D-3: Identify defining features of single-subject experimental designs (e.g., individuals serve as their own controls, repeated measures, prediction, verification, replication)©
Want this as a downloadable PDF? Click here!
Target terms (or phrases, in this case): Individuals serve as their own controls, repeated measures, prediction, verification, control
Two important things to know about single subject experimental designs:
(1) They are not the same as case studies. Case studies are clinical stories that just tell what happened without manipulation of the environment. Experimental designs involve deliberate manipulation of the environment to answer a particular question.
(2) The logic behind single subject experimental designs applies to the everyday work of programming for behavior change with clients. Even if we never publish a research study in a peer reviewed journal, or participate in any kind of formal research, we still need to be very familiar with the methodology in order to evaluate our work as behavior analysts. This is not optional – it’s an integral part of our work.
Individuals serve as their own controls
Please note that “subject” and “participant” are two words that can be used to describe someone (or a unit of people or animals) in a research study. “Participant” is often a preferred term because it emphasizes that people who take part in studies have rights and are actively part of the process.
Definition: Individuals serve as their own controls in a research study when the effects of an intervention are measured on the person themselves, not between one person who got the intervention (treatment) and one person who didn’t (control). In single subject methodology, the individual is essentially assigned to both treatment and control, because the research question is answered differently from other kinds of research. (See D-4 for more about what this all means.)
Example in clinical context: Tami is designing an intervention for her client Ariel, who needs help with remembering to complete her homework. Tami takes baseline data until stability is achieved, then introduces an intervention (series of alarms on Ariel’s phone) and continues to take data until stability is once again achieved. She then returns Ariel to the baseline condition and then introduces the intervention a second time, following the same process as before. The data depicting the dependent variable (Ariel’s homework completion behavior) show a clear relationship between the presence of the alarms and the completion of homework. In this example, no other person was used as a “control” for Ariel. That would not have been a great way to answer the question about how to help Ariel do her homework, since she might have special considerations and circumstances that are unique. Instead, she was the only subject, and the intervention was evaluated using Ariel herself in all phases.
Why it matters: Using subjects as their own controls matters a lot in behavior analysis. It enables us to take into account the unique idiosyncrasies (“weirdness”) of each individual person. We’re all different, and we’re all weird in some way. When we use big groups to answer research questions, one of the goals is to make the groups so big that the numbers “drown out” the differences between people by statistically canceling each other out. There’s nothing inherently wrong with this, but it doesn’t work for us as behavior analysts. We are interested in functional relations between individual behavior and experimental conditions! To do that, we need to study the individual and their environment, and how the two interact. The vast majority of research published in behavior analytic journals was conducted using single subject methodology.
Definition: When we use single subject experimental designs, we need to capture something to measure to see if our intervention is working. That thing we measure is called a dependent variable.
Examples in clinical context: Randi engages in swearing and property destruction. His team creates an intervention plan for him. In order to empirically answer questions about whether the intervention is working, the team carefully defines and records instances of Randi’s target behaviors.
This also works with skill acquisition (behavior we are teaching so that they will increase). For example, say your client Tanisha needs more skills related to asking for help. We could use “asking for help” as the dependent variable and measure it multiple times throughout the baseline and intervention.
Why it matters: Using repeated measures is super important, because if we only measure the dependent variable once or twice, we won’t be able to thoroughly see what our data points are telling us. Take a look at C-11 (interpret graph data) to understand more about how repeated measures help us analyze level, trend, and variability of data.
Definition: Prediction is looking at the data we have and making an informed guess about where it would go if we kept all variables the same (i.e. if we didn’t change anything). Take a look at C-11 (interpret graph data) for more on how to predict where data will head next.
Example in clinical context: Johnny is a client who is being assessed at a severe behavior clinic due to self injury. His team conducts observations and they highly suspect that his self injury is maintained by access to attention (connection with other people). The team conducts a functional assessment (baseline) condition in which they give Johnny attention every time his self injury occurs. Team members observe that Johnny engages in self injury every single time attention is withdrawn, and stops once he receives attention. After observing this multiple times, graphing the results, and engaging in visual analysis, the team predicts that, if they keep providing attention contingent on self injury, the target behavior will continue as before.
Why it matters: Behavior analysts need to predict what the dependent variable would look like if everything else stayed the same in order to design experiments that demonstrate that an independent variable can change the otherwise predicted outcome.
Definition: Verification is demonstrating that baseline levels of behavior would have remained without introducing the independent variable (intervention). Verification as a concept can take several forms within a research design, but the foundational idea is the same.
Example in clinical context: Let’s take the example of Johnny above. The team moved into the intervention phase. Now his team ignores self injury and they have taught Johnny to use an “I want attention” button instead, which is always reinforced with attention. Johnny’s levels of self injury are down significantly! To demonstrate that their intervention, rather than something else (such as medication), was responsible for the change in self injury, Johnny’s team could take the button away and start reinforcing self injury again. If Johnny’s behavior looks similar to what it was at the first baseline phase during the assessment, then that is evidence that the intervention was responsible for the change in behavior.
Why it matters: It’s important to go beyond educated guesses and actually demonstrate in a logical manner that our interventions are responsible for the change in behavior that the client needs.
Definition: Replication is strengthening the case that the independent variable is responsible for changes in behavior by demonstrating it multiple times.
Example in clinical context: Let’s keep talking about Johnny from before. His team could strengthen the probability that their intervention was responsible for the change in the self injury by implementing baseline and intervention/treatment conditions several more times. If the self injury stays low in each intervention phase and high in each baseline phase, each repetition of that change would lend further credibility to the functional relationship.
Replication can also happen in “smaller” or “bigger” ways. In terms of more micro replications, we can often see within-session replication, such as Johnny engaging in the target behavior every time attention was withdrawn, over and over again, within a single assessment session (e.g., 10 minutes). We also see replication on a broader scale, such as if the researchers utilized similar methodology for many other individuals who also engaged in attention-maintained self injury and found similar treatment results.
Why it matters: Every time a possible relationship between two variables is demonstrated, it becomes less and less likely that “chance” or some other factor was primarily responsible for the relationship between the dependent and independent variables. Replication as a concept can take several forms within a research design, but the foundational idea is the same. (See “D-2 distinguish between internal and external validity” for more on how replication ties into validity.)