, then reliability is established via statistical analysis for the appropriate type of data the measures yields. If the measure is to be used over time, a different type of reliability is employed—one that is more applied and typically relies on correlational analysis.

When creating an attitude measure or scale, the researcher attempts to predict how individuals feel or evaluate some abstract object such as credibility or persuasiveness. Although there are many different types of attitudinal measures, public relations typically employs what is known as the Likert-type scale, which requires people to respond to a series of statements as to whether they (5) strongly agree, (4) agree, (3) neither agree nor disagree, (2) disagree, or (1) strongly disagree with each statement. The responses are then summed across items. This assumes the data are continuous in nature and the computed score represents the participants’ evaluation on the item. Reliability for Likert-type scales is established using coefficient alpha statistics. A coefficient alpha of .80 or better is generally accepted as good to excellent, an alpha of .70 adequate, and an alpha less than .70 problematic. An example of this type of measure would be [Page 785]James E. Grunig and Linda Childers Hon's 1999 measure of relationships. A second attitudinal measure asks people to pick only the statements that they agree with. This type of measure yields categorical data, and its reliability is established statistically by the KR-20 statistic.

...