Types of Measurement Errors
Several factors can occur during the measurement process that can increase random error. These factors include (1) transient personal factors, such as fatigue, hunger, attention span, health, mood, mental status, and motivation; (2) situational factors, such as a hot stuffy room, distractions, the presence of significant others, rapport with the researcher, and the playfulness or seriousness of the situation; (3) variations in the administration of the measurement procedure, such as interviews in which wording or sequence of questions is varied, questions are added or deleted, or researchers code responses differently; and (4) processing of data, such as errors in coding, accidentally marking the wrong column, punching the wrong key when entering data into the computer, or incorrectly totaling instrument scores (Devon et al., 2007; Waltz et al., 2010).
If you were to measure a variable for three subjects and diagram the random error, it might appear as shown in Figure 16-2. The difference between the true score of subject 1 (T1) and the observed score (O1) is two positive measurement intervals. The difference between the true score (T2) and observed score (O2) for subject 2 is two negative measurement intervals. The difference between the true score (T3) and observed score (O3) for subject 3 is zero. The random error for these three subjects is zero (+2 − 2 + 0 = 0). In viewing this example, one must remember that this is only a means of conceptualizing random error.
Measurement error that is not random is referred to as systematic error. A scale that weighs subjects 3 pounds more than their true weights is an example of systematic error. All of the body weights would be higher, and, as a result, the mean would be higher than it should be. Systematic error occurs because something else is being measured in addition to the concept. A conceptualization of systematic error is presented in Figure 16-3. Systematic error (represented by the shaded area in the figure) is due to the part of A-1 that is outside of A. This part of A-1 measures factors other than A and biases scores in a particular direction.
Another effective means of diminishing systematic error is to use more than one measure of an attribute or a concept and to compare the measures. To make this comparison, researchers use various data collection methods, such as scale, interview, and observation. Campbell and Fiske (1959) developed a technique of using more than one method to measure a concept, referred to as the multimethod-multitrait technique. More recently, the technique has been described as a version of mixed methodology, as discussed in Chapter 10. These techniques allow researchers to measure more dimensions of abstract concepts, and the effect of the systematic error on the composite observed score decreases. Figure 16-4 illustrates how more dimensions of concept A are measured through the use of four instruments, designated A-1, A-2, A-3, and A-4.
In some studies, researchers use instruments to examine relationships. Consider a hypothesis that tests the relationship between concept A and concept B. In Figure 16-5, the shaded area enclosed in the dark lines represents the true relationship between concepts A and B, such as the relationship between anxiety and depression. For example, two instruments, A-1 (Spielberger State Anxiety Scale) and B-1 (Center for Epidemiological Studies Depression Scale, Radloff, 1977), are used to examine the relationship between concepts A and B. The part of the true relationship actually reflected by A-1 and B-1 measurement methods is represented by the colored area in Figure 16-6. Because two instruments provide a more accurate measure of concepts A and B, more of the true relationship between concepts A and B can be measured.
If additional instruments (A-2 and B-2) are used to measure concepts A and B, more of the true relationship might be reflected. Figure 16-7 demonstrates with different colors the parts of the true relationship between concepts A and B that is measured when concept A is measured with two instruments (A-1 and A-2) and concept B is measured with two instruments (B-1 and B-2).