Applied Behavior Analysis: Guidelines

3. ANALYTIC

Applied behavioral research is analytic to the extent that it believably demonstrates that certain events were responsible for an observed behavior change. Believability is enhanced with each successful replication. Replication is accomplished by various types of experimental manipulations, known as research designs. We obviously cannot do justice to all the research designs employed in applied behavior analysis (for an introduction, see Martin & Pear, 1998, Chapter 21); however, we will briefly mention two of the more common ones to illustrate how applied researchers attempt to improve our confidence in the effectiveness of a treatment variable.

Many behavior research programs track the target behavior over time. Typically, the target behavior is first measured in the absence of the treatment variable; this is called the baseline (or the A) phase. When a stable level is observed, the treatment variable is introduced; this is called the treatment (or the B) phase. Suppose the target behavior changes from baseline to treatment. Can we conclude that the treatment variable was responsible? We cannot say for certain because some other unknown variable may have been introduced at the same time. For this reason, behavioral researchers often employ an ABAB design. The treatment variable is withdrawn in a third phase (i.e., baseline conditions are reinstated) in order to determine if the target behavior returns to its pre-treatment level; if it does, then the treatment variable is applied once again in an attempt to recover the behavior change that was observed during the first intervention. Any given confounding variable introduced when the treatment variable was initially applied is unlikely to have been subsequently withdrawn and reintroduced, respectively, at the very same time as the treatment variable, and thus can be ruled out as a cause. In effect, in an ABAB design the target behavior is turned on and off by successive applications of the treatment variable, which constitutes a believable demonstration of control over that behavior (see Figure 1).

The ABAB design presents special problems for applied researchers. For example, it may not be ethically justified to return to baseline after eliminating a life-threatening behavior. Furthermore, desirable behavior change make become trapped by natural contingencies of reinforcement over which the researcher has little control. For example, a shy child may be reinforced with special treats for interacting with her peers. If her behavior improves, she is likely to meet new friends and have fun playing with them. These natural outcomes of increased social interaction may function to maintain her socializing even when the special treats are withdrawn as consequences. In other words, it may be practically impossible to reverse the behavior change to demonstrate control. For these and other reasons, multiple baseline designs are common in applied behavior analysis.

In one type of multiple baseline design, several behaviors are tracked over time concurrently until they show stability. The treatment variable is then applied successively to each of the behaviors. If each one changes only when the treatment variable is applied to it, then we have greater confidence in the treatment variable as the causative factor. To illustrate this, consider the following scenario (refer to Figure 2). Johnny's homework behavior in three subject areas (math, science, history) is recorded. After a baseline period, Johnny is allowed to stay up late an extra 15 minutes when he presents evidence to his parents that he has completed his math homework. His parents observe that the duration of time he spends each night doing his math homework increases. While it is tempting to attribute this behavior change to the "stay-up-late" consequence, we need also be aware that some other important events could happened at the same time. For example, Johnny's parents may have told him that if he didn't start doing better in school then he would have to quit the hockey team. One advantage of the multiple baseline is that it helps us discount such factors as causes. In this case, if the parental warning resulted in Johnny doing more math homework, we would also expect to see similar improvements in his other school work (i.e., his science and history homework)---our multiple baseline design clearly shows that this did not happen. The time he spent doing his science and history homework increased only later when treatment was successively applied to each of those behaviors. It is unlikely the warning occurred only for math when treatment was applied to it, only for science when treatment was later applied to it, and only for history when treatment was even later applied to it. Thus, the multiple baseline design, like the ABAB design, controls for confounding variables and replicates the treatment effect. The replication is this example occurs across behaviors. We are better assured that the treatment variable was the cause of the behavior change.

Besides replication, there are six other commonly used guidelines for assessing the effectiveness of a treatment variable: "the fewer the overlapping points between baseline and treatment phases, the sooner the effect is observed following the introduction of the treatment, the larger the effect in comparison to baseline, the more precisely the treatment procedures are specified (see Technological section), the more the reliable the response measures (see Behavioral section), and the more consistent the findings with existing data and accepted behavioral theory (see Conceptual section)" (Martin & Pear, 1998, pp. 280-281).

(For more information on controlling for confounding variables, check out our Internal Validity tutorial.)

Illustrative Example/Nonexample Pair

Nonexample

Reports by classroom teachers, therapists, and teachers indicated that autistic children attending biweekly therapy sessions were unmotivated and engaged in frequent self-stimulatory behavior (e.g., rocking back and forth). It was agreed that a reasonable goal was to increase correct task responding from each child's academic curriculum (e.g., "touch your nose" versus "touch my nose"). Previous research suggested that identifying reinforcers for autistic children can be difficult, in that these children often do not respond to stimuli that interest other children (e.g., toys) or to social reinforcers (e.g., praise). The researchers were aware of the Premack Principle, which states that the opportunity to engage in a behavior that occurs frequently can be used to reinforce a behavior that occurs less often. With this in mind, the intervention included prompting the children to engage in 3-5 seconds of self-stimulation following correct task responding. The children's parents were given special invitations to come and see this procedure in operation. To see the data for one child named Anna, see Figure 3.

Example

Reports by classroom teachers, therapists, and teachers indicated that autistic children attending biweekly therapy sessions were unmotivated and engaged in frequent self-stimulatory behavior (e.g., rocking back and forth). It was agreed that a reasonable goal was to increase correct task responding from each child's academic curriculum (e.g., "touch your nose" versus "touch my nose"). Previous research suggested that identifying reinforcers for autistic children can be difficult, in that these children often do not respond to stimuli that interest other children (e.g., toys) or to social reinforcers (e.g., praise). The researchers were aware of the Premack Principle, which states that the opportunity to engage in a behavior that occurs frequently can be used to reinforce a behavior that occurs less often. With this in mind, the intervention included prompting the children to engage in 3-5 seconds of self-stimulation following correct task responding. The children's parents were given special invitations to come and see this procedure in operation. To see the data for one child named Anna, see Figure 4.

Analysis

The first item is not analytic in the sense that it does not provide a convincing demonstration that the treatment variable was the reason for the behavior change. It is true that Anna's performance improved when the teacher first started following her correct responding with the opportunity to engage in self-stimulatory behavior. However, this is confounded with the fact that Anna's parents may have started attending the training sessions at that very same point in time. The research design illustrated in the figure does not allow us to decide which of these two new events in Anna's life altered her correct responding.

The second item is analytic because it does provide a convincing demonstration. If the presence of Anna's parents in the classroom contributed to increasing one type of correct responding ("touch your nose" versus "touch my nose") then we would expect it to also increase the two other types of correct responding. However, neither behavior was enhanced until the treatment variable was applied to it. The presence of Anna's parents is ruled out as a cause and the effectiveness of the treatment variable is replicated across behaviors.

Related Source: Charlop, Kurtz, & Casey (1990)